tracing: Fix free of probe entry by calling call_rcu_sched()
authorSteven Rostedt (Red Hat) <rostedt@goodmis.org>
Wed, 13 Mar 2013 15:15:19 +0000 (11:15 -0400)
committerBen Hutchings <ben@decadent.org.uk>
Wed, 27 Mar 2013 02:41:14 +0000 (02:41 +0000)
commit 740466bc89ad8bd5afcc8de220f715f62b21e365 upstream.

Because function tracing is very invasive, and can even trace
calls to rcu_read_lock(), RCU access in function tracing is done
with preempt_disable_notrace(). This requires a synchronize_sched()
for updates and not a synchronize_rcu().

Function probes (traceon, traceoff, etc) must be freed after
a synchronize_sched() after its entry has been removed from the
hash. But call_rcu() is used. Fix this by using call_rcu_sched().

Also fix the usage to use hlist_del_rcu() instead of hlist_del().

Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
kernel/trace/ftrace.c

index 6c880e8..0943d2a 100644 (file)
@@ -2725,8 +2725,8 @@ __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
                                        continue;
                        }
 
-                       hlist_del(&entry->node);
-                       call_rcu(&entry->rcu, ftrace_free_entry_rcu);
+                       hlist_del_rcu(&entry->node);
+                       call_rcu_sched(&entry->rcu, ftrace_free_entry_rcu);
                }
        }
        __disable_ftrace_function_probe();