Merge pull request open-mpi#7501 from awlauria/finalize_leaks_ggouaillardet_awlauria

Finalize memchecker calls and one memory leak

Author: awlauria

ompi/dpm/dpm.c

@@ -16,7 +16,8 @@
  * Copyright (c) 2011-2015 Los Alamos National Security, LLC.  All rights
  *                         reserved.
  * Copyright (c) 2013-2019 Intel, Inc.  All rights reserved.
- * Copyright (c) 2014-2017 Research Organization for Information Science
+ * Copyright (c) 2013-2017 Intel, Inc. All rights reserved.
+ * Copyright (c) 2014-2020 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2018      Amazon.com, Inc. or its affiliates.  All Rights reserved.
  * $COPYRIGHT$
@@ -1113,6 +1114,16 @@ int ompi_dpm_finalize(void)
     return OMPI_SUCCESS;
 }
 
+static void cleanup_dpm_disconnect_objs(ompi_dpm_disconnect_obj **objs, int count)
+{
+    for(int i = 0; i < count; i++) {
+        if (NULL != objs[i]->reqs) {
+            free(objs[i]->reqs);
+        }
+        free(objs[i]);
+    }
+    free(objs);
+}
 
 /**********************************************************************/
 /**********************************************************************/
@@ -1126,7 +1137,7 @@ int ompi_dpm_dyn_finalize(void)
     ompi_communicator_t *comm=NULL;
 
     if (1 <ompi_comm_num_dyncomm) {
-        objs = (ompi_dpm_disconnect_obj**)malloc(ompi_comm_num_dyncomm*
+        objs = (ompi_dpm_disconnect_obj**)malloc(ompi_comm_num_dyncomm *
                                sizeof(ompi_dpm_disconnect_obj*));
         if (NULL == objs) {
             return OMPI_ERR_OUT_OF_RESOURCE;
@@ -1140,13 +1151,12 @@ int ompi_dpm_dyn_finalize(void)
             }
         }
 
-        if (j != ompi_comm_num_dyncomm+1) {
-            free(objs);
+        if (j != ompi_comm_num_dyncomm) {
+            cleanup_dpm_disconnect_objs(objs, j);
             return OMPI_ERROR;
         }
 
         disconnect_waitall(ompi_comm_num_dyncomm, objs);
-        free(objs);
     }
 
     return OMPI_SUCCESS;
@@ -1265,13 +1275,7 @@ static int disconnect_waitall (int count, ompi_dpm_disconnect_obj **objs)
     ret = ompi_request_wait_all(2*totalcount, reqs, MPI_STATUSES_IGNORE);
 
     /* Finally, free everything */
-    for (i=0; i< count; i++ ) {
-        if (NULL != objs[i]->reqs ) {
-            free(objs[i]->reqs );
-        }
-        free(objs[i]);
-    }
-
+    cleanup_dpm_disconnect_objs(objs, count);
     free(reqs);
 
     return ret;

ompi/include/ompi/memchecker.h

@@ -6,11 +6,11 @@
  *                         reserved.
  * Copyright (c) 2009      Oak Ridge National Labs.  All rights reserved.
  * Copyright (c) 2012-2013 Inria.  All rights reserved.
- * Copyright (c) 2014-2017 Research Organization for Information Science
+ * Copyright (c) 2014-2018 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2014      Intel, Inc. All rights reserved.
  *
- * Copyright (c) 2016 Cisco Systems, Inc.  All rights reserved.
+ * Copyright (c) 2016      Cisco Systems, Inc.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -107,7 +107,7 @@ static inline int memchecker_call (int (*f)(void *, size_t), const void * addr,
 
     if( datatype->super.size == (size_t) (datatype->super.true_ub - datatype->super.true_lb) ) {
         /*  We have a contiguous type. */
-        f( (void*)addr , datatype->super.size * count );
+        f( (void*)((char *)addr+datatype->super.true_lb), datatype->super.size * count );
     } else {
         /* Now we got a noncontigous type. */
         uint32_t         elem_pos = 0, i;
@@ -128,7 +128,18 @@ static inline int memchecker_call (int (*f)(void *, size_t), const void * addr,
 
