Hi Julien, On 21/02/2026 12:26, Julien Grall wrote: > Hi Michal, > > On 18/02/2026 08:36, Orzel, Michal wrote: >>> But that's not correct for cases where >>> share_xen_page_with_privileged_guest() >>> is passed SHARE_ro. XENMAPSPACE_gmfn_foreign requests have to result in r/o >>> mappings in that case. >> Yes. Therefore, on Arm: >> - p2m_ram_ro is never stored in P2M tables for normal domains > > p2m_set_permission() is able to deal with p2m_ram_ro. So this could in > theory happen. Only in theory. As of today, this is the unreachable code. In fact I checked our coverage reports for safety and indeed it shows up as an unreachable code. There is no path in Xen that can lead to that point.
> >> - it's only used by get_page_from_gfn() for DOMID_XEN pages >> - it's used as a signal to install p2m_map_foreign_ro mappings >> >> The code should stay as is then and we could modify the comment to say: > > /* Read-only RAM; only used for DOMID_XEN */ > > With what I wrote above, I don't think we should add such comment. I think the goal here is to make the comment reflect the current situation (and as of now it's only used for DOMID_XEN). Taking what I wrote above, do you still think we should not update it? Once we have a use for RO for normal domains, we could then update the comment to reflect a new reality. ~Michal
