Skip to contentSkip to conversation

perf tools: Ensure event leader stays at head of evlist after sorting

PATCH Superseded Chun-Tse Shao 2 participants 3 messages May 1 – May 5 perf events

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 perf tools: Ensure event leader stays at head of evlist after sorting

From: Ian Rogers <[email protected]>

For evlist of a certain event/metric, the HEAD should be the event leader. In some scenarios where uncore_xxx_0 does not exist, the event leader is not the first element after sorting. For example, on my test machine uncore_iio_0 does not exist, the event leader is uncore_iio_2.

However, in evlist__cmp, it was reordered based on the PMU name, which makes uncore_iio_1 the HEAD of evlist, breaking the following merge logic in evsel__merge_aliases.

The patch adds a loop at the end of parse_events__sort_events_and_fix_groups to make sure the first wildcard match is the earliest entry in the list, updating pointers accordingly without breaking reordering detection.

Tested on device lacks uncore_iio_0, and perf test looks good.

---
v2:
Adding missing Signed-off-by from Ian Rogers.
Renaming variable l to orig_leader.

v1: lore.kernel.org/[email protected]

 tools/perf/util/parse-events.c | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)
tools/perf/util/parse-events.c +27 -0
index 1497e1f2a08c..be825617321d 100644
@@ -2251,6 +2251,33 @@ static int parse_events__sort_events_and_fix_groups(struct list_head *list)
 		}
 		last_event_was_forced_leader = (force_grouped_leader == pos);
 	}
+
+	/*
+	 * Make sure the first wildcard match is the earliest entry in the list.
+	 * Since list_sort might have reordered the aliases, the original leader
+	 * might not be at the head of the list anymore. We find the first
+	 * alias in the sorted list and make it the new leader, and redirect
+	 * all other aliases to it.
+	 */
+	list_for_each_entry(pos, list, core.node) {
+		struct evsel *l = pos->first_wildcard_match;
+
+		if (!l)
+			continue;
+
+		if (l->first_wildcard_match) {
+			/* Original leader was redirected to a new leader */
+			pos->first_wildcard_match = l->first_wildcard_match;
+		} else if (pos->core.idx < l->core.idx) {
+			/*
+			 * We are earlier than the original leader in sorted order,
+			 * and no earlier alias has claimed leadership yet.
+			 */
+			l->first_wildcard_match = pos;
+			pos->first_wildcard_match = NULL;
+		}
+	}
+
 	list_for_each_entry(pos, list, core.node) {
 		struct evsel *pos_leader = evsel__leader(pos);
 

--
2.54.0.545.g6539524ca2-goog

maintainer
#1 lore raw
Quoted 16 lines · Chun-Tse Shao

From: Ian Rogers <[email protected]>

For evlist of a certain event/metric, the HEAD should be the event leader. In some scenarios where uncore_xxx_0 does not exist, the event leader is not the first element after sorting. For example, on my test machine uncore_iio_0 does not exist, the event leader is uncore_iio_2.

However, in evlist__cmp, it was reordered based on the PMU name, which makes uncore_iio_1 the HEAD of evlist, breaking the following merge logic in evsel__merge_aliases.

The patch adds a loop at the end of parse_events__sort_events_and_fix_groups to make sure the first wildcard match is the earliest entry in the list, updating pointers accordingly without breaking reordering detection.

Quoted 8 lines

Tested on device lacks uncore_iio_0, and perf test looks good.

---
v2:
Adding missing Signed-off-by from Ian Rogers.
Renaming variable l to orig_leader.

I can't see this.

Thanks,
Namhyung

Quoted 47 lines

v1: lore.kernel.org/[email protected]

 tools/perf/util/parse-events.c | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)
tools/perf/util/parse-events.c +27 -1
index 1497e1f2a08c..be825617321d 100644
@@ -2251,6 +2251,33 @@ static int parse_events__sort_events_and_fix_groups(struct list_head *list)
 		}
 		last_event_was_forced_leader = (force_grouped_leader == pos);
 	}
+
+	/*
+	 * Make sure the first wildcard match is the earliest entry in the list.
+	 * Since list_sort might have reordered the aliases, the original leader
+	 * might not be at the head of the list anymore. We find the first
+	 * alias in the sorted list and make it the new leader, and redirect
+	 * all other aliases to it.
+	 */
+	list_for_each_entry(pos, list, core.node) {
+		struct evsel *l = pos->first_wildcard_match;
+
+		if (!l)
+			continue;
+
+		if (l->first_wildcard_match) {
+			/* Original leader was redirected to a new leader */
+			pos->first_wildcard_match = l->first_wildcard_match;
+		} else if (pos->core.idx < l->core.idx) {
+			/*
+			 * We are earlier than the original leader in sorted order,
+			 * and no earlier alias has claimed leadership yet.
+			 */
+			l->first_wildcard_match = pos;
+			pos->first_wildcard_match = NULL;
+		}
+	}
+
 	list_for_each_entry(pos, list, core.node) {
 		struct evsel *pos_leader = evsel__leader(pos);
 
--

2.54.0.545.g6539524ca2-goog

#2 lore raw

My apologies, resend the patch:
lore.kernel.org/[email protected]

Quoted 79 lines · Namhyung Kim
Quoted 16 lines · Chun-Tse Shao

From: Ian Rogers <[email protected]>

For evlist of a certain event/metric, the HEAD should be the event leader. In some scenarios where uncore_xxx_0 does not exist, the event leader is not the first element after sorting. For example, on my test machine uncore_iio_0 does not exist, the event leader is uncore_iio_2.

However, in evlist__cmp, it was reordered based on the PMU name, which makes uncore_iio_1 the HEAD of evlist, breaking the following merge logic in evsel__merge_aliases.

The patch adds a loop at the end of parse_events__sort_events_and_fix_groups to make sure the first wildcard match is the earliest entry in the list, updating pointers accordingly without breaking reordering detection.

Quoted 8 lines

Tested on device lacks uncore_iio_0, and perf test looks good.

---
v2:
Adding missing Signed-off-by from Ian Rogers.
Renaming variable l to orig_leader.

I can't see this.

Thanks,
Namhyung

Quoted 47 lines

v1: lore.kernel.org/[email protected]

 tools/perf/util/parse-events.c | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)
tools/perf/util/parse-events.c +27 -1
index 1497e1f2a08c..be825617321d 100644
@@ -2251,6 +2251,33 @@ static int parse_events__sort_events_and_fix_groups(struct list_head *list)
              }
              last_event_was_forced_leader = (force_grouped_leader == pos);
      }
+
+     /*
+      * Make sure the first wildcard match is the earliest entry in the list.
+      * Since list_sort might have reordered the aliases, the original leader
+      * might not be at the head of the list anymore. We find the first
+      * alias in the sorted list and make it the new leader, and redirect
+      * all other aliases to it.
+      */
+     list_for_each_entry(pos, list, core.node) {
+             struct evsel *l = pos->first_wildcard_match;
+
+             if (!l)
+                     continue;
+
+             if (l->first_wildcard_match) {
+                     /* Original leader was redirected to a new leader */
+                     pos->first_wildcard_match = l->first_wildcard_match;
+             } else if (pos->core.idx < l->core.idx) {
+                     /*
+                      * We are earlier than the original leader in sorted order,
+                      * and no earlier alias has claimed leadership yet.
+                      */
+                     l->first_wildcard_match = pos;
+                     pos->first_wildcard_match = NULL;
+             }
+     }
+
      list_for_each_entry(pos, list, core.node) {
              struct evsel *pos_leader = evsel__leader(pos);
 
--

2.54.0.545.g6539524ca2-goog