On Tue, Feb 12, 2019 at 10:42:03AM +0100, Harald Geyer wrote:
> Maxime Ripard writes:
> > On Mon, Feb 11, 2019 at 08:32:35PM +0100, Harald Geyer wrote:
> > > > We want to model this properly. I guess using a pinctrl driver
> > > > controlled through GPIO (similar to what regulator-gpio is) would be a
> > > > good first step.
> > > 
> > > I considered this too, but didn't like it:
> > > 
> > > 1) Seems like a bit of overkill.
> > >
> > > 2) The HW at hand is a rather different kind of multiplexer than
> > > what pinctrl assumes. We don't want two mutually exclusive devices,
> > > (Ie don't make the kernel unbind /dev/console for the sake of audio.)
> > > but we want switch the jack between two devices, that might both be
> > > active at the same time. This looks more like the channel multiplexers
> > > used with many ADCs and such. I guess, I could start a new subsystem
> > > around this. Seems like even more overkill.
> > 
> > I'm not quite sure about how that's different from what pinctrl
> > assumes. pinctrl assumes to handle devices that have multiple signals
> > as input, and one as output. Isn't that exactly what you have?
> 
> I think the pinctrl way would be to have the audio card device
> request the HP jack and the uart node request the HP jack and only
> once device could probe successfully. Ie it is about ressource
> allocation, not true multiplexing where both devices can use the
> ressource at the same time. Am I wrong?

By default, it's what happens yes, but you can definitely have more
complex behaviour that would support changing dynamically back and
forth that muxing.

> Or course we don't actually want true multiplexing for audio quality
> reasons, but I don't see how we could use pinctrl without doing nasty
> things to /dev/console ...

You can't have multiplexing, because that device doesn't allow you to:
it allows only a single signal out at a given time. If you're using
audio, you will not be able to get your logs out of the UART.

And /dev/console is always there as far as I'm aware.

> > And pinctrl can be used dynamically as well if you need to
> 
> Can you explain or point me to the relevant explanation in the docs?
> I don't seem to know about it.

pinctrl_select_state is what you would want to call, you would have a
good example of that runtime change in
drivers/i2c/muxes/i2c-mux-pinctrl.c.

> > > Instead I just got the original patch working, by implementing
> > > "output-high" DT property in sunxi-pinctrl. I'll send a patch for
> > > review soon.
> > 
> > What do you want to do with output-high exactly?
> 
> Exactly what I do in the patch that started this thread.
> (I'll resend when wens' cpvdd patch is available for me to rebase onto.)

There's a few issues with that approach as well:

  - We're actively trying to remove the pinctrl nodes for the GPIOs

  - It's completely static: if one only wants to use the UART all the
    time, they would have to change the DTB, which might or might not
    be possible. Note that this could be trickier: why would you
    prevent the console from working for example if the audio support
    isn't built in? or as a module that might or might not be loaded?

Maxime

-- 
Maxime Ripard, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com