Re: [RFC 10/11] iio: Add OF support

[Lars-Peter Clausen <lars-Qo5EllUWu/uELgA04lAiVw@public.gmane.org>]

On 02/03/2013 12:39 PM, Jonathan Cameron wrote:
> On 02/02/2013 04:10 PM, Guenter Roeck wrote:
>> On Sat, Feb 02, 2013 at 10:29:02AM +0000, Jonathan Cameron wrote:
>>> On 01/31/2013 09:43 PM, Guenter Roeck wrote:
>>>> Provide bindings, new API access functions, and parse OF data
>>>> during initialization.
>>>>
>>> Firstly thanks for working on this Guenter, it's been a big hole
>>> for a while largely because non of our largest developers were
>>> actually using development platforms with device tree support.
>>>
>>> Given my knowledge of device tree is based on the odd article
>>> and looking at similar sets of bindings this morning, my comments
>>> are likely to be somewhat superficial and uninformed ;)
>>>
>>> Mostly on this one I'll take a back seat and let those who
>>> know this stuff better come to a consensus.
>>>
>>> Jonathan
>>>
>>>> Signed-off-by: Guenter Roeck <linux-0h96xk9xTtrk1uMJSBkQmQ@public.gmane.org>
>>>> ---
>>>>  .../devicetree/bindings/iio/iio-bindings.txt       |   97 ++++++++
>>>>  drivers/iio/inkern.c                               |  241 ++++++++++++++++----
>>>>  include/linux/iio/consumer.h                       |    8 +
>>>>  3 files changed, 299 insertions(+), 47 deletions(-)
>>>>  create mode 100644 Documentation/devicetree/bindings/iio/iio-bindings.txt
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/iio/iio-bindings.txt b/Documentation/devicetree/bindings/iio/iio-bindings.txt
>>>> new file mode 100644
>>>> index 0000000..0f51c95
>>>> --- /dev/null
>>>> +++ b/Documentation/devicetree/bindings/iio/iio-bindings.txt
>>>> @@ -0,0 +1,97 @@
>>>> +This binding is a work-in-progress, and are based on clock bindings and
>>>> +suggestions from Lars-Peter Clausen [1].
>>>> +
>>>> +Sources of IIO channels can be represented by any node in the device
>>>> +tree.  Those nodes are designated as IIO providers.  IIO consumer
>>>> +nodes use a phandle and IIO specifier pair to connect IIO provider
>>>> +outputs to IIO inputs.  Similar to the gpio specifiers, an IIO
>>>> +specifier is an array of one more more cells identifying the IIO
>>>> +output on a device.  The length of an IIO specifier is defined by the
>>>> +value of a #io-channel-cells property in the clock provider node.
>>>> +
>>>> +[1] http://marc.info/?l=linux-iio&m=135902119507483&w=2
>>>> +
>>>> +==IIO providers==
>>>> +
>>>> +Required properties:
>>>> +#io-channel-cells: Number of cells in an IIO specifier; Typically 0 for nodes
>>>> +		   with a single IIO output and 1 for nodes with multiple
>>>> +		   IIO outputs.
>>>> +
>>>> +Optional properties:
>>>> +io-channel-output-names:
>>>> +		    Recommended to be a list of strings of IIO output signal
>>>> +		    names indexed by the first cell in the IIO specifier.
>>>> +		    However, the meaning of io-channel-output-names is domain
>>>> +		    specific to the IIO provider, and is only provided to
>>>> +		    encourage using the same meaning for the majority of IIO
>>>> +		    providers.  This format may not work for IIO providers
>>>> +		    using a complex IIO specifier format.  In those cases it
>>>> +		    is recommended to omit this property and create a binding
>>>> +		    specific names property.
>>>> +
>>>> +		    IIO consumer nodes must never directly reference
>>>> +		    the provider's io-channel-output-names property.
>>>> +
>>>> +For example:
>>>> +
>>>> +    adc: adc@35 {
>>>> +	compatible = "maxim,max1139";
>>>> +	reg = <0x35>;
>>>> +        #io-channel-cells = <1>;
>>>> +        io-channel-output-names = "adc1", "adc2";
>>>> +    };
>>>> +
>>>> +- this node defines a device with two named IIO outputs, the first named
>>>> +  "adc1" and the second named "adc2".  Consumer nodes always reference
>>>> +  IIO channels by index. The names should reflect the IIO output signal
>>>> +  names for the device.
>>>> +
>>>> +==IIO consumers==
>>>> +
>>>> +Required properties:
>>>> +io-channels:	List of phandle and IIO specifier pairs, one pair
>>>> +		for each IIO input to the device.  Note: if the
>>>> +		IIO provider specifies '0' for #clock-cells, then
>>>> +		only the phandle portion of the pair will appear.
>>>> +
>>>> +Optional properties:
>>>> +io-channel-names:
>>>> +		List of IIO input name strings sorted in the same
>>>> +		order as the io-channels property.  Consumers drivers
>>>> +		will use io-channel-names to match IIO input names
>>>> +		with IIO specifiers.
>>>> +io-channel-ranges:
>>>> +		Empty property indicating that child nodes can inherit named
>>>> +		IIO channels from this node. Useful for bus nodes to provide
>>>> +		and IIO channel to their children.
>>>> +
>>>> +For example:
>>>> +
