Re: [PATCH v3 4/4] mmc: sdhci-esdhc-imx: add device tree probe support

[Scott Wood <scottwood@freescale.com>]

On Wed, 13 Jul 2011 19:52:12 +0400
Anton Vorontsov <cbouatmailru@gmail.com> wrote:

> On Wed, Jul 06, 2011 at 03:05:18PM -0600, Grant Likely wrote:
> > On Thu, Jul 07, 2011 at 12:47:50AM +0800, Shawn Guo wrote:
> > > +- cd-gpios : Specify GPIOs for card detection
> > > +- wp-gpios : Specify GPIOs for write protection
> [...]
> > > +	cd-gpios = <&gpio0 6 0>; /* GPIO1_6 */
> > > +	wp-gpios = <&gpio0 5 0>; /* GPIO1_5 */
> 
> Is there any particular reason why we started using named GPIOs
> in the device tree? To me, that's quite drastic change... should
> we start using named regs and interrupts as well? Personally, I
> don't think that the idea is great, as now you don't know where
> to expect memory resources, interrupt resources and, eventually
> GPIO resources.

Just check the binding, and you'll know. :-)

It makes it easier to deal with cases where some resources are optional,
especially when dealing with a binding for a family of devices that grows
over time, and helps avoid resources being mismatched since order no longer
matters.

> Just a few disadvantages off the top of my head:
> 
> - You can't statically validate your device tree for correctness.
>   Today dtc checks #{address,size}-cells sanity for 'regs' and
>   'ranges';

The vast majority of stuff in the device tree already cannot be checked in
this manner, without adding binding knowledge to dtc.

Perhaps it could check things that end in "-gpios", "-regs", etc., if we
avoid adding new properties that match those patterns that aren't what they
appear to be, and let dtc know about any existing cases.

> - You can't automatically convert named resources into 'struct
>   resource *', as we do for platform devices now;

So add named resource support to platform devices. :-)

-Scott