IB/ipoib: Make sendonly multicast joins create the mcast group
[deliverable/linux.git] / drivers / infiniband / ulp / ipoib / ipoib_multicast.c
index 33131001fa244e8f3a1c73ca126f961f7021de94..136cbefe00f87aeb79b02d6508d42fdac5741069 100644 (file)
@@ -508,17 +508,19 @@ static void ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast)
                rec.hop_limit     = priv->broadcast->mcmember.hop_limit;
 
                /*
-                * Historically Linux IPoIB has never properly supported SEND
-                * ONLY join. It emulated it by not providing all the required
-                * attributes, which is enough to prevent group creation and
-                * detect if there are full members or not. A major problem
-                * with supporting SEND ONLY is detecting when the group is
-                * auto-destroyed as IPoIB will cache the MLID..
+                * Send-only IB Multicast joins do not work at the core
+                * IB layer yet, so we can't use them here.  However,
+                * we are emulating an Ethernet multicast send, which
+                * does not require a multicast subscription and will
+                * still send properly.  The most appropriate thing to
+                * do is to create the group if it doesn't exist as that
+                * most closely emulates the behavior, from a user space
+                * application perspecitive, of Ethernet multicast
+                * operation.  For now, we do a full join, maybe later
+                * when the core IB layers support send only joins we
+                * will use them.
                 */
-#if 1
-               if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
-                       comp_mask &= ~IB_SA_MCMEMBER_REC_TRAFFIC_CLASS;
-#else
+#if 0
                if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
                        rec.join_state = 4;
 #endif
This page took 0.024519 seconds and 5 git commands to generate.