Re: [Patch net v2] mlx5: fixup checksum for short ethernet frame padding

[Eric Dumazet]

On 11/28/2018 04:05 PM, Cong Wang wrote:

> While we are on this page, mlx5e_lro_update_hdr() incorrectly assumes
> TCP header is located right after struct iphdr, which is wrong if we could
> have IP options on this path.
> 
> It could the hardware which already verified this corner case though.
> 

GRO makes sure IPv4 header has no options, so I would not be surprised
that LRO on mlx5 has the same logic.