Re: [libvirt] [PATCH 1/2] Add support for preallocated memory - parse xml

On Thu, Jun 23, 2016 at 01:25:28PM +0100, Jaroslav Safka wrote: This is all fine since it matches what we discussed previously in the context of Nova. For benefit of others, here is the thinking behind this. QEMU has ability to map guest memory anonymously (the default) or via a file. Libvirt only exposes the ability to use a file when dealing with huge pages. Nova wants ability to use a file with non-hugepage memory, eg a plain /dev/shm file. IOW, huge pages implies source type=file, but source type=file does not imply huge pages - it could be /dev/hugepages or /dev/shm Requiring anonymous with hugepages is an error scenario. When configuring NUMA we have ability to mark the memory as shared, vs private. eg <numa> ... <cell id='1' cpus='4-7' memory='512000' unit='KiB' memAccess='shared'/> ... </numa> The <access mode=shared|private> gives us ability to control this for the VM as a whole, without defining NUMA. The per-cell setting should override this if present. Conversely, <cell> with no memAccess attribute should use the <access mode=> top level setting. Finally QEMU has ability to allocate guest pages on-demand (when the guest first touches a page) or immediate (guaranteed fully reserved upfront). Use of huge pages implies "immediate" mode since huge pages are not swappable or overcommittable. In the non-huge page case though, we don't have a way to request immedate allocate, so this is adding it. Requiring allocate=ondemand in combination with hugepages is an error scenario. Also need to update docs/formatdomain.html.in to document this. The QEMU test suite additions should be in the patch that adds the code to the QEMU driver. 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 :|