On 20/11/2017 10:55, David Hildenbrand wrote: > This looks good to me. > > General question: how do we treat KVM single stepping and concurrent TF > in the guest? IOW, shouldn't the "else" rather be a check for ctxt->tf ? > (or is that handled later on e.g. in user space?) At this time, not very well. What you suggest seems sensible, however note that this function is only used on the (slow and rare) emulation path, not when the guest is running in VMX (or SVM) non-root mode. Paolo