[PATCH] genirq: move mask_cache into struct irq_chip_type

[Thomas Gleixner]

On Wed, 6 Mar 2013, Simon Guinot wrote:

> On Tue, Mar 05, 2013 at 11:15:04AM +0100, Thomas Gleixner wrote:
> > On Mon, 4 Mar 2013, Simon Guinot wrote:
> > > On Mon, Mar 04, 2013 at 03:28:14PM +0100, Gerlando Falauto wrote:
> > > 
> > > AFAIK this issue is still here. I include the current Orion maintainers
> > > in the Cc list.
> > 
> > Can you please resend the patch?
> 
> Hi Thomas,
> 
> Do you agree with the original patch concept: moving mask_cache from
> struct irq_chip_generic into struct irq_chip_type ?

Not really. See below.
 
> This allows to handle the irq_chips which have separate mask registers
> for the edge and level interrupts. At the time of this patch, this was
> the only existing case.
> 
> Saeed objected that the patch was not correct because this was breaking
> "hypothetic" devices with multiple irq_chip_type and a single mask
> register.

Not so hypothetic. There _are_ irq controllers out there which use the
same mask register for both level and edge type irqs.

> The other option was to provide some dedicated irq_mask handlers for the
> Orion irq chip.

I'd rather refactor the core code so it uses a pointer to the
mask_cache. The default would be to let it point to gc->mask_cache and
optionally let it point to ct->mask_cache. We'd need to store the flag
in the gc struct so we can redirect the pointer to the ct->mask_cache
in irq_setup_alt_chip().

Thanks,

	tglx