2024-04-02 20:50:43

by Chris Packham

[permalink] [raw]
Subject: hwmon: label vs temp%d_label

Hi Guenter, Jean,

I've got a requirement to add some meaningful names to some hwmon
sensors (LM75 specifically) so that we can provide some indication of
where on a board the sensor is located (e.g. "Intake" vs "Exhaust" vs
"Near that really hot chip").

I see that the sysfs ABI documents both "label" for the chip and
"temp[1-*]_label" (as well as similar fan and Vin attributes). The
latter seem to be supported by the hwmon core but I don't see anything
for the former (I'm struggling to find any driver that supports a
chip-wide label).

Assuming I want to have a label added in the device tree to a lm75 would
something like the following be acceptable

      sensor@48 {
        compatible = "national,lm75";
        reg = <0x48>;
        label = "Intake";
      };

I'd then update the lm75 driver to grab that from the devicetree and use
it to provide the hwmon_temp_label attribute.

Thanks,
Chris


2024-04-02 20:59:53

by Guenter Roeck

[permalink] [raw]
Subject: Re: hwmon: label vs temp%d_label

On Tue, Apr 02, 2024 at 08:24:37PM +0000, Chris Packham wrote:
> Hi Guenter, Jean,
>
> I've got a requirement to add some meaningful names to some hwmon
> sensors (LM75 specifically) so that we can provide some indication of
> where on a board the sensor is located (e.g. "Intake" vs "Exhaust" vs
> "Near that really hot chip").
>
> I see that the sysfs ABI documents both "label" for the chip and
> "temp[1-*]_label" (as well as similar fan and Vin attributes). The
> latter seem to be supported by the hwmon core but I don't see anything
> for the former (I'm struggling to find any driver that supports a
> chip-wide label).
>
> Assuming I want to have a label added in the device tree to a lm75 would
> something like the following be acceptable
>
> ????? sensor@48 {
> ??????? compatible = "national,lm75";
> ??????? reg = <0x48>;
> ??????? label = "Intake";
> ????? };
>
> I'd then update the lm75 driver to grab that from the devicetree and use
> it to provide the hwmon_temp_label attribute.
>

Have you tried just declaring the label property as you suggested above
in your system without doing anything else, and looked at the generated
sysfs attributes ?

Guenter

2024-04-02 21:32:51

by Chris Packham

[permalink] [raw]
Subject: Re: hwmon: label vs temp%d_label


On 3/04/24 09:59, Guenter Roeck wrote:
> On Tue, Apr 02, 2024 at 08:24:37PM +0000, Chris Packham wrote:
>> Hi Guenter, Jean,
>>
>> I've got a requirement to add some meaningful names to some hwmon
>> sensors (LM75 specifically) so that we can provide some indication of
>> where on a board the sensor is located (e.g. "Intake" vs "Exhaust" vs
>> "Near that really hot chip").
>>
>> I see that the sysfs ABI documents both "label" for the chip and
>> "temp[1-*]_label" (as well as similar fan and Vin attributes). The
>> latter seem to be supported by the hwmon core but I don't see anything
>> for the former (I'm struggling to find any driver that supports a
>> chip-wide label).
>>
>> Assuming I want to have a label added in the device tree to a lm75 would
>> something like the following be acceptable
>>
>>       sensor@48 {
>>         compatible = "national,lm75";
>>         reg = <0x48>;
>>         label = "Intake";
>>       };
>>
>> I'd then update the lm75 driver to grab that from the devicetree and use
>> it to provide the hwmon_temp_label attribute.
>>
> Have you tried just declaring the label property as you suggested above
> in your system without doing anything else, and looked at the generated
> sysfs attributes ?

I have not. But in my defense I'm also using an older kernel LTS that
doesn't have commit e1c9d6d61ddf ("hwmon: Add "label" attribute"). But
now that I know it exists I can carry it as a local patch until we next
update.

