ASoC: qcom: qdsp6: clear stale channel_map entries
Subsystems (4)
PATCH 1/1 ASoC: qcom: qdsp6: clear stale channel_map entries
Since commit c663014e452d ("ASoC: qcom: audioreach: compute active channel maps from channel_map"), the codec DMA active_channels_mask is built from the non-zero entries of the channel_map instead of from num_channels.
audioreach_set_default_channel_mapping() only writes the first num_channels entries, and q6dma_set_channel_map() only the first tx_num/rx_num entries. The per-DAI module_config in q6apm-lpass-dais is persistent, so entries written by an earlier, wider configuration survive a later narrower one. E.g. after a 4 channel setup, a 2 channel capture on VA_CODEC_DMA_TX_0 keeps the map 01 02 04 05 and sends active_channels_mask = 0xf with num_channels = 2. The DSP rejects the graph start:
qcom-apm gprsvc:service:2:1: DSP returned error[1001002] 1 q6apm-lpass-dais ...: Failed to start APM port 110 MultiMedia4 Capture: ASoC error (-22): at dpcm_be_dai_trigger()
which breaks the built-in DMIC capture on the Lenovo Yoga Slim 7x (X1E80100).
Clear the whole channel_map before filling it in both places.
Note: this is my very first contribution to the linux kernel so I may be new to some of the processes. I apologize in advance.
---
Tested on a Lenovo Yoga Slim 7x (X1E80100): built-in DMIC capture on VA_CODEC_DMA_TX_0 (MultiMedia4 Capture) failed with the DSP error above; with this patch repeated arecord runs work. Based on broonie/sound.git for-next (d34a818a331f).
sound/soc/qcom/qdsp6/audioreach.c | 6 ++++++ sound/soc/qcom/qdsp6/q6apm-lpass-dais.c | 2 ++ 2 files changed, 8 insertions(+)
sound/soc/qcom/qdsp6/audioreach.c +6 -0
@@ -308,6 +308,12 @@ EXPORT_SYMBOL_GPL(audioreach_alloc_apm_cmd_pkt);void audioreach_set_default_channel_mapping (u8 *ch_map,int num_channels) { +/* +* The active channel mask is derived from the non-zero entries, so +* stale entries from a previous, wider configuration must not survive. +*/ +memset (ch_map,0 ,AR_PCM_MAX_NUM_CHANNEL ); +if (num_channels ==1 ) { ch_map[0 ] =PCM_CHANNEL_FL ; }else if (num_channels ==2 ) {
sound/soc/qcom/qdsp6/q6apm-lpass-dais.c +2 -0
@@ -99,6 +99,7 @@ static int q6dma_set_channel_map(struct snd_soc_dai *dai, tx_num);return -EINVAL ; } +memset (cfg->channel_map,0 ,sizeof (cfg->channel_map));for (i =0 ; i < tx_num; i++) cfg->channel_map[i] = tx_ch_mask[i];
@@ -123,6 +124,7 @@ static int q6dma_set_channel_map(struct snd_soc_dai *dai, rx_num);return -EINVAL ; } +memset (cfg->channel_map,0 ,sizeof (cfg->channel_map));for (i =0 ; i < rx_num; i++) cfg->channel_map[i] = rx_ch_mask[i];
base-commit: d34a818a331f7e270fa9dc3db3d96f6a0d4fa7ab
Signature · 1 line
2.55.0
Quoted 18 lines · Fabian Renn-Giles
Since commit
c663014e452d("ASoC: qcom: audioreach: compute active channel maps fromchannel_map"), the codec DMAactive_channels_maskis built from the non-zero entries of thechannel_mapinstead of fromnum_channels.
audioreach_set_default_channel_mapping()only writes the firstnum_channelsentries, andq6dma_set_channel_map()only the first tx_num/rx_num entries. The per-DAImodule_configin q6apm-lpass-dais is persistent, so entries written by an earlier, wider configuration survive a later narrower one. E.g. after a 4 channel setup, a 2 channel capture onVA_CODEC_DMA_TX_0keeps the map 01 02 04 05 and sendsactive_channels_mask=0xfwithnum_channels= 2. The DSP rejects the graph start:qcom-apm gprsvc:service:2:1: DSP returned error[1001002] 1 q6apm-lpass-dais ...: Failed to start APM port 110 MultiMedia4 Capture: ASoC error (-22): at
dpcm_be_dai_trigger()
Quoted 7 lines
which breaks the built-in DMIC capture on the Lenovo Yoga Slim 7x (X1E80100).
Clear the whole
channel_mapbefore filling it in both places.Note: this is my very first contribution to the linux kernel so I may be new to some of the processes. I apologize in advance.
Thanks for sending the patch and Welcome.
You can move this comment after --- as it does not make sense to be in the commit log.
Quoted 19 lines
Fixes c663014e452d ("ASoC: qcom: audioreach: compute active channel maps from channel_map")Signed-off-by Fabian Renn-Giles <[email protected]>Assisted-by Claude:claude-sonnet-5-5---
Tested on a Lenovo Yoga Slim 7x (X1E80100): built-in DMIC capture onVA_CODEC_DMA_TX_0(MultiMedia4 Capture) failed with the DSP error above; with this patch repeated arecord runs work. Based on broonie/sound.git for-next (d34a818a331f).sound/soc/qcom/qdsp6/audioreach.c | 6 ++++++ sound/soc/qcom/qdsp6/q6apm-lpass-dais.c | 2 ++ 2 files changed, 8 insertions(+)sound/soc/qcom/qdsp6/audioreach.c +0
-0
sound/soc/qcom/qdsp6/audioreach.c EXPORT_SYMBOL_GPL(audioreach_alloc_apm_cmd_pkt); · line 309
Quoted 2 lines
sound/soc/qcom/qdsp6/audioreach.c +0
-0@@ -308,6 +308,12 @@ EXPORT_SYMBOL_GPL(audioreach_alloc_apm_cmd_pkt);void audioreach_set_default_channel_mapping (u8 *ch_map,int num_channels)
there is patch on the list to remove this function, could you please take a look at
https://lore.kernel.org/all/[email protected]/
q6dsp_map_channels() already does the memset, pl recheck with that patch.
If your patch might be still relevant, please send patch against it.
--srini
Quoted 32 lines
sound/soc/qcom/qdsp6/audioreach.c +6
-0{ +/* +* The active channel mask is derived from the non-zero entries, so +* stale entries from a previous, wider configuration must not survive. +*/ +memset (ch_map,0 ,AR_PCM_MAX_NUM_CHANNEL ); +if (num_channels ==1 ) { ch_map[0 ] =PCM_CHANNEL_FL ; }else if (num_channels ==2 ) {sound/soc/qcom/qdsp6/q6apm-lpass-dais.c +2
-0@@ -99,6 +99,7 @@ static int q6dma_set_channel_map(struct snd_soc_dai *dai, tx_num);return -EINVAL ; } +memset (cfg->channel_map,0 ,sizeof (cfg->channel_map));for (i =0 ; i < tx_num; i++) cfg->channel_map[i] = tx_ch_mask[i];@@ -123,6 +124,7 @@ static int q6dma_set_channel_map(struct snd_soc_dai *dai, rx_num);return -EINVAL ; } +memset (cfg->channel_map,0 ,sizeof (cfg->channel_map));for (i =0 ; i < rx_num; i++) cfg->channel_map[i] = rx_ch_mask[i];base-commit:
d34a818a331f7e270fa9dc3db3d96f6a0d4fa7ab