>>> On 09.11.18 at 16:26, <andrew.coop...@citrix.com> wrote: > This is a minimal backport of pieces of: > > c/s 28d9a9a2d41759b9e5163037b759ac557aea767c > c/s 4c5d78a10dc89427140a50a1df5a0b8e9f073e82 > > to fix a PV shadowing problem which I hadn't anticipated at the time these > fixes were first accepted. > > Having opt_allow_superpage disabled causes guest_supports_superpages() to > return false for PV guests. Returning false causes guest_walk_tables() to > ignore L2 superpages, and read under them. > > This ignoring behaviour is correct for 2-level paging when CR4.PSE is clear, > but isn't correct for 3- or 4-level paging. > > When opt_allow_superpage is clear, PV domU's can't have superpages, but dom0 > will still have its initial P2M constructed with 2M superpages. > > The end result is that, if dom0 becomes shadowed (e.g. PV-L1TF), the next > memory access touching a P2M superpage will cause the shadow code to read > under the P2M superpage and attempt to shadow junk. > > Signed-off-by: Andrew Cooper <andrew.coop...@citrix.com>
Reviewed-by: Jan Beulich <jbeul...@suse.com> _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
