diff --git a/drivers/devfreq/arm-memlat-mon.c b/drivers/devfreq/arm-memlat-mon.c
index f79b61fdb018..1d9062f9dae0 100644
--- a/drivers/devfreq/arm-memlat-mon.c
+++ b/drivers/devfreq/arm-memlat-mon.c
@@ -28,6 +28,7 @@ enum ev_index {
 	INST_IDX,
 	CM_IDX,
 	CYC_IDX,
+	STALL_CYC_IDX,
 	NUM_EVENTS
 };
 #define INST_EV		0x08
@@ -92,12 +93,19 @@ static void read_perf_counters(int cpu, struct cpu_grp_info *cpu_grp)
 {
 	struct cpu_pmu_stats *cpustats = to_cpustats(cpu_grp, cpu);
 	struct dev_stats *devstats = to_devstats(cpu_grp, cpu);
-	unsigned long cyc_cnt;
+	unsigned long cyc_cnt, stall_cnt;
 
 	devstats->inst_count = read_event(&cpustats->events[INST_IDX]);
 	devstats->mem_count = read_event(&cpustats->events[CM_IDX]);
 	cyc_cnt = read_event(&cpustats->events[CYC_IDX]);
 	devstats->freq = compute_freq(cpustats, cyc_cnt);
+	if (cpustats->events[STALL_CYC_IDX].pevent) {
+		stall_cnt = read_event(&cpustats->events[STALL_CYC_IDX]);
+		stall_cnt = min(stall_cnt, cyc_cnt);
+		devstats->stall_pct = mult_frac(100, stall_cnt, cyc_cnt);
+	} else {
+		devstats->stall_pct = 100;
+	}
 }
 
 static unsigned long get_cnt(struct memlat_hwmon *hw)
@@ -117,7 +125,10 @@ static void delete_events(struct cpu_pmu_stats *cpustats)
 
 	for (i = 0; i < ARRAY_SIZE(cpustats->events); i++) {
 		cpustats->events[i].prev_count = 0;
-		perf_event_release_kernel(cpustats->events[i].pevent);
+		if (cpustats->events[i].pevent) {
+			perf_event_release_kernel(cpustats->events[i].pevent);
+			cpustats->events[i].pevent = NULL;
+		}
 	}
 }
 
@@ -135,6 +146,7 @@ static void stop_hwmon(struct memlat_hwmon *hw)
 		devstats->inst_count = 0;
 		devstats->mem_count = 0;
 		devstats->freq = 0;
+		devstats->stall_pct = 0;
 	}
 }
 
@@ -159,6 +171,7 @@ static int set_events(struct cpu_grp_info *cpu_grp, int cpu)
 	struct perf_event *pevent;
 	struct perf_event_attr *attr;
 	int err, i;
+	unsigned int event_id;
 	struct cpu_pmu_stats *cpustats = to_cpustats(cpu_grp, cpu);
 
 	/* Allocate an attribute for event initialization */
@@ -167,7 +180,11 @@ static int set_events(struct cpu_grp_info *cpu_grp, int cpu)
 		return -ENOMEM;
 
 	for (i = 0; i < ARRAY_SIZE(cpustats->events); i++) {
-		attr->config = cpu_grp->event_ids[i];
+		event_id = cpu_grp->event_ids[i];
+		if (!event_id)
+			continue;
+
+		attr->config = event_id;
 		pevent = perf_event_create_kernel_counter(attr, cpu, NULL,
 							  NULL, NULL);
 		if (IS_ERR(pevent))
@@ -282,6 +299,13 @@ static int arm_memlat_mon_driver_probe(struct platform_device *pdev)
 	}
 	cpu_grp->event_ids[INST_IDX] = event_id;
 
