Query on workqueue

[Richard Maciel Costa]

Hi Victor,

I don't think it's a good idea to block an entire IRQ line for so much time
(i.e. until a work queue runs), unless you're sure you're the sole user of
it.

How about you try a more traditional approach and make the IRQ handler
schedule a work queue after it completes?

Or, if you need to run as fast as possible, why don't you create a tasklet?
Remember that you can't sleep in this case.

- Richard

2015-01-13 11:51 GMT-02:00 Victor Ascroft <victorascroft at gmail.com>:

> Hello,
>
> Is it ok to use wait_for_completion in a workqueue?
>
> static void my_work(struct work_struct *work)
> {
>
>         while (true)
>         {
>                 wait_for_completion(completion);
>
>                 // Do something here
>
>                 reinit_completion(completion);
>                 enable_irq(irq);
>         }
> }
>
> static irqreturn_t my_irq_handler(int irq, void *dev)
> {
>         disable_irq(irq);
>
>         complete(completion);
>
>         return IRQ_HANDLED;
> }
>
> Something like the above is what I have. Is it a correct way to
> do things or complete idiotic brainfart. Workqueues can sleep so
> I thought of the above code, but, was not sure.
>
> And I am not using _interruptible or _interruptible_timeout because
> I absolutely want it to wait for the IRQ. Now as such though this
> works I get stack traces with hung task complaining and I have
> to set /proc/sys/kernel/hung_task_timeout_secs to 0. And were the IRQ
> not generated I do get NMI watchdog hang.
>
> I will accept I have only little knowledge of these things.
>
> Regards,
> Victor.
>
> _______________________________________________
> Kernelnewbies mailing list
> Kernelnewbies at kernelnewbies.org
> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20150113/7d9f9e92/attachment.html