[snip]
> > Additionally, has_msr_vmx_vmfunc has the similar compat issue. I think
> > it deserves a fix, too.
> >
> > -Zhao
> Thanks for your reply. In fact, I've tried to process has_msr_vmx_vmfunc in
> the same
> way as has_msr_vmx_procbased_ctls in this patch, but when I tested on Linux
> kernel
> 4.19.67, I encountered an "error: failed to set MSR 0x491 to 0x***".
>
> This issue is due to Linux kernel commit 27c42a1bb ("KVM: nVMX: Enable
> VMFUNC
> for the L1 hypervisor", 2017-08-03) exposing VMFUNC to the QEMU guest
> without
> corresponding VMFUNC MSR modification code, leading to an error when QEMU
> attempts
> to set the VMFUNC MSR. This bug affects kernels from 4.14 to 5.2, with a fix
> introduced
> in 5.3 by Paolo (e8a70bd4e "KVM: nVMX: allow setting the VMFUNC controls
> MSR", 2019-07-02).
It looks like this fix was not ported to the 4.19 stable kernel.
> So the fix for has_msr_vmx_vmfunc is clearly different from
> has_msr_vmx_procbased_ctls2.
> However, due to the different kernel support situations, I have not yet come
> up with a suitable
> way to handle the compatibility of has_msr_vmx_procbased_ctls2 across
> different kernel versions.
>
> Therefore, should we consider only fixing has_msr_vmx_procbased_ctls2 this
> time and addressing
> has_msr_vmx_vmfunc in a future patch when the timing is more appropriate?
>
I agree this fix should focus on MSR_IA32_VMX_PROCBASED_CTLS2.
But I think at least we need a comment (maybe a TODO) to note the case of
has_msr_vmx_vmfunc in a followup patch.
Let's wait and see what Paolo will say.
-Zhao
