[mpich2-commits] r7972 - in mpich2/trunk/src: include mpid/common/sched

[goodell at mcs.anl.gov]

Author: goodell
Date: 2011-02-15 15:58:56 -0600 (Tue, 15 Feb 2011)
New Revision: 7972

Modified:
   mpich2/trunk/src/include/mpiimpl.h
   mpich2/trunk/src/mpid/common/sched/mpid_sched.c
Log:
add/release refs in the NBC schedule code to prevent premature frees

Otherwise a user might initiate an MPIX_Ifoo request, free a datatype
or op, and then wait on the request, which would blow up.

Reviewed by buntinas at .

Modified: mpich2/trunk/src/include/mpiimpl.h
===================================================================
--- mpich2/trunk/src/include/mpiimpl.h	2011-02-15 21:58:52 UTC (rev 7971)
+++ mpich2/trunk/src/include/mpiimpl.h	2011-02-15 21:58:56 UTC (rev 7972)
@@ -1766,6 +1766,8 @@
 extern MPID_Op MPID_Op_direct[];
 extern MPIU_Object_alloc_t MPID_Op_mem;
 
+#define MPIR_Op_add_ref(_op) \
+    do { MPIU_Object_add_ref(_op); } while (0)
 #define MPIR_Op_release_ref( _op, _inuse ) \
     do { MPIU_Object_release_ref( _op, _inuse ); } while (0)
 

Modified: mpich2/trunk/src/mpid/common/sched/mpid_sched.c
===================================================================
--- mpich2/trunk/src/mpid/common/sched/mpid_sched.c	2011-02-15 21:58:52 UTC (rev 7971)
+++ mpich2/trunk/src/mpid/common/sched/mpid_sched.c	2011-02-15 21:58:56 UTC (rev 7972)
@@ -129,6 +129,11 @@
             mpi_errno = MPIR_Reduce_local_impl(e->u.reduce.inbuf, e->u.reduce.inoutbuf, e->u.reduce.count,
                                                e->u.reduce.datatype, e->u.reduce.op);
             if (mpi_errno) MPIU_ERR_POP(mpi_errno);
+            if (HANDLE_GET_KIND(e->u.reduce.op) != HANDLE_KIND_BUILTIN) {
+                MPID_Op *op_ptr = NULL;
+                MPID_Op_get_ptr(e->u.reduce.op, op_ptr);
+                MPIR_Op_release(op_ptr);
+            }
             e->status = MPIDU_SCHED_ENTRY_STATUS_COMPLETE;
             break;
         case MPIDU_SCHED_ENTRY_COPY:
@@ -380,6 +385,16 @@
     e->u.send.sreq = NULL; /* will be populated by _start_entry */
     e->u.send.comm = comm;
 
+    /* the user may free the comm & type after initiating but before the
+     * underlying send is actually posted, so we must add a reference here and
+     * release it at entry completion time */
+    MPIR_Comm_add_ref(comm);
+    if (HANDLE_GET_KIND(datatype) != HANDLE_KIND_BUILTIN) {
+        MPID_Datatype *dtp = NULL;
+        MPID_Datatype_get_ptr(datatype, dtp);
+        MPID_Datatype_add_ref(dtp);
+    }
+
 fn_exit:
     return mpi_errno;
 fn_fail:
@@ -409,6 +424,13 @@
     e->u.recv.rreq = NULL; /* will be populated by _start_entry */
     e->u.recv.comm = comm;
 
+    MPIR_Comm_add_ref(comm);
+    if (HANDLE_GET_KIND(datatype) != HANDLE_KIND_BUILTIN) {
+        MPID_Datatype *dtp = NULL;
+        MPID_Datatype_get_ptr(datatype, dtp);
+        MPID_Datatype_add_ref(dtp);
+    }
+
 fn_exit:
     return mpi_errno;
 fn_fail:
@@ -439,6 +461,17 @@
     reduce->datatype = datatype;
     reduce->op = op;
 
+    if (HANDLE_GET_KIND(datatype) != HANDLE_KIND_BUILTIN) {
+        MPID_Datatype *dtp = NULL;
+        MPID_Datatype_get_ptr(datatype, dtp);
+        MPID_Datatype_add_ref(dtp);
+    }
+    if (HANDLE_GET_KIND(op) != HANDLE_KIND_BUILTIN) {
+        MPID_Op *op_ptr = NULL;
+        MPID_Op_get_ptr(op, op_ptr);
+        MPIR_Op_add_ref(op_ptr);
+    }
+
 fn_exit:
     return mpi_errno;
 fn_fail:
@@ -473,6 +506,17 @@
     copy->outcount = outcount;
     copy->outtype = outtype;
 
+    if (HANDLE_GET_KIND(intype) != HANDLE_KIND_BUILTIN) {
+        MPID_Datatype *dtp = NULL;
+        MPID_Datatype_get_ptr(intype, dtp);
+        MPID_Datatype_add_ref(dtp);
+    }
+    if (HANDLE_GET_KIND(outtype) != HANDLE_KIND_BUILTIN) {
+        MPID_Datatype *dtp = NULL;
+        MPID_Datatype_get_ptr(outtype, dtp);
+        MPID_Datatype_add_ref(dtp);
+    }
+
 fn_exit:
     return mpi_errno;
 fn_fail:
@@ -558,6 +602,12 @@
                         dprintf(stderr, "completed SEND entry %d, sreq=%p\n", i, e->u.send.sreq);
                         e->status = MPIDU_SCHED_ENTRY_STATUS_COMPLETE;
                         MPID_Request_release(e->u.send.sreq);
+                        MPIR_Comm_release(e->u.send.comm, /*isDisconnect=*/FALSE);
+                        if (HANDLE_GET_KIND(e->u.send.datatype) != HANDLE_KIND_BUILTIN) {
+                            MPID_Datatype *dtp = NULL;
+                            MPID_Datatype_get_ptr(e->u.send.datatype, dtp);
+                            MPID_Datatype_release(dtp);
+                        }
                         e->u.send.sreq = NULL;
                     }
                     break;
@@ -567,6 +617,12 @@
                         e->status = MPIDU_SCHED_ENTRY_STATUS_COMPLETE;
                         MPID_Request_release(e->u.recv.rreq);
                         e->u.recv.rreq = NULL;
+                        MPIR_Comm_release(e->u.recv.comm, /*isDisconnect=*/FALSE);
+                        if (HANDLE_GET_KIND(e->u.recv.datatype) != HANDLE_KIND_BUILTIN) {
+                            MPID_Datatype *dtp = NULL;
+                            MPID_Datatype_get_ptr(e->u.recv.datatype, dtp);
+                            MPID_Datatype_release(dtp);
+                        }
                     }
                     break;
                 default: