Re: [Xen-devel] [PATCH for-4.10] x86/hvm: Don't ignore unknown MSRs in the migration stream

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 16.11.17 at 20:15, <andrew.cooper3@xxxxxxxxxx> wrote:
> Doing so amounts to silent state corruption, and must be avoided.

I think a little more explanation is needed on why the current code
is insufficient. Note specifically this

    for ( i = 0; !err && i < ctxt->count; ++i )
    {
        switch ( ctxt->msr[i].index )
        {
        default:
            if ( !ctxt->msr[i]._rsvd )
                err = -ENXIO;
            break;
        }
    }

in hvm_load_cpu_msrs(), intended to give vendor code a first
shot, but allowing for vendor independent MSRs to be handled
here.

> --- a/xen/arch/x86/hvm/svm/svm.c
> +++ b/xen/arch/x86/hvm/svm/svm.c
> @@ -450,7 +450,7 @@ static int svm_load_msr(struct vcpu *v, struct hvm_msr 
> *ctxt)
>              break;
>  
>          default:
> -            continue;
> +            err = -ENXIO;
>          }

If the change was nevertheless needed, please add break
statements here (and in VMX code as well), despite this
currently being the last label in the switch() block.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel