Skip to contentSkip to conversation

ASoC: bdw-rt5677: enable runtime channel merge

PATCH In review Brent Lu 2 participants 5 messages Oct 2, 2019 – Nov 5, 2019

Subsystems (3)
lore
Share & tools
lore

Brent Lu

author
#0 lore raw

PATCH 1/1 ASoC: bdw-rt5677: enable runtime channel merge

In the DAI link "Capture PCM", the FE DAI "Capture Pin" supports 4-channel capture but the BE DAI supports only 2-channel capture. To fix the channel mismatch, we need to enable the runtime channel merge for this DAI link.

---
 sound/soc/intel/boards/bdw-rt5677.c | 1 +
 1 file changed, 1 insertion(+)
sound/soc/intel/boards/bdw-rt5677.c +1 -0
index 4a4d335..8d07038 100644
@@ -273,6 +273,7 @@ static struct snd_soc_dai_link bdw_rt5677_dais[] = {
 		},
 		.dpcm_capture = 1,
 		.dpcm_playback = 1,
+		.dpcm_merged_chan = 1,
 		SND_SOC_DAILINK_REG(fe, dummy, platform),
 	},
 
Signature · 1 line
2.7.4
author
#1 lore raw
Quoted 6 lines

Subject: [PATCH] ASoC: bdw-rt5677: enable runtime channel merge

In the DAI link "Capture PCM", the FE DAI "Capture Pin" supports 4-channel capture but the BE DAI supports only 2-channel capture. To fix the channel mismatch, we need to enable the runtime channel merge for this DAI link.

Hi Pierre,

This patch is for the same issue discussed in the following thread:
https://patchwork.kernel.org/patch/11134167/

We enable the runtime channel merge for the DMIC DAI instead of adding a machine driver constraint. It's working good on chrome's 3.14 branch (which requires some backport for the runtime channel merge feature). Please let me know if this implementation is correct for the FE/BE mismatch problem. Thanks.

Regards,
Brent

#2 lore raw
Quoted 8 lines
Quoted 5 lines

In the DAI link "Capture PCM", the FE DAI "Capture Pin" supports 4-channel capture but the BE DAI supports only 2-channel capture. To fix the channel mismatch, we need to enable the runtime channel merge for this DAI link.

Hi Pierre,

Quoted 7 lines

This patch is for the same issue discussed in the following thread:
https://patchwork.kernel.org/patch/11134167/

We enable the runtime channel merge for the DMIC DAI instead of adding a machine driver constraint. It's working good on chrome's 3.14 branch (which requires some backport for the runtime channel merge feature). Please let me know if this implementation is correct for the FE/BE mismatch problem.

Sorry, I don't fully understand your points, and it's the first time I see anyone use this .dpcm_merged_chan field for an Intel platform.

If I look at the code I see that the core would limit the number of channels to two. But that code needs the CPU_DAI to use 2 channels, which I don't see. So is this patch self-contained or do we need an additional constraint on the FE?

Thanks
-Pierre

author
#3 lore raw
Quoted 20 lines
Quoted 3 lines
Quoted 3 lines

In the DAI link "Capture PCM", the FE DAI "Capture Pin" supports 4-channel capture but the BE DAI supports only 2-channel capture. To fix the channel mismatch, we need to enable the runtime channel merge

for this DAI link.

Quoted 11 lines
Quoted 1 line

Hi Pierre,

This patch is for the same issue discussed in the following thread:
https://patchwork.kernel.org/patch/11134167/

We enable the runtime channel merge for the DMIC DAI instead of adding a machine driver constraint. It's working good on chrome's 3.14 branch (which requires some backport for the runtime channel merge feature). Please let me know if this implementation is correct for the FE/BE mismatch

problem.

Sorry, I don't fully understand your points, and it's the first time I see anyone use this .dpcm_merged_chan field for an Intel platform.

Quoted 7 lines

If I look at the code I see that the core would limit the number of channels to two. But that code needs the CPU_DAI to use 2 channels, which I don't see. So is this patch self-contained or do we need an additional constraint on the FE?

Thanks
-Pierre

Hi Pierre,

We don't need constraint on FE because dpcm_runtime_merge_chan() modifies the channel number of snd_pcm_hardware structure directly. The structure will be used to initialize the snd_pcm_hw_constraints structure later in the snd_pcm_hw_constraints_complete() function. Since the channel number is already modified, we don't need a constraint to install an extra rule for it.

The result of using dpcm_merged_chan flag and machine driver constraint should be the same when user space programs calling HW_REFINE ioctl call but I think the flag is more elegant for machine driver code.

Regards,
Brent

#4 lore raw
Quoted 36 lines · Brent
Quoted 20 lines
Quoted 3 lines
Quoted 3 lines

In the DAI link "Capture PCM", the FE DAI "Capture Pin" supports 4-channel capture but the BE DAI supports only 2-channel capture. To fix the channel mismatch, we need to enable the runtime channel merge

for this DAI link.

Quoted 11 lines
Quoted 1 line

Hi Pierre,

This patch is for the same issue discussed in the following thread:
https://patchwork.kernel.org/patch/11134167/

We enable the runtime channel merge for the DMIC DAI instead of adding a machine driver constraint. It's working good on chrome's 3.14 branch (which requires some backport for the runtime channel merge feature). Please let me know if this implementation is correct for the FE/BE mismatch

problem.

Sorry, I don't fully understand your points, and it's the first time I see anyone use this .dpcm_merged_chan field for an Intel platform.

Quoted 7 lines

If I look at the code I see that the core would limit the number of channels to two. But that code needs the CPU_DAI to use 2 channels, which I don't see. So is this patch self-contained or do we need an additional constraint on the FE?

Thanks
-Pierre

Hi Pierre,

We don't need constraint on FE because dpcm_runtime_merge_chan() modifies the channel number of snd_pcm_hardware structure directly. The structure will be used to initialize the snd_pcm_hw_constraints structure later in the snd_pcm_hw_constraints_complete() function. Since the channel number is already modified, we don't need a constraint to install an extra rule for it.

Quoted 3 lines

The result of using dpcm_merged_chan flag and machine driver constraint should be the same when user space programs calling HW_REFINE ioctl call but I think the flag is more elegant for machine driver code.

I could use the opposite argument: by using a capability that no one else uses, you make the solution more obscure and less intuitive. All other machine drivers use constraints, so if the two solutions are equivalent let's follow the more common solution.