Starting the app on KDE Wayland can lock up KWin and freeze the whole desktop (NVIDIA)

[ismailivanov]

Hi, I ran into a complete desktop freeze immediately after launching the app. I am not sure whether the root cause is in the app/Electron, KWin, or the NVIDIA driver, but the timestamps line up very closely with the app launch.

What happened

I was in a normal KDE Plasma Wayland session. Discord was open, and a Godot/OpenGL application had been stopped a couple of seconds before this. I launched ChatGPT/Codex Desktop from the application menu.

The window did not appear normally. The mouse, keyboard, and the rest of the desktop became unresponsive, and I eventually had to power-cycle the machine.

This has happened once so far and is not reliably reproducible. After rebooting, the updater installed a newer build and that build currently launches normally.

Timeline

Local timestamps:

21:52:03  systemd: Started ChatGPT.
21:52:04  Electron spawned.
21:52:06  Electron GPU process logged the Wayland/Vulkan warning below.
21:52:16  KWin logged its first "main thread was hanging" message.
21:52-21:57  KWin continued logging temporary main-thread hangs.
21:53:36  Mouse event processing was reported as lagging.
21:55:12  Keyboard event processing was 510 ms behind.
21:57:33  KWin atomic commit failed.
21:57:36  KWin could not find a working output layer configuration.
21:57:38  Applying the output configuration failed.
21:59:04  Last journal entry before the forced power-off.

Relevant app launcher output:

Electron launch mode: rendering_mode=default wslg_detected=0
ozone_platform=default ozone_hint=auto gpu_enabled=1
gpu_disable_arg=0 gpu_compositing_disabled=1

ERROR: ui/ozone/platform/wayland/gpu/wayland_surface_factory.cc:252
'--ozone-platform=wayland' is not compatible with Vulkan.
Consider switching to '--ozone-platform=x11' or disabling Vulkan

The primary window reached ready-to-show after 38,386 ms during the affected launch. With the newer build after reboot, the same step took 579 ms.

Relevant KWin journal entries:

kwin_wayland: The main thread was hanging temporarily!
kwin_wayland: Libinput: event processing lagging behind, your system is too slow
kwin_wayland: atomic commit failed: Permission denied
kwin_wayland: Failed to find a working output layer configuration!
kwin_wayland: Applying output configuration failed!

The app backend remained alive and continued serving assets/responding to requests during the freeze, so this does not look like a normal application crash. There was no coredump.

I also checked the kernel journal around the incident:

• no OOM event in the incident window
• no NVIDIA Xid
• no GPU fault reported by the kernel
• no soft/hard lockup or hung-task warning
• no storage I/O error

System information

OS: CachyOS Linux (rolling)
Kernel: 7.1.3-2-cachyos
Desktop: KDE Plasma 6.7.2
KWin: 6.7.2
Session: Wayland
GPU: NVIDIA GeForce RTX 4060 Ti
NVIDIA driver: 610.43.03
CPU: AMD Ryzen 7 7700 (8 cores / 16 threads)
RAM: 16 GiB
Electron: 42.1.0
Displays:
  - 2560x1440 @ 180 Hz
  - 1920x1080 @ 165 Hz

Affected package:

codex-desktop 2026.07.14.004718+c243c94f-1
App release reported in the log: 26.707.62119

Installed automatically after reboot:

codex-desktop 2026.07.15.130125+40e34814-1
App release reported in the log: 26.707.72221

One caveat: the Wayland/Vulkan warning also appears on successful launches, including the new build, so I do not think that warning alone proves the cause. It may still be relevant to the KWin/NVIDIA interaction.

Would it make sense for the launcher to disable Vulkan explicitly on native Wayland, or fall back to X11/software rendering if startup stalls? I can collect additional targeted diagnostics if there is a preferred way to reproduce this safely.

[ismailivanov]

I went through the current launcher code and the relevant history after filing this.

The affected launch used the launcher's normal Wayland profile. In that profile the launcher deliberately adds --disable-gpu-compositing for side-panel stability while leaving Ozone selection on auto. The wayland-gpu profile does the opposite and keeps GPU compositing enabled. There is also already a supported persistent X11 override in ~/.config/codex-desktop/electron-flags.conf.

I do not think there is enough evidence for a broad code change yet:

• this is still a single non-reproducible freeze;
• the same Wayland/Vulkan warning appears on successful launches;
• there was no NVIDIA Xid or kernel GPU fault;
• forcing all KDE/Wayland users onto X11 would have scaling and IME trade-offs;
• disabling Vulkan for every native Wayland launch would be speculative for the same reason.

For now I enabled the existing persistent workaround on the affected machine:

--ozone-platform=x11

XWayland is available on this system, and the flags file now has exactly that one active option. I will keep using it and report back if the freeze happens again. If another KDE/NVIDIA user can reproduce this, the useful A/B comparison would be default Wayland versus persistent X11, and possibly CODEX_LINUX_RENDERING_MODE=wayland-gpu, with the KWin and launcher timestamps from the same boot. That would give enough evidence to decide whether the launcher needs a narrowly scoped automatic profile.