Re: [patch, -rc5-mm1] genirq: add chip->eoi(), fastack -> fasteoi

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Wed, 2006-05-31 at 23:30 +0200, Ingo Molnar wrote:
> * Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:
> 
> > Hrm... ok. Not sure I agree with adding one more callback but it 
> > doesn't matter much.
> > 
> > Thing is, end() isn't used anymore at all now. Thus it's just 
> > basically renaming end() to eoi() except that end() is still there for 
> > whoever uses __do_IRQ() and ... handle_percpu_irq(). Doesn't make that 
> > much sense to me. So I suppose you should also change 
> > handle_percpu_irq() to use eoi() then and consider end() to be 
> > "legacy" (to be used only by __do_IRQ) ?
> 
> ok, that works with me. I did not want to reuse ->end() just to have a 
> clean migration path. ->eoi() is in fact quite descriptive as well, so 
> i'm not worried about the name.

Ok, I'll send a patch changing percpu to also use eoi() later from work
unless you beat me to it.

> > > sounds like a plan? The patch below works fine for me.
> > 
> > The patch is _almost_ right to me :) I don't need the
> > 
> > 	if (unlikely(desc->status & IRQ_DISABLED))
> >  		desc->chip->mask(irq);
> > 
> > At all. I suppose it won't harm, but it shouldn't be necessary for me 
> > and I'm not sure why it's necessary on IO_APIC neither (but then I 
> > don't know those very well).
> 
> hm, i dont think it's necessary either. I'll run a few experiments. 
> Thomas, do you remember why we have that masking there?

Ben.