Re: [PATCH v4 0/6] UFFD write-tracking migration/snapshots

[Dr. David Alan Gilbert]

* Peter Xu (peterx@redhat.com) wrote:
> On Tue, Dec 01, 2020 at 06:40:55PM +0000, Dr. David Alan Gilbert wrote:
> > > Yep, seems that current userfaultfd supports hugetlbfs and shared memory 
> > > for
> > > missing pages but not for wr-protected..
> > 
> > For hugepages, you'd need kernel support - but also you'd want to make
> > sure you write the whole hugepage at once.
> 
> Or we can do similar things by splitting the huge pages just like when we
> migrate.
> 
> I should have overlooked these facts when I replied previusly - we do need the
> same logic, but also special care on these special memory types.
> 
> > 
> > For shared, there's a harder problem to ask; what happens if RAM is
> > written by the other process - for postcopy, we get the other process
> > to send us a userfaultfd that they have registered with their VM.
> 
> Good point... so we should need similar things too.
> 
> Looks like we'd better explicitly disable shmem/hugetlbfs for now from qemu
> background snapshots before we have prepared these utilities, just in case it
> got run on some "future" kernels and accidentally got enabled, so the snapshot
> files could be corrupted ones.
> 
> Is shmem used a lot in libvirt, or is it even a default configuration?

No, but it's used with vhost-user applications; like dpdk.

Dave

> -- 
> Peter Xu
> 
-- 
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK