freebsd-skq/share/man/man9/mutex.9
Konstantin Belousov 90b581f2cc Implement mtx_trylock_spin(9).
Discussed with:	bde
Reviewed by:	jhb
Sponsored by:	The FreeBSD Foundation
MFC after:	1 week
Differential revision:	https://reviews.freebsd.org/D7192
2016-07-23 05:30:55 +00:00

561 lines
15 KiB
Groff

.\"
.\" 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 $
.\" $FreeBSD$
.\"
.Dd July 18, 2016
.Dt MUTEX 9
.Os
.Sh NAME
.Nm mutex ,
.Nm mtx_init ,
.Nm mtx_destroy ,
.Nm mtx_lock ,
.Nm mtx_lock_spin ,
.Nm mtx_lock_flags ,
.Nm mtx_lock_spin_flags ,
.Nm mtx_trylock ,
.Nm mtx_trylock_flags ,
.Nm mtx_trylock_spin ,
.Nm mtx_trylock_spin_flags ,
.Nm mtx_unlock ,
.Nm mtx_unlock_spin ,
.Nm mtx_unlock_flags ,
.Nm mtx_unlock_spin_flags ,
.Nm mtx_sleep ,
.Nm mtx_initialized ,
.Nm mtx_owned ,
.Nm mtx_recursed ,
.Nm mtx_assert ,
.Nm MTX_SYSINIT
.Nd kernel synchronization primitives
.Sh SYNOPSIS
.In sys/param.h
.In sys/lock.h
.In sys/mutex.h
.Ft void
.Fn mtx_init "struct mtx *mutex" "const char *name" "const char *type" "int opts"
.Ft void
.Fn mtx_destroy "struct mtx *mutex"
.Ft void
.Fn mtx_lock "struct mtx *mutex"
.Ft void
.Fn mtx_lock_spin "struct mtx *mutex"
.Ft void
.Fn mtx_lock_flags "struct mtx *mutex" "int flags"
.Ft void
.Fn mtx_lock_spin_flags "struct mtx *mutex" "int flags"
.Ft int
.Fn mtx_trylock "struct mtx *mutex"
.Ft int
.Fn mtx_trylock_flags "struct mtx *mutex" "int flags"
.Ft void
.Fn mtx_trylock_spin "struct mtx *mutex"
.Ft int
.Fn mtx_trylock_spin_flags "struct mtx *mutex" "int flags"
.Ft void
.Fn mtx_unlock "struct mtx *mutex"
.Ft void
.Fn mtx_unlock_spin "struct mtx *mutex"
.Ft void
.Fn mtx_unlock_flags "struct mtx *mutex" "int flags"
.Ft void
.Fn mtx_unlock_spin_flags "struct mtx *mutex" "int flags"
.Ft int
.Fn mtx_sleep "void *chan" "struct mtx *mtx" "int priority" "const char *wmesg" "int timo"
.Ft int
.Fn mtx_initialized "const struct mtx *mutex"
.Ft int
.Fn mtx_owned "const struct mtx *mutex"
.Ft int
.Fn mtx_recursed "const struct mtx *mutex"
.Pp
.Cd "options INVARIANTS"
.Cd "options INVARIANT_SUPPORT"
.Ft void
.Fn mtx_assert "const struct mtx *mutex" "int what"
.In sys/kernel.h
.Fn MTX_SYSINIT "name" "struct mtx *mtx" "const char *description" "int opts"
.Sh DESCRIPTION
Mutexes are the most basic and primary method of thread synchronization.
The major design considerations for mutexes are:
.Bl -enum
.It
Acquiring and releasing uncontested mutexes should be as cheap
as possible.
.It
They must have the information and storage space to support
priority propagation.
.It
A thread must be able to recursively acquire a mutex,
provided that the mutex is initialized to support recursion.
.El
.Pp
There are currently two flavors of mutexes, those that context switch
when they block and those that do not.
.Pp
By default,
.Dv MTX_DEF
mutexes will context switch when they are already held.
As an optimization,
they may spin for some amount
of time before context switching.
It is important to remember that since a thread may be preempted at any time,
the possible context switch introduced by acquiring a mutex is guaranteed
to not break anything that is not already broken.
.Pp
Mutexes which do not context switch are
.Dv MTX_SPIN
mutexes.
These should only be used to protect data shared with primary interrupt
code.
This includes interrupt filters and low level scheduling code.
In 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 either blocked or deferred on a processor
while holding a spin lock.
It is permissible to hold multiple spin mutexes.
.Pp
Once a spin mutex has been acquired it is not permissible to acquire a
blocking mutex.
.Pp
The storage needed to implement a mutex is provided by a
.Vt struct mtx .
In general this should be treated as an opaque object and
referenced only with the mutex primitives.
.Pp
The
.Fn mtx_init
function must be used to initialize a mutex
before it can be passed to any of the other mutex functions.
