ALSA 1.0.25 broke AC3 passthrough?

ALSA 1.0.25 broke AC3 passthrough?

[Jean-Yves Avenard]

Hi

Today, I installed 1.0.25 drivers and libraries...
AC3 passthrough is completely broken.

My hardware: NVidia GT430, hdmi output
Device used:
hdmi:CARD=NVidia,DEV=0,AES0=6,AES1=0x82,AES2=0x00,AES3=0x01
on Ubuntu 10.04 LTS (but using manually built alsa)

It makes the application hangs or behaves incorrectly. All the time
broken, no matter which app.
Reverting back to 1.0.24 fixes the problem

I haven't delved much into it yet,
MythTV (I am the audio developer) after opening the audio device in
passthrough mode, trying to write audio on the pcm just hangs and
result in an I/O error after a little while
snd_pcm_writei returns an error (not immediately)
snd_pcm_state(pcm_handle) returns 3, snd_strerror(err) returns
"Input/output error"

mplayer just hangs...

nothing obvious in the logs, audio device opens okay, i can't say I
notice much difference in opening the device, allocating fragment and
the lot.

Anyone else has seen this?

Jean-Yves

Re: ALSA 1.0.25 broke AC3 passthrough?

[Stephen Warren]

On 02/06/2012 04:38 AM, Jean-Yves Avenard wrote:
> Today, I installed 1.0.25 drivers and libraries...
> AC3 passthrough is completely broken.
> 
> My hardware: NVidia GT430, hdmi output
> Device used:
> hdmi:CARD=NVidia,DEV=0,AES0=6,AES1=0x82,AES2=0x00,AES3=0x01
> on Ubuntu 10.04 LTS (but using manually built alsa)

Are you able to regress this to a smaller window? I did do some rework a
while back on the HDA driver to support newer GeForce GPUs' audio HW
better, and this potentially had impact on bitstream pass-through,
although it seemed to work at the time. Commits were:

384a48d ALSA: hda: HDMI: Support codecs with fewer cvts than pins
2def817 ALSA: hda: hdmi_eld_update_pcm_info: update a stream in place
3aaf898 ALSA: hda: Separate generic and non-generic implementations
74b654c ALSA: hda: Virtualize SPDIF out controls
7c93597 ALSA: hda: Allow multple SPDIF controls per codec

-- 
nvpublic

Re: ALSA 1.0.25 broke AC3 passthrough?

[Jean-Yves Avenard]

[Re-send to include distribution list]

Hi

On 7 February 2012 16:46, Stephen Warren <swarren@nvidia.com> wrote:

> Are you able to regress this to a smaller window? I did do some rework a
> while back on the HDA driver to support newer GeForce GPUs' audio HW
> better, and this potentially had impact on bitstream pass-through,
> although it seemed to work at the time. Commits were:

I'm not sure what you mean by regress to a smaller window.

I use a quite "oldish" version of nvidia drivers: 260.19.36 (though
same issue with latest 290.10).

At least one other mythtv user has reported the same problem with
1.0.25, not sure what his video/audio card is ...

>
> 384a48d ALSA: hda: HDMI: Support codecs with fewer cvts than pins
> 2def817 ALSA: hda: hdmi_eld_update_pcm_info: update a stream in place
> 3aaf898 ALSA: hda: Separate generic and non-generic implementations
> 74b654c ALSA: hda: Virtualize SPDIF out controls
> 7c93597 ALSA: hda: Allow multple SPDIF controls per codec

Do you want me to compile alsa back to those release and see if
anything makes any difference?

Can certainly do so...

Jean-Yves

Re: ALSA 1.0.25 broke AC3 passthrough?

[Stephen Warren]

On 02/09/2012 03:01 AM, Jean-Yves Avenard wrote:
> [Re-send to include distribution list]
> 
> Hi
> 
> On 7 February 2012 16:46, Stephen Warren <swarren@nvidia.com> wrote:
> 
>> Are you able to regress this to a smaller window? I did do some rework a
>> while back on the HDA driver to support newer GeForce GPUs' audio HW
>> better, and this potentially had impact on bitstream pass-through,
>> although it seemed to work at the time. Commits were:
> 
> I'm not sure what you mean by regress to a smaller window.

You'd mentioned that this problem was present in ALSA 1.0.25, but wasn't
in some earlier release. Potentially a lot changed between those two
releases. It'd be nice if you could try compiling a number of
intermediate versions of the ALSA driver to try and more precisely
locate the change that caused the issue. "git bisect" on the ALSA driver
repository might be useful here.

-- 
nvpublic

Re: ALSA 1.0.25 broke AC3 passthrough?

[Jean-Yves Avenard]

Hi

On 10 February 2012 15:09, Stephen Warren <swarren@nvidia.com> wrote:

> You'd mentioned that this problem was present in ALSA 1.0.25, but wasn't
> in some earlier release. Potentially a lot changed between those two
> releases. It'd be nice if you could try compiling a number of
> intermediate versions of the ALSA driver to try and more precisely
> locate the change that caused the issue. "git bisect" on the ALSA driver
> repository might be useful here.

Ok...

This is interesting..

Compiling from git, using tag v1.0.25, the problem is there...
Compiling master, it's not

However, the issue doesn't show with v1.0.25, when the ALSA hardware
buffer is set to its default 64kB.
If I set it to a higher value (echo 4096 | sudo tee
/proc/asound/card1/pcm3p/sub0/prealloc) ; then it shows up.

so obviously the issue has been fixed between the time 1.0.25 was
released, and now.

I will try to identify which change between 1.0.24 and 1.0.25 broke it
however... Which would lead precisely to which change in master fixed
it. So a 1.0.25.2 can be released soon fixing that problem..

JY