diff --git a/include/spdk/bdev.h b/include/spdk/bdev.h
index d4da408b2..af094002a 100644
--- a/include/spdk/bdev.h
+++ b/include/spdk/bdev.h
@@ -115,6 +115,9 @@ struct spdk_bdev_io_stat {
 	uint64_t num_read_ops;
 	uint64_t bytes_written;
 	uint64_t num_write_ops;
+	uint64_t read_latency_ticks;
+	uint64_t write_latency_ticks;
+	uint64_t ticks_rate;
 };
 
 typedef void (*spdk_bdev_init_cb)(void *cb_arg, int rc);
diff --git a/include/spdk_internal/bdev.h b/include/spdk_internal/bdev.h
index 6d4b5c692..d58e3f10c 100644
--- a/include/spdk_internal/bdev.h
+++ b/include/spdk_internal/bdev.h
@@ -385,6 +385,9 @@ struct spdk_bdev_io {
 	/** It may be used by modules to put the bdev_io into its own list. */
 	TAILQ_ENTRY(spdk_bdev_io) module_link;
 
+	/** Current tsc at submit time. Used to calculate latency at completion. */
+	uint64_t submit_tsc;
+
 	/**
 	 * Per I/O context for use by the bdev module.
 	 */
diff --git a/lib/bdev/bdev.c b/lib/bdev/bdev.c
index 7a2404bbf..25b4c082b 100644
--- a/lib/bdev/bdev.c
+++ b/lib/bdev/bdev.c
@@ -807,6 +807,7 @@ spdk_bdev_io_submit(struct spdk_bdev_io *bdev_io)
 
 	assert(bdev_io->status == SPDK_BDEV_IO_STATUS_PENDING);
 
+	bdev_io->submit_tsc = spdk_get_ticks();
 	shared_ch->io_outstanding++;
 	bdev_io->in_submit_request = true;
 	if (spdk_likely(bdev_ch->flags == 0)) {
@@ -1664,6 +1665,7 @@ spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
 
 	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
 
+	channel->stat.ticks_rate = spdk_get_ticks_hz();
 	*stat = channel->stat;
 	memset(&channel->stat, 0, sizeof(channel->stat));
 }
@@ -1916,10 +1918,12 @@ spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status sta
 		case SPDK_BDEV_IO_TYPE_READ:
 			bdev_ch->stat.bytes_read += bdev_io->u.bdev.num_blocks * bdev->blocklen;
 			bdev_ch->stat.num_read_ops++;
+			bdev_ch->stat.read_latency_ticks += (spdk_get_ticks() - bdev_io->submit_tsc);
 			break;
 		case SPDK_BDEV_IO_TYPE_WRITE:
 			bdev_ch->stat.bytes_written += bdev_io->u.bdev.num_blocks * bdev->blocklen;
 			bdev_ch->stat.num_write_ops++;
+			bdev_ch->stat.write_latency_ticks += (spdk_get_ticks() - bdev_io->submit_tsc);
 			break;
 		default:
 			break;