Hi, thanks for your reply. Do you have details on the unclear definition? I would go with the RFC definition that adds additional 802.1Q header in front while the main difference is the different ethertype (0x88a8 instead of 0x8100).
Do you see any chances that a patch added to the drivers might be accepted in upstream that tries to cover the RFC definition of QinQ? Thanks Andreas On 7/18/19 5:47 PM, Fujinaka, Todd wrote: > As far as I know you're correct that it's not supported in the driver at this > time. It's questionable if it's supported in the hardware as not everything > in the datasheet has been validated completely. > > The other problem I've heard is that there is no clear definition of QinQ. > There are many different use cases, and several implementations asked for by > different customers. > > Todd Fujinaka > Software Application Engineer > Datacenter Engineering Group > Intel Corporation > todd.fujin...@intel.com > > -----Original Message----- > From: Andreas Herz [mailto:e1000-de...@herzandreas.de] > Sent: Wednesday, July 17, 2019 8:02 AM > To: e1000-devel@lists.sourceforge.net > Subject: [E1000-devel] Double VLAN (QinQ) results in single instead of multi > queue usage with ixgbe and i40e > > Hi, > > I have some setups with Intel X710 and Intel X520 cards (10GE) which are used > for monitoring (IDS) purpose. For improved performance I used the possibility > that the NICs offer several RX queues which I can later use with the IDS. > With rx-flow-hash and RSS settings I see a solid distribution of the incoming > packets when I run 'ethtool -S $NIC' even in scenarios with *one* VLAN tag > added. But in scenarios where I have VLAN QinQ (double vlan) it all comes > down to one queue being used. > > I read a lot and it seems that it's supported by the hardware but not by the > drivers yet. Please correct me if I'm wrong. > > I also tried different firmware and i40e/ixgbe driver versions but the > problem still persists with the most recent versions. I played around with > all the offloading settings, rx-flow-hash settings, RSS keys but didn't > succeed. > > I also was wondering why setting 'rx-flow-hash' with 'v' doesn't work, I get > 'invalid command' and not 'operation not supported' but looks like it only > supports 'sdfn' on those NICs/drivers. > > I saw some older posts like this > https://sourceforge.net/p/e1000/mailman/e1000-devel/thread/loom.20140128T201406-960%40post.gmane.org/ > and patches like > https://github.com/serhepopovych/linux/commit/8ddd8e870c3d60d971dfae49c0fd89ec3e1d3ac9 > where I got the assumption that it might be possible to solve this issue. > > So first of all can anyone confirm my findings? And if yes is there a fix or > just a wrong configuration on my side or is that something that needs to be > patched? Or even worse. is this something that can't be solved with those > hardware cards? > > Thanks > > Andreas > > > > _______________________________________________ > E1000-devel mailing list > E1000-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/e1000-devel > To learn more about Intel® Ethernet, visit > http://communities.intel.com/community/wired _______________________________________________ E1000-devel mailing list E1000-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/e1000-devel To learn more about Intel® Ethernet, visit http://communities.intel.com/community/wired
