Re: [PATCH] iio: adc: add exynos5 adc driver under iio framwork

[Lars-Peter Clausen <lars@metafoo.de>]

On 01/24/2013 08:15 PM, Tomasz Figa wrote:
> Hi,
> 
> On Thursday 24 of January 2013 19:19:57 Lars-Peter Clausen wrote:
>> On 01/24/2013 05:12 PM, Doug Anderson wrote:
>>> Lars,
>>>
>>> Thank you for your comments / thoughts...
>>
>> Hi,
>>
>>> On Thu, Jan 24, 2013 at 1:54 AM, Lars-Peter Clausen <lars@metafoo.de> 
> wrote:
>>>> adc: adc@12D10000 {
>>>>
>>>>         #io-channel-cells = <1>;
>>>>         io-channel-output-names = "adc1", "adc2", ...;
>>>>         
>>>>         ncp15wb473@0 {
>>>>         
>>>>                 compatible = "ntc,ncp15wb473";
>>>>                 ...
>>>>                 io-channels = <&adc 0>; // First ADC channel
>>>
>>> I'm not an expert, but I think the typical way is:
>>> * No need to include a handle to &adc.  It's logically our parent.  In
>>> a similar way i2c devices don't specify their parent bus--they are
>>> just listed under it.
>>> * The "0" should be specified with reg = <0>
>>
>> The relationship between the IIO sensor device and the consumer device
>> is not always a parent child relationship. In this case it makes sense
>> to have the ADC as the parent for the thermistors. But for other cases
>> this may not be true. E.g. take a touchscreen or power monitoring
>> platform device which uses the IIO device to do measurements.
> 
> The policy is to use children with reg property only inside a node 
> representing a bus controller through which the child device is being 
> accessed (like I2C, SPI).
> 
> I would see IIO bindings similar to what we have with GPIOs, interrupts or 
> regulators, so io-channels = <&iio-controller channel> seems fine (or 
> rather iio-channels) with the node under appropriate parent.

IIO is a very Linux specific term, the device tree bindings should be as OS
agnostic as possible, so io-channels is probably the better term.


> 
>>> To implement this I'd imagine that we'll need a new API call, right?
>>> In this case the thermistor driver won't know the name of the channel.
>>>
>>>  It can find the ADC (the struct device and probably other things) and
>>>
>>> knows a channel index.  Am I understanding properly?
>>
>> This can be done by adding a new api call, but it would be best if both
>> dt and non-dt based consumers can use the same function. I outlined one
>> possible solution how this can be done in the previous mail to Naveen.
> 
> In case of the solution I mentioned, implementation would be almost 
> identical to what is done with GPIOs (see drivers/gpio/gpiolib-of.c).

Although similar to the GPIO bindings, the clk bindings are in my opinion an
even better example.

- Lars