Re: 2 CPU Conformance Issue in KVM/x86

["Radim Krčmář" <rkrcmar@redhat.com>]

2015-03-03 12:18+0200, Nadav Amit:
> Paolo Bonzini <pbonzini@redhat.com> wrote:
> > On 03/03/2015 09:34, Nadav Amit wrote:
> >> I got two conformance issues in x86/KVM. For the first one I have no
> >> solution. For the latter, my solution is not “great”. Ideas and feedback
> >> would be appreciated.
> >> 
> >> The first problem is caused by the deprecating of FPU CS/DS in new Intel
> >> CPUs. Assume the VM executes a floating point instruction in real mode (when
> >> CS != 0), and later KVM exits to userspace, causing XSAVE/XRSTOR to save and
> >> restore the FPU state. At this point FPU CS/DS in new CPUs are zero. If the
> >> VM then executes FSAVE in real-mode the save FPU IP would be wrong, since it
> >> is actually calculated by the CPU as [FPU CS] * 4 + [FPU IP].
> > 
> > I think this was analyzed a couple years ago and we decided that this bit
> > was not virtualizable.
> 
> I am fully aware of the previous reports ( https://lkml.org/lkml/2013/10/16/258 ).
> 
> However, one might have expected the conformance problem to be fully
> resolved now, since [FPU CS] and [FPU DS] are deprecated in new CPUs.
> Indeed, the problem is resolved in all modes, but not in real-mode.

Since we can't get the CS/DS from real FPUs, we'd have to do ugly and
slow hacks to virtualize it (checking all possible changes of FPU
pointers and code segments).

I think that forcing the deprecation bit to guest CPUID, if set on host,
would be the best compromise between correctness and sanity.