The
.Fa name
option is used to identify the lock in debugging output etc.
The
.Fa type
option is used by the witness code to classify a mutex when doing checks
of lock ordering.
If
.Fa type
is
.Dv NULL ,
.Fa name
is used in its place.
The pointer passed in as
.Fa name
and
.Fa type
is saved rather than the data it points to.
The data pointed to must remain stable
until the mutex is destroyed.
The
.Fa opts
argument is used to set the type of mutex.
It may contain either
.Dv MTX_DEF
or
.Dv MTX_SPIN
but not both.
If the kernel has been compiled with
.Cd "option INVARIANTS" ,
.Fn mtx_init
will assert that the
.Fa mutex
has not been initialized multiple times without intervening calls to
.Fn mtx_destroy
unless the
.Dv MTX_NEW
option is specified.
See below for additional initialization options.
.Pp
The
.Fn mtx_lock
function acquires a
.Dv MTX_DEF
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 block).
.Pp
The
.Fn mtx_lock_spin
function acquires a
.Dv MTX_SPIN
mutual exclusion lock
on behalf of the currently running kernel thread.
If another kernel thread is holding the mutex,
the caller will spin until the mutex becomes available.
Interrupts are disabled during the spin and remain disabled
following the acquiring of the lock.
.Pp
It is possible for the same thread to recursively acquire a mutex
with no ill effects, provided that the
.Dv MTX_RECURSE
bit was passed to
.Fn mtx_init
during the initialization of the mutex.
.Pp
The
.Fn mtx_lock_flags
and
.Fn mtx_lock_spin_flags
functions acquire a
.Dv MTX_DEF
or
.Dv MTX_SPIN
lock, respectively, and also accept a
.Fa flags
argument.
In both cases, the only flags presently available for lock acquires are
.Dv MTX_QUIET
and
.Dv MTX_RECURSE .
If the
.Dv MTX_QUIET
bit is turned on in the
.Fa flags
argument, then if
.Dv KTR_LOCK
tracing is being done,
it will be silenced during the lock acquire.
If the
.Dv MTX_RECURSE
bit is turned on in the
.Fa flags
argument, then the mutex can be acquired recursively.
.Pp
The
.Fn mtx_trylock
and
.Fn mtx_trylock_spin
functions attempt to acquire a
.Dv MTX_DEF
or
.Dv MTX_SPIN
mutex, respectively, pointed to by
.Fa mutex .
If the mutex cannot be immediately acquired, the functions will return 0,
otherwise the mutex will be acquired and a non-zero value will be returned.
.Pp
The
.Fn mtx_trylock_flags
and
.Fn mtx_trylock_spin_flags
functions have the same behavior as
.Fn mtx_trylock
and
.Fn mtx_trylock_spin
respectively, but should be used when the caller desires to pass in a
.Fa flags
value.
Presently, the only valid value in the
.Fn mtx_trylock
and
.Fn mtx_trylock_spin
cases is
.Dv MTX_QUIET ,
and its effects are identical to those described for
.Fn mtx_lock
above.
.Pp
The
.Fn mtx_unlock
function releases a
.Dv MTX_DEF
mutual exclusion lock.
The current thread may be preempted if a higher priority thread is waiting
for the mutex.
.Pp
The
.Fn mtx_unlock_spin
function releases a
.Dv MTX_SPIN
mutual exclusion lock.
.Pp
The
.Fn mtx_unlock_flags
and
.Fn mtx_unlock_spin_flags
functions behave in exactly the same way as do the standard mutex
unlock routines above, while also allowing a
.Fa flags
argument which may specify
.Dv MTX_QUIET .
The behavior of
.Dv MTX_QUIET
is identical to its behavior in the mutex lock routines.
.Pp
The
.Fn mtx_destroy
function is used to destroy
.Fa 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 thread blocked on the mutex
when it is destroyed.
.Pp
The
.Fn mtx_sleep
function is used to atomically release
.Fa mtx
while waiting for an event.
For more details on the parameters to this function,
see
.Xr sleep 9 .
.Pp
The
.Fn mtx_initialized
function returns non-zero if
.Fa mutex
has been initialized and zero otherwise.
.Pp
The
.Fn mtx_owned
function returns non-zero
if the current thread holds
.Fa mutex .
If the current thread does not hold
.Fa mutex
zero is returned.
.Pp
The
.Fn mtx_recursed
function returns non-zero if the
.Fa mutex
is recursed.
This check should only be made if the running thread already owns
.Fa mutex .
.Pp
The
.Fn mtx_assert
function allows assertions specified in
.Fa what
to be made about
.Fa mutex .
If the assertions are not true and the kernel is compiled with
.Cd "options INVARIANTS"
and
.Cd "options INVARIANT_SUPPORT" ,
the kernel will panic.
