ASoc: soc_core.c stream direction from snd_soc_dai

ASoc: soc_core.c stream direction from snd_soc_dai

[Matt Flax]

Hi there,

A large number of audio codecs allow different formats for playback and 
capture. This becomes very useful when there are different latencies 
between playback and capture hardware data lines. For example digital 
isolation chips typically have a 1 bit delay in propagation as the bit 
clock rate gets faster for higher sample rates. By setting the capture 
and playback formats to differ by one or two bit clock cycles, the delay 
problem is solved.

There doesn't seem to be a simple way to detect stream direction in the 
codec driver's set_fmt function.

The snd_soc_runtime_set_dai_fmt function :

https://github.com/torvalds/linux/blob/master/sound/soc/soc-core.c#L1480

calls the snd_soc_dai_set_fmt function :

https://github.com/torvalds/linux/blob/master/sound/soc/soc-dai.c#L101

which calls the set_fmt function :

https://github.com/torvalds/linux/blob/master/include/sound/soc-dai.h#L189


The snd_soc_dai_ops set_fmt function is defined as :

     int (*set_fmt)(struct snd_soc_dai *dai, unsigned int fmt);


Is there a simple way to find the stream direction from a snd_soc_dai ?

If the stream direction can be detected then the playback and capture 
formats can be set independently for the codec.

It there a different way to set the playback and capture formats for the 
codec independently at runtime, depending on the sample rate ?

Matt

Re: ASoc: soc_core.c stream direction from snd_soc_dai

[Pierre-Louis Bossart]

On 3/11/20 5:54 PM, Matt Flax wrote:
> Hi there,
> 
> A large number of audio codecs allow different formats for playback and 
> capture. This becomes very useful when there are different latencies 
> between playback and capture hardware data lines. For example digital 
> isolation chips typically have a 1 bit delay in propagation as the bit 
> clock rate gets faster for higher sample rates. By setting the capture 
> and playback formats to differ by one or two bit clock cycles, the delay 
> problem is solved.
> 
> There doesn't seem to be a simple way to detect stream direction in the 
> codec driver's set_fmt function.
> 
> The snd_soc_runtime_set_dai_fmt function :
> 
> https://github.com/torvalds/linux/blob/master/sound/soc/soc-core.c#L1480
> 
> calls the snd_soc_dai_set_fmt function :
> 
> https://github.com/torvalds/linux/blob/master/sound/soc/soc-dai.c#L101
> 
> which calls the set_fmt function :
> 
> https://github.com/torvalds/linux/blob/master/include/sound/soc-dai.h#L189
> 
> 
> The snd_soc_dai_ops set_fmt function is defined as :
> 
>      int (*set_fmt)(struct snd_soc_dai *dai, unsigned int fmt);
> 
> 
> Is there a simple way to find the stream direction from a snd_soc_dai ?
> 
> If the stream direction can be detected then the playback and capture 
> formats can be set independently for the codec.
> 
> It there a different way to set the playback and capture formats for the 
> codec independently at runtime, depending on the sample rate ?

FWIW I remember a discussion in the past on how to deal with interfaces 
that may have different clocks sources for capture and playback 
(typically with the 6-pin version of I2S/TDM), and the answer was: use 
two dais, with one dealing with capture and the other with playback.

I would bet this applies for the format as well. If you use a DAI that 
can do both directions, then indeed there's no obvious way to specify 
that formats or clock ownership could be different between the two 
directions.

It would probably make sense anyway to have a representation with two 
dais, e.g. the codec capture dai receives data from somewhere and the 
codec playback dai forwards it to another destination.

My 2 cents
-Pierre

Re: ASoc: soc_core.c stream direction from snd_soc_dai

[Matt Flax]

