[Sound-open-firmware] What is the correct loopback configuration for ASoC machine driver DAI links?

[Thomas Preston]

Hi,
Thanks for getting back so quick

On 30/05/2019 17:29, Pierre-Louis Bossart wrote:
> On 5/30/19 10:14 AM, Thomas Preston wrote:
>> Hi,
>> I have a question about configuring a physical loopback on a custom
>> board for the Intel Skylake platform driver. I have also asked on
>> freenode/#alsa-dev but they directed me here.
> 
> When you write "Intel Skylake driver", you are referring to the code in sound/soc/intel/skylake? If yes, this is a different driver based on closed-source firmware and I am afraid very few people on this list can really help you.
> 

That's right.

> Also the reference to A2B and the reference to SSP5 make me think this is an automotive product based on ApolloLake? The board-level loopback to synchronize two SSPs was enabled in products I was exposed to (maybe two years ago), but this was quite tricky and you will most likely have to contact your Intel support channels to get the clocks and SSP setup right.
> 

Okay thanks. We have fixed DSP firmware and SSP blobs, so I think the hardware is setup correctly. I also think this because the loopback works when I split SSP5 CPU DAI in two (see 3 below). I just don't understand what's happening in ASoC when this happens.

>>
>> We have the following physical SSP layout:
>>
>> A2B axillary audio
>> + SSP0 Tx (connected to SSP5 Rx, for debug)
>> + SSP0 Rx
>>
>> Main speaker output + loopback from A2B
>> + SSP5 Tx
>> + SSP5 Rx (connected to SSP0 Tx)
>>
>> We have a board-level hardware loopback where the output signals and
>> clock (slave) on SSP0 are connected to the input signals and clock
>> (master) on SSP5.
>>
>> I'm trying to configure the DAI links in the machine driver so that they
>> link against SSP0 Pin CPU DAI correctly but I have three problematic
>> configurations:
>>
>> 1. Separate BE DAI links for SSP5 playback and capture, linking against
>> the same Skylake CPU DAI "SSP5 Tx". This crashes the kernel with a NULL
>> pointer dereference (I'm looking into this). Presumably this is an
>> invalid configuration because I have multiple DAI links linking against
>> one CPU DAI.  See machine1.c
>>
>> 2. One BE DAI link for SSP5 playback + capture, linking against the same
>> Skylake CPU DAI. This silences my SSP5 Tx main audio output (which
>> appears to be muted) when I try to arecord on the SSP5 Rx A2B debug
>> loopback. The recording is a perfect copy of SSP0 Tx so the loopback
>> part works. I can recover by restarting my whole audio stack.  See
>> machine2.c
> 
> This seems like the logical configuration, I don't think we've ever enabled 2 dailinks for capture and playback on the same SSP. it wouldn't make a lot of sense since the SSP format needs to be the same for the two directions (same bclk, fsync, number of slots and data format)
> 
> It's likely that you are still configured as SSP0 in master mode or have some sort of clocking issue. You will also need to enable the SSP5 clocks by default in your machine driver so that the SSP0 TX is always clocked.
> 

Okay that's the next thing to check. Although, if that's the case I'd expect it to break with the work around below.

Thanks

>>
>> 3. Separate BE DAI links for SSP playback and loopback, linking against
>> a separated SSP5 Skylake CPU DAI (SSP5 Pin with ssp5 Tx, SSP5 Pin
>> Loopback with ssp5 Rx). This configuration works but requires a
>> modification to the Skylake platform driver - which I'd rather avoid.
>> See machine3.c, skylake.diff
>>
>> Configuration 2 is closest to what I want, and configuration 3 is my
>> work-around for now.
>>
>> What is the correct way to link CPU DAIs for this kind of loopback? I
>> have attached example code.