>From: Kevin Traynor [mailto:[email protected]] >Sent: Thursday, December 7, 2017 3:29 PM >To: Jan Scheurich <[email protected]>; Kavanagh, Mark B ><[email protected]>; Ilya Maximets <[email protected]>; Stokes, >Ian <[email protected]>; [email protected] >Cc: [email protected]; Mooney, Sean K <[email protected]>; >Fischetti, Antonio <[email protected]>; Bie, Tiwei ><[email protected]>; Mcnamara, John <[email protected]>; Guoshuai Li ><[email protected]>; Loftus, Ciara <[email protected]>; Yuanhan Liu ><[email protected]> >Subject: Re: [ovs-dev] [PATCH V4 2/2] netdev-dpdk: vHost IOMMU support > >On 12/07/2017 03:09 PM, Jan Scheurich wrote: >>>> I think the point that both yourself and Sean has made is completely >valid, which puts option a) back on the table. >>>> >>> >>> a) Sounds ok to me. I think an early DPDK17.11.1 before OVS 2.9 would be >>> good in addition though. It is nicer that an OVS 2.9 user doesn't have >>> to know they can't use the latest DPDK in the guest. >>> >> >> Would the virtio PMD bug in DPDK 17.11 in the guest actually be mitigated by >running DPDK 17.05 or a fixed 17.11.1 as vhostuser backend inside OVS on the >host? >> >> If not, I would prefer if we decoupled the DPDK life cycle of OVS and DPDK >applications in the guest. Guests should update their DPDK version if it >contains a critical bug. >> > >I don't think there is any documented coupling between host and guest >DPDK versions. I doubt anyone tests lots of combinations but hopefully >virtio provides the necessary means to run multiple combos. I think it's >reasonable in this case to document a warning for an OVS user about a >known bad combination that is likely to be selected (i.e. latest >upstream/releases). > >Kevin. > >> BR, Jan >>
Hi Jan, DPDK v17.11 in the host is fine; the observed issue is present in the guest's virtio driver in DPDK v17.11 (which is not present in v17.05.2). As Kevin mentioned, I don't think there is any explicit coupling - or expectation of same - regarding the combination of DPDK versions used in guest and host; however, in this case I think it's certainly sensible to document the issue presented by the inclusion of 17.11 DPDK in the guest (at least when DPDK applications are used therein). Thanks, Mark >> _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
