Return to task_add.9 CVS log

Up to [local] / src / share / man / man9

Document a locking constraint that applies to barriers.

OK cheloha@

.\"	$OpenBSD: task_add.9,v 1.23 2022/06/22 14:10:49 visa Exp $
.\"
.\" Copyright (c) 2013 David Gwynne <dlg@openbsd.org>
.\"
.\" Permission to use, copy, modify, and distribute this software for any
.\" purpose with or without fee is hereby granted, provided that the above
.\" copyright notice and this permission notice appear in all copies.
.\"
.\" THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
.\" WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
.\" MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
.\" ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
.\" WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
.\" ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
.\" OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
.\"
.Dd $Mdocdate: June 22 2022 $
.Dt TASK_ADD 9
.Os
.Sh NAME
.Nm taskq_create ,
.Nm taskq_destroy ,
.Nm taskq_barrier ,
.Nm taskq_del_barrier ,
.Nm task_set ,
.Nm task_add ,
.Nm task_del ,
.Nm task_pending ,
.Nm TASK_INITIALIZER
.Nd task queues
.Sh SYNOPSIS
.In sys/task.h
.Ft struct taskq *
.Fo taskq_create
.Fa "const char *name"
.Fa "unsigned int nthreads"
.Fa "int ipl"
.Fa "unsigned int flags"
.Fc
.Ft void
.Fn taskq_destroy "struct taskq *tq"
.Ft void
.Fn taskq_barrier "struct taskq *tq"
.Ft void
.Fn taskq_del_barrier "struct taskq *tq" "struct task *t"
.Ft void
.Fn task_set "struct task *t" "void (*fn)(void *)" "void *arg"
.Ft int
.Fn task_add "struct taskq *tq" "struct task *t"
.Ft int
.Fn task_del "struct taskq *tq" "struct task *t"
.Ft int
.Fn task_pending "struct task *t"
.Vt extern struct taskq *const systq;
.Vt extern struct taskq *const systqmp;
.Fn TASK_INITIALIZER "void (*fn)(void *)" "void *arg"
.Sh DESCRIPTION
The
taskq
API provides a mechanism to defer work to a process context.
.Pp
.Fn taskq_create
allocates a taskq and a set of threads to be used to complete work
that would be inappropriate for the shared system taskq.
The
.Fa name
argument specifies the name of the kernel threads that are created
to service the work on the taskq.
.Fa nthreads
specifies the number of threads that will be created to handle the work.
.Fa ipl
specifies the highest interrupt protection level at which
.Fn task_add
and
.Fn task_del
will be called against the created taskq.
See
.Xr spl 9
for a list of the IPLs.
The operational characteristics of the taskq
can be altered by OR'ing the following defines into the
.Fa flags
argument:
.Bl -tag -width xxx -offset indent
.It Dv TASKQ_MPSAFE
The threads servicing the taskq will be run without the kernel big lock.
.El
.Pp
.Fn taskq_destroy
causes the resources associated with a previously created taskq to be freed.
It will wait till all the tasks in the work queue are completed before
returning.
Calling
.Fn taskq_destroy
against the system taskq is an error and will lead to undefined
behaviour or a system fault.
.Pp
.Fn taskq_barrier
guarantees that any task that was running on the
.Fa tq
taskq when the barrier was called has finished by the time the barrier
returns.
.Pp
.Fn taskq_del_barrier
either removes
.Fa t
from the list of pending tasks on the
.Fa tq
taskq, or waits until any running task has completed.
.Pp
The caller of
.Fn taskq_barrier
or
.Fn taskq_del_barrier
must not hold locks that can block the taskq.
Otherwise, the system will deadlock.
.Pp
It is the responsibility of the caller to provide the
.Fn task_set ,
.Fn task_add ,
and
.Fn task_del
functions with pre-allocated task structures.
.Pp
.Fn task_set
prepares the task structure
.Fa t
to be used in future calls to
.Fn task_add
and
.Fn task_del .
.Fa t
will be prepared to call the function
.Fa fn
with the argument specified by
.Fa arg .
Once initialised, the
.Fa t
structure can be used repeatedly in calls to
.Fn task_add
and
.Fn task_del
and does not need to be reinitialised unless the function called
and/or its argument must change.
.Pp
.Fn task_add
schedules the execution of the work specified by the
task structure
.Fa t
on the
.Fa tq
taskq.
The task structure must already be initialised by
.Fn task_set .
.Pp
.Fn task_del
will remove the task structure
.Fa t
from the taskq
.Fa tq .
If the work was already executed or has not been added to the taskq,
the call will have no effect.
Calling
.Fn task_del
against a different taskq than the one given in a previous call to
.Fn task_add
is an error and will lead to undefined behaviour.
.Pp
The kernel provides two system taskqs:
.Va systq ,
which executes while holding the kernel lock, and
.Va systqmp ,
which does not hold the kernel lock during execution.
They can both be used by any subsystem for short lived tasks.
They are serviced by a single thread and can therefore provide predictable
ordering of work.
Work can be scheduled on the system taskqs from callers at or below IPL_HIGH.
.Pp
The
.Fn task_pending
macro can be used to check if a task is scheduled to run.
.Pp
A task declaration can be initialised with the
.Fn TASK_INITIALIZER
macro.
The task will be prepared to call the function specified by the
.Fa fn
argument with the
.Fa void *
argument given in
.Fa arg .
.Sh CONTEXT
.Fn taskq_create
and
.Fn taskq_destroy
can be called during autoconf, or from process context.
.Fn taskq_barrier
and
.Fn taskq_del_barrier
can be called from process context.
.Fn task_set ,
.Fn task_add ,
.Fn task_del ,
and
.Fn task_pending
can be called during autoconf, from process context, or from interrupt context.
.Sh RETURN VALUES
.Fn taskq_create
returns a pointer to a taskq structure on success or
.Dv NULL
on failure.
.Pp
.Fn task_add
will return 1 if the task
.Fa t
was added to the taskq
.Fa tq
or 0 if the task was already queued.
.Pp
.Fn task_del
will return 1 if the task
.Fa t
was removed from the taskq
.Fa tq
or 0 if the task was not already on the queue.
.Pp
.Fn task_pending
will return non-zero if the task is queued to run, or 0 if the task
is not queued.
.Sh SEE ALSO
.Xr autoconf 9 ,
.Xr spl 9
.Sh HISTORY
The task API was originally written by
.An David Gwynne Aq Mt dlg@openbsd.org .
The task API first appeared in
.Ox 5.5 .