perf record: Generate tracking events for process forked by perf
authorWang Nan <wangnan0@huawei.com>
Wed, 20 Apr 2016 18:59:54 +0000 (18:59 +0000)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Thu, 28 Apr 2016 12:58:59 +0000 (09:58 -0300)
With 'perf record --switch-output' without -a, record__synthesize() in
record__switch_output() won't generate tracking events because there's
no thread_map in evlist. Which causes newly created perf.data doesn't
contain map and comm information.

This patch creates a fake thread_map and directly call
perf_event__synthesize_thread_map() for those events.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1461178794-40467-8-git-send-email-wangnan0@huawei.com
Signed-off-by: He Kuang <hekuang@huawei.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/builtin-record.c

index 178b49ecd05ff600182802e464c86b920ceb0732..f3679c44d3f3d4b7c51bbb627375a315099b974d 100644 (file)
@@ -500,6 +500,23 @@ record__finish_output(struct record *rec)
        return;
 }
 
+static int record__synthesize_workload(struct record *rec)
+{
+       struct {
+               struct thread_map map;
+               struct thread_map_data map_data;
+       } thread_map;
+
+       thread_map.map.nr = 1;
+       thread_map.map.map[0].pid = rec->evlist->workload.pid;
+       thread_map.map.map[0].comm = NULL;
+       return perf_event__synthesize_thread_map(&rec->tool, &thread_map.map,
+                                                process_synthesized_event,
+                                                &rec->session->machines.host,
+                                                rec->opts.sample_address,
+                                                rec->opts.proc_map_timeout);
+}
+
 static int record__synthesize(struct record *rec);
 
 static int
@@ -532,9 +549,21 @@ record__switch_output(struct record *rec, bool at_exit)
                        file->path, timestamp);
 
        /* Output tracking events */
-       if (!at_exit)
+       if (!at_exit) {
                record__synthesize(rec);
 
+               /*
+                * In 'perf record --switch-output' without -a,
+                * record__synthesize() in record__switch_output() won't
+                * generate tracking events because there's no thread_map
+                * in evlist. Which causes newly created perf.data doesn't
+                * contain map and comm information.
+                * Create a fake thread_map and directly call
+                * perf_event__synthesize_thread_map() for those events.
+                */
+               if (target__none(&rec->opts.target))
+                       record__synthesize_workload(rec);
+       }
        return fd;
 }