[openib-general] [PATCH] osm: PathRecord prefer 1K MTU for MT23108 - not for MTU Sel=3

[Eitan Zahavi]

Hi Hal

We have reviewed the patch for the above and figured out there is an
issue with it:
Currently when MTU_SEL=3 the quirk applies.
We think this is wrong behavior as MTU_SEL=3 means "max possible MTU" by 
the IBTA spec. So if an application/ULP would like to get the max MTU possible 
the correct answer is 2K for Tavor by the spec.
So this patch fxies the quirk and when MTU_SEL=3 it does not apply the MTU
limit quirk for Tavor devices.

Thanks

Eitan

Signed-off-by:  Eitan Zahavi <eitan at mellanox.co.il>

Index: 1.1/src/userspace/management/osm/opensm/osm_sa_multipath_record.c
===================================================================
--- 1.1/src/userspace/management/osm/opensm/osm_sa_multipath_record.c	(revision 9500)
+++ 1.1/src/userspace/management/osm/opensm/osm_sa_multipath_record.c	(working copy)
@@ -203,9 +203,13 @@ boolean_t
       break;
 
     case 1:    /* must be less than */
-    case 3:    /* largest available */
 	       /* can't be disqualified by this one */
       break;
+    case 3:    /* largest available */
+               /* the ULP intentionally requested */
+               /* the largest MTU possible */
+      return(FALSE);
+      break;
 			
     default:
       /* if we're here, there's a bug in ib_multipath_rec_mtu_sel() */
Index: opensm/osm_sa_path_record.c
===================================================================
--- 1.1/src/userspace/management/osm/opensm/osm_sa_path_record.c	(revision 9500)
+++ 1.1/src/userspace/management/osm/opensm/osm_sa_path_record.c	(working copy)
@@ -204,9 +204,13 @@ static boolean_t
       break;
 
     case 1:    /* must be less than */
-    case 3:    /* largest available */
                /* can't be disqualified by this one */
       break;
+    case 3:    /* largest available */
+               /* the ULP intentionally requested */
+               /* the largest MTU possible */
+      return(FALSE);
+      break;
 			
     default:
       /* if we're here, there's a bug in ib_path_rec_mtu_sel() */