smcuda: fixes when using enable-mca-dso

related to open-mpi#11354 Signed-off-by: Howard Pritchard <howardp@lanl.gov>
yli137 · Jun 20, 2023 · a279541 · a279541
1 parent ba85e65
commit a279541
Show file tree

Hide file tree

Showing 3 changed files with 34 additions and 16 deletions.
diff --git a/opal/mca/btl/smcuda/btl_smcuda_accelerator.c b/opal/mca/btl/smcuda/btl_smcuda_accelerator.c
@@ -1,6 +1,8 @@
 /* 
  * Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * Copyright (c) 2022      IBM Corporation.  All rights reserved.
+ * Copyright (c) 2023      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -35,6 +37,8 @@ static int accelerator_event_max = 400;
 static int accelerator_event_ipc_most = 0;
 static bool smcuda_accelerator_initialized = false;
 
+static void mca_btl_smcuda_accelerator_fini(void);
+
 int mca_btl_smcuda_accelerator_init(void)
 {
     int rc = OPAL_SUCCESS;
@@ -79,6 +83,14 @@ int mca_btl_smcuda_accelerator_init(void)
         goto cleanup_and_error;
     }
 
+    /*
+     * add smcuda acclerator fini code to opal's list of cleanup functions.
+     * Cleanups are called before all the MCA frameworks are closed, so by
+     * adding this function to the callback list, we avoid issues with ordering
+     * of the closing of the BTL framework with the accelerator framework, etc. etc.
+     */
+    opal_finalize_register_cleanup(mca_btl_smcuda_accelerator_fini);
+
     smcuda_accelerator_initialized = true;
 
 cleanup_and_error:
@@ -103,7 +115,7 @@ int mca_btl_smcuda_accelerator_init(void)
     return rc;
 }
 
-void mca_btl_smcuda_accelerator_fini(void)
+static void mca_btl_smcuda_accelerator_fini(void)
 {
     int i;
 

diff --git a/opal/mca/btl/smcuda/btl_smcuda_accelerator.h b/opal/mca/btl/smcuda/btl_smcuda_accelerator.h
@@ -1,6 +1,8 @@
 /*
  * Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights
  *                         reserved.
+ * Copyright (c) 2023      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -17,7 +19,6 @@
 #include "opal/mca/btl/btl.h"
 
 OPAL_DECLSPEC int mca_btl_smcuda_accelerator_init(void);
-OPAL_DECLSPEC void mca_btl_smcuda_accelerator_fini(void);
 OPAL_DECLSPEC int mca_btl_smcuda_progress_one_ipc_event(struct mca_btl_base_descriptor_t **frag);
 OPAL_DECLSPEC int mca_btl_smcuda_memcpy(void *dst, void *src, size_t amount, char *msg,
                            struct mca_btl_base_descriptor_t *frag);

diff --git a/opal/mca/btl/smcuda/btl_smcuda_component.c b/opal/mca/btl/smcuda/btl_smcuda_component.c
@@ -18,6 +18,8 @@
  * Copyright (c) 2014      Intel, Inc. All rights reserved.
  * Copyright (c) 2018-2022 Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * Copyright (c) 2022      IBM Corporation.  All rights reserved.
+ * Copyright (c) 2023      Triad National Security, LLC. All rights
+ *                         reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -69,6 +71,8 @@ static int smcuda_register(void);
 static mca_btl_base_module_t **
 mca_btl_smcuda_component_init(int *num_btls, bool enable_progress_threads, bool enable_mpi_threads);
 
+static void mca_btl_smcuda_component_fini(void);
+
 typedef enum {
     MCA_BTL_SM_RNDV_MOD_SM = 0,
     MCA_BTL_SM_RNDV_MOD_MPOOL
@@ -214,7 +218,7 @@ static int smcuda_register(void)
     if (0 == mca_btl_smcuda.super.btl_accelerator_eager_limit) {
         mca_btl_smcuda.super.btl_accelerator_eager_limit = SIZE_MAX; /* magic number */
     }
-#endif
+#endif /* OPAL_CUDA_SUPPORT */
     return mca_btl_smcuda_component_verify();
 }
 
@@ -260,16 +264,24 @@ static int mca_btl_smcuda_component_open(void)
     OBJ_CONSTRUCT(&mca_btl_smcuda_component.sm_frags_max, opal_free_list_t);
     OBJ_CONSTRUCT(&mca_btl_smcuda_component.sm_frags_user, opal_free_list_t);
     OBJ_CONSTRUCT(&mca_btl_smcuda_component.pending_send_fl, opal_free_list_t);
+
+    opal_finalize_register_cleanup(mca_btl_smcuda_component_fini);
+
+    return OPAL_SUCCESS;
+}
+
+static int mca_btl_smcuda_component_close(void)
+{
     return OPAL_SUCCESS;
 }
 
 /*
  * component cleanup - sanity checking of queue lengths
  */
 
-static int mca_btl_smcuda_component_close(void)
+static void mca_btl_smcuda_component_fini(void)
 {
-    int return_value = OPAL_SUCCESS;
+    int rc;
 
     OBJ_DESTRUCT(&mca_btl_smcuda_component.sm_lock);
     /**
@@ -282,11 +294,10 @@ static int mca_btl_smcuda_component_close(void)
 
     /* unmap the shared memory control structure */
     if (mca_btl_smcuda_component.sm_seg != NULL) {
-        return_value = mca_common_sm_fini(mca_btl_smcuda_component.sm_seg);
-        if (OPAL_SUCCESS != return_value) {
-            return_value = OPAL_ERROR;
+        rc = mca_common_sm_fini(mca_btl_smcuda_component.sm_seg);
+        if (OPAL_SUCCESS != rc) {
             opal_output(0, " mca_common_sm_fini failed\n");
-            goto CLEANUP;
+            return;
         }
 
         /* unlink file, so that it will be deleted when all references
@@ -310,13 +321,7 @@ static int mca_btl_smcuda_component_close(void)
         unlink(mca_btl_smcuda_component.sm_fifo_path);
     }
 #endif
-
-CLEANUP:
-
-    mca_btl_smcuda_accelerator_fini();
-
-    /* return */
-    return return_value;
+    return;
 }
 
 /*