Re: [PATCH] ARM: dts: document pinctrl-single,pins when #pinctrl-cells = 2

[Tony Lindgren]

* Drew Fustini <drew@xxxxxxxxxxxxxxx> [200917 10:39]:
> On Thu, Sep 17, 2020 at 03:00:36AM -0700, Trent Piepho wrote:
> > On Thu, Sep 17, 2020 at 2:20 AM Drew Fustini <drew@xxxxxxxxxxxxxxx> wrote:
> > >
> > > On Thu, Sep 17, 2020 at 02:03:46AM -0700, Trent Piepho wrote:
> > > > On Mon, Sep 14, 2020 at 3:44 AM Drew Fustini <drew@xxxxxxxxxxxxxxx> wrote:
> > > > >
> > > > > +
> > > > > +When #pinctrl-cells = 2, then setting a pin for a device could be done with:
> > > > > +
> > > > > +       pinctrl-single,pins = <0xdc 0x30 0x07>;
> > > > > +
> > > > > +Where 0x30 is the pin configuration value and 0x07 is the pin mux mode value.
> > > > > +See the device example and static board pins example below for more information.
> > > >
> > > > Pin configuration and mux mode don't mean anything in pinctrl-single.
> > > > On another machine, mux mode might not be programmed this way or even
> > > > exist.  Or the location of bits would probably be different, and this
> > > > would seem to imply the 0x07 would get shifted to the correct location
> > > > for where the pin mux setting was on that machine's pinctrl registers.
> > > >
> > > > It seems like it would be better to explain the values are ORed together.
> > >
> > > I descirbed it as seoerate values as I did not want to prescribe what
> > > the pcs driver would do with those values.  But, yes, it is a just an OR
> > > operation, so I could change the language to reflect tat.
> > 
> > If you don't say what the pinctrl-single driver does with the values,
> > how would anyone know how to use it?
> > 
> > > > What is the purpose of this change anyway?  It seems like in the end
> > > > it just does what it did before.  The data is now split into two cells
> > > > in the device tree, but why?
> > >
> > > These changes were a result of desire to seperate pinconf and pinmux.
> > > Tony raised the idea in a thread at the end of May [1].
> > >
> > > Tony wrote:
> > > > Only slightly related, but we should really eventually move omaps to use
> > > > #pinctrl-cells = <2> (or 3) instead of 1, and pass the pinconf seprately
> > > > from the mux mode. We already treat them separately with the new
> > > > AM33XX_PADCONF macro, so we'd only have to change one SoC at a time to
> > > > use updated #pinctrl-cells. But I think pinctrl-single might need some
> > > > changes before we can do that.
> > 
> > I still don't see what the goal is here.  Support generic pinconf?
> 
> My interest is came out of my desire to turn on generic pinconf for AM3358
> and I had to fix a bug that was breaking compatible "pinconf,single":
> f46fe79ff1b6 ("pinctrl-single: fix pcs_parse_pinconf() return value")
> 
> > Also note that while AM33XX_PADCONF() is changed, there is an in tree
> > board that doesn't use it, so it's broken now.  I found this change
> > when it broke my out of tree board, due to the dtsi change not being
> > reflected in my board's pinctrl values.
> 
> Thanks, that is a good point that arch/arm/boot/dts/am335x-guardian.dts
> needs to be converted from AM33XX_IOPAD to AM33XX_PADCONF.  I'll submit
> a patch for that.
> 
> Regarding AM33XX_PADCONF() restructuring, the change to have seperate
> arguments for direction and mux in AM33XX_PADCONF() predates my
> invovlement, so I've CC'd Christina Quast.
> 
>     commit f1ff9be7652b716c7eea67c9ca795027d911f148
>     Author: Christina Quast <cquast@xxxxxxxxxxxxxxxxxxx>
>     Date:   Mon Apr 8 10:01:51 2019 -0700
> 
>     ARM: dts: am33xx: Added AM33XX_PADCONF macro
>     
>     AM33XX_PADCONF takes three instead of two parameters, to make
>     future changes to #pinctrl-cells easier.
>     
>     For old boards which are not mainlined, we left the AM33XX_IOPAD
>     macro.
>     
>     Signed-off-by: Christina Quast <cquast@xxxxxxxxxxxxxxxxxxx>
>     Reviewed-by: Rob Herring <robh@xxxxxxxxxx>
>     Signed-off-by: Tony Lindgren <tony@xxxxxxxxxxx>
> 
> Hopefully, Tony can also chime in.

Also FYI, folks have also complained for a long time that the pinctrl-single
binding mixes mux and conf values while they should be handled separately.

Regards,

Tony