[PATCH RFC] trace: correct off by one preempt counter while recording the trace-event

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

[PATCH RFC] trace: correct off by one preempt counter while recording the trace-event

Sebastian Andrzej Siewior-4
Trace events like raw_syscalls show always a preempt code of one. The
reason is that on PREEMPT kernels rcu_read_lock_sched_notrace()
increases the preemption counter and the function recording the counter
is caller within the RCU section.

Signed-off-by: Sebastian Andrzej Siewior <[hidden email]>
---
I am not sure if this is the only spot that needs correction of the
preemption counter. If it is it could be corrected directly in
trace_event_buffer_reserve().

 include/linux/tracepoint.h  | 13 +++++++++++++
 kernel/trace/trace_events.c |  2 +-
 2 files changed, 14 insertions(+), 1 deletion(-)

diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h
index be586c632a0c..12cb3bb40c1c 100644
--- a/include/linux/tracepoint.h
+++ b/include/linux/tracepoint.h
@@ -33,6 +33,19 @@ struct trace_enum_map {
 
 #define TRACEPOINT_DEFAULT_PRIO 10
 
+/*
+ * The preempt count recorded in trace_event_raw_event_# are off by one due to
+ * rcu_read_lock_sched_notrace() in __DO_TRACE. This is corrected here.
+ */
+static inline int event_preempt_count(void)
+{
+#ifdef CONFIG_PREEMPT
+ return preempt_count() - 1;
+#else
+ return 0;
+#endif
+}
+
 extern int
 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
 extern int
diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
index 52c4fffaddcd..90b40cf6ec98 100644
--- a/kernel/trace/trace_events.c
+++ b/kernel/trace/trace_events.c
@@ -245,7 +245,7 @@ void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
  return NULL;
 
  local_save_flags(fbuffer->flags);
- fbuffer->pc = preempt_count();
+ fbuffer->pc = event_preempt_count();
  fbuffer->trace_file = trace_file;
 
  fbuffer->event =
--
2.8.1