Merge branch 'mlx4-next'

Or Gerlitz says:

====================
mlx4: Extend statistics gathering and display

This series from Eran and Matan extends the statistics collected and later
reported to the user via ethtool for native mode and SRIOV PF.

More Packet statistics, PFC statistics and global pause statistics
are now displayed via ethtool.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

9 files changed, 546 insertions, 151 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_dcb_nl.c b/drivers/net/ethernet/mellanox/mlx4/en_dcb_nl.c
index cde14fa2f742..8e3260c0eaa5 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_dcb_nl.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_dcb_nl.c
@@ -226,6 +226,10 @@ static int mlx4_en_dcbnl_ieee_setpfc(struct net_device *dev,
 				    prof->rx_ppp);
 	if (err)
 		en_err(priv, "Failed setting pause params\n");
+	else
+		mlx4_en_update_pfc_stats_bitmap(mdev->dev, &priv->stats_bitmap,
+						prof->rx_ppp, prof->rx_pause,
+						prof->tx_ppp, prof->tx_pause);
 
 	return err;
 }
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
index a7b58ba8492b..eba969b08dd1 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
@@ -38,6 +38,7 @@
 #include <linux/mlx4/device.h>
 #include <linux/in.h>
 #include <net/ip.h>
+#include <linux/bitmap.h>
 
 #include "mlx4_en.h"
 #include "en_port.h"
@@ -104,6 +105,7 @@ static const char mlx4_en_priv_flags[][ETH_GSTRING_LEN] = {
 };
 
 static const char main_strings[][ETH_GSTRING_LEN] = {
+	/* main statistics */
 	"rx_packets", "tx_packets", "rx_bytes", "tx_bytes", "rx_errors",
 	"tx_errors", "rx_dropped", "tx_dropped", "multicast", "collisions",
 	"rx_length_errors", "rx_over_errors", "rx_crc_errors",
@@ -117,14 +119,76 @@ static const char main_strings[][ETH_GSTRING_LEN] = {
 	"queue_stopped", "wake_queue", "tx_timeout", "rx_alloc_failed",
 	"rx_csum_good", "rx_csum_none", "rx_csum_complete", "tx_chksum_offload",
 
+	/* priority flow control statistics rx */
+	"rx_pause_prio_0", "rx_pause_duration_prio_0",
+	"rx_pause_transition_prio_0",
+	"rx_pause_prio_1", "rx_pause_duration_prio_1",
+	"rx_pause_transition_prio_1",
+	"rx_pause_prio_2", "rx_pause_duration_prio_2",
+	"rx_pause_transition_prio_2",
+	"rx_pause_prio_3", "rx_pause_duration_prio_3",
+	"rx_pause_transition_prio_3",
+	"rx_pause_prio_4", "rx_pause_duration_prio_4",
+	"rx_pause_transition_prio_4",
+	"rx_pause_prio_5", "rx_pause_duration_prio_5",
+	"rx_pause_transition_prio_5",
+	"rx_pause_prio_6", "rx_pause_duration_prio_6",
+	"rx_pause_transition_prio_6",
+	"rx_pause_prio_7", "rx_pause_duration_prio_7",
+	"rx_pause_transition_prio_7",
+
+	/* flow control statistics rx */
+	"rx_pause", "rx_pause_duration", "rx_pause_transition",
+
+	/* priority flow control statistics tx */
+	"tx_pause_prio_0", "tx_pause_duration_prio_0",
+	"tx_pause_transition_prio_0",
+	"tx_pause_prio_1", "tx_pause_duration_prio_1",
+	"tx_pause_transition_prio_1",
+	"tx_pause_prio_2", "tx_pause_duration_prio_2",
+	"tx_pause_transition_prio_2",
+	"tx_pause_prio_3", "tx_pause_duration_prio_3",
+	"tx_pause_transition_prio_3",
+	"tx_pause_prio_4", "tx_pause_duration_prio_4",
+	"tx_pause_transition_prio_4",
+	"tx_pause_prio_5", "tx_pause_duration_prio_5",
+	"tx_pause_transition_prio_5",
+	"tx_pause_prio_6", "tx_pause_duration_prio_6",
+	"tx_pause_transition_prio_6",
+	"tx_pause_prio_7", "tx_pause_duration_prio_7",
+	"tx_pause_transition_prio_7",
+
+	/* flow control statistics tx */
+	"tx_pause", "tx_pause_duration", "tx_pause_transition",
+
 	/* packet statistics */
-	"broadcast", "rx_prio_0", "rx_prio_1", "rx_prio_2", "rx_prio_3",
-	"rx_prio_4", "rx_prio_5", "rx_prio_6", "rx_prio_7", "tx_prio_0",
-	"tx_prio_1", "tx_prio_2", "tx_prio_3", "tx_prio_4", "tx_prio_5",
-	"tx_prio_6", "tx_prio_7",
+	"rx_multicast_packets",
+	"rx_broadcast_packets",
+	"rx_jabbers",
+	"rx_in_range_length_error",
+	"rx_out_range_length_error",
+	"tx_multicast_packets",
+	"tx_broadcast_packets",
+	"rx_prio_0_packets", "rx_prio_0_bytes",
+	"rx_prio_1_packets", "rx_prio_1_bytes",
+	"rx_prio_2_packets", "rx_prio_2_bytes",
+	"rx_prio_3_packets", "rx_prio_3_bytes",
+	"rx_prio_4_packets", "rx_prio_4_bytes",
+	"rx_prio_5_packets", "rx_prio_5_bytes",
+	"rx_prio_6_packets", "rx_prio_6_bytes",
+	"rx_prio_7_packets", "rx_prio_7_bytes",
+	"rx_novlan_packets", "rx_novlan_bytes",
+	"tx_prio_0_packets", "tx_prio_0_bytes",
+	"tx_prio_1_packets", "tx_prio_1_bytes",
+	"tx_prio_2_packets", "tx_prio_2_bytes",
+	"tx_prio_3_packets", "tx_prio_3_bytes",
+	"tx_prio_4_packets", "tx_prio_4_bytes",
+	"tx_prio_5_packets", "tx_prio_5_bytes",
+	"tx_prio_6_packets", "tx_prio_6_bytes",
+	"tx_prio_7_packets", "tx_prio_7_bytes",
+	"tx_novlan_packets", "tx_novlan_bytes",
+
 };
-#define NUM_MAIN_STATS	21
-#define NUM_ALL_STATS	(NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PKT_STATS + NUM_PERF_STATS)
 
