Ok，thanks for your opinions, I will try to correct it.
Best regards，Dexia -----邮件原件----- 发件人: Ilya Maximets <[email protected]> 发送时间: 2023年7月15日 0:33 收件人: Dexia Li <[email protected]>; [email protected] 抄送: [email protected] 主题: Re: [ovs-dev] [PATCH] [PATCH] userspace: support vxlan and geneve tunnel tso On 7/13/23 13:01, Dexia Li via dev wrote: > Signed-off-by: Dexia Li <[email protected]> > --- > lib/dp-packet.h | 7 +++--- > lib/netdev-dpdk.c | 45 +++++++++++++++++++++++++++++--------- > lib/netdev-native-tnl.c | 48 +++++++++++++++++++++++++++++++++++++++++ > lib/netdev.c | 17 +++++++++------ > 4 files changed, 97 insertions(+), 20 deletions(-) Hi, Dexia Li. Thanks for the patch! It's lacking a few important parts though. First of all, as you may notice, most of the tests failed in CI. This is because it must be possible to build the code without DPDK. More generally, DPDK-specific code should not appear in files that do not have DPDK in the name. In the exception of dp-packet headers, since we add rte_mbuf as part of the dp-packet impleentation. It means that there should be a generic implementation for this functionality that is using DPDK under the hood, if available. Also, not all devices support segmentation of encapsulated packets. It should be checked that particular netdev supports offloads of this type before packets can be send to it. And we may need a generic software fallback in case device doesn't support this functionality. There should be a way to send these packets to netdev-linux or dpdk vhost-user ports, for example, left physical DPDK ports that do not support this kind of offloading. Best regards, Ilya Maximets. _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
