On Mon, Jul 25, 2016 at 5:49 PM, Pravin B Shelar <pshe...@ovn.org> wrote: > OVS compat layer can handle tunnel GSO packets. but it does > keep skb encapsulation on for packet handled in GSO. This can > confuse some NIC drivers. I have seen this issue on intel devices: > > i40e 0000:42:00.0: TX driver issue detected, PF reset issued > > Following patch resets this bit in case compat layer handles the packet. > > VMware-BZ: 1698877 > Signed-off-by: Pravin B Shelar <pshe...@ovn.org>
In upstream, this is done as part of the GSO code (for example, in __skb_udp_tunnel_segment()) so that probably makes more sense and is safer if this is GSO specific. There is already code in ovs_iptunnel_handle_offloads() that will clear the encapsulation bit in the case of checksum offload on the outer header. One minor comment - skb->encapsulation is actually a u8 field instead of a bool, so setting it to 0 rather than false is more consistent with the rest of the kernel. Something that I noticed while looking at this is it looks like the recent patch that moved the check for gso_type_mask into a GSO-only block in ovs_iptunnel_handle_offloads() might cause a bit of a performance regression. Even though that field is GSO-specific, it is also used to control whether we resolve partial checksums. Even in cases where we do need to use the OVS offload compat code, I suspect we could take better advantage of hardware offloads - not computing the checksum when !skb->encapsulation (since every kernel can do UDP checksum offload), using scatter/gather in GSO, checking for backport support when clearing type in rpl_udp_tunnel_handle_offloads(). _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