+	ret = of_property_read_u32(dev->of_node, "qcom,stall-cycle-ev",
+				   &event_id);
+	if (ret)
+		dev_dbg(dev, "Stall cycle event not specified. Event ignored.\n");
+	else
+		cpu_grp->event_ids[STALL_CYC_IDX] = event_id;
+
 	for_each_cpu(cpu, &cpu_grp->cpus)
 		to_devstats(cpu_grp, cpu)->id = cpu;
 
diff --git a/drivers/devfreq/governor_memlat.c b/drivers/devfreq/governor_memlat.c
index 3bd4fce79f99..34f05c079155 100644
--- a/drivers/devfreq/governor_memlat.c
+++ b/drivers/devfreq/governor_memlat.c
@@ -27,6 +27,7 @@
 
 struct memlat_node {
 	unsigned int ratio_ceil;
+	unsigned int stall_floor;
 	bool mon_started;
 	bool already_zero;
 	struct list_head list;
@@ -239,9 +240,11 @@ static int devfreq_memlat_get_freq(struct devfreq *df,
 					hw->core_stats[i].id,
 					hw->core_stats[i].inst_count,
 					hw->core_stats[i].mem_count,
-					hw->core_stats[i].freq, ratio);
+					hw->core_stats[i].freq,
+					hw->core_stats[i].stall_pct, ratio);
 
 		if (ratio <= node->ratio_ceil
+		    && hw->core_stats[i].stall_pct >= node->stall_floor
 		    && hw->core_stats[i].freq > max_freq) {
 			lat_dev = i;
 			max_freq = hw->core_stats[i].freq;
@@ -267,9 +270,11 @@ static int devfreq_memlat_get_freq(struct devfreq *df,
 }
 
 gov_attr(ratio_ceil, 1U, 10000U);
+gov_attr(stall_floor, 0U, 100U);
 
 static struct attribute *dev_attr[] = {
 	&dev_attr_ratio_ceil.attr,
+	&dev_attr_stall_floor.attr,
 	&dev_attr_freq_map.attr,
 	NULL,
 };
diff --git a/drivers/devfreq/governor_memlat.h b/drivers/devfreq/governor_memlat.h
index 1eeb75b62e39..a52ef25291f9 100644
--- a/drivers/devfreq/governor_memlat.h
+++ b/drivers/devfreq/governor_memlat.h
@@ -21,6 +21,7 @@ struct dev_stats {
 	unsigned long inst_count;
 	unsigned long mem_count;
 	unsigned long freq;
+	unsigned long stall_pct;
 };
 
 struct core_dev_map {
diff --git a/include/trace/events/power.h b/include/trace/events/power.h
index 4c287fcb40bf..dad9a5cd1bab 100644
--- a/include/trace/events/power.h
+++ b/include/trace/events/power.h
@@ -730,9 +730,10 @@ TRACE_EVENT(cache_hwmon_update,
 TRACE_EVENT(memlat_dev_meas,
 
 	TP_PROTO(const char *name, unsigned int dev_id, unsigned long inst,
-		 unsigned long mem, unsigned long freq, unsigned int ratio),
+		 unsigned long mem, unsigned long freq, unsigned int stall,
+		 unsigned int ratio),
 
-	TP_ARGS(name, dev_id, inst, mem, freq, ratio),
+	TP_ARGS(name, dev_id, inst, mem, freq, stall, ratio),
 
 	TP_STRUCT__entry(
 		__string(name, name)
@@ -740,6 +741,7 @@ TRACE_EVENT(memlat_dev_meas,
 		__field(unsigned long, inst)
 		__field(unsigned long, mem)
 		__field(unsigned long, freq)
+		__field(unsigned int, stall)
 		__field(unsigned int, ratio)
 	),
 
@@ -749,15 +751,17 @@ TRACE_EVENT(memlat_dev_meas,
 		__entry->inst = inst;
 		__entry->mem = mem;
 		__entry->freq = freq;
+		__entry->stall = stall;
 		__entry->ratio = ratio;
 	),
 
-	TP_printk("dev: %s, id=%u, inst=%lu, mem=%lu, freq=%lu, ratio=%u",
+	TP_printk("dev: %s, id=%u, inst=%lu, mem=%lu, freq=%lu, stall=%u, ratio=%u",
 		__get_str(name),
 		__entry->dev_id,
 		__entry->inst,
 		__entry->mem,
 		__entry->freq,
+		__entry->stall,
 		__entry->ratio)
 );