> I have used the latest master which includes the patch. Ok thanks. To confirm this is an issue with the zero-copy mode, can you try to create the af_xdp interface in copy mode by adding the 'no-zero-copy' keyword in the create command, eg. ~# vppctl create int af_xdp host-if <linux iface> no-zero-copy
If it works in copy mode (as I think it will), it means the zero-copy mode does not work correctly for some reason. If it is an offset issue, you can look into the incoming data by breaking in af_xdp_device_input_bufs() and see if the extracted offset matches the start of the data in the packet. Best ben > commit 1a06e53341224bb08cff2fb6d218415c50f947d3 > Author: Benoît Ganne <bga...@cisco.com> > Date: Wed May 19 16:41:03 2021 +0200 > > af_xdp: use desc offset on rx > > Instead of pre-programming the data offset on rx, use offset passed in > the descriptor. This is more robust and future-proof. > > Type: fix > > Change-Id: I2bd910d92b8b03d17be5be85a24108be711dc7b9 > Signed-off-by: Benoît Ganne <bga...@cisco.com> > > Thx, Xuekun > > -----Original Message----- > From: Benoit Ganne (bganne) <bga...@cisco.com> > Sent: Tuesday, May 25, 2021 9:18 PM > To: Hu, Xuekun <xuekun...@intel.com>; vpp-dev@lists.fd.io > Subject: RE: AF_XDP usage help in VPP21.01 > > Hi Xuekun, > > Can you try with the latest master which now contains > https://gerrit.fd.io/r/c/vpp/+/32384 if you did not already tried it? > I suspect that could be an offset issue and this change should be more > robust for those. > > Best > ben > > > -----Original Message----- > > From: Hu, Xuekun <xuekun...@intel.com> > > Sent: mardi 25 mai 2021 14:21 > > To: Benoit Ganne (bganne) <bga...@cisco.com>; vpp-dev@lists.fd.io > > Subject: RE: AF_XDP usage help in VPP21.01 > > > > Hi, Ben. > > > > I figured out the problem in my setup that on the ixgbe driver. > > 1. ixgbe 5.1.0 (which don't have XDP enabled yet) + VPP master (that > > have your below patch) : work 2. ixgbe 5.8.1 (which have XDP enabled) > > + VPP master : doesn't work > > > > Strange. Have any ideas? > > Thanks. > > > > Thx, Xuekun > > > > -----Original Message----- > > From: Benoit Ganne (bganne) <bga...@cisco.com> > > Sent: Thursday, May 20, 2021 12:50 AM > > To: Hu, Xuekun <xuekun...@intel.com>; vpp-dev@lists.fd.io > > Subject: RE: AF_XDP usage help in VPP21.01 > > > > Hi Xuekun, > > > > > Thanks for reply. However my kernel is 5.7.19, and I checked that > > > only the first patch was included. > > > So, any more thoughts? > > > > Can you try https://gerrit.fd.io/r/c/vpp/+/32384 and see if it fixes > > you issue? > > > > Best > > ben > > > > > > > -----Original Message----- > > > From: Benoit Ganne (bganne) <bga...@cisco.com> > > > Sent: Monday, May 17, 2021 4:26 PM > > > To: Hu, Xuekun <xuekun...@intel.com>; vpp-dev@lists.fd.io > > > Subject: RE: AF_XDP usage help in VPP21.01 > > > > > > Hi Xuekun, > > > > > > I suspect you are using a Linux kernel >= 5.9. There was an ABI > > > change between 5.8 and 5.9, the issue is discussed here: > > > https://lore.kernel.org/bpf/BYAPR11MB365382C5DB1E5FCC53242609C1549@B > > > YA > > > PR11 > > > MB3653.namprd11.prod.outlook.com/ > > > I'd recommend using a Linux kernel between 5.6 and 5.9 for now until > > > those issues are resolved. > > > > > > Best > > > ben > > > > > > > -----Original Message----- > > > > From: vpp-dev@lists.fd.io <vpp-dev@lists.fd.io> On Behalf Of > > > > Xuekun > > > > Sent: jeudi 13 mai 2021 15:17 > > > > To: vpp-dev@lists.fd.io > > > > Subject: [vpp-dev] AF_XDP usage help in VPP21.01 > > > > > > > > Hi, All > > > > > > > > > > > > > > > > I used VPP21.01 and AF_XDP on a physical nic. > > > > > > > > I followed the guide in > > > > https://docs.fd.io/vpp/21.01/d8/d44/af_xdp_doc.html > > > > <https://docs.fd.io/vpp/21.01/d8/d44/af_xdp_doc.html> that > > > > > > > > 1. ip link set dev <iface> promisc on > > > > 2. vppctl create int af_xdp host-if ... // load the > > default > > > > one, which I think should forward all received packets from kernel > > > > to > > > VPP > > > > 3. vppctl set int mac address <iface> <mac> // set to the NIC > > MAC. > > > > 4. vppctl set int ip addr ... > > > > > > > > > > > > > > > > However "vppctl ping" still failed with "l3 mac mismatch". > > > > > > > > > > > > > > > > vpp# show errors > > > > > > > > Count Node Reason > > > > Severity > > > > > > > > 20 ip4-glean ARP requests > > sent > > > > error > > > > > > > > 20 enp9s0f0/0-tx sendto > required > > > > error > > > > > > > > 82 ethernet-input l3 mac > mismatch > > > > error > > > > > > > > From the remote ping target machine, tcpdump did show receiving > > > > the arp request and sending the arp reply with the correct dmac. > > > > > > > > > > > > > > > > However in VPP, "trace add af_xdp-input 10", and "show trace" > > > > > > > > Packet 1 > > > > > > > > > > > > > > > > 00:14:16:539735: af_xdp-input > > > > > > > > af_xdp: enp9s0f0/0 (2) next-node ethernet-input > > > > > > > > 00:14:16:539744: ethernet-input > > > > > > > > frame: flags 0x1, hw-if-index 2, sw-if-index 2 > > > > > > > > 0x0000: 00:00:00:00:00:00 -> 00:00:00:00:00:00 > > > > > > > > 00:14:16:539748: error-punt > > > > > > > > rx:enp9s0f0/0 > > > > > > > > 00:14:16:539750: punt > > > > > > > > ethernet-input: l3 mac mismatch > > > > > > > > > > > > > > > > Does anyone have any ideas? > > > > > > > > Many thanks. > > > > > > > > > > > > > > > > Thx, Xuekun > > > > > > > >
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#19461): https://lists.fd.io/g/vpp-dev/message/19461 Mute This Topic: https://lists.fd.io/mt/82798670/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
