Il 13/06/2013 21:16, Wanlong Gao ha scritto: > On 06/14/2013 09:05 AM, Anthony Liguori wrote: >> Paolo Bonzini <pbonz...@redhat.com> writes: >> >>> Il 13/06/2013 08:50, Eduardo Habkost ha scritto: >>>> I believe an interface based on guest physical memory addresses is more >>>> flexible (and even simpler!) than one that only allows binding of whole >>>> virtual NUMA nodes. >>> >>> And "-numa node" is already one, what about just adding "mem-path=/foo" >>> or "host_node=NN" suboptions? Then "-mem-path /foo" would be a shortcut >>> for "-numa node,mem-path=/foo". >>> >>> I even had patches to convert -numa to QemuOpts, I can dig them out if >>> your interested. >> >> Ack. This is a very reasonable thing to add. > > How about making "-numa node,membind=0" like options, and also provide a > QMP interface "numa_set guest_node_id mempolicy". So that we can set > the mempolicy not only for file backed memory but also for anon mapped > guest numa node. This is full numa support in QEMU as you said. I'm making > the patches now.
Yup, that's exactly what I called "host_node". membind also makes sense, but make it mem-bind or mem-host-node for consistency with mem-path. Paolo
