Re: [PATCHv3 06/13] perf record: Store data mmaps for dwarf unwind

[Arnaldo Carvalho de Melo]

Em Thu, Jan 07, 2016 at 02:30:22PM +0100, Jiri Olsa escreveu:
> On Thu, Jan 07, 2016 at 10:14:14PM +0900, Namhyung Kim wrote:
> > On Thu, Jan 07, 2016 at 01:39:28PM +0100, Jiri Olsa wrote:
> > > Currently we don't synthesize data mmap by default. It depends
> > > on -d option, that enables data address sampling.
> > > 
> > > But we've seen cases (softice) where DWARF unwinder went through
> > > non executable mmaps, which we need to lookup in MAP__VARIABLE tree.
> > > 
> > > Making data mmaps to be synthesized for dwarf unwind as well.
> > > 
> > > Reported-by: Noel Grandin <noelgrandin@xxxxxxxxx>
> > > Link: http://lkml.kernel.org/n/tip-lh02yir6qfycn8zr892rmlgg@xxxxxxxxxxxxxx
> > > Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
> > > ---
> > >  tools/perf/builtin-record.c | 5 ++++-
> > >  1 file changed, 4 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> > > index 9c5cdc2c4471..465bde77f49b 100644
> > > --- a/tools/perf/builtin-record.c
> > > +++ b/tools/perf/builtin-record.c
> > > @@ -815,8 +815,11 @@ int record_parse_callchain_opt(const struct option *opt,
> > >  	}
> > >  
> > >  	ret = parse_callchain_record_opt(arg, &callchain_param);
> > > -	if (!ret)
> > > +	if (!ret) {
> > > +		/* Enable data address sampling for DWARF unwind. */
> > > +		record->sample_address = true;
> > 
> > How does it guarantee that it enables data address sampling only for
> > DWARF unwind?  Maybe you want this?
> > 
> > 		if (callchain_param.record_mode == CALLCHAIN_DWARF)
> > 			record->sample_address = true;
> 
> ugh.. too much coofee.. :-\ third time's a charm, v3 attached

Noel, have you tested this latest patch? Does it fix things for you in
this latest iteration?

If so, can I stick a "Reported-and-Tested-by: Noel Grandin <>" to this?
 
> thanks,
> jirka
> 
> 
> ---
> Currently we don't synthesize data mmap by default. It depends
> on -d option, that enables data address sampling.
> 
> But we've seen cases (softice) where DWARF unwinder went through
> non executable mmaps, which we need to lookup in MAP__VARIABLE tree.
> 
> Making data mmaps to be synthesized for dwarf unwind as well.
> 
> Reported-by: Noel Grandin <noelgrandin@xxxxxxxxx>
> Link: http://lkml.kernel.org/n/tip-lh02yir6qfycn8zr892rmlgg@xxxxxxxxxxxxxx
> Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
> ---
>  tools/perf/builtin-record.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 9c5cdc2c4471..dc4e0adf5c5b 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -815,8 +815,12 @@ int record_parse_callchain_opt(const struct option *opt,
>  	}
>  
>  	ret = parse_callchain_record_opt(arg, &callchain_param);
> -	if (!ret)
> +	if (!ret) {
> +		/* Enable data address sampling for DWARF unwind. */
> +		if (callchain_param.record_mode == CALLCHAIN_DWARF)
> +			record->sample_address = true;
>  		callchain_debug();
> +	}
>  
>  	return ret;
>  }
> -- 
> 2.4.3