Clean up OFI common code and delay patcher initialization until needed

config/opal_check_ofi.m4

@@ -126,19 +126,26 @@ AC_DEFUN([_OPAL_CHECK_OFI],[
     CPPFLAGS="$CPPFLAGS $opal_ofi_CPPFLAGS"
 
     AS_IF([test $opal_ofi_happy = yes],
-          [AC_CHECK_MEMBER([struct fi_info.nic],
+          [AC_CHECK_HEADERS([rdma/fi_ext.h])
+
+           AC_CHECK_MEMBER([struct fi_info.nic],
                            [opal_check_fi_info_pci=1],
                            [opal_check_fi_info_pci=0],
-                           [[#include <rdma/fabric.h>]])])
+                           [[#include <rdma/fabric.h>]])
+
+           AC_DEFINE_UNQUOTED([OPAL_OFI_PCI_DATA_AVAILABLE],
+                              [$opal_check_fi_info_pci],
+                              [check if pci data is available in ofi])
 
-    AC_DEFINE_UNQUOTED([OPAL_OFI_PCI_DATA_AVAILABLE],
-                       [$opal_check_fi_info_pci],
-                       [check if pci data is available in ofi])
+           AC_CHECK_DECLS([PMIX_PACKAGE_RANK],
+                          [],
+                          [],
+                          [#include <pmix.h>])
 
-    AC_CHECK_DECLS([PMIX_PACKAGE_RANK],
-                   [],
-                   [],
-                   [#include <pmix.h>])
+           AC_CHECK_TYPES([struct fi_ops_mem_monitor], [], [],
+                          [#ifdef HAVE_RDMA_FI_EXT_H
+#include <rdma/fi_ext.h>
+#endif])])
 
     CPPFLAGS=$opal_check_ofi_save_CPPFLAGS
     LDFLAGS=$opal_check_ofi_save_LDFLAGS
@@ -157,18 +164,6 @@ AC_DEFUN([_OPAL_CHECK_OFI],[
                  [AC_MSG_WARN([OFI libfabric support requested (via --with-ofi or --with-libfabric), but not found.])
                   AC_MSG_ERROR([Cannot continue.])])
            ])
-    opal_ofi_import_monitor=no
-    AS_IF([test $opal_ofi_happy = "yes"],
-          [OPAL_CHECK_OFI_VERSION_GE([1,14],
-                                     [opal_ofi_import_monitor=yes],
-                                     [opal_ofi_import_monitor=no])])
-
-
-if test "$opal_ofi_import_monitor" = "yes"; then
-    AC_DEFINE_UNQUOTED([OPAL_OFI_IMPORT_MONITOR_SUPPORT],1,
-                       [Whether libfabric supports monitor import])
-fi
-
 ])dnl
 
 

ompi/mca/mtl/ofi/mtl_ofi_component.c

@@ -254,9 +254,7 @@ ompi_mtl_ofi_component_register(void)
                                     MCA_BASE_VAR_SCOPE_READONLY,
                                     &ompi_mtl_ofi.num_ofi_contexts);
 
-    opal_common_ofi_register_mca_variables(&mca_mtl_ofi_component.super.mtl_version);
-
-    return OMPI_SUCCESS;
+    return opal_common_ofi_mca_register(&mca_mtl_ofi_component.super.mtl_version);
 }
 
 
@@ -285,7 +283,7 @@ ompi_mtl_ofi_component_open(void)
             "provider_exclude")) {
         return OMPI_ERR_NOT_AVAILABLE;
     }
-    return opal_common_ofi_init();
+    return opal_common_ofi_open();
 }
 
 static int
@@ -302,9 +300,7 @@ ompi_mtl_ofi_component_close(void)
 #if OPAL_CUDA_SUPPORT
     mca_common_cuda_fini();
 #endif
-    opal_common_ofi_mca_deregister();
-    opal_common_ofi_fini();
-    return OMPI_SUCCESS;
+    return opal_common_ofi_close();
 }
 
 int
@@ -582,8 +578,6 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
     int universe_size;
     char *univ_size_str;
 
