Hi,
Changes since v1:
- the hw.formats mask preparation moved to dmaengine_pcm_set_runtime_hwparams()
- we prepare the mask for all supported sample formats based on the physical
width of the sample and beased on the slave_caps information.
I have noticed while switching the daVinci audio to dmaengine_pcm that the _3LE
and _3BE support is not working correctly via dmaengine_pcm:
In case of _3LE/_3BE formats the samples are stored in 3 consecutive bytes
without padding it to 4 bytes. This means that the DMA needs to be able to
support 3 bytes word length in order to read/write the samples from memory
correctly. Originally the code treated 24 bits physical length samples as
they were 32 bits which leads to corruption when playing or recording audio.
To fix the support for 3 bytes physical samples the dma driver also need to have
support for such data arrangement. eDMA does have support for it in HW and the
legacy davinci-pcm platform driver w/o dmaengine was able to support such
formats.
First step is to add DMA_SLAVE_BUSWIDTH_3_BYTES to dma_slave_buswidth for
engines and users to select 3 bytes as bus width.
Followed by:
In dmaengine_pcm_set_runtime_hwparams() we should check the slave_caps of the
dma if it supports the give sample physical width of the sample. Based on this
information we initialize the hw.formats: masking out all non supported formats
based on the physical width.
In case the dma driver does not implement the slave_caps the default assumption
is that it supports 1, 2 and 4 bytes widths.
Regards,
Peter
---
Peter Ujfalusi (5):
dma: Support for 3 bytes word size
ALSA: pcm_dmaengine: Use the available wrapper to get physical width
ASoC: generic-dmaengine-pcm: Prepare formats mask for valid physical
sample sizes
ALSA: pcm_dmaengine: Correct support for 24bits physical sample widths
dma: edma: Declare DMA_SLAVE_BUSWIDTH_3_BYTES as supported buswidth
drivers/dma/edma.c | 1 +
include/linux/dmaengine.h | 1 +
sound/core/pcm_dmaengine.c | 4 +++-
sound/soc/soc-generic-dmaengine-pcm.c | 26 +++++++++++++++++++++++++-
4 files changed, 30 insertions(+), 2 deletions(-)
--
2.0.0
Add DMA_SLAVE_BUSWIDTH_3_BYTES to dma_slave_buswidth for engines and users
to select 3 bytes as bus width.
For example eDMA can be configured to use 3bytes mode and in audio we have
formats stored on 3bytes in memory (_XXX_3LE) where this new bus width can
be used.
Signed-off-by: Peter Ujfalusi <[email protected]>
---
include/linux/dmaengine.h | 1 +
1 file changed, 1 insertion(+)
diff --git a/include/linux/dmaengine.h b/include/linux/dmaengine.h
index d2c5cc7c583c..3d1c2aa51530 100644
--- a/include/linux/dmaengine.h
+++ b/include/linux/dmaengine.h
@@ -299,6 +299,7 @@ enum dma_slave_buswidth {
DMA_SLAVE_BUSWIDTH_UNDEFINED = 0,
DMA_SLAVE_BUSWIDTH_1_BYTE = 1,
DMA_SLAVE_BUSWIDTH_2_BYTES = 2,
+ DMA_SLAVE_BUSWIDTH_3_BYTES = 3,
DMA_SLAVE_BUSWIDTH_4_BYTES = 4,
DMA_SLAVE_BUSWIDTH_8_BYTES = 8,
};
--
2.0.0
In case of _3LE/_3BE formats the samples are stored in 3 consecutive bytes
without padding it to 4 bytes. This means that the DMA needs to be able to
support 3 bytes word length in order to read/write the samples from memory
correctly. Originally the code treated 24 bits physical length samples as
they were 32 bits which leads to corruption when playing or recording audio.
The hw.formats field has already been prepared to exclude formats not
supported by the DMA engine in use, which means that only on platforms where
3 bytes is supported by the DMA will be able to use this format.
