v3.0.x: hook madvaise to fix threading memory allocation issues

opal/mca/memory/patcher/memory_patcher_component.c

@@ -59,11 +59,6 @@ static int patcher_query (int *);
 
 static int mca_memory_patcher_priority;
 
-/* NTH: we can't currently allow madvise to be intercepted due to a
- * deadlock when running with glibc. In the future, we may re-enable
- * this hook if the deadlock can be resolved. */
-#define WANT_INTERCEPT_MADVISE 0
-
 opal_memory_patcher_component_t mca_memory_patcher_component = {
     .super = {
         .memoryc_version = {
@@ -249,7 +244,6 @@ static void *intercept_mremap (void *start, size_t oldlen, void *new_address, si
 
 #endif
 
-#if WANT_INTERCEPT_MADVISE
 #if defined (SYS_madvise)
 
 static int (*original_madvise) (void *, size_t, int);
@@ -284,7 +278,6 @@ static int intercept_madvise (void *start, size_t length, int advice)
 }
 
 #endif
-#endif // WANT_INTERCEPT_MADVISE
 
 #if defined SYS_brk
 
@@ -503,14 +496,12 @@ static int patcher_open (void)
     }
 #endif
 
-#if WANT_INTERCEPT_MADVISE
 #if defined (SYS_madvise)
     rc = opal_patcher->patch_symbol ("madvise", (uintptr_t)intercept_madvise, (uintptr_t *) &original_madvise);
     if (OPAL_SUCCESS != rc) {
         return rc;
     }
 #endif
-#endif // WANT_INTERCEPT_MADVISE
 
 #if defined(SYS_shmdt) && defined(__linux__)
     rc = opal_patcher->patch_symbol ("shmdt", (uintptr_t) intercept_shmdt, (uintptr_t *) &original_shmdt);

opal/mca/rcache/base/base.h

@@ -58,6 +58,12 @@ OPAL_DECLSPEC mca_rcache_base_component_t *mca_rcache_base_component_lookup(cons
 OPAL_DECLSPEC mca_rcache_base_module_t *mca_rcache_base_module_lookup (const char *name);
 OPAL_DECLSPEC int mca_rcache_base_module_destroy(mca_rcache_base_module_t *module);
 
+extern opal_free_list_t mca_rcache_base_vma_tree_items;
+extern bool mca_rcache_base_vma_tree_items_inited;
+extern unsigned int mca_rcache_base_vma_tree_items_min;
+extern int mca_rcache_base_vma_tree_items_max;
+extern unsigned int mca_rcache_base_vma_tree_items_inc;
+
 /* only used within base -- no need to DECLSPEC */
 extern int mca_rcache_base_used_mem_hooks;
 

opal/mca/rcache/base/rcache_base_create.c

@@ -29,6 +29,7 @@
 #include "opal/mca/rcache/rcache.h"
 #include "opal/mca/rcache/base/base.h"
 #include "opal/mca/rcache/base/rcache_base_mem_cb.h"
+#include "rcache_base_vma_tree.h"
 #include "opal/util/show_help.h"
 #include "opal/util/proc.h"
 

opal/mca/rcache/base/rcache_base_frame.c

@@ -27,12 +27,17 @@
 
 #include "opal/mca/mca.h"
 #include "opal/mca/base/base.h"
+#include "opal/mca/base/mca_base_pvar.h"
 #include "opal/mca/rcache/rcache.h"
 #include "opal/mca/rcache/base/base.h"
 #include "opal/memoryhooks/memory.h"
 #include "opal/constants.h"
 #include "rcache_base_mem_cb.h"
 
+/* two-level macro for stringifying a number */
+#define STRINGIFYX(x) #x
+#define STRINGIFY(x) STRINGIFYX(x)
+
 /*
  * The following file was created by configure.  It contains extern
  * statements and the definition of an array of pointers to each
@@ -59,12 +64,19 @@ static void mca_rcache_base_registration_constructor( mca_rcache_base_registrati
 OBJ_CLASS_INSTANCE(mca_rcache_base_registration_t, opal_free_list_item_t,
                    mca_rcache_base_registration_constructor, NULL);
 
+#define TREE_ITEMS_MIN 2048
+#define TREE_ITEMS_MAX 16384
+#define TREE_ITEMS_INC 2048
 
 /*
  * Global variables
  */
 opal_list_t mca_rcache_base_modules = {{0}};
-
+opal_free_list_t mca_rcache_base_vma_tree_items = {{0}};
+bool mca_rcache_base_vma_tree_items_inited = false;
+unsigned int mca_rcache_base_vma_tree_items_min = TREE_ITEMS_MIN;
+int mca_rcache_base_vma_tree_items_max = TREE_ITEMS_MAX;
+unsigned int mca_rcache_base_vma_tree_items_inc = TREE_ITEMS_INC;
 
 OBJ_CLASS_INSTANCE(mca_rcache_base_selected_module_t, opal_list_item_t, NULL, NULL);
 
@@ -100,8 +112,11 @@ static int mca_rcache_base_close(void)
            time between now and end of application (even post main()!) */
         (void) mca_base_framework_close (&opal_memory_base_framework);
     }
-    /* All done */
 
