On 08/01/2018 20:36, Jim Mattson wrote: > Shouldn't there be an IBPB on *any* context switch away from a VCPU > thread, even if it is to a non-VCPU thread?
Yes, but that's the task of patches to the generic Linux context switching code. As mentioned in the cover letter, this isn't yet a full solution---just the KVM bits, which is what Radim and I have full control on. Hence the hacks with cpuid in local bools. Paolo
