Re: Setting the priority of an IRQ thread

[Martin Shepherd <mcs@astro.caltech.edu>]

On Wed, 17 Jun 2009, Thomas Gleixner wrote:
> The irq/softirq threads call sys_sched_setscheduler() already to
> change their priorities.
>
> The right way to do this is to have an interface
> set_irq_thread_prio(irq, prio)

I just wrote such a function, and it appears to work. I couldn't get
it to work with sys_sched_setscheduler(), since this returned -EFAULT,
indicating that it didn't like the provenance of the sched_param
argument. I thus instead used the non-syscall variant,
sched_setscheduler(), protecting the target task structure from
deletion by bracketing the code with calls to
read_lock(&tasklist_lock), and read_unlock(&tasklist_lock). Hopefully
that is the right lock to take. The implemenation, which I added to
kernel/irq/manage.c, is as follows:

#ifdef CONFIG_PREEMPT_HARDIRQS

int set_irq_thread_prio(unsigned int irq, int prio)
{
         struct irq_desc *desc = irq_to_desc(irq);
         struct sched_param param = { 0, };
         int status;

         param.sched_priority = prio;

         if(desc) {
                 read_lock(&tasklist_lock);
                 if(desc->thread)
                         status = sched_setscheduler(desc->thread, SCHED_FIFO,
                                                     &param);
                 else
                         status = -ENODEV;
                 read_unlock(&tasklist_lock);
         } else {
                 status = -ENODEV;
         }
         return status;
}

EXPORT_SYMBOL(set_irq_thread_prio);

#endif

I also put the corresponding function prototype in
include/linux/irq.h. I called this from my device-driver, and ran my
test program with it a few times, and it worked fine.

Should I submit this as a patch as-is, or is there anything
problematic that I should fix first?

Thanks,

Martin