tracing: generate RCU warnings even when tracepoints are disabled
authorDave Hansen <dave.hansen@linux.intel.com>
Thu, 7 Aug 2014 17:52:04 +0000 (10:52 -0700)
committerSteven Rostedt <rostedt@goodmis.org>
Wed, 10 Sep 2014 14:48:46 +0000 (10:48 -0400)
Dave Jones reported seeing a bug from one of my TLB tracepoints:

http://lkml.kernel.org/r/20140806181801.GA4605@redhat.com

I've been running these patches for months and never saw this.
But, a big chunk of my testing, especially with all the debugging
enabled, was in a vm where intel_idle doesn't work.  On the
systems where I was using intel_idle, I never had lockdep enabled
and this tracepoint on at the same time.

This patch ensures that whenever we have lockdep available, we do
_some_ RCU activity at the site of the tracepoint, despite
whether the tracepoint's condition matches or even if the
tracepoint itself is completely disabled.  This is a bit of a
hack, but it is pretty self-contained.

I confirmed that with this patch plus lockdep I get the same
splat as Dave Jones did, but without enabling the tracepoint
explicitly.

Link: http://lkml.kernel.org/p/20140807175204.C257CAC5@viggo.jf.intel.com
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Dave Hansen <dave@sr71.net>
Cc: Dave Jones <davej@redhat.com>,
Cc: paulmck@linux.vnet.ibm.com
Cc: Ingo Molnar <mingo@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
include/linux/tracepoint.h

index b1293f15f5920c80d9ebaaa102f91df15a9ab1b7..e08e21e5f601bcec81ea21131c4e58e69c611850 100644 (file)
@@ -157,6 +157,12 @@ extern void syscall_unregfunc(void);
  * Make sure the alignment of the structure in the __tracepoints section will
  * not add unwanted padding between the beginning of the section and the
  * structure. Force alignment to the same alignment as the section start.
+ *
+ * When lockdep is enabled, we make sure to always do the RCU portions of
+ * the tracepoint code, regardless of whether tracing is on or we match the
+ * condition.  This lets us find RCU issues triggered with tracepoints even
+ * when this tracepoint is off.  This code has no purpose other than poking
+ * RCU a bit.
  */
 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
        extern struct tracepoint __tracepoint_##name;                   \
@@ -167,6 +173,11 @@ extern void syscall_unregfunc(void);
                                TP_PROTO(data_proto),                   \
                                TP_ARGS(data_args),                     \
                                TP_CONDITION(cond),,);                  \
+               if (IS_ENABLED(CONFIG_LOCKDEP)) {                       \
+                       rcu_read_lock_sched_notrace();                  \
+                       rcu_dereference_sched(__tracepoint_##name.funcs);\
+                       rcu_read_unlock_sched_notrace();                \
+               }                                                       \
        }                                                               \
        __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args),          \
                PARAMS(cond), PARAMS(data_proto), PARAMS(data_args))    \