[Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[javier Martin]

Hi,
I am trying to configure a DAI interface between a i.MX27 SSI and a
wm8974 codec.

The definition of SND_SOC_DAIFMT_NB_NF specifies 'normal' bit clock
and frame for the specified format.
http://lxr.linux.no/#linux+v3.2.7/include/sound/soc-dai.h#L48

As I wasn't sure about the meaning of 'normal'  I've dug into the code
of some platforms to see what they do. What I've found, however, is
that it doesn't seem to be an agreement about it.

Let me expose some examples:

1. Frame clock is active High:

i.MX: http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/imx-ssi.c#L131 +
MCIMX27RM.pdf (p 1606)    (default)
PXA:  http://lxr.linux.no/#linux+v3.2.7/sound/soc/pxa/pxa-ssp.c#L498
pxa-ssp.c + PXA3xx_DM_Vol_IV.pdf (p 332)  (not default)

2. Frame clock is active Low:

OMAP: http://lxr.linux.no/#linux+v3.2.7/sound/soc/omap/omap-mcbsp.c#L478
+ sprugn4h.pdf (p 3165) (not default)

Some chips like s6000 seem to provide a more original use for this flag:

http://lxr.linux.no/#linux+v3.2.7/sound/soc/s6000/s6000-i2s.c#L255

Similar behavior has been found for bit clock.

ALso, it seems that there is no possibility to specify different
configurations for RX and TX parts.

-- 
Javier Martin
Vista Silicon S.L.
CDTUC - FASE C - Oficina S-345
Avda de los Castros s/n
39005- Santander. Cantabria. Spain
+34 942 25 32 60
www.vista-silicon.com

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[Mark Brown]

[-- Attachment #1.1: Type: text/plain, Size: 628 bytes --]

On Thu, Feb 23, 2012 at 01:40:40PM +0100, javier Martin wrote:

> As I wasn't sure about the meaning of 'normal'  I've dug into the code
> of some platforms to see what they do. What I've found, however, is
> that it doesn't seem to be an agreement about it.

It's whatever is standard for the given format - the Wolfson datasheets
are a pretty good reference, they have clear diagrams.

> ALso, it seems that there is no possibility to specify different
> configurations for RX and TX parts.

This is correct, if the configurations are different you're really
looking at separate audio interfaces so should have multiple DAIs.

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[javier Martin]

On 23 February 2012 16:19, Mark Brown
<broonie@opensource.wolfsonmicro.com> wrote:
> On Thu, Feb 23, 2012 at 01:40:40PM +0100, javier Martin wrote:
>
>> As I wasn't sure about the meaning of 'normal'  I've dug into the code
>> of some platforms to see what they do. What I've found, however, is
>> that it doesn't seem to be an agreement about it.
>
> It's whatever is standard for the given format - the Wolfson datasheets
> are a pretty good reference, they have clear diagrams.

So, let's see if I understood properly. If I want to configure the
i.MX SSI into standard, I2S format I should do:

dai_format =  SND_SOC_DAIFMT_I2S | SND_SOC_DAIFMT_NB_NF;
snd_soc_dai_set_fmt(cpu_dai, dai_format);

However, this doesn't work with the imx-ssi driver which does the following:
http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/imx-ssi.c#L83

	case SND_SOC_DAIFMT_I2S:
		/* data on rising edge of bclk, frame low 1clk before data */
		strcr |= SSI_STCR_TFSI | SSI_STCR_TEFS | SSI_STCR_TXBIT0;
		scr |= SSI_SCR_NET;
		if (ssi->flags & IMX_SSI_USE_I2S_SLAVE) {
			scr &= ~SSI_I2S_MODE_MASK;
			scr |= SSI_SCR_I2S_MODE_SLAVE;
		}
		break;
[...]
	case SND_SOC_DAIFMT_NB_NF:
		strcr &= ~SSI_STCR_TFSI;
		strcr |= SSI_STCR_TSCKP;
		break;

While initially TFSI bit is set because of I2S flag, it is then
cleared because of the NB_NF.

Does this mean this is a bug then? If it is, I would gladly fix it but
there is at least one platform relying on this misleading behavior
that I can't test:
http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/wm1133-ev1.c#L81


-- 
Javier Martin
Vista Silicon S.L.
CDTUC - FASE C - Oficina S-345
Avda de los Castros s/n
39005- Santander. Cantabria. Spain
+34 942 25 32 60
www.vista-silicon.com

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[Mark Brown]

[-- Attachment #1.1: Type: text/plain, Size: 1004 bytes --]

On Thu, Feb 23, 2012 at 04:58:09PM +0100, javier Martin wrote:

> So, let's see if I understood properly. If I want to configure the
> i.MX SSI into standard, I2S format I should do:

> dai_format =  SND_SOC_DAIFMT_I2S | SND_SOC_DAIFMT_NB_NF;
> snd_soc_dai_set_fmt(cpu_dai, dai_format);

Yes.