>>>> +    device {
>>>> +        io-channels = <&adc 1>, <&ref 0>;
>>>> +        io-channel-names = "vcc", "vdd";
>>>> +    };
>>>> +
>>>> +This represents a device with two IIO inputs, named "vcc" and "vdd".
>>>> +The vcc channel is connected to output 1 of the &adc device, and the
>>>> +vdd channel is connected to output 0 of the &ref device.
>>>> +
>>>> +==Example==
>>>> +
>>>> +	adc: max1139@35 {
>>>> +		compatible = "maxim,max1139";
>>>> +		reg = <0x35>;
>>>> +		#io-channel-cells = <1>;
>>>> +	};
>>>> +
>>>> +	...
>>>> +
>>>> +	iio_hwmon {
>>>> +		compatible = "iio-hwmon";
>>>> +		io-channels = <&adc 0>, <&adc 1>, <&adc 2>,
>>>> +			<&adc 3>, <&adc 4>, <&adc 5>,
>>>> +			<&adc 6>, <&adc 7>, <&adc 8>,
>>>> +			<&adc 9>, <&adc 10>, <&adc 11>;
>>>> +		io-channel-names = "vcc", "vdd", "vref", "1.2V";
>>> Having different numbers of channels and channel names seems
>>> unusual... Deliberate or you got bored making up channel names?
>>>
>>> Why use indexed values for <&adc 0> etc rather than the output
>>> channel names on adc?  For the iio_map stuff we initialy used
>>> indexes but got a lot of responses that it was a silly idea and
>>> naming was much more consistent and easy to follow.
>>>
>>> Is there a fundamental reason for it here?
>>>
>>> (note I don't mind either way as this seems more compact and cleaner
>>> in some ways)
>>>
>>
>> It follows the structure used by clocks, which uses the provided name(s) to
>> calculate an index into io-channels. This way, the provider does not have to
>> provide the mapping, the consumer does not have to know the io-channel index,
>> and the consumer code can call something like
>>
>> 	channel = iio_get_channel(dev, "vcc");
>>
>> In the above example, "vcc" will map to "<&adc, 0>", and "vref" to "<&adc, 2>".
>>
>> This works for both platform data and OF data (though platform data will
>> still need provider-based mapping, at least for now).
>>
>> This lets the code use a static name (eg "vcc"), and the mapping to the actual
>> provider happens through devicetree. Since the name is only used locally and
>> consumer driver specific, there is no need to define globally unique names.
>>
>> With this approach, the io channel map is not needed at all for the OF case.
>> I had used it in this version of the patch set, but got rid of it now.
>>
>> Actually, provider based mapping doesn't even work. If the consumer is
>> instantiated before the provider, the mapping doesn't exist yet, and the
>> call to iio_channel_get_all will fail. There is no way to prevent this,
>> as providers can come online at any time and there is no means to enforce that
>> all providers are already active by the time the consumers are instantiated.
>> Even if a mapping exists, there is no way to know if it is complete, if a
>> consumer is mapped to multiple providers.
>>
>> With the consumer based mapping, iio_channel_get_all 'knows' that not all
>> requested providers are available and can return -EPROBEDEFER in that case.
> Thanks. That makes sense.  At the moment iio_hwmon is the only case that
> does a 'get all'. Clearly things are easier when the driver is requesting a
> specific set and we can do the back off much more easily.
> 
>>
>> As a side effect, we can also use the names - if provided - as channel
>> labels in iio_hwmon.
>>
>> Note this will require the iio_get_channel API to change from taking the
>> consumer device name to taking the consumer device pointer as argument.
>> This will enable it to work for both OF and non-OF cases, should address Lars'
>> concerns about duplicate API functions, and synchronize the code to match how
>> the clock framework works.
> 
> Agreed, doing this gives us a cleaner syntax as well.  Note there are other
> users of that function in tree so be sure to get them all!
> 
>>
>> Thanks,
>> Guenter
> Thanks for the explanation.  What I was actually suggesting was something
> like:
> 
> adc: max1139@35 {
> 		compatible = "maxim,max1139";
> 		reg = <0x35>;
> 		#io-channel-cells = <1>;
> 		io-channel-output-names = "adc1", "adc2", "adc3"				
> 	};
> 
> iio_hwmon {
> 	compatible = "iio-hwmon";
> 	io-channels = <&adc "adc1">, <&adc "adc2">, <&adc "adc3">,
> 	io-channel-names = "vcc", "vdd", "vref";
> }
> 
> Having taken a look at the available syntax, those <> pairs have
> to be unsigned integers?  Hence the additional level of indirection?

Yea, I think mixing phandles and strings simply doesn't work, due how
devicetree stores things.

> 
> (sorry, I'm getting you to give me a tutorial on device tree syntax rather
> than the actual issue here!)
> 
> I guess it was desirable to keep the syntax relatively simple but that occasionally
> adds the requirement for a bit of indirection.
> 
> 
> 
> Jonathan