Re: [PATCH v4] gpio: Add support for Intel SoC PMIC (Crystal Cove)

[Mika Westerberg <mika.westerberg@linux.intel.com>]

On Thu, May 29, 2014 at 05:22:05PM +0200, Linus Walleij wrote:
> On Thu, May 29, 2014 at 6:03 PM, Grygorii Strashko
> <grygorii.strashko@ti.com> wrote:
> 
> > Also, I'd like to note that GPIO IRQs can be accessible not only
> > when GPIO chips is added, but also when IRQ domain is registered
> > (at least it's valid for DT cases). In these cases gpiod_to_irq()
> > might be not used at all.
> 
> Yes. We concluded some time back that gpio_chip:s and
> irq_chip:s are orthogonal abstractions: you should be able
> to use one of them without paying any respect to the other.
> 
> We only added the ability to flag GPIO lines as used for
> IRQs so they would not be set to output by mistake...
> (Straightening up the semantics.)
> 
> The only real semantic dependence that really makes sense
> is .to_irq() which leads to this semantic registration ordering.

acpi_gpiochip_request_interrupts() depends on ->to_irq() to be set
before acpi_gpiochip_add() is called. Since the ordering changes this
won't work anymore.

I'm thinking that could we solve this so that we call
acpi_gpiochip_request_interrupts() at the end of gpiochip_irqchip_add()
and convert both pinctrl-baytrail and gpio-lynxpoint to use
gpiochip_irqchip_add()?