IB/BASE: Fixes #8595

The code move RDMA transport check to MD level and prevents MD creation if the underlying NIC is not supported.
openucx · Oct 18, 2022 · b3252f0 · b3252f0
1 parent a133f2d
commit b3252f0
Show file tree

Hide file tree

Showing 3 changed files with 40 additions and 12 deletions.
diff --git a/src/uct/ib/base/ib_device.c b/src/uct/ib/base/ib_device.c
@@ -743,11 +743,6 @@ static unsigned long uct_ib_device_get_ib_gid_index(uct_ib_md_t *md)
     }
 }
 
-static int uct_ib_device_is_iwarp(uct_ib_device_t *dev)
-{
-    return dev->ibv_context->device->transport_type == IBV_TRANSPORT_IWARP;
-}
-
 ucs_status_t uct_ib_device_port_check(uct_ib_device_t *dev, uint8_t port_num,
                                       unsigned flags)
 {
@@ -774,12 +769,6 @@ ucs_status_t uct_ib_device_port_check(uct_ib_device_t *dev, uint8_t port_num,
         return UCS_ERR_UNREACHABLE;
     }
 
-    if (uct_ib_device_is_iwarp(dev)) {
-        /* TODO: enable it when support is ready */
-        ucs_debug("iWarp device %s is not supported", uct_ib_device_name(dev));
-        return UCS_ERR_UNSUPPORTED;
-    }
-
     if (!uct_ib_device_is_port_ib(dev, port_num) && (flags & UCT_IB_DEVICE_FLAG_LINK_IB)) {
         ucs_debug("%s:%d is not IB link layer", uct_ib_device_name(dev),
                   port_num);

diff --git a/src/uct/ib/base/ib_md.c b/src/uct/ib/base/ib_md.c
@@ -1230,6 +1230,42 @@ static uct_md_ops_t UCS_V_UNUSED uct_ib_md_global_odp_ops = {
     .detect_memory_type = ucs_empty_function_return_unsupported,
 };
 
+static const char *uct_ib_device_transport_type_name(struct ibv_device *device)
+{
+    switch (device->transport_type) {
+    case IBV_TRANSPORT_IB:
+        return "InfiniBand";
+    case IBV_TRANSPORT_IWARP:
+        return "iWARP";
+#if HAVE_DECL_IBV_TRANSPORT_USNIC
+    case IBV_TRANSPORT_USNIC:
+        return "usNIC";
+#endif
+#if HAVE_DECL_IBV_TRANSPORT_USNIC_UDP
+    case IBV_TRANSPORT_USNIC_UDP:
+        return "usNIC UDP";
+#endif
+#if HAVE_DECL_IBV_TRANSPORT_UNSPECIFIED
+    case IBV_TRANSPORT_UNSPECIFIED:
+        return "Unspecified";
+#endif
+    default:
+        return "Unknown";
+    }
+}
+
+static int uct_ib_device_is_supported(struct ibv_device *device)
+{
+    /* TODO: enable additional transport types when ready */
+    int ret = device->transport_type == IBV_TRANSPORT_IB;
+    if (!ret) {
+        ucs_debug("device %s of type %s is not supported",
+                  device->dev_name, uct_ib_device_transport_type_name(device));
+    }
+
+    return ret;
+}
+
 int uct_ib_device_is_accessible(struct ibv_device *device)
 {
     /* Enough place to hold the full path */
@@ -1248,7 +1284,7 @@ int uct_ib_device_is_accessible(struct ibv_device *device)
         return 0;
     }
 
-    return 1;
+    return uct_ib_device_is_supported(device);
 }
 
 static ucs_status_t uct_ib_query_md_resources(uct_component_t *component,

diff --git a/src/uct/ib/configure.m4 b/src/uct/ib/configure.m4
@@ -195,6 +195,9 @@ AS_IF([test "x$with_ib" = "xyes"],
        AC_CHECK_DECLS([IBV_LINK_LAYER_INFINIBAND,
                        IBV_LINK_LAYER_ETHERNET,
                        IBV_EVENT_GID_CHANGE,
+                       IBV_TRANSPORT_USNIC,
+                       IBV_TRANSPORT_USNIC_UDP,
+                       IBV_TRANSPORT_UNSPECIFIED,
                        ibv_create_qp_ex,
                        ibv_create_cq_ex,
                        ibv_create_srq_ex],