[PATCH] drivers/perf: riscv_pmu: Add riscv pmu pm notifier

[Eric Lin]

Currently, when the CPU is doing suspend to ram, we don't
save pmu counter register and its content will be lost.

To ensure perf profiling is not affected by suspend to ram,
this patch is based on arm_pmu CPU_PM notifier and implements riscv
pmu pm notifier. In the pm notifier, we stop the counter and update
the counter value before suspend and start the counter after resume.

Signed-off-by: Eric Lin <eric.lin at sifive.com>
---
 drivers/perf/riscv_pmu.c       |  4 +-
 drivers/perf/riscv_pmu_sbi.c   | 81 ++++++++++++++++++++++++++++++++--
 include/linux/perf/riscv_pmu.h |  4 ++
 3 files changed, 83 insertions(+), 6 deletions(-)

diff --git a/drivers/perf/riscv_pmu.c b/drivers/perf/riscv_pmu.c
index b2b8d2074ed0..2c961839903d 100644
--- a/drivers/perf/riscv_pmu.c
+++ b/drivers/perf/riscv_pmu.c
@@ -121,7 +121,7 @@ u64 riscv_pmu_event_update(struct perf_event *event)
 	return delta;
 }
 
-static void riscv_pmu_stop(struct perf_event *event, int flags)
+void riscv_pmu_stop(struct perf_event *event, int flags)
 {
 	struct hw_perf_event *hwc = &event->hw;
 	struct riscv_pmu *rvpmu = to_riscv_pmu(event->pmu);
@@ -175,7 +175,7 @@ int riscv_pmu_event_set_period(struct perf_event *event)
 	return overflow;
 }
 
-static void riscv_pmu_start(struct perf_event *event, int flags)
+void riscv_pmu_start(struct perf_event *event, int flags)
 {
 	struct hw_perf_event *hwc = &event->hw;
 	struct riscv_pmu *rvpmu = to_riscv_pmu(event->pmu);
diff --git a/drivers/perf/riscv_pmu_sbi.c b/drivers/perf/riscv_pmu_sbi.c
index dca3537a8dcc..adc910e0f1e2 100644
--- a/drivers/perf/riscv_pmu_sbi.c
+++ b/drivers/perf/riscv_pmu_sbi.c
@@ -17,6 +17,7 @@
 #include <linux/irqdomain.h>
 #include <linux/of_irq.h>
 #include <linux/of.h>
+#include <linux/cpu_pm.h>
 
 #include <asm/sbi.h>
 #include <asm/hwcap.h>
@@ -693,6 +694,73 @@ static int pmu_sbi_setup_irqs(struct riscv_pmu *pmu, struct platform_device *pde
 	return 0;
 }
 
+#ifdef CONFIG_CPU_PM
+static int riscv_pm_pmu_notify(struct notifier_block *b, unsigned long cmd,
+				void *v)
+{
+	struct riscv_pmu *rvpmu = container_of(b, struct riscv_pmu, riscv_pm_nb);
+	struct cpu_hw_events *cpuc = this_cpu_ptr(rvpmu->hw_events);
+	int enabled = bitmap_weight(cpuc->used_hw_ctrs, RISCV_MAX_COUNTERS);
+	struct perf_event *event;
+	int idx;
+
+	if (!enabled)
+		return NOTIFY_OK;
+
+	for (idx = 0; idx < RISCV_MAX_COUNTERS; idx++) {
+		event = cpuc->events[idx];
+		if (!event)
+			continue;
+
+		switch (cmd) {
+		case CPU_PM_ENTER:
+			/*
+			 * Stop and update the counter
+			 */
+			riscv_pmu_stop(event, PERF_EF_UPDATE);
+			break;
+		case CPU_PM_EXIT:
+		case CPU_PM_ENTER_FAILED:
+			/*
+			 * Restore and enable the counter.
+			 *
+			 * Requires RCU read locking to be functional,
+			 * wrap the call within RCU_NONIDLE to make the
+			 * RCU subsystem aware this cpu is not idle from
+			 * an RCU perspective for the riscv_pmu_start() call
+			 * duration.
+			 */
+			RCU_NONIDLE(riscv_pmu_start(event, PERF_EF_RELOAD));
+			break;
+		default:
+			break;
+		}
+	}
+
+	return NOTIFY_OK;
+}
+
+static int riscv_pm_pmu_register(struct riscv_pmu *pmu)
+{
+	pmu->riscv_pm_nb.notifier_call = riscv_pm_pmu_notify;
+	return cpu_pm_register_notifier(&pmu->riscv_pm_nb);
+}
+
+static void riscv_pm_pmu_unregister(struct riscv_pmu *pmu)
+{
+	cpu_pm_unregister_notifier(&pmu->riscv_pm_nb);
+}
+#else
+static inline int riscv_pm_pmu_register(struct riscv_pmu *pmu) { return 0; }
+static inline void riscv_pm_pmu_unregister(struct riscv_pmu *pmu) { }
+#endif
+
+static void riscv_pmu_destroy(struct riscv_pmu *pmu)
+{
+	riscv_pm_pmu_unregister(pmu);
+	cpuhp_state_remove_instance(CPUHP_AP_PERF_RISCV_STARTING, &pmu->node);
+}
+
 static int pmu_sbi_device_probe(struct platform_device *pdev)
 {
 	struct riscv_pmu *pmu = NULL;
@@ -733,14 +801,19 @@ static int pmu_sbi_device_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
+	ret = riscv_pm_pmu_register(pmu);
+	if (ret)
+		goto out_unregister;
+
 	ret = perf_pmu_register(&pmu->pmu, "cpu", PERF_TYPE_RAW);
-	if (ret) {
-		cpuhp_state_remove_instance(CPUHP_AP_PERF_RISCV_STARTING, &pmu->node);
-		return ret;
-	}
+	if (ret)
+		goto out_unregister;
 
 	return 0;
 
+out_unregister:
+	riscv_pmu_destroy(pmu);
+
 out_free:
 	kfree(pmu);
 	return ret;
diff --git a/include/linux/perf/riscv_pmu.h b/include/linux/perf/riscv_pmu.h
index 46f9b6fe306e..bf66fe011fa8 100644
--- a/include/linux/perf/riscv_pmu.h
+++ b/include/linux/perf/riscv_pmu.h
@@ -56,9 +56,13 @@ struct riscv_pmu {
 
 	struct cpu_hw_events	__percpu *hw_events;
 	struct hlist_node	node;
+	struct notifier_block   riscv_pm_nb;
 };
 
 #define to_riscv_pmu(p) (container_of(p, struct riscv_pmu, pmu))
+
+void riscv_pmu_start(struct perf_event *event, int flags);
+void riscv_pmu_stop(struct perf_event *event, int flags);
 unsigned long riscv_pmu_ctr_read_csr(unsigned long csr);
 int riscv_pmu_event_set_period(struct perf_event *event);
 uint64_t riscv_pmu_ctr_get_width_mask(struct perf_event *event);
-- 
2.36.1