diff --git a/block/blk-core.c b/block/blk-core.c
index 4c090a0d8812..b2e57642dbb9 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -3583,76 +3583,43 @@ int __init blk_dev_init(void)
  * TODO : If necessary, we can make the histograms per-cpu and aggregate
  * them when printing them out.
  */
-void
-blk_zero_latency_hist(struct io_latency_state *s)
-{
-	memset(s->latency_y_axis_read, 0,
-	       sizeof(s->latency_y_axis_read));
-	memset(s->latency_y_axis_write, 0,
-	       sizeof(s->latency_y_axis_write));
-	s->latency_reads_elems = 0;
-	s->latency_writes_elems = 0;
-}
-EXPORT_SYMBOL(blk_zero_latency_hist);
-
 ssize_t
-blk_latency_hist_show(struct io_latency_state *s, char *buf)
+blk_latency_hist_show(char* name, struct io_latency_state *s, char *buf,
+		int buf_size)
 {
 	int i;
 	int bytes_written = 0;
 	u_int64_t num_elem, elem;
 	int pct;
+	u_int64_t average;
 
-	num_elem = s->latency_reads_elems;
-	if (num_elem > 0) {
-		bytes_written += scnprintf(buf + bytes_written,
-			   PAGE_SIZE - bytes_written,
-			   "IO svc_time Read Latency Histogram (n = %llu):\n",
-			   num_elem);
-		for (i = 0;
-		     i < ARRAY_SIZE(latency_x_axis_us);
-		     i++) {
-			elem = s->latency_y_axis_read[i];
-			pct = div64_u64(elem * 100, num_elem);
-			bytes_written += scnprintf(buf + bytes_written,
-						   PAGE_SIZE - bytes_written,
-						   "\t< %5lluus%15llu%15d%%\n",
-						   latency_x_axis_us[i],
-						   elem, pct);
-		}
-		/* Last element in y-axis table is overflow */
-		elem = s->latency_y_axis_read[i];
-		pct = div64_u64(elem * 100, num_elem);
-		bytes_written += scnprintf(buf + bytes_written,
-					   PAGE_SIZE - bytes_written,
-					   "\t> %5dms%15llu%15d%%\n", 10,
-					   elem, pct);
-	}
-	num_elem = s->latency_writes_elems;
-	if (num_elem > 0) {
-		bytes_written += scnprintf(buf + bytes_written,
-			   PAGE_SIZE - bytes_written,
-			   "IO svc_time Write Latency Histogram (n = %llu):\n",
-			   num_elem);
-		for (i = 0;
-		     i < ARRAY_SIZE(latency_x_axis_us);
-		     i++) {
-			elem = s->latency_y_axis_write[i];
-			pct = div64_u64(elem * 100, num_elem);
-			bytes_written += scnprintf(buf + bytes_written,
-						   PAGE_SIZE - bytes_written,
-						   "\t< %5lluus%15llu%15d%%\n",
-						   latency_x_axis_us[i],
-						   elem, pct);
-		}
-		/* Last element in y-axis table is overflow */
-		elem = s->latency_y_axis_write[i];
-		pct = div64_u64(elem * 100, num_elem);
-		bytes_written += scnprintf(buf + bytes_written,
-					   PAGE_SIZE - bytes_written,
-					   "\t> %5dms%15llu%15d%%\n", 10,
-					   elem, pct);
+       num_elem = s->latency_elems;
+       if (num_elem > 0) {
+	       average = div64_u64(s->latency_sum, s->latency_elems);
+	       bytes_written += scnprintf(buf + bytes_written,
+			       buf_size - bytes_written,
+			       "IO svc_time %s Latency Histogram (n = %llu,"
+			       " average = %llu):\n", name, num_elem, average);
+	       for (i = 0;
+		    i < ARRAY_SIZE(latency_x_axis_us);
+		    i++) {
+		       elem = s->latency_y_axis[i];
+		       pct = div64_u64(elem * 100, num_elem);
+		       bytes_written += scnprintf(buf + bytes_written,
+				       PAGE_SIZE - bytes_written,
+				       "\t< %6lluus%15llu%15d%%\n",
+				       latency_x_axis_us[i],
+				       elem, pct);
+	       }
+	       /* Last element in y-axis table is overflow */
+	       elem = s->latency_y_axis[i];
+	       pct = div64_u64(elem * 100, num_elem);
+	       bytes_written += scnprintf(buf + bytes_written,
+			       PAGE_SIZE - bytes_written,
+			       "\t>=%6lluus%15llu%15d%%\n",
+			       latency_x_axis_us[i - 1], elem, pct);
 	}
+
 	return bytes_written;
 }
 EXPORT_SYMBOL(blk_latency_hist_show);
diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
index 40ddc3e69a4d..00b9d7ef392f 100644
--- a/drivers/mmc/core/core.c
+++ b/drivers/mmc/core/core.c
@@ -209,9 +209,10 @@ void mmc_request_done(struct mmc_host *host, struct mmc_request *mrq)
 				completion = ktime_get();
 				delta_us = ktime_us_delta(completion,
 							  mrq->io_start);
-				blk_update_latency_hist(&host->io_lat_s,
-					(mrq->data->flags & MMC_DATA_READ),
-					delta_us);
+				blk_update_latency_hist(
+					(mrq->data->flags & MMC_DATA_READ) ?
+					&host->io_lat_read :
+					&host->io_lat_write, delta_us);
 			}
 #endif
 		}
@@ -3100,8 +3101,14 @@ static ssize_t
 latency_hist_show(struct device *dev, struct device_attribute *attr, char *buf)
 {
 	struct mmc_host *host = cls_dev_to_mmc_host(dev);
+	size_t written_bytes;
 
