On Mon, Jun 30, 2014 at 10:39:05AM +0200, Martin Kletzander wrote: > On Fri, Jun 27, 2014 at 10:53:10PM +0200, Michele Paolino wrote: > >This patch enables the possibility to run a qemu virtual machine with the > >share option for the memory-backend-file. The xml description looks like: > > > ><memdev type='file' share='on'> > > <name>ram0</name> > > <source mem-path='/hugetlbfs'/> > > <capacity unit='MiB'>1024</capacity> > ></memdev> > > > >This work is based on the previous work of Chen Fan[1]. > >We are aware of the existing conflict with the previous numa patches[2]. > >We are sharing this because it is a dependency for some use cases of the > >qemu vhost-user support(e.g. snabbswitch). > > > > There is one important thing left out. What qemu uses as a memdev > should not be exposed as such (a device) in the XML, that way the > definition makes sense only for people aware of how qemu works, but > we are trying to keep the XML as hypervisor-independent as possible. > Of course only where applicable, many places must have > hypervisor-dependent structures, but I don't think memory device is > one of these exceptions. All the options that qemu aspose as memdev > options should be abstracted in the XML. For example3, we already > have an element in the XML that says "use hugepages as a backing for > memory" and that element should be used for that. I just skimmed the > patches a bit, but it looks like none of the series works with that > information or changes the behaviour based on that. Agreed, we really don't want to expose memdev as a user visible device, since it really isn't a guest device - this is entirely host-side config information. Regards, Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :| -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list