 static const char mlx4_en_test_names[][ETH_GSTRING_LEN]= {
 	"Interrupt Test",
@@ -224,14 +288,50 @@ static int mlx4_en_set_wol(struct net_device *netdev,
 	return err;
 }
 
+struct bitmap_iterator {
+	unsigned long *stats_bitmap;
+	unsigned int count;
+	unsigned int iterator;
+	bool advance_array; /* if set, force no increments */
+};
+
+static inline void bitmap_iterator_init(struct bitmap_iterator *h,
+					unsigned long *stats_bitmap,
+					int count)
+{
+	h->iterator = 0;
+	h->advance_array = !bitmap_empty(stats_bitmap, count);
+	h->count = h->advance_array ? bitmap_weight(stats_bitmap, count)
+		: count;
+	h->stats_bitmap = stats_bitmap;
+}
+
+static inline int bitmap_iterator_test(struct bitmap_iterator *h)
+{
+	return !h->advance_array ? 1 : test_bit(h->iterator, h->stats_bitmap);
+}
+
+static inline int bitmap_iterator_inc(struct bitmap_iterator *h)
+{
+	return h->iterator++;
+}
+
+static inline unsigned int
+bitmap_iterator_count(struct bitmap_iterator *h)
+{
+	return h->count;
+}
+
 static int mlx4_en_get_sset_count(struct net_device *dev, int sset)
 {
 	struct mlx4_en_priv *priv = netdev_priv(dev);
-	int bit_count = hweight64(priv->stats_bitmap);
+	struct bitmap_iterator it;
+
+	bitmap_iterator_init(&it, priv->stats_bitmap.bitmap, NUM_ALL_STATS);
 
 	switch (sset) {
 	case ETH_SS_STATS:
-		return (priv->stats_bitmap ? bit_count : NUM_ALL_STATS) +
+		return bitmap_iterator_count(&it) +
 			(priv->tx_ring_num * 2) +
 #ifdef CONFIG_NET_RX_BUSY_POLL
 			(priv->rx_ring_num * 5);
@@ -253,34 +353,45 @@ static void mlx4_en_get_ethtool_stats(struct net_device *dev,
 {
 	struct mlx4_en_priv *priv = netdev_priv(dev);
 	int index = 0;
-	int i, j = 0;
+	int i;
+	struct bitmap_iterator it;
+
+	bitmap_iterator_init(&it, priv->stats_bitmap.bitmap, NUM_ALL_STATS);
 
 	spin_lock_bh(&priv->stats_lock);
 
