Re: [Xen-devel] [PATCH v4 3/6] x86emul: conditionally clear BNDn for branches

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 10/01/17 09:04, Jan Beulich wrote:
> @@ -1836,6 +1840,34 @@ static int inject_swint(enum x86_swint_t
>      generate_exception(fault_type, error_code);
>  }
>  
> +static void adjust_bnd(struct x86_emulate_ctxt *ctxt,
> +                       const struct x86_emulate_ops *ops, enum vex_pfx pfx)
> +{
> +    uint64_t bndcfg;
> +    int rc;
> +
> +    if ( pfx == vex_f2 || !vcpu_has_mpx() )
> +        return;

I'm sorry, but I am still going to argue over this.  This needs to be a
host_and_vcpu check, because we are actually using real host
state/operations to perform the emulation.

At the moment, given junk in the vcpu cpuid information, we could still
hit some assertions.  Furthermore, this is the litmus test I use to
chose between the two form.  I.e. "If the vcpu information has junk in
it, will Xen crash when it comes to execute this path?".

At the point we think the vcpu information is strictly a subset of
hardware, we could turn the host_and_* side of the check into an
ASSERT(cpu_has_*) instead, but the difference between the two is still
semantic relevant.

At some point, if we ever get around supporting features while lacking
hardware support, the host_and_* checks document the areas which need to
split into

vcpu_has_*
if ( cpu_has_* )
    /* optimised emulation path with hardware support. */
else
    /* unoptimised path without any support. */

Everything else looks fine however.

~Andrew

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