Summary
On a pure-AMD machine (no hybrid/NVIDIA), cosmic-comp starts failing to commit page flips and the entire desktop becomes severely stalled: the screen barely updates and mouse/keyboard input, while still technically working, lags badly — effective update interval > 1 second. It does not recover on its own. Audio (PipeWire) keeps running normally throughout, which points at a compositor/display-pipeline stall rather than a full GPU hang.
The single most reliable signal: this only ever happens while a Zoom call is running, and killall -9 zoom clears it instantly — the desktop returns to full responsiveness immediately. I have not been able to trigger it with any other application so far.
The kernel logs show no amdgpu reset, ring timeout, or GPU fault around these events — only cosmic-comp's submit failure.
This appears distinct from the existing suspend/resume reports (#872, #1332) and the external-display-won't-activate reports (#2601 / #1086): here the displays stay active, there is no suspend/resume involved, and it is reproducibly tied to a running Zoom client.
It is also not the amdgpu kernel-panic/GPU-hang regression from #2338: I see no dm_handle_vmin_vmax_update workqueue hogging, no dc_stream use-after-free, no kernel panic and no GPU hang — nothing in the kernel log at all, only cosmic-comp's userspace page-flip rejection. I am also on a newer cosmic-comp build (0.1~1781629539~24.04~836ccba) than the one bisected as bad in #2338.
Error
cosmic-comp[2722]: Failed to submit rendering: Failed to submit result for display
Caused by:
0: The underlying drm surface encountered an error: DRM access error: Page flip commit failed on device `Some("/dev/dri/card1")` (Invalid argument (os error 22))
1: DRM access error: Page flip commit failed on device `Some("/dev/dri/card1")` (Invalid argument (os error 22))
2: Invalid argument (os error 22)
How it manifests
- Triggered while running a Zoom video call. Zoom runs under XWayland (
QT_QPA_PLATFORM=xcb), with its own GPU disabled (software/SwiftShader rendering). The failing component is cosmic-comp, not Zoom itself — but Zoom is a necessary precondition.
- Whole screen stops updating smoothly; mouse and keyboard remain usable but extremely sluggish (input/refresh latency > 1 s). The state persists indefinitely.
killall -9 zoom resolves it immediately and fully. Audio continues normally the entire time.- No amdgpu/DRM error in the kernel log at the moment of the stall; only the cosmic-comp
Page flip commit failed messages.
- Recurs on multiple days, always the same signature.
Environment
|
|
| OS |
Pop!_OS 24.04 LTS |
| Kernel |
6.18.7-76061807-generic |
| cosmic-comp |
0.1~1781629539~24.04~836ccba (apt package; binary reports "1.0.0", git commit 836ccba) |
| GPU |
AMD Radeon 860M (Ryzen AI 7 PRO 450), amdgpu, PCI 0000:c4:00.0 → card1 |
| Mesa |
25.2.8-0ubuntu0.24.04.2 |
| libdrm |
2.4.125-1ubuntu0.1~24.04.2 |
| PipeWire |
1.5.85 |
| Zoom |
7.1.0 (3715), running under XWayland |
Displays (3, multi-monitor):
card1-eDP-1 — internal panel (connected)card1-DP-11 — external (connected, via Lenovo ThinkPad Thunderbolt 5 dock)card1-DP-14 — external (connected, via the same dock)
The two externals are driven over DisplayPort-MST through the Thunderbolt 5 dock.
Notes / possibly relevant
Summary
On a pure-AMD machine (no hybrid/NVIDIA), cosmic-comp starts failing to commit page flips and the entire desktop becomes severely stalled: the screen barely updates and mouse/keyboard input, while still technically working, lags badly — effective update interval > 1 second. It does not recover on its own. Audio (PipeWire) keeps running normally throughout, which points at a compositor/display-pipeline stall rather than a full GPU hang.
The single most reliable signal: this only ever happens while a Zoom call is running, and
killall -9 zoomclears it instantly — the desktop returns to full responsiveness immediately. I have not been able to trigger it with any other application so far.The kernel logs show no amdgpu reset, ring timeout, or GPU fault around these events — only cosmic-comp's submit failure.
This appears distinct from the existing suspend/resume reports (#872, #1332) and the external-display-won't-activate reports (#2601 / #1086): here the displays stay active, there is no suspend/resume involved, and it is reproducibly tied to a running Zoom client.
It is also not the amdgpu kernel-panic/GPU-hang regression from #2338: I see no
dm_handle_vmin_vmax_updateworkqueue hogging, nodc_streamuse-after-free, no kernel panic and no GPU hang — nothing in the kernel log at all, only cosmic-comp's userspace page-flip rejection. I am also on a newer cosmic-comp build (0.1~1781629539~24.04~836ccba) than the one bisected as bad in #2338.Error
How it manifests
QT_QPA_PLATFORM=xcb), with its own GPU disabled (software/SwiftShader rendering). The failing component is cosmic-comp, not Zoom itself — but Zoom is a necessary precondition.killall -9 zoomresolves it immediately and fully. Audio continues normally the entire time.Page flip commit failedmessages.Environment
0.1~1781629539~24.04~836ccba(apt package; binary reports "1.0.0", git commit 836ccba)Displays (3, multi-monitor):
card1-eDP-1— internal panel (connected)card1-DP-11— external (connected, via Lenovo ThinkPad Thunderbolt 5 dock)card1-DP-14— external (connected, via the same dock)The two externals are driven over DisplayPort-MST through the Thunderbolt 5 dock.
Notes / possibly relevant
EINVALon an atomic page-flip commit, on the samecard1device path reported in "Failed to submit rendering: Failed to submit result for display" #872 and External monitor dies on hybrid graphics laptop when going to sleep #1332 — but my case is AMD-only and involves neither suspend/resume nor hybrid graphics.killall -9 zoominstantly restores responsiveness, it looks like a Zoom client surface/buffer is putting cosmic-comp's atomic commit into a state the kernel keeps rejecting (e.g. an XWayland/overlay or direct-scanout buffer with an invalid configuration), rather than a transient GPU hiccup.