[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <YaY8eBB+PnklIxZA@kernel.org>
Date: Tue, 30 Nov 2021 12:00:08 -0300
From: Arnaldo Carvalho de Melo <acme@...nel.org>
To: Jiri Olsa <jolsa@...hat.com>
Cc: Ian Rogers <irogers@...gle.com>,
Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Namhyung Kim <namhyung@...nel.org>,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
Stephane Eranian <eranian@...gle.com>
Subject: Re: [PATCH] perf map: Fix namespace memory leak
Em Sun, Nov 28, 2021 at 04:40:26PM +0100, Jiri Olsa escreveu:
> On Thu, Nov 18, 2021 at 11:37:14AM -0800, Ian Rogers wrote:
> > This leak was happening reliably with test "Lookup mmap thread" with
> > stack traces like:
> >
> > Direct leak of 5504 byte(s) in 172 object(s) allocated from:
> > #0 0x7f4685e47987 in __interceptor_calloc
> > #1 0x56063b974c2a in nsinfo__new util/namespaces.c:142
> > #2 0x56063b9781ff in thread__new util/thread.c:70
> > #3 0x56063b944953 in ____machine__findnew_thread util/machine.c:543
> > #4 0x56063b944ac6 in __machine__findnew_thread util/machine.c:574
> > #5 0x56063b944b36 in machine__findnew_thread util/machine.c:584
> > #6 0x56063b94c892 in machine__process_fork_event util/machine.c:1954
> > #7 0x56063b94cc1f in machine__process_event util/machine.c:2019
> > #8 0x56063b894f18 in perf_event__process util/event.c:567
> > #9 0x56063ba17951 in perf_tool__process_synth_event util/synthetic-events.c:65
> > #10 0x56063ba19086 in perf_event__synthesize_fork util/synthetic-events.c:287
> > #11 0x56063ba1c39d in __event__synthesize_thread util/synthetic-events.c:775
> > #12 0x56063ba1cf6f in __perf_event__synthesize_threads util/synthetic-events.c:929
> > #13 0x56063ba1d4ab in perf_event__synthesize_threads util/synthetic-events.c:1000
> > #14 0x56063b821a3d in synth_all tests/mmap-thread-lookup.c:136
> > #15 0x56063b821c86 in mmap_events tests/mmap-thread-lookup.c:174
> > #16 0x56063b8221b7 in test__mmap_thread_lookup tests/mmap-thread-lookup.c:230
> >
> > The dso->nsinfo is overwritten, but without a nsinfo__put this can leak
> > the overwritten nsinfo.
> >
> > Signed-off-by: Ian Rogers <irogers@...gle.com>
>
> nice catch!
>
> Acked-by: Jiri Olsa <jolsa@...hat.com>
This one is tricky, I have to retrieve the details from last time this
surfaced, but try:
# perf top -F 10000
leave it for a while and press 'q':
perf: /home/acme/git/perf/tools/include/linux/refcount.h:131: refcount_sub_and_test: Assertion `!(new > val)' failed.
Aborted (core dumped)
I reproduced the above assertion a few times, now I'm not being able,
probably some namespace related activity took place when it hit.
- Arnaldo
> thanks,
> jirka
>
> > ---
> > tools/perf/util/map.c | 1 +
> > 1 file changed, 1 insertion(+)
> >
> > diff --git a/tools/perf/util/map.c b/tools/perf/util/map.c
> > index 8af693d9678c..ceed8f407bc0 100644
> > --- a/tools/perf/util/map.c
> > +++ b/tools/perf/util/map.c
> > @@ -192,6 +192,7 @@ struct map *map__new(struct machine *machine, u64 start, u64 len,
> > if (!(prot & PROT_EXEC))
> > dso__set_loaded(dso);
> > }
> > + nsinfo__put(dso->nsinfo);
> > dso->nsinfo = nsi;
> >
> > if (build_id__is_defined(bid))
> > --
> > 2.34.0.rc2.393.gf8c9666880-goog
> >
--
- Arnaldo
Powered by blists - more mailing lists