Git SHA: Not retrievable from this patch format
Author: David Marchand <[email protected]>
Subject: dpif-netdev: Enhance checksum coverage for tunnels.
This commit enhances checksum offload coverage for VXLAN tunnels by
adding support for outer checksums in the netdev-dummy driver and
extensive unit tests covering various traffic types and offload
scenarios for both IPv4 and IPv6 VXLAN tunnels.
> diff --git a/lib/netdev-dummy.c b/lib/netdev-dummy.c
> index b72820fcc5..0da7195940 100644
> --- a/lib/netdev-dummy.c
> +++ b/lib/netdev-dummy.c
> @@ -172,6 +172,16 @@ struct netdev_dummy {
> /* Disable L4 Tx csum offload. */
> bool ol_l4_tx_csum_disabled OVS_GUARDED;
>
> + /* Announce netdev outer IP Tx csum offload. */
> + bool ol_out_ip_tx_csum OVS_GUARDED;
> + /* Disable outer IP Tx csum offload. */
> + bool ol_out_ip_tx_csum_disabled OVS_GUARDED;
> +
> + /* Announce netdev outer UDP Tx csum offload. */
> + bool ol_out_udp_tx_csum OVS_GUARDED;
> + /* Disable outer UDP Tx csum offload. */
> + bool ol_out_udp_tx_csum_disabled OVS_GUARDED;
> +
> /* Set the segment size for netdev TSO support. */
> int ol_tso_segsz OVS_GUARDED;
> };
The new fields follow the existing pattern consistently.
> @@ -1340,6 +1392,11 @@ netdev_dummy_send(struct netdev *netdev, int qid,
> }
>
> if (VLOG_IS_DBG_ENABLED()) {
> + bool inner_ip_csum_good;
> + bool inner_l4_csum_good;
> + bool inner_ip_csum_bad;
> + bool inner_l4_csum_bad;
> + const char *tunnel;
> bool ip_csum_good;
> bool l4_csum_good;
> bool ip_csum_bad;
Does this code cause any alignment issues with the new variable
declarations mixed between inner and outer variables?
> @@ -1349,16 +1406,38 @@ netdev_dummy_send(struct netdev *netdev, int qid,
> ip_csum_bad = !!(packet->offloads & DP_PACKET_OL_IP_CKSUM_BAD);
> l4_csum_good = !!(packet->offloads & DP_PACKET_OL_L4_CKSUM_GOOD);
> l4_csum_bad = !!(packet->offloads & DP_PACKET_OL_L4_CKSUM_BAD);
> - VLOG_DBG("Tx: packet with csum IP %s, L4 %s, segsz %"PRIu16,
> + inner_ip_csum_good =
> + !!(packet->offloads & DP_PACKET_OL_INNER_IP_CKSUM_GOOD);
> + inner_ip_csum_bad =
> + !!(packet->offloads & DP_PACKET_OL_INNER_IP_CKSUM_BAD);
> + inner_l4_csum_good =
> + !!(packet->offloads & DP_PACKET_OL_INNER_L4_CKSUM_GOOD);
> + inner_l4_csum_bad =
> + !!(packet->offloads & DP_PACKET_OL_INNER_L4_CKSUM_BAD);
> + tunnel = ! dp_packet_tunnel(packet) ? "none"
> + : (dp_packet_tunnel_vxlan(packet) ? "vxlan"
> + : (dp_packet_tunnel_geneve(packet) ? "geneve"
> + : "gre"));
Might the space after the '!' operator create a style violation?
> @@ -1367,8 +1447,15 @@ netdev_dummy_send(struct netdev *netdev, int qid,
> }
>
> if (dp_packet_ip_checksum_partial(packet)
> - || dp_packet_l4_checksum_partial(packet)) {
> + || dp_packet_l4_checksum_partial(packet)
> + || dp_packet_inner_ip_checksum_partial(packet)
> + || dp_packet_inner_l4_checksum_partial(packet)) {
> dp_packet_ol_send_prepare(packet, flags);
> }
The additional conditions for inner checksums look correct.
> diff --git a/tests/dpif-netdev.at b/tests/dpif-netdev.at
> index bbce1ccee4..43e844c1ed 100644
> --- a/tests/dpif-netdev.at
> +++ b/tests/dpif-netdev.at
[ ... ]
The test additions span over 2,000 lines and cover multiple scenarios
for IPv4/IPv6 VXLAN tunnels with different offload configurations.
While comprehensive, this represents a significant test suite expansion.
Would it be beneficial to organize these tests into separate files or
use helper functions to reduce duplication between similar test cases?
The tests follow a consistent pattern but there's substantial repetition
in the Python packet generation and expected output validation.
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
