[PATCH v2] mfd: DT bindings for the palmas family MFD

Stephen Warren swarren at wwwdotorg.org
Fri Jun 7 01:53:41 EST 2013


On 06/05/2013 09:34 PM, J, KEERTHY wrote:
> Hi Stephen,
> 
> Thanks for the quick review.
> 
> Stephen Warren wrote at Wednesday, June 05, 2013 10:44 PM:
>> On 06/04/2013 02:41 AM, J Keerthy wrote:
>>> From: Graeme Gregory <gg at slimlogic.co.uk>
>>>
>>> Add the various binding files for the palmas family of chips. There is
>>> a top level MFD binding then a seperate binding for regulators IP
>> blocks on chips.
...
>> Oh, one question though: How does the regulator driver determine the
>> register address of the regulator sub-device within the overall PMIC?
>> Presumably if these are pluggable independent modules, that could
>> change depending on which overall chip the PMIC device is plugged into.
>> don't you need a reg property to specify that?
> 
> The variants have identical register addresses. These are not pluggable
> Independent modules. All the variants come with all the regulators
> Listed above in general. The driver today has a statically defined
> Array of all the above mentioned regulators with their addresses.
>  
> drivers/regulator/palmas-regulator.c
> 
> Line 38.

I meant the I2C address used to communicate with the regulator registers
really, and I suppose the base address of the regulator register block.

In the driver, I see this is handled by the top-level Palmas driver
creating a regmap object which the regulator driver used. This keeps the
regulator driver completely unaware of these issues, only the top-level
chip driver cares about this, which is fine.

While that justification is in terms of OS-specific code, the basic
argument can be applied to the HW itself (the top-level chip implies the
I2C address and any register offset), so this really is a HW-driven
argument, so I guess it's fine not having a reg property in the
top-level regulator node.

One question though: I wonder why if the HW IP blocks aren't completely
independent modules that can be mixed/matched together to form new
chips, there's even a need for a separate regulator node with its own
compatible value. Still, I suppose it's a valid way to construct the DT
either way, so it's fine.


More information about the devicetree-discuss mailing list