Re: [PATCH 35/40] perf record: Synthesize COMM event for a command line workload

From: Namhyung Kim
Date: Tue May 19 2015 - 21:06:01 EST


On Tue, May 19, 2015 at 09:22:10PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Wed, May 20, 2015 at 08:56:58AM +0900, Namhyung Kim escreveu:
> > On Tue, May 19, 2015 at 05:18:54PM -0300, Arnaldo Carvalho de Melo wrote:
> > > Em Tue, May 19, 2015 at 09:49:03PM +0200, Jiri Olsa escreveu:
> > > > > > Humm, you're thinking about where you managed to reproduce the problem,
> > > > > > I am thinking outside indexing, etc, i.e. by definition we either enable
> > > > > > the event before we fork, so that we get the PERF_RECORD_FORK/COMM or we
> > > > > > synthesize it either from /proc or directly (preferred) if we decide to
> > > > > > do it after the fork/exec, right?
>
> > > > > But as I said before, later COMM event will override thread->comm to a
> > > > > proper string as long as it can find a matching thread. So I think it
> > > > > has no problem in the current code.
>
> > > > I can see the issue in the current script code and the patch cured it ;-)
>
> > > Exactly, this is my point, this is not something new :-)
>
> > Ah, okay. The perf script shows samples before processing comm events
> > while perf report shows after processing all events.
>
> I.e. 'perf script' behaves like 'perf trace' and 'perf top'. 'perf
> report' is the odd one out, and I think it should be not, i.e. you
> should try to think more about the non 'report' use cases when thinking
> about how to improve report :-)

I'll keep it in mind.

>
> But I digress, lets get back to the question at hand...
>
> > But to move it under generic place like perf_evlist__{prepare,start}_
> > workload(), it seems we need to pass an additional callback and data.
>
> Only if you want to do it with perf_event__synthesize_comm(). I
> suggested writing a new synthesize routine that doesn't parses /proc, as
> we have all that we need, no?

Agreed.

>
> I think that just doing something like:
>
> thread = machine__findnew_thread(evlist->workload.pid, evlist->workload.pid);
> if (thread)
> thread__set_comm(thread, argv[0], timestamp);
>
> Should be enough, no? I.e. no need for setting up a PERF_RECORD_FORK and
> a PERF_RECORD_COMM, read /proc, etc, just do it directly with the info
> we used to do the fork in perf_evlist__prepare_workload(), etc.

For non-record use case it'd be enough. But for record, it needs to
synthesize/write an event to data file so that perf report can
recognize it later. That's why I think it needs callback.

Thanks,
Namhyung
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/