rt-patches/0349-trace-correct-off-by-one-while-recording-the-trace-e.patch

   1 From 8072afcff6ac98ce8a2068dd406d9ad1332c47ce Mon Sep 17 00:00:00 2001
   2 From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
   3 Date: Wed, 25 May 2016 14:03:50 +0200
   4 Subject: [PATCH 349/366] trace: correct off by one while recording the
   5  trace-event
   6
   7 Trace events like raw_syscalls show always a preempt code of one. The
   8 reason is that on PREEMPT kernels rcu_read_lock_sched_notrace()
   9 increases the preemption counter and the function recording the counter
  10 is caller within the RCU section.
  11
  12 Cc: stable-rt@vger.kernel.org
  13 Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
  14 [ Changed this to upstream version. See commit e947841c0dce ]
  15 Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
  16 ---
  17  kernel/trace/trace_events.c | 8 ++++++++
  18  1 file changed, 8 insertions(+)
  19
  20 diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
  21 index 4a48f97..5bd79b3 100644
  22 --- a/kernel/trace/trace_events.c
  23 +++ b/kernel/trace/trace_events.c
  24 @@ -246,6 +246,14 @@ void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
  25
  26         local_save_flags(fbuffer->flags);
  27         fbuffer->pc = preempt_count();
  28 +       /*
  29 +        * If CONFIG_PREEMPT is enabled, then the tracepoint itself disables
  30 +        * preemption (adding one to the preempt_count). Since we are
  31 +        * interested in the preempt_count at the time the tracepoint was
  32 +        * hit, we need to subtract one to offset the increment.
  33 +        */
  34 +       if (IS_ENABLED(CONFIG_PREEMPT))
  35 +               fbuffer->pc--;
  36         fbuffer->trace_file = trace_file;
  37
  38         fbuffer->event =
  39 --
  40 1.9.1
  41