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-
> [email protected]'; [email protected]
> 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 <[email protected]>
> > >>> 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
Hi Peter,
On 10/21/2014 09:05 PM, Peter Rosin wrote:
> I did some further tests, and the following program fails without the patch:
With the patch, it is OK?
> #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
This return from codec or from atmel_ssc_dai?
> (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
>
Best Regards,
Bo Shen
Hi!
> Hi Peter,
>
> On 10/21/2014 09:05 PM, Peter Rosin wrote:
> > I did some further tests, and the following program fails without the patch:
>
> With the patch, it is OK?
Yes.
> > #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
>
> This return from codec or from atmel_ssc_dai?
This -EBUSY definitely comes from atmel_ssc_set_dai_sysclk, when my
card-driver tries to set ATMEL_SSC_CMR_DIV. With the patch, it works.
(the codec is spdif-transmitter, since the i2c interface of the actual tfa9879
codec is not directly reachable from the linux cpu, but that has nothing to
do with this issue).
> > (I admin to having edited the above code slightly in this mail, so I
s/admin/admit/
> > 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
Hi Peter,
On 10/22/2014 12:47 PM, Peter Rosin wrote:
> Hi!
>
>> Hi Peter,
>>
>> On 10/21/2014 09:05 PM, Peter Rosin wrote:
>>> I did some further tests, and the following program fails without the patch:
>>
>> With the patch, it is OK?
>
> Yes.
>
>>> #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
>>
>> This return from codec or from atmel_ssc_dai?
>
> This -EBUSY definitely comes from atmel_ssc_set_dai_sysclk, when my
> card-driver tries to set ATMEL_SSC_CMR_DIV. With the patch, it works.
> (the codec is spdif-transmitter, since the i2c interface of the actual tfa9879
> codec is not directly reachable from the linux cpu, but that has nothing to
> do with this issue).
I try to reproduce it (using the code your pasted directly) on atmel
sama5d3xek with wm8904 code, don't meet this error.
I also go through the OSS code, I still don't find this is related with
atmel_ssc_set_dai_sysclk.
So, am I missing something or something else?
>>> (I admin to having edited the above code slightly in this mail, so I
> s/admin/admit/
>>> 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
>
Best Regards,
Bo Shen
Hi!
> >> With the patch, it is OK?
> >
> > Yes.
> >
> >>> #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
> >>
> >> This return from codec or from atmel_ssc_dai?
> >
> > This -EBUSY definitely comes from atmel_ssc_set_dai_sysclk, when my
> > card-driver tries to set ATMEL_SSC_CMR_DIV. With the patch, it works.
> > (the codec is spdif-transmitter, since the i2c interface of the actual
> > tfa9879 codec is not directly reachable from the linux cpu, but that
> > has nothing to do with this issue).
>
> I try to reproduce it (using the code your pasted directly) on atmel
> sama5d3xek with wm8904 code, don't meet this error.
>
> I also go through the OSS code, I still don't find this is related with
> atmel_ssc_set_dai_sysclk.
>
> So, am I missing something or something else?
The sama5d3xek/wm9804 combo, as implemented in the kernel, has the
ssc dai in slave mode, and therefore don't need to fiddle with any
ssc dai dividers (atmel_9804.c :atmel_asoc_wm9804_hw_params() only
sets things in the wm9804 codec dai driver and leaves the ssc dai to itself).
Instead, try the above code on your code with the ssc dai in master mode
that you pointed at previously.
https://github.com/Android4SAM/linux-at91/commit/33db8ebd3e75632c482dda271340f4d2adcfd320
If that happens to not hit -EBUSY (which it might not, since the wm9804 codec
will only allow stereo, so the SNDCTL_DSP_CHANNELS ioctl might not need
to make any change for any ssc divider) add code to also set a non-default
sample rate, e.g.:
speed = 22050;
if (ioctl(fd, SNDCTL_DSP_SPEED, &speed) == -1) {
perror("SNDCTL_DSP_SPEED");
return 1;
}
Cheers,
Peter
Hi Peter,
On 10/22/2014 04:33 PM, Peter Rosin wrote:
> The sama5d3xek/wm9804 combo, as implemented in the kernel, has the
> ssc dai in slave mode, and therefore don't need to fiddle with any
> ssc dai dividers (atmel_9804.c :atmel_asoc_wm9804_hw_params() only
> sets things in the wm9804 codec dai driver and leaves the ssc dai to itself).
>
> Instead, try the above code on your code with the ssc dai in master mode
> that you pointed at previously.
Yes, I try the ssc working as master mode.
> https://github.com/Android4SAM/linux-at91/commit/33db8ebd3e75632c482dda271340f4d2adcfd320
>
> If that happens to not hit -EBUSY (which it might not, since the wm9804 codec
> will only allow stereo, so the SNDCTL_DSP_CHANNELS ioctl might not need
> to make any change for any ssc divider) add code to also set a non-default
> sample rate, e.g.:
>
> speed = 22050;
> if (ioctl(fd, SNDCTL_DSP_SPEED, &speed) == -1) {
> perror("SNDCTL_DSP_SPEED");
> return 1;
> }
with this piece of code, I reproduce your issue.
Now, I know the reason of this issue, work in oss mode, it will set the
default clock to 8KHz, and then if change to other sample rate, for
example 48KHz, the div is different, then it reports -EBUSY.
So, I think we won't change the ATMEL_SSC_CMR_DIV to ATMEL_SSC_TCMR_DIV
and ATMEL_SSC_RCMR_DIV, as it will affect other users. We just deal with
this situation in ATMEL_SSC_CMR_DIV block, check the direction, if the
same direction change the div, then accept the change, otherwise, return
-EBUSY.
>
> Cheers,
> Peter
Best Regards,
Bo Shen
Bo Chen wrote:
> with this piece of code, I reproduce your issue.
>
> Now, I know the reason of this issue, work in oss mode, it will set the default
> clock to 8KHz, and then if change to other sample rate, for example 48KHz,
> the div is different, then it reports -EBUSY.
Indeed.
> So, I think we won't change the ATMEL_SSC_CMR_DIV to
> ATMEL_SSC_TCMR_DIV and ATMEL_SSC_RCMR_DIV, as it will affect other
> users. We just deal with this situation in ATMEL_SSC_CMR_DIV block, check
> the direction, if the same direction change the div, then accept the change,
> otherwise, return -EBUSY.
Ok.
But I'm not sure it is possible to dig out the current direction in the
.set_clkdiv callback? Perhaps the correct fix is to set the bits
.symmetric_rates, .symmetric_channels and .symmetric_samplebits
in the atmel_ssc_dai struct when the ssc dai is master? Then I
expect that other mechanisms will kick in that will render the current
CMR_DIV check pointless?
Is a dai driver allowed to change these symmetry bits after registration?
Can they be set in the .set_sysclk callback? Perhaps in the
ATMEL_SSC_CMR_DIV block itself? That callback should only be
called when the dai is master, so that would be perfect...
Yes, the limitation would be a little bit more strict than today, but is it
really common to require different modes on playback and capture?
Cheers,
Peter