On Wed, Aug 18, 2021 at 10:10:51AM -0700, Dan Williams wrote: > > Sounds like a nice solution. I think I can add an is_notify_supported() interface in dax_holder_ops and check it when register dax_holder. > > Shouldn't the fs avoid registering a memory failure handler if it is > not prepared to take over? For example, shouldn't this case behave > identically to ext4 that will not even register a callback? Yes.
