On Tue, Mar 13, 2018 at 10:30:12AM +0000, Marc Zyngier wrote:
> >> +
> >> + if (cpus_have_const_cap(ARM64_HARDEN_BRANCH_PREDICTOR)) {
> >
> > Moving the create_hyp_mappings() under this cap check looks like a fix
> > that could be posted separately?
>
> We could. Not sure that's a big deal though.
Nah, it isn't worth the fuss.
> >> +
> >> + if (cpus_have_const_cap(ARM64_HARDEN_EL2_VECTORS)) {
> >> + __kvm_harden_el2_vector_slot = atomic_inc_return(&arm64_el2_vector_last_slot);
> >
> > If I understood the logic in the above function correctly, then we won't
> > be using this slot when we have the ARM64_HARDEN_BRANCH_PREDICTOR cap.
> > Should we even bother allocating it when we don't intend to use it?
>
> You could be on a system with a mix of affected and non-affected CPUs,
> and the capability only tells you that at least one of the CPUs is
> affected. For the non affected CPUs, we won't have a slot (since there
> is no workaround to provide), unless we actually allocated one.
>
Ah yes, the heterogeneous considerations that I always forget to consider.
Thanks,
drew
