Le Tue, Oct 29, 2024 at 02:52:31PM +0100, Sebastian Andrzej Siewior a écrit :
> On 2024-10-28 15:01:55 [+0100], Frederic Weisbecker wrote:
> > > diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h
> > > index 457151f9f263d..9637af78087f3 100644
> > > --- a/include/linux/interrupt.h
> > > +++ b/include/linux/interrupt.h
> > > @@ -616,6 +616,50 @@ extern void __raise_softirq_irqoff(unsigned int nr);
> > > extern void raise_softirq_irqoff(unsigned int nr);
> > > extern void raise_softirq(unsigned int nr);
> > >
> > > +/*
> > > + * Handle timers in a dedicated thread at a low SCHED_FIFO priority
> > > instead in
> > > + * ksoftirqd as to be prefred over SCHED_NORMAL tasks.
> > > + */
> >
> > This doesn't parse. How about, inspired by your changelog:
> …
>
> What about this essay instead:
>
> | With forced-threaded interrupts enabled a raised softirq is deferred to
> | ksoftirqd unless it can be handled within the threaded interrupt. This
> | affects timer_list timers and hrtimers which are explicitly marked with
> | HRTIMER_MODE_SOFT.
> | With PREEMPT_RT enabled more hrtimers are moved to softirq for processing
> | which includes all timers which are not explicitly marked HRTIMER_MODE_HARD.
> | Userspace controlled timers (like the clock_nanosleep() interface) is
> divided
> | into two categories: Tasks with elevated scheduling policy including
> | SCHED_{FIFO|RR|DL} and the remaining scheduling policy. The tasks with the
> | elevated scheduling policy are woken up directly from the HARDIRQ while all
> | other wake ups are delayed to so softirq and so to ksoftirqd.
First "so" is intended?
> |
> | The ksoftirqd runs at SCHED_OTHER policy at which it should remain since it
> | handles the softirq in an overloaded situation (not handled everything
> | within its last run).
> | If the timers are handled at SCHED_OTHER priority then they competes with
> all
> | other SCHED_OTHER tasks for CPU resources are possibly delayed.
> | Moving timers softirqs to a low priority SCHED_FIFO thread instead ensures
> | that timer are performed before scheduling any SCHED_OTHER thread.
Works for me!
> And with this piece of text I convinced myself to also enable this in
> the forced-threaded case.
Yes, that makes sense.
Thanks.
> > Thanks.
>
> Sebastian