             while( pElem->elem.common.flags & OPAL_DATATYPE_FLAG_DATA ) {
                 /* now here we have a basic datatype */
-                f( (void *)(source_base + pElem->elem.disp), pElem->elem.count*pElem->elem.extent );
+                size_t blength = opal_datatype_basicDatatypes[pElem->elem.common.type]->size;
+                if ((size_t)pElem->elem.extent == blength) {
+                    /* block is made of contiguous basic datatype */
+                    f( (void *)(source_base + pElem->elem.disp), pElem->elem.count*blength );
+                } else {
+                    uint32_t j;
+                    ptrdiff_t offset;
+                    for (j = 0, offset=0; j < pElem->elem.count; j++) {
+                        f( (void *)(source_base + pElem->elem.disp + offset), blength);
+                        offset += pElem->elem.extent;
+                    }
+                }
                 elem_pos++;       /* advance to the next data */
                 pElem = &(description[elem_pos]);
                 continue;

ompi/mca/osc/pt2pt/osc_pt2pt_frag.h

@@ -3,6 +3,8 @@
  * Copyright (c) 2012      Sandia National Laboratories.  All rights reserved.
  * Copyright (c) 2014-2017 Los Alamos National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2018      Research Organization for Information Science
+ *                         and Technology (RIST). All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -91,11 +93,18 @@ static inline ompi_osc_pt2pt_frag_t *ompi_osc_pt2pt_frag_alloc_non_buffered (omp
 
     curr->header->base.type = OMPI_OSC_PT2PT_HDR_TYPE_FRAG;
     curr->header->base.flags = OMPI_OSC_PT2PT_HDR_FLAG_VALID;
+#if OPAL_ENABLE_DEBUG
+    curr->header->padding[0] = 0;
+    curr->header->padding[1] = 0;
+#endif
     if (module->passive_target_access_epoch) {
         curr->header->base.flags |= OMPI_OSC_PT2PT_HDR_FLAG_PASSIVE_TARGET;
     }
     curr->header->source = ompi_comm_rank(module->comm);
     curr->header->num_ops = 1;
+#if OPAL_ENABLE_DEBUG
+    curr->header->pad = 0;
+#endif
 
     return curr;
 }

ompi/mca/osc/pt2pt/osc_pt2pt_header.h

@@ -13,7 +13,7 @@
  * Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2010      Oracle and/or its affiliates.  All rights reserved.
  * Copyright (c) 2012-2013 Sandia National Laboratories.  All rights reserved.
- * Copyright (c) 2015      Research Organization for Information Science
+ * Copyright (c) 2015-2018 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * $COPYRIGHT$
  *
@@ -100,7 +100,7 @@ typedef struct ompi_osc_pt2pt_header_get_t ompi_osc_pt2pt_header_get_t;
 
 struct ompi_osc_pt2pt_header_complete_t {
     ompi_osc_pt2pt_header_base_t base;
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t padding[2];
 #endif
     int frag_count;
@@ -123,7 +123,7 @@ typedef struct ompi_osc_pt2pt_header_post_t ompi_osc_pt2pt_header_post_t;
 
 struct ompi_osc_pt2pt_header_lock_t {
     ompi_osc_pt2pt_header_base_t base;
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t padding[2];
 #endif
     int32_t lock_type;
@@ -133,14 +133,17 @@ typedef struct ompi_osc_pt2pt_header_lock_t ompi_osc_pt2pt_header_lock_t;
 
 struct ompi_osc_pt2pt_header_lock_ack_t {
     ompi_osc_pt2pt_header_base_t base;
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
+    uint8_t padding[2];
+#endif
     uint32_t source;
     uint64_t lock_ptr;
 };
 typedef struct ompi_osc_pt2pt_header_lock_ack_t ompi_osc_pt2pt_header_lock_ack_t;
 
