[pve-devel] kernel 4.10 : mellanox connectx-5 nic bug with lacp

[Alexandre DERUMIER]

Hi, 
Seem than mellanox connectx-5 card are buggy on kernel 4.10.

works fine with kernel 4.4 and 4.12.

I'll try to backport last patches from 4.12 or 4.9.
Too bad that 4.10 is not lts :(



[ 38.920764] mlx5_core 0000:04:00.1 eth5: TX timeout detected 
[ 38.920786] mlx5_core 0000:04:00.1 eth5: TX timeout on queue: 0, SQ: 0xed, CQ: 0x12, SQ Cons: 0x26 SQ Prod: 0x34 
[ 38.920838] mlx5_core 0000:04:00.0 eth4: TX timeout detected 
[ 38.920852] mlx5_core 0000:04:00.0 eth4: TX timeout on queue: 0, SQ: 0xed, CQ: 0x12, SQ Cons: 0x27 SQ Prod: 0x31 
[ 39.128889] mlx5_core 0000:04:00.1 eth5: Link up 
[ 39.136625] bond1: link status up again after 0 ms for interface eth5 
[ 39.342022] mlx5_core 0000:04:00.0 eth4: Link up 
[ 39.342332] mlx5_core 0000:04:00.0 eth4: speed changed to 0 for port eth4 
[ 39.348653] bond1: link status up again after 0 ms for interface eth4 
[ 55.048676] mlx5_core 0000:04:00.0 eth4: TX timeout detected 
[ 55.048707] mlx5_core 0000:04:00.0 eth4: TX timeout on queue: 0, SQ: 0x14d, CQ: 0x12, SQ Cons: 0x36 SQ Prod: 0x5d 
[ 55.048761] mlx5_core 0000:04:00.1 eth5: TX timeout detected 
[ 55.048776] mlx5_core 0000:04:00.1 eth5: TX timeout on queue: 0, SQ: 0x14d, CQ: 0x12, SQ Cons: 0x43 SQ Prod: 0x6d 
[ 55.250931] mlx5_core 0000:04:00.0 eth4: Link up 
[ 55.256659] bond1: link status up again after 0 ms for interface eth4 
[ 55.451522] mlx5_core 0000:04:00.1 eth5: Link up 
[ 55.451803] mlx5_core 0000:04:00.1 eth5: speed changed to 0 for port eth5 
[ 55.456657] bond1: link status up again after 0 ms for interface eth5 
[ 115.976786] mlx5_core 0000:04:00.0 eth4: TX timeout detected 
[ 115.976835] mlx5_core 0000:04:00.0 eth4: TX timeout on queue: 0, SQ: 0x1ad, CQ: 0x12, SQ Cons: 0x526 SQ Prod: 0x552 
[ 116.185610] mlx5_core 0000:04:00.0 eth4: Link up 
[ 116.185806] mlx5_core 0000:04:00.0 eth4: speed changed to 0 for port eth4