On 09/08/2017 22:31, David Hildenbrand wrote: > On 09.08.2017 17:14, Jim Mattson wrote: >> The only thing that makes me unhappy about this is that the >> KVM_SET_SREGS ioctl may modify some VCPU state before returning >> -EINVAL. I could hoist the call to kvm_set_apic_base, but that only >> works for one mutator. If this doesn't bother anyone else, I'll just >> leave it where it is. > > Good point, but the question is if the caller is even able to recover > from this failure? Likely not, but being cleaner is usually better... > If we care, you might have to move kvm_set_apic_base() to the very top > in kvm_arch_vcpu_ioctl_set_sregs. Or just do the check at that point. > > Guess Paolo knows the answer to our question, just as always :) Not sure I do, but I am (though only slightly) worried about not doing the kvm_mmu_reset_context if EFER as changed and also about missing update_cr8_intercept. Moving kvm_set_apic_base early is harmless, so why not move that to the beginning. Paolo