-	if (!(priv->stats_bitmap)) {
-		for (i = 0; i < NUM_MAIN_STATS; i++)
-			data[index++] =
-				((unsigned long *) &priv->stats)[i];
-		for (i = 0; i < NUM_PORT_STATS; i++)
+	for (i = 0; i < NUM_MAIN_STATS; i++, bitmap_iterator_inc(&it))
+		if (bitmap_iterator_test(&it))
+			data[index++] = ((unsigned long *)&priv->stats)[i];
+
+	for (i = 0; i < NUM_PORT_STATS; i++, bitmap_iterator_inc(&it))
+		if (bitmap_iterator_test(&it))
+			data[index++] = ((unsigned long *)&priv->port_stats)[i];
+
+	for (i = 0; i < NUM_FLOW_PRIORITY_STATS_RX;
+	     i++, bitmap_iterator_inc(&it))
+		if (bitmap_iterator_test(&it))
 			data[index++] =
-				((unsigned long *) &priv->port_stats)[i];
-		for (i = 0; i < NUM_PKT_STATS; i++)
+				((u64 *)&priv->rx_priority_flowstats)[i];
+
+	for (i = 0; i < NUM_FLOW_STATS_RX; i++, bitmap_iterator_inc(&it))
+		if (bitmap_iterator_test(&it))
+			data[index++] = ((u64 *)&priv->rx_flowstats)[i];
+
+	for (i = 0; i < NUM_FLOW_PRIORITY_STATS_TX;
+	     i++, bitmap_iterator_inc(&it))
+		if (bitmap_iterator_test(&it))
 			data[index++] =
-				((unsigned long *) &priv->pkstats)[i];
-	} else {
-		for (i = 0; i < NUM_MAIN_STATS; i++) {
-			if ((priv->stats_bitmap >> j) & 1)
-				data[index++] =
-				((unsigned long *) &priv->stats)[i];
-			j++;
-		}
-		for (i = 0; i < NUM_PORT_STATS; i++) {
-			if ((priv->stats_bitmap >> j) & 1)
-				data[index++] =
-				((unsigned long *) &priv->port_stats)[i];
-			j++;
-		}
-	}
+				((u64 *)&priv->tx_priority_flowstats)[i];
+
+	for (i = 0; i < NUM_FLOW_STATS_TX; i++, bitmap_iterator_inc(&it))
+		if (bitmap_iterator_test(&it))
+			data[index++] = ((u64 *)&priv->tx_flowstats)[i];
+
+	for (i = 0; i < NUM_PKT_STATS; i++, bitmap_iterator_inc(&it))
+		if (bitmap_iterator_test(&it))
+			data[index++] = ((unsigned long *)&priv->pkstats)[i];
+
 	for (i = 0; i < priv->tx_ring_num; i++) {
 		data[index++] = priv->tx_ring[i]->packets;
 		data[index++] = priv->tx_ring[i]->bytes;
@@ -309,7 +420,10 @@ static void mlx4_en_get_strings(struct net_device *dev,
 {
 	struct mlx4_en_priv *priv = netdev_priv(dev);
 	int index = 0;
-	int i;
+	int i, strings = 0;
+	struct bitmap_iterator it;
+
+	bitmap_iterator_init(&it, priv->stats_bitmap.bitmap, NUM_ALL_STATS);
 
 	switch (stringset) {
 	case ETH_SS_TEST:
@@ -322,29 +436,30 @@ static void mlx4_en_get_strings(struct net_device *dev,
 
 	case ETH_SS_STATS:
 		/* Add main counters */
-		if (!priv->stats_bitmap) {
-			for (i = 0; i < NUM_MAIN_STATS; i++)
+		for (i = 0; i < NUM_MAIN_STATS; i++, strings++,
+		     bitmap_iterator_inc(&it))
+			if (bitmap_iterator_test(&it))
 				strcpy(data + (index++) * ETH_GSTRING_LEN,
-					main_strings[i]);
-			for (i = 0; i < NUM_PORT_STATS; i++)
+				       main_strings[strings]);
+
+		for (i = 0; i < NUM_PORT_STATS; i++, strings++,
+		     bitmap_iterator_inc(&it))
+			if (bitmap_iterator_test(&it))
 				strcpy(data + (index++) * ETH_GSTRING_LEN,
-					main_strings[i +
-					NUM_MAIN_STATS]);
-			for (i = 0; i < NUM_PKT_STATS; i++)
+				       main_strings[strings]);
+
+		for (i = 0; i < NUM_FLOW_STATS; i++, strings++,
+		     bitmap_iterator_inc(&it))
+			if (bitmap_iterator_test(&it))
 				strcpy(data + (index++) * ETH_GSTRING_LEN,
-					main_strings[i +
-					NUM_MAIN_STATS +
-					NUM_PORT_STATS]);
-		} else
-			for (i = 0; i < NUM_MAIN_STATS + NUM_PORT_STATS; i++) {
-				if ((priv->stats_bitmap >> i) & 1) {
-					strcpy(data +
-					       (index++) * ETH_GSTRING_LEN,
-					       main_strings[i]);
-				}
-				if (!(priv->stats_bitmap >> i))
-					break;
-			}
+				       main_strings[strings]);
+
+		for (i = 0; i < NUM_PKT_STATS; i++, strings++,
+		     bitmap_iterator_inc(&it))
+			if (bitmap_iterator_test(&it))
+				strcpy(data + (index++) * ETH_GSTRING_LEN,
+				       main_strings[strings]);
+
 		for (i = 0; i < priv->tx_ring_num; i++) {
 			sprintf(data + (index++) * ETH_GSTRING_LEN,
 				"tx%d_packets", i);
@@ -885,6 +1000,12 @@ static int mlx4_en_set_pauseparam(struct net_device *dev,
 				    priv->prof->rx_ppp);
 	if (err)
 		en_err(priv, "Failed setting pause params\n");
+	else
+		mlx4_en_update_pfc_stats_bitmap(mdev->dev, &priv->stats_bitmap,
+						priv->prof->rx_ppp,
+						priv->prof->rx_pause,
+						priv->prof->tx_ppp,
+						priv->prof->tx_pause);
 