> While initially TFSI bit is set because of I2S flag, it is then
> cleared because of the NB_NF.

> Does this mean this is a bug then? If it is, I would gladly fix it but

I've no idea what any of the register settings mean here...

> there is at least one platform relying on this misleading behavior
> that I can't test:
> http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/wm1133-ev1.c#L81

That was correct at the time it was written, though the code has changed
since then and I don't know when it was last retested.  Of course
depending on the signal it may interoperate well even if the setup isn't
actually correct.  I'd just fix the driver and if it works for yours
it's probably OK for that board too.

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[javier Martin]

On 23 February 2012 17:14, Mark Brown
<broonie@opensource.wolfsonmicro.com> wrote:
> On Thu, Feb 23, 2012 at 04:58:09PM +0100, javier Martin wrote:
>
>> So, let's see if I understood properly. If I want to configure the
>> i.MX SSI into standard, I2S format I should do:
>
>> dai_format =  SND_SOC_DAIFMT_I2S | SND_SOC_DAIFMT_NB_NF;
>> snd_soc_dai_set_fmt(cpu_dai, dai_format);
>
> Yes.
>
>> While initially TFSI bit is set because of I2S flag, it is then
>> cleared because of the NB_NF.
>
>> Does this mean this is a bug then? If it is, I would gladly fix it but
>
> I've no idea what any of the register settings mean here...

If TFSI bit is 1, frame clock is active high.
If TFSI bit is 0, frame clock is active low.

In the piece of code I've shown to you, when someone configures
(SND_SOC_DAIFMT_I2S | SND_SOC_DAIFMT_NB_NF) he's really selecting an
active high frame clock when the standard says just the opposite
(wm8974 datasheet p50).

>> there is at least one platform relying on this misleading behavior
>> that I can't test:
>> http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/wm1133-ev1.c#L81
>
> That was correct at the time it was written, though the code has changed
> since then and I don't know when it was last retested.  Of course
> depending on the signal it may interoperate well even if the setup isn't
> actually correct.  I'd just fix the driver and if it works for yours
> it's probably OK for that board too.

It is not one but two boards that are affected:
http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/wm1133-ev1.c#L81
http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/eukrea-tlv320.c#L32

If I fixed the bug in the SSI I would break both of them, since I
would be changing frame clock polarity.

Regards.
-- 
Javier Martin
Vista Silicon S.L.
CDTUC - FASE C - Oficina S-345
Avda de los Castros s/n
39005- Santander. Cantabria. Spain
+34 942 25 32 60
www.vista-silicon.com

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[Peter Ujfalusi]

On 02/24/2012 09:53 AM, javier Martin wrote:
> If TFSI bit is 1, frame clock is active high.
> If TFSI bit is 0, frame clock is active low.
> 
> In the piece of code I've shown to you, when someone configures
> (SND_SOC_DAIFMT_I2S | SND_SOC_DAIFMT_NB_NF) he's really selecting an
> active high frame clock when the standard says just the opposite
> (wm8974 datasheet p50).

We had similar issues on OMAP McBSP in the past. If you take a look at
the omap-mcbsp.c: omap_mcbsp_dai_set_dai_fmt() function we have inv_fs
variable to handle such a case.
We do the FS/BCLK configuration only in DAIFMT_INV config phase. The FS
active level is different in these formats. We invert the FS based on
the used format (LEFT/RIGHT just, DSP). The DAIFMT_INV switch configures
according to I2S, but we invert the FS if we are in DSP, Justified modes
afterwards.

>>> there is at least one platform relying on this misleading behavior
>>> that I can't test:
>>> http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/wm1133-ev1.c#L81
>>
>> That was correct at the time it was written, though the code has changed
>> since then and I don't know when it was last retested.  Of course
>> depending on the signal it may interoperate well even if the setup isn't
>> actually correct.  I'd just fix the driver and if it works for yours
>> it's probably OK for that board too.
> 
> It is not one but two boards that are affected:
> http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/wm1133-ev1.c#L81
> http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/eukrea-tlv320.c#L32
> 
> If I fixed the bug in the SSI I would break both of them, since I
> would be changing frame clock polarity.

I don't think it is going to break. I would guess that the channels are
swapped on these machines at the moment. If you correct the CPU dai
driver the only thing that will happen is that the channels will jump in
their correct place.

