On Mon, Jul 13, 2015 at 02:07:45PM +0200, Linus Walleij wrote:

On Tue, Jun 30, 2015 at 11:17 AM, Nicolas Ferre [off-list ref] wrote:

quoted

- no agreement on 3 points:
1/ ways to use groups in generic pinctrl
2/ ways to describe a comprehensive configuration in device tree
3/ readability of a sysfs information

- no way out on the generic pinctrl little changes that Ludovic proposed
as Linus W. never gave his point of view (RFC posts on April the 2nd).

Yeah I know. I am battling with this, it is one of those topics
where you feel "eeeehhhh" and try to avoid looking at it. Sadly
I have to...

I refer to Documentation/ManagementStyle, chapter 1,
"Decisions".

What makes me feel uneasy about these things is that the decisions
are irreversible due to the nature of the device tree bindings.

I am not a manager, but a maintainer...

quoted

Ludovic explained at length our point of view and gave detailed
technical arguments. We don't intend to convince you, we just would like
the harmless modifications to be integrated.

So is it universally agreed that the changes are harmless?

If they are harmless, I would be able to revert the patch and
nothing breaks in the world, I don't think that is the case. The
case is a piece of code and functionality that is not AT91-specific
but has to be maintained in the core pinctrl code forever.

If this change turns out to be wrong it doesn't necessarily have to stay
in the core pinctrl code. It can be pushed back into the AT91 driver
(and the others using it), won't be used for new bindings and after some
years can be pushed into some dark corner.

That is basically the big problem with anything device tree.
It etches stuff in stone so it can't be changed, like ever.
Further, the decision on whether to etch this or that is
pushed to Linux subsystem maintainers, who are clearly
unsuited for the task. :(

quoted

As we preferred to give a chance to the generic pinconf/pinctrl for our
use by adding a little bit of flexibility, we are now in a situation
where we are nearly obliged to give up this approach and write a new
driver without the use of the generic facilities: what a pity!
We lost several months of useless work to match what we thought the
maintainer would prefer.

So Linus, do you confirm that we won't go further with this approach?

We are pretty disappointed by the way this interaction with the pinctrl
sub-system went.

I'm sorry, I am just trying to wait for consensus but it seems to
be hard for that to happen.

Things I need:

- More DT bindings people to look at this patch.

- More other driver maintainers to look at this patch.
  Sacha is one of those I'd like to get an opinion from
  for example. ACKs, Reviewed-by's are always good.

I welcome the addition of the combined pin number / mux value
properties like we have in the Mediatek driver. Making the pinctrl
core aware of these defines like Ludovic did is a good idea IMO.

Sascha with two 's'

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |