Re: tasklets and finalization

Re: tasklets and finalization

[Duncan Sands]

On Monday 12 November 2001 8:57 am, you wrote:
> On 11-Nov-2001 Duncan Sands wrote:
> > PS: My first thought was to use tasklet_kill, but according
> > to "Linux device drivers" (2nd ed) that may hang if the
> > tasklet isn't pending.
>
> true.
>
> > PPS: Another thought was to call schedule(), hoping that
> > all pending asklets will get run then, but is that reliable?
>
> if you are sure, the tasklet isn't rescheduled within the tasklet
> i THINK this will do it:
>
>         current->policy |= SCHED_YIELD;
>         schedule();
>
>         me

Thanks for thinking about this.  Here's a thought:

...
tasklet_schedule(&my_tasklet);
tasklet_kill(&my_tasklet);
...

Since (as far as I can see) there is no way the
tasklet will run before calling tasklet_kill, this
should just kill any pending tasklets.

Duncan.

Re: tasklets and finalization

[Mathijs Mohlmann]

On 12-Nov-2001 Duncan Sands wrote:
> ...
> tasklet_schedule(&my_tasklet);
> tasklet_kill(&my_tasklet);
> ...
> 
> Since (as far as I can see) there is no way the
> tasklet will run before calling tasklet_kill, this
> should just kill any pending tasklets.

cpu#1           cpu#2
tasklet_schedule
                tasklet_schedule
run tasklet
                tasklet_kill
                loop


-- 
        me

Re: tasklets and finalization

[Duncan Sands]

On Monday 12 November 2001 9:35 am, Mathijs Mohlmann wrote:
> On 12-Nov-2001 Duncan Sands wrote:
> > ...
> > tasklet_schedule(&my_tasklet);
> > tasklet_kill(&my_tasklet);
> > ...
> >
> > Since (as far as I can see) there is no way the
> > tasklet will run before calling tasklet_kill, this
> > should just kill any pending tasklets.
>
> cpu#1           cpu#2
> tasklet_schedule
>                 tasklet_schedule
> run tasklet
>                 tasklet_kill
>                 loop

Aaargh!  I should think before typing.  It is too
early in the morning here...

Duncan.

tasklets and finalization

[Duncan Sands]

Hi!  In the driver I'm working on, a tasklet is scheduled
from time to time.  Are there any guarantees as to when
it will run?  I'm worried, for example, about module
unloading: how to be sure that the scheduled tasklet runs
before the module is unloaded?

Thanks,

Duncan.

PS: My first thought was to use tasklet_kill, but according
to "Linux device drivers" (2nd ed) that may hang if the
tasklet isn't pending.

PPS: Another thought was to call schedule(), hoping that
all pending asklets will get run then, but is that reliable?