On 13/3/20 9:55 am, Pierre-Louis Bossart wrote:
>
>
> On 3/11/20 5:54 PM, Matt Flax wrote:
>> Hi there,
>>
>> A large number of audio codecs allow different formats for playback 
>> and capture. This becomes very useful when there are different 
>> latencies between playback and capture hardware data lines. For 
>> example digital isolation chips typically have a 1 bit delay in 
>> propagation as the bit clock rate gets faster for higher sample 
>> rates. By setting the capture and playback formats to differ by one 
>> or two bit clock cycles, the delay problem is solved.
>>
>> There doesn't seem to be a simple way to detect stream direction in 
>> the codec driver's set_fmt function.
>>
>> The snd_soc_runtime_set_dai_fmt function :
>>
>> https://github.com/torvalds/linux/blob/master/sound/soc/soc-core.c#L1480
>>
>> calls the snd_soc_dai_set_fmt function :
>>
>> https://github.com/torvalds/linux/blob/master/sound/soc/soc-dai.c#L101
>>
>> which calls the set_fmt function :
>>
>> https://github.com/torvalds/linux/blob/master/include/sound/soc-dai.h#L189 
>>
>>
>>
>> The snd_soc_dai_ops set_fmt function is defined as :
>>
>>      int (*set_fmt)(struct snd_soc_dai *dai, unsigned int fmt);
>>
>>
>> Is there a simple way to find the stream direction from a snd_soc_dai ?
>>
>> If the stream direction can be detected then the playback and capture 
>> formats can be set independently for the codec.
>>
>> It there a different way to set the playback and capture formats for 
>> the codec independently at runtime, depending on the sample rate ?
>
> FWIW I remember a discussion in the past on how to deal with 
> interfaces that may have different clocks sources for capture and 
> playback (typically with the 6-pin version of I2S/TDM), and the answer 
> was: use two dais, with one dealing with capture and the other with 
> playback.
>
> I would bet this applies for the format as well. If you use a DAI that 
> can do both directions, then indeed there's no obvious way to specify 
> that formats or clock ownership could be different between the two 
> directions.
>
> It would probably make sense anyway to have a representation with two 
> dais, e.g. the codec capture dai receives data from somewhere and the 
> codec playback dai forwards it to another destination.
>
I think I get it ...

This approach would keep extra stream selective functionality out of 
soc-dai.c. That is probably a good thing for the simplicity of the core.

A machine driver could then call snd_soc_dai_set_fmt passing in the 
correct codec_dai from the codec_dais array for the stream they want to 
operate on.

Matt


> My 2 cents
> -Pierre

Re: ASoc: soc_core.c stream direction from snd_soc_dai

[Matt Flax]

On 13/3/20 10:19 am, Matt Flax wrote:
>
> On 13/3/20 9:55 am, Pierre-Louis Bossart wrote:
>>
>>
>> On 3/11/20 5:54 PM, Matt Flax wrote:
>>> Hi there,
>>>
>>> A large number of audio codecs allow different formats for playback 
>>> and capture. This becomes very useful when there are different 
>>> latencies between playback and capture hardware data lines. For 
>>> example digital isolation chips typically have a 1 bit delay in 
>>> propagation as the bit clock rate gets faster for higher sample 
>>> rates. By setting the capture and playback formats to differ by one 
>>> or two bit clock cycles, the delay problem is solved.
>>>
>>> There doesn't seem to be a simple way to detect stream direction in 
>>> the codec driver's set_fmt function.
>>>
>>> The snd_soc_runtime_set_dai_fmt function :
>>>
>>> https://github.com/torvalds/linux/blob/master/sound/soc/soc-core.c#L1480 
>>>
>>>
>>> calls the snd_soc_dai_set_fmt function :
>>>
>>> https://github.com/torvalds/linux/blob/master/sound/soc/soc-dai.c#L101
>>>
>>> which calls the set_fmt function :
>>>
>>> https://github.com/torvalds/linux/blob/master/include/sound/soc-dai.h#L189 
>>>
>>>
>>>
>>> The snd_soc_dai_ops set_fmt function is defined as :
>>>
>>>      int (*set_fmt)(struct snd_soc_dai *dai, unsigned int fmt);
>>>
>>>
>>> Is there a simple way to find the stream direction from a snd_soc_dai ?
>>>
>>> If the stream direction can be detected then the playback and 
>>> capture formats can be set independently for the codec.
>>>
>>> It there a different way to set the playback and capture formats for 
>>> the codec independently at runtime, depending on the sample rate ?
>>
>> FWIW I remember a discussion in the past on how to deal with 
>> interfaces that may have different clocks sources for capture and 
>> playback (typically with the 6-pin version of I2S/TDM), and the 
>> answer was: use two dais, with one dealing with capture and the other 
>> with playback.
>>
>> I would bet this applies for the format as well. If you use a DAI 
>> that can do both directions, then indeed there's no obvious way to 
>> specify that formats or clock ownership could be different between 
>> the two directions.
>>
>> It would probably make sense anyway to have a representation with two 
>> dais, e.g. the codec capture dai receives data from somewhere and the 
>> codec playback dai forwards it to another destination.
>>
> I think I get it ...
>
> This approach would keep extra stream selective functionality out of 
> soc-dai.c. That is probably a good thing for the simplicity of the core.
>
> A machine driver could then call snd_soc_dai_set_fmt passing in the 
> correct codec_dai from the codec_dais array for the stream they want 
> to operate on.
>
In an example case, cs4271 ... how do we enforce symmetric rates ?

