On Wed, Apr 24, 2024 at 10:47:11PM +0300, Andy Shevchenko wrote:
> On Wed, Apr 24, 2024 at 08:51:23PM +0200, Marek Behún wrote:
> > On Wed, 24 Apr 2024 21:33:44 +0300
> > Andy Shevchenko <andy@xxxxxxxxxx> wrote:
> > > On Wed, Apr 24, 2024 at 07:38:05PM +0200, Marek Behún wrote:
...
> > > > +static void omnia_irq_mapping_drop(void *res)
> > > > +{
> > > > + irq_dispose_mapping((unsigned int)(unsigned long)res);
> > > > +}
> > >
> > > Leftover?
> >
> > What do you mean? I dropped the devm-helpers.h changes, now I do
> > devm_add_action_or_reset() manually, with this function as the action.
>
> But why?
...
> > > > + irq_idx = omnia_int_to_gpio_idx[__bf_shf(INT_TRNG)];
> > > > + irq = gpiod_to_irq(gpiochip_get_desc(&mcu->gc, irq_idx));
> > > > + if (irq < 0)
> > > > + return dev_err_probe(dev, irq, "Cannot get TRNG IRQ\n");
>
> > > > + err = devm_add_action_or_reset(dev, omnia_irq_mapping_drop,
> > > > + (void *)(unsigned long)irq);
> > > > + if (err)
> > > > + return err;
> > >
> > > Are you sure it's correct now?
> >
> > Yes, why wouldn't it?
>
> For what purpose? I don't see drivers doing that. Are you expecting that
> the same IRQ mapping will be reused for something else? Can you elaborate
> how? (I can imagine one theoretical / weird case how to achieve that,
> but impractical.)
>
> Besides above, this is asymmetrical call to gpiod_to_irq(). If we really care
> about this, it should be provided by GPIO library.
FWIW, gpiochip_irqchip_remove() disposes mappings for internally allocated IRQ
domains. So with your code it even might be a double-disposal.
--
With Best Regards,
Andy Shevchenko
