[PATCH v2 01/12] ARM: Orion: DT support for IRQ and GPIO Controllers

[Arnd Bergmann]

On Thursday 05 July 2012, Andrew Lunn wrote:

> > 
> > I'm wondering about this one.  The other platforms usually put the secondary
> > interrupt controllers into the same match table, while you call orion_gpio_of_init
> > from orion_add_irq_domain. Can you explain why you do this? Does it have
> > any disadvantages?
> 
> The issue is knowing what IRQ number to use for the secondary
> interrupts.
> 
> Orion use generic chip interrupts, both for the main interrupts and
> the GPIO interrupts. This does not yet support irq domain, so i have
> to layer a legacy domain on top. The legacy domain needs to know the
> first IRQ and the number of IRQs. For the primary IRQs that is
> easy. However, GPIO IRQ is not so easy, it depends on how many primary
> IRQs there are. This is not fixed. Orion5x has 32, Dove 64, kirkwood,
> 64, and mv78xx0 has 96. I need to know this number when adding the
> GPIO secondary IRQ legacy domain. By calling orion_gpio_of_init() in
> the orion_add_irq_domain() i have this number to hand. If i used to
> entries in the match table, i would have to put this number into some
> global variable, or somehow ask the IRQ subsystem what the next free
> IRQ number is.

But couldn't you store the number of interrupts for the primary controller
in a local variable? I think the of_irq code already guarantees that
the parent is probed first.

> As for disadvantages, humm. Dove has yet more interrupts, from the
> PMU. They are currently unsupported in DT. When we add support for the
> PMU interrupt controller, we are going to have the same problem, what
> IRQ base should it use. Either we extend the chaining, calling
> dove_pmu_of_init from orion_gpio_of_init(), where we know the next
> free IRQ. Or we find out how to ask the IRQ subsystem for the next
> available. Better still, the work to make generic chip interrupts irq
> domain aware would get completed, and we can swap all this code to irq
> domain linear and this whole probable probably goes away.

Yes, that makes sense. Using the linear domain should solve all these
nicely.

	Arnd