Possible rendering/flickering issue with niri after upgrading to Wayland 1.26 on Arch Linux

[mingerfan]

Environment

• OS: Arch Linux
• Kernel: linux-zen 7.1.5-zen1-2-zen
• WM: niri

Description

Yesterday, after upgrading my system to Wayland 1.26, I started experiencing severe screen flickering/rendering glitches while using niri.

Initially, I suspected that the issue might be related to the monitor, niri itself, the GPU, or the kernel.

I tried switching to a different kernel, including the LTS kernel, but the issue was still present. This seems to rule out a kernel-specific problem. I also tested the same system under KDE, where I could not reproduce the issue, so the GPU itself is less likely to be the cause.

Eventually, I rolled back my system using a Btrfs snapshot, and the problem disappeared.

After checking the packages that had been upgraded, I noticed that Wayland had been updated to version 1.26 at around the same time the issue started.

At the moment, my suspicion is that there may be some kind of rendering regression or compatibility issue between Wayland 1.26 and niri. However, I have not yet been able to identify the exact cause, so this still needs further investigation.

Since both niri and Arch Linux are rolling-release projects, I'm posting this here in case anyone else has encountered similar behavior after upgrading to Wayland 1.26.

A photo of the issue is attached below.

What I have tested

• Rebooted the system
• Switched from linux-zen to another kernel
• Tested with the LTS kernel
• Issue still occurred with different kernels
• Tested KDE on the same machine: no issue observed
• Rolled back the system using Btrfs: issue disappeared

[HigherOrderLogic]

If your GPU is AMD, maybe it's an issue with the kernel. There have been known bugs with AMD GPU driver in recent kernel versions.

[YaLTeR]

This looks like the kernel 7.1.x amdgpu bug fixed in 7.1.8

[mingerfan]

yes，maybe this is the root cause

[mingerfan]

After looking into this a bit more, I found something that may be relevant.

linux-zen 7.1.6/7.1.7 and linux-lts 6.18.43 all include the same AMD DCN page-flip change. There is a confirmed race condition associated with this change: the kernel may report a page flip as completed to the compositor before the hardware has actually finished the flip. This can cause the compositor to reuse a buffer that is still being scanned out, resulting in display corruption.

The upstream fix explicitly mentions this kind of “display corruption”. The fix is included in 7.1.8 and 6.18.44, but is not present in the kernel versions where I encountered the issue.

So this may actually be a kernel-side AMD DCN regression rather than an issue introduced by Wayland 1.26 or niri itself. The timing of the Wayland update may have just made it look related initially.