On Tue, Feb 5, 2019 at 6:37 PM Sean Christopherson <sean.j.christopherson@xxxxxxxxx> wrote: > > On Tue, Feb 05, 2019 at 04:06:21PM -0800, Jim Mattson wrote: > > This code is fine, but it makes me wonder if the nVMX implications of > > !X86_FEATURE_LM are completely botched. > > Probably, but does it even it matter? I doubt anyone has hardware to > compare against :-) AFAIK, Yonah[1] was the only production CPU to > support VMX but not 64-bit, and I'm guessing there aren't many of them > left in the wild. > > Could we do something like state that KVM always emulates a 64-bit CPU > for VMX instructions? It'd save a lot of (IMO) pointless checks. > > [1] https://en.wikipedia.org/wiki/Xeon#LV_(ULV),_%22Sossaman%22 It might be reasonable to enforce the rule that the guest CPUID.80000001H:EDX.LM[bit 29] must match the host CPUID.80000001H:EDX.LM[bit 29]. No doubt, someone would complain, though.
