On Tue, Sep 24, 2019 at 10:29 AM Nadav Amit <nadav.amit@xxxxxxxxx> wrote: > > > On Sep 20, 2019, at 12:44 PM, Jim Mattson <jmattson@xxxxxxxxxx> wrote: > > > > On Fri, Sep 20, 2019 at 12:36 PM Krish Sadhukhan > > <krish.sadhukhan@xxxxxxxxxx> wrote: > >> On 9/19/19 4:02 PM, Jim Mattson wrote: > >>> Ensure that support for RDPRU is not enumerated in the guest's CPUID > >>> and that the RDPRU instruction raises #UD. > >> > >> > >> The AMD spec says, > >> > >> "When the CPL>0 with CR4.TSD=1, the RDPRUinstruction will > >> generate a #UD fault." > >> > >> So we don't need to check the CR4.TSD value here ? > > > > KVM should set CPUID Fn8000_0008_EBX[RDPRU] to 0. > > > > However, I should modify the test so it passes (or skips) on hardware. :-) > > Thanks for making this exception. Just wondering: have you or anyone else > used this functionality - of running tests on bare-metal? I have not. However, if there is a simple way to add this testing to our workflow, I would be happy to ask the team to do so before sending submissions upstream. > I ask because it would also save me the trouble of checking (occasionally) > that nothing broke. >
