<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40"> <head> <meta http-equiv="Content-Type" content="text/html; charset=us-ascii"> <meta name="Generator" content="Microsoft Word 15 (filtered medium)"> <style></style> </head> <body lang="EN-US" link="#0563C1" vlink="#954F72"> <div class="WordSection1"> Hi<o:p></o:p> <o:p> </o:p> I work mostly in OpenStack on enabling ovs with dpdk.<o:p></o:p> <o:p> </o:p> When deploying vms on host running ovs with dpdk vms are booted utilizing <o:p></o:p> Vhost-user interfaces.<o:p></o:p> Qemu support creating vms with vhost-user network interfaces as of v2.1.<o:p></o:p> Libvirt currently has support for requesting the use of vhost-user interfaces by added the following xml fragment<o:p></o:p> <o:p> </o:p> <interface type='vhostuser'><o:p></o:p> <mac address='fa:16:3e:ea:2a:08'/><o:p></o:p> <source type='unix' path='/var/run/openvswitch/vhuf1204e0c-98' mode='client'/><o:p></o:p> <model type='virtio'/><o:p></o:p> <alias name='net0'/><o:p></o:p> <address type='pci' domain='0x0000' bus='0x00' slot='0x03' function='0x0'/><o:p></o:p> </interface><o:p></o:p> <o:p> </o:p> Send traffic via a vhost-user interface, the vhost-user frontend (provided by qemu) must share <o:p></o:p> the virtios ring to the vhost-backed via passing a file descript to the memory backing object<o:p></o:p> of the qemu instance as part of the port creation.<o:p></o:p> <o:p> </o:p> Today the only native way to request Libvirt to create a vm who’s memory is backed by a memdev which can be<o:p></o:p> Accessed via a file descriptor is to request hugepage backed memory. This requires the operator to<o:p></o:p> Manage and configure hugepage on each of their compute hosts and take special care to ensure that vms are<o:p></o:p> Not placed on host were vhost-user interface are used if they do not request hugepages.<o:p></o:p> <o:p> </o:p> Today it is possible to use Libvirt to spawn a vm without hugepage memory and a file descriptor backed memdev<o:p></o:p> Via the use of the qemu:commandline element.<o:p></o:p> <o:p> </o:p> <qemu:commandline><o:p></o:p> <qemu:arg value='-object'/><o:p></o:p> <qemu:arg value='memory-backend-file,id=mem,size=1024M,mem-path=/var/lib/libvirt/qemu,share=on'/><o:p></o:p> <qemu:arg value='-numa'/><o:p></o:p> <qemu:arg value='node,memdev=mem'/><o:p></o:p> <qemu:arg value='-mem-prealloc'/><o:p></o:p> </qemu:commandline><o:p></o:p> <o:p> </o:p> I created a proof of concept patch to nova to demonstrate that this works however to support this usecase in<o:p></o:p> Nova a new xml element is required.<o:p></o:p> <a href="https://review.openstack.org/#/c/309565/1">https://review.openstack.org/#/c/309565/1</a><o:p></o:p> <o:p> </o:p> I would like to propose the introduction of a new subelemnt to the memorybacking element to request file discrptro backed memory<o:p></o:p> <o:p> </o:p> <memoryBacking><o:p></o:p> <filedescriptor size_mb=”1024” path=”/var/lib/libvirt/qemu” prealloc=”true” shared=”on” /> <o:p></o:p> </memoryBacking><o:p></o:p> <o:p> </o:p> The above filedescriptor xml fragment above would then be parsed to generate the same qemu argument as the qemu:commandline fragment<o:p></o:p> Therefor allowing the creation of a vm with vhost-user interface without hugepage memory backing.<o:p></o:p> <o:p> </o:p> Before I start looking at the Libvirt code base I wanted to ask if the Libvirt community would be open to this change and what would be the best way <o:p></o:p> To approach enabling this feature.<o:p></o:p> <o:p> </o:p> Regards<o:p></o:p> Sean.<o:p></o:p> <o:p> </o:p> </div> </body> </html>