On Tue, Aug 2, 2016 at 4:55 PM, Jesse Gross <je...@kernel.org> wrote: > On Tue, Aug 2, 2016 at 3:55 PM, pravin shelar <pshe...@ovn.org> wrote: >> On Tue, Aug 2, 2016 at 3:11 PM, Jesse Gross <je...@kernel.org> wrote: >>> On Tue, Aug 2, 2016 at 2:17 PM, Pravin B Shelar <pshe...@ovn.org> wrote: >>>> Signed-off-by: Pravin B Shelar <pshe...@ovn.org> >>>> --- >>>> datapath/linux/compat/include/net/udp.h | 2 +- >>>> datapath/linux/compat/udp.c | 19 ++----------------- >>>> datapath/linux/compat/udp_tunnel.c | 17 +---------------- >>>> 3 files changed, 4 insertions(+), 34 deletions(-) >>> >>> Can you give some more information on this? Is it a bug fix, >>> performance improvement, or cleanup? >> >> How about following commit msg. >> Following patch simplifies UDP-checksum routine by unconditionally >> using checksum offload for non GSO packets. We might get some >> performance improvement due to code simplification. > > This is just the check for whether the device has the > NETIF_F_V[46]_CSUM feature? I wonder whether that is really enough of > a benefit to deviate from upstream. If it is noticeable then it seems > like this should be an upstream patch. >
In UDP xmit dst_entry is set after calling this function, so for non goo packets this function always calculate checksum in software. > I think the biggest checksum related improvement for compat code would > be to make the checksum computation conditional on skb->encapsulation > being set in rpl_ip_local_out. I am not sure how that will help, In rpl_ip_local_out() GSO packet we can not generate checksum partial packets, udp_set_csum() can not handle it. non GSO packet's checksum is calculated in ovs_iptunnel_handle_offloads() and the skb is not touched in rpl_ip_local_out(). _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
