[ofa-general] [PATCH] opensm/osm_node_info_rcv.c: create physp for the newly discovered port of the known node

[Yevgeny Kliteynik]

Hi Sasha,

This patch fixes bugzilla issue #1515:

Topology:
                 |---------------|
                 |      SW2      |
                 |---------------|
                   |x |y    |z |v
              |----|  |     |  |----|
              |       |     |       |
              |  |----|     |----|  |
              |  |               |  |
             a| b|              c| d|
      |---------------|     |---------------|
      |       SW1     |     |     SW3       |
      |---------------|     |---------------|
          |                             |
          |                             |
       HCA with SM                      HCA

During the discovery:

SM sends NodeInfo request to SW1
SM sends NodeInfo request to SW2 through link a->x
SM discovers new node SW2:
  - updates DR to SW2 to go through link a->x
  - creates physp x
SM sends NodeInfo request to SW2 through link b->y
SM discovers a known node SW2
  - DOES NOT create physp y
  - updates DR to SW2 to go through link b->y

>From now on, the DR to SW2 is going through port y, so OpenSM won't deal with
port y any more, leaving it uninitialized (no physp object for this port).

The fix is to create physp for the newly discovered port of the known
switch node, same way as it is done for HCAs.
I also added one log message for the case that showed the problem - when
one of the link sides is uninitialized (no valid ports check). Perhaps
this log message should be an error message instead?

Signed-off-by: Yevgeny Kliteynik <kliteyn at dev.mellanox.co.il>
---
 opensm/opensm/osm_node_info_rcv.c |   24 +++++++++++++++++++++++-
 1 files changed, 23 insertions(+), 1 deletions(-)

diff --git a/opensm/opensm/osm_node_info_rcv.c b/opensm/opensm/osm_node_info_rcv.c
index c52c0d5..7da3103 100644
--- a/opensm/opensm/osm_node_info_rcv.c
+++ b/opensm/opensm/osm_node_info_rcv.c
@@ -164,8 +164,12 @@ __osm_ni_rcv_set_links(IN osm_sm_t * sm,
 	 */
 	if (!osm_node_link_has_valid_ports(p_node, port_num,
 					   p_neighbor_node,
-					   p_ni_context->port_num))
+					   p_ni_context->port_num)) {
+		OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
+			"Link at node 0x%" PRIx64 ", port %u - no valid ports\n",
+			cl_ntoh64(osm_node_get_node_guid(p_node)), port_num);
 		goto _exit;
+	}

 	if (osm_node_link_exists(p_node, port_num,
 				 p_neighbor_node, p_ni_context->port_num)) {
@@ -537,8 +541,26 @@ __osm_ni_rcv_process_existing_switch(IN osm_sm_t * sm,
 				     IN osm_node_t * const p_node,
 				     IN const osm_madw_t * const p_madw)
 {
+
+	ib_smp_t *p_smp;
+	ib_node_info_t *p_ni;
+	uint8_t port_num;
+
 	OSM_LOG_ENTER(sm->p_log);

+	p_smp = osm_madw_get_smp_ptr(p_madw);
+	p_ni = (ib_node_info_t *) ib_smp_get_payload_ptr(p_smp);
+	port_num = ib_node_info_get_local_port_num(p_ni);
+
+	if (!osm_node_get_physp_ptr(p_node, port_num)) {
+		OSM_LOG(sm->p_log, OSM_LOG_DEBUG,
+			"Creating physp for node GUID:0x%"
+			PRIx64 ", port %u\n",
+			cl_ntoh64(osm_node_get_node_guid(p_node)),
+			port_num);
+		osm_node_init_physp(p_node, p_madw);
+	}
+
 	/*
 	   If this switch has already been probed during this sweep,
 	   then don't bother reprobing it.
-- 
1.5.1.4