On 14.01.21 00:57, Alex Williamson wrote: > On Thu, 7 Jan 2021 14:34:23 +0100 > David Hildenbrand <da...@redhat.com> wrote: > >> We support coordinated discarding of RAM using the RamDiscardMgr. Let's >> unlock support for coordinated discards, keeping uncoordinated discards >> (e.g., via virtio-balloon) disabled. >> >> This unlocks virtio-mem + vfio. Note that vfio used via "nvme://" by the >> block layer has to be implemented/unlocked separately. For now, >> virtio-mem only supports x86-64 - spapr IOMMUs are not tested/affected. > > I think SPAPR always uses a vIOMMU, but I won't claim to understand it.
I heard about that, but I also know that they are using virtio-balloon. Theoretically, virtio-balloon does currently not support vIOMMUs (iommu=on results in the balloon device essentially being disable). > Is there anything other than testing that restricts it to x86-64? ARM I am not aware, but well, I guess we'll learn as we go/unlock virtio-mem on these architectures. Jonathan is trying to enable arm64 support for virtio-mem. I asked him to test vfio. I'll have to cleanup my s390x prototype and find a machine to test with vfio. > and S390 will use the same type1 IOMMU backend as far as vfio support > is concerned. Should we do something more than put it in a commit log > if we really want to prevent it elsewhere? Thanks, I think it should be sufficient to have it in the logs for now. I'll be involved when unlocking virtio-mem on other archs and will request to test vfio. In case we find issues, we should try fixing them first (instead of fencing it / bailing out). Obviously, this won't stop other RamDiscardMgr users from popping up, but I guess we'll stumble over that as well. -- Thanks, David / dhildenb
