https://bugs.kde.org/show_bug.cgi?id=517716
Bug ID: 517716
Summary: After upgrading to Plasma 6.6.0 (Arch Linux), all
X11-based 3D/Vulkan/OpenGL applications (vkcube --wsi
xlib, glxinfo -B on GBM/X11 platform, any app using
glamor/DRI3) fall back to llvmpipe software rendering.
Native Wayland apps and KWin itself continue
Classification: Plasma
Product: kwin
Version First 6.6.0
Reported In:
Platform: Arch Linux
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: egl
Assignee: [email protected]
Reporter: [email protected]
Target Milestone: ---
SUMMARY
After upgrading to Plasma 6.6.0 (Arch Linux), all X11-based 3D/Vulkan/OpenGL
applications (vkcube --wsi xlib, glxinfo -B on GBM/X11 platform, any app using
glamor/DRI3) fall back to llvmpipe software rendering. Native Wayland apps and
KWin itself continue to use full hardware acceleration on the AMD Radeon RX
5700 XT.
The exact error in kwin_wayland_wrapper:
_amdgpu_device_initialize: amdgpu_query_info(ACCEL_WORKING) failed (-13)
amdgpu: amdgpu_device_initialize failed.
This is a regression introduced exactly by kwin-6.5.5-3 → kwin-6.6.0-1. All
other packages (Mesa 25.3.5, kernel 6.18.9, libdrm, plasma-workspace, etc.)
stayed the same or were also upgraded to 6.6, but downgrading **only kwin** to
6.5.5-3 immediately restores hardware-accelerated XWayland/glamor/GBM (glxinfo
shows AMD renderer, vkcube --wsi xlib works).
STEPS TO REPRODUCE
1. Arch Linux with RX 5700 XT (Navi10), Mesa 25.3.5, Plasma 6.6.0.
2. Log in via Wayland (direct TTY + startplasma-wayland or any display
manager).
3. Run: eglinfo -B (GBM platform shows llvmpipe)
4. Run: vkcube --wsi xlib (crashes or software)
OBSERVED RESULT
XWayland/glamor/GBM path falls back to llvmpipe because DRM_IOCTL_AMDGPU_INFO
returns EACCES (-13) inside kwin_wayland_wrapper.
EXPECTED RESULT
X11 apps get full hardware acceleration (AMD Radeon RX 5700 XT, radeonsi, same
as on Plasma 6.5).
SOFTWARE/OS VERSIONS
- GPU: AMD Radeon RX 5700 XT (Navi10)
- Mesa: 25.3.5-arch1.1 (unchanged)
- Kernel: 6.18.9-arch1-2
- Session: Plasma Wayland (XDG_SESSION_TYPE=wayland)
- Device nodes:
crw-rw----+ root video 226,1 card1 (explicit user:isaac:rw- ACL)
crw-rw-rw- root render 226,128 renderD128
- User is in video + render groups.
- Happens with SDDM, plasma-login-manager, AND direct TTY startplasma-wayland
(rules out display manager).
ADDITIONAL INFORMATION
**Bisected cause**
Test: everything upgraded to 6.6 EXCEPT kwin downgraded to 6.5.5-3 → problem
disappears immediately.
Therefore the regression is in kwin 6.6 (specifically the new/refactored
kwin_wayland_wrapper XWayland handling).
- Discuss thread with same symptoms:
https://discuss.kde.org/t/failed-to-use-gpu-render-after-upgrade-from-6-5-to-6-6/45349
I am trying to build from the Kwin source, and check via `env
QT_PLUGIN_PATH="$(pwd)":"$QT_PLUGIN_PATH" dbus-run-session ./kwin_wayland
--xwayland "vkcube --wsi xlib"`, v6.5.0, v6.5.5. v6.5.6 all passed, but v6.6.0
failed. I try to git bisect with
```
Bashgit bisect start
git bisect bad v6.6.0
git bisect good v6.5.6
```
and then
```
Bisecting: a merge base must be tested
[67917b995e27148c71202e268450be1e2a6d3d60] Update version for new release
6.4.90
➜ 12:22AM isaac bin (67917b995e) ✗ git bisect bad
The merge base 67917b995e27148c71202e268450be1e2a6d3d60 is bad.
This means the bug has been fixed between
67917b995e27148c71202e268450be1e2a6d3d60 and
[8276c53534a41b3dbd1a55b92026c1276f8a2946].
```
I am willing to test on any potential problematic commit if it is helpful to
identify the cause.
--
You are receiving this mail because:
You are watching all bug changes.
