Re: [PATCH v2] x86/traps: Rework #PF[Rsvd] bit handling

[Jan Beulich <jbeulich@suse.com>]

On 21.05.2020 17:43, Andrew Cooper wrote:
> @@ -1439,6 +1418,21 @@ void do_page_fault(struct cpu_user_regs *regs)
>      if ( unlikely(fixup_page_fault(addr, regs) != 0) )
>          return;
>  
> +    /*
> +     * Xen doesn't have reserved bits set in its pagetables, nor do we permit
> +     * PV guests to write any.  Such entries would generally be vulnerable to
> +     * the L1TF sidechannel.
> +     *
> +     * The shadow pagetable logic may use reserved bits as part of
> +     * SHOPT_FAST_FAULT_PATH.  Pagefaults arising from these will be resolved
> +     * via the fixup_page_fault() path.
> +     *
> +     * Anything remaining is an error, constituting corruption of the
> +     * pagetables and probably an L1TF vulnerable gadget.
> +     */
> +    if ( error_code & PFEC_reserved_bit )
> +        goto fatal;
> +
>      if ( unlikely(!guest_mode(regs)) )
>      {
>          enum pf_type pf_type = spurious_page_fault(addr, regs);
> @@ -1457,13 +1451,12 @@ void do_page_fault(struct cpu_user_regs *regs)
>          if ( likely((fixup = search_exception_table(regs)) != 0) )

While I continue to not fully agree with not trying to fix up such
faults if the fault location has recovery code attached, I realize
that we're not going to reach agreement here, so somewhat hesitantly
Acked-by: Jan Beulich <jbeulich@suse.com>

Jan