Merge pull request #7975 from wckzhang/btlcommonlist

btl/ofi: Use common provider include/exclude list

ompi/mca/mtl/ofi/mtl_ofi_component.c

@@ -306,26 +306,6 @@ ompi_mtl_ofi_progress_no_inline(void)
 	return ompi_mtl_ofi_progress();
 }
 
-static int
-is_in_list(char **list, char *item)
-{
-    int i = 0;
-
-    if ((NULL == list) || (NULL == item)) {
-        return 0;
-    }
-
-    while (NULL != list[i]) {
-        if (0 == strncasecmp(item, list[i], strlen(list[i]))) {
-            return 1;
-        } else {
-            i++;
-        }
-    }
-
-    return 0;
-}
-
 static struct fi_info*
 select_ofi_provider(struct fi_info *providers,
                     char **include_list, char **exclude_list)
@@ -334,7 +314,7 @@ select_ofi_provider(struct fi_info *providers,
 
     if (NULL != include_list) {
         while ((NULL != prov) &&
-               (!is_in_list(include_list, prov->fabric_attr->prov_name))) {
+               (!opal_common_ofi_is_in_list(include_list, prov->fabric_attr->prov_name))) {
             opal_output_verbose(1, opal_common_ofi.output,
                                 "%s:%d: mtl:ofi: \"%s\" not in include list\n",
                                 __FILE__, __LINE__,
@@ -343,7 +323,7 @@ select_ofi_provider(struct fi_info *providers,
         }
     } else if (NULL != exclude_list) {
         while ((NULL != prov) &&
-               (is_in_list(exclude_list, prov->fabric_attr->prov_name))) {
+               (opal_common_ofi_is_in_list(exclude_list, prov->fabric_attr->prov_name))) {
             opal_output_verbose(1, opal_common_ofi.output,
                                 "%s:%d: mtl:ofi: \"%s\" in exclude list\n",
                                 __FILE__, __LINE__,
@@ -733,8 +713,8 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
      * this logic if the user specifies an include list without EFA or adds EFA
      * to the exclude list.
      */
-    if ((include_list && is_in_list(include_list, "efa")) ||
-        (exclude_list && !is_in_list(exclude_list, "efa"))) {
+    if ((include_list && opal_common_ofi_is_in_list(include_list, "efa")) ||
+        (exclude_list && !opal_common_ofi_is_in_list(exclude_list, "efa"))) {
         hints_dup = fi_dupinfo(hints);
         hints_dup->caps &= ~(FI_LOCAL_COMM | FI_REMOTE_COMM);
         hints_dup->fabric_attr->prov_name = strdup("efa");

opal/mca/btl/ofi/btl_ofi_component.c

@@ -53,10 +53,25 @@ static int mca_btl_ofi_init_device(struct fi_info *info);
 
 /* validate information returned from fi_getinfo().
  * return OPAL_ERROR if we dont have what we need. */
-static int validate_info(struct fi_info *info, uint64_t required_caps)
+static int validate_info(struct fi_info *info, uint64_t required_caps,
+                         char **include_list, char **exclude_list)
 {
     int mr_mode;
 
+    if (NULL != include_list && !opal_common_ofi_is_in_list(include_list, info->fabric_attr->prov_name)) {
+        opal_output_verbose(1, opal_common_ofi.output,
+                            "%s:%d: btl:ofi: \"%s\" not in include list\n",
+                            __FILE__, __LINE__,
+                            info->fabric_attr->prov_name);
+        return OPAL_ERROR;
+    } else if (NULL != exclude_list && opal_common_ofi_is_in_list(exclude_list, info->fabric_attr->prov_name)) {
+        opal_output_verbose(1, opal_common_ofi.output,
+                            "%s:%d: btl:ofi: \"%s\" in exclude list\n",
+                            __FILE__, __LINE__,
+                            info->fabric_attr->prov_name);
+        return OPAL_ERROR;
+    }
+
     BTL_VERBOSE(("validating device: %s", info->domain_attr->name));
 
