Re: [PATCH v2 2/3] ftrace: add tracepoint for hrtimer

[Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>]

Hi Thomas,

Thomas Gleixner wrote:
>
>   Here I'd like to see a callback_done entry as well, so we can check
>   for long lasting callbacks.
> 

I am trying to address your comment that move trace_timer_expires()
form hrtimer_interrupt()/hrtimer_run_queues() to __run_hrtimer() to
trace the execution time of hrtimer's callback, like below:
 
 __run_hrtimer()
 {
 	.....
   +	trace_hrtimer_expires();
 	restart = fn(timer);
   +	trace_hrtimer_callback_done();
 	......
 }

But I meet a problem here that I can't get the timestamps when hrtimer
expires in __run_hrtimer(), which is used to calculate the latency of
hrtimer. I think it's Ok to not move trace_hrtimer_expires(), that is:

 hrtimer_interrupt()
 {
 	 ......
   +	 trace_hrtimer_expires();
 	 __run_hrtimer(timer);
 	 ......
  }
 	 
  __run_hrtimer()
  {
 	......
	restart = fn(timer);
   +	trace_hrtimer_callback_done();
 	......
 }
 
But if more accurate execution time is a require, we'd better use the
tracepoints added by Anton Blanchard <anton@samba.org>. That is:

 hrtimer_interrupt()
 {
 	 ......
   +	 trace_hrtimer_expires();
 	 __run_hrtimer(timer);
 	 ......
 }
 
 __run_hrtimer()
 {
 	......
  + 	trace_hrtimer_callback_entry()
 	restart = fn(timer);
  +	trace_hrtimer_callback_exit();
 	......
 }


Thanks,
Xiao