I'm prepared for the "how" aspect of the 1st patch here to end up controversial. Since the observed quirk will imo want dealing with, I'd appreciate any objection to the proposed change to be accompanied by an alternative suggestion. An intention of mine was to not further increase the number of arch hooks needed. I further realize that this change conflicts with Oleksandr's "xen/gnttab: Store frame GFN in struct page_info on Arm", at the very least contextually.
1: gnttab: remove guest_physmap_remove_page() call from gnttab_map_frame() 2: memory: XENMEM_add_to_physmap (almost) wrapping checks Jan
