> -----Original Message----- > From: Ilya Maximets <[email protected]> > Sent: Thursday 17 March 2022 13:10 > To: Eli Britstein <[email protected]>; [email protected]; Finn, Emma > <[email protected]> > Cc: [email protected]; Stokes, Ian <[email protected]> > Subject: Re: [ovs-dev] [PATCH V4 1/2] netdev-offload-dpdk: Use has_vlan match > attribute > > On 3/16/22 14:56, Eli Britstein wrote: > > > > > >> -----Original Message----- > >> From: Ilya Maximets <[email protected]> > >> Sent: Wednesday, March 16, 2022 2:43 PM > >> To: Eli Britstein <[email protected]>; [email protected]; Emma Finn > >> <[email protected]> > >> Cc: [email protected]; Ian Stokes <[email protected]> > >> Subject: Re: [ovs-dev] [PATCH V4 1/2] netdev-offload-dpdk: Use > >> has_vlan match attribute > >> > >> External email: Use caution opening links or attachments > >> > >> > >> On 2/7/22 17:56, Eli Britstein via dev wrote: > >>> DPDK 20.11 introduced an ability to specify existance/non-existance > >>> of VLAN tag by [1]. > >>> Use this attribute. > >>> > >>> [1]: 09315fc83861 ("ethdev: add VLAN attributes to ethernet and VLAN > >>> items") > >> > >> Hi, Eli. I'm afraid we still can't use the 'has_vlan' item until > >> there are drivers that silently ignore it. And, unfortunately, there > >> are may of them. I created a DPDK bug for that issue: > > AFAIU, the problem is not about drivers silently ignoring, but with drivers > > that > fail validation when using this flag. > > If a driver silently ignores, the same behavior as if not using this flag > > at all. > > I believe that i40e driver silently ignores the field, but it was reported > that > offloading is broken with that driver for some reason. > > Emma, Ian, could you, please, give more information on what exactly happens > with i40e driver if this patch set applied? > > If the validation actually fails, I don't see a problem with applying the > patch set, > because correctness is more important. But if there are drivers that accept > the > flow, but ignores the flag, we are getting into "what is more broken?" > territory, > so it might be OK to get the patches anyway, I'm not sure. This part we need > to > discuss, but we need more information, as what exactly is happening with the > i40e driver. >
With these patches applied, only partial offload of VLAN flows will fail to match on the inner_type. |WARN|dpdk0: rte_flow creation failed: 13 (Unsupported inner_type.). |WARN|dpdk0: Failed flow: flow create 3 ingress priority 0 group 0 pattern eth src is f6:e5:d4:c3:b2:a1 dst is 1a:2b:3c:4d:5e:6f has_vlan is 1 / vlan inner_type is 0x800 tci spec 0x7b tci mask 0xefff / ipv4 fragment_offset is 0x0 / end actions mark id 1 / rss / end I tested this with other flows (Eth, IPv4) i.e when has_vlan is 0, and those flows will offload correctly when this series is applied. +CC Beilei, maintainer of the i40e PMD. Link to these patches here - https://patchwork.ozlabs.org/project/openvswitch/list/?series=284866 Link to DPDK bug report being tracked here - https://bugs.dpdk.org/show_bug.cgi?id=958 Could you look into improving support for this within the driver itself in DPDK? Thanks, Emma <snip> _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