static struct snd_soc_dai_driver cs4271_dai[] = {
     {
         .name = "cs4271-hifi-p",
         .playback = {
             .stream_name    = "Playback",
             .channels_min    = 2,
             .channels_max    = 2,
             .rates        = CS4271_PCM_RATES,
             .formats    = CS4271_PCM_FORMATS,
         },
         .ops = &cs4271_dai_ops,
         .symmetric_rates = 1,
     },
     {
         .name = "cs4271-hifi-c",
         .capture = {
             .stream_name    = "Capture",
             .channels_min    = 2,
             .channels_max    = 2,
             .rates        = CS4271_PCM_RATES,
             .formats    = CS4271_PCM_FORMATS,
         },
         .ops = &cs4271_dai_ops,
         .symmetric_rates = 1,
     }
};



> Matt
>
>
>> My 2 cents
>> -Pierre

Re: ASoc: soc_core.c stream direction from snd_soc_dai

[Lars-Peter Clausen]

On 3/13/20 4:56 AM, Matt Flax wrote:
>
> On 13/3/20 10:19 am, Matt Flax wrote:
>>
>> On 13/3/20 9:55 am, Pierre-Louis Bossart wrote:
>>>
>>>
>>> On 3/11/20 5:54 PM, Matt Flax wrote:
>>>> Hi there,
>>>>
>>>> A large number of audio codecs allow different formats for playback 
>>>> and capture. This becomes very useful when there are different 
>>>> latencies between playback and capture hardware data lines. For 
>>>> example digital isolation chips typically have a 1 bit delay in 
>>>> propagation as the bit clock rate gets faster for higher sample 
>>>> rates. By setting the capture and playback formats to differ by one 
>>>> or two bit clock cycles, the delay problem is solved.
>>>>
>>>> There doesn't seem to be a simple way to detect stream direction in 
>>>> the codec driver's set_fmt function.
>>>>
>>>> The snd_soc_runtime_set_dai_fmt function :
>>>>
>>>> https://github.com/torvalds/linux/blob/master/sound/soc/soc-core.c#L1480 
>>>>
>>>>
>>>> calls the snd_soc_dai_set_fmt function :
>>>>
>>>> https://github.com/torvalds/linux/blob/master/sound/soc/soc-dai.c#L101
>>>>
>>>> which calls the set_fmt function :
>>>>
>>>> https://github.com/torvalds/linux/blob/master/include/sound/soc-dai.h#L189 
>>>>
>>>>
>>>>
>>>> The snd_soc_dai_ops set_fmt function is defined as :
>>>>
>>>>      int (*set_fmt)(struct snd_soc_dai *dai, unsigned int fmt);
>>>>
>>>>
>>>> Is there a simple way to find the stream direction from a 
>>>> snd_soc_dai ?
>>>>
>>>> If the stream direction can be detected then the playback and 
>>>> capture formats can be set independently for the codec.
>>>>
>>>> It there a different way to set the playback and capture formats 
>>>> for the codec independently at runtime, depending on the sample rate ?
>>>
>>> FWIW I remember a discussion in the past on how to deal with 
>>> interfaces that may have different clocks sources for capture and 
>>> playback (typically with the 6-pin version of I2S/TDM), and the 
>>> answer was: use two dais, with one dealing with capture and the 
>>> other with playback.
>>>
>>> I would bet this applies for the format as well. If you use a DAI 
>>> that can do both directions, then indeed there's no obvious way to 
>>> specify that formats or clock ownership could be different between 
>>> the two directions.
>>>
>>> It would probably make sense anyway to have a representation with 
>>> two dais, e.g. the codec capture dai receives data from somewhere 
>>> and the codec playback dai forwards it to another destination.
>>>
>> I think I get it ...
>>
>> This approach would keep extra stream selective functionality out of 
>> soc-dai.c. That is probably a good thing for the simplicity of the core.
>>
>> A machine driver could then call snd_soc_dai_set_fmt passing in the 
>> correct codec_dai from the codec_dais array for the stream they want 
>> to operate on.
>>
> In an example case, cs4271 ... how do we enforce symmetric rates ?

