freebsd-skq/share/man/man9/sleep.9

.\"
.\" Copyright (c) 1996 Joerg Wunsch
.\"
.\" 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.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE DEVELOPERS ``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 DEVELOPERS 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 December 17, 1998
.Os
.Dt SLEEP 9
.Sh NAME
.Nm sleep ,
.Nm msleep ,
.Nm tsleep ,
.Nm asleep ,
.Nm await ,
.Nm wakeup
.Nd wait for events
.Sh SYNOPSIS
.Fd #include <sys/param.h>
.Fd #include <sys/systm.h>
.Fd #include <sys/proc.h>
.Ft int
.Fn tsleep "void *ident" "int priority" "const char *wmesg" "int timo"
.Ft int
.Fn msleep "void *ident" "struct mtx *mtx" "int priority" "const char *wmesg" "int timo"
.Ft int
.Fn asleep "void *ident" "int priority" "const char *wmesg" "int timo"
.Ft int
.Fn await "int priority" "int timo"
.Ft void
.Fn wakeup "void *ident"
.Ft void
.Fn wakeup_one "void *ident"
.Sh DESCRIPTION
The functions
.Fn tsleep
and
.Fn wakeup
handle event-based process blocking.  If a process must wait for an
external event, it is put on sleep by
.Nm tsleep .
The parameter
.Ar ident
is an arbitrary address that uniquely identifies the event on which
the process is being asleep.  All processes sleeping on a single
.Ar ident
are woken up later by
.Nm wakeup ,
often called from inside an interrupt routine, to indicate that the
resource the process was blocking on is available now.
.Pp
The parameter
.Ar wmesg
is a string describing the sleep condition for tools like
.Xr ps 1 .
Due to the limited space of those programs to display arbitrary strings,
this message should not be longer than 6 characters.
.Pp
The
.Fn wakeup_one
function is used to make the first process in the queue that is
sleeping on the parameter
.Fa ident
runnable.  This can prevent the system from becoming saturated 
when a large number of processes are sleeping on the same address,
but only one of them can actually do any useful work when made
runnable.
.Pp
.Nm Tsleep
is the general sleep call.  Suspends the current process until a wakeup is
performed on the specified identifier.  The process will then be made
runnable with the specified
.Ar priority .
Sleeps at most
.Ar timo
\&/ hz seconds (0 means no timeout).  If
.Ar pri
includes the
.Dv PCATCH
flag, signals are checked before and after sleeping, else signals are
not checked.  Returns 0 if awakened,
.Er EWOULDBLOCK
if the timeout expires.  If
.Dv PCATCH
is set and a signal needs to be delivered,
.Er ERESTART
is returned if the current system call should be restarted if
possible, and
.Er EINTR
is returned if the system call should be interrupted by the signal
.Pq return Er EINTR .
.Pp
.Nm Msleep
is a variation on tsleep.  The parameter
.Ar mtx
is a mutex, which will be exited before sleeping, and entered before
.Nm msleep
returns.  If
.Ar pri
includes the
.Dv PDROP
flag, the
.Ar mtx
parameter will not be entered before returning.  The mutex is
used to ensure that a condition can be checked atomicly, and
that the current process can be suspended without missing a
change to the condition, or an associated wakeup.
.Pp
.Nm Asleep
implements the new asynchronous sleep function.  It takes the same arguments
as
.Fn tsleep
and places the process on the appropriate wait queue, but
.Fn asleep
leaves the process runnable and returns immediately.  The caller is then
expected to, at some point in the future, call
.Fn await
to actually wait for the previously queued wait condition.
If
.Fn asleep
is called several times, only the most recent call is effective.
.Fn asleep
may be called with an
.Ar ident
value of NULL
to remove any previously queued condition.  
.Pp
.Nm Await
implements the new asynchronous wait function.  When 
.Fn asleep
is called on an identifier it associates the process with that
identifier but does not block.
.Fn await
will actually block the process until
.Fn wakeup
is called on that identifier any time after the
.Fn asleep . 
If 
.Fn wakeup
is called after you
.Fn asleep
but before you 
.Fn await
then the 
.Fn await
call is effectively a NOP.
If
.Fn await
is called multiple times without an intervening
.Fn asleep ,
the 
.Fn await
is effectively a NOP but will also call
.Fn mi_switch
for safety.  The
.Fn await
function allows you to override the priority and timeout values to be used.
If the value -1 is specified for an argument, the value is taken from the
previous
.Fn asleep
call.  If -1 is passed for the priority you must be prepared to catch signal
conditions if the prior call to
.Fn asleep
specified it in its priority.  If -1 is passed for the timeout you must be
prepared to catch a timeout condition if the prior call to 
.Fn asleep
specified a timeout.  When you use -1, it is usually a good idea to not make
assumptions as to the arguments used by the prior
.Fn asleep
call.
.Pp
The
.Fn asleep
and
.Fn await
functions are mainly used by the kernel to shift the burden of blocking
away from extremely low level routines and to push it onto their callers.
This in turn allows more complex interlocking code to
.Em backout 
of a temporary resource failure
(such as lack of memory) in order to release major locks prior to actually
blocking, and to then retry the operation on wakeup.  This key feature is
expected to be heavily used in SMP situations in order to allow code to make
better use of spinlocks.  A spinlock, by its very nature, cannot be used 
around code that might block.  It is hoped that these capabilities will
make it easier to migrate the SMP master locks deeper into the kernel.
.Pp
These routines may also be used to avoid nasty spl*() calls to get around
race conditions with simple conditional test/wait interlocks.  You simply
call
.Fn asleep
prior to your test, then conditionally
.Fn await
only if the test fails.  It is usually a good idea to cancel an
.Fn asleep
if you wind up never calling the related
.Fn await ,
but it is not required.  If you do not want to waste cpu calling
.Fn asleep
unnecessarily, you can surround the whole thing with a second test.  The
race condition is still handled by the inside
.Fn asleep
call.
.Sh RETURN VALUES
See above.
.Sh SEE ALSO
.Xr ps 1 ,
.Xr malloc 9 ,
.Xr mi_switch 9
.Sh HISTORY
The sleep/wakeup process synchronization mechanism is very old.  It
appeared in a very early version of Unix.
.Pp
.Nm Tsleep
appeared in
.Bx 4.4 .
.Pp
.Nm Asleep Ns / Ns Nm await 
first appeared in
.Fx 3.0
and is designed to shift the burden of blocking
away from extremely low level routines and push it up to their callers.
.Pp
.Nm Sleep
used to be the traditional form.  It doesn't let you specify a timeout or a
.Ar wmesg ,
hence it has been discontinued.
.Sh AUTHORS
.An -nosplit
This man page was written by
.An J\(:org Wunsch .
.Nm Asleep
and
.Nm await
were designed and written by
.An Matthew Dillon .