 	return err;
 }
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
index ebc93a101c93..354e254b53cf 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
@@ -1888,6 +1888,12 @@ static void mlx4_en_clear_stats(struct net_device *dev)
 	memset(&priv->pstats, 0, sizeof(priv->pstats));
 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
+	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
+	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
+	memset(&priv->rx_priority_flowstats, 0,
+	       sizeof(priv->rx_priority_flowstats));
+	memset(&priv->tx_priority_flowstats, 0,
+	       sizeof(priv->tx_priority_flowstats));
 
 	for (i = 0; i < priv->tx_ring_num; i++) {
 		priv->tx_ring[i]->bytes = 0;
@@ -2648,6 +2654,82 @@ int mlx4_en_netdev_event(struct notifier_block *this,
 	return NOTIFY_DONE;
 }
 
+void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
+				     struct mlx4_en_stats_bitmap *stats_bitmap,
+				     u8 rx_ppp, u8 rx_pause,
+				     u8 tx_ppp, u8 tx_pause)
+{
+	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS;
+
+	if (!mlx4_is_slave(dev) &&
+	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
+		mutex_lock(&stats_bitmap->mutex);
+		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
+
+		if (rx_ppp)
+			bitmap_set(stats_bitmap->bitmap, last_i,
+				   NUM_FLOW_PRIORITY_STATS_RX);
+		last_i += NUM_FLOW_PRIORITY_STATS_RX;
+
+		if (rx_pause && !(rx_ppp))
+			bitmap_set(stats_bitmap->bitmap, last_i,
+				   NUM_FLOW_STATS_RX);
+		last_i += NUM_FLOW_STATS_RX;
+
+		if (tx_ppp)
+			bitmap_set(stats_bitmap->bitmap, last_i,
+				   NUM_FLOW_PRIORITY_STATS_TX);
+		last_i += NUM_FLOW_PRIORITY_STATS_TX;
+
+		if (tx_pause && !(tx_ppp))
+			bitmap_set(stats_bitmap->bitmap, last_i,
+				   NUM_FLOW_STATS_TX);
+		last_i += NUM_FLOW_STATS_TX;
+
+		mutex_unlock(&stats_bitmap->mutex);
+	}
+}
+
+void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
+			      struct mlx4_en_stats_bitmap *stats_bitmap,
+			      u8 rx_ppp, u8 rx_pause,
+			      u8 tx_ppp, u8 tx_pause)
+{
+	int last_i = 0;
+
+	mutex_init(&stats_bitmap->mutex);
+	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
+
+	if (mlx4_is_slave(dev)) {
+		bitmap_set(stats_bitmap->bitmap, last_i +
+					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
+		bitmap_set(stats_bitmap->bitmap, last_i +
+					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
+		bitmap_set(stats_bitmap->bitmap, last_i +
+					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
+		bitmap_set(stats_bitmap->bitmap, last_i +
+					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
+		bitmap_set(stats_bitmap->bitmap, last_i +
+					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
+		bitmap_set(stats_bitmap->bitmap, last_i +
+					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
+	} else {
+		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
+	}
+	last_i += NUM_MAIN_STATS;
+
+	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
+	last_i += NUM_PORT_STATS;
+
+	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
+					rx_ppp, rx_pause,
+					tx_ppp, tx_pause);
+	last_i += NUM_FLOW_STATS;
+
+	if (!mlx4_is_slave(dev))
+		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
+}
+
 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
 			struct mlx4_en_port_profile *prof)
 {
@@ -2881,7 +2963,11 @@ int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
 		queue_delayed_work(mdev->workqueue, &priv->service_task,
 				   SERVICE_TASK_DELAY);
 
