Re: [PATCH leds v5 02/12] leds: turris-omnia: Use command execution functions from the MCU driver

[Andy Shevchenko <andy@kernel.org>]

On Tue, Nov 05, 2024 at 10:07:51AM +0100, Marek Behún wrote:
> On Mon, Nov 04, 2024 at 04:51:24PM +0200, Andy Shevchenko wrote:
> > On Mon, Nov 04, 2024 at 03:19:14PM +0100, Marek Behún wrote:

...

> > I'm wondering why the MCU driver (and node) is not represented as syscon
> > (with some regmap beneath it).
> > 
> > In such a case it would be something like
> > 
> >   foo = syscon_regmap_lookup_by_compatible();
> > 
> > here instead of all these dances.
> 
> Hi Andy,
> 
> the MCU interface is command-reply oriented. It is incompatible with
> regmap.

I'm not sure I understand the impediment here. There are plenty of hardware
that uses the similar approach and well compatible with regmap (assuming
custom ->read*() / ->write*() callbacks)...

> I investigated this back in 2019 and explained to Jacek why it
> is not possible, but can't find the e-mail on mailing lists, so I am
> attaching it.

...but I'm not insisting you to revisit this right now, just maybe
you can think more about this again at some point.

> So regmap is most probably not possible, unless things changed.
> 
> It is possible to add MCU node to the DT binding and find the device
> that way. But if the device-tree does not contain the MCU node, the
> driver would still have to fall back to this dance, for backwards
> compatibility. Otherwise it would not be able to determine whether gamma
> correction is supported with old device tree, as it does currently.
> 
> I guess I could break backwards compatibility with old device tree with
> this small feature. I don't think there are any users that don't use
> TurrisOS, do upgrade the kernel, but don't upgrade the device-tree...

-- 
With Best Regards,
Andy Shevchenko