Audio not working: "snd_hda_intel: No response from codec, resetting bus"

[Claudia]

Audio works fine in Fedora 25-1.3 livecd and codec is detected as
"ALC3234 Analog", alsa version k4.8.6-300.fc25.x86_64. It also works in
F30.

On Qubes 4.0.2-rc2, I can't get audio to work at all. Codec shows up as
"Generic Analog" in `aplay -l`, in alsamixer chip (codec) is "Generic
ffff Generic", and kernel logs show

snd_hda_intel 0000:03:00.1: no codecs initialized
snd_hda_intel 0000:03:00.6: azx_get_response timeout, switching to
polling mode: last cmd=0x000f0000
snd_hda_intel 0000:03:00.6: No response from codec, disabling MSI: last
cmd=0x000f0000
snd_hda_intel 0000:03:00.6: Codec #0 probe error; disabling it...
snd_hda_intel 0000:03:00.6: No response from codec, resetting bus: last
cmd=0x000f0000

... followed by a stack trace.

0000:03:00.6 is the speaker/headphone interface, and .1 is HDMI. I'm not
concerned with HDMI at the moment.

I get a variety of different errors from aplay in dom0, or it just hangs
indefinitely; it seems random. And I obviously can't hear anything
playing in VMs.

ALSA version is k4.19.81-1.pvops.qubes.x86_64

I tried playing around with some modprobe options, such as probe_mask=1,
probe_mask=8, model=auto, and index=1, but none of them get rid of the
errors or cause the codec to be detected as anything other than "Generic
Analog".

I get the same result when booting Qubes with VT-x and VT-d disabled. I
also tried booting the Qubes installer, but it appears the installer
doesn't attempt to load any sound drivers (no success or failure
messages) and it doesn't appear to have alsa-utils.

Any idea why audio would work in Fedora 25 with 4.8.6, but not Qubes
R4.0.2 with 4.18.81?

-------------------------------------------------
This free account was provided by VFEmail.net - report spam to ab...@vfemail.net

ONLY AT VFEmail! - Use our Metadata Mitigator to keep your email out of the NSA's hands!
$24.95 ONETIME Lifetime accounts with Privacy Features!
15GB disk! No bandwidth quotas!
Commercial and Bulk Mail Options!

[Claudia]

(cc xen-users; see original message below)

So I was able to boot Qubes as a regular kernel, without Xen, and the
codec is properly detected as Realtek ALC3234. Same exact kernel, same
commandline, same userspace, just without Xen. (Xen 4.8.5-11.fc25)

Any ideas on what might be causing this, or how to debug further?

Here's some useful information I found on snd_hda_intel:
https://help.ubuntu.com/community/HdaIntelSoundHowto#Playing_with_probe_mask
https://www.kernel.org/doc/html/v4.18/sound/hd-audio/notes.html#codec-probing-problem
https://www.kernel.org/doc/html/v4.18/sound/alsa-configuration.html#module-snd-hda-intel

Note: I found a *lot* of info telling me to set the model= parameter,
however the documentation states that model is specific to the codec
driver, e.g. snd_hda_intel_realtek. The problem here is that
snd_hda_intel cannot even detect the make/model of the codec chip, and
thus the realtek driver doesn't even get loaded under Xen. Just in case,
I tried several model= options, including model=auto, and none of them
had any effect.

[Claudia]

Claudia:

As a workaround, I was able to get Xen to detect the codec and get rid
of the error by using single_cmd=1. However the snd_hda_intel
documentation warns that single_cmd is meant for debugging purposes only
and is generally not recommended.

However, even with the error gone and the codec driver loading now, I
still can't get any sound in Xen, even after reordering the cards,
disabling the HDMI card, trying various model= options, disabling power
save, and trying speaker-test with pulseaudio disabled. At best I get a
quiet static from headphones, and can hear a slight pop when the channel
is unmuted. Additionally, speaker-test and aplay occasionally seem to
hang indefinitely, as if the device is blocking, but not always. Audio
works fine using the same configuration when booted without Xen.

So, to recap what I've found so far:
Fedora 25 live cd: works
Qubes without Xen: works
Qubes with Xen: codec probe error
Qubes with Xen, VT-x & VT-d disabled: codec probe error
Qubes with Xen, single_cmd=1: codec detected, but no sound

I suppose I could try installing F25 with Xen and see if I get sound.
But I'm pretty well convinced it's Xen (not Qubes) at this point.

