freebsd/share/man/man9/mtx_pool.9

.\"
.\" Copyright (C) 2002 Garrett Rooney <rooneg@electricjellyfish.net>.
.\" 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(s), this list of conditions and the following disclaimer as
.\"    the first lines of this file unmodified other than the possible
.\"    addition of one or more copyright notices.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\"    notice(s), this list of conditions and the following disclaimer in the
.\"    documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``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 THE COPYRIGHT HOLDER(S) 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.
.\"
.\" $FreeBSD$
.\"
.Dd February 6, 2010
.Dt MTX_POOL 9
.Os
.Sh NAME
.Nm mtx_pool ,
.Nm mtx_pool_alloc ,
.Nm mtx_pool_find ,
.Nm mtx_pool_lock ,
.Nm mtx_pool_lock_spin ,
.Nm mtx_pool_unlock ,
.Nm mtx_pool_unlock_spin ,
.Nm mtx_pool_create ,
.Nm mtx_pool_destroy
.Nd "mutex pool routines"
.Sh SYNOPSIS
.In sys/param.h
.In sys/lock.h
.In sys/mutex.h
.Ft "struct mtx *"
.Fn mtx_pool_alloc "struct mtx_pool *pool"
.Ft "struct mtx *"
.Fn mtx_pool_find "struct mtx_pool *pool" "void *ptr"
.Ft void
.Fn mtx_pool_lock "struct mtx_pool *pool" "void *ptr"
.Ft void
.Fn mtx_pool_lock_spin "struct mtx_pool *pool" "void *ptr"
.Ft void
.Fn mtx_pool_unlock "struct mtx_pool *pool" "void *ptr"
.Ft void
.Fn mtx_pool_unlock_spin "struct mtx_pool *pool" "void *ptr"
.Ft "struct mtx_pool *"
.Fn mtx_pool_create "const char *mtx_name" "int pool_size" "int opts"
.Ft "void"
.Fn mtx_pool_destroy "struct mtx_pool **poolp"
.Sh DESCRIPTION
Mutex pools are designed to be used as short term leaf mutexes;
i.e., the last mutex one might acquire before calling
.Xr mtx_sleep 9 .
They operate using a shared pool of mutexes.
A mutex may be chosen from the pool based on a supplied pointer,
which may or may not point to anything valid,
or the caller may allocate an arbitrary shared mutex from the pool
and save the returned mutex pointer for later use.
.Pp
The shared mutexes in the
.Va mtxpool_sleep
mutex pool,
which is created by default,
are standard, non-recursive,
blockable mutexes, and should only be used in appropriate situations.
The mutexes in the
.Va mtxpool_lockbuilder
mutex pool are similar, except that they are initialized with the MTX_NOWITNESS
flag so that they may be used to build higher-level locks.
Other mutex pools may be created that contain mutexes with different
properties, such as spin mutexes.
.Pp
The caller can lock and unlock mutexes returned by the pool routines, but
since the mutexes are shared, the caller should not attempt to destroy them
or modify their characteristics.
While pool mutexes are normally leaf mutexes
(meaning that one cannot depend on any ordering guarantees
after obtaining one),
one can still obtain other mutexes under carefully controlled circumstances.
Specifically, if one has a private mutex
(one that was allocated and initialized by the caller),
one can obtain it after obtaining a pool mutex if ordering issues are
carefully accounted for.
In these cases the private mutex winds up being the true leaf mutex.
.Pp
Pool mutexes have the following advantages:
.Pp
.Bl -enum -offset indent -compact
.It
No structural overhead;
i.e., they can be associated with a structure without adding bloat to it.
.It
Mutexes can be obtained for invalid pointers, which is useful when one uses
mutexes to interlock destructor operations.
.It
No initialization or destruction overhead.
.It
Can be used with
.Xr mtx_sleep 9 .
.El
.Pp
And the following disadvantages:
.Pp
.Bl -enum -offset indent -compact
.It
Should generally only be used as leaf mutexes.
.It
Pool/pool dependency ordering cannot be guaranteed.
.It
Possible L1 cache mastership contention between CPUs.
.El
.Pp
.Fn mtx_pool_alloc
obtains a shared mutex from the specified pool.
This routine uses a simple rover to choose one of the shared mutexes managed
by the
.Nm
subsystem.
.Pp
.Fn mtx_pool_find
returns the shared mutex associated with the specified address.
This routine will create a hash out of the pointer passed into it
and will choose a shared mutex from the specified pool based on that hash.
The pointer does not need to point to anything real.
.Pp
.Fn mtx_pool_lock ,
.Fn mtx_pool_lock_spin ,
.Fn mtx_pool_unlock ,
and
.Fn mtx_pool_unlock_spin
lock and unlock the shared mutex from the specified pool
associated with the specified address;
they are a combination of
.Fn mtx_pool_find
and
.Xr mtx_lock 9 ,
.Xr mtx_lock_spin 9 ,
.Xr mtx_unlock 9 ,
and
.Xr mtx_unlock_spin 9 ,
respectively.
Since these routines must first find the mutex to operate on,
they are not as fast as directly using the mutex pointer returned by
a previous invocation of
.Fn mtx_pool_find
or
.Fn mtx_pool_alloc .
.Pp
.Fn mtx_pool_create
allocates and initializes a new mutex pool of the
specified size.
The pool size must be a power of two.
The
.Fa opts
argument is passed to
.Xr mtx_init 9
to set the options for each mutex in the pool.
.Pp
.Fn mtx_pool_destroy
calls
.Xr mtx_destroy 9
on each mutex in the specified pool,
deallocates the memory associated with the pool,
and assigns NULL to the pool pointer.
.Sh SEE ALSO
.Xr locking 9 ,
.Xr mutex 9
.Sh HISTORY
These routines first appeared in
.Fx 5.0 .