On Mon, Mar 11, 2024 at 10:08:56AM -0700, Anish Moorthy wrote: > I think patches 10/11/12 are useful changes to the selftest that make > sense to merge even with KVM_MEM_EXIT_ON_MISSING being mothballed- > they should rebase without any issues. And the annotations on the > stage-2 fault handlers seem like they should still be added, but I > suppose David can do that with his series. Yeah, let's fold the vCPU exit portions of the UAPI into the overall KVM userfault series. In that case there is sufficient context at the time of the "memory fault" to generate a 'precise' fault context (this GFN failed since it isn't marked as present). Compare that to the current implementation, which actually annotates _any_ __gfn_to_pfn_memslot() failures on the way out to userspace. I haven't seen anyone saying their userspace wants to use this, and I'd rather not take a new feature without a user, even if it is comparably trivial. -- Thanks, Oliver