 struct ompi_osc_pt2pt_header_unlock_t {
     ompi_osc_pt2pt_header_base_t base;
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t padding[2];
 #endif
     int32_t lock_type;
@@ -151,7 +154,7 @@ typedef struct ompi_osc_pt2pt_header_unlock_t ompi_osc_pt2pt_header_unlock_t;
 
 struct ompi_osc_pt2pt_header_unlock_ack_t {
     ompi_osc_pt2pt_header_base_t base;
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t padding[6];
 #endif
     uint64_t lock_ptr;
@@ -160,7 +163,7 @@ typedef struct ompi_osc_pt2pt_header_unlock_ack_t ompi_osc_pt2pt_header_unlock_a
 
 struct ompi_osc_pt2pt_header_flush_t {
     ompi_osc_pt2pt_header_base_t base;
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t padding[2];
 #endif
     uint32_t frag_count;
@@ -170,7 +173,7 @@ typedef struct ompi_osc_pt2pt_header_flush_t ompi_osc_pt2pt_header_flush_t;
 
 struct ompi_osc_pt2pt_header_flush_ack_t {
     ompi_osc_pt2pt_header_base_t base;
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t padding[6];
 #endif
     uint64_t lock_ptr;
@@ -179,6 +182,9 @@ typedef struct ompi_osc_pt2pt_header_flush_ack_t ompi_osc_pt2pt_header_flush_ack
 
 struct ompi_osc_pt2pt_frag_header_t {
     ompi_osc_pt2pt_header_base_t base;
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
+    uint8_t padding[2];
+#endif
     uint32_t source; /* rank in window of source process */
     opal_atomic_int32_t num_ops; /* number of operations in this buffer */
     uint32_t pad; /* ensure the fragment header is a multiple of 8 bytes */

ompi/mca/pml/ob1/pml_ob1_cuda.c

@@ -36,7 +36,6 @@
 #include "pml_ob1_recvreq.h"
 #include "pml_ob1_sendreq.h"
 #include "ompi/mca/bml/base/base.h"
-#include "ompi/memchecker.h"
 
