Re: [PATCH 2/2] perf record: Synthesize COMM event for a command line workload

[Namhyung Kim <namhyung@kernel.org>]

Hi Arnaldo,

On Mon, Sep 21, 2015 at 04:39:26PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Mon, Sep 21, 2015 at 09:26:49AM +0900, Namhyung Kim escreveu:
> > When perf creates a new child to profile, the events are enabled on
> > exec().  And in this case, it doesn't synthesize any event for the
> > child since they'll be generated during exec().  But there's an window
> > between the enabling and the event generation.
> > 
> > It used to be overcome since samples are only in kernel (so we always
> > have the map) and the comm is overridden by a later COMM event.
> > However it won't work if events are processed and displayed before the
> > COMM event overrides like in 'perf script'.  This leads to those early
> > samples (like native_write_msr_safe) not having a comm but pid (like
> > ':15328').
> > 
> > So it needs to synthesize COMM event for the child explicitly before
> > enabling so that it can have a correct comm.  But at this time, the
> > comm will be "perf" since it's not exec-ed yet.
> > 
> > Acked-by: Jiri Olsa <jolsa@redhat.com>
> > Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> > ---
> >  tools/perf/builtin-record.c | 41 ++++++++++++++++++++++++++++++++++++++++-
> >  1 file changed, 40 insertions(+), 1 deletion(-)
> > 
> > diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> > index 142eeb341b29..b83373adb9f8 100644
> > --- a/tools/perf/builtin-record.c
> > +++ b/tools/perf/builtin-record.c
> > @@ -469,6 +469,43 @@ static void workload_exec_failed_signal(int signo __maybe_unused,
> >  	child_finished = 1;
> >  }
> >  
> > +static int synthesize_workload_comm_event(struct perf_evlist *evlist, void *arg)
> > +{
> > +	union perf_event *event;
> > +	struct record *rec = arg;
> > +	struct machine *machine = &rec->session->machines.host;
> > +	int pid = evlist->workload.pid;
> > +	const char *comm_str = program_invocation_short_name;
> > +	size_t comm_size, total_size;
> > +	int ret;
> > +
> > +	comm_size = PERF_ALIGN(strlen(comm_str) + 1, sizeof(u64));
> > +	total_size = sizeof(event->comm) + machine->id_hdr_size;
> > +	/*
> > +	 * (aligned) comm size might be smaller than expected size
> > +	 * (i.e.  size of event->comm.comm[]), in that case it needs
> > +	 * to shrink the total size.
> > +	 */
> > +	if (comm_size < sizeof(event->comm.comm))
> > +		total_size -= sizeof(event->comm.comm) - comm_size;
> > +
> > +	event = zalloc(total_size);
> > +	if (event == NULL)
> > +		return -ENOMEM;
> > +
> > +	event->comm.header.type = PERF_RECORD_COMM;
> > +	event->comm.header.size = total_size;
> > +
> > +	event->comm.pid = pid;
> > +	event->comm.tid = pid;
> > +	strncpy(event->comm.comm, comm_str, comm_size);
> > +
> > +	ret = record__write(rec, event, total_size);
> > +
> > +	free(event);
> > +	return ret;
> > +}
> > +
> >  static void snapshot_sig_handler(int sig);
> >  
> >  static int __cmd_record(struct record *rec, int argc, const char **argv)
> > @@ -637,7 +674,9 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
> >  	 * Let the child rip
> >  	 */
> >  	if (forks)
> > -		perf_evlist__start_workload(rec->evlist);
> > +		perf_evlist__start_workload_ex(rec->evlist,
> > +					       synthesize_workload_comm_event,
> > +					       rec);
> 
> Why not call it directly? I.e.:
> 
> 	if (forks) {
> 		err = synthesize_workload_comm_event(evlist, rec);
> 		if (!err)
> 			err = perf_evlist__start_workload(rec->evlist);
> 	}
> 
> Since, from what I saw, the very first thing that
> perf_evlist__start_workload_ex() does is to call the callback?

I originally thought that it'd be used by other commands too.  I
checked that 'perf trace' has similar code so I generalized it with
callbacks.  But then I realized the perf trace generates events only
after exec() so I dropped the patch for it.


> 
> Also, don't we have already a synthesize_comm routine? I.e. can't
> perf_event__prepare_comm() be used here?

Ok, it'd be better exporting perf_event__synthesize_comm() for
consistency then.  Will send v2 soon.

Thanks,
Namhyung