Currently the following assertions are supported:
.Bl -tag -width MA_NOTRECURSED
.It Dv MA_OWNED
Assert that the current thread
holds the mutex
pointed to by the first argument.
.It Dv MA_NOTOWNED
Assert that the current thread
does not hold the mutex
pointed to by the first argument.
.It 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 conjunction with
.Dv MA_OWNED .
.It 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 conjunction with
.Dv MA_OWNED .
.El
.Pp
The
.Fn MTX_SYSINIT
macro is used to generate a call to the
.Fn mtx_sysinit
routine at system startup in order to initialize a given mutex lock.
The parameters are the same as
.Fn mtx_init
but with an additional argument,
.Fa name ,
that is used in generating unique variable names for the related structures associated with the lock and the sysinit routine.
.Ss The Default Mutex Type
Most kernel code should use the default lock type,
.Dv MTX_DEF .
The default lock type will allow the thread
to be disconnected from the CPU
if the lock is already held by another thread.
The 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
.Dv MTX_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 another thread interrupted the thread which held a mutex
and then tried to acquire the mutex.
For this reason spin locks disable all interrupts on the local CPU.
.Pp
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 other synchronization primitives such as default mutexes,
thread scheduling, and interrupt threads.
.Ss Initialization Options
The options passed in the
.Fa opts
argument of
.Fn mtx_init
specify the mutex type.
One of the
.Dv MTX_DEF
or
.Dv MTX_SPIN
options is required and only one of those two options may be specified.
The possibilities are:
.Bl -tag -width MTX_NOWITNESS
.It Dv MTX_DEF
Default mutexes
will always allow the current thread to be suspended
to avoid deadlock conditions against interrupt threads.
The implementation of this lock type
may spin for a while before suspending the current thread.
.It Dv MTX_SPIN
Spin mutexes
will never relinquish the CPU.
All interrupts are disabled on the local CPU
while any spin lock is held.
.It Dv MTX_RECURSE
Specifies that the initialized mutex is allowed to recurse.
This bit must be present if the mutex is permitted to recurse.
.Pp
Note that neither
.Fn mtx_trylock
nor
.Fn mtx_trylock_spin
support recursion;
that is, attempting to acquire an already-owned mutex fails.
.It Dv MTX_QUIET
Do not log any mutex operations for this lock.
.It Dv MTX_NOWITNESS
Instruct
.Xr witness 4
to ignore this lock.
.It Dv MTX_DUPOK
Witness should not log messages about duplicate locks being acquired.
.It Dv MTX_NOPROFILE
Do not profile this lock.
.It Dv MTX_NEW
Do not check for double-init.
.El
.Ss Lock and Unlock Flags
The flags passed to the
.Fn mtx_lock_flags ,
.Fn mtx_lock_spin_flags ,
.Fn mtx_unlock_flags ,
and
.Fn mtx_unlock_spin_flags
functions provide some basic options to the caller,
and are often used only under special circumstances to modify lock or
unlock behavior.
Standard locking and unlocking should be performed with the
.Fn mtx_lock ,
.Fn mtx_lock_spin ,
.Fn mtx_unlock ,
and
.Fn mtx_unlock_spin
functions.
Only if a flag is required should the corresponding
flags-accepting routines be used.
.Pp
Options that modify mutex behavior:
.Bl -tag -width MTX_QUIET
.It Dv MTX_QUIET
This option is used to quiet logging messages during individual mutex
operations.
This can be used to trim superfluous logging messages for debugging purposes.
.El
.Ss Giant
If
.Va Giant
must be acquired, it must be acquired prior to acquiring
other mutexes.
Put another way: it is impossible to acquire
.Va Giant
non-recursively while
holding another mutex.
It is possible to acquire other mutexes while holding
.Va Giant ,
and it is possible to acquire
.Va Giant
recursively while holding other mutexes.
.Ss Sleeping
Sleeping while holding a mutex (except for
.Va Giant )
is never safe
and should be avoided.
There are numerous assertions which will fail if this is attempted.
.Ss Functions Which Access Memory in Userspace
No mutexes should be held (except for
.Va Giant )
across functions which
access memory in userspace, such as
.Xr copyin 9 ,
.Xr copyout 9 ,
.Xr uiomove 9 ,
.Xr fuword 9 ,
etc.
No locks are needed when calling these functions.
.Sh SEE ALSO
.Xr condvar 9 ,
.Xr LOCK_PROFILING 9 ,
.Xr locking 9 ,
.Xr mtx_pool 9 ,
.Xr panic 9 ,
.Xr rwlock 9 ,
.Xr sema 9 ,
.Xr sleep 9 ,
.Xr sx 9
.Sh HISTORY
These
functions appeared in
.Bsx 4.1
and
.Fx 5.0 .
The
.Fn mtx_trylock_spin
function was added in
.Fx 12.0 .