Clemens, thanks for your help.

I am using Fedora 21 (Ubuntu 14.04.1 has the same problem).

'rmmod snd-hda-intel' shows this module is in use too.

'modinfo snd-hda-intel', nor file /lib/modules/.../modules.dep shows that any kernel modules depends on snd-hda-intel or snd_hda_codec_hdmi.

Is it possible that a kernel thread is using these 2 modules?

Any ideas? Thanks.

On Tue, 2014-12-23 at 17:21 +0100, Clemens Ladisch wrote:

Xuebing Wang wrote:

...

rmmod: ERROR: Module snd_hda_codec_hdmi is in use

rmmod snd-hda-intel first.

Regards, Clemens

On Wed, 2014-12-24 at 11:57 +0200, Anssi Hannula wrote:

24.12.2014, 11:49, Xuebing Wang kirjoitti:

...

On Tue, 2014-12-23 at 17:21 +0100, Clemens Ladisch wrote:

...

Xuebing Wang wrote:

...

rmmod: ERROR: Module snd_hda_codec_hdmi is in use

rmmod snd-hda-intel first.

Clemens, thanks for your help.

I am using Fedora 21 (Ubuntu 14.04.1 has the same problem).

'rmmod snd-hda-intel' shows this module is in use too.

'modinfo snd-hda-intel', nor file /lib/modules/.../modules.dep shows that any kernel modules depends on snd-hda-intel or snd_hda_codec_hdmi.

Is it possible that a kernel thread is using these 2 modules?

Any ideas? Thanks.

Run this to see if some userspace process is using the ALSA devices: fuser -v /dev/snd/*

Anssi, thanks for your help.

1) In Fedora 21 (kernel 3.17.4), 'fuser -v /dev/snd/*' shows below: /dev/snd/controlC0: xbing 2400 F.... pulseaudio /dev/snd/controlC1: xbing 2400 F.... pulseaudio /dev/snd/pcmC0D3p: xbing 2400 F...m pulseaudio /dev/snd/pcmC1D0c: xbing 2400 F...m pulseaudio /dev/snd/pcmC1D0p: xbing 2400 F...m pulseaudio

After 'pulseaudio --kill' to kill pulseaudio (after disable autospawn), lsmod shows refcount of snd_hda_intel is down from 7 to 2 as below: $ lsmod | grep hda_intel snd_hda_intel 30479 2

'rmmod snd_hda_intel' still shows this module is in use.

2) But, in Ubuntu 14.04.1 (kernel 3.13.0) DOES work after 'pulseaudio --kill', because lsmod shows refcount of snd_hda_intel is 5.

Any suggestions? Could it be because of kernel difference 3.17.4 vs 3.13.0?

Thanks.

On 三, 2014-12-24 at 21:28 +0900, Takashi Sakamoto wrote:

On Dec 24 2014 20:54, Xuebing Wang wrote:

...

Anssi, thanks for your help.

1. In Fedora 21 (kernel 3.17.4),

'fuser -v /dev/snd/*' shows below: /dev/snd/controlC0: xbing 2400 F.... pulseaudio /dev/snd/controlC1: xbing 2400 F.... pulseaudio /dev/snd/pcmC0D3p: xbing 2400 F...m pulseaudio /dev/snd/pcmC1D0c: xbing 2400 F...m pulseaudio /dev/snd/pcmC1D0p: xbing 2400 F...m pulseaudio

After 'pulseaudio --kill' to kill pulseaudio (after disable autospawn), lsmod shows refcount of snd_hda_intel is down from 7 to 2 as below: $ lsmod | grep hda_intel snd_hda_intel 30479 2

'rmmod snd_hda_intel' still shows this module is in use.

2. But, in Ubuntu 14.04.1 (kernel 3.13.0) DOES work after 'pulseaudio --kill', because lsmod shows refcount of snd_hda_intel is 5.

Any suggestions? Could it be because of kernel difference 3.17.4 vs 3.13.0?

In Ubuntu, pulseaudio process re-run even if you kill the process. See: https://wiki.ubuntu.com/PulseAudio/Log

But in this case, you don't need to kill pulseaudio. You just set 'off' for card profile, or unload alsa-card module. Then these character devices are released.

The way to set card profile: $ pactl list cards | grep Card Card #3 $ pactl set-card-profile 3 off

Or use 'configuration' tab in pavucontrol GUI. In Ubuntu, this application is released by 'pavucontrol' package.

The way to unload alsa-card module: $ pactl list modules | grep alsa (... you can see module-alsa-card is loaded.) $ pactl unload-module module-alsa-card $ pactl list modules | grep alsa (... you can see module-alsa-card is unloaded.)

Sakamoto-san, thank you very much for your sophisticated way, it helps.

1) I am sorry that I forgot to mention that I disabled 'pulseaudio autospawn' in Ubuntu too, as I disabled 'pulseaudio autospawn' in Fedora. Both simply 'pulseaudio --kill' (after disable autospawn) and your sophisticated way can get down refcount of snd_hda_intel from 5 to 0, then I can rmmod snd_hda_intel.

2) However, in Fedora, 'pulseaudio --kill' or your way can get refcount of snd_hda_intel from 7 to 2 (5 decreased). Any suggestions how to decrease the rest 2 refcount?

3) I did try to build tag 3.13 from mainline kernel (using Fedora 3.17.4 config file) and tried 3.13 kernel on Fedora, and got the same result (refcount of snd_hda_intel is down from 7 to 2).

It seems something in Fedora (either Fedora kernel or application) is using snd_hda_intel.

Thanks again.

Regards

Takashi Sakamoto o-takashi@sakamocchi.jp

On Wed, 2014-12-24 at 22:21 +0900, Takashi Sakamoto wrote:

On Dec 24 2014 22:07, Xuebing Wang wrote:

...

2. However, in Fedora, 'pulseaudio --kill' or your way can get refcount

of snd_hda_intel from 7 to 2 (5 decreased). Any suggestions how to decrease the rest 2 refcount?

3. I did try to build tag 3.13 from mainline kernel (using Fedora 3.17.4

config file) and tried 3.13 kernel on Fedora, and got the same result (refcount of snd_hda_intel is down from 7 to 2).

Hm. In this case that the refcount is down to 2, how is the output of 'fuser' command? If any applications use these character devices, we can see the name of command.

Sakamoto-san, thanks for your help.

I found out that it is alsactl in Fedora. After kill alsactl, refcnt is now down to 0, and both snd_hda_intel and snd_hda_codec_hdmi can be rmmod'ed. 'fuser -v /dev/snd' does not reveal alsactl.

Thanks again.

Regards

Takashi Sakamoto o-takashi@sakamocchi.jp