> On 8/7/2020 7:55 AM, Eli Britstein wrote: > > > > On 8/6/2020 8:28 PM, Stokes, Ian wrote: > >>> On 8/6/2020 6:17 PM, Emma Finn wrote: > >>>> The following 2 commits introduced changes which caused a > >>>> regression for XL710 devices and functionality ceases for partial > >>>> offload as a result. > >>>> 864852a0624a ("netdev-offload-dpdk: Fix Ethernet matching for type > >>>> only.") > >>>> a79eae87abe4 ("netdev-offload-dpdk: Remove pre-validate of > patterns > >>>> function.") > >>> OVS is vendor agnostic. That kind of workaround belongs in intel PMD > >>> in dpdk, not in OVS. > >> Hi Eli, > >> > >> Yes OVS looks to be vendor agnostic, but this code I believe was > >> already in place and working for this usecase. As such it's removal > >> introduced a regression from an OVS point of view between the releases. > >> > >> We have had examples in the past where workarounds are permissible if > >> there is a clear path to fixing this in the future on the DPDK side > >> (which is what I suggest here) (for example scatter gather support > >> for MTUs in the past raised similar issue where we hand to handle > >> specific NIC until the next DPDK LTS release). > >> > >> So my suggestion is to re-instate the original workaround and remove > >> its when fixed in the next DPDK LTS which supports the change for > >> i40e at the PMD layer or if it's backported to the next 19.11 stable > >> release which would be validated for use with OVS. > > > > Hi, > > > > There was a bug with this WA. > > > > Please see > > https://patchwork.ozlabs.org/project/openvswitch/patch/1587180266- > 28632-1-git-send-email-jackerd...@gmail.com/. > > > > Is it possible to address it in DPDK instead of reverting in OVS and > > later re-reverting? > > > > Thanks, > > > > Eli > > > >> > >> I've included the i40e DPDK maintainers here for their thoughts also. > >> > >> @Beilei/Jia Is this something we can look at to introduce in the i40e > >> PMD? > > Hello, > > Please take a look at: > > http://mails.dpdk.org/archives/dev/2020-May/166272.html > > For MLX it was only an optimization. For i40e something similar may be a > workaround for this issue.
Thanks for this Eli, let me sync with Beilei on this. If it's something we can resolve in the PMD then I think we can add an errata or known issue for the 2.14 release (and possibly the 2.13 release as I think the same issue is present there). If it was fixed in the future we could then remove the issue notice. Regards Ian > > Thanks, > > Eli > > >> > >> Regards > >> Ian > >> > >>>> Fixed by partial reversion of these changes. > >>>> > >>>> Signed-off-by: Emma Finn<emma.f...@intel.com> _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev