Fix race condition in coll_han_alltoall.c

Add a barrier in cases where reading memory directly from neighbors
to fix an issue where the send buffer is overwritten prematurely in
coll_han_alltoall.c

Bug discovered and patched at AccelCom, Barcelona Supercomputing Center

Signed-off-by: Petter Sandas <petter.sandas@bsc.es>

Author: Petter-Programs

ompi/mca/coll/han/coll_han_alltoall.c

@@ -1,6 +1,8 @@
 /*
  * Copyright (c) 2024      Amazon.com, Inc. or its affiliates. All Rights Reserved.
  *
+ * Copyright (c) 2025      Barcelona Supercomputing Center (BSC-CNS). All Rights Reserved.
+ *
  * Additional copyrights may follow
  *
  * $HEADER$
@@ -369,6 +371,13 @@ int mca_coll_han_alltoall_using_smsc(
     ompi_request_wait_all(inter_recv_count, inter_recv_reqs, MPI_STATUS_IGNORE);
 
 cleanup:
+
+    /* we may still have neighbors reading directly from our buffer, so we must ensure it is not modified */
+    if (!ii_push_data)
+    {
+        low_comm->c_coll->coll_barrier(low_comm, low_comm->c_coll->coll_barrier_module);
+    }
+
     for (int jlow=0; jlow<low_size; jlow++) {
         if (jlow != low_rank ) {
             mca_smsc->unmap_peer_region(sbuf_map_ctx[jlow]);