Re: [PATCH 1/1] softirq: Use a dedicated thread for timer wakeups on PREEMPT_RT.

[Frederic Weisbecker <frederic@kernel.org>]

Le Wed, Oct 23, 2024 at 12:52:57PM +0200, Sebastian Andrzej Siewior a écrit :
> On 2024-10-23 08:30:18 [+0200], To Frederic Weisbecker wrote:
> > > > > > +void raise_timer_softirq(void)
> > > > > > +{
> > > > > > +	unsigned long flags;
> > > > > > +
> > > > > > +	local_irq_save(flags);
> > > > > > +	raise_ktimers_thread(TIMER_SOFTIRQ);
> > > > > > +	wake_timersd();
> > > > > 
> > > > > This is supposed to be called from hardirq only, right?
> > > > > Can't irq_exit_rcu() take care of it? Why is it different
> > > > > from HRTIMER_SOFTIRQ ?
> > > > 
> > > > Good question. This shouldn't be any different compared to the hrtimer
> > > > case. This is only raised in hardirq, so yes, the irq_save can go away
> > > > and the wake call, too.
> > > 
> > > Cool. You can add lockdep_assert_in_irq() within raise_ktimers_thread() for
> > > some well deserved relief :-)
> > 
> > If you want to, sure. I would add them to both raise functions.
> 
> That function (run_local_timers()) was in past also called from other
> places like the APIC IRQ but all this is gone now. The reason why I
> added the wake and the local_irq_save() is because it uses
> raise_softirq() instead raise_softirq_irqoff(). And raise_softirq() was
> used since it was separated away from tasklets.
> 
> Now, raise_timer_softirq() is a function within softirq.c because it
> needs to access task_struct timersd which was only accessible there. It
> has been made public later due to the rcutorture bits so it could be
> very much be made inline and reduced to just raise_ktimers_thread().
> I tend to make TIMER_SOFTIRQ use also raise_softirq_irqoff() to make it
> look the same.

Sounds good!

> That lockdep_assert_in_irq() is probably cheap but it
> might look odd why RT needs or just TIMER and not HRTIMER.

I guess adding the same test on inline !RT functions in bottom_half.h
will be challening... Perhaps forget about that idea...

Thanks.