Bugzilla – Bug 1161152
No sound in qemu openQA since update to kernel 5.4
Last modified: 2020-01-20 11:00:09 UTC
There is no sound in openQA tests within qemu since update to kernel 5.4. Kernel 5.3 was fine.
Related kernel traces:
[ 16.337715] snd_hda_intel 0000:00:01.0: CORB reset timeout#1, CORBRP = 0
[ 16.338620] snd_hda_intel 0000:00:01.0: no codecs found!
Full serial log:
Interesting, I didn't know that qemu emulates the HD-audio device even on arm64.
Does the problem appear not specifically on openQA but generally with qemu and linux-5.4 on arm?
Yes, this is a problem in qemu on aarch64, not only within openQA.
There were some updates on snd_hda_intel between kernel 5.3 and 5.4.
Right, and the question is which change hitting this. I suspect the change of the register accessor to the direct MMIO read/write, but not quite sure.
What's the easiest procedure to reproduce and debug locally, preferably without the dedicated hardware?
The easiest way, would be use JeOS image from:
uncompress it and boot it manually with qemu-system-aarch64.
Info at: https://en.opensuse.org/openSUSE:AArch64#Running_openSUSE_from_raw_image
Do not forget to add '-soundhw hda' option for qemu-system-aarch64.
Thanks, I could reproduce it locally. Will debug / test further later.
(BTW, qemu-uef-aarch64 is missing on Leap 15.1 or TW?)
(In reply to Takashi Iwai from comment #5)
> Thanks, I could reproduce it locally. Will debug / test further later.
> (BTW, qemu-uef-aarch64 is missing on Leap 15.1 or TW?)
It is a known problem. qemu-uefi-aarch64 is only in ARM repos because there is no export to x86_64 repo (built in aarch64 only).
Found out that it's a regression due to the recent simplification of HD-audio helper code. The aligned MMIO access is enforced for ARM, not only for snd-hda-tegra but also for snd-hda-intel, and that broke QEMU by some unknown reason.
I'm going to submit the fix patch to upstream and backport to relevant branches.
Created attachment 827816 [details]
(In reply to Takashi Iwai from comment #7)
> Found out that it's a regression due to the recent simplification of
> HD-audio helper code. The aligned MMIO access is enforced for ARM, not only
> for snd-hda-tegra but also for snd-hda-intel, and that broke QEMU by some
> unknown reason.
> I'm going to submit the fix patch to upstream and backport to relevant
Thanks a lot!
The fix was pushed to stable for-next branch. It'll be merged soon later (hopefully).