Signed-off-by: Peter Ujfalusi <[email protected]>
---
sound/core/pcm_dmaengine.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/sound/core/pcm_dmaengine.c b/sound/core/pcm_dmaengine.c
index d5611ec80381..6542c4083594 100644
--- a/sound/core/pcm_dmaengine.c
+++ b/sound/core/pcm_dmaengine.c
@@ -72,6 +72,8 @@ int snd_hwparams_to_dma_slave_config(const struct snd_pcm_substream *substream,
buswidth = DMA_SLAVE_BUSWIDTH_1_BYTE;
else if (bits == 16)
buswidth = DMA_SLAVE_BUSWIDTH_2_BYTES;
+ else if (bits == 24)
+ buswidth = DMA_SLAVE_BUSWIDTH_3_BYTES;
else if (bits <= 32)
buswidth = DMA_SLAVE_BUSWIDTH_4_BYTES;
else
--
2.0.0
eDMA can be configured for 3bytes word size for source and destination.
Signed-off-by: Peter Ujfalusi <[email protected]>
---
drivers/dma/edma.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/dma/edma.c b/drivers/dma/edma.c
index d08c4dedef35..b512caf46944 100644
--- a/drivers/dma/edma.c
+++ b/drivers/dma/edma.c
@@ -982,6 +982,7 @@ static void __init edma_chan_init(struct edma_cc *ecc,
#define EDMA_DMA_BUSWIDTHS (BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
+ BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
BIT(DMA_SLAVE_BUSWIDTH_4_BYTES))
static int edma_dma_device_slave_caps(struct dma_chan *dchan,
--
2.0.0
Based on the dma_slave_caps's addr_widths queried from the dma driver
prepare the hw.formats mask to include only formats which is supported by
the DMA engine.
In case the dma driver does not implement the slave_caps the default
assumption is that it supports 1, 2 and 4 bytes widths.
Signed-off-by: Peter Ujfalusi <[email protected]>
---
sound/soc/soc-generic-dmaengine-pcm.c | 26 +++++++++++++++++++++++++-
1 file changed, 25 insertions(+), 1 deletion(-)
diff --git a/sound/soc/soc-generic-dmaengine-pcm.c b/sound/soc/soc-generic-dmaengine-pcm.c
index 5bace124ef43..c295e1754d1a 100644
--- a/sound/soc/soc-generic-dmaengine-pcm.c
+++ b/sound/soc/soc-generic-dmaengine-pcm.c
@@ -119,7 +119,10 @@ static int dmaengine_pcm_set_runtime_hwparams(struct snd_pcm_substream *substrea
struct snd_dmaengine_dai_dma_data *dma_data;
struct dma_slave_caps dma_caps;
struct snd_pcm_hardware hw;
- int ret;
+ u32 addr_widths = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
+ BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
+ BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
+ int i, ret;
if (pcm->config && pcm->config->pcm_hardware)
return snd_soc_set_runtime_hwparams(substream,
@@ -146,6 +149,27 @@ static int dmaengine_pcm_set_runtime_hwparams(struct snd_pcm_substream *substrea
hw.info |= SNDRV_PCM_INFO_PAUSE | SNDRV_PCM_INFO_RESUME;
if (dma_caps.residue_granularity <= DMA_RESIDUE_GRANULARITY_SEGMENT)
hw.info |= SNDRV_PCM_INFO_BATCH;
+
+ if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
+ addr_widths = dma_caps.dstn_addr_widths;
+ else
+ addr_widths = dma_caps.src_addr_widths;
+ }
+
+ /*
+ * Prepare formats mask for valid/allowed sample types. If the dma does
+ * not have support for the given physical word size, it needs to be
+ * masked out so user space can not use the format which produces
+ * corrupted audio.
+ * In case the dma driver does not implement the slave_caps the default
+ * assumption is that it supports 1, 2 and 4 bytes widths.
