Re: [PATCH 3/3] soundwire: bus_type: add sdw_master_device support

[Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>]

On 5/12/20 11:08 AM, Pierre-Louis Bossart wrote:
> 
> 
> On 5/12/20 10:59 AM, Vinod Koul wrote:
>> On 12-05-20, 09:36, Pierre-Louis Bossart wrote:
>>> On 5/11/20 10:30 PM, Vinod Koul wrote:
>>>> On 11-05-20, 14:00, Pierre-Louis Bossart wrote:
>>>>>>> +    md = &bus->md;
>>>>>>> +    md->dev.bus = &sdw_bus_type;
>>>>>>> +    md->dev.type = &sdw_master_type;
>>>>>>> +    md->dev.parent = parent;
>>>>>>> +    md->dev.of_node = parent->of_node;
>>>>>>> +    md->dev.fwnode = fwnode;
>>>>>>> +    md->dev.dma_mask = parent->dma_mask;
>>>>>>> +
>>>>>>> +    dev_set_name(&md->dev, "sdw-master-%d", bus->link_id);
>>>>>>
>>>>>> This give nice sdw-master-0. In DT this comes from reg property. I 
>>>>>> dont
>>>>>> seem to recall if the ACPI/Disco spec treats link_id as unique across
>>>>>> the system, can you check that please, if not we would need to update
>>>>>> this.
>>>>> Table 3 in the Disco for Soundwire 1.0 spec: "all LinkID values are 
>>>>> relative
>>>>> to the immediate parent Device."
>>>>>
>>>>> There isn't any known implementation with more than one controller.
>>>>
>>>> But then it can come in "future" right. So lets try to make it future
>>>> proof by not using the link_id (we can expose that as a sysfs if people
>>>> want to know). So a global unique id needs to allocated (hint: idr or
>>>> equivalent) and used as master_id
>>>
>>> Can you clarify if you are asking for a global ID for Intel/ACPI 
>>> platforms,
>>> or for DT as well? I can't figure out from the soundwire-controller.yaml
>>> definitions if there is already a notion of unique ID.
>>
>> If ACPI was unique, then I was planning to update the definition below
>> to include that. Given that it is not the case, let's make it agnostic to
>> underlying firmware.
> 
> I am not sure I understand how this would be done.
> 
> The call sequence is
> 
> sdw_bus_master_add(bus)
>      sdw_master_device_add(bus, parent, fw_node)
> 
> At the bus level, we don't have any information on which controller the 
> bus is related to.
> 
> We'd need to add an argument to sdw_bus_master_add() and have the 
> controller unique ID be allocated outside of the SoundWire core, hence 
> my question on whether the DT definition should not be extended.

And btw I don't think it makes sense to add a new definition for Intel. 
We already have a notion of HDaudio bus->idx that's set to zero since we 
don't have a case for multiple HDaudio controllers.

if we ever do have more than once controller, then we should rely on 
HDaudio bus->idx as the identifier and not create one specifically for 
SoundWire - which means as I mentioned above passing an argument and not 
defining a controller ID in the SoundWire core.