Error when powering on a vm with a host device under Ubuntu 24.04 and kernel 6.8.0-64

[gmiranda]

Describe the bug
A server running Ubuntu 24.04.2 with linux 6.8.0-63 was working flawlessly until it was updated to 6.8.0-64.

We have 2 nvidia L4s split into 4 vGPUs, running nVidia driver KVM-570.158.02.

I get the following errors in ravada:

• (Just once per boot): qemu doesn't support rtc-reset-reinjection command
• libvirt error code: 1, message: internal error: process exited while connecting to monitor: /usr/bin/kvm-spice: W: this is an old compat wrapper script for kvm-spice /usr/bin/kvm-spice: W: please use qemu-system-x86_64 instead of /usr/bin/kvm-spice 2025-07-22T08:53:49.631244Z qemu-system-x86_64: -device {"driver":"vfio-pci","host":"0000:c3:00.4","id":"hostdev0","bus":"pci.1","addr":"0x0","rombar":1}: vfio 0000:c3:00.4: error getting device info: Input/output error

To Reproduce
Steps to reproduce the behavior:

1. Update kernel to 6.8.0-64
2. Start a machine with a host device attached to
3. After a few seconds, Ravada should show this error in the message stream

Expected behavior
No errors

Screenshots
If applicable, add screenshots to help explain your problem.

Server (please complete the following information):

• OS: Ubuntu Server 24.04.2
• Browser: Firefox
• Ravada Version: 2.3.3-rc1

Client (please complete the following information):

• Browser: Firefox
• OS: Linux Mint 22.1
• Spice client version: latest?

Additional context

[frankiejol]

Hello @gmiranda , thank you for your detailed bug report. It may be a longshot, but, could you try disabling the forced time setting procedure for the KVM virtual machines ?

At /usr/share/perl5/Ravada/Domain/KVM.pm around line 700 add a return statement at the beginning of the set_time function:

sub set_time($self) {
    return;

This code forced the time on virtual machines after start or resume and it was required to sync guests time. It may not be necessary now and I think is what triggers the problem.

[frankiejol]

@gmiranda we found issues to running Ubuntu 24 and kernel 6.8.0-64 . We downgraded to 63. In our case we spotted this in kern.log

rvd_back[5096]: segfault at 7 ip 000072c75b19b8dc sp 00007ffc4886bdb8 error 4 in libc.so.6[72c75b028000+188000] likely on CPU 7 (core 19, socket 0)

[gmiranda]

@frankiejol we still have this issue (the segfaults) with Linux 6.8.0-100-generic and Ravada 2.4.4. It only happens when there are virtual machines running.

[frankiejol]

@gmiranda I have a server with 6.8.0-100 but with other types of nvidia: rtx3090 . In another server with partitioned nvidias we have 6.8.0-63-generic.
I think your best option is to downgrade to 6.8.0-63 because it looks like an issue with nvidia drivers and kernel.