Re: [PATCH 1/2] linux/interrupt.h: allow "guard" notation to disable and reenable IRQ with valid IRQ check

[Marek Vasut]

On 1/27/26 10:14 AM, Thomas Gleixner wrote:
> On Thu, Jan 22 2026 at 00:23, Marek Vasut wrote:
>
> $Subject: linux/interrupt.h is not a proper prefix. Just use 'genirq:'
>
> Also what means 'allow ....'? Subject lines should provide a concise
> summary of the change, i.e. something like:
>
>          Provide conditional disable_irq guard variant

I did admittedly replicate the subject and tags from prior commit

c76494768761 ("linux/interrupt.h: allow "guard" notation to disable and 
reenable IRQ")

Fixed, thanks.

> > Introduce disable_valid_irq scoped guard. This is an extension
> > of disable_irq scoped guard, which disables and enables IRQs
> > around a scope. The disable_valid_irq scoped guard does almost
> > the same, except it handles the case where IRQ is not valid,
> > in which case it does not do anything. This is meant to be used
> > by for example touch controller drivers, which can do both IRQ
> > driven and polling mode of operation, and this makes their code
> > slighly simpler.
>
> How many of those have this pattern? From a cursory look I found not
> even one as your example is neither applicable to upstream nor to next.

One. I think I have two-ish more candidates for conversion.

> > +static inline void disable_valid_irq(unsigned int irq)
> > +{
> > +	if (irq > 0)
> > +		disable_irq(irq);
> > +}
> > +
> > +static inline void enable_valid_irq(unsigned int irq)
> > +{
> > +	if (irq > 0)
> > +		enable_irq(irq);
> > +}
> > +
> > +DEFINE_LOCK_GUARD_1(disable_valid_irq, int,
>
> 'int' because it matches the function argument type of the inlines,
> right?
>
> > +		    disable_valid_irq(*_T->lock), enable_valid_irq(*_T->lock))
>
> disable_valid_irq is a pretty non-intuitive name if you look at it just
> by reading a usage site. It's not really improving the readability of
> the code, it's in fact obscuring it as the reader has to actually look
> up what the hell this means and then stumble upon a completely
> undocumented lock guard define.
>
> I'm all for using guards, but using guards just for the sake of using
> guards is not a really good approach.
I wouldn't even be opposed to converting the ili2xxx driver (the piece 
of code in patch 2/2 of this series) back to simple enable/disable_irq() 
. I am not particularly on board even with the disable_irq lock guard, 
or more specifically, lock guard used for non-lock things like this.