[RFC] hwmon: (tps23861) PSE-PD support, extend hwmon driver or move to pse-pd?

[RFC] hwmon: (tps23861) PSE-PD support, extend hwmon driver or move to pse-pd?

[Roman Kuzmitskii]

hi,

TP-Link TL-SG2210P v3 (realtek rtl838x, openwrt) has two tps23861 for PoE.
driver is hwmon only. ports can be enabled via in*_enable sysfs, rest is 
only reachable by poking i2c directly.
i would like to control them through PSE-PD instead (ethtool --set-pse).

i made patch for openwrt which registers pse_controller_dev from hwmon 
driver when node has pse-pis.
it does c33 admin state, power status, enable/disable, voltage and power 
class.
events are polled.
only ports with pse-pi node go to auto mode, and in*_enable is hidden 
while PSE is registered.
https://github.com/openwrt/openwrt/pull/26330 (tested on openwrt 6.18 
kernel)

openwrt reviewers asked me to check with you first which way you prefer:
1. extend hwmon driver with PSE-PD (like it is done right now)
2. move driver to drivers/net/pse-pd and extend it with PSE-PD
3. something else

thanks,
Roman

Re: [RFC] hwmon: (tps23861) PSE-PD support, extend hwmon driver or move to pse-pd?

[Guenter Roeck]

On 11/1/26 10:02, Roman Kuzmitskii wrote:
> hi,
> 
> TP-Link TL-SG2210P v3 (realtek rtl838x, openwrt) has two tps23861 for PoE.
> driver is hwmon only. ports can be enabled via in*_enable sysfs, rest is only reachable by poking i2c directly.
> i would like to control them through PSE-PD instead (ethtool --set-pse).
> 
> i made patch for openwrt which registers pse_controller_dev from hwmon driver when node has pse-pis.
> it does c33 admin state, power status, enable/disable, voltage and power class.
> events are polled.
> only ports with pse-pi node go to auto mode, and in*_enable is hidden while PSE is registered.
> https://github.com/openwrt/openwrt/pull/26330 (tested on openwrt 6.18 kernel)
> 
> openwrt reviewers asked me to check with you first which way you prefer:
> 1. extend hwmon driver with PSE-PD (like it is done right now)
> 2. move driver to drivers/net/pse-pd and extend it with PSE-PD
> 3. something else
> 
The hardware monitoring "enable" attribute is supposed to enable monitoring,
not to enable some power delivery or network (or any other) port. That "port
enable" feature using the hwmon ABI was sneaked in, and I am inclined to just
remove the upstream driver for severe ABI violation.

Guenter

Re: [RFC] hwmon: (tps23861) PSE-PD support, extend hwmon driver or move to pse-pd?

[Roman Kuzmitskii]

On 11/2/26 1:38 AM, Guenter Roeck wrote:
> On 11/1/26 10:02, Roman Kuzmitskii wrote:
>> openwrt reviewers asked me to check with you first which way you prefer:
>> 1. extend hwmon driver with PSE-PD (like it is done right now)
>> 2. move driver to drivers/net/pse-pd and extend it with PSE-PD
>> 3. something else
>>
> The hardware monitoring "enable" attribute is supposed to enable monitoring,
> not to enable some power delivery or network (or any other) port. That "port
> enable" feature using the hwmon ABI was sneaked in, and I am inclined to just
> remove the upstream driver for severe ABI violation.
> 
> Guenter

thanks, does second option look okay with taking tps23861 to drivers/net/pse-pd
with enable attribute going away?

Roman

Re: [RFC] hwmon: (tps23861) PSE-PD support, extend hwmon driver or move to pse-pd?

[Guenter Roeck]

On 11/1/26 10:50, Roman Kuzmitskii wrote:
> On 11/2/26 1:38 AM, Guenter Roeck wrote:
>> On 11/1/26 10:02, Roman Kuzmitskii wrote:
>>> openwrt reviewers asked me to check with you first which way you prefer:
>>> 1. extend hwmon driver with PSE-PD (like it is done right now)
>>> 2. move driver to drivers/net/pse-pd and extend it with PSE-PD
>>> 3. something else
>>>
>> The hardware monitoring "enable" attribute is supposed to enable monitoring,
>> not to enable some power delivery or network (or any other) port. That "port
>> enable" feature using the hwmon ABI was sneaked in, and I am inclined to just
>> remove the upstream driver for severe ABI violation.
>>
>> Guenter
> 
> thanks, does second option look okay with taking tps23861 to drivers/net/pse-pd
> with enable attribute going away?
> 
A perfect solution would be to move the driver to drivers/net/pse-pd (or elsewhere)
and implement the hwmon part as auxiliary driver, but I am also ok with moving it all.
Note that I do not review hwmon drivers outside drivers/hwmon, so if you move the
driver to drivers/net/pse-pd you can abuse the hwmon ABI as much as you want.

Guenter

Re: [RFC] hwmon: (tps23861) PSE-PD support, extend hwmon driver or move to pse-pd?

