One of the issues with running GPU's in and out of PCIe passthrough mode, running different models of GPU (older), and meeting the needs of host-native applications for GPU compute (like hashing) is the complexity and possible conflicts that arise in the host which aren't anticipated or quickly isolated. It's at this point that I should say, avoid using the host for native applications mixed with a guest VM's PCIe passthrough needs even when you don't feel the host native applications are particularly taxing.

One such issue which can occur during an upgrade, is graphical subsystem/driver changes, particularly module blacklisting. Mixing of proprietary and non-proprietary drivers or use of the ROCM drivers can present difficulties. Symptoms of the issue include the following:

  1. Multiple monitor configurations presenting in mirrored mode with a lower resolution like the maximum VGA resolution of 1600x1200. This indicates use of the fbdev driver in a failsafe config.
  2. xrandr --list-monitors shows a single default display in the aforementioned resolution as minimum and maximum.
  3. Your driver module in /var/log/Xorg.0.log shows as failing after a failure to load DRI due to the missing /dev/dri/card0 device.

A simple grep in the modprobe.d directory should locate the offending blacklist entry

Example: xorg radeon driver

~# grep -r "radeon" /etc/modprobe.d/

/etc/modprobe.d/blacklist-radeon.conf:blacklist radeon

The solution on Ubuntu is simply to comment out the blacklist entry, update-initramfs and restart:

update-initramfs -u

Previous Post Next Post