-	return blk_latency_hist_show(&host->io_lat_s, buf);
+	written_bytes = blk_latency_hist_show("Read", &host->io_lat_read,
+			buf, PAGE_SIZE);
+	written_bytes += blk_latency_hist_show("Write", &host->io_lat_write,
+			buf + written_bytes, PAGE_SIZE - written_bytes);
+
+	return written_bytes;
 }
 
 /*
@@ -3119,9 +3126,10 @@ latency_hist_store(struct device *dev, struct device_attribute *attr,
 
 	if (kstrtol(buf, 0, &value))
 		return -EINVAL;
-	if (value == BLK_IO_LAT_HIST_ZERO)
-		blk_zero_latency_hist(&host->io_lat_s);
-	else if (value == BLK_IO_LAT_HIST_ENABLE ||
+	if (value == BLK_IO_LAT_HIST_ZERO) {
+		memset(&host->io_lat_read, 0, sizeof(host->io_lat_read));
+		memset(&host->io_lat_write, 0, sizeof(host->io_lat_write));
+	} else if (value == BLK_IO_LAT_HIST_ENABLE ||
 		 value == BLK_IO_LAT_HIST_DISABLE)
 		host->latency_hist_enabled = value;
 	return count;
diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index f0a015aca608..fe893f23d693 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -3622,10 +3622,10 @@ static void __ufshcd_transfer_req_compl(struct ufs_hba *hba,
 					completion = ktime_get();
 					delta_us = ktime_us_delta(completion,
 						  req->lat_hist_io_start);
-					/* rq_data_dir() => true if WRITE */
-					blk_update_latency_hist(&hba->io_lat_s,
-						(rq_data_dir(req) == READ),
-						delta_us);
+					blk_update_latency_hist(
+						(rq_data_dir(req) == READ) ?
+						&hba->io_lat_read :
+						&hba->io_lat_write, delta_us);
 				}
 			}
 			/* Do not touch lrbp after scsi done */
@@ -6371,9 +6371,10 @@ latency_hist_store(struct device *dev, struct device_attribute *attr,
 
 	if (kstrtol(buf, 0, &value))
 		return -EINVAL;
-	if (value == BLK_IO_LAT_HIST_ZERO)
-		blk_zero_latency_hist(&hba->io_lat_s);
-	else if (value == BLK_IO_LAT_HIST_ENABLE ||
+	if (value == BLK_IO_LAT_HIST_ZERO) {
+		memset(&hba->io_lat_read, 0, sizeof(hba->io_lat_read));
+		memset(&hba->io_lat_write, 0, sizeof(hba->io_lat_write));
+	} else if (value == BLK_IO_LAT_HIST_ENABLE ||
 		 value == BLK_IO_LAT_HIST_DISABLE)
 		hba->latency_hist_enabled = value;
 	return count;
@@ -6384,8 +6385,14 @@ latency_hist_show(struct device *dev, struct device_attribute *attr,
 		  char *buf)
 {
 	struct ufs_hba *hba = dev_get_drvdata(dev);
+	size_t written_bytes;
 
