2019-10-02 15:55:18

by Matt

[permalink] [raw]
Subject: possible audio regression for usb-audio with 5.3 kernel (5.3.1), sounds like variants of 8-bit audio

Hi Takashi,

there appears to be a sound regression for plug-n-play usb-audio DACs
with 5.3 kernel.

I'm using the following one:

https://www.aliexpress.com/item/33012416525.html?spm=a2g0s.9042311.0.0.4f314c4dv3EF0p

which includes SA9023A + ES9018K2M on the chip.

When listening to parts of the Unreal Tournament 3 soundtrack the
output resembles a bit like:

https://www.youtube.com/watch?v=9gq4R6acnBQ Rush - Tom Sawyer (8-bit
NES audio render)

it immediately made me think of 8-bit audio and the NES.

First thought of misconfiguration in the kernel config but then went
back to 5.2.17-rt9 (copying the config and only enabling full
preemption) and there the output is correct and crystal clear.

For each kernel build I need to jump through a few hoops (nvidia
driver, zfsonlinux modules, luks, genkernel, etc.) so it'll take a
while to build and get it up and running.

System is:
cat /etc/lsb-release
DISTRIB_ID="Gentoo"

~amd64

gcc version 9.2.0 (Gentoo Hardened 9.2.0-r1 p2)
GNU ld (Gentoo 2.32 p2) 2.32.0

PCIe related sound doesn't seem to be affected (Xonar DX), it works
fine both on 5.3.1 and 5.2.17-rt9

Kind Regards

Matthew


2019-10-18 23:18:19

by Takashi Iwai

[permalink] [raw]
Subject: Re: possible audio regression for usb-audio with 5.3 kernel (5.3.1), sounds like variants of 8-bit audio

On Wed, 02 Oct 2019 17:01:29 +0200,
Matt wrote:
>
> Hi Takashi,
>
> there appears to be a sound regression for plug-n-play usb-audio DACs
> with 5.3 kernel.
>
> I'm using the following one:
>
> https://www.aliexpress.com/item/33012416525.html?spm=a2g0s.9042311.0.0.4f314c4dv3EF0p
>
> which includes SA9023A + ES9018K2M on the chip.
>
> When listening to parts of the Unreal Tournament 3 soundtrack the
> output resembles a bit like:
>
> https://www.youtube.com/watch?v=9gq4R6acnBQ Rush - Tom Sawyer (8-bit
> NES audio render)
>
> it immediately made me think of 8-bit audio and the NES.
>
> First thought of misconfiguration in the kernel config but then went
> back to 5.2.17-rt9 (copying the config and only enabling full
> preemption) and there the output is correct and crystal clear.
>
> For each kernel build I need to jump through a few hoops (nvidia
> driver, zfsonlinux modules, luks, genkernel, etc.) so it'll take a
> while to build and get it up and running.

I don't know of a similar regression, and have currently no idea of
possible cause. The git bisection would be the best option to spot
out the cause, I suppose.

For that, it'd be maybe helpful to reduce the reproducible condition,
e.g. try to boot with the stock Linus kernel without Nvidia graphics
and reduced configuration. Basically you don't need the graphics
environment for reproducing the audio problem. This will make the git
bisection much easier.


thanks,

Takashi

>
> System is:
> cat /etc/lsb-release
> DISTRIB_ID="Gentoo"
>
> ~amd64
>
> gcc version 9.2.0 (Gentoo Hardened 9.2.0-r1 p2)
> GNU ld (Gentoo 2.32 p2) 2.32.0
>
> PCIe related sound doesn't seem to be affected (Xonar DX), it works
> fine both on 5.3.1 and 5.2.17-rt9
>
> Kind Regards
>
> Matthew
>