perf stat: Fix adding multiple event groups
authorAndi Kleen <ak@linux.intel.com>
Thu, 14 Sep 2017 20:57:35 +0000 (13:57 -0700)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Thu, 21 Sep 2017 16:12:58 +0000 (13:12 -0300)
The -M metric group parser threw away the events of earlier groups when
multiple groups were specified. Fix this here by not overwriting the
string incorrectly.

Now this works correctly:

% perf stat -M Summary,SMT --metric-only -a sleep 1

 Performance counter stats for 'system wide':

Instructions CPI CLKS         CPU_Utilization GFLOPs SMT_2T_Utilization SMT_2T_Utilization Kernel_Utilization CoreIPC CORE_CLKS
900907376.0  2.7 2398954144.0 0.1             0.0    0.2                0.2                0.1                0.4     2080822855.5

while previously it would only show the SMT metrics.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20170914205735.18431-1-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/metricgroup.c

index fa37ef7..0ddd9c1 100644 (file)
@@ -373,9 +373,6 @@ static int metricgroup__add_metric(const char *metric, struct strbuf *events,
        int ret = -EINVAL;
        int i, j;
 
-       strbuf_init(events, 100);
-       strbuf_addf(events, "%s", "");
-
        if (!map)
                return 0;
 
@@ -433,6 +430,10 @@ static int metricgroup__add_metric_list(const char *list, struct strbuf *events,
        if (!nlist)
                return -ENOMEM;
        llist = nlist;
+
+       strbuf_init(events, 100);
+       strbuf_addf(events, "%s", "");
+
        while ((p = strsep(&llist, ",")) != NULL) {
                ret = metricgroup__add_metric(p, events, group_list);
                if (ret == -EINVAL) {