bond0: hw csum failure

[digrouz]

Hello,

I've upgraded my 18.04 to use the 4.18 hwe kernel, since then I'm seeing this kind of message:

[ 125.502876] bond0: hw csum failure
[ 125.503965] CPU: 36 PID: 0 Comm: swapper/36 Tainted: P O 4.18.0-13-generic #14~18.04.1-Ubuntu
[ 125.503966] Hardware name: Dell Inc. PowerEdge R840/08XR9M, BIOS 1.3.9 11/21/2018
[ 125.503967] Call Trace:
[ 125.503969] <IRQ>
[ 125.503974] dump_stack+0x63/0x85
[ 125.503978] netdev_rx_csum_fault+0x38/0x40
[ 125.503982] __skb_checksum_complete+0xbc/0xd0
[ 125.503986] tcp_v4_rcv+0x12f/0xae0
[ 125.503990] ip_local_deliver_finish+0x62/0x200
[ 125.503993] ip_local_deliver+0x6f/0xf0
[ 125.503996] ? ip_route_input_noref+0x28/0x40
[ 125.503999] ip_rcv_finish+0x126/0x420
[ 125.504002] ip_rcv+0x28f/0x360
[ 125.504007] __netif_receive_skb_core+0x48c/0xb70
[ 125.504010] ? __build_skb+0x2b/0xf0
[ 125.504013] ? tcp4_gro_receive+0x137/0x1a0
[ 125.504017] __netif_receive_skb+0x18/0x60
[ 125.504021] ? __netif_receive_skb+0x18/0x60
[ 125.504025] netif_receive_skb_internal+0x45/0xe0
[ 125.504029] napi_gro_receive+0xc5/0xf0
[ 125.504050] mlx5e_handle_rx_cqe+0x1a6/0x510 [mlx5_core]
[ 125.504070] mlx5e_poll_rx_cq+0xd3/0x990 [mlx5_core]
[ 125.504088] mlx5e_napi_poll+0x9b/0xc60 [mlx5_core]
[ 125.504093] net_rx_action+0x140/0x3a0
[ 125.504098] __do_softirq+0xe4/0x2bb
[ 125.504103] irq_exit+0xbc/0xd0
[ 125.504107] do_IRQ+0x8a/0xd0
[ 125.504110] common_interrupt+0xf/0xf
[ 125.504112] </IRQ>
[ 125.504116] RIP: 0010:cpuidle_enter_state+0xa5/0x2c0
[ 125.504116] Code: 8b 3d cf cc 9d 76 e8 0a 16 89 ff 48 89 c3 0f 1f 44 00 00 31 ff e8 6b 21 89 ff 45 84 ff 0f 85 c8 01 00 00 fb 66 0f 1f 44 00 00 <48> 2b 5d d0 48 ba cf f7 53 e3 a5 9b c4 20 48 89 d8 48 c1 fb 3f 48
[ 125.504170] RSP: 0018:ffffb32098bc7e50 EFLAGS: 00000246 ORIG_RAX: ffffffffffffffd7
[ 125.504173] RAX: ffff91fdefe62c40 RBX: 0000001d38785887 RCX: 000000000000001f
[ 125.504175] RDX: 0000001d38785887 RSI: 000000002ac44ebf RDI: 0000000000000000
[ 125.504177] RBP: ffffb32098bc7e90 R08: 0000000000000f33 R09: ffffd20140645240
[ 125.504178] R10: ffffb32098bc7e20 R11: 0000000000000f0c R12: 0000000000000003
[ 125.504180] R13: ffffd20140645240 R14: ffffffff8a384918 R15: 0000000000000000
[ 125.504185] cpuidle_enter+0x17/0x20
[ 125.504188] call_cpuidle+0x23/0x40
[ 125.504191] do_idle+0x204/0x280
[ 125.504195] cpu_startup_entry+0x73/0x80
[ 125.504199] start_secondary+0x1ab/0x200
[ 125.504202] secondary_startup_64+0xa5/0xb0

I seems to be related to a change in the 4.18 kernel that does not trust anymore the checksum response from the mlx5 driver. This bug seems to appears on other distro too: https://access.redhat.com/solutions/3425461

Is there a way to fix this without disabling the hardware checksum offloading ?