Re: [PATCH] KVM: SVM: Ensure SPEC_CTRL[63:32] is context switched between guest and host

[Sean Christopherson <seanjc@google.com>]

On Thu, Nov 06, 2025, Uros Bizjak wrote:
> On Thu, Nov 6, 2025 at 2:13 AM Sean Christopherson <seanjc@google.com> wrote:
> >  998:
> > -
> >         /* Now restore the host value of the MSR if different from the guest's.  */
> > -       movl PER_CPU_VAR(x86_spec_ctrl_current), %eax
> > -       cmp SVM_spec_ctrl(%_ASM_DI), %eax
> > +#ifdef CONFIG_X86_64
> > +       mov SVM_spec_ctrl(%rdi), %rdx
> > +       cmp PER_CPU_VAR(x86_spec_ctrl_current), %rdx
> >         je 901b
> > -       xor %edx, %edx
> > +       mov PER_CPU_VAR(x86_spec_ctrl_current), %rdx
> > +       movl %edx, %eax
> > +       shr $32, %rdx
> 
> The above code can be written as:
> 
> mov PER_CPU_VAR(x86_spec_ctrl_current), %rdx
> cmp SVM_spec_ctrl(%rdi), %rdx

Gah, that's obvious in hindsight.

> je 901b
> movl %edx, %eax
> shr $32, %rdx
> 
> The improved code will save a memory read from x86_spec_ctrl_current.
> 
> > +#else
> > +       mov SVM_spec_ctrl(%edi), %esi
> > +       mov PER_CPU_VAR(x86_spec_ctrl_current), %eax
> 
> Can the above two instructions be swapped, just to be consistent with
> x86_64 code?
> 
> > +       xor %eax, %esi
> 
> > +       mov SVM_spec_ctrl + 4(%edi), %edi
> > +       mov PER_CPU_VAR(x86_spec_ctrl_current + 4), %edx
> 
> ... and the above two insns.

Ya, will do.  Thanks!