Re: [PATCH] perf: POSIX CLOCK_PERF to report current time value

[Ingo Molnar <mingo@kernel.org>]

* Ingo Molnar <mingo@kernel.org> wrote:

> 
> * David Ahern <dsahern@gmail.com> wrote:
> 
> > From: Pawel Moll <pawel.moll@arm.com>
> > 
> > To co-relate user space events with the perf events stream
> > a current (as in: "what time(stamp) is it now?") time value
> > must be made available.
> 
> So I'm wondering about your earlier approach posted here:
> 
>   https://lkml.org/lkml/2011/6/7/636
> 
> I'd modify that patch the following way: instead of tracking each 
> separate reason, perhaps only track timekeeping_update().
> 
> Such a tracepoint, combined with PERF_SAMPLE_TIME, would very 
> accurately track external changes to GTOD (xtime).
> 
> That leaves us with tracking/correlating the regular flow of time, 
> which could be achieved by another tracepoint in:
> 
>    kernel/time/timekeeping.c::do_timer()
> 
> So only two new tracepoints are needed AFAICS - and the tracepoints 
> would obviously be useful for other (debugging) purposes as well. 
> Would that solve the wall-clock correlation problem adequately?

I forgot to mention what data the new tracepoints would trace:

   - timekeeping_update() would trace the 'before' and 'after' 
     (==current) GTOD values

   - do_timer() would trace 'ticks' and the current GTOD value.

'GTOD value' is what gettimeofday() would return, which in the 
do_timer() case ought to be something like this:

   tk->xtime_sec
  (tk->xtime_nsec >> tk->shift) + get_arch_timeoffset()

(Note that this is simple and fast as there's no need to read the 
clock once again, it was just read.)

Thanks,

	Ingo