atomic_compare_exchange_strong(src, exp, dst) returns false if
*exp != src and stores src in *exp.
This was overwriting mcast_sw_info->table_size with 0 every time the
first igmp group was processed, causing all others to be skipped.
Fixes: 74daa0607c7f ("ovn-northd: Introduce parallel lflow build")
Reported-at: https://bugzilla.redhat.com/show_bug.cgi?id=1979870
Signed-off-by: Dumitru Ceara <[email protected]>
---
northd/ovn-northd.c | 8 +++-----
1 file changed, 3 insertions(+), 5 deletions(-)
diff --git a/northd/ovn-northd.c b/northd/ovn-northd.c
index 570c6a3ef..952b664df 100644
--- a/northd/ovn-northd.c
+++ b/northd/ovn-northd.c
@@ -7439,6 +7439,7 @@ build_lswitch_ip_mcast_igmp_mld(struct ovn_igmp_group
*igmp_group,
struct mcast_switch_info *mcast_sw_info =
&igmp_group->datapath->mcast_info.sw;
+ uint64_t table_size = mcast_sw_info->table_size;
if (IN6_IS_ADDR_V4MAPPED(&igmp_group->address)) {
/* RFC 4541, section 2.1.2, item 2: Skip groups in the 224.0.0.X
@@ -7449,10 +7450,8 @@ build_lswitch_ip_mcast_igmp_mld(struct ovn_igmp_group
*igmp_group,
if (ip_is_local_multicast(group_address)) {
return;
}
-
if (atomic_compare_exchange_strong(
- &mcast_sw_info->active_v4_flows,
- (uint64_t *) &mcast_sw_info->table_size,
+ &mcast_sw_info->active_v4_flows, &table_size,
mcast_sw_info->table_size)) {
return;
}
@@ -7467,8 +7466,7 @@ build_lswitch_ip_mcast_igmp_mld(struct ovn_igmp_group
*igmp_group,
return;
}
if (atomic_compare_exchange_strong(
- &mcast_sw_info->active_v6_flows,
- (uint64_t *) &mcast_sw_info->table_size,
+ &mcast_sw_info->active_v6_flows, &table_size,
mcast_sw_info->table_size)) {
return;
}
--
2.27.0
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
