https://bugs.kde.org/show_bug.cgi?id=528237

            Bug ID: 528237
           Summary: kwin_wayland crashes in
                    GpuManager::findCompatibleRenderDevice when a second
                    GPU is unbound for VFIO passthrough
    Classification: Plasma
           Product: kwin
      Version First 6.8.2
       Reported In:
          Platform: CachyOS
                OS: Linux
            Status: REPORTED
          Severity: crash
          Priority: NOR
         Component: generic-crash
          Assignee: [email protected]
          Reporter: [email protected]
  Target Milestone: ---

Created attachment 197982
  --> https://bugs.kde.org/attachment.cgi?id=197982&action=edit
Crashlog

DESCRIPTION
- Two GPUs. Radeon 9060 XT drives all displays; Arc A770 drives none and is
used only for QSV encode.
 - Starting a libvirt VM with <hostdev managed='yes'> on the Arc makes libvirt
rebind it from i915 to vfio-pci mid-session. kwin_wayland segfaults the same
second. Session restarts, all Wayland clients lost. 100% reproducible, happened
three times with identical library offsets  which means handleUdevEvent reacts
to a device excluded from both explicit lists.
- Not bug 527984 / commit c9d720793bbc — that crashes at compositor.cpp:481 in
prepareDirectScanout. This one is inside GpuManager, reached from
handleUdevEvent, nowhere near a scanout path.
- KWIN_DRM_DEVICES alone → crash
- KWIN_RENDER_NODES alone → crash
- Both set, KWin reporting GPU 0 only and holding neither of the Arc's nodes
and that  still crash.

SOFTWARE/OS VERSIONS
Operating System: CachyOS Linux
KDE Plasma Version: 6.8.2
KDE Frameworks Version: 6.31.0
Qt Version: 6.12.0
Kernel Version: 7.3.0-2-cachyos-bore (64-bit)
Graphics Platform: Wayland
Processors: 20 × 12th Gen Intel Core i7-12700K
Memory: 32 GiB of RAM
Graphics Processor: AMD Radeon RX 9060 XT

ADDITIONAL INFORMATION
- Related but not duplicates: 521814 (KWin holds a secondary GPU, unbind
stalls, no crash) and 524535 (KWIN_DRM_DEVICES can't express by-path names).
512707 in your Possible Duplicates list is unrelated — that's
DrmGpu::pageFlipHandler inside a guest.
- I have removed the passthrough, so I'm no longer affected, but I can re-add
it and reproduce on request if you want a backtrace with debug symbols.

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to