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

Anton Vorontsov cbouatmailru at gmail.com
Wed Jul 13 13:32:45 EDT 2011


On Thu, Jul 14, 2011 at 01:36:39AM +0900, Grant Likely wrote:
[...]
> Only for irqs and regs.  gpios have never been automatically loaded
> into resources.

Which doesn't mean we wouldn't want it sooner or later.

> > - Any pros for using named resources in the device tree? I don't
> >  see any.
> 
> Human readability.  To know exactly what a gpio is intended to be used
> for.  Particularly for the case where a device might not use all the
> gpios that it could use.  Yes, the gpios property can have 'holes' in
> it, but the observation was made by several people that it is easy to
> get wrong.  I for one thing the concern was well justified.

The GPIO bindings are no harder to deal with than PCI memory bindings,
not even close to that complexity. So I don't really see why you try
to simplify GPIOs, but disagree on making the same for memory and
interrupt resources.

For example arch/powerpc/boot/dts/ebony.dts, 'mcmal' node has five
interrupts (txeob, rxeob, serr, txde, rxde). Or, gianfar nodes have
either three interrupts (tx, rx, err) or just one.

The average user of 'gpios' has 1-2 entries (the noticeable exception
is USB FHCI, which has 8 GPIOs).

I.e., I don't see how GPIOs are special. I'm all for consistency,
that's it. If that doesn't work for IRQs, then I want to understand
why so. And if you explain why named resources are no good for IRQs,
maybe I could use the same argument against named GPIOs? :-)

Or it could be otherwise: we agree that named resources are good, and
we should explicitly write when to use named and when to use anonymous
resources.

> > So, I suggest to at least discuss this stuff a little bit more
> > before polluting device trees with dubious ideas.
> 
> It was discussed on list quite a while ago.

I probably wasn't Cc'ed, can you point me to that thread?

The last time I was Cc'ed on a such discussion, we (well who cared
enough to 'vote') agreed* that we should wait with deploying named
GPIOs scheme, and discuss it later. And here we are.

The patch that added of_get_named_gpio() triggered no discussion
at all, but I wasn't Cc'ed either.

* http://lists.ozlabs.org/pipermail/linuxppc-dev/2008-October/064701.html

-- 
Anton Vorontsov
Email: cbouatmailru at gmail.com



More information about the linux-arm-kernel mailing list