-- 
Péter

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[javier Martin]

On 24 February 2012 09:34, Peter Ujfalusi <peter.ujfalusi@ti.com> wrote:
> On 02/24/2012 09:53 AM, javier Martin wrote:
>> If TFSI bit is 1, frame clock is active high.
>> If TFSI bit is 0, frame clock is active low.
>>
>> In the piece of code I've shown to you, when someone configures
>> (SND_SOC_DAIFMT_I2S | SND_SOC_DAIFMT_NB_NF) he's really selecting an
>> active high frame clock when the standard says just the opposite
>> (wm8974 datasheet p50).
>
> We had similar issues on OMAP McBSP in the past. If you take a look at
> the omap-mcbsp.c: omap_mcbsp_dai_set_dai_fmt() function we have inv_fs
> variable to handle such a case.
> We do the FS/BCLK configuration only in DAIFMT_INV config phase. The FS
> active level is different in these formats. We invert the FS based on
> the used format (LEFT/RIGHT just, DSP). The DAIFMT_INV switch configures
> according to I2S, but we invert the FS if we are in DSP, Justified modes
> afterwards.

I intend to do something similar for imx-ssi. What keeps me awake is
that there are some other platforms as well that don't respect the
convention.

>>>> there is at least one platform relying on this misleading behavior
>>>> that I can't test:
>>>> http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/wm1133-ev1.c#L81
>>>
>>> That was correct at the time it was written, though the code has changed
>>> since then and I don't know when it was last retested.  Of course
>>> depending on the signal it may interoperate well even if the setup isn't
>>> actually correct.  I'd just fix the driver and if it works for yours
>>> it's probably OK for that board too.
>>
>> It is not one but two boards that are affected:
>> http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/wm1133-ev1.c#L81
>> http://lxr.linux.no/#linux+v3.2.7/sound/soc/imx/eukrea-tlv320.c#L32
>>
>> If I fixed the bug in the SSI I would break both of them, since I
>> would be changing frame clock polarity.
>
> I don't think it is going to break. I would guess that the channels are
> swapped on these machines at the moment. If you correct the CPU dai
> driver the only thing that will happen is that the channels will jump in
> their correct place.

That would be true for stereo. However, mono should not be working right now.

-- 
Javier Martin
Vista Silicon S.L.
CDTUC - FASE C - Oficina S-345
Avda de los Castros s/n
39005- Santander. Cantabria. Spain
+34 942 25 32 60
www.vista-silicon.com

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[Mark Brown]

[-- Attachment #1.1: Type: text/plain, Size: 852 bytes --]

On Fri, Feb 24, 2012 at 12:05:09PM +0100, javier Martin wrote:
> On 24 February 2012 09:34, Peter Ujfalusi <peter.ujfalusi@ti.com> wrote:

> > I don't think it is going to break. I would guess that the channels are
> > swapped on these machines at the moment. If you correct the CPU dai
> > driver the only thing that will happen is that the channels will jump in
> > their correct place.

> That would be true for stereo. However, mono should not be working right now.

Mono isn't particularly defined for I2S at the best of times; the
clocking is inherently stereo.  You may just pick out the left channel
or you may duplicate things on left and right channels (at which point a
polarity change won't be noticed).  I wouldn't worry too much about it,
it's better to have things the right way round and fix up configuration
for machines that need it.

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[Peter Ujfalusi]

On 02/24/2012 01:05 PM, javier Martin wrote:
>> I don't think it is going to break. I would guess that the channels are
>> swapped on these machines at the moment. If you correct the CPU dai
>> driver the only thing that will happen is that the channels will jump in
>> their correct place.
> 
> That would be true for stereo. However, mono should not be working right now.

The notion of mono in I2S, Left/Right Justified formats is not really
valid. They are stereo protocols by design. DSP_A/B is used with mono
audio (which often referred as PCM).
It is the responsibility of the codec/cpu drivers to interpret the
standards correctly, and configure themselves accordingly. Take a look
at the Wolfson codec documentation they are really good in this matter
(what I2S, DSP/PCM, and other coding means in real life).
You can look at the omap-mcbsp driver for an example, but configuring
the FS/BC polarity in the format selection (without taking a look at the
INV mask) does not seam to be correct.

-- 
Péter

Re: [Q] ASoC: What's the real meaning of SND_SOC_DAIFMT_NB_NF?

[Mark Brown]

[-- Attachment #1.1: Type: text/plain, Size: 673 bytes --]

On Fri, Feb 24, 2012 at 08:53:11AM +0100, javier Martin wrote:

> In the piece of code I've shown to you, when someone configures
> (SND_SOC_DAIFMT_I2S | SND_SOC_DAIFMT_NB_NF) he's really selecting an
> active high frame clock when the standard says just the opposite
> (wm8974 datasheet p50).

> If I fixed the bug in the SSI I would break both of them, since I
> would be changing frame clock polarity.

No, with I2S the main effect of inverting the frame clock is to cause a
left/right swap which is a relatively subtle issue, it's not obvious on
an immediate listening test.  It's not really active high or active low
but rather indicating which channel the sample is.

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]