Re: [PATCH 11/13] mfd: pm8008: rework driver

[Andy Shevchenko <andy.shevchenko@xxxxxxxxx>]

On Wed, May 22, 2024 at 9:49 AM Johan Hovold <johan@xxxxxxxxxx> wrote:
> On Fri, May 10, 2024 at 04:15:43PM +0300, Andy Shevchenko wrote:
> > On Thu, May 9, 2024 at 12:42 PM Johan Hovold <johan@xxxxxxxxxx> wrote:
> > > On Mon, May 06, 2024 at 10:18:58PM +0300, Andy Shevchenko wrote:
> > > > Mon, May 06, 2024 at 05:08:28PM +0200, Johan Hovold kirjoitti:

...

> > > > > +static void devm_irq_domain_fwnode_release(void *res)
> > > > > +{
> > > >
> > > > > +   struct fwnode_handle *fwnode = res;
> > > >
> > > > Unneeded line, can be
> > > >
> > > > static void devm_irq_domain_fwnode_release(void *fwnode)
> > > >
> > > > > +   irq_domain_free_fwnode(fwnode);
> > > > > +}
> > >
> > > I think I prefer it this way for clarity and for type safety in the
> > > unlikely even that the argument to irq_domain_free_fwnode() would ever
> > > change.
> >
> > If it ever changes, the allocation part most likely would need an
> > update and since devm_add_action() takes this type of function, I
> > don't believe the argument would ever change from void * to something
> > else. With this it just adds an additional burden on the conversion.
>
> I was referring to the irq_domain_free_fwnode() prototype.

And I also referred to that one. The release callback, i.e. the type
of the parameter, is solely defined by a caller of devm_add_action()
end friends, and in this case it means that if ever the type changes
(this is your argument why you want to have explicit line for that,
necessity of which I oppose) the devm_add_action() arguments also has
to be changed, it can't be done _just_ there, in
irq_domain_free_fwnode().

> > > > > +   name = devm_kasprintf(dev, GFP_KERNEL, "%pOF-internal", dev->of_node);
> > > >
> > > > You are using fwnode for IRQ domain and IRQ domain core uses fwnode, why OF here?
> > > >
> > > >       name = devm_kasprintf(dev, GFP_KERNEL, "%pfw-internal", dev_fwnode(dev));
> > >
> > > This driver only support OF so why bother.
> >
> > Sure, but it makes a bit of inconsistency.
>
> No, I don't consider this an inconsistency. Again, *this* is an OF
> driver, other subsystems need to deal with ACPI and use fwnode.

OK.


-- 
With Best Regards,
Andy Shevchenko