+ */
+ for (i = 0; i <= SNDRV_PCM_FORMAT_LAST; i++) {
+ int bits = snd_pcm_format_physical_width(i);
+
+ /* Enable only samples with DMA supported physical widths */
+ if (addr_widths & BIT(bits / 8))
+ hw.formats |= (1LL << i);
}
return snd_soc_set_runtime_hwparams(substream, &hw);
--
2.0.0
params_physical_width() is available via pcm_params.h
Signed-off-by: Peter Ujfalusi <[email protected]>
Acked-by: Lars-Peter Clausen <[email protected]>
---
sound/core/pcm_dmaengine.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/sound/core/pcm_dmaengine.c b/sound/core/pcm_dmaengine.c
index 76cbb9ec953a..d5611ec80381 100644
--- a/sound/core/pcm_dmaengine.c
+++ b/sound/core/pcm_dmaengine.c
@@ -65,7 +65,7 @@ int snd_hwparams_to_dma_slave_config(const struct snd_pcm_substream *substream,
enum dma_slave_buswidth buswidth;
int bits;
- bits = snd_pcm_format_physical_width(params_format(params));
+ bits = params_physical_width(params);
if (bits < 8 || bits > 64)
return -EINVAL;
else if (bits == 8)
--
2.0.0
At Wed, 2 Jul 2014 16:29:21 +0300,
Peter Ujfalusi wrote:
>
> Based on the dma_slave_caps's addr_widths queried from the dma driver
> prepare the hw.formats mask to include only formats which is supported by
> the DMA engine.
> In case the dma driver does not implement the slave_caps the default
> assumption is that it supports 1, 2 and 4 bytes widths.
>
> Signed-off-by: Peter Ujfalusi <[email protected]>
> ---
> sound/soc/soc-generic-dmaengine-pcm.c | 26 +++++++++++++++++++++++++-
> 1 file changed, 25 insertions(+), 1 deletion(-)
>
> diff --git a/sound/soc/soc-generic-dmaengine-pcm.c b/sound/soc/soc-generic-dmaengine-pcm.c
> index 5bace124ef43..c295e1754d1a 100644
> --- a/sound/soc/soc-generic-dmaengine-pcm.c
> +++ b/sound/soc/soc-generic-dmaengine-pcm.c
> @@ -119,7 +119,10 @@ static int dmaengine_pcm_set_runtime_hwparams(struct snd_pcm_substream *substrea
> struct snd_dmaengine_dai_dma_data *dma_data;
> struct dma_slave_caps dma_caps;
> struct snd_pcm_hardware hw;
> - int ret;
> + u32 addr_widths = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
> + BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
> + BIT(DMA_SLAVE_BUSWIDTH_4_BYTES);
> + int i, ret;
>
> if (pcm->config && pcm->config->pcm_hardware)
> return snd_soc_set_runtime_hwparams(substream,
> @@ -146,6 +149,27 @@ static int dmaengine_pcm_set_runtime_hwparams(struct snd_pcm_substream *substrea
> hw.info |= SNDRV_PCM_INFO_PAUSE | SNDRV_PCM_INFO_RESUME;
> if (dma_caps.residue_granularity <= DMA_RESIDUE_GRANULARITY_SEGMENT)
> hw.info |= SNDRV_PCM_INFO_BATCH;
> +
> + if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
> + addr_widths = dma_caps.dstn_addr_widths;
> + else
> + addr_widths = dma_caps.src_addr_widths;
> + }
> +
> + /*
> + * Prepare formats mask for valid/allowed sample types. If the dma does
> + * not have support for the given physical word size, it needs to be
> + * masked out so user space can not use the format which produces
> + * corrupted audio.
> + * In case the dma driver does not implement the slave_caps the default
> + * assumption is that it supports 1, 2 and 4 bytes widths.
> + */
> + for (i = 0; i <= SNDRV_PCM_FORMAT_LAST; i++) {
> + int bits = snd_pcm_format_physical_width(i);
Better to add a negative value check. Some formats return -EINVAL.
> +
> + /* Enable only samples with DMA supported physical widths */
> + if (addr_widths & BIT(bits / 8))
Just nitpicking: Using BIT() for 32bit int isn't always a wise choice
since BIT() is defined as unsigned long. This is a generic problem,
though, as we see many codes using BIT() wrongly (and thus get
compiler warnings on 64bit machines). We should introduce a new macro
for 32bit int...
Takashi
On 07/02/2014 03:47 PM, Takashi Iwai wrote:
[...]
>> + /*
>> + * Prepare formats mask for valid/allowed sample types. If the dma does
>> + * not have support for the given physical word size, it needs to be
>> + * masked out so user space can not use the format which produces
>> + * corrupted audio.
>> + * In case the dma driver does not implement the slave_caps the default
>> + * assumption is that it supports 1, 2 and 4 bytes widths.