-	mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
+	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
+				 mdev->profile.prof[priv->port].rx_ppp,
+				 mdev->profile.prof[priv->port].rx_pause,
+				 mdev->profile.prof[priv->port].tx_ppp,
+				 mdev->profile.prof[priv->port].tx_pause);
 
 	return 0;
 
diff --git a/drivers/net/ethernet/mellanox/mlx4/en_port.c b/drivers/net/ethernet/mellanox/mlx4/en_port.c
index 6cb80072af6c..54f0e5ab2e55 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_port.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_port.c
@@ -128,9 +128,29 @@ out:
 	return err;
 }
 
+/* Each counter set is located in struct mlx4_en_stat_out_mbox
+ * with a const offset between its prio components.
+ * This function runs over a counter set and sum all of it's prio components.
+ */
+static unsigned long en_stats_adder(__be64 *start, __be64 *next, int num)
+{
+	__be64 *curr = start;
+	unsigned long ret = 0;
+	int i;
+	int offset = next - start;
+
+	for (i = 0; i <= num; i++) {
+		ret += be64_to_cpu(*curr);
+		curr += offset;
+	}
+
+	return ret;
+}
+
 int mlx4_en_DUMP_ETH_STATS(struct mlx4_en_dev *mdev, u8 port, u8 reset)
 {
 	struct mlx4_en_stat_out_mbox *mlx4_en_stats;
+	struct mlx4_en_stat_out_flow_control_mbox *flowstats;
 	struct mlx4_en_priv *priv = netdev_priv(mdev->pndev[port]);
 	struct net_device_stats *stats = &priv->stats;
 	struct mlx4_cmd_mailbox *mailbox;
@@ -183,22 +203,25 @@ int mlx4_en_DUMP_ETH_STATS(struct mlx4_en_dev *mdev, u8 port, u8 reset)
 		priv->port_stats.xmit_more         += ring->xmit_more;
 	}
 
