https://bugs.kde.org/show_bug.cgi?id=528172
Bug ID: 528172
Summary: KWin 6.8.2: blurry scanout on nouveau (GK107) when
using the legacy / non-atomic modesetting path
Classification: Plasma
Product: kwin
Version First 6.8.2
Reported In:
Platform: Fedora RPMs
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: platform-drm
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected]
Target Milestone: ---
Created attachment 197931
--> https://bugs.kde.org/attachment.cgi?id=197931&action=edit
Attachments as described
SUMMARY
After upgrading Plasma 6.7.5 → 6.8.2 output on an NVIDIA GeForce GT 740 (GK107)
driven by nouveau renders blurry (NB1. nouveau genuinely lacks atomic) (NB2. no
scaling involved on these monitors)
Two monitors on that GPU are affected; three monitors on the Intel iGPU remain
sharp. With Kwin 6.7.5 until today morning everything was working fine
SYSTEM
- Fedora 44, kernel 7.2.13-200.fc44.x86_64
- KDE Plasma 6.8.2, KWin 6.8.2-2.fc44 (Wayland)
- GPU 1: NVIDIA GK107 [GeForce GT 740] — nouveau 1.4.2, 10de:0fc8,
/dev/dri/card0 (minor 0)
- GPU 2: Intel Raptor Lake-S UHD — i915 1.6.0, 8086:a782, /dev/dri/card1 (minor
1)
(NB: card numbers are not stable across boots)
- Displays: 5 total; 2 on GT740 (both DVI), 3 on Intel
REGRESSION WINDOW
kwin-6.7.5-1 → kwin-6.8.2-2
kwin-common, kwin-libs likewise
plasma-workspace* likewise
Cannot bisect locally: 6.7.5 is not in any enabled Fedora repo or cache, and
`dnf history undo 208` fails to resolve 6.7.5.
OBSERVED
- Both GT740 outputs are blurry — like a resolution/scale mismatch, but OSD
reports correct 1920x1080@60, PC mode, Wide.
- Swap cables: blur follows the GT740, not the monitor or cable.
- Intel outputs stay sharp with the same monitors and cables.
DIAGNOSTICS THAT ISOLATE THE FAULT
- Spectacle screenshot of a GT740 output: SHARP. → framebuffer content is
correct; fault is at scanout/present.
- Raw scanout via modetest on the same nouveau connectors:
sudo modetest -M nouveau -s 47@62:1920x1080
→ SHARP. → nouveau KMS/modesetting itself is fine.
- KWIN_COMPOSE=Q (software rendering): STILL BLURRY.
→ fault is NOT in the GPU render/compositing stage.
- Possible cause: fault may be in KWin's present/scanout stage on nouveau's
non-atomic path.
- Plane state shows 1:1 scanout. nouveau-state.txt has crtc-pos=1920x1080+0+0
with src-pos=1920.000000x1080.000000+0.000000+0.000000 on both active planes —
no scaling (scaling is ruled out by the state dump.)
- Dmabuf has unsupported modifier for format XR24 little-endian (0x34325258) —
appears twice in kwin.log at 09:46:33 (that's directly in the
present/buffer-import path).
KWIN LOG
journalctl --user -u plasma-kwin_wayland:
kwin_wayland[2375]: No backend specified, automatically choosing drm
kwin_wayland[2375]: drmSetClientCap for Atomic Mode Setting failed.
Using legacy mode on GPU "/dev/dri/card0"
nouveau on GK107 exposes no atomic modesetting; KWin falls back to the legacy
path. Suspect a change in the legacy/non-atomic present path in 6.8.2.
ATTACHMENTS
- screenshot of a blurry monitor (looks fine viewed on an intel monitor)
- actual photo of the blurry monitor (= also what the same screenshot looks
like on that monitor)
- drm_info
- /sys/kernel/debug/dri/0/state (nouveau)
- /sys/kernel/debug/dri/1/state excerpt (Intel, for contrast)
- kwin log (journalctl --user -u plasma-kwin_wayland --since today >
/tmp/kwin.log)
- kernel log (journalctl -b -k)
--
You are receiving this mail because:
You are watching all bug changes.
