Received-SPF: pass (google.com: domain of linux-kernel+bounces-787-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
From: Thomas Gleixner <tglx@linutronix.de>
To: Serge Semin <fancer.lancer@gmail.com>
Cc: Linus Walleij <linus.walleij@linaro.org>, Luo Jiaxing
 <luojiaxing@huawei.com>, Marc Zyngier <maz@kernel.org>, Bartosz
 Golaszewski <bgolaszewski@baylibre.com>, Serge Semin
 <Sergey.Semin@baikalelectronics.ru>, Andy Shevchenko
 <andy.shevchenko@gmail.com>, Andy Shevchenko
 <andriy.shevchenko@linux.intel.com>, "open list:GPIO SUBSYSTEM"
 <linux-gpio@vger.kernel.org>, "linux-kernel@vger.kernel.org"
 <linux-kernel@vger.kernel.org>, Linuxarm <linuxarm@huawei.com>
Subject: Re: [PATCH v1] gpio: dwapb: mask/unmask IRQ when disable/enable it
In-Reply-To: <cbgdeksaulqu65nbnz7l4wpuwensllkdlyi5babhpngclze4df@zxetbohgpfvg>
References: <1606728979-44259-1-git-send-email-luojiaxing@huawei.com>
 <CACRpkdYHAecm3ygaze80SsXzNMYSA9p+p6JY4BKO2D+vArb-WA@mail.gmail.com>
 <87fs03opju.ffs@tglx>
 <cbgdeksaulqu65nbnz7l4wpuwensllkdlyi5babhpngclze4df@zxetbohgpfvg>
Date: Fri, 15 Dec 2023 11:56:02 +0100
Message-ID: <87a5qbohtp.ffs@tglx>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain

On Fri, Dec 15 2023 at 13:24, Serge Semin wrote:
> On Fri, Dec 15, 2023 at 09:09:09AM +0100, Thomas Gleixner wrote:
>> On Sat, Dec 05 2020 at 22:58, Linus Walleij wrote:
>> > Sorry for top posting but I need the help of the irqchip maintainer
>> > Marc Z to hash this out.
>> >
>> > The mask/unmask/disable/enable semantics is something that
>> > you need to work with every day to understand right.
>> 
>> The patch is correct.
>> 
>> The irq_enable() callback is required to be a superset of
>> irq_unmask(). I.e. the core code expects it to do:
>> 
>>   1) Some preparatory work to enable the interrupt line
>> 
>>   2) Unmask the interrupt, which is why the masked state is cleared
>>      by the core after invoking the irq_enable() callback.
>> 
>> #2 is pretty obvious because if an interrupt chip does not implement the
>> irq_enable() callback the core defaults to irq_unmask()
>> 
>> Correspondingly the core expects from the irq_disable() callback:
>> 
>>    1) To mask the interrupt
>> 
>>    2) To do some extra work to disable the interrupt line
>> 
>> Same reasoning as above vs. #1 as the core fallback is to invoke the
>> irq_unmask() callback when the irq_disable() callback is not
>> implemented.
>
> Just curious. Wouldn't that be more correct/portable for the core to
> call both callbacks when it's required and if both are provided? So
> the supersetness requirement would be no longer applied to the
> IRQ enable/disable callbacks implementation thus avoiding the code
> duplications in the low-level drivers.

We could do that, but there are chips which require atomicity of the
operations (#1/#2). Not sure whether it safes much.

Thanks,

        tglx