fix for deadlock in connect/accept

This commit was SVN r3315.

src/communicator/comm_dyn.c

@@ -223,10 +223,12 @@ ompi_process_name_t *ompi_comm_get_rport (ompi_process_name_t *port, int send_fi
     int rc;
     ompi_process_name_t *rport, tbuf;
     ompi_proc_t *rproc=NULL;
+    bool isnew = false;
 
     if ( send_first ) {
         ompi_buffer_t sbuf;
 
+        rproc = ompi_proc_find_and_add(port, &isnew);
         ompi_buffer_init(&sbuf, sizeof(ompi_process_name_t));
         ompi_pack(sbuf, &(proc->proc_name), 1, OMPI_NAME);
         rc = mca_oob_send_packed(port, sbuf, tag, 0);
@@ -240,11 +242,12 @@ ompi_process_name_t *ompi_comm_get_rport (ompi_process_name_t *port, int send_fi
         rc = mca_oob_recv_packed(MCA_OOB_NAME_ANY, &rbuf, &tag);
         ompi_unpack(rbuf, &tbuf, 1, OMPI_NAME);
         ompi_buffer_free(rbuf);
-
+        rproc = ompi_proc_find_and_add(&tbuf, &isnew);
-	rproc = ompi_proc_find_and_add(&tbuf);
         rport = &(rproc->proc_name);
-    }
 
+    }
+    if(isnew)
+        mca_pml.pml_add_procs(&rproc, 1);
     return rport;
 }
 

src/mca/base/mca_base_module_exchange.c

@@ -16,6 +16,7 @@
 #include "mca/gpr/base/base.h"
 #include "mca/ns/ns.h"
 #include "mca/ns/base/base.h"
+#include "mca/pml/pml.h"
 #include "mca/base/mca_base_module_exchange.h"
 #include "runtime/runtime.h"
 
@@ -187,6 +188,12 @@ static void mca_base_modex_registry_callback(
     void* cbdata)
 {
     ompi_list_item_t* item;
+    ompi_proc_t** new_procs = NULL;
+    size_t new_proc_count = 0;
+
+    if(ompi_list_get_size(&msg->data)) {
+        new_procs = malloc(sizeof(ompi_proc_t*) * ompi_list_get_size(&msg->data));
+    }
 
     /* process the callback */
     while((item = ompi_list_remove_first(&msg->data)) != NULL) {
@@ -201,6 +208,7 @@ static void mca_base_modex_registry_callback(
         mca_base_component_t component;
         void* bptr;
         int32_t bsize;
+        bool isnew = false;
                                                                                                        
         /* transfer ownership of registry object to buffer and unpack */
         ompi_buffer_init_preallocated(&buffer, value->object, value->object_size);
@@ -212,9 +220,13 @@ static void mca_base_modex_registry_callback(
          * Lookup the process.
          */
         ompi_unpack(buffer, &proc_name, 1, OMPI_NAME);
-        proc = ompi_proc_find_and_add(&proc_name);
+        proc = ompi_proc_find_and_add(&proc_name, &isnew);
         if(NULL == proc)
             continue;
+        if(isnew) {
+            new_procs[new_proc_count] = proc;
+            new_proc_count++;
+        }
 
         /*
          * Lookup the modex data structure.
@@ -277,6 +289,13 @@ static void mca_base_modex_registry_callback(
         /* release buffer */
         ompi_buffer_free(buffer);
         OMPI_THREAD_UNLOCK(&proc->proc_lock);
+
+        /* update the pml/ptls with new proc */
+    }
+
+    if(NULL != new_procs) {
+        mca_pml.pml_add_procs(new_procs, new_proc_count);
+        free(new_procs);
     }
 }
 

src/proc/proc.c

@@ -170,11 +170,12 @@ ompi_proc_t * ompi_proc_find ( const ompi_process_name_t * name )
 }
 
 
-ompi_proc_t * ompi_proc_find_and_add ( const ompi_process_name_t * name )
+ompi_proc_t * ompi_proc_find_and_add ( const ompi_process_name_t * name, bool* isnew )
 {
     ompi_proc_t *proc, *rproc=NULL;
     ompi_ns_cmp_bitmask_t mask;
 
+    fprintf(stderr, "ompi_proc_find_and_add\n");
     /* return the proc-struct which matches this jobid+process id */
     mask = OMPI_NS_CMP_CELLID | OMPI_NS_CMP_JOBID | OMPI_NS_CMP_VPID;
     OMPI_THREAD_LOCK(&ompi_proc_lock);
@@ -183,6 +184,7 @@ ompi_proc_t * ompi_proc_find_and_add ( const ompi_process_name_t * name )
         proc =  (ompi_proc_t*)ompi_list_get_next(proc)) {
         if (0 == ompi_name_server.compare(mask, &proc->proc_name, name))
             { 
+                *isnew = false;
                 rproc = proc;
                 break;
             }
@@ -193,9 +195,7 @@ ompi_proc_t * ompi_proc_find_and_add ( const ompi_process_name_t * name )
 	    ompi_proc_t *tproc = OBJ_NEW(ompi_proc_t);
 	    rproc = tproc;
 	    rproc->proc_name = *name;
-
+        *isnew = true;
-        /* downcall into pml to notify ptls of new proc */
-        mca_pml.pml_add_procs(&rproc,1);
     }
     return rproc;
 }
@@ -222,6 +222,7 @@ int ompi_proc_get_proclist (ompi_buffer_t buf, int proclistsize, ompi_proc_t ***
     int i;
     ompi_proc_t **plist=NULL;
     ompi_process_name_t name;
+    bool isnew = false;
 
     /* do not free plist *ever*, since it is used in the remote group structure
        of a communicator */
@@ -232,7 +233,10 @@ int ompi_proc_get_proclist (ompi_buffer_t buf, int proclistsize, ompi_proc_t ***
 
     for ( i=0; i<proclistsize; i++ ){
 	ompi_unpack(buf, &name, 1, OMPI_NAME);
-        plist[i] = ompi_proc_find_and_add ( &name);
+        plist[i] = ompi_proc_find_and_add ( &name, &isnew );
+        if(isnew) {
+            mca_pml.pml_add_procs(&plist[i], 1);
+        }
     }
     *proclist = plist;
 

src/proc/proc.h

@@ -72,7 +72,7 @@ ompi_proc_t * ompi_proc_find ( const ompi_process_name_t* name );
  * Returns the proc instance for a given name. If process is not known,
  * we add it to the process list.
 */
-ompi_proc_t * ompi_proc_find_and_add ( const ompi_process_name_t* name );
+ompi_proc_t * ompi_proc_find_and_add ( const ompi_process_name_t* name, bool* isnew );
 
 /**
  * INPUT: ompi_proc_t **proclist : list of process pointers