Re: irq 16: nobody cared

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Apr 22, 2013 at 11:18:47AM +0200, Borislav Petkov wrote:
> On Mon, Apr 22, 2013 at 10:01:36AM +0200, Ingo Molnar wrote:
> > Hm, this really smells like a workaround: treating the symptom, not
> > the cause.
> 
> Well, I just tested Takashi's add missing synchronize_irq() to the
> suspend path of snd_hda_intel and it doesn't help.
> 
> So it could be an issue with this driver or only this hw/driver can
> exacerbate RCU this much to trigger the spurious irq machinery.
> 
> > How can an increase in grace-periods break drivers and suspend? Do we
> > understand exactly what happens there?
> 
> I'd rather look into Paul's direction here. :)

I believe that we need both -- a fix that prevents the stray irq for
normal CONFIG_RCU_FAST_NO_HZ grace periods and also expediting grace
periods during suspend/resume.  I am greedy that way.  ;-)

						Thanx, Paul

> > If we do, can we fix that instead of tweaking the RCU timeouts?
> 
> Well, I was under the impression yesterday that we actually want
> to switch to expedited grace periods for suspend/resume *anyway*,
> regardless of drivers.
> 
> Paul?
> 
> -- 
> Regards/Gruss,
>     Boris.
> 
> Sent from a fat crate under my desk. Formatting is fine.
> --
>