perf parse-events: Extra care around force grouped events
authorIan Rogers <irogers@google.com>
Wed, 19 Jul 2023 00:18:34 +0000 (17:18 -0700)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Thu, 27 Jul 2023 13:27:12 +0000 (10:27 -0300)
Perf metric (topdown) events on Intel Icelake+ machines require a
group, however, they may be next to events that don't require a group.
Consider:

  cycles,slots,topdown-fe-bound

The cycles event needn't be grouped but slots and topdown-fe-bound need
grouping.

Prior to this change, as slots and topdown-fe-bound need a group forcing
and all events share the same PMU, slots and topdown-fe-bound would be
forced into a group with cycles.

This is a bug on two fronts, cycles wasn't supposed to be grouped and
cycles can't be a group leader with a perf metric event.

This change adds recognition that cycles isn't force grouped and so it
shouldn't be force grouped with slots and topdown-fe-bound.

Fixes: a90cc5a9eeab45ea ("perf evsel: Don't let evsel__group_pmu_name() traverse unsorted group")
Signed-off-by: Ian Rogers <irogers@google.com>
Tested-by: Andi Kleen <ak@linux.intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Xing Zhengjun <zhengjun.xing@linux.intel.com>
Link: https://lore.kernel.org/r/20230719001836.198363-2-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/parse-events.c

index acde097..af46835 100644 (file)
@@ -2149,7 +2149,7 @@ static int parse_events__sort_events_and_fix_groups(struct list_head *list)
        int idx = 0, unsorted_idx = -1;
        struct evsel *pos, *cur_leader = NULL;
        struct perf_evsel *cur_leaders_grp = NULL;
-       bool idx_changed = false;
+       bool idx_changed = false, cur_leader_force_grouped = false;
        int orig_num_leaders = 0, num_leaders = 0;
        int ret;
 
@@ -2190,7 +2190,7 @@ static int parse_events__sort_events_and_fix_groups(struct list_head *list)
                const struct evsel *pos_leader = evsel__leader(pos);
                const char *pos_pmu_name = pos->group_pmu_name;
                const char *cur_leader_pmu_name, *pos_leader_pmu_name;
-               bool force_grouped = arch_evsel__must_be_in_group(pos);
+               bool pos_force_grouped = arch_evsel__must_be_in_group(pos);
 
                /* Reset index and nr_members. */
                if (pos->core.idx != idx)
@@ -2206,7 +2206,8 @@ static int parse_events__sort_events_and_fix_groups(struct list_head *list)
                        cur_leader = pos;
 
                cur_leader_pmu_name = cur_leader->group_pmu_name;
-               if ((cur_leaders_grp != pos->core.leader && !force_grouped) ||
+               if ((cur_leaders_grp != pos->core.leader &&
+                    (!pos_force_grouped || !cur_leader_force_grouped)) ||
                    strcmp(cur_leader_pmu_name, pos_pmu_name)) {
                        /* Event is for a different group/PMU than last. */
                        cur_leader = pos;
@@ -2216,9 +2217,14 @@ static int parse_events__sort_events_and_fix_groups(struct list_head *list)
                         * group.
                         */
                        cur_leaders_grp = pos->core.leader;
+                       /*
+                        * Avoid forcing events into groups with events that
+                        * don't need to be in the group.
+                        */
+                       cur_leader_force_grouped = pos_force_grouped;
                }
                pos_leader_pmu_name = pos_leader->group_pmu_name;
-               if (strcmp(pos_leader_pmu_name, pos_pmu_name) || force_grouped) {
+               if (strcmp(pos_leader_pmu_name, pos_pmu_name) || pos_force_grouped) {
                        /*
                         * Event's PMU differs from its leader's. Groups can't
                         * span PMUs, so update leader from the group/PMU