rcu: Avoid rcu_print_detail_task_stall_rnp() segfault
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Fri, 10 Aug 2012 23:00:11 +0000 (16:00 -0700)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Sun, 23 Sep 2012 14:42:50 +0000 (07:42 -0700)
The rcu_print_detail_task_stall_rnp() function invokes
rcu_preempt_blocked_readers_cgp() to verify that there are some preempted
RCU readers blocking the current grace period outside of the protection
of the rcu_node structure's ->lock.  This means that the last blocked
reader might exit its RCU read-side critical section and remove itself
from the ->blkd_tasks list before the ->lock is acquired, resulting in
a segmentation fault when the subsequent code attempts to dereference
the now-NULL gp_tasks pointer.

This commit therefore moves the test under the lock.  This will not
have measurable effect on lock contention because this code is invoked
only when printing RCU CPU stall warnings, in other words, in the common
case, never.

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

index 70b33bf780f00ad6ceff97c524c734df3ee3de4b..df47014e129d302c4927f11e4c282c01a825d874 100644 (file)
@@ -422,9 +422,11 @@ static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
        unsigned long flags;
        struct task_struct *t;
 
-       if (!rcu_preempt_blocked_readers_cgp(rnp))
-               return;
        raw_spin_lock_irqsave(&rnp->lock, flags);
+       if (!rcu_preempt_blocked_readers_cgp(rnp)) {
+               raw_spin_unlock_irqrestore(&rnp->lock, flags);
+               return;
+       }
        t = list_entry(rnp->gp_tasks,
                       struct task_struct, rcu_node_entry);
        list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry)