[SRU][N:linux-aws][PATCH 1/2] perf/core: Optimize perf_adjust_freq_unthr_context()

[Philip Cox]

From: Namhyung Kim <namhyung at kernel.org>

BugLink: https://bugs.launchpad.net/bugs/2106004

It was unnecessarily disabling and enabling PMUs for each event.  It
should be done at PMU level.  Add pmu_ctx->nr_freq counter to check it
at each PMU.  As PMU context has separate active lists for pinned group
and flexible group, factor out a new function to do the job.

Another minor optimization is that it can skip PMUs w/ CAP_NO_INTERRUPT
even if it needs to unthrottle sampling events.

Signed-off-by: Namhyung Kim <namhyung at kernel.org>
Signed-off-by: Ingo Molnar <mingo at kernel.org>
Tested-by: Mingwei Zhang <mizhang at google.com>
Reviewed-by: Ian Rogers <irogers at google.com>
Reviewed-by: Kan Liang <kan.liang at linux.intel.com>
Link: https://lore.kernel.org/r/20240207050545.2727923-1-namhyung@kernel.org
(cherry picked from commit 0259bf63f71e2accfeca4a4e346ede8edcc86aab)
Signed-off-by: Philip Cox <philip.cox at canonical.com>
---
 include/linux/perf_event.h |  6 ++++
 kernel/events/core.c       | 70 ++++++++++++++++++++++++--------------
 2 files changed, 50 insertions(+), 26 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 676bcc5382ea..f5cafc6342eb 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -884,6 +884,7 @@ struct perf_event_pmu_context {
 
 	unsigned int			nr_events;
 	unsigned int			nr_cgroups;
+	unsigned int			nr_freq;
 
 	atomic_t			refcount; /* event <-> epc */
 	struct rcu_head			rcu_head;
@@ -898,6 +899,11 @@ struct perf_event_pmu_context {
 	int				rotate_necessary;
 };
 
+static inline bool perf_pmu_ctx_is_active(struct perf_event_pmu_context *epc)
+{
+	return !list_empty(&epc->flexible_active) || !list_empty(&epc->pinned_active);
+}
+
 struct perf_event_groups {
 	struct rb_root	tree;
 	u64		index;
diff --git a/kernel/events/core.c b/kernel/events/core.c
index bb2d2790c33b..5546618bab3e 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2308,8 +2308,10 @@ event_sched_out(struct perf_event *event, struct perf_event_context *ctx)
 
 	if (!is_software_event(event))
 		cpc->active_oncpu--;
-	if (event->attr.freq && event->attr.sample_freq)
+	if (event->attr.freq && event->attr.sample_freq) {
 		ctx->nr_freq--;
+		epc->nr_freq--;
+	}
 	if (event->attr.exclusive || !cpc->active_oncpu)
 		cpc->exclusive = 0;
 
@@ -2564,9 +2566,10 @@ event_sched_in(struct perf_event *event, struct perf_event_context *ctx)
 
 	if (!is_software_event(event))
 		cpc->active_oncpu++;
-	if (event->attr.freq && event->attr.sample_freq)
+	if (event->attr.freq && event->attr.sample_freq) {
 		ctx->nr_freq++;
-
+		epc->nr_freq++;
+	}
 	if (event->attr.exclusive)
 		cpc->exclusive = 1;
 
@@ -4133,30 +4136,14 @@ static void perf_adjust_period(struct perf_event *event, u64 nsec, u64 count, bo
 	}
 }
 
-/*
- * combine freq adjustment with unthrottling to avoid two passes over the
- * events. At the same time, make sure, having freq events does not change
- * the rate of unthrottling as that would introduce bias.
- */
-static void
-perf_adjust_freq_unthr_context(struct perf_event_context *ctx, bool unthrottle)
+static void perf_adjust_freq_unthr_events(struct list_head *event_list)
 {
 	struct perf_event *event;
 	struct hw_perf_event *hwc;
 	u64 now, period = TICK_NSEC;
 	s64 delta;
 
-	/*
-	 * only need to iterate over all events iff:
-	 * - context have events in frequency mode (needs freq adjust)
-	 * - there are events to unthrottle on this cpu
-	 */
-	if (!(ctx->nr_freq || unthrottle))
-		return;
-
-	raw_spin_lock(&ctx->lock);
-
-	list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
+	list_for_each_entry(event, event_list, active_list) {
 		if (event->state != PERF_EVENT_STATE_ACTIVE)
 			continue;
 
@@ -4164,8 +4151,6 @@ perf_adjust_freq_unthr_context(struct perf_event_context *ctx, bool unthrottle)
 		if (!event_filter_match(event))
 			continue;
 
-		perf_pmu_disable(event->pmu);
-
 		hwc = &event->hw;
 
 		if (hwc->interrupts == MAX_INTERRUPTS) {
@@ -4175,7 +4160,7 @@ perf_adjust_freq_unthr_context(struct perf_event_context *ctx, bool unthrottle)
 		}
 
 		if (!event->attr.freq || !event->attr.sample_freq)
-			goto next;
+			continue;
 
 		/*
 		 * stop the event and update event->count
@@ -4197,8 +4182,41 @@ perf_adjust_freq_unthr_context(struct perf_event_context *ctx, bool unthrottle)
 			perf_adjust_period(event, period, delta, false);
 
 		event->pmu->start(event, delta > 0 ? PERF_EF_RELOAD : 0);
-	next:
-		perf_pmu_enable(event->pmu);
+	}
+}
+
+/*
+ * combine freq adjustment with unthrottling to avoid two passes over the
+ * events. At the same time, make sure, having freq events does not change
+ * the rate of unthrottling as that would introduce bias.
+ */
+static void
+perf_adjust_freq_unthr_context(struct perf_event_context *ctx, bool unthrottle)
+{
+	struct perf_event_pmu_context *pmu_ctx;
+
+	/*
+	 * only need to iterate over all events iff:
+	 * - context have events in frequency mode (needs freq adjust)
+	 * - there are events to unthrottle on this cpu
+	 */
+	if (!(ctx->nr_freq || unthrottle))
+		return;
+
+	raw_spin_lock(&ctx->lock);
+
+	list_for_each_entry(pmu_ctx, &ctx->pmu_ctx_list, pmu_ctx_entry) {
+		if (!(pmu_ctx->nr_freq || unthrottle))
+			continue;
+		if (!perf_pmu_ctx_is_active(pmu_ctx))
+			continue;
+		if (pmu_ctx->pmu->capabilities & PERF_PMU_CAP_NO_INTERRUPT)
+			continue;
+
+		perf_pmu_disable(pmu_ctx->pmu);
+		perf_adjust_freq_unthr_events(&pmu_ctx->pinned_active);
+		perf_adjust_freq_unthr_events(&pmu_ctx->flexible_active);
+		perf_pmu_enable(pmu_ctx->pmu);
 	}
 
 	raw_spin_unlock(&ctx->lock);
-- 
2.43.0