On Thu, Sep 28, 2023 at 08:53:19AM +0200, Luca Ceresoli wrote:
> Hello Linus, Prathamesh,
>
> On Wed, 27 Sep 2023 10:54:15 +0200
> Linus Walleij <linus.walleij@xxxxxxxxxx> wrote:
>
> > On Mon, Sep 25, 2023 at 6:30 PM Luca Ceresoli <luca.ceresoli@xxxxxxxxxxx> wrote:
> >
> > > The symptom is that i2c3 is not working anymore, the I2C lines being
> > > always high. No other known issues at the moment.
> >
> > Hm....
> >
> > >> + {"nvidia,function", TEGRA_PINCONF_PARAM_FUNCTION},
> > >
> > > FYI, I reduced your patch to only this line plus the one in the
> > > pinctrl-tegra.h and the problem appears as well.
> >
> > I think there is a conflict now, that the pinconf is "stealing" the function
> > assignment from the pinmux call.
> >
> > It's just a debugprint, I will revert the patch, Luca can investigate and you
>
> Thanks for the quick revert Linus.
>
> > can test a new patch then we will merge that.
>
> Prathamesh, if you send a new patch it would be great if you can Cc: me so I
> can test it.
I was able to reproduce this on tegra20-trimslice, and it looks indeed
that the "shortcut" of supporting this through the pinconf "framework"
doesn't work. In addition to the pinmux now no longer getting applied
(exactly why that is I don't think I understand), it also leads to weird
things in other parts of the debugfs output. For example the code now
ends up trying to read a u32 from the nvidia,function property, which is
actually a string. That seems to be fine, but obviously it doesn't yield
the value that is expected for a function and messes up the config param
which will then later on fail to properly display.
Anyway, turns out the correct implementation is even shorter. I've sent
out a patch.
Thierry
Attachment:
signature.asc
Description: PGP signature