[Andrew Lunn]

On Sun, Nov 01, 2026 at 11:09:42AM -0800, Guenter Roeck wrote:
> On 11/1/26 10:50, Roman Kuzmitskii wrote:
> > On 11/2/26 1:38 AM, Guenter Roeck wrote:
> > > On 11/1/26 10:02, Roman Kuzmitskii wrote:
> > > > openwrt reviewers asked me to check with you first which way you prefer:
> > > > 1. extend hwmon driver with PSE-PD (like it is done right now)
> > > > 2. move driver to drivers/net/pse-pd and extend it with PSE-PD
> > > > 3. something else
> > > > 
> > > The hardware monitoring "enable" attribute is supposed to enable monitoring,
> > > not to enable some power delivery or network (or any other) port. That "port
> > > enable" feature using the hwmon ABI was sneaked in, and I am inclined to just
> > > remove the upstream driver for severe ABI violation.
> > > 
> > > Guenter
> > 
> > thanks, does second option look okay with taking tps23861 to drivers/net/pse-pd
> > with enable attribute going away?
> > 
> A perfect solution would be to move the driver to drivers/net/pse-pd (or elsewhere)
> and implement the hwmon part as auxiliary driver, but I am also ok with moving it all.
> Note that I do not review hwmon drivers outside drivers/hwmon, so if you move the
> driver to drivers/net/pse-pd you can abuse the hwmon ABI as much as you want.

Although some network maintainers do have a basic understanding of
hwmon, and might push back on such abuse.

Does the tps23861 share anything with the tps23881? It already has a
pse-pd driver.

       Andrew

Re: [RFC] hwmon: (tps23861) PSE-PD support, extend hwmon driver or move to pse-pd?

[Robert Marko]

On Sun, Nov 1, 2026 at 7:38 PM Guenter Roeck <linux@roeck-us.net> wrote:
>
> On 11/1/26 10:02, Roman Kuzmitskii wrote:
> > hi,
> >
> > TP-Link TL-SG2210P v3 (realtek rtl838x, openwrt) has two tps23861 for PoE.
> > driver is hwmon only. ports can be enabled via in*_enable sysfs, rest is only reachable by poking i2c directly.
> > i would like to control them through PSE-PD instead (ethtool --set-pse).
> >
> > i made patch for openwrt which registers pse_controller_dev from hwmon driver when node has pse-pis.
> > it does c33 admin state, power status, enable/disable, voltage and power class.
> > events are polled.
> > only ports with pse-pi node go to auto mode, and in*_enable is hidden while PSE is registered.
> > https://github.com/openwrt/openwrt/pull/26330 (tested on openwrt 6.18 kernel)
> >
> > openwrt reviewers asked me to check with you first which way you prefer:
> > 1. extend hwmon driver with PSE-PD (like it is done right now)
> > 2. move driver to drivers/net/pse-pd and extend it with PSE-PD
> > 3. something else
> >
> The hardware monitoring "enable" attribute is supposed to enable monitoring,
> not to enable some power delivery or network (or any other) port. That "port
> enable" feature using the hwmon ABI was sneaked in, and I am inclined to just
> remove the upstream driver for severe ABI violation.

Hi Guenther, I am the original driver author for tps23861.

Sorry for the ABI violation, that wasn't really clear back when it was
being written, especially with no PSE subsystem at all.

We can always just remove the enable attribute, the IC works in auto
mode by default anyway, instead of dropping the driver
as it is still useful for monitoring.

That being said, ideally it would just be moved to the PSE subsystem.

Regards,
Robert

>
> Guenter
>


-- 
Robert Marko
Staff Embedded Linux Engineer
Sartura d.d.
Lendavska ulica 16a
10000 Zagreb, Croatia
Email: robert.marko@sartura.hr
Web: www.sartura.hr

Re: [RFC] hwmon: (tps23861) PSE-PD support, extend hwmon driver or move to pse-pd?

[Kory Maincent]

On 11/1/26 20:57, Andrew Lunn wrote:
> On Sun, Nov 01, 2026 at 11:09:42AM -0800, Guenter Roeck wrote:
>> On 11/1/26 10:50, Roman Kuzmitskii wrote:
>>> On 11/2/26 1:38 AM, Guenter Roeck wrote:
>>>> On 11/1/26 10:02, Roman Kuzmitskii wrote:
>>>>> openwrt reviewers asked me to check with you first which way you prefer:
>>>>> 1. extend hwmon driver with PSE-PD (like it is done right now)
>>>>> 2. move driver to drivers/net/pse-pd and extend it with PSE-PD
>>>>> 3. something else

I am currently working on this topic by extending the tps23881 driver.
Please don't do it yourself to avoid working in parallel on the same 
topic. On the other hand, it would be nice if you can test it.
Roman I will add you in CC of my series.

>>>>>
>>>> The hardware monitoring "enable" attribute is supposed to enable monitoring,
>>>> not to enable some power delivery or network (or any other) port. That "port
>>>> enable" feature using the hwmon ABI was sneaked in, and I am inclined to just
>>>> remove the upstream driver for severe ABI violation.

Do you think it would be possible to remove the current tps23861 driver 
and binding? This would break the backward compatibility.
I am currently using "tps23861-poe" compatible name binding.

Regards,
-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com