https://bugs.kde.org/show_bug.cgi?id=523485
Bug ID: 523485
Summary: Intermittent Plasma Wayland login hang on AMD
hybrid-GPU laptop when the dGPU runtime-resumes during
session startup
Classification: Plasma
Product: plasma-login-manager
Version First 6.7.80
Reported In:
Platform: KDE Linux
OS: Linux
Status: REPORTED
Severity: grave
Priority: NOR
Component: general
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected], [email protected]
Target Milestone: ---
## Summary
On KDE Linux, logging into a Plasma Wayland session through Plasma Login
Manager intermittently hangs at the Plasma splash screen. Authentication
succeeds and the user session is created, but the desktop never becomes
ready.
On every clearly captured failed login, the discrete AMD GPU starts a runtime
resume approximately 1.4–1.7 seconds after systemd-logind creates the user
session. plasma-kded6.service and plasma-plasmashell.service then time out
repeatedly.
On successful boots, the dGPU does not runtime-resume during the login
window,
and both services become ready in about 1.5 seconds.
This looks like a race during the transition from the plasmalogin greeter
session to the user KWin/Wayland session on a hybrid-GPU system. This is a
correlation, not yet a proven root cause.
## Environment
- Distribution: KDE Linux
- KDE Linux VERSION_ID: 2026-07-24
- Image: 202607240254
- KDE_LINUX_COMMIT_SHA: unknownSHA
- Kernel: 7.1.4-arch1-1
- KWin: 6.7.80
- Qt: 6.11.1
- systemd: 261
- Display manager: Plasma Login Manager
- Session: Plasma Wayland
- Machine: Lenovo 82UH / Legion R9000K ARHA7
- Mainboard: LNVNB161216
- BIOS: K9CN46WW
Graphics:
- dGPU: AMD Radeon RX 6850M XT
- PCI: 0000:03:00.0
- PCI ID: 1002:73df
- DRM node: /dev/dri/card1
- render node: /dev/dri/renderD128
- boot_vga=0
- power/control=auto
- normally runtime-suspended/D3cold
- iGPU: AMD Radeon 680M
- PCI: 0000:37:00.0
- PCI ID: 1002:1681
- DRM node: /dev/dri/card2
- render node: /dev/dri/renderD129
- boot_vga=1
- drives the internal eDP display
No external monitor was connected.
## Steps to reproduce
1. Boot the laptop in its normal hybrid-graphics mode.
2. Wait for the Plasma Login Manager greeter.
3. Authenticate and start the Plasma Wayland session.
4. Repeat across several reboots.
The failure is intermittent.
## Expected result
The greeter exits and the Plasma Wayland desktop appears normally.
## Actual result
Authentication succeeds and the user session is created, but the system
remains
at the Plasma splash screen. TTY access still works.
kded6 and plasmashell are launched but do not register their D-Bus names
before the systemd startup timeout. They produce no application log messages
and no coredumps.
## Failed boot example
Relevant monotonic timestamps:
- 47.499978: logind creates the Wayland user session
- 49.174729: dGPU runtime resume starts (PSP is resuming)
- 49.249474: user plasma-kwin_wayland.service starts
- 49.305681: plasma-kded6.service starts
- 49.422863: dGPU resume completes
- 49.430283: plasma-plasmashell.service starts
- 49.549382: greeter KWin reports permission denied opening /dev/dri/card1
- 65.319532: plasma-kded6.service startup timeout
- 89.569533: plasma-plasmashell.service startup timeout
The kded6 timeout then repeats.
## Successful boot example
- 37.401363: logind creates the Wayland user session
- 39.381513: user KWin starts
- 39.422737: kded6 starts
- 39.490622: plasmashell starts
- 40.873108: kded6 is ready
- 40.877750: plasmashell is ready
There is no dGPU runtime resume around this login.
## Correlation across retained boots
Failed logins:
Boot dGPU resume relative to user-session creation
━━━━━━ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
-12 +1.669 s
────── ───────────────────────────────────────────────
-9 +1.703 s
────── ───────────────────────────────────────────────
-8 +1.422 s
────── ───────────────────────────────────────────────
-6 +1.394 s
────── ───────────────────────────────────────────────
-5 +1.431 s
────── ───────────────────────────────────────────────
-1 +1.676 s
Successful logins did not have a dGPU resume during the corresponding login
window.
## Other observations
- No amdgpu GPU fault, ring timeout, reset, or runtime-resume failure is
logged.
- The greeter's Failed to open drm node /dev/dri/card1: Permission denied
message also occasionally appears on successful boots, so that message
alone
is not sufficient to explain the failure.
- The failed boot followed a normal power-off, not hibernation or suspend.
- Both DRM nodes receive an ACL for the active user.
- No KWIN_DRM_DEVICES, DRI_PRIME, or similar override was active when the
failures occurred.
- KWIN_DRM_DEVICES restricted to the iGPU is a proposed diagnostic
workaround,
but it has not yet been tested.
--
You are receiving this mail because:
You are watching all bug changes.
