On 9/13/21 2:11 PM, Sean Christopherson wrote: > My argument against handling this fully in the kernel is that to handle a vNUMA > setup with multiple vEPC sections, the ioctl() would need to a take a set of file > descriptors to handle the case where an SECS is pinned by a child page in a > diferent vEPC. Bah, I'm always forgetting about the multiple vepc fd's case. I completely agree that there's no sane way to do this with a per-vepc ioctl() when the EREMOVE failures can originate from other vepc instances. The only other possible thing would be keep an mm_list for vepc instances and have this ioctl() (or another interface) blast them all. But that's going to be a heck of a lot more complicated than this is. OK... you two are wearing me down on this one. Let's just get this all documented in the changelogs, especially the retry behavior.
