Re: [PATCH v10 2/2] ASoC: fsl: Add S/PDIF machine driver

[Stephen Warren <swarren@wwwdotorg.org>]

On 08/21/2013 12:54 PM, Tomasz Figa wrote:
> On Wednesday 21 of August 2013 12:30:59 Stephen Warren wrote:
>> On 08/20/2013 09:13 PM, Nicolin Chen wrote:
>>> This patch implements a device-tree-only machine driver for Freescale
>>> i.MX series Soc. It works with spdif_transmitter/spdif_receiver and
>>> fsl_spdif.c drivers.
>>>
>>> diff --git
>>> a/Documentation/devicetree/bindings/sound/imx-audio-spdif.txt
>>> b/Documentation/devicetree/bindings/sound/imx-audio-spdif.txt
>>>
>>> +Optional properties:
>>> +
>>> +  - spdif-transmitter : The phandle of the spdif-transmitter codec
>>> +
>>> +  - spdif-receiver : The phandle of the spdif-receiver codec
>>> +
>>> +* Note: At least one of these two properties should be set in the DT
>>> binding.
>> I still don't think those two properties are correct.
>>
>> Exactly what node will those phandles point at?
> 
> Imagine following setup:
> 
>   ________              ________________
>  |        |     RX     | Microphone DSP |   Analog mic input
>  | S/PDIF | <--------< |________________| <-------------------
>  |        |             ________________
>  |  DAI   | >--------> | Amplifier      | >-------------------
>  |________|     TX     |________________|   Speakers output
> 
> As you see in the diagram, the S/PDIF interface of the SoC can be 
> connected to some external devices that can perform sound processing or 
> simply handle the physical layer.
> 
> I'd say that normally both RX and TX lines would be connected to a single 
> codec chip that has multiple blocks inside, like sound processing, 
> amplifier, mixer, etc., but nothing stops you from making a crazy setup, 
> when RX and TX lines are connected to different chips.

That's much rarer with S/PDIF than I2S though right? Usually I'd expect
the S/PDIF controller to simply be routed out to a jack/connector on the
board, or perhaps to an internal HDMI encoder.

But the point of my question was more that the binding should fully
describe the type of object/node that the phandle should reference. The
type of the node would then imply what kind of operations could be
performed on that other node's driver by this node's driver. If the set
of operations is undefined, that's bad. If the set of operations is
NULL, there's probably no need to reference the node.

>> There definitely should not be a DT node for any "dummy CODEC",
>> irrespective of whether this binding calls the other node a "CODEC" or a
>> "dummy CODEC".
> 
> I agree. Instead if no chip connected to particular line is specified in 
> device tree, it's responsibility of Linux sound core to handle this 
> properly by adding a dummy codec or whatever.
> 
>> If these properties are to contain phandles, it would be acceptable for
>> the referenced node to be:
>>
>> * A node representing the physical connector/jack on the board.
>>
>> * A node representing some other IP block on the board, such as an HDMI
>> encoder/display-controller
>>
>> I think those options are unlikely in general
> 
> Why? You usually codec SoC DAIs to some external chips.

It's unlikely the phandle would reference a connector/jack since we
(thus far at least) haven't created DT nodes for them. If we do put
jacks/connectors into DT, then referencing them is fine.

In the "other IP block" case, it may be worth referencing the device,
although as I mentioned above, we need to describe exactly what is
expected from that other device interface-wise.