I also did some tracing on the QEMU side of my KVM instance and was not able to see any spurious EINVALs from the host. Clearly there is a timing component as replacing the HOST kernel does have an effect. But as noted in comment #28 as it can trigger on a bare metal install it must be in the GUEST side.
I also did some tracing on the QEMU side of my KVM instance and was not able to see any spurious EINVALs from the host. Clearly there is a timing component as replacing the HOST kernel does have an effect. But as noted in comment #28 as it can trigger on a bare metal install it must be in the GUEST side.