Re: [v3.4-rc1] ACPI regression bisected

[Thomas Gleixner <tglx@linutronix.de>]

On Thu, 12 Apr 2012, Andi Kleen wrote:

> 
> > > I feared that. So now is the question whether there is a way to mask
> > > the interrupt at the "device" level.
> 
> Probably not. Len?
> 
> > If not, I could be persuaded to provide a mechanism to force thread
> > all interrupts which are on that line, if there's a good argument to
> > do so.
> 
> Hmm, we could probably just use a work queue for this instead. I guess i drank
> the threaded
> interrupt kool aid too much when I did the patch :-)
> 
> I'll switch over to a workqueue.

That still requires that you can silence the interrupt at the device
level in the first place unless you play silly games with
disable_irq_nosync(), which is even worse than having a
IRQF_FORCE_ONESHOT mechanism in the core.

If you go the workqueue way, then you basically split the interrupt
into a hard irq context and a thread handler. So it's the same to
request a real threaded interrupt with a primary and a thread handler
and return IRQF_WAKE_THREAD conditionally from the primary handler,
when you need the thread context for further processing.

Thoughts?

	tglx