+	/* net device stats */
 	stats->rx_errors = be64_to_cpu(mlx4_en_stats->PCS) +
-			   be32_to_cpu(mlx4_en_stats->RdropLength) +
 			   be32_to_cpu(mlx4_en_stats->RJBBR) +
 			   be32_to_cpu(mlx4_en_stats->RCRC) +
-			   be32_to_cpu(mlx4_en_stats->RRUNT);
-	stats->tx_errors = be32_to_cpu(mlx4_en_stats->TDROP);
-	stats->multicast = be64_to_cpu(mlx4_en_stats->MCAST_prio_0) +
-			   be64_to_cpu(mlx4_en_stats->MCAST_prio_1) +
-			   be64_to_cpu(mlx4_en_stats->MCAST_prio_2) +
-			   be64_to_cpu(mlx4_en_stats->MCAST_prio_3) +
-			   be64_to_cpu(mlx4_en_stats->MCAST_prio_4) +
-			   be64_to_cpu(mlx4_en_stats->MCAST_prio_5) +
-			   be64_to_cpu(mlx4_en_stats->MCAST_prio_6) +
-			   be64_to_cpu(mlx4_en_stats->MCAST_prio_7) +
-			   be64_to_cpu(mlx4_en_stats->MCAST_novlan);
+			   be32_to_cpu(mlx4_en_stats->RRUNT) +
+			   be64_to_cpu(mlx4_en_stats->RInRangeLengthErr) +
+			   be64_to_cpu(mlx4_en_stats->ROutRangeLengthErr) +
+			   be32_to_cpu(mlx4_en_stats->RSHORT) +
+			   en_stats_adder(&mlx4_en_stats->RGIANT_prio_0,
+					  &mlx4_en_stats->RGIANT_prio_1,
+					  NUM_PRIORITIES);
+	stats->tx_errors = en_stats_adder(&mlx4_en_stats->TGIANT_prio_0,
+					  &mlx4_en_stats->TGIANT_prio_1,
+					  NUM_PRIORITIES);
+	stats->multicast = en_stats_adder(&mlx4_en_stats->MCAST_prio_0,
+					  &mlx4_en_stats->MCAST_prio_1,
+					  NUM_PRIORITIES);
 	stats->collisions = 0;
+	stats->rx_dropped = be32_to_cpu(mlx4_en_stats->RDROP);
 	stats->rx_length_errors = be32_to_cpu(mlx4_en_stats->RdropLength);
 	stats->rx_over_errors = be32_to_cpu(mlx4_en_stats->RdropOvflw);
 	stats->rx_crc_errors = be32_to_cpu(mlx4_en_stats->RCRC);
@@ -210,33 +233,116 @@ int mlx4_en_DUMP_ETH_STATS(struct mlx4_en_dev *mdev, u8 port, u8 reset)
 	stats->tx_fifo_errors = 0;
 	stats->tx_heartbeat_errors = 0;
 	stats->tx_window_errors = 0;
+	stats->tx_dropped = be32_to_cpu(mlx4_en_stats->TDROP);
+
+	/* RX stats */
+	priv->pkstats.rx_multicast_packets = stats->multicast;
+	priv->pkstats.rx_broadcast_packets =
+			en_stats_adder(&mlx4_en_stats->RBCAST_prio_0,
+				       &mlx4_en_stats->RBCAST_prio_1,
+				       NUM_PRIORITIES);
+	priv->pkstats.rx_jabbers = be32_to_cpu(mlx4_en_stats->RJBBR);
+	priv->pkstats.rx_in_range_length_error =
+		be64_to_cpu(mlx4_en_stats->RInRangeLengthErr);
+	priv->pkstats.rx_out_range_length_error =
+		be64_to_cpu(mlx4_en_stats->ROutRangeLengthErr);
+
+	/* Tx stats */
+	priv->pkstats.tx_multicast_packets =
+		en_stats_adder(&mlx4_en_stats->TMCAST_prio_0,
+			       &mlx4_en_stats->TMCAST_prio_1,
+			       NUM_PRIORITIES);
+	priv->pkstats.tx_broadcast_packets =
+		en_stats_adder(&mlx4_en_stats->TBCAST_prio_0,
+			       &mlx4_en_stats->TBCAST_prio_1,
+			       NUM_PRIORITIES);
+
+	priv->pkstats.rx_prio[0][0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_0);
+	priv->pkstats.rx_prio[0][1] = be64_to_cpu(mlx4_en_stats->ROCT_prio_0);
+	priv->pkstats.rx_prio[1][0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_1);
+	priv->pkstats.rx_prio[1][1] = be64_to_cpu(mlx4_en_stats->ROCT_prio_1);
+	priv->pkstats.rx_prio[2][0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_2);
+	priv->pkstats.rx_prio[2][1] = be64_to_cpu(mlx4_en_stats->ROCT_prio_2);
+	priv->pkstats.rx_prio[3][0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_3);
+	priv->pkstats.rx_prio[3][1] = be64_to_cpu(mlx4_en_stats->ROCT_prio_3);
+	priv->pkstats.rx_prio[4][0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_4);
+	priv->pkstats.rx_prio[4][1] = be64_to_cpu(mlx4_en_stats->ROCT_prio_4);
+	priv->pkstats.rx_prio[5][0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_5);
+	priv->pkstats.rx_prio[5][1] = be64_to_cpu(mlx4_en_stats->ROCT_prio_5);
+	priv->pkstats.rx_prio[6][0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_6);
+	priv->pkstats.rx_prio[6][1] = be64_to_cpu(mlx4_en_stats->ROCT_prio_6);
+	priv->pkstats.rx_prio[7][0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_7);
+	priv->pkstats.rx_prio[7][1] = be64_to_cpu(mlx4_en_stats->ROCT_prio_7);
+	priv->pkstats.rx_prio[8][0] = be64_to_cpu(mlx4_en_stats->RTOT_novlan);
+	priv->pkstats.rx_prio[8][1] = be64_to_cpu(mlx4_en_stats->ROCT_novlan);
+	priv->pkstats.tx_prio[0][0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_0);
+	priv->pkstats.tx_prio[0][1] = be64_to_cpu(mlx4_en_stats->TOCT_prio_0);
+	priv->pkstats.tx_prio[1][0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_1);
+	priv->pkstats.tx_prio[1][1] = be64_to_cpu(mlx4_en_stats->TOCT_prio_1);
+	priv->pkstats.tx_prio[2][0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_2);
+	priv->pkstats.tx_prio[2][1] = be64_to_cpu(mlx4_en_stats->TOCT_prio_2);
+	priv->pkstats.tx_prio[3][0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_3);
+	priv->pkstats.tx_prio[3][1] = be64_to_cpu(mlx4_en_stats->TOCT_prio_3);
+	priv->pkstats.tx_prio[4][0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_4);
+	priv->pkstats.tx_prio[4][1] = be64_to_cpu(mlx4_en_stats->TOCT_prio_4);
+	priv->pkstats.tx_prio[5][0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_5);
+	priv->pkstats.tx_prio[5][1] = be64_to_cpu(mlx4_en_stats->TOCT_prio_5);
+	priv->pkstats.tx_prio[6][0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_6);
+	priv->pkstats.tx_prio[6][1] = be64_to_cpu(mlx4_en_stats->TOCT_prio_6);
+	priv->pkstats.tx_prio[7][0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_7);
+	priv->pkstats.tx_prio[7][1] = be64_to_cpu(mlx4_en_stats->TOCT_prio_7);
+	priv->pkstats.tx_prio[8][0] = be64_to_cpu(mlx4_en_stats->TTOT_novlan);
+	priv->pkstats.tx_prio[8][1] = be64_to_cpu(mlx4_en_stats->TOCT_novlan);
+
+	spin_unlock_bh(&priv->stats_lock);
+
+	/* 0xffs indicates invalid value */
+	memset(mailbox->buf, 0xff, sizeof(*flowstats) * MLX4_NUM_PRIORITIES);
+
+	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN) {
+		memset(mailbox->buf, 0,
+		       sizeof(*flowstats) * MLX4_NUM_PRIORITIES);
+		err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma,
+				   in_mod | MLX4_DUMP_ETH_STATS_FLOW_CONTROL,
+				   0, MLX4_CMD_DUMP_ETH_STATS,
+				   MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
+		if (err)
+			goto out;
+	}
+
+	flowstats = mailbox->buf;
+
+	spin_lock_bh(&priv->stats_lock);
+
+	for (i = 0; i < MLX4_NUM_PRIORITIES; i++)	{
+		priv->rx_priority_flowstats[i].rx_pause =
+			be64_to_cpu(flowstats[i].rx_pause);
+		priv->rx_priority_flowstats[i].rx_pause_duration =
+			be64_to_cpu(flowstats[i].rx_pause_duration);
+		priv->rx_priority_flowstats[i].rx_pause_transition =
+			be64_to_cpu(flowstats[i].rx_pause_transition);
+		priv->tx_priority_flowstats[i].tx_pause =
+			be64_to_cpu(flowstats[i].tx_pause);
+		priv->tx_priority_flowstats[i].tx_pause_duration =
+			be64_to_cpu(flowstats[i].tx_pause_duration);
+		priv->tx_priority_flowstats[i].tx_pause_transition =
+			be64_to_cpu(flowstats[i].tx_pause_transition);
+	}
+
+	/* if pfc is not in use, all priorities counters have the same value */
+	priv->rx_flowstats.rx_pause =
+		be64_to_cpu(flowstats[0].rx_pause);
+	priv->rx_flowstats.rx_pause_duration =
+		be64_to_cpu(flowstats[0].rx_pause_duration);
+	priv->rx_flowstats.rx_pause_transition =
+		be64_to_cpu(flowstats[0].rx_pause_transition);
+	priv->tx_flowstats.tx_pause =
+		be64_to_cpu(flowstats[0].tx_pause);
+	priv->tx_flowstats.tx_pause_duration =
+		be64_to_cpu(flowstats[0].tx_pause_duration);
+	priv->tx_flowstats.tx_pause_transition =
+		be64_to_cpu(flowstats[0].tx_pause_transition);
 