 size_t mca_pml_ob1_rdma_cuda_btls(
     mca_bml_base_endpoint_t* bml_endpoint,

ompi/mca/pml/ob1/pml_ob1_hdr.h

@@ -13,6 +13,8 @@
  * Copyright (c) 2009      IBM Corporation.  All rights reserved.
  * Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2018      Research Organization for Information Science
+ *                         and Technology (RIST). All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -84,11 +86,11 @@ struct mca_pml_ob1_match_hdr_t {
     int32_t  hdr_src;                      /**< source rank */
     int32_t  hdr_tag;                      /**< user tag */
     uint16_t hdr_seq;                      /**< message sequence number */
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t  hdr_padding[2];               /**< explicitly pad to 16 bytes.  Compilers seem to already prefer to do this, but make it explicit just in case */
 #endif
 };
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
 #define OMPI_PML_OB1_MATCH_HDR_LEN  16
 #else
 #define OMPI_PML_OB1_MATCH_HDR_LEN  14
@@ -104,7 +106,7 @@ static inline void mca_pml_ob1_match_hdr_prepare (mca_pml_ob1_match_hdr_t *hdr,
     hdr->hdr_src = hdr_src;
     hdr->hdr_tag = hdr_tag;
     hdr->hdr_seq = hdr_seq;
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT && OPAL_ENABLE_DEBUG
+#if OPAL_ENABLE_DEBUG
     hdr->hdr_padding[0] = 0;
     hdr->hdr_padding[1] = 0;
 #endif
@@ -169,7 +171,7 @@ static inline void mca_pml_ob1_rendezvous_hdr_prepare (mca_pml_ob1_rendezvous_hd
  */
 struct mca_pml_ob1_rget_hdr_t {
     mca_pml_ob1_rendezvous_hdr_t hdr_rndv;
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t hdr_padding[4];
 #endif
     opal_ptr_t hdr_frag;                      /**< source fragment (for fin) */
@@ -185,7 +187,7 @@ static inline void mca_pml_ob1_rget_hdr_prepare (mca_pml_ob1_rget_hdr_t *hdr, ui
 {
     mca_pml_ob1_rendezvous_hdr_prepare (&hdr->hdr_rndv, MCA_PML_OB1_HDR_TYPE_RGET, hdr_flags,
                                         hdr_ctx, hdr_src, hdr_tag, hdr_seq, hdr_msg_length, hdr_src_req);
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT && OPAL_ENABLE_DEBUG
+#if OPAL_ENABLE_DEBUG
     hdr->hdr_padding[0] = 0;
     hdr->hdr_padding[1] = 0;
     hdr->hdr_padding[2] = 0;
@@ -215,7 +217,7 @@ static inline void mca_pml_ob1_rget_hdr_prepare (mca_pml_ob1_rget_hdr_t *hdr, ui
  */
 struct mca_pml_ob1_frag_hdr_t {
     mca_pml_ob1_common_hdr_t hdr_common;     /**< common attributes */
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t hdr_padding[6];
 #endif
     uint64_t hdr_frag_offset;                /**< offset into message */
@@ -229,7 +231,7 @@ static inline void mca_pml_ob1_frag_hdr_prepare (mca_pml_ob1_frag_hdr_t *hdr, ui
                                                  uint64_t hdr_dst_req)
 {
     mca_pml_ob1_common_hdr_prepare (&hdr->hdr_common, MCA_PML_OB1_HDR_TYPE_FRAG, hdr_flags);
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT && OPAL_ENABLE_DEBUG
+#if OPAL_ENABLE_DEBUG
     hdr->hdr_padding[0] = 0;
     hdr->hdr_padding[1] = 0;
     hdr->hdr_padding[2] = 0;
@@ -260,7 +262,7 @@ static inline void mca_pml_ob1_frag_hdr_prepare (mca_pml_ob1_frag_hdr_t *hdr, ui
 
 struct mca_pml_ob1_ack_hdr_t {
     mca_pml_ob1_common_hdr_t hdr_common;      /**< common attributes */
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
     uint8_t hdr_padding[6];
 #endif
     opal_ptr_t hdr_src_req;                   /**< source request */
@@ -275,7 +277,7 @@ static inline void mca_pml_ob1_ack_hdr_prepare (mca_pml_ob1_ack_hdr_t *hdr, uint
                                                 uint64_t hdr_send_offset, uint64_t hdr_send_size)
 {
     mca_pml_ob1_common_hdr_prepare (&hdr->hdr_common, MCA_PML_OB1_HDR_TYPE_ACK, hdr_flags);
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT && OPAL_ENABLE_DEBUG
+#if OPAL_ENABLE_DEBUG
     hdr->hdr_padding[0] = 0;
     hdr->hdr_padding[1] = 0;
     hdr->hdr_padding[2] = 0;
@@ -313,8 +315,8 @@ static inline void mca_pml_ob1_ack_hdr_prepare (mca_pml_ob1_ack_hdr_t *hdr, uint
 
