The former logic directly do shift operation on big endian data, fix
it by convert the big endian data into CPU endian firstly.
Fixes: c3b7254093c2 ("net/nfp: support IPv6 VXLAN encap flow action")
Cc: sta...@dpdk.org
Signed-off-by: Chaoyong He <chaoyong...@corigine.com>
Reviewed-by: Niklas Söderlund <niklas.soderl...@corigine.com>
---
drivers/net/nfp/nfp_flow.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/net/nfp/nfp_flow.c b/drivers/net/nfp/nfp_flow.c
index 937fb390df..914afb85e3 100644
--- a/drivers/net/nfp/nfp_flow.c
+++ b/drivers/net/nfp/nfp_flow.c
@@ -2743,6 +2743,7 @@ nfp_flow_action_vxlan_encap_v6(struct nfp_app_fw_flower
*app_fw_flower,
struct nfp_fl_rule_metadata *nfp_flow_meta,
struct nfp_fl_tun *tun)
{
+ uint8_t tos;
uint64_t tun_id;
struct nfp_fl_act_pre_tun *pre_tun;
struct nfp_fl_act_set_tun *set_tun;
@@ -2763,9 +2764,9 @@ nfp_flow_action_vxlan_encap_v6(struct nfp_app_fw_flower
*app_fw_flower,
set_tun = (struct nfp_fl_act_set_tun *)(act_data + act_pre_size);
memset(set_tun, 0, act_set_size);
tun_id = rte_be_to_cpu_32(vxlan->hdr.vx_vni);
+ tos = rte_be_to_cpu_32(ipv6->hdr.vtc_flow) >> RTE_IPV6_HDR_TC_SHIFT;
nfp_flow_set_tun_process(set_tun, NFP_FL_TUN_VXLAN, tun_id,
- ipv6->hdr.hop_limits,
- (ipv6->hdr.vtc_flow >> RTE_IPV6_HDR_TC_SHIFT) & 0xff);
+ ipv6->hdr.hop_limits, tos);
set_tun->tun_flags = vxlan->hdr.vx_flags;
/* Send the tunnel neighbor cmsg to fw */
--
2.39.1
