Re: [alsa-devel] [PATCH v2 3/5] ASoC: core: add support to snd_soc_dai_get_sdw_stream()

From: Srinivas Kandagatla
Date: Thu Oct 10 2019 - 04:50:35 EST

On 09/10/2019 19:53, Pierre-Louis Bossart wrote:

On 10/9/19 11:01 AM, Srinivas Kandagatla wrote:

On 09/10/2019 15:29, Pierre-Louis Bossart wrote:

On 10/9/19 3:32 AM, Srinivas Kandagatla wrote:
Hi Pierre,

On 14/08/2019 15:09, Pierre-Louis Bossart wrote:

On 8/13/19 11:11 PM, Vinod Koul wrote:
On 13-08-19, 20:58, Mark Brown wrote:
On Tue, Aug 13, 2019 at 02:38:53PM -0500, Pierre-Louis Bossart wrote:

Indeed. I don't have a full understanding of that part to be honest, nor why
we need something SoundWire-specific. We already abused the set_tdm_slot API
to store an HDaudio stream, now we have a rather confusing stream
information for SoundWire and I have about 3 other 'stream' contexts in
SOF... I am still doing basic cleanups but this has been on my radar for a

There is something to be said for not abusing the TDM slot API if it can
make things clearer by using bus-idiomatic mechanisms, but it does mean
everything needs to know about each individual bus :/ .

Here ASoC doesn't need to know about sdw bus. As Srini explained, this
helps in the case for him to get the stream context and set the stream
context from the machine driver.

Nothing else is expected to be done from this API. We already do a set
using snd_soc_dai_set_sdw_stream(). Here we add the snd_soc_dai_get_sdw_stream() to query

I didn't see a call to snd_soc_dai_set_sdw_stream() in Srini's code?

There is a snd_soc_dai_get_sdw_stream() to get stream context and we add slave streams(amplifier in this case) to that context using sdw_stream_add_slave() in machine driver[1].

Without this helper there is no way to link slave streams to stream context in non dai based setup like smart speaker amplifiers.

Currently this driver is blocked on this patch, If you think there are other ways to do this, am happy to try them out.

So to be clear, you are *not* using snd_soc_dai_set_sdw_stream?
Yes, am not using snd_soc_dai_set_sdw_stream().

It's been a while since this thread started, and I still don't quite get the concepts or logic.

First, I don't understand what the problem with "aux" devices is. All the SoundWire stuff is based on the concept of DAI, so I guess I am

That is the actual problem! Some aux devices does not have dais.

missing why introducing the "aux" device changes anything.

Problem is that aux devices(amplifier) are dai less but connected via SoundWire. So question is how do we attach those SoundWire streams to SoundWire master stream?

My Idea was to get handle to the SoundWire stream from controller dai and adding these aux SoundWire slave devices as slave to them in machine driver. This was much less intrusive than other solutions.

Is there a better way to associate a dai-less codecs to SoundWire master stream?

Next, a 'stream' connects multiple DAIs to transmit information from sources to sinks. A DAI in itself is created without having any notion of which stream it will be associated with. This can only be done with machine level information.

If you query a DAI to get a stream context, then how is this stream context allocated in the first place? It looks like a horse and cart problem. Or you are assuming that a specific DAI provides the context, and that all other DAIs do not expose this .get_sdw_stream()? What if more that 1 DAI provide a stream context?

In this particular board setup. Soundwire master is allocating the stream runtime for each of it dais, and this runtime is used in machine driver to attach Auxdev Soundwire slaves.

Other setups where the codec has dai should work as expected!

And last, I am even more lost since w/ the existing codec drivers we have, sdw_stream_add_slave() is called from the dai .hw_params callback, once you have information on formats/rates, etc. using sdw_stream_add_slave() from a machine driver looks like an even bigger stretch. I really thought the machine driver would only propagate the notion of stream to all DAIs that are part of the same dailink.

I am not trying to block the Qualcomm implementation, just would like to make sure the assumptions are clear and that we are not using the same API in completely different ways.

Am open for any suggestions on dealing with dai less setups like what we have on our board.