perf metricgroups: Add NO_THRESHOLD_AND_NMI constraint

Thresholds can increase the number of counters a metric needs. The NMI
watchdog can take away a counter (hopefully the buddy watchdog will
become the default and this will no longer be true). Add a new
constraint for the case that a metric and its thresholds would fit in
counters but only if the NMI watchdog isn't enabled. Either the
threshold or the NMI watchdog should be disabled to make the metric
fit. Wire this up into the metric__group_events logic.

Signed-off-by: Ian Rogers <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Namhyung Kim <[email protected]>

Author: captain5050

tools/perf/pmu-events/jevents.py

@@ -235,6 +235,7 @@ def convert_metric_constraint(metric_constraint: str) -> Optional[str]:
           'NO_GROUP_EVENTS_NMI': '2',
           'NO_NMI_WATCHDOG': '2',
           'NO_GROUP_EVENTS_SMT': '3',
+          'NO_THRESHOLD_AND_NMI': '4',
       }
       return metric_constraint_to_enum[metric_constraint]
 

tools/perf/pmu-events/pmu-events.h

@@ -25,15 +25,21 @@ enum metric_event_groups {
 	 */
 	MetricNoGroupEvents = 1,
 	/**
-	 * @MetricNoGroupEventsNmi: Don't group events for the metric if the NMI
-	 *                          watchdog is enabled.
+	 * @MetricNoGroupEventsNmi:
+	 * Don't group events for the metric if the NMI watchdog is enabled.
 	 */
 	MetricNoGroupEventsNmi = 2,
 	/**
-	 * @MetricNoGroupEventsSmt: Don't group events for the metric if SMT is
-	 *                          enabled.
+	 * @MetricNoGroupEventsSmt:
+	 * Don't group events for the metric if SMT is enabled.
 	 */
 	MetricNoGroupEventsSmt = 3,
+	/**
+	 * @MetricNoGroupEventsThresholdAndNmi:
+	 * Don't group events for the metric thresholds and if the NMI watchdog
+	 * is enabled.
+	 */
+	MetricNoGroupEventsThresholdAndNmi = 4,
 };
 /*
  * Describe each PMU event. Each CPU has a table of PMU events.

tools/perf/util/metricgroup.c

@@ -179,7 +179,7 @@ static void metric__watchdog_constraint_hint(const char *name, bool foot)
 		   "    echo 1 > /proc/sys/kernel/nmi_watchdog\n");
 }
 
-static bool metric__group_events(const struct pmu_metric *pm)
+static bool metric__group_events(const struct pmu_metric *pm, bool metric_no_threshold)
 {
 	switch (pm->event_grouping) {
 	case MetricNoGroupEvents:
@@ -191,6 +191,13 @@ static bool metric__group_events(const struct pmu_metric *pm)
 		return false;
 	case MetricNoGroupEventsSmt:
 		return !smt_on();
+	case MetricNoGroupEventsThresholdAndNmi:
+		if (metric_no_threshold)
+			return true;
+		if (!sysctl__nmi_watchdog_enabled())
+			return true;
+		metric__watchdog_constraint_hint(pm->metric_name, /*foot=*/false);
+		return false;
 	case MetricGroupEvents:
 	default:
 		return true;
@@ -212,6 +219,7 @@ static void metric__free(struct metric *m)
 static struct metric *metric__new(const struct pmu_metric *pm,
 				  const char *modifier,
 				  bool metric_no_group,
+				  bool metric_no_threshold,
 				  int runtime,
 				  const char *user_requested_cpu_list,
 				  bool system_wide)
@@ -246,7 +254,7 @@ static struct metric *metric__new(const struct pmu_metric *pm,
 	}
 	m->pctx->sctx.runtime = runtime;
 	m->pctx->sctx.system_wide = system_wide;
-	m->group_events = !metric_no_group && metric__group_events(pm);
+	m->group_events = !metric_no_group && metric__group_events(pm, metric_no_threshold);
 	m->metric_refs = NULL;
 	m->evlist = NULL;
 
@@ -831,8 +839,8 @@ static int __add_metric(struct list_head *metric_list,
 		 * This metric is the root of a tree and may reference other
 		 * metrics that are added recursively.
 		 */
-		root_metric = metric__new(pm, modifier, metric_no_group, runtime,
-					  user_requested_cpu_list, system_wide);
+		root_metric = metric__new(pm, modifier, metric_no_group, metric_no_threshold,
+					  runtime, user_requested_cpu_list, system_wide);
 		if (!root_metric)
 			return -ENOMEM;