On Thu, 2014-09-11 at 11:53 -0400, Eduardo Valentin wrote:

Arnd,

On Thu, Sep 11, 2014 at 02:32:08PM +0200, Arnd Bergmann wrote:

quoted

On Thursday 11 September 2014 08:18:43 Eduardo Valentin wrote:

quoted

quoted

As what we want is to make thermal driver have a chance to configure the
hardware shutdown registers, I'm thinking if we can do this without
representing the hardware shutdown value as a trip point.
Say,
1. parse DT, and get the hardware shutdown temperature value, and store
it somewhere, e.g. struct __thermal_zone.
2. introduce a new parameter, int (*set_hardware_trip)(void *, long *),
in thermal_zone_of_sensor_register().
3. invoke set_hard_trip(tz, hardware_shutdown_temperature_value) in
thermal_zone_of_sensor_register().

The only issue I have with the above proposal is that not all platforms
use DT. Some still boot with boardfiles, for instance. Thus, the
parameter to configure hardware thermal shutdown needs to be common on
thermal core, not specific to of-thermal. Do you agree?

Do you know of a machine that can't yet be converted to DT and that
needs this driver? In case of rockchips that is certainly not the
case, and we don't care about anybody trying to use board files out
of tree, they can just hack the thermal support as well.

I see. Again, the only concern I have is to produce thermal framework APIs
that would be only in the of-thermal. My point is not specific to this
patch, or this platform, but with a detail in the above proposal. 

While I agree to have a trip specific to configurable hardware triggered
thermal shutdown, I just don't see why it needs to be a feature
implemented only via of-thermal. It has to be properly defined in
thermal core.

The proposal of of-thermal is not to become a separate/competing thermal
framework.

Agreed.
And I think we can have such feature in thermal core.
But again I don't think we should represent it as an trip point. 

thanks,
rui

quoted

	Arnd

Cheers,

Eduardo