[libvirt] [PATCH] Support CPU less NUMA node.

Feng, Shaohe shaohe.feng at intel.com
Wed Apr 26 05:24:45 UTC 2017


Thanks Peter.

Let's try, can it works on Phi platform firstly.

BR
Shaohe.

> -----Original Message-----
> From: Peter Krempa [mailto:pkrempa at redhat.com]
> Sent: Tuesday, April 25, 2017 4:33 PM
> To: Feng, Shaohe <shaohe.feng at intel.com>
> Cc: libvir-list at redhat.com; Du, Dolpher <dolpher.du at intel.com>; Zyskowski, Robert <robert.zyskowski at intel.com>;
> Uminski, Piotr <Piotr.Uminski at intel.com>; Daniluk, Lukasz <lukasz.daniluk at intel.com>; Zang, Rui <rui.zang at intel.com>;
> Chen, He <he.chen at intel.com>; jdenemar at redhat.com
> Subject: Re: [libvirt] [PATCH] Support CPU less NUMA node.
> 
> On Thu, Apr 20, 2017 at 19:22:49 +0800, Feng, Shaohe wrote:
> > Thanks Peter for the comments.
> > Will update it in the next version.
> >
> > On 2017年04月20日 19:17, Peter Krempa wrote:
> > > On Thu, Apr 20, 2017 at 03:49:29 +0000, Shaohe Feng wrote:
> > > > Some platform NUMA node does not include cpu, such as Phi.
> > > >
> > > > This patch support CPU less NUMA node.
> > > > But there must be one CPU cell node for a guest.
> > > > Also must assign the host nodeset for this guest cell node.
> > > Additionally to the review below:
> > >
> > > Did qemu always support this? If not, you need to add code which
> > > will allow such configuration only if qemu supports this (adding a
> > > capability bit etc ...).
> > need to check it.
> > > Also please elaborat how is this better than adding a regular numa
> > > node whith cpus which are disabled. In such case you can still add
> > > more cpus via the hotplug API in case you desire so.
> > do you means add an extra attribute "disable" for cpus?
> > such as
> >
> > <cell id='1' cpus="" cpu_disable="true" "memory='512000' unit='KiB'/>
> 
> Why would you do that? You can specify cpus="some-sane-cpu-set" and then using the <vcpus> element disable the given
> set.
> 
> See http://libvirt.org/formatdomain.html#elementsCPUAllocation





More information about the libvir-list mailing list