Re: [PATCH 3/3] gpio: regmap: Support a custom ->to_irq() hook

From: Linus Walleij <linus.walleij@linaro.org>
To: Aidan MacDonald <aidanmacdonald.0x0@gmail.com>,
	Marc Zyngier <maz@kernel.org>
Cc: michael@walle.cc, brgl@bgdev.pl, linux-gpio@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH 3/3] gpio: regmap: Support a custom ->to_irq() hook
Date: Mon, 11 Jul 2022 13:48:50 +0200	[thread overview]
Message-ID: <CACRpkdYEeNY=PmGaNfV=K7cJPJu1vyUa3wREbCU8hCJVJb5YfA@mail.gmail.com> (raw)
In-Reply-To: <ZTG1etjWj47jkqTxIRgriV6pQhW9dasf@localhost>

On Wed, Jul 6, 2022 at 4:21 PM Aidan MacDonald
<aidanmacdonald.0x0@gmail.com> wrote:

> If moving to hierarchical IRQ chips is the plan, could we add a note
> to say .to_irq() is discouraged and shouldn't be used in new code?
> Based on what you're saying (which I agree makes sense) it sounds
> like there's really no reason to ever use .to_irq().

There is a reason to use .to_itq() to get the irq associated with
a GPIO, in a Linux kernel context. Especially when doing exactly
what IRQ keys are doing: finding the IRQ for a GPIO line, if any.

But it is not intended to provide a random IRQ from anywhere
else in the system. Maybe I could clarify that.

There are many old boardfile-instances of GPIO where .to_irq()
may be doing something really funky, these are not really up to
date with the current idea of how we model systems in hardware
description languages such as DT or ACPI DSDT.

There may be archs that do not even have DT or ACPI or similar,
there it would be fine to use it in a more liberal way. This may
be the case for m68k or something so I cannot speak about
everything here.

I understand that it might be annoying that this is a bit fuzzy
around the edges.

> Okay, that sounds like a good plan. I'll look more carefully at the
> existing drivers and see if I can use existing gpiolib helpers.
>
> One potential issue (from reading the code) is that hierarchical IRQ
> domains seemingly can't have a non-hierarchical domain as the parent:
> irq_domain_alloc_irqs_parent() calls irq_domain_alloc_irqs_hierarchy()
> and the latter fails with -ENOSYS for a non-hierarchical domain.

Yes that is a characteristic of hierarchical irq domains, they have to
be hierarchical all the way down. It is very often what we want
as well.

> In my case I'm using a regmap IRQ chip, which is non-hierarchical,
> so perhaps that will need to be expanded?

Yes I'm sorry if it adds work to your plate :(

However I think the end result would be very useful for everyone.
Maybe Marc Z has something to add here?

Yours,
Linus Walleij