perf stat: Make metric event lookup more robust
authorAndi Kleen <ak@linux.intel.com>
Mon, 24 Jun 2019 19:37:08 +0000 (12:37 -0700)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 2 Jul 2019 01:50:41 +0000 (22:50 -0300)
After setting up metric groups through the event parser, the metricgroup
code looks them up again in the event list.

Make sure we only look up events that haven't been used by some other
metric. The data structures currently cannot handle more than one metric
per event. This avoids problems with multiple events partially
overlapping.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Link: http://lkml.kernel.org/r/20190624193711.35241-2-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/stat-shadow.c

index 027b09aaa4cf33effef65206b0d0359b6227d603..3f8fd127d31e0a8053462e54fa25073291603ee1 100644 (file)
@@ -304,7 +304,7 @@ static struct perf_evsel *perf_stat__find_event(struct perf_evlist *evsel_list,
        struct perf_evsel *c2;
 
        evlist__for_each_entry (evsel_list, c2) {
-               if (!strcasecmp(c2->name, name))
+               if (!strcasecmp(c2->name, name) && !c2->collect_stat)
                        return c2;
        }
        return NULL;
@@ -343,7 +343,8 @@ void perf_stat__collect_metric_expr(struct perf_evlist *evsel_list)
                        if (leader) {
                                /* Search in group */
                                for_each_group_member (oc, leader) {
-                                       if (!strcasecmp(oc->name, metric_names[i])) {
+                                       if (!strcasecmp(oc->name, metric_names[i]) &&
+                                               !oc->collect_stat) {
                                                found = true;
                                                break;
                                        }