https://bugs.kde.org/show_bug.cgi?id=507799
--- Comment #1 from Graham G <[email protected]> --- Sent my logs and steps over to ChatGPT's o3, and had ran a few other diagnostic things it suggested. Here's that output, incase it's helpful. Title: plasmashell crash during SceneWallpaper teardown; render thread wedged in NVIDIA driver (vkQueueSubmit via GLX), dual-GPU X11 Summary Plasma (X11, kwin_x11) occasionally crashes while tearing down a wallpaper scene (wallpaper-engine-kde-plugin). The render/looper thread blocks inside the NVIDIA userspace driver during vkQueueSubmit, while the main thread attempts to stop/join it. Looks like a GLX↔Vulkan interop deadlock at shutdown, possibly exacerbated by a headless second NVIDIA GPU. Environment Distro: Pop!_OS 22.04 (Jammy base) Plasma session: X11 (kwin_x11) GPUs: RTX 3080 (display) + RTX 3060 (headless) NVIDIA: 570.172.08 (kernel + userspace match) nvidia-smi → 570.172.08, CUDA 12.8 /proc/driver/nvidia/version → 570.172.08 OpenGL: 4.6 (NVIDIA 570.172.08) Vulkan: 1.3.280, NVIDIA ICD present Plugin involved: wallpaper-engine-kde-plugin (scene backend; Vulkan path) Frequency Very rare; occurs during teardown/shutdown of wallpaper scene / plasmashell. What happened Thread 7 (render): blocks in libnvidia-glcore.so.570.172.08 via libGLX_nvidia.so.0 during vvk::Queue::Submit(...) → VulkanRender::Impl::drawFrameOffscreen() → drawFrame(Scene&). Thread 1 (main): wallpaper::looper::Looper::stop() → std::thread::join() while QML/scene objects are being destroyed (SceneWallpaper, MainHandler, SceneObject), then KCrash. Secondary 3060 is headless; kernel dmesg shows nvidia ... Cannot find any crtc or sizes for 02:00.0. Expected Plasmashell / wallpaper teardown completes cleanly; no crash or hang even if a render thread is active. Actual Render thread never returns from driver submit; plasmashell tears down QML/scene graph while blocked thread is being joined → crash. -- You are receiving this mail because: You are watching all bug changes.
