Re: [PATCH 1/3] tick/nohz: Re-evaluate the tick on polling idle and on user entry

[Frederic Weisbecker]

Le Tue, Nov 10, 2026 at 06:35:49PM +0100, Marco Crivellari a écrit :
> On Tue, Nov 10, 2026 at 2:33 PM Frederic Weisbecker <frederic@xxxxxxxxxx> wrote:
> > [...]
> > > With the hunk, the stopped tick firing for the first timer sets
> > > TIF_NEED_RESCHED, the polling loop exits and the tick is reprogrammed
> > > for the second timer within a few microseconds; the 10 iterations
> > > captured take 0.92s instead of ~18s. Without that hunk, the CPU stops
> > > spinning on an unrelated event.
> > >
> > > So in my understanding tick_nohz_idle_exit() program the tick only for
> > > the earliest timer; when it expire, the tick is stopped and not
> > > rearmed in tick_nohz_handler() (HRTIMER_NORESTART). Without setting
> > > NEED_RESCHED if another timer has been queued it has to wait for
> > > another event that set NEED_RESCHED.
> >
> > Hehe, you're right :)
> >
> > Let's try another strategy then. TS_FLAG_INIDLE is what tells nohz_fool whether
> > to manage the tick through the idle loop or through the IRQ exit.
> >
> > So why not calling tick_nohz_idle_enter() on each idle loop iterations? That is,
> > from cpuidle_idle_call(). And call tick_nohz_idle_exit() right before
> > cpu_idle_poll()?
> 
> Aha, sounds good! :-) So INIDLE is set only when really "in idle" (eg
> no polling).
> 
> > Then you can remove tick_nohz_idle_restart_tick().
> 
> Should we remove only the call or also the function and its declaration?
> I'm asking because the call we're going to remove is the only one present.

You can remove also the function.

> 
> > Just be careful: tick_nohz_idle_enter() and tick_nohz_idle_exit() may now be
> > unbalanced. If TS_FLAG_INIDLE is already set, tick_nohz_idle_enter() should
> > return early and tick_nohz_idle_exit() as well if !TS_FLAG_INIDLE.
> >
> > Also IRQs may now be disabled while calling tick_nohz_idle_exit(), so it should
> > use local_irq_save() and not local_irq_disable(). OTOH tick_nohz_idle_enter()
> > should always expect IRQs disabled.
> >
> > So the purpose is to move cpu_idle_poll() outside the dynticks idle logic, then
> > it's safe against nohz_full timer management, the tick keeps being stopped if
> > possible and it's only restarted for dynticks idle.
> 
> Ok, I think it's clear.
> 
> I've already started implementing the changes. I will send v2 with all
> the changes.

Thanks a lot!

> 
> Thank you!
> 
> -- 
> 
> Marco Crivellari
> 
> SUSE Labs

-- 
Frederic Weisbecker
SUSE Labs