SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

gen2arch · 2025-11-30 10:40:53

Hi,

since last update I get uninterrupted errors in journal, every GUI action (e.g. switching workspace) triggers a new error:

kwin_wayland[1620]: 0x500: GL_INVALID_ENUM error generated. Invalid <face>.
kwin_wayland[1620]: Invalid framebuffer status:  "GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT"
kwin_wayland[1620]: 0x502: GL_INVALID_OPERATION error generated. <image> and <target> are incompatible
kwin_wayland[1620]: Invalid framebuffer status:  "GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT"
kwin_wayland[1620]: Failed to create framebuffer: Invalid argument
kwin_wayland[1620]: Failed to create framebuffer: Invalid argument
kwin_wayland[1620]: 0x500: GL_INVALID_ENUM error generated. Invalid <face>.

Display manager is SDDM, in kernel cmd line we have:

nvidia_drm.modeset=1 nvidia.NVreg_PreserveVideoMemoryAllocations=1

There seems to be no degradation of functionality, at least no immediately evident one; just the errors in the log.

This Arch install is on a notebook; here is some hw and Graphics stack version info:

inxi -Fxz

System:
Kernel: 6.17.9-zen1-1-zen arch: x86_64 bits: 64 compiler: gcc v: 15.2.1
Desktop: KDE Plasma v: 6.5.3 Distro: Arch Linux
CPU:
Info: 24-core (8-mt/16-st) model: 13th Gen Intel Core i9-13980HX bits: 64
type: MST AMCP arch: Raptor Lake rev: 1 cache: L1: 2.1 MiB L2: 32 MiB
L3: 36 MiB
Graphics:
Device-1: Intel Raptor Lake-S UHD Graphics vendor: Micro-Star MSI
driver: i915 v: kernel arch: Xe bus-ID: 0000:00:02.0
Device-2: NVIDIA AD103M / GN21-X11 [GeForce RTX 4090 Laptop GPU]
vendor: Micro-Star MSI driver: nvidia v: 580.105.08 arch: Lovelace
bus-ID: 0000:01:00.0
Display: wayland server: X.org v: 1.21.1.21 with: Xwayland v: 24.1.9
compositor: kwin_wayland driver:
gpu: i915,nv_platform,nvidia,nvidia-nvswitch resolution: 1: 5120x2880~60Hz
2: 3840x2160
API: EGL v: 1.5 drivers: iris,nvidia,swrast platforms:
active: gbm,wayland,x11,surfaceless,device inactive: device-2
API: OpenGL v: 4.6.0 compat-v: 4.5 vendor: intel mesa v: 25.2.7-arch1.1
glx-v: 1.4 direct-render: yes renderer: Mesa Intel Graphics (RPL-S)
API: Vulkan v: 1.4.328 drivers: nvidia surfaces: N/A devices: 1
Info: Tools: api: clinfo, eglinfo, glxinfo, vulkaninfo
de: kscreen-console,kscreen-doctor gpu: nvidia-smi wl: wayland-info
x11: xdpyinfo, xprop, xrandr
Info:
Memory: total: 128 GiB note: est. available: 125.49 GiB
used: 6.58 GiB (5.2%)
Processes: 513 Uptime: 2h 5m Init: systemd
Packages: 1523 Compilers: clang: 21.1.6 gcc: 15.2.1 Shell: Zsh v: 5.9
inxi: 3.3.40

Thanks.

Last edited by gen2arch (2025-12-02 13:57:59)

gen2arch · 2025-11-30 13:00:28

@admin: Sorry for the noise, but would this post better go in "Applications & Desktop Environments"?

seth · 2025-11-30 18:32:32

You can https://bbs.archlinux.org/misc.php?report=2275308 your post if you want it to be moved.
Which versions where actually changed?
The errors show up in https://bbs.archlinux.org/viewtopic.php?id=303829 and quite some other google hits, doesn't seem to be a particularly new thing.
Any chance you've a HDR monitor?

gen2arch · 2025-12-01 03:15:51

seth wrote:

Any chance you've a HDR monitor?

In fact I do: the 4k built-in screen as well as the external (5k) one are HDR. How might this affect the error?

seth · 2025-12-01 08:26:34

HDR showed up a lot along those errors, I guess the conversion causes "<image> and <target> are incompatible"
=> what if you disable it, https://wiki.archlinux.org/title/KDE#HDR ?

gen2arch · 2025-12-02 13:52:34

seth: sorry for the delayed reply. took a while to sort it all out, but it is solved now:

1. On my machine the errors in the journal are independent of HDR: HDR active on internal display > no error; external *non-HDR* display attached, the error returns!
2. Main reasons seems to be a combination of Multi-GPU + Multi-monitor setup.

