On Fri Feb 20, 2026 at 11:50 AM CET, Andy Shevchenko wrote:
> On Fri, Feb 20, 2026 at 10:19:07AM +0100, Luca Weiss wrote:
>> If the devicetree node is not referenced in a zone under /thermal-zones,
>> devm_thermal_of_zone_register will fail with -ENODEV.
>
> devm_thermal_of_zone_register()
Ack
>
>> Since the driver is now also registering as an IIO device[0], allow the
>> probe to continue without the thermal zone.
>
> Isn't it dangerous?
Why? The idea is that generic-adc-thermal is the middleman to convert
from one IIO input to one IIO output, and is purely informational, so
that user space can get some temperature value to display somewhere.
How thermal management will be hooked up in the future to charger
drivers is a bit out of scope here I'd say. There's not even any cooling
support in the power supply core anymore, that was ripped out a while
ago if I'm not mistaken.
>
>> We also can't use gadc_thermal_get_temp anymore because we haven't
>
> gadc_thermal_get_temp()
Ack
>
>> necessarily initialized tz_dev.
>
>> [0] commit 3762f5851ac5 ("thermal/drivers/thermal-generic-adc: Add
>> temperature sensor channel")
>
> You can make it a Link tag:
>
> Link: https://git.kernel.org/torvalds/c/3762f5851ac5 [0]
Ack
>
>> Signed-off-by: Luca Weiss <[email protected]>
>
> ...
>
>> if (IS_ERR(gti->tz_dev)) {
>> ret = PTR_ERR(gti->tz_dev);
>> - if (ret != -EPROBE_DEFER)
>> - dev_err(dev,
>> - "Thermal zone sensor register failed: %d\n",
>> - ret);
>> - return ret;
>> - }
>> + if (ret == -EPROBE_DEFER)
>> + return ret;
>
> I believe it's better to return all possible errors. If you see the
> similarities with regulator APIs, then use an explicit check for -ENODEV.
Sure, will update this.
Regards
Luca
>
>> - devm_thermal_add_hwmon_sysfs(dev, gti->tz_dev);
>> + dev_info(dev, "Thermal zone sensor register failed: %d\n",
>> + ret);
>> + } else {
>> + devm_thermal_add_hwmon_sysfs(dev, gti->tz_dev);
>> + }
