Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Thomas Gleixner <tglx@linutronix.de>
To:     Randy Dunlap <rdunlap@infradead.org>, linux-kernel@vger.kernel.org
Cc:     Randy Dunlap <rdunlap@infradead.org>,
        Marc Zyngier <maz@kernel.org>, Arnd Bergmann <arnd@arndb.de>,
        linux-arch@vger.kernel.org, linux-gpio@vger.kernel.org
Subject: Re: [PATCH v2] irq domain: drop IRQ_DOMAIN_HIERARCHY option, make
 it always on
In-Reply-To: <20230313023935.31037-1-rdunlap@infradead.org>
References: <20230313023935.31037-1-rdunlap@infradead.org>
Date:   Thu, 23 Mar 2023 21:53:04 +0100
Message-ID: <877cv78a1b.ffs@tglx>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

On Sun, Mar 12 2023 at 19:39, Randy Dunlap wrote:
> In preparation for dropping the IRQ_DOMAIN Kconfig option (effectively
> making it always set/on), first drop IRQ_DOMAIN_HIERARCHY as an option,
> making its code always set/on.
>
> This has been built successfully on all ARCHes except hexagon,
> both 32-bit and 64-bit where applicable.

I really like where this is going, but reviewing this is a pain. I tried
to split it up into more digestable pieces:

   https://tglx.de/~tglx/patches.tar

That's not completely equivalent to your patch as I did some of the
changes below. It builds on various oddball architectures with
IRQ_DOMAIN=n, but is otherwise completely untested.

It should be actually trivial after that to make IRQ_DOMAIN def_bool y
and then gradually remove the IRQ_DOMAIN selects and ifdeffery.

> v2: add stubs in include/linux/irqdomain.h for the config case of
> IRQ_DOMAIN is not set. If these are not added, there will be plenty
> of build errors (not so much for modern arches as for older ones).

I'm not really convinced that all of these stubs are required. Why would
there suddenly be a requirement to expose stubs for functions which
depend on CONFIG_IRQ_DOMAIN=y already today just by removing the
hierarchy config?

Even exposing stubs for functions which have been only available via
CONFIG_IRQ_DOMAIN_HIERARCHY is questionable simply because there cannot
be any code which invokes them unconditionally if
CONFIG_IRQ_DOMAIN_HIERARCHY=n today.

IOW, the sum of required stubs cannot be larger than number of stubs
required today.

If there is code which has a #ifdef CONFIG_IRQ_DOMAIN_HIERARCHY then
this needs to be changed to CONFIG_IRQ_DOMAIN or the required functions
have to be exposed unconditionally, right?

> diff -- a/drivers/gpio/Kconfig b/drivers/gpio/Kconfig
> --- a/drivers/gpio/Kconfig
> +++ b/drivers/gpio/Kconfig
> @@ -361,7 +361,7 @@ config GPIO_IXP4XX
>  	depends on OF
>  	select GPIO_GENERIC
>  	select GPIOLIB_IRQCHIP
> -	select IRQ_DOMAIN_HIERARCHY
> +	select IRQ_DOMAIN

IRQ_DOMAIN is already selected by GPIOLIB_IRQCHIP, so this select is
redundant for all GPIO configs which select GPIOLIB_IRQCHIP.

Thanks,

        tglx