On Fri, Jul 28, 2017 at 05:48:07PM +0200, David Hildenbrand wrote: > In general, a paravirtualized interface (for detection of PMEM regions) > might have one big advantage: not limited to certain architectures. What follows is a generic rant, and slightly offtopic -sorry about that. I thought it's worth replying to above since people sometimes propose random PV devices and portability is often the argument. I'd claim if its the only argument - its not a very good one. One of the points of KVM is to try and reuse the infrastructure in Linux that runs containers/bare metal anyway. The more paravirtualized interfaces you build, the more effort you get to spend to maintain various aspects of the system. As optimizations force more and more paravirtualization into the picture, our solution has been to try to localize their effect, so you can mix and match paravirtualization and emulation, as well as enable a subset of PV things that makes sense. For example, virtio devices look more or less like PCI devices on systems that have PCI. It's not clear it applies here - memory overcommit on bare metal is kind of different. -- MST