-    opal_common_ofi_mca_register();
-
     opal_output_verbose(1, opal_common_ofi.output,
                         "%s:%d: mtl:ofi:provider_include = \"%s\"\n",
                         __FILE__, __LINE__, *opal_common_ofi.prov_include);
@@ -893,6 +887,20 @@ ompi_mtl_ofi_component_init(bool enable_progress_threads,
         }
     }
 
+    /* this must be called during single threaded part of the code and
+     * before Libfabric configures its memory monitors.  Easiest to do
+     * that before domain open.  Silently ignore not-supported errors,
+     * as they are not critical to program correctness, but only
+     * indicate that LIbfabric will have to pick a different, possibly
+     * less optimial, monitor. */
+    ret = opal_common_ofi_export_memory_monitor();
+    if (0 != ret && -FI_ENOSYS != ret) {
+        opal_output_verbose(1, opal_common_ofi.output,
+                            "Failed to inject Libfabric memory monitor: %s",
+                             fi_strerror(-ret));
+    }
+
+
     /**
      * Open fabric
      * The getinfo struct returns a fabric attribute struct that can be used to

opal/mca/btl/ofi/btl_ofi_component.c

@@ -123,6 +123,7 @@ static int validate_info(struct fi_info *info, uint64_t required_caps, char **in
 /* Register the MCA parameters */
 static int mca_btl_ofi_component_register(void)
 {
+    int ret;
     char *msg;
     mca_btl_ofi_module_t *module = &mca_btl_ofi_module_template;
 
@@ -191,27 +192,30 @@ static int mca_btl_ofi_component_register(void)
     /* for now we want this component to lose to the MTL. */
     module->super.btl_exclusivity = MCA_BTL_EXCLUSIVITY_HIGH - 50;
 
-    opal_common_ofi_register_mca_variables(&mca_btl_ofi_component.super.btl_version);
+    ret = opal_common_ofi_mca_register(&mca_btl_ofi_component.super.btl_version);
+    if (OPAL_SUCCESS != ret) {
+        return ret;
+    }
 
     return mca_btl_base_param_register(&mca_btl_ofi_component.super.btl_version, &module->super);
 }
 
 static int mca_btl_ofi_component_open(void)
 {
     mca_btl_ofi_component.module_count = 0;
-    return opal_common_ofi_init();
+    return opal_common_ofi_open();
 }
 
 /*
  * component cleanup - sanity checking of queue lengths
  */
 static int mca_btl_ofi_component_close(void)
 {
-    opal_common_ofi_mca_deregister();
-    opal_common_ofi_fini();
+    int ret;
+    ret = opal_common_ofi_close();
     /* If we don't sleep, sockets provider freaks out. Ummm this is a scary comment */
     sleep(1);
-    return OPAL_SUCCESS;
+    return ret;
 }
 
 void mca_btl_ofi_exit(void)
@@ -259,8 +263,6 @@ static mca_btl_base_module_t **mca_btl_ofi_component_init(int *num_btl_modules,
     struct fi_domain_attr domain_attr = {0};
     uint64_t required_caps;
 
-    opal_common_ofi_mca_register();
-
     switch (mca_btl_ofi_component.mode) {
 
     case MCA_BTL_OFI_MODE_TWO_SIDED:
@@ -444,6 +446,19 @@ static int mca_btl_ofi_init_device(struct fi_info *info)
      * to prevent races. */
     mca_btl_ofi_rcache_init(module);
 
+    /* for similar reasons to the rcache call, this must be called
+     * during single threaded part of the code and before Libfabric
+     * configures its memory monitors.  Easiest to do that before
+     * domain open.  Silently ignore not-supported errors, as they
+     * are not critical to program correctness, but only indicate
+     * that LIbfabric will have to pick a different, possibly less
+     * optimial, monitor. */
+    rc = opal_common_ofi_export_memory_monitor();
+    if (0 != rc && -FI_ENOSYS != rc) {
+        BTL_VERBOSE(("Failed to inject Libfabric memory monitor: %s",
+                     fi_strerror(-rc)));
+    }
+
     linux_device_name = info->domain_attr->name;
     BTL_VERBOSE(
         ("initializing dev:%s provider:%s", linux_device_name, info->fabric_attr->prov_name));