[PATCH] pinctrl: Add generic pinctrl-simple driver that supports omap2+ padconf

[Tony Lindgren]

* Stephen Warren <swarren at wwwdotorg.org> [120511 14:08]:
> On 05/11/2012 01:51 PM, Tony Lindgren wrote:
> > * Stephen Warren <swarren at wwwdotorg.org> [120511 12:21]:
> >>
> >> The mapping of GPIO to pinctrl pins would presumably be driven solely by
> >> the HW design of the pin controller and GPIO, and not by the mux
> >> selection in the pin controller (otherwise, I'd argue this isn't a
> >> simple case that should be handled by pinctrl-simple).
> >>
> >> As such, I'd expect some properties/table at the top-level of the pin
> >> controller object to describe the GPIO mapping. In turn, that implies
> >> that the individual per-pin mux-selection/configuration nodes don't need
> >> to describe any GPIO-related information.
> > 
> > Yes good point. I agree it's a HW design issue, and could be in the properties
> > for the pin controller object.
> > 
> > Just to summarize, the things to consider with the GPIO to mux mapping are:
> > 
> > 1. Having this table as static data in the driver is is not a nice
> >    solution as it seems that we'd currently need six mapping tables for
> >    omap2+ alone.
> > 
> > 2. This table is not needed for most of the (hundreds of) pins, it's
> >    only needed for a few selected pins, let's say ten or so on an average
> >    device. So there's no need to stuff the kernel with information about
> >    the unused GPIO pins.
> > 
> > It seems that the conclusion here is that we don't need to worry about
> > GPIOs in the pinctrl-simple binding for now, and it can be added later
> > without having to change the basic binding.
> 
> The one thing I wanted to resolve here wasn't so much the binding for
> GPIO interaction here, but the following comment:
> 
> You wrote:
> > I wrote:
> >> From a binding perspective, I don't see why you'd want to allow two cases:
> >>
> >> 1) One node with multiple entries in pinctrl-simple,cells
> >> 2) Multiple nodes each with a single entry in pinctrl-simple,cells
> >>
> >> Why not only allow (1)?
> >
> > Because we need to specify GPIO for some pins. There may be additional flags
> > too, we do have external DMA request lines for few pins available.. I'm not
> > saying pinctrl fwk should know about that, but it's a similar mapping of pins
> > to GPIO lines.
> 
> I'm asserting that since any GPIO mapping information would be at the
> top-level of the pinctrl-simple binding, we can in fact only allow
> option (1) above for the individual pin configuration nodes.

Yes I agree. Based on what we have discussed #2 binding can now be dropped.
At least I don't have any other reasons for pin specific flags that would
need to be passed in the binding #2 above.

The only additional data #2 binding provides are the real function names
for the mux signals. But those can be eventually be shown by userspace debug
tools, so no need for the kernel to care about them. From kernel and device
driver point of view the named states are much more generic.

Regards,

Tony