Arch Linux

HCF

Member

Registered: 2016-04-10

Posts: 33

[SOLVED] Segfault with i915-sriov-dkms on Guest Shutdown

Hoping 'Applications & Desktop Environments' is the right place for this.

I have set up https://github.com/strongtz/i915-sriov-dkms with a 12700 as per readme, and it's working great with virt-manager (Windows 11 IoT). Latest Arch running Wayland/GNOME.

Until I power off the virtual machine. Clean shutdown or forced doesn't matter.

That's when the entire host system locks up. Either completely (requires power reset), or everything but the cursor followed by a crash to login screen.

The only thing I can see in the logs that might possibly be related is this:

[77784.484710] gnome-shell[1736]: segfault at 5638d0202918 ip 00007fe7d957933d sp 00007ffc052b7fe0 error 4 in libgobject-2.0.so.0.8400.2[3a33d,7fe7d954b000+37000] likely on CPU 10 (core 20, socket 0)
[77784.484717] Code: 0d 19 93 00 00 e9 0c fd ff ff 0f 1f 00 f3 0f 1e fa 55 48 89 e5 48 85 ff 0f 84 8f 00 00 00 48 8b 07 48 85 c0 0f 84 a3 00 00 00 <48> 8b 10 48 81 fa fc 03 00 00 76 47 48 89 d6 b8 01 00 00 00 48 83
[77786.213617] WebExtensions[2794]: segfault at 0 ip 00007f8f941bf182 sp 00007ffc3e3f7240 error 6
[77786.213617] Web Content[2874]: segfault at 0 ip 00007f49c2bbf182 sp 00007fff942dfa60 error 6
[77786.213622]  in libxul.so[5fbe182,7f8f90226000+6ed7000]
[77786.213623]  in libxul.so[5fbe182,7f49bec26000+6ed7000]
[77786.213624]  likely on CPU 12 (core 24, socket 0)
[77786.213626]  likely on CPU 17 (core 37, socket 0)

Has anybody else run into this?

//UPDATE
Disabling all extensions results in less dmesg errors, but end result is the same.

I'm wondering if GNOME just doesn't deal well with releasing vGPUs?

gnome-shell[314668]: segfault at 5636cc64357d ip 00007f7c50fbf33d sp 00007ffcb576b840 error 4 in libgobject-2.0.so.0.8400.2[3a33d,7f7c50f91000+37000] likely on CPU 10 (core 20, socket 0)

Last edited by HCF (2025-06-28 03:16:01)

HCF

Member

Registered: 2016-04-10

Posts: 33

Re: [SOLVED] Segfault with i915-sriov-dkms on Guest Shutdown

https://github.com/strongtz/i915-sriov- … 1567465036

Wayland gnome session crashes when you power off VM with passthrough vGPU, but no issues with Gnome on Xorg. The root cause - mutter doesn't handle vGPU hotadd well when monitoring /dev/dri/, so when virt-manager detach vGPU and reattach it back to host it triggers gnome session crash. The workaround is simple - just change parameter of assigned vGPU device from managed="yes" to "no". Example for attached 0000:00:02.7:

<hostdev mode="subsystem" type="pci" managed="no">
  <source>
    <address domain="0x0000" bus="0x00" slot="0x02" function="0x7"/>
  </source>
  <address type="pci" domain="0x0000" bus="0x04" slot="0x00" function="0x0"/>
</hostdev>

Also you need manually (or via qemu pre-boot hook) detach vGPU from host, just once:
virsh nodedev-detach pci_0000_00_02_7

This way it will disappear from /dev/dri and virt-manager will not do nodedev-reattach, confusing Wayland.