Paolo Bonzini <pbonzini@xxxxxxxxxx> writes: > On 10/07/20 13:40, Vitaly Kuznetsov wrote: >> Hm, it seems I missed svm_set_nested_state() path >> completely. Surprisingly, state_test didn't fail) >> >> I'm struggling a bit to understand why we don't have kvm_set_cr3() on >> svm_set_nested_state() path: enter_svm_guest_mode() does it through >> nested_prepare_vmcb_save() but it is skipped in svm_set_nested_state(). >> Don't we need it at least for !npt_enabled case? > > In svm_set_nested_state you'll have CR3 already set to the right value. > On the source, KVM_GET_SREGS returns the vmcb12's CR3 and it is already > restored with KVM_SET_SREGS on the destination before set_nested_state. > > So, only the nested_cr3 has to be set. > Ah, thanks, so it seems there is no need to merge nested_svm_load_cr3() (former kvm_set_cr3()) with nested_svm_init_mmu_context() in a new nested_svm_init_mmu() as we'll only need the latter on svm_set_nested_state(). We can just move nested_svm_load_cr3() out of nested_prepare_vmcb_save() to enter_svm_guest_mode(). -- Vitaly
