Fix pbm_join kernel panic caused by early parent exit

which exposed the following precondition for calling
pbm_join_and_print_graph_self(): ALL childs, grandchilds etc. of the
root process of a measured program must be exited since otherwise the
informations in the model to modify and print are not complete.

kernel/behave.c

@@ -390,10 +390,12 @@ int pbm_task_end(PBM* pbm) {
     // end task of parent process
 	parent_pbm = get_pbm_by_pid(parent_proc->pid);
 	if(!parent_pbm) {
-		printk(KERN_WARNING "COULD NOT FIND PARENT-PBM!\n");
-		//TODO Since this will happen right at the first call to pbm_fork()
-		//TODO since no parent process has been initialized (since this is
-		//TODO the first relevant parent process)
+        // this should only happen right at the first call to pbm_fork() since at this point no
+        // parent process has been initialized yet since this is the first relevant parent process
+        if (current->pid != root_proc) {
+            printk(KERN_WARNING "ERROR: COULD NOT FIND PARENT-PBM!\n");
+            BUG();
+        }
         return NULL;
 	}
 	pbm_task_end(parent_pbm);
@@ -488,7 +490,9 @@ int pbm_fork(struct task_struct* proc, pid_t parent_pid, pbm_NODE* fork_node) {
 			child_pbm->next_sib = parent_pbm->children;
 		}
 		parent_pbm->children = child_pbm;
-	}
+	} else {
+        BUG_ON(current->pid != root_proc);
+    }
 
     // We have to know WHEN the exit happens relative to the parent. So every child remembers the
     // current fork-task-node of the parent on exit (so that the join can happen at the correct
@@ -531,7 +535,15 @@ int pbm_exit(pid_t pid, pid_t parent_pid) {
     // set current parent task as the exit task of this child where the join
     // gets inserted
     if(parent_pbm) {
+        if (parent_pbm->last == NULL) {
+            printk(KERN_WARNING "ERROR: ParentPBM->LAST is NULL! ppid: %u\n", parent_pid);
+        }
         pbm->exit_date = parent_pbm->last;
+    } else {
+        if (current->pid != root_proc) {
+            printk(KERN_WARNING "ERROR: COULD NOT FIND PARENT! child: %u, parent: %u\n", pid, parent_pid);
+            BUG();
+        }
     }
 
     return 1;
@@ -550,6 +562,10 @@ int pbm_join(PBM* child_pbm) {
     if(!child_pbm)
         return 0;
 
+    if (!child_pbm->exit_date) {
+        printk(KERN_WARNING "Exit date missing for pid %u. Has the process really been exited?\n", child_pbm->last->thread_id);
+    }
+
     printk(KERN_WARNING "Joining thread %u with fork_node: %lli\n", child_pbm->last->thread_id, task_2_index(child_pbm->fork_date));
     fork_node = child_pbm->fork_date;
 
@@ -605,6 +621,14 @@ void pbm_post_processing(PBM* pbm) {
 * PBM graph output functions
 */
 
+/**
+ * BEWARE:
+ * This function must not be called before ALL childs, grandchilds etc. of the root process of a
+ * measured program are exited since otherwise the informations in the model are not complete and
+ * therefore this function will throw an error!
+ * In this case most probably pbm_join() will generate the error since the exit_date of the not yet
+ * exited process will not be set and also the perf counter values will probably be wrong.
+ */
 void pbm_join_and_print_graph_self(pid_t pid) {
     PBM* pbm;
 

kernel/exit.c

@@ -771,6 +771,13 @@ void __noreturn do_exit(long code)
 	if (is_relevant_process(tsk)) {
 		pbm_exit(tsk->pid, tsk->real_parent->pid);
 	}
+
+	/**
+	 * BEWARE:
+	 * It is necessary that the root process of a measured program does not exit before ALL its
+	 * childs, grandchilds etc. are exited since otherwise the informations in the model are not
+	 * complete and therefore pbm_join_and_print_graph_self() will throw an error!
+	 */
 	if (is_root_process(tsk)) {
 		pbm_join_and_print_graph_self(tsk->pid);
 	}

pb_utils/pb_submitter/test_prog.c

@@ -17,10 +17,11 @@ int main(void)
     // grandchild.
     pid_t n2 = fork();
 
-    if (n1 > 0 && n2 > 0) {
-        // barrier: let the father wait for all the child processes
-        while ((wpid = wait(&status)) > 0) {};
+    // barrier: make all fathers wait for all its child processes (since all processes use that line)
+    while ((wpid = wait(&status)) > 0) {};
 
+    if (n1 > 0 && n2 > 0)
+    {
         fprintf(stderr, "Parent %d of %d %d\n", getpid(), n1, n2);
 
         // check if program runs && syscall to switch tasks