On Mon, Jun 1, 2020 at 6:21 PM Sean Christopherson
<sean.j.christopherson@xxxxxxxxx> wrote:
>
> On Mon, Jun 01, 2020 at 03:24:15PM -0700, Jim Mattson wrote:
> > As we already do in svm, record the last logical processor on which a
> > vCPU has run, so that it can be communicated to userspace for
> > potential hardware errors.
> >
> > Signed-off-by: Jim Mattson <jmattson@xxxxxxxxxx>
> > Reviewed-by: Oliver Upton <oupton@xxxxxxxxxx>
> > Reviewed-by: Peter Shier <pshier@xxxxxxxxxx>
> > ---
> > arch/x86/kvm/vmx/vmx.c | 1 +
> > arch/x86/kvm/vmx/vmx.h | 3 +++
> > 2 files changed, 4 insertions(+)
> >
> > diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> > index 170cc76a581f..42856970d3b8 100644
> > --- a/arch/x86/kvm/vmx/vmx.c
> > +++ b/arch/x86/kvm/vmx/vmx.c
> > @@ -6730,6 +6730,7 @@ static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
> > if (vcpu->arch.cr2 != read_cr2())
> > write_cr2(vcpu->arch.cr2);
> >
> > + vmx->last_cpu = vcpu->cpu;
>
> This is redundant in the EXIT_FASTPATH_REENTER_GUEST case. Setting it
> before reenter_guest is technically wrong if emulation_required is true, but
> that doesn't seem like it'd be an issue in practice.
I really would like to capture the last logical processor to execute
VMLAUNCH/VMRESUME (or VMRUN on the AMD side) on behalf of this vCPU.
> > vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
> > vmx->loaded_vmcs->launched);
> >
> > diff --git a/arch/x86/kvm/vmx/vmx.h b/arch/x86/kvm/vmx/vmx.h
> > index 672c28f17e49..8a1e833cf4fb 100644
> > --- a/arch/x86/kvm/vmx/vmx.h
> > +++ b/arch/x86/kvm/vmx/vmx.h
> > @@ -302,6 +302,9 @@ struct vcpu_vmx {
> > u64 ept_pointer;
> >
> > struct pt_desc pt_desc;
> > +
> > + /* which host CPU was used for running this vcpu */
> > + unsigned int last_cpu;
>
> Why not put this in struct kvm_vcpu_arch? I'd also vote to name it
> last_run_cpu, as last_cpu is super misleading.
I think last_run_cpu may also be misleading, since in the cases of
interest, nothing actually 'ran.' Maybe last_attempted_vmentry_cpu?
> And if it's in arch, what about setting it vcpu_enter_guest?
As you point out above, this isn't entirely accurate. (But that's the
way we roll in kvm, isn't it? :-)
> > };
> >
> > enum ept_pointers_status {
> > --
> > 2.27.0.rc2.251.g90737beb825-goog
> >