>> + */
>> + for (i = 0; i <= SNDRV_PCM_FORMAT_LAST; i++) {
>> + int bits = snd_pcm_format_physical_width(i);
>
> Better to add a negative value check. Some formats return -EINVAL.
>
So what should we do with those formats? Not support them?
>> +
>> + /* Enable only samples with DMA supported physical widths */
>> + if (addr_widths & BIT(bits / 8))
>
> Just nitpicking: Using BIT() for 32bit int isn't always a wise choice
> since BIT() is defined as unsigned long. This is a generic problem,
> though, as we see many codes using BIT() wrongly (and thus get
> compiler warnings on 64bit machines). We should introduce a new macro
> for 32bit int...
At Wed, 02 Jul 2014 16:10:09 +0200,
Lars-Peter Clausen wrote:
>
> On 07/02/2014 03:47 PM, Takashi Iwai wrote:
> [...]
> >> + /*
> >> + * Prepare formats mask for valid/allowed sample types. If the dma does
> >> + * not have support for the given physical word size, it needs to be
> >> + * masked out so user space can not use the format which produces
> >> + * corrupted audio.
> >> + * In case the dma driver does not implement the slave_caps the default
> >> + * assumption is that it supports 1, 2 and 4 bytes widths.
> >> + */
> >> + for (i = 0; i <= SNDRV_PCM_FORMAT_LAST; i++) {
> >> + int bits = snd_pcm_format_physical_width(i);
> >
> > Better to add a negative value check. Some formats return -EINVAL.
> >
>
> So what should we do with those formats? Not support them?
Yes, these are special formats that can't be handled generically.
Takashi
On 07/02/2014 04:12 PM, Takashi Iwai wrote:
> At Wed, 02 Jul 2014 16:10:09 +0200,
> Lars-Peter Clausen wrote:
>>
>> On 07/02/2014 03:47 PM, Takashi Iwai wrote:
>> [...]
>>>> + /*
>>>> + * Prepare formats mask for valid/allowed sample types. If the dma does
>>>> + * not have support for the given physical word size, it needs to be
>>>> + * masked out so user space can not use the format which produces
>>>> + * corrupted audio.
>>>> + * In case the dma driver does not implement the slave_caps the default
>>>> + * assumption is that it supports 1, 2 and 4 bytes widths.
>>>> + */
>>>> + for (i = 0; i <= SNDRV_PCM_FORMAT_LAST; i++) {
>>>> + int bits = snd_pcm_format_physical_width(i);
>>>
>>> Better to add a negative value check. Some formats return -EINVAL.
>>>
>>
>> So what should we do with those formats? Not support them?
>
> Yes, these are special formats that can't be handled generically.
>
Looking through the list in pcm_misc.c there are also some formats where the
physical width is not a multiple of 8. We should probably also skip those.
- Lars
At Wed, 02 Jul 2014 16:13:59 +0200,
Lars-Peter Clausen wrote:
>
> On 07/02/2014 04:12 PM, Takashi Iwai wrote:
> > At Wed, 02 Jul 2014 16:10:09 +0200,
> > Lars-Peter Clausen wrote:
> >>
> >> On 07/02/2014 03:47 PM, Takashi Iwai wrote:
> >> [...]
> >>>> + /*
> >>>> + * Prepare formats mask for valid/allowed sample types. If the dma does
> >>>> + * not have support for the given physical word size, it needs to be
> >>>> + * masked out so user space can not use the format which produces
> >>>> + * corrupted audio.
> >>>> + * In case the dma driver does not implement the slave_caps the default
> >>>> + * assumption is that it supports 1, 2 and 4 bytes widths.
> >>>> + */
> >>>> + for (i = 0; i <= SNDRV_PCM_FORMAT_LAST; i++) {
> >>>> + int bits = snd_pcm_format_physical_width(i);
> >>>
> >>> Better to add a negative value check. Some formats return -EINVAL.
> >>>
> >>
> >> So what should we do with those formats? Not support them?
> >
> > Yes, these are special formats that can't be handled generically.
> >
>
> Looking through the list in pcm_misc.c there are also some formats where the
> physical width is not a multiple of 8. We should probably also skip those.
Right, ADPCM and G723 are such ones.
Takashi