Thanks,
Chris

2024-04-02 21:58:21

by Chris Packham

[permalink] [raw]
Subject: Re: hwmon: label vs temp%d_label


On 3/04/24 10:22, Chris Packham wrote:
>
> On 3/04/24 09:59, Guenter Roeck wrote:
>> On Tue, Apr 02, 2024 at 08:24:37PM +0000, Chris Packham wrote:
>>> Hi Guenter, Jean,
>>>
>>> I've got a requirement to add some meaningful names to some hwmon
>>> sensors (LM75 specifically) so that we can provide some indication of
>>> where on a board the sensor is located (e.g. "Intake" vs "Exhaust" vs
>>> "Near that really hot chip").
>>>
>>> I see that the sysfs ABI documents both "label" for the chip and
>>> "temp[1-*]_label" (as well as similar fan and Vin attributes). The
>>> latter seem to be supported by the hwmon core but I don't see anything
>>> for the former (I'm struggling to find any driver that supports a
>>> chip-wide label).
>>>
>>> Assuming I want to have a label added in the device tree to a lm75
>>> would
>>> something like the following be acceptable
>>>
>>>         sensor@48 {
>>>           compatible = "national,lm75";
>>>           reg = <0x48>;
>>>           label = "Intake";
>>>         };
>>>
>>> I'd then update the lm75 driver to grab that from the devicetree and
>>> use
>>> it to provide the hwmon_temp_label attribute.
>>>
>> Have you tried just declaring the label property as you suggested above
>> in your system without doing anything else, and looked at the generated
>> sysfs attributes ?
>
> I have not. But in my defense I'm also using an older kernel LTS that
> doesn't have commit e1c9d6d61ddf ("hwmon: Add "label" attribute"). But
> now that I know it exists I can carry it as a local patch until we
> next update.

Related is there an lm-sensors change that uses this attribute for
display purposes?

I do have a couple of PRs open on the lm-sensors github project I'd like
to see merged but given recent events this should absolutely not be
construed as a criticism of anyone maintaining lm-sensors merely a query
as to whether PRs are the right path for changes or if they should be
sent to a mailing list somewhere.

2024-04-02 22:19:11

by Guenter Roeck

[permalink] [raw]
Subject: Re: hwmon: label vs temp%d_label

On Tue, Apr 02, 2024 at 09:57:44PM +0000, Chris Packham wrote:
>
> On 3/04/24 10:22, Chris Packham wrote:
> >
> > On 3/04/24 09:59, Guenter Roeck wrote:
> >> On Tue, Apr 02, 2024 at 08:24:37PM +0000, Chris Packham wrote:
> >>> Hi Guenter, Jean,
> >>>
> >>> I've got a requirement to add some meaningful names to some hwmon
> >>> sensors (LM75 specifically) so that we can provide some indication of
> >>> where on a board the sensor is located (e.g. "Intake" vs "Exhaust" vs
> >>> "Near that really hot chip").
> >>>
> >>> I see that the sysfs ABI documents both "label" for the chip and
> >>> "temp[1-*]_label" (as well as similar fan and Vin attributes). The
> >>> latter seem to be supported by the hwmon core but I don't see anything
> >>> for the former (I'm struggling to find any driver that supports a
> >>> chip-wide label).
> >>>
> >>> Assuming I want to have a label added in the device tree to a lm75
> >>> would
> >>> something like the following be acceptable
> >>>
> >>> ? ????? sensor@48 {
> >>> ? ??????? compatible = "national,lm75";
> >>> ? ??????? reg = <0x48>;
> >>> ? ??????? label = "Intake";
> >>> ? ????? };
> >>>
> >>> I'd then update the lm75 driver to grab that from the devicetree and
> >>> use
> >>> it to provide the hwmon_temp_label attribute.
> >>>
> >> Have you tried just declaring the label property as you suggested above
> >> in your system without doing anything else, and looked at the generated
> >> sysfs attributes ?
> >
> > I have not. But in my defense I'm also using an older kernel LTS that
> > doesn't have commit e1c9d6d61ddf ("hwmon: Add "label" attribute"). But
> > now that I know it exists I can carry it as a local patch until we
> > next update.
>
> Related is there an lm-sensors change that uses this attribute for
> display purposes?
>
Sorry, I don't know. I stopped paying attention to the lm-sensors package
a long time ago, and I don't know its status. I just don't have the time.

> I do have a couple of PRs open on the lm-sensors github project I'd like
> to see merged but given recent events this should absolutely not be
> construed as a criticism of anyone maintaining lm-sensors merely a query
> as to whether PRs are the right path for changes or if they should be
> sent to a mailing list somewhere.
>
I have no idea, sorry.

Guenter