ASoC: mxs-saif: Fix channel swap for 24-bit format
Subsystems (3)
Fabio Estevam
reviewerPATCH 1/1 ASoC: mxs-saif: Fix channel swap for 24-bit format
From: Fabio Estevam <[email protected]>
Playing 24-bit format file leads to channel swap on mx28 and the reason is that the current driver performs one write/read to/from the SAIF_DATA register to trigger the transfer.
This approach works fine for S16_LE case because SAIF_DATA is a 32-bit register and thus is capable of storing the 16-bit left and right channels, but for the S24_LE case it can only store one channel, so in order to not lose the FIFO sync an extra read/write is needed.
--- sound/soc/mxs/mxs-saif.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-)
sound/soc/mxs/mxs-saif.c +10 -4
@@ -523,16 +523,22 @@ static int mxs_saif_trigger(struct snd_pcm_substream *substream, int cmd,if (substream->stream ==SNDRV_PCM_STREAM_PLAYBACK ) {/* -* write a data to saif data register to trigger -* the transfer +* write data to saif data register to trigger +* the transfer. +* For 24-bit format the 32-bit FIFO register stores +* only one channel, so we need to write twice. */ __raw_writel (0 , saif->base +SAIF_DATA ); +__raw_writel (0 , saif->base +SAIF_DATA ); }else {/* -* read a data from saif data register to trigger -* the receive +* read data from saif data register to trigger +* the receive. +* For 24-bit format the 32-bit FIFO register stores +* only one channel, so we need to read twice. */ __raw_readl (saif->base +SAIF_DATA ); +__raw_readl (saif->base +SAIF_DATA ); } master_saif->ongoing =1 ;
Signature · 1 line
1.7.9.5
Hi Fabio,
Quoted 25 lines · Fabio Estevam
From: Fabio Estevam <[email protected]>
Playing 24-bit format file leads to channel swap on mx28 and the reason is that the current driver performs one write/read to/from the
SAIF_DATAregister to trigger the transfer.This approach works fine for
S16_LEcase becauseSAIF_DATAis a 32-bit register and thus is capable of storing the 16-bit left and right channels, but for theS24_LEcase it can only store one channel, so in order to not lose the FIFO sync an extra read/write is needed.Reported-by Dan Winner <[email protected]>Signed-off-by Fabio Estevam <[email protected]>Tested-by Dan Winner <[email protected]>--- sound/soc/mxs/mxs-saif.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-)sound/soc/mxs/mxs-saif.c +0
-0@@ -523,16 +523,22 @@ static int mxs_saif_trigger(struct snd_pcm_substream *substream, int cmd,if (substream->stream ==SNDRV_PCM_STREAM_PLAYBACK ) {
sound/soc/mxs/mxs-saif.c static int mxs_saif_trigger(struct snd_pcm_substream *substream, int cmd, · line 532
Quoted 10 lines
sound/soc/mxs/mxs-saif.c +5
-2/* -* write a data to saif data register to trigger -* the transfer +* write data to saif data register to trigger +* the transfer. +* For 24-bit format the 32-bit FIFO register stores +* only one channel, so we need to write twice. */ __raw_writel (0 , saif->base +SAIF_DATA ); +__raw_writel (0 , saif->base +SAIF_DATA );
This probably could a workaround for the customer, but i'm not sure Mark could accept it
since the code is a bit confusing. (Mark, what's your suggestion on this?) Maybe the correct way to do is to get rid of this write for playback and the read for capture,
since I did not see any place in datasheet to specify we must write/read to trigger the transfer,
But the test result seems we indeed need this to trigger.
And this code also exists in Freescale internal BSP for a long time. Maybe we need spend some more time to dig into why the trigger is needed, however, one bad thing is that the original owner of this IC module has left Freescale,
i have no one to ask about hw details now.
Regards
Dong Aisheng
Quoted 17 lines
sound/soc/mxs/mxs-saif.c +5
-3}else {/* -* read a data from saif data register to trigger -* the receive +* read data from saif data register to trigger +* the receive. +* For 24-bit format the 32-bit FIFO register stores +* only one channel, so we need to read twice. */ __raw_readl (saif->base +SAIF_DATA ); +__raw_readl (saif->base +SAIF_DATA ); } master_saif->ongoing =1 ; --1.7.9.5
Quoted 10 lines · Dong Aisheng
sound/soc/mxs/mxs-saif.c static int mxs_saif_trigger(struct snd_pcm_substream *substream, int cmd, · line 532
Quoted 7 lines · Fabio Estevam
sound/soc/mxs/mxs-saif.c +5
-0+* write data to saif data register to trigger +* the transfer. +* For 24-bit format the 32-bit FIFO register stores +* only one channel, so we need to write twice. */ __raw_writel (0 , saif->base +SAIF_DATA ); +__raw_writel (0 , saif->base +SAIF_DATA );
Quoted 3 lines
This probably could a workaround for the customer, but i'm not sure Mark could accept it
since the code is a bit confusing. (Mark, what's your suggestion on this?)
I'm not really concerned about the code, the comment is fairly clear except that it'd be good to mention that this is safe for data sizes other than 24 bit. It's certainly really much more confusing than the existing code and if it works better that's obviously good!
Quoted 11 lines · Mark Brown
Quoted 9 lines · Dong Aisheng
Quoted 7 lines · Fabio Estevam
sound/soc/mxs/mxs-saif.c +5
-0+* write data to saif data register to trigger +* the transfer. +* For 24-bit format the 32-bit FIFO register stores +* only one channel, so we need to write twice. */ __raw_writel (0 , saif->base +SAIF_DATA ); +__raw_writel (0 , saif->base +SAIF_DATA );
Quoted 8 lines
Quoted 3 lines
This probably could a workaround for the customer, but i'm not sure Mark could accept it
since the code is a bit confusing. (Mark, what's your suggestion on this?)I'm not really concerned about the code, the comment is fairly clear except that it'd be good to mention that this is safe for data sizes other than 24 bit. It's certainly really much more confusing than the existing code and if it works better that's obviously good!
If you're fine with it, i'm also ok.
This probably is a suitable solution for now since finding out the strange IC behavior of this issue
may need much longer time without IC people's help and it's still not on the plan.
Fabio,
You can update the patch with Mark's suggestion and resend it.
Regards
Dong Aisheng