executor/x86: [CRITICAL] Full System Hang When Using VM Actors on Intel Coffee Lake

[OleksiiOleksenko]

Environment

• Component: executor
• OS: Ubuntu 24.04.2 LTS
• Kernel: 6.8.0-53
• CPU: Intel Xeon E-2288G
• Affected branches: both main and dev

NOTE: the issue affects only Intel CPUs, as it is caused by VMX instructions.

Issue Description

The system completely hangs when Revizor is configured with at least one VM actor and at least one entry into this actor within the test case template.

Steps to Reproduce

1. Run the acceptance test called Feature: VM test case
2. Observe system hang requiring hard reboot

Impact

• Complete system hang
• Requires hard reboot to recover
• Blocks all functionality involving VM actors on Coffee Lake CPUs and older

Root Cause

I was able to narrow down the issue to the vmlaunch instruction that is a part of the switch macro. Further investigation is necessary

[OleksiiOleksenko]

This one is shaping to be a harder problem than expected. No solution has been found so far as I still couldn't find the root cause of the crash.

If somebody is encountering this issue, there is a weird work-around. I observed that once this test (or any other execution with VM actors) passes successfully, it won't crash again until you reboot.

To work around the issue:

• Reboot the machine
• Run the test
• If it crashes, repeat steps 1-2 until it passes
• Once it passes, VMX will work reliably until the next reboot