https://bugs.kde.org/show_bug.cgi?id=522800
Bug ID: 522800
Summary: Plasma 6.7.x Wayland — hard freeze at login on RX 6700
XT (RDNA2) with 4 monitors (6.6.5 works,
6.7.0/6.7.1/6.7.2 all freeze)
Classification: Plasma
Product: kwin
Version First 6.7.0
Reported In:
Platform: CachyOS
OS: Linux
Status: REPORTED
Severity: critical
Priority: NOR
Component: wayland-generic
Assignee: [email protected]
Reporter: [email protected]
Target Milestone: ---
### Summary
Since the Plasma 6.6.5 → 6.7.0 update, my Wayland session hard-freezes right
after I enter my password at the login screen: all screens go black, the cursor
freezes on the primary monitor, and the keyboard becomes unresponsive (Num-Lock
LED no longer toggles). A short press of the power button still triggers a
clean ACPI shutdown, so the kernel is alive — only the graphical session is
wedged. The problem is identical on 6.7.0, 6.7.1 and 6.7.2. Rolling back to
6.6.5 (Btrfs/Snapper snapshot) restores a fully working desktop with the exact
same hardware, cabling and configuration.
### System
- Distro: CachyOS (rolling), Btrfs + Snapper + Limine
- CPU: AMD Ryzen 9 5950X
- GPU: AMD Radeon RX 6700 XT (RDNA2 / Navi 22), discrete, PCI 0000:0e:00.0,
amdgpu
- RAM: 128 GB
- Kernel: linux-cachyos 7.1.3-1 — unchanged across working and broken states;
no kernel/firmware update was part of the breaking transaction. *(please verify
with `uname -r`)*
- Session: KDE Plasma on Wayland (kwin_wayland)
- Broken on: kwin 6.7.0-3.1, 6.7.1-1.1, 6.7.2-1.1
- Works on: Plasma/kwin 6.6.5
- Mesa: 3:26.1.4-1 (also froze on 2:26.1.2 and 3:26.1.2 — see below)
- Displays: 4 connected outputs — DP-1, DP-2, DP-3, HDMI-A-1 (one is a small
touch panel used as a stream-deck). All report `connected` in
`/sys/class/drm/`.
### Symptom (precise)
- Login screen (SDDM/plasmalogin) displays and accepts input normally.
- Immediately after entering the password, all outputs go black; a cursor is
briefly visible on the primary output, then frozen.
- Keyboard completely unresponsive (Num-Lock LED does not toggle).
- Short power-button press → clean shutdown (ACPI still working).
- Hard freeze: `journalctl -b -1` contains no kwin/drm fatal message — journald
cannot flush the final line before the session locks up. Last persisted entries
are unrelated (bluetooth, NetworkManager dispatcher).
### What I have already tested (elimination)
1. **Mesa is NOT the cause.** Held mesa at the previous working version via
`IgnorePkg` and updated only the rest to Plasma 6.7.0 → still froze. Mesa later
moved 2:26.1.2 → 3:26.1.4 → still froze. Independent of mesa version.
2. **Kernel/firmware unchanged**, so this is a Plasma 6.7 userspace regression,
not a kernel one.
3. **amdgpu Overdrive is a trigger.** I run `options amdgpu
ppfeaturemask=0xFFF7FFFF` (Overdrive enabled) via `/etc/modprobe.d/`. With
Overdrive enabled, 6.7.x freezes instantly at login. Disabling Overdrive (mask
becomes `0xfff7bfff`, i.e. bit `0x4000` cleared) + `mkinitcpio -P` allowed
exactly ONE successful login to 6.7.2 — but the desktop came up at a fallback
1024×768 / scale 2.7 in the top-left corner, because no valid kwin output
config had ever been saved (all previous logins froze first).
4. **The deeper trigger is KWin's multi-monitor atomic modeset.** When I moved
`~/.config/kwinoutputconfig.json` aside to force KWin to rebuild the 4-output
config from scratch, it hard-froze AGAIN — even with Overdrive disabled. So
Overdrive-off only lets it limp up while reusing a minimal fallback config; a
fresh multi-output atomic modeset in 6.7 wedges the GPU.
5. **Workarounds that did NOT help:** `KWIN_DRM_NO_AMS=1` and
`KWIN_FORCE_SW_CURSOR=1` (set via `~/.config/plasma-workspace/env/`) — still
froze.
6. Possibly related dmesg line (may be benign): `amdgpu 0000:0e:00.0: [drm]
Failed to setup vendor infoframe on connector DP-2: -22`.
### What works
Plasma 6.6.5 with the identical hardware, all four monitors, and Overdrive
ENABLED — rock solid.
### Logs
- `cachyos-bugreport.sh` paste: https://paste.cachyos.org/p/30f21e4.log
- Note: because the freeze is hard, the local journal does not capture the
fatal event (`journalctl -b -1` shows no kwin/drm error).
### Question
Is this a known KWin 6.7 regression around AMD (RDNA2) multi-output atomic
modeset? Is there a config/env workaround to bring up a 4-monitor RDNA2 setup
on 6.7.x, or should I stay on 6.6.5 until a fix lands? Happy to capture a
`drm.debug` log or test patches.
--
You are receiving this mail because:
You are watching all bug changes.
