rcu: Make expedited RCU-preempt stall warnings count accurately
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Wed, 13 Jan 2016 18:52:35 +0000 (10:52 -0800)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Thu, 31 Mar 2016 20:34:03 +0000 (13:34 -0700)
Currently, synchronize_sched_expedited_wait() simply sets the ndetected
variable to the rcu_print_task_exp_stall() return value.  This means
that if the last rcu_node structure has no stalled tasks, record of
any stalled tasks in previous rcu_node structures is lost, which can
in turn result in failure to dump out the blocking rcu_node structures.
Or could, had the test been correct.

This commit therefore adds the return value of rcu_print_task_exp_stall()
to ndetected and corrects the later test for ndetected.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
kernel/rcu/tree.c

index 5f4336f..687d8a5 100644 (file)
@@ -3778,7 +3778,7 @@ static void synchronize_sched_expedited_wait(struct rcu_state *rsp)
                       rsp->name);
                ndetected = 0;
                rcu_for_each_leaf_node(rsp, rnp) {
-                       ndetected = rcu_print_task_exp_stall(rnp);
+                       ndetected += rcu_print_task_exp_stall(rnp);
                        mask = 1;
                        for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask <<= 1) {
                                struct rcu_data *rdp;
@@ -3797,7 +3797,7 @@ static void synchronize_sched_expedited_wait(struct rcu_state *rsp)
                pr_cont(" } %lu jiffies s: %lu root: %#lx/%c\n",
                        jiffies - jiffies_start, rsp->expedited_sequence,
                        rnp_root->expmask, ".T"[!!rnp_root->exp_tasks]);
-               if (!ndetected) {
+               if (ndetected) {
                        pr_err("blocking rcu_node structures:");
                        rcu_for_each_node_breadth_first(rsp, rnp) {
                                if (rnp == rnp_root)