diff --git a/sys/ofed/drivers/net/mlx4/en_netdev.c b/sys/ofed/drivers/net/mlx4/en_netdev.c
index ac1214ab73cb..3c77385cac74 100644
--- a/sys/ofed/drivers/net/mlx4/en_netdev.c
+++ b/sys/ofed/drivers/net/mlx4/en_netdev.c
@@ -2681,6 +2681,8 @@ static void mlx4_en_sysctl_stat(struct mlx4_en_priv *priv)
 	SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_chksum_offload",
 	    CTLFLAG_RD, &priv->port_stats.tx_chksum_offload,
 	    "TX checksum offloads");
+	SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "defrag_attempts", CTLFLAG_RD,
+	    &priv->port_stats.defrag_attempts, "Oversized chains defragged");
 
 	/* Could strdup the names and add in a loop.  This is simpler. */
 	SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_bytes", CTLFLAG_RD,
@@ -2774,6 +2776,10 @@ static void mlx4_en_sysctl_stat(struct mlx4_en_priv *priv)
 		    CTLFLAG_RD, &tx_ring->packets, "TX packets");
 		SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "bytes",
 		    CTLFLAG_RD, &tx_ring->bytes, "TX bytes");
+		SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "tso_packets",
+		    CTLFLAG_RD, &tx_ring->tso_packets, "TSO packets");
+		SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "defrag_attempts",
+		    CTLFLAG_RD, &tx_ring->defrag_attempts, "Oversized chains defragged");
 	}
 
 	for (i = 0; i < priv->rx_ring_num; i++) {
diff --git a/sys/ofed/drivers/net/mlx4/en_port.c b/sys/ofed/drivers/net/mlx4/en_port.c
index 518337d48278..56bbf320249f 100644
--- a/sys/ofed/drivers/net/mlx4/en_port.c
+++ b/sys/ofed/drivers/net/mlx4/en_port.c
@@ -224,11 +224,16 @@ int mlx4_en_DUMP_ETH_STATS(struct mlx4_en_dev *mdev, u8 port, u8 reset)
 	priv->port_stats.tx_chksum_offload = 0;
 	priv->port_stats.queue_stopped = 0;
 	priv->port_stats.wake_queue = 0;
+	priv->port_stats.oversized_packets = 0;
+	priv->port_stats.tso_packets = 0;
+	priv->port_stats.defrag_attempts = 0;
 	for (i = 0; i < priv->tx_ring_num; i++) {
 		priv->port_stats.tx_chksum_offload += priv->tx_ring[i]->tx_csum;
 		priv->port_stats.queue_stopped += priv->tx_ring[i]->queue_stopped;
 		priv->port_stats.wake_queue += priv->tx_ring[i]->wake_queue;
 		priv->port_stats.oversized_packets += priv->tx_ring[i]->oversized_packets;
+		priv->port_stats.tso_packets += priv->tx_ring[i]->tso_packets;
+		priv->port_stats.defrag_attempts += priv->tx_ring[i]->defrag_attempts;
 	}
 	/* RX Statistics */
 	priv->pkstats.rx_packets = be64_to_cpu(mlx4_en_stats->RTOT_prio_0) +
diff --git a/sys/ofed/drivers/net/mlx4/en_tx.c b/sys/ofed/drivers/net/mlx4/en_tx.c
index 1fc82e29a56b..0fe4cf770a31 100644
--- a/sys/ofed/drivers/net/mlx4/en_tx.c
+++ b/sys/ofed/drivers/net/mlx4/en_tx.c
@@ -788,7 +788,7 @@ static int mlx4_en_xmit(struct mlx4_en_priv *priv, int tx_ind, struct mbuf **mbp
 			num_pkts = DIV_ROUND_UP(payload_len, mss);
 		ring->bytes += payload_len + (num_pkts * ihs);
 		ring->packets += num_pkts;
-		priv->port_stats.tso_packets++;
+		ring->tso_packets++;
 		/* store pointer to inline header */
 		dseg_inline = dseg;
 		/* copy data inline */
@@ -809,20 +809,11 @@ static int mlx4_en_xmit(struct mlx4_en_priv *priv, int tx_ind, struct mbuf **mbp
 	}
 	m_adj(mb, ihs);
 
-	/* trim off empty mbufs */
-	while (mb->m_len == 0) {
-		mb = m_free(mb);
-		/* check if all data has been inlined */
-		if (mb == NULL) {
-			nr_segs = 0;
-			goto skip_dma;
-		}
-	}
-
 	err = bus_dmamap_load_mbuf_sg(ring->dma_tag, tx_info->dma_map,
 	    mb, segs, &nr_segs, BUS_DMA_NOWAIT);
 	if (unlikely(err == EFBIG)) {
 		/* Too many mbuf fragments */
+		ring->defrag_attempts++;
 		m = m_defrag(mb, M_NOWAIT);
 		if (m == NULL) {
 			ring->oversized_packets++;
@@ -838,11 +829,18 @@ static int mlx4_en_xmit(struct mlx4_en_priv *priv, int tx_ind, struct mbuf **mbp
 		ring->oversized_packets++;
 		goto tx_drop;
 	}
-	/* make sure all mbuf data is written to RAM */
-	bus_dmamap_sync(ring->dma_tag, tx_info->dma_map,
-	    BUS_DMASYNC_PREWRITE);
+	/* If there were no errors and we didn't load anything, don't sync. */
+	if (nr_segs != 0) {
+		/* make sure all mbuf data is written to RAM */
+		bus_dmamap_sync(ring->dma_tag, tx_info->dma_map,
+		    BUS_DMASYNC_PREWRITE);
+	} else {
+		/* All data was inlined, free the mbuf. */
+		bus_dmamap_unload(ring->dma_tag, tx_info->dma_map);
+		m_freem(mb);
+		mb = NULL;
+	}
 
-skip_dma:
 	/* compute number of DS needed */
 	ds_cnt = (dseg - ((volatile struct mlx4_wqe_data_seg *)tx_desc)) + nr_segs;
 
diff --git a/sys/ofed/drivers/net/mlx4/mlx4_en.h b/sys/ofed/drivers/net/mlx4/mlx4_en.h
index 54d8b158c4b6..be566f4ddc69 100644
--- a/sys/ofed/drivers/net/mlx4/mlx4_en.h
+++ b/sys/ofed/drivers/net/mlx4/mlx4_en.h
@@ -278,6 +278,8 @@ struct mlx4_en_tx_ring {
 	unsigned long queue_stopped;
 	unsigned long oversized_packets;
 	unsigned long wake_queue;
+	unsigned long tso_packets;
+	unsigned long defrag_attempts;
 	struct mlx4_bf bf;
 	bool bf_enabled;
 	int hwtstamp_tx_type;
diff --git a/sys/ofed/drivers/net/mlx4/mlx4_stats.h b/sys/ofed/drivers/net/mlx4/mlx4_stats.h
index f70a83358839..3b86ea18cbd0 100644
--- a/sys/ofed/drivers/net/mlx4/mlx4_stats.h
+++ b/sys/ofed/drivers/net/mlx4/mlx4_stats.h
@@ -126,6 +126,7 @@ struct mlx4_en_port_stats {
 	unsigned long rx_chksum_good;
 	unsigned long rx_chksum_none;
 	unsigned long tx_chksum_offload;
+	unsigned long defrag_attempts;
 };
 
 struct mlx4_en_perf_stats {