Re: [PATCH 3/3] thermal: of: notify sensor driver on trip updates

[navneet kumar <navneetk@nvidia.com>]

On 12/01/2014 01:23 PM, Eduardo Valentin wrote:
> * PGP Signed by an unknown key
> 
> On Mon, Dec 01, 2014 at 12:45:52PM -0800, navneet kumar wrote:
>> Hi Eduardo,
>>
>> On 11/27/2014 06:32 AM, Eduardo Valentin wrote:
>>>> Old Signed by an unknown key
>>>
>>> Hello Navneet,
>>>
>>> On Wed, Nov 26, 2014 at 05:16:29PM -0800, Navneet Kumar wrote:
>>>> From: navneet kumar <navneetk@nvidia.com>
>>>>
>>>> some thermal sensor hardwares include logic which
>>>> can raise interrupts at certain programmed temperature
>>>> thresholds.
>>>>
>>>> Drivers for such sensors should be able to learn the
>>>> appropriate threshold temperatures for interrupts by querying
>>>> the thermal framework.
>>>>
>>>> This change provides a mechanism to allow a sensor driver to
>>>> update it's thresholds when userspace changes a trip point
>>>> temperature.
>>>>
>>>> While this behavior may not make sense in thermal zones
>>>> with more than one sensor, no such examples exist in
>>>> the kernel.
>>>>
>>>> Signed-off-by: navneet kumar <navneetk@nvidia.com>
>>>> ---
>>>>  drivers/thermal/of-thermal.c | 7 +++++++
>>>>  include/linux/thermal.h      | 1 +
>>>>  2 files changed, 8 insertions(+)
>>>>
>>>> diff --git a/drivers/thermal/of-thermal.c b/drivers/thermal/of-thermal.c
>>>> index 3d47a0cf3825..3568e4a586dc 100644
>>>> --- a/drivers/thermal/of-thermal.c
>>>> +++ b/drivers/thermal/of-thermal.c
>>>> @@ -258,6 +258,9 @@ static int of_thermal_set_trip_temp(struct thermal_zone_device *tz, int trip,
>>>>  	/* thermal framework should take care of data->mask & (1 << trip) */
>>>>  	data->trips[trip].temperature = temp;
>>>>  
>>>> +	if (data->sops.trip_update)
>>>> +		data->sops.trip_update(data->sensor_data, trip);
>>>> +
>>>>  	return 0;
>>>>  }
>>>>  
>>>> @@ -285,6 +288,9 @@ static int of_thermal_set_trip_hyst(struct thermal_zone_device *tz, int trip,
>>>>  	/* thermal framework should take care of data->mask & (1 << trip) */
>>>>  	data->trips[trip].hysteresis = hyst;
>>>>  
>>>> +	if (data->sops.trip_update)
>>>> +		data->sops.trip_update(data->sensor_data, trip);
>>>> +
>>>>  	return 0;
>>>>  }
>>>>  
>>>> @@ -500,6 +506,7 @@ void thermal_zone_of_sensor_unregister(struct device *dev,
>>>>  
>>>>  	tz->sops.get_temp = NULL;
>>>>  	tz->sops.get_trend = NULL;
>>>> +	tz->sops.trip_update = NULL;
>>>>  	tz->sensor_data = NULL;
>>>>  	mutex_unlock(&tzd->lock);
>>>>  }
>>>> diff --git a/include/linux/thermal.h b/include/linux/thermal.h
>>>> index 58341c56a01f..b93e65815175 100644
>>>> --- a/include/linux/thermal.h
>>>> +++ b/include/linux/thermal.h
>>>> @@ -292,6 +292,7 @@ struct thermal_genl_event {
>>>>  struct thermal_of_sensor_ops {
>>>>  	int (*get_temp)(void *, long *);
>>>>  	int (*get_trend)(void *, long *);
>>>> +	int (*trip_update)(void *, int);
>>>
>>> First thing I ask you is to update your work on top of my -linus branch,
>>> as I already mentioned. Reasoning is that part of the changes you are
>>> sending is already there.
>> will do.
>>>
>>> As for this new callback, I am fine with it as long as it is also
>>> available for drivers that do not use of-thermal. Once again, of-thermal
>>> is not a competitor of thermal core. It will never be. It is not a new
>>> thermal API. 
>> I agree that this callback is not a part of the thermal_core functionality.
>> However, when a driver registers directly with the thermal_core (doesn't use
>> of-thermal), it 'owns' the set_trip_XX callbacks in the first place; which is
>> the sole purpose of using the 'trip_update' callback in of-thermal.
>>
>> Adding an additional 'update' to the thermal_core ops would be a no-op. right?
> 
> Yes, you are right. Now I understand your point. 
> 
> Can we then re-use the .set_trips nomenclature?
Sorry, I fail to understand. Are you suggesting to re-use the interface for set_trip 'temp' as well as 'hyst'?
If so, is it just to maintain the commonality across thermal_core and of-thermal interfaces?

The way i see it, the driver just needs to get some kind of 'update' that 'something' changed with
a trip point; and can later query the trips from of-thermal. (Lukasz's patch helps with that).
Functionality-wise, using two callbacks seems excessive. But i may be wrong :-)

> 
> Cheers,
> 
>>>
>>> That said, it does not make sense to have functionality in of-thermal that
>>> do not belong to thermal core. Exceptions are, of course, for helping
>>> doing the same operations we already have in thermal core.
>>>
>>> All the best,
>>>
>>> Eduardo Valentin
>>>
>>>>  };
>>>>  
>>>>  /* Function declarations */
>>>> -- 
>>>> 1.8.1.5
>>>>
>>>
>>> * Unknown Key
>>> * 0x7DA4E256
>>>
> 
> * Unknown Key
> * 0x7DA4E256
>