I tried pci=noacpi but the keyboard wouldn't work, so I couldn't test
sound. The console said something about "couldn't find IRQ ... please
try using pci=biosirq". Exact same result when I replaced it with
acpi=noirq. pci=biosirq had no effect as far as I could tell.

I don't know what else to try at the moment. Any ideas would be highly
appreciated.

[Claudia]

Claudia:
> [...]

> So, to recap what I've found so far:
> Fedora 25 live cd: works
> Qubes without Xen: works
> Qubes with Xen: codec probe error
> Qubes with Xen, VT-x & VT-d disabled: codec probe error
> Qubes with Xen, single_cmd=1: codec detected, but no sound
>

> [...]

Just following up. Audio works flawlessly out of the box in
R4.1-20181013 with kernel 4.19.76-1. Didn't work in R4.0.2 with 4.19 or
even 5.0 kernels even after days of tinkering with it. So it could be
due to the Xen version (4.8 -> 4.12), Fedora userland version (25 ->
29), or changes to qubes-specific code/config (4.0 -> 4.1). I was hoping
to get it working in stable, but I'll probably just keep using 4.1 for now.

[fiftyfour...@gmail.com]

Hi Claudia,

I'm suffering from the same problem as you, and 'dmesg | grep -i snd_hda_intel' brought me to this thread. 

My sound was originally working fine after I updated dom0 and domus (tested via Youtube) but somewhere along the line something broke and the next time I did the same thing, no sound. I did everything in my non-technical power but it's dead, bub. No idea what caused it.

dmesg output in short:

snd_hda_intel [Audio device PCI code thing]: azx_get_response timeout, switching to polling mode: last cmd=0xXXXXXXXXX

snd_hda_intel [Audio device PCI code thing]: No response from codec, resetting bus: last cmd=0xXXXXXXXXX    (this line keeps repeating)

snd_hda_intel [Audio device PCI code thing]: azx_get_response timeout, switching to single_cmd mode: last cmd=0xXXXXXXXXX

snd_hda_codec_realtek hdaudioC0D0: Unable to sync to register 0xXXXXXXX. -5

snd_hda_codc_hdmi hdaudioC0D2: HDMI: invalid ELD buf size   (note I'm not using HDMI)

[54th Parallel]

On Monday, 27 July 2020 at 23:52:20 UTC+8 fiftyfour...@gmail.com wrote:

Hi Claudia,

I'm suffering from the same problem as you, and 'dmesg | grep -i snd_hda_intel' brought me to this thread. 

My sound was originally working fine after I updated dom0 and domus (tested via Youtube) but somewhere along the line something broke and the next time I did the same thing, no sound. I did everything in my non-technical power but it's dead, bub. No idea what caused it.

dmesg output in short:

snd_hda_intel [Audio device PCI code thing]: azx_get_response timeout, switching to polling mode: last cmd=0xXXXXXXXXX

snd_hda_intel [Audio device PCI code thing]: No response from codec, resetting bus: last cmd=0xXXXXXXXXX    (this line keeps repeating)

snd_hda_intel [Audio device PCI code thing]: azx_get_response timeout, switching to single_cmd mode: last cmd=0xXXXXXXXXX

snd_hda_codec_realtek hdaudioC0D0: Unable to sync to register 0xXXXXXXX. -5

snd_hda_codc_hdmi hdaudioC0D2: HDMI: invalid ELD buf size   (note I'm not using HDMI)

Update for the benefit thread readers:

On Dell Inspiron 5593, updating dom0 to kernel 5.7 solved the audio problem (audio starts unreliably after boot--sometimes it works tens of minutes or hours after boot, making it hard to narrow down which actions caused it to start, if any did), but left hibernation and suspension inoperable.  On the current kernel (4.19) the problem for some reason seems to have solved itself. Maybe it's because I don't use audio on that machine near boot (unless for testing).

Oh, and this is important to those who value their hearing--keeping your headphones plugged into the headphone jack when booting on 4.19 leads to loud screeching noises resembling the death throes of autistic parrots suffocating on helium. The sound is apparently correlated with CPU use. Not sure what's going on there,  but my guess is the kernel confused the output with some other channel or there's some leaking somewhere.