https://bugs.kde.org/show_bug.cgi?id=525029
Bug ID: 525029
Summary: Plasma Wayland loses all outputs on NVIDIA RTX 3060 –
GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT
Classification: Plasma
Product: kwin
Version First 6.7.4
Reported In:
Platform: CachyOS
OS: Linux
Status: REPORTED
Severity: crash
Priority: NOR
Component: wayland-generic
Assignee: [email protected]
Reporter: [email protected]
Target Milestone: ---
On an installed CachyOS system, Plasma Wayland consistently loses signal to all
monitors immediately after login. The display/login manager works normally, but
after starting the Plasma Wayland session both monitors report no signal.
KWin logs repeated OpenGL/framebuffer errors and then fails to configure any
outputs:
kwin_wayland: 0x502: GL_INVALID_OPERATION error generated. and are incompatible
kwin_wayland: Invalid framebuffer status:
“GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT”
kwin_wayland: Applying output configuration failed!
kcminit_startup: There are no outputs - creating placeholder screen
ksplashqml: There are no outputs - creating placeholder screen
The kernel log shows no corresponding NVIDIA Xid, GPU fault or timeout.
Important additional observation:
X11 is much more usable and does not fail immediately at login, but the user
has now also experienced at least one random display-signal loss while running
X11. Therefore the intermittent signal-loss problem may be broader than the
reproducible KWin/Wayland login failure, and the two symptoms should not be
assumed to have the same cause.
Hardware / system
GPU: NVIDIA GeForce RTX 3060 (GA106, PCI ID 10de:2503)
CPU: AMD Ryzen 7 5700X3D
RAM: 32 GiB
Two monitors: primary 240 Hz, secondary 120 Hz
CachyOS
KDE Plasma / KWin 6.7.4 series
Normal/current kernel: linux-cachyos 7.2.2-1
Current NVIDIA driver tested: 610.57.04
NVIDIA open kernel modules
Wayland reproduction
Boot normally.
Login/greeter screen works.
Select Plasma (Wayland).
Login.
KWin starts.
Both displays lose signal almost immediately.
KWin logs repeated GL_INVALID_OPERATION /
GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT errors.
KWin reports: Applying output configuration failed!
Plasma reports that there are no outputs and creates a placeholder screen.
During the failure the displays have no signal and SSH has also become
inaccessible, so logs were collected after reboot from the failed boot using
journalctl -b -1.
Relevant Wayland log
Started KDE Wayland Compositor
kwin_wayland: No backend specified, automatically choosing drm
followed by repeated:
kwin_wayland: 0x502: GL_INVALID_OPERATION error generated. and are incompatible
kwin_wayland: Invalid framebuffer status:
“GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT”
and then:
kwin_wayland: Applying output configuration failed!
kcminit_startup: There are no outputs - creating placeholder screen
ksplashqml: There are no outputs - creating placeholder screen
The framebuffer errors then continue and output configuration fails again.
Kernel / NVIDIA behaviour
With NVIDIA 610.57.04, the open kernel module initializes normally:
NVRM: loading NVIDIA UNIX Open Kernel Module for x86_64 610.57.04
nvidia-modeset: Loading NVIDIA UNIX Open Kernel Mode Setting Driver
[drm] [nvidia-drm] Loading driver
[drm] Initialized nvidia-drm
Kernel command line includes:
nvidia-drm.modeset=1
nvidia-drm.fbdev=1
No NVIDIA Xid, GPU fault or timeout was found corresponding to the immediate
Wayland failure.
Tests already performed
The immediate Plasma Wayland failure has persisted across extensive A/B
testing, including:
NVIDIA 610.57.04
NVIDIA 595.71.05
NVIDIA 580.xx
linux-cachyos 7.2.x
linux-cachyos-lts 6.18.42
KWin 6.7.4-1 and later/current 6.7.4 builds
Mesa 26.1.6 and 26.2.1
One monitor connected instead of two
Both monitors tested individually
Lower refresh rates (120/144 Hz) instead of 240 Hz
Adaptive Sync disabled
HDR disabled
Fresh KWin output configuration (~/.config/kwinoutputconfig.json moved
away)
KWIN_DRM_PREFER_COLOR_DEPTH=24
CachyOS prebuilt linux-cachyos-nvidia-open kernel modules instead of
relying on nvidia-open-dkms
NVIDIA 595 was successfully tested with the 6.18.42 LTS kernel: nvidia-smi
worked normally, but Plasma Wayland still lost all displays immediately after
login.
The user later also tested NVIDIA 580.xx on recommendation from CachyOS
support/Discord. The immediate Wayland failure remained unchanged.
The prebuilt CachyOS NVIDIA-open module path was also tried and did not resolve
the problem.
Useful comparison: EndeavourOS live environment
The same physical computer, GPU and monitors were booted into an EndeavourOS
live environment.
The user switched the live session to Wayland and confirmed:
echo $XDG_SESSION_TYPE
wayland
The Wayland live session worked normally.
This makes a basic hardware incompatibility with Wayland less likely and
suggests that something specific to the installed software/configuration stack
is involved.
X11 observation
Earlier testing showed X11 working while Wayland failed immediately after
login, making X11 the practical fallback.
However, after further use the user reported a random loss of display signal
under X11 as well.
This X11 event differs from the deterministic Wayland failure:
Wayland: displays lose signal immediately after login, reproducibly.
X11: normally usable, but at least one later random signal-loss event
occurred.
Physical display/power/GPU/DP/HDMI connections were unplugged and reseated, but
the issue subsequently occurred again.
This additional X11 symptom may or may not share the same root cause as the
KWin/Wayland framebuffer failure.
Current assessment
The most reproducible and well-logged failure is still the Plasma Wayland
startup path:
KWin starts
→ repeated GL_INVALID_OPERATION
→ GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT
→ Applying output configuration failed!
→ no usable outputs
→ displays lose signal
There is no corresponding NVIDIA Xid/GPU fault/timeout in the captured kernel
log.
Because the problem persists across NVIDIA 580, 595 and 610, multiple kernels,
KWin/Mesa versions, display configurations, color-depth testing and both
DKMS/prebuilt NVIDIA-open module approaches, further random package downgrades
do not appear useful.
The later intermittent X11 signal loss should be noted separately and means
this report should not claim that KWin alone necessarily explains every
display-loss event.
A very similar NVIDIA/KWin Wayland symptom has been discussed in NVIDIA
open-gpu-kernel-modules issue #1008, including
GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT and black output without a useful Xid in
some cases.
Suggested upstream destination
https://github.com/NVIDIA/open-gpu-kernel-modules/issues/1008
--
You are receiving this mail because:
You are watching all bug changes.
