On 2/20/19 10:26 AM, Mark Brown wrote:

On Wed, Feb 20, 2019 at 08:35:16AM -0600, Pierre-Louis Bossart wrote:

...

I see your point and agree the current direction isn't scalable, but I am not sure about the definition of 'first class device'. Did you mean 'component' at the same level as the 'platforms' and 'codecs', with all the DSP stuff nicely packaged and maintained in a self-contained way? Or a subcomponent that might be part of those platforms and codecs?

As far as I can see for the Intel systems it's looking like a first class component - there will be some systems where the DSP is embedded in something (like the big CODECs) but even there it's easier as it's contained within the device rather than affecting multiple devices which is the big issue here.

Sorry Mark, I am struggling with the 'first class' idea.

Allow me to go back to the practical issue we faced here, and maybe/hopefully this will help us converge on a consensus.

The only reason for extending the struct snd_soc_pcm_runtime was to enable the use of a 'context' in PCM operations, see e.g in patch 5/14 the repeated pattern to get an 'SOF PCM' context:

struct snd_sof_pcm *spcm = rtd->private;

Managing a context for PCM ops is typically done in other drivers by using the substream->private_data aka snd_pcm_substream_chip(substream).

In ASoC/DPCM, unfortunately this substream->private_data field was reclaimed to store a pointer to a 'struct snd_soc_pcm_runtime', see e.g. in soc-pcm.c:

pcm->private_data = rtd;

if (playback) pcm->streams[SNDRV_PCM_STREAM_PLAYBACK].substream->private_data = rtd;         if (capture) pcm->streams[SNDRV_PCM_STREAM_CAPTURE].substream->private_data = rtd;

So ironically in ASoC we don't have a means to use a private_data field for the substreams since it's used for a 'standard' mechanism.

The extension suggested in this patch 1/14 was really trying to come back to a practical means to store a context in a substream, and make the PCM operations manageable. I totally agree that extending the PCM runtime might not recommended or safe, but at this point I don't have any slack in the existing data structures to add a context in a substream.

Does this help clarify the problem statement?