OK, I think the following change (already committed) should fix IPoIB
with PCI Express HCAs. Thanks to Josh England for giving me access to
a couple of his machines for debugging, and to Tziporet Koren for
pointing out this Arbel "feature."
- Roland
Index: infiniband/hw/mthca/mthca_av.c
===================================================================
--- infiniband/hw/mthca/mthca_av.c (revision 1331)
+++ infiniband/hw/mthca/mthca_av.c (working copy)
@@ -97,6 +97,9 @@
cpu_to_be32((ah_attr->grh.traffic_class << 20) |
ah_attr->grh.flow_label);
memcpy(av->dgid, ah_attr->grh.dgid.raw, 16);
+ } else {
+ /* Arbel workaround -- low byte of GID must be 2 */
+ av->dgid[3] = cpu_to_be32(2);
}
if (0) {
_______________________________________________
openib-general mailing list
[EMAIL PROTECTED]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
