Comment 0 for bug 1922472

Revision history for this message
Amir Tzin (amirtz) wrote :

[Impact]

On mlx5 device, When creating geneve tunnel in a Virtual Function and setting it up error are logged in kernel buffer.

[test case]

using two setups connected back to back
on both sides create vf's
$ echo 1 > /sys/class/net/ens5f0/device/sriov_numvfs
add ip-es to vf's interfaces on both sides (13.194.5.1/16, 13.194.6.1/16)
$ ip a add 13.194.5.1/16 dev ens5f0v0
set interfaces on both sides up
$ ip l set dev ens5f0v0 up
check connectivity
$ ping 13.194.6.1 -I 13.194.5.1 -c 6
on both sides define geneve tunnel with id 300 over the vf-s
$ ip link add name gen_vf type geneve id 300 remote 13.194.6.1
add ip addresses to geneve interfaces on both sides (14.194.5.1/16, 14.194.6.1/16)
$ ip a add 14.194.5.1/16 dev gen_vf
set geneve interfaces up
$ ip l set dev gen_vf up
check log
$ dmesg
[ 1221.501048] mlx5_core 0000:24:00.2 ens5f0v0: Error cqe on cqn 0x48a, ci 0x9, sqn 0x116, opcode 0xd, syndrome 0x2, vendor syndrome 0x68
[ 1221.501179] 00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[ 1221.501183] 00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[ 1221.501185] 00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[ 1221.501188] 00000030: 00 00 00 00 6a 10 68 02 0a 00 01 16 00 09 20 d2
[ 1221.501240] mlx5_core 0000:24:00.2 ens5f0v0: ERR CQE on SQ: 0x116
[ 1222.930608] mlx5_core 0000:24:00.2 ens5f0v0: Error cqe on cqn 0x48e, ci 0x5, sqn 0x11b, opcode 0xd, syndrome 0x2, vendor syndrome 0x68
[ 1222.930733] 00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[ 1222.930736] 00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[ 1222.930739] 00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
[ 1222.930741] 00000030: 00 00 00 00 6a 10 68 02 0a 00 01 1b 00 05 2d d2
[ 1222.930791] mlx5_core 0000:24:00.2 ens5f0v0: ERR CQE on SQ: 0x11b

[Fix]

The issue was solved upstream v5.12-rc1 with
e1c3940c6003 net/mlx5e: Check tunnel offload is required before setting SWP
The attached patch is modification of e1c3940c6003 for focal kernel.

[Regression Potential]
Regression risk is low as It is a very small fix which was also tested thoroughly on upstream
setups.