[PATCH] soundwire: intel: move to auxiliary bus

Pierre-Louis Bossart pierre-louis.bossart at linux.intel.com
Wed Mar 24 16:03:42 CET 2021



On 3/24/21 5:50 AM, Vinod Koul wrote:
> On 23-03-21, 12:29, Pierre-Louis Bossart wrote:
>> Thanks Greg and Vinod for the reviews
>>
>>>>> -static int intel_master_probe(struct platform_device *pdev)
>>>>> +static int intel_link_probe(struct auxiliary_device *auxdev, const struct auxiliary_device_id *id)
>>>>>    {
>>>>> -	struct device *dev = &pdev->dev;
>>>>> +	struct device *dev = &auxdev->dev;
>>>>> +	struct sdw_intel_link_dev *ldev = auxiliary_dev_to_sdw_intel_link_dev(auxdev);
>>>>
>>>> Do we need another abstractions for resources here, why not aux dev
>>>> creation set the resources required and we skip this step...
>>
>> Not sure what resources you are referring to?
> 
> Resources in the sdw_intel_link_dev which are sdw_intel_link_res. They
> should be resources for auxdev and if you do that lets you get rid of
> this abstraction.

Sorry Vinod, I am not following your line of thought. We must be talking 
of different things or having a different understanding of what the 
auxiliary device is.

The auxiliary device is deliberately minimal by design and does not 
contain domain-specific information/resources/pointers/pdata as the 
platform_device does. You extend it by defining a larger structure that 
contains an auxiliary device and whatever domain-specific 
fields/structures/domains are needed, then use container_of to access it.

It's not just Intel doing this, the first example from Mellanox uses the 
same pattern, albeit with a single pointer instead of the structure we used.

see e.g. 
https://elixir.bootlin.com/linux/latest/source/include/linux/mlx5/driver.h#L545

So I am not sure what you mean by 'rid of this abstraction' when this 
abstraction is pretty much the way things were designed?

Maybe an example of what sort of structure you had in mind would help?


>> this is just a container_of() and the documented way of extending the auxbus
>> (see https://www.kernel.org/doc/html/latest/driver-api/auxiliary_bus.html#example-usage)
>>
>>
>> struct sdw_intel_link_dev {
>> 	struct auxiliary_device auxdev;
>> 	struct sdw_intel_link_res link_res;
>> };
>>
>> #define auxiliary_dev_to_sdw_intel_link_dev(auxiliary_dev) \
>> 	container_of(auxiliary_dev, struct sdw_intel_link_dev, auxdev)
>>
>>>>>    	struct sdw_intel *sdw;
>>>>>    	struct sdw_cdns *cdns;
>>>>>    	struct sdw_bus *bus;
>>>>> @@ -1346,14 +1347,14 @@ static int intel_master_probe(struct platform_device *pdev)
>>>>>    	cdns = &sdw->cdns;
>>>>>    	bus = &cdns->bus;
>>>>> -	sdw->instance = pdev->id;
>>>>> -	sdw->link_res = dev_get_platdata(dev);
>>>>> +	sdw->instance = auxdev->id;
>>>>
>>>> so auxdev has id and still we pass id as argument :( Not sure if folks
>>>> can fix this now
>>>
>>> That's odd, yeah, it should be fixed.
>>
>> I think we are talking about different things?
>>
>> this is defined in mod_devicetable.h:
>>
>> struct auxiliary_device_id {
>> 	char name[AUXILIARY_NAME_SIZE];
>> 	kernel_ulong_t driver_data;
>> };
>>
>> and used for the driver probe:
>>
>> 	ret = auxdrv->probe(auxdev, auxiliary_match_id(auxdrv->id_table, auxdev));
>>
>> but there is a separate id:
>>
>> struct auxiliary_device {
>> 	struct device dev;
>> 	const char *name;
>> 	u32 id;
>> };
>>
>> which is set during the device initialization in intel_init.c
>>
>> 	/* we don't use an IDA since we already have a link ID */
>> 	auxdev->id = link_id;
>>
>> In the auxiliary bus design, the parent has to take care of managing this
>> id, be it with an IDA or as we do here with a physical link ID that is
>> unique.
> 
> Aha, maybe both of them should not be 'id' to avoid this confusion!

the function definition follows the expected prototype

struct auxiliary_driver {
         int (*probe)(struct auxiliary_device *,
                      const struct auxiliary_device_id *id);

we can rename the argument to e.g. dev_id if that helps. Suggestions 
welcome.

> That also reminds me that we have duplicate info:
> 
> +       sdw->instance = auxdev->id;
> +       bus->link_id = auxdev->id;
> 
> drop the local driver instance and use bus->link_id please

if you are referring to sdw->instance, it could probably be removed, but 
that would need to be a separate cleanup changing cadence_master.c as 
well. this patch only changes pdev->id with auxdev->id and provides only 
the transition from platform device to auxiliary device.



More information about the Alsa-devel mailing list