Bug#1112288: amdgpu: GPU hang with 6.12 & 6.16sid, 6.16 liquorix works.

[Salvatore Bonaccorso <carnil@debian.org>]

Hi

On Thu, Aug 28, 2025 at 11:57:39AM +0300, kuu wrote:
> Package: linux-image-amd64
> Version: 6.16.3-1
> Severity: important
> Tags: upstream
> X-Debbugs-Cc: kuu@qcm.org
> 
> Dear Maintainer,
> 
> *** Reporter, please consider answering these questions, where appropriate ***
> 
>    * What led up to the situation?
> The user is having system freezes with their AMD Radeon RX 7600M XT GPU on a
> Debian trixie system. The problem manifests as a full system freeze, typically
> after a few hours or days of uptime. The freeze is preceded by specific errors
> at boot time.
> 
>    * What exactly did you do (or not do) that was effective (or
>      ineffective)?
> 
> Ineffective actions:
> 
>     Using the default Debian trixie kernel (6.12 series). This resulted in a
> system freeze.
> 
>     Upgrading to the latest official unstable (sid) kernel (6.16.3+1) and the
> latest firmware-amd-graphics package. This also resulted in a system freeze.
> 
>     Adding the kernel parameters amdgpu.powerplay=0 and amdgpu.noretry=0 did
> not resolve the issue.
> 
> Effective action:
> 
>     Installing and using the Liquorix kernel (6.16.3-1-liquorix-amd64). This
> kernel works perfectly and has not resulted in any system freezes.
> 
> 
>    * What was the outcome of this action?
> 
> When using the official Debian kernels (6.12 or 6.16), the following errors
> appear at boot:
> 
>     REG_WAIT timeout
> 
>     hub6-0:1.0: config failed, hub doesn't have any ports
> 
> These boot errors serve as a direct precursor to the system freeze that occurs
> later.
> When using the Liquorix kernel (6.16.3-1-liquorix-amd64), these boot errors do
> not appear, and the system remains stable indefinitely.
> 
> 
>    * What outcome did you expect instead?
> 
> The user expected the official Debian 6.16 kernel, paired with the latest
> firmware, to work as reliably as the Liquorix kernel of the same version, and
> for the boot errors and subsequent freezes to be gone. The expectation was that
> the bug would be fixed in a newer kernel version, which was not the case with
> the official build.

A side note: liquorix seems to have a quite patched kernel (with a
single big patch?) I think the best you can do here is to ask the
liquorix developers to forward their fixes so they land in upstream.

Regards,
Salvatore