Re: [PATCH v3 1/2] Documentation: devicetree: update bindings for tps23861

From: Guenter Roeck
Date: Fri Aug 26 2022 - 08:22:04 EST


On Fri, Aug 26, 2022 at 09:56:29AM +0300, Krzysztof Kozlowski wrote:
> On 25/08/2022 18:31, Robert Marko wrote:
> > On Thu, Aug 25, 2022 at 5:29 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> >>
> >> On Thu, Aug 25, 2022 at 05:07:45PM +0200, Robert Marko wrote:
> >>> On Thu, Aug 25, 2022 at 5:02 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> >>>>
> >>>> On Thu, Aug 25, 2022 at 04:37:36PM +0200, Andreas Böhler wrote:
> >>>>> The tps23861 driver does not initialize the chip and relies on it being
> >>>>> in auto-mode by default. On some devices, these controllers default to
> >>>>> OFF-Mode and hence cannot be used at all.
> >>>>>
> >>>>> This brings minimal support for initializing the controller in a user-
> >>>>> defined mode.
> >>>>>
> >>>>> Signed-off-by: Andreas Böhler <dev@xxxxxxxxxxx>
> >>>>
> >>>> nack for the series, sorry. The suggested properties are not hardware
> >>>> monitoring but phy properties. There should be a separate phy driver
> >>>> to manage those.
> >>>>
> >>>> Also, as mentioned, the hwmon 'enable' attribute is abused to control
> >>>> port functionality and should be removed.
> >>>
> >>> Hi Guenter,
> >>> Are you referring to an ethernet PHY driver or the generic PHY framework?
> >>>
> >>
> >> Could be both, though ethernet phy sounds about right for me.
> >> I don't know where/how similar chips are handled. hwmon is most definitey
> >> the wrong place.
> >
> > Hi,
> >
> > Well, that is the thing, this is definitively not an ethernet PHY nor
> > a PHY of any other kind.
> > I dont see where it would fit if not hwmon, there is no more specific
> > subsystem in the
> > kernel.
>
> It's not hwmon. The device has monitoring capabilities, but it's only
> one piece and calling something hwmon just because can provide sensor
> data is like calling a plane a car, because it has wheels.
>
> Maybe this is similar to these series:
> https://lore.kernel.org/linux-devicetree/20220825130211.3730461-1-o.rempel@xxxxxxxxxxxxxx/
> ?
>
> The datasheet says it is a "PSE Controller" so looks similar to the
> problem solved above...

Excellent find. That infrastructure is exactly what the driver for this chip
needs to tie into. I would suggest to get in touch with the author of that
series - it is quite likely that they are working on adding support for one
or more real PSE chips.

The only open question is if the hwmon driver should be retained as a
separate driver or be implemented as part of the PSE networking driver.
I am open to both.

Thanks,
Guenter