On Wed Jul 1, 2026 at 12:55 PM CEST, Andy Shevchenko wrote:
> On Wed, Jul 01, 2026 at 01:01:10PM +0300, Andy Shevchenko wrote:
>> On Wed, Jul 1, 2026 at 11:44 AM Michael Walle <mwalle@kernel.org> wrote:
>> > On Fri Jun 19, 2026 at 11:08 PM CEST, Linus Walleij wrote:
>> > > On Mon, Jun 8, 2026 at 4:41 PM Michael Walle <mwalle@kernel.org> wrote:
>> > >
>> > >> >>> Without an accessor like gpio_regmap_get_gpiochip(), we cannot retrieve the
>> > >> >>> gpio_chip instantiated inside gpio-regmap.c to fulfill these requirements in our
>> > >> >>> map() function.
>> > >>
>> > >> Why is gpiochip_irq_reqres() called in the first place? Isn't that
>> > >> only called if the irq handling is set up via gc->irq.chip and not
>> > >> via gpiochip_irqchip_add_domain() like in gpio-regmap?
>> > >
>> > > Not really, the gpiochip_irq_reqres() is called to mark that a
>> > > GPIO line is used for IRQ, so the gpiolib cannot turn this
>> > > GPIO into an output line, gpiod_direction_out() will fail
>> > > on lines used for IRQ. So it's a failsafe.
>> > >
>> > > You can live without it of course, but then you don't get
>> > > this failsafe.
>> >
>> > Thanks for the explanation! So did I make a mistake years ago by
>> > adding the gpiochip_irqchip_add_domain(), see commit 6a45b0e2589f
>> > ("gpiolib: Introduce gpiochip_irqchip_add_domain()")
>> >
>> > As Yu-Chun found, gpiochip_irq_reqres() expect the irq chip data
>> > to be a gpio_chip, which isn't the case (in general) for an
>> > externally allocated domain, is it?
>> 
>> So the whole issue comes from the fact that the IRQ chip is not marked
>> as immutable. For immutable IRQ chips (which all GPIO provides should
>> have) there is no such issue to begin with, id est there is no
>> gpiochip_irq_reqres() callback assigned (and respective _relres).
>
> Ah, for immutable chips we put either custom ones or
> GPIOCHIP_IRQ_RESOURCE_HELPERS which actually refers to those callbacks.
>
> So, if the domain is external, it should also provide irq_request_resources
> and release callbacks. In the custom case we can wrap gpiochip_reqres_irq()
> and gpiochip_relres_irq() respectively.

Exactly. And its seems that this should have been possible with this
series, too.

Apart from that, most drivers use regmap-irq with gpio-regmap. For
this we'd probably have to add something to regmap-irq because that
module owns the irq_chip.

> But we need to have a struct gpio_chip pointer for them. And note, the
> IRQ chip data can be anything in that case, so it's not a requirement.

That's what I've meant. Conceptionally, it should be part of
gpiochip_irqchip_add_domain()/gpiolib, so a user doesn't have to
have that knowledge. But I don't see how this could be achieved.

-michael