Re: [PATCH] of/irq: introduce of_has_named_irqs helper

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

On Fri, Jul 24, 2015 at 03:59:39PM -0500, Rob Herring wrote:
> On Fri, Jul 24, 2015 at 2:26 PM, Dmitry Torokhov
> <dmitry.torokhov@gmail.com> wrote:
> > On Fri, Jul 24, 2015 at 02:14:57PM -0500, Rob Herring wrote:
> >> On Fri, Jul 24, 2015 at 1:26 PM, Dmitry Torokhov
> >> <dmitry.torokhov@gmail.com> wrote:
> >> > Sometimes drivers might wish to transition from index-based to named
> >> > interrupt descriptions. To aid in decision-making when parsing device
> >> > tree data let's provide a helper that will indicate the scheme that is
> >> > being used.
> >>
> >> Generally, IRQs are retrieved by platform_get_irq or
> >> platform_get_irq_byname. Drivers should not call the of_irq_*
> >> functions directly in most cases.
> >
> > That would be true for platform drivers, but not all devices are
> > platform devices.
> 
> I should say for drivers, IRQs are retrieved by a bus specific method,
> not DT functions. There are 3 cases of drivers using them:
> 
> arch/arm/mach-omap2/prm3xxx.c:                  irq_num = of_irq_get(np, 0);
> arch/arm/mach-omap2/prm44xx.c:  irq_num = of_irq_get(prm_init_data->np, 0);
> drivers/pinctrl/sirf/pinctrl-atlas7.c:          ret = of_irq_get(np, idx);

Right, but I think with the newer wakeirq infrastructure we may start
seeing them for non-platform devices unless we push it into bus code (as
I've attempted to do in the patch I sent in the other email). 

> 
> >> > Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
> >> > ---
> >> >
> >> > The intent is to it like this:
> >> >
> >> >         if (of_has_named_irqs(np) {
> >> >                 /* Wake IRQ is optional */
> >> >                 dev->wakeirq = of_irq_get_byname(np, "wakeup");
> >> >                 if (dev->wakeirq < 0 && dev->wakeirq != -ENODATA)
> >> >                         return dev->wakeirq;
> >> >         }
> >>
> >> of_irq_get_byname will already return an error if the property is not
> >> present. Use that.
> >
> > I do not like that it returns -EINVAL when property is missing, can we
> > change it to return -ENODATA (so it is the same as when the property is
> > defined but such name is missing)?
> 
> It may be useful to distinguish between those cases. If you want audit
> all the callers we could possibly change it though.

It looks like the only user of of_irq_get_byname() is
platform_get_irq_byname() and it doe snot care about anything but
-EPROBE_DEFER, so my audit effort is complete ;)

> 
> Really, I don't understand why you would care about any error code
> other than EPROBE_DEFER. If it is optional, then you would want to
> continue no matter what is returned.

No, I only want to proceed if it is not there. If there are errors (in
DTS, elsewhere in the kernel) I want to let the user know and abort.

Thanks.

-- 
Dmitry