You'd have to do this manually in your driver. The core itself does not 
support synchronizing streams on different DAIs.

You can do this by saving the rate when it is set on the first stream 
and then apply a constraint to the second stream using 
snd_pcm_hw_constraint_single() it the driver's startup() callback.

Have a look at the uda134x.c or twl4030.c driver as an example.

But I think Pierre was mainly talking about the case where there are 
separate clocks for each direction and the rates don't have to be the 
same. I believe long term it might make sense to extend the core to 
allow different formats for input and output direction on the same DAI.

- Lars

Re: ASoc: soc_core.c stream direction from snd_soc_dai

[Shengjiu Wang]

On Fri, Mar 13, 2020 at 6:01 PM Lars-Peter Clausen <lars@metafoo.de> wrote:
>
> On 3/13/20 4:56 AM, Matt Flax wrote:
> >
> > On 13/3/20 10:19 am, Matt Flax wrote:
> >>
> >> On 13/3/20 9:55 am, Pierre-Louis Bossart wrote:
> >>>
> >>>
> >>> On 3/11/20 5:54 PM, Matt Flax wrote:
> >>>> Hi there,
> >>>>
> >>>> A large number of audio codecs allow different formats for playback
> >>>> and capture. This becomes very useful when there are different
> >>>> latencies between playback and capture hardware data lines. For
> >>>> example digital isolation chips typically have a 1 bit delay in
> >>>> propagation as the bit clock rate gets faster for higher sample
> >>>> rates. By setting the capture and playback formats to differ by one
> >>>> or two bit clock cycles, the delay problem is solved.
> >>>>
> >>>> There doesn't seem to be a simple way to detect stream direction in
> >>>> the codec driver's set_fmt function.
> >>>>
> >>>> The snd_soc_runtime_set_dai_fmt function :
> >>>>
> >>>> https://github.com/torvalds/linux/blob/master/sound/soc/soc-core.c#L1480
> >>>>
> >>>>
> >>>> calls the snd_soc_dai_set_fmt function :
> >>>>
> >>>> https://github.com/torvalds/linux/blob/master/sound/soc/soc-dai.c#L101
> >>>>
> >>>> which calls the set_fmt function :
> >>>>
> >>>> https://github.com/torvalds/linux/blob/master/include/sound/soc-dai.h#L189
> >>>>
> >>>>
> >>>>
> >>>> The snd_soc_dai_ops set_fmt function is defined as :
> >>>>
> >>>>      int (*set_fmt)(struct snd_soc_dai *dai, unsigned int fmt);
> >>>>
> >>>>
> >>>> Is there a simple way to find the stream direction from a
> >>>> snd_soc_dai ?
> >>>>
> >>>> If the stream direction can be detected then the playback and
> >>>> capture formats can be set independently for the codec.
> >>>>
> >>>> It there a different way to set the playback and capture formats
> >>>> for the codec independently at runtime, depending on the sample rate ?
> >>>
> >>> FWIW I remember a discussion in the past on how to deal with
> >>> interfaces that may have different clocks sources for capture and
> >>> playback (typically with the 6-pin version of I2S/TDM), and the
> >>> answer was: use two dais, with one dealing with capture and the
> >>> other with playback.
> >>>
> >>> I would bet this applies for the format as well. If you use a DAI
> >>> that can do both directions, then indeed there's no obvious way to
> >>> specify that formats or clock ownership could be different between
> >>> the two directions.
> >>>
> >>> It would probably make sense anyway to have a representation with
> >>> two dais, e.g. the codec capture dai receives data from somewhere
> >>> and the codec playback dai forwards it to another destination.
> >>>
> >> I think I get it ...
> >>
> >> This approach would keep extra stream selective functionality out of
> >> soc-dai.c. That is probably a good thing for the simplicity of the core.
> >>
> >> A machine driver could then call snd_soc_dai_set_fmt passing in the
> >> correct codec_dai from the codec_dais array for the stream they want
> >> to operate on.
> >>
> > In an example case, cs4271 ... how do we enforce symmetric rates ?
>
> You'd have to do this manually in your driver. The core itself does not
> support synchronizing streams on different DAIs.
>
> You can do this by saving the rate when it is set on the first stream
> and then apply a constraint to the second stream using
> snd_pcm_hw_constraint_single() it the driver's startup() callback.
>
> Have a look at the uda134x.c or twl4030.c driver as an example.
>
> But I think Pierre was mainly talking about the case where there are
> separate clocks for each direction and the rates don't have to be the
> same. I believe long term it might make sense to extend the core to
> allow different formats for input and output direction on the same DAI.
>

Sorry for resuming this old thread. I have the same requirement for
supporting different formats for input and output on the same DAI.

