rcu: Fix for rcuo online-time-creation reorganization bug
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Fri, 24 Oct 2014 18:11:08 +0000 (11:11 -0700)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Wed, 29 Oct 2014 17:20:08 +0000 (10:20 -0700)
Commit 35ce7f29a44a (rcu: Create rcuo kthreads only for onlined CPUs)
contains checks for the case where CPUs are brought online out of
order, re-wiring the rcuo leader-follower relationships as needed.
Unfortunately, this rewiring was broken.  This apparently went undetected
due to the tendency of systems to bring CPUs online in order.  This commit
nevertheless fixes the rewiring.

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

index 6d07fb4..5053dae 100644 (file)
@@ -2586,9 +2586,13 @@ static void rcu_spawn_one_nocb_kthread(struct rcu_state *rsp, int cpu)
                        rdp->nocb_leader = rdp_spawn;
                        if (rdp_last && rdp != rdp_spawn)
                                rdp_last->nocb_next_follower = rdp;
-                       rdp_last = rdp;
-                       rdp = rdp->nocb_next_follower;
-                       rdp_last->nocb_next_follower = NULL;
+                       if (rdp == rdp_spawn) {
+                               rdp = rdp->nocb_next_follower;
+                       } else {
+                               rdp_last = rdp;
+                               rdp = rdp->nocb_next_follower;
+                               rdp_last->nocb_next_follower = NULL;
+                       }
                } while (rdp);
                rdp_spawn->nocb_next_follower = rdp_old_leader;
        }