-	priv->pkstats.broadcast =
-				be64_to_cpu(mlx4_en_stats->RBCAST_prio_0) +
-				be64_to_cpu(mlx4_en_stats->RBCAST_prio_1) +
-				be64_to_cpu(mlx4_en_stats->RBCAST_prio_2) +
-				be64_to_cpu(mlx4_en_stats->RBCAST_prio_3) +
-				be64_to_cpu(mlx4_en_stats->RBCAST_prio_4) +
-				be64_to_cpu(mlx4_en_stats->RBCAST_prio_5) +
-				be64_to_cpu(mlx4_en_stats->RBCAST_prio_6) +
-				be64_to_cpu(mlx4_en_stats->RBCAST_prio_7) +
-				be64_to_cpu(mlx4_en_stats->RBCAST_novlan);
-	priv->pkstats.rx_prio[0] = be64_to_cpu(mlx4_en_stats->RTOT_prio_0);
-	priv->pkstats.rx_prio[1] = be64_to_cpu(mlx4_en_stats->RTOT_prio_1);
-	priv->pkstats.rx_prio[2] = be64_to_cpu(mlx4_en_stats->RTOT_prio_2);
-	priv->pkstats.rx_prio[3] = be64_to_cpu(mlx4_en_stats->RTOT_prio_3);
-	priv->pkstats.rx_prio[4] = be64_to_cpu(mlx4_en_stats->RTOT_prio_4);
-	priv->pkstats.rx_prio[5] = be64_to_cpu(mlx4_en_stats->RTOT_prio_5);
-	priv->pkstats.rx_prio[6] = be64_to_cpu(mlx4_en_stats->RTOT_prio_6);
-	priv->pkstats.rx_prio[7] = be64_to_cpu(mlx4_en_stats->RTOT_prio_7);
-	priv->pkstats.tx_prio[0] = be64_to_cpu(mlx4_en_stats->TTOT_prio_0);
-	priv->pkstats.tx_prio[1] = be64_to_cpu(mlx4_en_stats->TTOT_prio_1);
-	priv->pkstats.tx_prio[2] = be64_to_cpu(mlx4_en_stats->TTOT_prio_2);
-	priv->pkstats.tx_prio[3] = be64_to_cpu(mlx4_en_stats->TTOT_prio_3);
-	priv->pkstats.tx_prio[4] = be64_to_cpu(mlx4_en_stats->TTOT_prio_4);
-	priv->pkstats.tx_prio[5] = be64_to_cpu(mlx4_en_stats->TTOT_prio_5);
-	priv->pkstats.tx_prio[6] = be64_to_cpu(mlx4_en_stats->TTOT_prio_6);
-	priv->pkstats.tx_prio[7] = be64_to_cpu(mlx4_en_stats->TTOT_prio_7);
 	spin_unlock_bh(&priv->stats_lock);
 