One of the suggestions is to use two DAIs.  but sometimes the
CPU/CODEC may have the same format for playback and capture,
then one DAI is enough.  it means that we need to define 3 DAIs
for the CPU/CODEC,  one supports playback and capture, another
two support capture and playback separately, is it some kind of
duplicate?

So I'd like to extend the set_fmt() interface, but this impacts all
the drivers.

Best regards
Wang shengjiu

Re: ASoc: soc_core.c stream direction from snd_soc_dai

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 964 bytes --]

On Sat, Feb 20, 2021 at 05:29:03PM +0800, Shengjiu Wang wrote:

> One of the suggestions is to use two DAIs.  but sometimes the
> CPU/CODEC may have the same format for playback and capture,
> then one DAI is enough.  it means that we need to define 3 DAIs
> for the CPU/CODEC,  one supports playback and capture, another
> two support capture and playback separately, is it some kind of
> duplicate?

> So I'd like to extend the set_fmt() interface, but this impacts all
> the drivers.

I'm not sure having two DAIs is an issue if you have them running the
smae format - does it cause any practical problems or is it just that it
doesn't seem elegant to you?  There were quite a few devices that pretty
much just had two unidirectional DAIs, in those cases it seems like a
sensible representation for the hardware.  If you can set unrelated
formats on transmit and receive then it's not clear that it's actually
the same DAI in anything except logical labelling.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: ASoc: soc_core.c stream direction from snd_soc_dai

[Shengjiu Wang]

On Tue, Feb 23, 2021 at 9:58 PM Mark Brown <broonie@kernel.org> wrote:
>
> On Sat, Feb 20, 2021 at 05:29:03PM +0800, Shengjiu Wang wrote:
>
> > One of the suggestions is to use two DAIs.  but sometimes the
> > CPU/CODEC may have the same format for playback and capture,
> > then one DAI is enough.  it means that we need to define 3 DAIs
> > for the CPU/CODEC,  one supports playback and capture, another
> > two support capture and playback separately, is it some kind of
> > duplicate?
>
> > So I'd like to extend the set_fmt() interface, but this impacts all
> > the drivers.
>
> I'm not sure having two DAIs is an issue if you have them running the
> smae format - does it cause any practical problems or is it just that it
> doesn't seem elegant to you?  There were quite a few devices that pretty
> much just had two unidirectional DAIs, in those cases it seems like a
> sensible representation for the hardware.  If you can set unrelated
> formats on transmit and receive then it's not clear that it's actually
> the same DAI in anything except logical labelling.

When an i2s device, Sometimes it is connected as async mode, there is
different clock for tx and rx then there should be two DAIs,  Sometimes
it is connected as sync mode, then there should be one DAI.

So we need to register different DAIs according to the async or sync mode
when the driver probes. right?

best regards
wang shengjiu

Re: ASoc: soc_core.c stream direction from snd_soc_dai

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 1078 bytes --]

On Fri, Feb 26, 2021 at 01:58:12PM +0800, Shengjiu Wang wrote:
> On Tue, Feb 23, 2021 at 9:58 PM Mark Brown <broonie@kernel.org> wrote:

> > I'm not sure having two DAIs is an issue if you have them running the
> > smae format - does it cause any practical problems or is it just that it
> > doesn't seem elegant to you?  There were quite a few devices that pretty
> > much just had two unidirectional DAIs, in those cases it seems like a
> > sensible representation for the hardware.  If you can set unrelated
> > formats on transmit and receive then it's not clear that it's actually
> > the same DAI in anything except logical labelling.

> When an i2s device, Sometimes it is connected as async mode, there is
> different clock for tx and rx then there should be two DAIs,  Sometimes
> it is connected as sync mode, then there should be one DAI.

> So we need to register different DAIs according to the async or sync mode
> when the driver probes. right?

You should just be able to connect the same device twice for the other
end of the link, once for Tx and once for Rx.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]