+    OBJ_DESTRUCT(&mca_rcache_base_vma_tree_items);
+    mca_rcache_base_vma_tree_items_inited = false;
+
+    /* All done */
     /* Close all remaining available components */
     return mca_base_framework_components_close(&opal_rcache_base_framework, NULL);
 }
@@ -117,11 +132,42 @@ static int mca_rcache_base_open(mca_base_open_flag_t flags)
 
     OBJ_CONSTRUCT(&mca_rcache_base_modules, opal_list_t);
 
+    /* the free list is only initialized when a VMA tree is created */
+    OBJ_CONSTRUCT(&mca_rcache_base_vma_tree_items, opal_free_list_t);
+
      /* Open up all available components */
     return mca_base_framework_components_open(&opal_rcache_base_framework, flags);
 }
 
-MCA_BASE_FRAMEWORK_DECLARE(opal, rcache, "OPAL Registration Cache", NULL,
+static int mca_rcache_base_register_mca_variables (mca_base_register_flag_t flags)
+{
+
+    mca_rcache_base_vma_tree_items_min = TREE_ITEMS_MIN;
+    (void) mca_base_framework_var_register (&opal_rcache_base_framework, "vma_tree_items_min",
+                                            "Minimum number of VMA tree items to allocate (default: "
+                                            STRINGIFY(TREE_ITEMS_MIN) ")", MCA_BASE_VAR_TYPE_UNSIGNED_INT,
+                                            NULL, MCA_BASE_VAR_BIND_NO_OBJECT, 0, OPAL_INFO_LVL_6,
+                                            MCA_BASE_VAR_SCOPE_READONLY, &mca_rcache_base_vma_tree_items_min);
+
+    mca_rcache_base_vma_tree_items_max = TREE_ITEMS_MAX;
+    (void) mca_base_framework_var_register (&opal_rcache_base_framework, "vma_tree_items_max",
+                                            "Maximum number of VMA tree items to allocate (default: "
+                                            STRINGIFY(TREE_ITEMS_MAX) ", -1: unlimited)", MCA_BASE_VAR_TYPE_INT,
+                                            NULL, MCA_BASE_VAR_BIND_NO_OBJECT, 0, OPAL_INFO_LVL_6,
+                                            MCA_BASE_VAR_SCOPE_READONLY, &mca_rcache_base_vma_tree_items_max);
+
+    mca_rcache_base_vma_tree_items_inc = TREE_ITEMS_INC;
+    (void) mca_base_framework_var_register (&opal_rcache_base_framework, "vma_tree_items_inc",
+                                            "Number of VMA tree items to allocate at a time (default: "
+                                            STRINGIFY(TREE_ITEMS_INC) ")", MCA_BASE_VAR_TYPE_UNSIGNED_INT,
+                                            NULL, MCA_BASE_VAR_BIND_NO_OBJECT, 0, OPAL_INFO_LVL_6,
+                                            MCA_BASE_VAR_SCOPE_READONLY, &mca_rcache_base_vma_tree_items_inc);
+
+    return OPAL_SUCCESS;
+}
+
+MCA_BASE_FRAMEWORK_DECLARE(opal, rcache, "OPAL Registration Cache",
+                           mca_rcache_base_register_mca_variables,
                            mca_rcache_base_open, mca_rcache_base_close,
                            mca_rcache_base_static_components, 0);
 

opal/mca/rcache/base/rcache_base_vma.c

@@ -14,7 +14,7 @@
  * Copyright (c) 2009-2013 Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2009      IBM Corporation.  All rights reserved.
  * Copyright (c) 2013      NVIDIA Corporation.  All rights reserved.
- * Copyright (c) 2015-2016 Los Alamos National Security, LLC. All rights
+ * Copyright (c) 2015-2017 Los Alamos National Security, LLC. All rights
  *                         reserved.
  *
  * $COPYRIGHT$
@@ -31,6 +31,7 @@
 #include "opal/mca/rcache/rcache.h"
 #include "rcache_base_vma.h"
 #include "rcache_base_vma_tree.h"
+#include "opal/mca/rcache/base/base.h"
 
 /**
  * Initialize the rcache
@@ -52,6 +53,14 @@ OBJ_CLASS_INSTANCE(mca_rcache_base_vma_module_t, opal_object_t,
 
 mca_rcache_base_vma_module_t *mca_rcache_base_vma_module_alloc (void)
 {
+    if (!mca_rcache_base_vma_tree_items_inited) {
+        opal_free_list_init (&mca_rcache_base_vma_tree_items, sizeof (mca_rcache_base_vma_item_t),
+                             8, OBJ_CLASS(mca_rcache_base_vma_item_t), 0, 8,
+                             mca_rcache_base_vma_tree_items_min, mca_rcache_base_vma_tree_items_max,
+                             mca_rcache_base_vma_tree_items_inc, NULL, 0, NULL, NULL, NULL);
+        mca_rcache_base_vma_tree_items_inited = true;
+    }
+
     return OBJ_NEW(mca_rcache_base_vma_module_t);
 }