Re: [PATCH] i2c: core: dont change pinmux state to GPIO during recovery setup

[Linus Walleij]

On Thu, Nov 9, 2023 at 9:30 PM Russell King (Oracle)
<linux@xxxxxxxxxxxxxxx> wrote:
> On Thu, Nov 09, 2023 at 09:04:29PM +0100, Linus Walleij wrote:

> > > After it was converted to it, the I2C bus completely stopped working
> > > on Armada 3720
> > > if I2C recovery is enabled by making the recovery pinctrl available in DTS.
> >
> > Shouldn't we just revert that patch until we can figure this out then?
>
> Note that when I wrote the i2c-pxa recovery code (which was developed
> and tested on Armada 3720 - the uDPU) it had to work... when the
> suggestion came up to implement generic recovery, I stated:
>
> http://archive.lwn.net:8080/linux-kernel/20200705210942.GA1055@kunai/T/#mf7f862fcd53245f14fb650d33c29cf139d41039d

Makes me even more convinced that we should just revert this. i.e.
commit 0b01392c18b9993a584f36ace1d61118772ad0ca
i2c: pxa: move to generic GPIO recovery

There is even:
https://lore.kernel.org/linux-i2c/20201209204645.GF3499@kunai/

"In case we missed a glitch, we can still revert the patch later."
Well this is later.

Robert can you see if it possible to revert, that things work after a
revert and send a revert patch?

> > > I then spent quite a while trying to bisect the exact change that
> > > causes this issue
> > > in the conversion as code is almost identical to what the driver was
> > > doing previously,
> > > and have bisected it down to pinctrl_select_state(bri->pinctrl,
> > > bri->pins_gpio) being
> > > called before SDA and SCL pins are obtained via devm_gpiod_get().
>
> Yes, indeed. That's because the pinctrl internals get confused. I sent
> you an email about it on 6th December 2019
>
> "pinctrl states vs pinmux vs gpio (i2c bus recovery)"

I found it:
https://lore.kernel.org/all/20191206173343.GX25745@xxxxxxxxxxxxxxxxxxxxx/

Sadly I had no good advice for any simple elegant solutions
to the problem, but the more complicated solution does
work so let's go for that.

> which is why i2c-pxa did things the way it did in my commit
> "i2c: pxa: implement generic i2c bus recovery".

I think we need to go back to this.

It's nice with the ambition to create generic code of course, but
sometimes it is better to just roll something IP-unique.

Yours,
Linus Walleij