     /* we need exactly all the required bits */
@@ -219,7 +234,7 @@ static mca_btl_base_module_t **mca_btl_ofi_component_init (int *num_btl_modules,
     uint64_t progress_mode;
     unsigned resource_count = 0;
     struct mca_btl_base_module_t **base_modules;
-    char **include_list = NULL;
+    char **include_list = NULL, **exclude_list = NULL;
 
     BTL_VERBOSE(("initializing ofi btl"));
 
@@ -264,10 +279,18 @@ static mca_btl_base_module_t **mca_btl_ofi_component_init (int *num_btl_modules,
     }
 
     fabric_attr.prov_name = NULL;
-    /* Select the provider - sort of.  we just take first element in list for now */
+
+    opal_output_verbose(1, opal_common_ofi.output,
+                        "%s:%d: btl:ofi:provider_include = \"%s\"\n",
+                        __FILE__, __LINE__, *opal_common_ofi.prov_include);
+    opal_output_verbose(1, opal_common_ofi.output,
+                        "%s:%d: btl:ofi:provider_exclude = \"%s\"\n",
+                        __FILE__, __LINE__, *opal_common_ofi.prov_exclude);
+
     if (NULL != *opal_common_ofi.prov_include) {
         include_list = opal_argv_split(*opal_common_ofi.prov_include, ',');
-        fabric_attr.prov_name = include_list[0];
+    } else if (NULL != *opal_common_ofi.prov_exclude) {
+        exclude_list = opal_argv_split(*opal_common_ofi.prov_exclude, ',');
     }
 
     domain_attr.mr_mode = MCA_BTL_OFI_REQUESTED_MR_MODE;
@@ -331,7 +354,7 @@ static mca_btl_base_module_t **mca_btl_ofi_component_init (int *num_btl_modules,
     info = info_list;
 
     while(info) {
-        rc = validate_info(info, required_caps);
+        rc = validate_info(info, required_caps, include_list, exclude_list);
         if (OPAL_SUCCESS == rc) {
             /* Device passed sanity check, let's make a module.
              *

opal/mca/common/ofi/common_ofi.c

@@ -33,6 +33,25 @@ OPAL_DECLSPEC opal_common_ofi_module_t opal_common_ofi = {
 
 static const char default_prov_exclude_list[] = "shm,sockets,tcp,udp,rstream";
 
+OPAL_DECLSPEC int opal_common_ofi_is_in_list(char **list, char *item)
+{
+    int i = 0;
+
+    if ((NULL == list) || (NULL == item)) {
+        return 0;
+    }
+
+    while (NULL != list[i]) {
+        if (0 == strncasecmp(item, list[i], strlen(list[i]))) {
+            return 1;
+        } else {
+            i++;
+        }
+    }
+
+    return 0;
+}
+
 OPAL_DECLSPEC int opal_common_ofi_register_mca_variables(const mca_base_component_t *component)
 {
     static int registered = 0;

opal/mca/common/ofi/common_ofi.h

@@ -38,6 +38,21 @@ OPAL_DECLSPEC void opal_common_ofi_mca_register(void);
 OPAL_DECLSPEC void opal_common_ofi_mca_deregister(void);
 OPAL_DECLSPEC struct fi_info* opal_common_ofi_select_ofi_provider(struct fi_info *providers, 
                                                                   char *framework_name);
+/*
+ * @param list (IN)    List of strings corresponding to lower providers.
+ * @param item (IN)    Single string corresponding to a provider.
+ *
+ * @return 0           The lower provider of the item string is not in
+ *                     list or an input was NULL
+ * @return 1           The lower provider of the item string matches
+ *                     a string in the item list.
+ *
+ * This function will take a provider name string and a list of lower
+ * provider name strings as inputs. It will return true if the lower
+ * provider in the item string matches a lower provider in the list.
+ *
+ */
+OPAL_DECLSPEC int opal_common_ofi_is_in_list(char **list, char *item);
 
 END_C_DECLS