On 06/12/2018 03:35 PM, Razvan Cojocaru wrote: >>> @@ -210,7 +211,7 @@ static int p2m_init_altp2m(struct domain *d) >>> return -ENOMEM; >>> } >>> p2m->p2m_class = p2m_alternate; >>> - p2m->access_required = 1; >>> + p2m->access_required = hostp2m->access_required; >> There must have been a reason to have it start out as 1. You >> mention the fact in the description, but not why it is okay (or >> even necessary) to change it. > If there was one, I can't see what it was. It in any case make altp2ms > behave differently than the hostp2m. The mem_access system has been > designed so that the user can say "this domain should now no longer be > able to continue if there are EPT violatios and there's no vm_event > subscriber attached". > > Our application, for one, explicitly _disables_ access_required - so > that it is able to detach from the domain and reattach later. Hence the > surprise when an altp2m-enabled domain crashed when the introspection > agent detached. > > Basically, the altp2m behaviour can't be controlled at all here. Maybe > Tamas remembers / knows more about why access_required ended up being 1 > always, but it was probably just the quickest way to write the original > patch.
CCd Tamas. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
