Integration tests failures

[mtjhrc]

So recently the integration tests in the CI seem to have started randomly failing as mentioned in #296 @tylerfanelli. I don't really now what's going on.

The tests previously worked on my x86_64 laptop, but now also fail. Also I have noticed that just running the chroot_vm example sometimes (seems to be very rare though) fails with:

[2025-04-16T12:08:26Z ERROR vmm::linux::vstate] Failure during vcpu run: Cannot allocate memory (os error 12)
[2025-04-16T12:08:26Z ERROR vmm::linux::vstate] Failure during vcpu run: Cannot allocate memory (os error 12)
[2025-04-16T12:08:26Z INFO  vmm] Vmm is stopping.
[2025-04-16T12:08:26Z ERROR vmm::linux::vstate] Failure during vcpu run: Cannot allocate memory (os error 12)

The issue seems to be related to the kernel version, I'm using Fedora 40, and have only tried the packaged kernels so far, but I managed to narrow down the issue somewhat:

• kernel-6.12.13-100.fc40.x86_64 (and older) seem to work fine
• kernel-6.12.15-100.fc40.x86_64 (and newer 6.13.x) seem to break our integration tests

OS: Fedora Linux 40 (Workstation Edition) x86_64 
Host: 20TJS2F45T ThinkPad P1 Gen 3 
CPU: Intel i7-10850H (12) @ 5.100GHz

[jakecorrenti]

FWIW #313 doesn't seem to be hitting any integration test issues in CI

[tylerfanelli]

In #296, once I made the sleep in KVM_RUN contingent on the tee feature, the integration tests succeed.

[slp]

I can confirm this is a kernel issue. It was introduced by torvalds/linux@43fb96a and fixed by torvalds/linux@916b7f4

This fix has landed in 6.14, which is the kernel F42 ships with. The bad news is that the broken patches made their way into stable branches (6.12 and 6.13) of the kernel. Now that we are aware of the problem, we could devise a workaround for this or just sit this one out. I'm slightly in favor of the latter.