Solving the AMD Reset Bug in oVirt and KVM Environments

If you’re navigating the complexities of virtualization management systems, you might share my recent challenge with Red Hat’s oVirt, the open-source virtualization management platform. For those unfamiliar, oVirt orchestrates virtual machines, manages resources, and provides a centralized interface for distributed computing environments. It’s worth noting that while my experience is with oVirt, the insights I’ve gained may also benefit those working with KVM or XEN—especially KVM, since it’s the underlying technology in oVirt.

The Challenge: AMD Graphics Cards Passthrough Post-Reinstallation

I had a smooth operation with graphics passthrough for four AMD graphics cards until I reinstalled one of my compute nodes. Post-reinstallation, the notorious AMD reset bug resurfaced, throwing a wrench into my setup.

The Discovery: Missing VDSM Hooks

Upon investigation, I discovered that the VDSM Hooks I had crafted to tweak the KVM XML configurations were missing. These hooks are crucial—they ensure that certain features are enabled or disabled when the virtual machine starts. Specifically, it seemed that enabling APIC (Advanced Programmable Interrupt Controllers) was part of the solution. Additionally, I reintroduced an SMBios removal hook, which strips the SMBios line from the VM’s XML configuration. While I can’t pinpoint which modification was the definitive fix, I’ve reinstated both to positive effect.

Sharing the Solution: VDSM Hooks Repository

For those in the oVirt community, I’ve made my VDSM hooks available on GitHub. You can find them at gentoorax/vdsm-hook-scripts. These scripts are a resource for anyone looking to modify their VM configurations to potentially overcome similar issues.

For the Non-oVirt Users: Debugging with Python Scripts

If oVirt isn’t part of your toolset, these Python scripts can still serve a purpose. By running them with the --debug command line parameter, you can generate output that’s ripe for comparison against your KVM configurations. This could be a valuable step in diagnosing and resolving issues akin to the AMD reset bug.

Closing Thoughts

Virtualization management is an evolving field, and the challenges we face often lead to community-driven solutions. Whether you’re a seasoned pro or new to the scene, I hope my experiences and resources can make your virtualization journey a bit smoother. If you’ve faced similar issues or have insights to share, I’d love to hear from you in the comments below.