[alsa-devel] [PATCH] ASoC: atmel_ssc_dai: Track playback and capture CMR dividers separately.

[Peter Rosin]

Hi again,

> -----Original Message-----
> From: Peter Rosin
> Sent: Tuesday, October 21, 2014 13:05
> To: 'Bo Shen'
> Cc: Liam Girdwood; Mark Brown; Jaroslav Kysela; Takashi Iwai; 'alsa-
> devel at alsa-project.org'; linux-kernel at vger.kernel.org
> Subject: RE: [alsa-devel] [PATCH] ASoC: atmel_ssc_dai: Track playback and
> capture CMR dividers separately.
> 
> Hi again!
> 
> > Hi Peter,
> >
> > On 10/21/2014 03:55 PM, Peter Rosin wrote:
> > > Hi!
> > >
> > > (and thank you for the pointer to the example with the ssc-dai in
> > > master mode)
> > >
> > >> Hi Peter,
> > >>
> > >> On 10/20/2014 09:45 PM, Peter Rosin wrote:
> > >>>   From 1e5621d7b9887c648d1a66238dc82d715c1e2cad Mon Sep 17
> > 00:00:00
> > >>> 2001
> > >>> From: Peter Rosin <peda at axentia.se>
> > >>> Date: Mon, 20 Oct 2014 14:38:04 +0200
> > >>> Subject: [PATCH] ASoC: atmel_ssc_dai: Track playback and capture
> CMR
> > >> dividers
> > >>>    separately.
> > >>>
> > >>> The CMR divider register is shared by playback and capture. The SSC
> > >>> driver therefore tries to enforce rules so that the needed register
> > >>> content do not conflict during simultaneous playback/capture.
> > >>> However, the implementation also prevents changing the register
> > >>> content in a half-duplex scenario, which is needed when changing
> > sample rates.
> > >>
> > >> I am not fully get what you mean here, do you mean:
> > >>     - when playback, first playback 48kHz, and then playback 8Khz,
> > >> the 8Khz still playback in 48kHz mode?
> > >>
> > >> or other things?
> > >
> > > I do not know exactly why the problem occurs, but it might be
> > > connected to the library (proprietary) we are using. It apparently
> > > uses the OSS interface and somehow it opens a stream and changes the
> > > playback sample-rate a couple of time before it settles on something.
> > > I don't know why this is done, and I have no control over it. The end
> > > result is that without this patch, the ssc-dai driver returns -EBUSY
> > > when the library changes the playback sample-rate (the driver returns
> > > -EBUSY because it thinks that the new sample rate does not match a
> > > previously given capture sample-rate, but that is of course bogus when
> no
> > capture is going on at all).
> >
> > If this issue is caused by your proprietary library, please fix in the library.
> 
> Ah, but it's not "our" proprietary library to fix, it's simply a library that we
> are using (speech synthesis, not our area of expertise). So, I'm not in a
> position to simply fix it, as I have no source code.
> 
> > I don't know how this code can fix your issue and also there is no
> > improvement to the code and it absolutely increase work (choose which
> > clock to configure: tcmr or rcmr) for the SSC work in master. And also this
> > patch may confuse the end user, let them thinking the clock for tcmr and
> > rcmr can configure seperately.
> 
> I added some traces to our hw_params and got the following (without the
> patch):
> 
> [  161.170000] atmel ssc startup
> [  161.170000]  TFA9879: axentia_asoc_tfa9879_hw_params - mclk rate
> 66000000
> [  161.180000]  TFA9879: axentia_asoc_tfa9879_hw_params - bclk rate 128000
> [  161.180000]  TFA9879: axentia_asoc_tfa9879_hw_params - bclk divider 128
> [  161.190000]  TFA9879: axentia_asoc_tfa9879_hw_params - period 7
> [  161.200000]  TFA9879: axentia_asoc_tfa9879_hw_params - mclk rate
> 66000000
> [  161.200000]  TFA9879: axentia_asoc_tfa9879_hw_params - bclk rate 128000
> [  161.210000]  TFA9879: axentia_asoc_tfa9879_hw_params - bclk divider 128
> [  161.220000]  TFA9879: axentia_asoc_tfa9879_hw_params - period 7
> [  161.230000]  TFA9879: axentia_asoc_tfa9879_hw_params - mclk rate
> 66000000
> [  161.230000]  TFA9879: axentia_asoc_tfa9879_hw_params - bclk rate 256000
> [  161.240000]  TFA9879: axentia_asoc_tfa9879_hw_params - bclk divider 64
> [  161.250000]  TFA9879: axentia_asoc_tfa9879_hw_params - Failed to set cpu
> dai clk divider
> [  161.260000] axentia-tfa9879-audio sound.9: ASoC: machine hw_params
> failed: -16

I did some further tests, and the following program fails without the patch:

#include <sys/ioctl.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/soundcard.h>

int
main(void)
{
	int fd;
	int format;
	int channels;

	if ((fd = open("/dev/dsp", O_WRONLY, 0)) == -1) {
		perror("open");
		return 1;
	}
	format = AFMT_S16_LE;
	if (ioctl(fd, SNDCTL_DSP_SETFMT, &format) == -1) {
		perror("SNDCTL_DSP_SETFMT");
		return 1;
	}
	channels = 2;
	if (ioctl(fd, SNDCTL_DSP_CHANNELS, &channels) == -1) {
		perror("SNDCTL_DSP_CHANNELS");
		return 1;
	}
	return 0;
}

Output:
SNDCTL_DSP_CHANNELS: Device or resource busy

(I admin to having edited the above code slightly in this mail, so I might have
introduced some silly bug, but you get what I mean, just open the device
and request some parameters, and boom: -EBUSY)

Cheers,
Peter

> > So, I think keep the code as is (without this patch applied), what's your
> > opinion?
> 
> In that case we'll simply carry the patch ourselves. I don't know why the
> above
> happens, or if it is caused by bad behavior in the library, or what. But I do
> know
> that the library is unusable for us without some action.
> 
> I thought it was an improvement, and therefore sent the patch...
> 
> Cheers,
> Peter