 struct mca_pml_ob1_rdma_hdr_t {
     mca_pml_ob1_common_hdr_t hdr_common;      /**< common attributes */
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
-    uint8_t hdr_padding[2];                   /** two to pad out the hdr to a 4 byte alignment.  hdr_req will then be 8 byte aligned after 4 for hdr_seg_cnt */
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
+    uint8_t hdr_padding[6];                   /** two to pad out the hdr to a 4 byte alignment.  hdr_req will then be 8 byte aligned after 4 for hdr_seg_cnt */
 #endif
     /* TODO: add real support for multiple destination segments */
     opal_ptr_t hdr_req;                       /**< destination request */
@@ -334,9 +336,13 @@ static inline void mca_pml_ob1_rdma_hdr_prepare (mca_pml_ob1_rdma_hdr_t *hdr, ui
                                                  size_t local_handle_size)
 {
     mca_pml_ob1_common_hdr_prepare (&hdr->hdr_common, MCA_PML_OB1_HDR_TYPE_PUT, hdr_flags);
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT && OPAL_ENABLE_DEBUG
+#if OPAL_ENABLE_DEBUG
     hdr->hdr_padding[0] = 0;
     hdr->hdr_padding[1] = 0;
+    hdr->hdr_padding[2] = 0;
+    hdr->hdr_padding[3] = 0;
+    hdr->hdr_padding[4] = 0;
+    hdr->hdr_padding[5] = 0;
 #endif
     hdr->hdr_req.lval = hdr_req;
     hdr->hdr_frag.pval = hdr_frag;
@@ -371,8 +377,8 @@ static inline void mca_pml_ob1_rdma_hdr_prepare (mca_pml_ob1_rdma_hdr_t *hdr, ui
 
 struct mca_pml_ob1_fin_hdr_t {
     mca_pml_ob1_common_hdr_t hdr_common;      /**< common attributes */
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
-    uint8_t hdr_padding[2];
+#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT || OPAL_ENABLE_DEBUG
+    uint8_t hdr_padding[6];
 #endif
     int64_t hdr_size;                         /**< number of bytes completed (positive), error code (negative) */
     opal_ptr_t hdr_frag;                      /**< completed RDMA fragment */
@@ -383,9 +389,13 @@ static inline void mca_pml_ob1_fin_hdr_prepare (mca_pml_ob1_fin_hdr_t *hdr, uint
                                                 uint64_t hdr_frag, int64_t hdr_size)
 {
     mca_pml_ob1_common_hdr_prepare (&hdr->hdr_common, MCA_PML_OB1_HDR_TYPE_FIN, hdr_flags);
-#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT && OPAL_ENABLE_DEBUG
+#if OPAL_ENABLE_DEBUG
     hdr->hdr_padding[0] = 0;
     hdr->hdr_padding[1] = 0;
+    hdr->hdr_padding[2] = 0;
+    hdr->hdr_padding[3] = 0;
+    hdr->hdr_padding[4] = 0;
+    hdr->hdr_padding[5] = 0;
 #endif
     hdr->hdr_frag.lval = hdr_frag;
     hdr->hdr_size      = hdr_size;

ompi/mca/pml/ob1/pml_ob1_recvfrag.c

@@ -40,7 +40,6 @@
 #include "ompi/communicator/communicator.h"
 #include "ompi/mca/pml/pml.h"
 #include "ompi/peruse/peruse-internal.h"
-#include "ompi/memchecker.h"
 #include "ompi/runtime/ompi_spc.h"
 
 #include "pml_ob1.h"

ompi/mca/pml/ob1/pml_ob1_recvreq.c

@@ -40,7 +40,7 @@
 #include "pml_ob1_sendreq.h"
 #include "pml_ob1_rdmafrag.h"
 #include "ompi/mca/bml/base/base.h"
-#include "ompi/memchecker.h"
+
 #if OPAL_CUDA_SUPPORT
 #include "opal/datatype/opal_datatype_cuda.h"
 #include "opal/mca/common/cuda/common_cuda.h"

ompi/mca/pml/ob1/pml_ob1_sendreq.c

@@ -39,7 +39,6 @@
 #include "pml_ob1_rdmafrag.h"
 #include "pml_ob1_recvreq.h"
 #include "ompi/mca/bml/base/base.h"
-#include "ompi/memchecker.h"
 
 OBJ_CLASS_INSTANCE(mca_pml_ob1_send_range_t, opal_free_list_item_t,
         NULL, NULL);