]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/commitdiff
perf data: Fix releasing event_class
authorWang Nan <wangnan0@huawei.com>
Fri, 5 Feb 2016 14:01:30 +0000 (14:01 +0000)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 12 Feb 2016 20:27:48 +0000 (17:27 -0300)
A new patch in libbabeltrace [1] reveals a object leak problem in
'perf data' CTF support: perf code never releases the event_class
which is allocated in add_event() and stored in evsel's private field.

If libbabeltrace has the above patch applied, leaking event_class
prevents the writer from being destroyed and flushing metadata. For
example:

  $ perf record ls
  perf.data
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.012 MB perf.data (12 samples) ]
  $ perf data convert --to-ctf ./out.ctf
  [ perf data convert: Converted 'perf.data' into CTF data './out.ctf' ]
  [ perf data convert: Converted and wrote 0.000 MB (12 samples) ]
  $ cat ./out.ctf/metadata
  $ ls -l  ./out.ctf/metadata
  -rw-r----- 1 w00229757 mm 0 Jan 27 10:49 ./out.ctf/metadata

The correct result should be:
  ...
  $ cat ./out.ctf/metadata
  /* CTF 1.8 */

  trace {
  [SNIP]

  $ ls -l  ./out.ctf/metadata
  -rw-r----- 1 w00229757 mm 2446 Jan 27 10:52 ./out.ctf/metadata

The full story is:

Patch [1] of babeltrace redesigns its reference counting scheme. In that
patch:

 * writer <- trace (bt_ctf_writer_create)
 * trace <- stream_class (bt_ctf_trace_add_stream_class)
 * stream_class <- event_class (bt_ctf_stream_class_add_event_class)
 ('<-' means 'is a parent of')

Holding of event_class causes reference count of corresponding 'writer'
to increase through parent chain. Perf expects that 'writer' is released
(so metadata is flushed) through bt_ctf_writer_put() in
ctf_writer__cleanup(). However, since it never releases event_class, the
reference of 'writer' won't be dropped, so bt_ctf_writer_put() won't
lead to the release of writer.

Before this CTF patch, !(writer <- trace). Even with event_class leaking,
the writer ends up being released.

[1] https://github.com/efficios/babeltrace/commit/e6a8e8e4744633807083a077ff9f101eb97d9801

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Cody P Schafer <dev@codyps.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jeremie Galarneau <jeremie.galarneau@efficios.com>
Cc: Kirill Smelkov <kirr@nexedi.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1454680939-24963-6-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/data-convert-bt.c

index 34cd1e4039d35e05be0460ff0a259e7d8b2bb80a..b722e57d5a877131df36634d983eff7606b9a1f5 100644 (file)
@@ -858,6 +858,23 @@ static int setup_events(struct ctf_writer *cw, struct perf_session *session)
        return 0;
 }
 
+static void cleanup_events(struct perf_session *session)
+{
+       struct perf_evlist *evlist = session->evlist;
+       struct perf_evsel *evsel;
+
+       evlist__for_each(evlist, evsel) {
+               struct evsel_priv *priv;
+
+               priv = evsel->priv;
+               bt_ctf_event_class_put(priv->event_class);
+               zfree(&evsel->priv);
+       }
+
+       perf_evlist__delete(evlist);
+       session->evlist = NULL;
+}
+
 static int setup_streams(struct ctf_writer *cw, struct perf_session *session)
 {
        struct ctf_stream **stream;
@@ -1171,6 +1188,7 @@ int bt_convert__perf2ctf(const char *input, const char *path, bool force)
                (double) c.events_size / 1024.0 / 1024.0,
                c.events_count);
 
+       cleanup_events(session);
        perf_session__delete(session);
        ctf_writer__cleanup(cw);