BTW: exactly matching Bug report!
https://bugs.kde.org/show_bug.cgi?id=511852

Solution:

1. identify the GPU enumeration on your machine:

lspci -k | grep -A3 -E "VGA|3D"

compare with output from

ls -l /dev/dri/by-path

Mine has: "card1" = Nividia, "card2"=Intel.

2. tell KWin in which order to prefer DRM devices for displays / compositing. I want Nvidia="card1" to be used preferably. The goal here is to keep compositor and scan-out on one GPU, or at least a path the driver likes better: define a systemd user variable:

systemctl --user edit plasma-kwin_wayland.service

with this content:

[Service]
# Prefer NVIDIA (card1) over Intel (card2) for KWin because of external-monitor bug
Environment=KWIN_DRM_DEVICES=/dev/dri/card1:/dev/dri/card2

The corresponding file is: "~/.config/systemd/user/plasma-kwin_wayland.service.d/override.conf"
Then: systemctl --user daemon-reload; logout, re-login.
Errors in the journal are gone.
If I swap preferences (Intel first): errors return.
Here is an explanation GTP5.1 Thinking was offering:

The more plausible picture now:
Before: KWin probably defaulted to Intel as primary, but external 5K was hanging off NVIDIA → KWin had to do cross-GPU buffer sharing (render on Intel → scan out via NVIDIA) → that’s where the buggy FBO/scanout path got hit and spammed the logs.
Now: KWin renders and scans out on NVIDIA (for the external screen at least), so the weird cross-GPU path is no longer used → no more GL/FB errors.
This matches your experiments perfectly: the problem appears only when an external monitor is active (very likely on the NVIDIA side), and vanishes when you force KWin to treat NVIDIA as the main GPU.

evorster · 2026-03-11 04:41:11

While this workaround MAY work, it has a very real downside.
Dual GPU laptops have two GPU for exactly one reason: to save power when you don't need it.

By shifting composing to the dGPU, you will erase all benefits of the Hybrid screen mode, as the dGPU will be active all the time, and cannot be turned off to save power.

ajweiss · 2026-04-14 19:42:56

On a hybrid GPU laptop (Intel Iris Xe iGPU + NVIDIA RTX 3080 Ti dGPU) running KDE Plasma Wayland, KWin emits a continuous flood of GL_INVALID_ENUM and GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT errors when external monitors are connected via the NVIDIA dGPU.

KWin renders on the Intel iGPU and composites to the NVIDIA-attached external displays via the cross-GPU buffer sharing path. The errors appear as pairs: GL_INVALID_ENUM ("Invalid <face>") followed by GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT, repeating at the compositor repaint rate.

The errors only occur in the cross-GPU path, disconnecting the external monitors stops them immediately and downgrade to nvidia 590 reduces frequency to a few times a minute.

Impact

~390,000 error lines per boot session (~1,750 per 5 minutes), scaling with compositor repaint rate
Fills the systemd journal rapidly
Errors are cosmetic on this system (no freezes), but other users with similar setups report system freezes (Bug 491751, Bug 512792)

Reproduction

Hybrid GPU laptop: Intel iGPU (primary/rendering) + NVIDIA dGPU (external display output)
Run KDE Plasma on Wayland (KWin compositing on iGPU)
Connect external monitor(s) via NVIDIA dGPU (DisplayPort)
Move the mouse cursor across browser tabs (e.g. Chrome) on an external display — this triggers rapid partial repaints and a burst of errors

Observe journal:

journalctl _COMM=kwin_wayland -g GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT

Errors occur on any repaint of the NVIDIA-attached displays. Moving the cursor across Chrome tabs is an easy way to trigger a high rate of them. Disconnecting the external monitors stops the errors immediately.

Disabling the KWin blur desktop effect has no impact — the errors come from the base compositing path, not blur.

Regression notes

The errors were not present on this system prior to the KWin 6.6.3→6.6.4 / KDE Frameworks 6.24→6.25 upgrade (Apr 10, 2026). The same NVIDIA driver version (595.58.03, installed Mar 27) ran without these errors for 5 boot sessions across two weeks with the same external displays connected. The errors first appeared Apr 12, in the first boot after the KDE upgrade.

KWin 6.6.4 includes a direct scanout fix that changed how overlay items (cursors) interact with scanout candidates on multi-screen setups — this may have altered which rendering path is used for the NVIDIA-attached outputs.

