On 21/02/2023 18:03, Pierre-Louis Bossart wrote:
+ card->components); + if (!card->components) + return -ENOMEM;
+ ret = snd_soc_dapm_new_controls(&card->dapm, + cs35l56_sof_widgets, ARRAY_SIZE(cs35l56_sof_widgets)); + if (ret) { + dev_err(card->dev, "Widgets add failed: %d\n", ret); + return ret; + }
+ ret = snd_soc_dapm_add_routes(&card->dapm, cs35l56_sof_map, count); + if (ret) { + dev_err(card->dev, "Map add %d failed: %d\n", count, ret); + return ret; + }
+ /* Enable one feedback TX per amp on different slots */ + for_each_rtd_codec_dais(rtd, i, codec_dai) { + ret = snd_soc_dai_set_tdm_slot(codec_dai, 0x3, 1 << i, 4, 16);
TDM slots? Not getting how this would work with SoundWire?
Strictly speaking Soundwire is TDM (the frame time is divided up into slots for each sample...).
The problem is if you have N amps on the dailink all feeding back audio on the same bus. Their DP slots are all programmed to the same positions in the frame, same as for the playback. So you have 4 amps all trying to send 6 audio channels in the same positions in the frame and you'll just get a ton of bus clash interrupts.
So we use the set_tdm_slot() like we do with I2S TDM to set which slots are active for each amp.
I can't see that there's any obvious "generic" way that the manager code can automatically figure out how many channels to enable on each amp and what order to map them, so we do it here. Just as with I2S TDM - you have many slots and many codecs but the machine driver has to tell it how to map those.
IIRC Bard did the same thing recently, and the order of the feedback channels is really defined by the order in which the peripheral devices are added in the dailink. See https://github.com/thesofproject/linux/pull/4108
I don't really see how it is determining how to split the channels. I see there is a discussion about that but I didn't see any explanation that I could understand.
Also that only seems to have been done for IPCv4. What if my system uses IPCv3 ?
Anyway, I'll drop the machine driver patches for now until the IPC code is stable and consistent behavior for v3 and v4.
There's also another open related to the number of channels, we need to patch what the CPU DAI can handle, see https://github.com/thesofproject/linux/pull/4136 or https://github.com/thesofproject/linux/pull/4134