-	return blk_latency_hist_show(&hba->io_lat_s, buf);
+	written_bytes = blk_latency_hist_show("Read", &hba->io_lat_read,
+			buf, PAGE_SIZE);
+	written_bytes += blk_latency_hist_show("Write", &hba->io_lat_write,
+			buf + written_bytes, PAGE_SIZE - written_bytes);
+
+	return written_bytes;
 }
 
 static DEVICE_ATTR(latency_hist, S_IRUGO | S_IWUSR,
diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
index dbe68b27c6cb..b35a5b999412 100644
--- a/drivers/scsi/ufs/ufshcd.h
+++ b/drivers/scsi/ufs/ufshcd.h
@@ -564,7 +564,8 @@ struct ufs_hba {
 	enum bkops_status urgent_bkops_lvl;
 	bool is_urgent_bkops_lvl_checked;
 	int latency_hist_enabled;
-	struct io_latency_state io_lat_s;
+	struct io_latency_state io_lat_read;
+	struct io_latency_state io_lat_write;
 };
 
 /* Returns true if clocks can be gated. Otherwise false */
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index e47a7f7025a0..fe8dd2797b1f 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1738,43 +1738,26 @@ static const u_int64_t latency_x_axis_us[] = {
 #define BLK_IO_LAT_HIST_ZERO            2
 
 struct io_latency_state {
-	u_int64_t	latency_y_axis_read[ARRAY_SIZE(latency_x_axis_us) + 1];
-	u_int64_t	latency_reads_elems;
-	u_int64_t	latency_y_axis_write[ARRAY_SIZE(latency_x_axis_us) + 1];
-	u_int64_t	latency_writes_elems;
+	u_int64_t	latency_y_axis[ARRAY_SIZE(latency_x_axis_us) + 1];
+	u_int64_t	latency_elems;
+	u_int64_t	latency_sum;
 };
 
 static inline void
-blk_update_latency_hist(struct io_latency_state *s,
-			int read,
-			u_int64_t delta_us)
+blk_update_latency_hist(struct io_latency_state *s, u_int64_t delta_us)
 {
 	int i;
 
-	for (i = 0; i < ARRAY_SIZE(latency_x_axis_us); i++) {
-		if (delta_us < (u_int64_t)latency_x_axis_us[i]) {
-			if (read)
-				s->latency_y_axis_read[i]++;
-			else
-				s->latency_y_axis_write[i]++;
+	for (i = 0; i < ARRAY_SIZE(latency_x_axis_us); i++)
+		if (delta_us < (u_int64_t)latency_x_axis_us[i])
 			break;
-		}
-	}
-	if (i == ARRAY_SIZE(latency_x_axis_us)) {
-		/* Overflowed the histogram */
-		if (read)
-			s->latency_y_axis_read[i]++;
-		else
-			s->latency_y_axis_write[i]++;
-	}
-	if (read)
-		s->latency_reads_elems++;
-	else
-		s->latency_writes_elems++;
+	s->latency_y_axis[i]++;
+	s->latency_elems++;
+	s->latency_sum += delta_us;
 }
 
-void blk_zero_latency_hist(struct io_latency_state *s);
-ssize_t blk_latency_hist_show(struct io_latency_state *s, char *buf);
+ssize_t blk_latency_hist_show(char* name, struct io_latency_state *s,
+		char *buf, int buf_size);
 
 #else /* CONFIG_BLOCK */
 
diff --git a/include/linux/mmc/host.h b/include/linux/mmc/host.h
index fac3b5c27f4f..1808b7895dac 100644
--- a/include/linux/mmc/host.h
+++ b/include/linux/mmc/host.h
@@ -409,7 +409,8 @@ struct mmc_host {
 
 #ifdef CONFIG_BLOCK
 	int			latency_hist_enabled;
-	struct io_latency_state io_lat_s;
+	struct io_latency_state io_lat_read;
+	struct io_latency_state io_lat_write;
 #endif
 
 	unsigned long		private[0] ____cacheline_aligned;