Journal output

Apr 14 10:28:52 host kwin_wayland[4597]: 0x500: GL_INVALID_ENUM error generated. Invalid <face>.
Apr 14 10:28:52 host kwin_wayland[4597]: Invalid framebuffer status:  "GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT"
Apr 14 10:28:53 host kwin_wayland[4597]: 0x500: GL_INVALID_ENUM error generated. Invalid <face>.
Apr 14 10:28:53 host kwin_wayland[4597]: Invalid framebuffer status:  "GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT"
Apr 14 10:28:53 host kwin_wayland[4597]: 0x500: GL_INVALID_ENUM error generated. Invalid <face>.
Apr 14 10:28:53 host kwin_wayland[4597]: Invalid framebuffer status:  "GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT"

(Repeats continuously at repaint rate)

System information

iGPU: Intel Alder Lake-P Iris Xe (i915, card1)
dGPU: NVIDIA GeForce RTX 3080 Ti Mobile / GA103M (nvidia, card0)
NVIDIA driver: 595.58.03 (nvidia-open-dkms)
egl-wayland: 1.1.21-1
Mesa: 26.0.4-1
KWin: 6.6.4
Plasma: 6.6.4
Qt: 6.11.0
Kernel: 6.19.11-arch1-1
Distro: Arch Linux

Display configuration

eDP-1 (Intel iGPU, card1): 2560x1600@60Hz — laptop panel
DP-5 (NVIDIA dGPU, card0): 2560x1440@60Hz — external monitor
DP-6 (NVIDIA dGPU, card0): 2560x1440@60Hz — external monitor

KWin renders on the Intel iGPU and uses the cross-GPU path to scan out on the NVIDIA-attached DP-5 and DP-6.

EGL info

GBM platform:
EGL API version: 1.5
EGL vendor string: NVIDIA
EGL version string: 1.5

Related reports

AI
Triaged and report generated with claude-opus-4.6

seth · 2026-04-14 20:03:53

Triaged and report generated with claude-opus-4.6

What is this supposed to be? PSA? Did you visit any of the links the AI looked up for you?
export

QT_LOGGING_RULES=kwin_scene_opengl=false

ajweiss · 2026-04-14 21:11:18

What is this supposed to be? PSA?

Just an FYI, seems proper netiquette to disclose tool use.

Did you visit any of the links the AI looked up for you?
export
QT_LOGGING_RULES=kwin_scene_opengl=false

Yeah I did a fair bit of research myself. FWIW I wasn't really interested in a silencing band-aid and wanted to try and push things towards a real resolution. I've been really pleased with how far kwin+wayland has come on optimus and want to help make it stronger.

I was actually going to open this as an issue on the nvidia-egl GitHub issue tracker, but I wasn't confident enough that it was the right place. It seems clear to me that direct scanout fix in kwin exacerbates it, but unclear to me if the fundamental issue is in kwin, nvidia-egl, wayland, etc. I was hoping by posting here someone could help me get the bug report to the right place.

Thank you!

seth · 2026-04-15 07:23:59

I meant the entire post, what's the point of it?

wanted to try and push things towards a real resolution

The bug is either with nvidia or KDE, since it seems to only affect hybrid systems I'd not so easily buy that "NVidia's EGL driver reporting formats as supported that can't really be used" and caution that kwin_wayland might not sufficiently filter the formats on hybrid systems for overlapping output. You can try whether reverting the highlighted patch fixes this, but posting this *here* won't get you any closer to a fix and in any event you should™ have hooked onto the existing thread.

Please report your post to be merged w/ https://bbs.archlinux.org/viewtopic.php?id=310531

Lone_Wolf · 2026-04-15 15:27:26

Triaged and report generated with claude-opus-4.6

Thank you for making that clear.

Moderator Note
merging as asked by OP and suggested by Seth.

Arch Linux

#1 2025-11-30 10:40:53

SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#2 2025-11-30 13:00:28

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#3 2025-11-30 18:32:32

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#4 2025-12-01 03:15:51

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#5 2025-12-01 08:26:34

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#6 2025-12-02 13:52:34

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#7 2026-03-11 04:41:11

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#8 2026-04-14 19:42:56

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#9 2026-04-14 20:03:53

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#10 2026-04-14 21:11:18

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#11 2026-04-15 07:23:59

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

#12 2026-04-15 15:27:26

Re: SOLVED: kwin_wayland regression in Multi-GPU Multi-monitor setups

Board footer