 out:
diff --git a/drivers/net/ethernet/mellanox/mlx4/fw.c b/drivers/net/ethernet/mellanox/mlx4/fw.c
index 4a471f5d1b56..209a6171e59b 100644
--- a/drivers/net/ethernet/mellanox/mlx4/fw.c
+++ b/drivers/net/ethernet/mellanox/mlx4/fw.c
@@ -145,7 +145,8 @@ static void dump_dev_cap_flags2(struct mlx4_dev *dev, u64 flags)
 		[20] = "Recoverable error events support",
 		[21] = "Port Remap support",
 		[22] = "QCN support",
-		[23] = "QP rate limiting support"
+		[23] = "QP rate limiting support",
+		[24] = "Ethernet Flow control statistics support"
 	};
 	int i;
 
@@ -672,6 +673,7 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 #define QUERY_DEV_CAP_RSVD_XRC_OFFSET		0x66
 #define QUERY_DEV_CAP_MAX_XRC_OFFSET		0x67
 #define QUERY_DEV_CAP_MAX_COUNTERS_OFFSET	0x68
+#define QUERY_DEV_CAP_PORT_FLOWSTATS_COUNTERS_OFFSET	0x70
 #define QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET	0x70
 #define QUERY_DEV_CAP_FLOW_STEERING_IPOIB_OFFSET	0x74
 #define QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET	0x76
@@ -773,6 +775,9 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_VL_PORT_OFFSET);
 	dev_cap->num_ports = field & 0xf;
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_MSG_SZ_OFFSET);
+	MLX4_GET(field, outbox, QUERY_DEV_CAP_PORT_FLOWSTATS_COUNTERS_OFFSET);
+	if (field & 0x10)
+		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN;
 	dev_cap->max_msg_sz = 1 << (field & 0x1f);
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_FLOW_STEERING_RANGE_EN_OFFSET);
 	if (field & 0x80)
@@ -1088,6 +1093,7 @@ out:
 	return err;
 }
 
+#define DEV_CAP_EXT_2_FLAG_PFC_COUNTERS	(1 << 28)
 #define DEV_CAP_EXT_2_FLAG_VLAN_CONTROL (1 << 26)
 #define DEV_CAP_EXT_2_FLAG_80_VFS	(1 << 21)
 #define DEV_CAP_EXT_2_FLAG_FSM		(1 << 20)
@@ -1177,7 +1183,7 @@ int mlx4_QUERY_DEV_CAP_wrapper(struct mlx4_dev *dev, int slave,
 	/* turn off host side virt features (VST, FSM, etc) for guests */
 	MLX4_GET(field32, outbox->buf, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
 	field32 &= ~(DEV_CAP_EXT_2_FLAG_VLAN_CONTROL | DEV_CAP_EXT_2_FLAG_80_VFS |
-		     DEV_CAP_EXT_2_FLAG_FSM);
+		     DEV_CAP_EXT_2_FLAG_FSM | DEV_CAP_EXT_2_FLAG_PFC_COUNTERS);
 	MLX4_PUT(outbox->buf, field32, QUERY_DEV_CAP_EXT_2_FLAGS_OFFSET);
 
 	/* turn off QCN for guests */
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h b/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
index 8ad241bbcf25..67eeea244eff 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
@@ -55,6 +55,7 @@
 #include <linux/mlx4/cmd.h>
 
 #include "en_port.h"
+#include "mlx4_stats.h"
 
 #define DRV_NAME	"mlx4_en"
 #define DRV_VERSION	"2.2-1"
@@ -171,7 +172,6 @@ enum {
 /* Number of samples to 'average' */
 #define AVG_SIZE			128
 #define AVG_FACTOR			1024
-#define NUM_PERF_STATS			NUM_PERF_COUNTERS
 
 #define INC_PERF_COUNTER(cnt)		(++(cnt))
 #define ADD_PERF_COUNTER(cnt, add)	((cnt) += (add))
@@ -182,7 +182,6 @@ enum {
 
 #else
 
-#define NUM_PERF_STATS			0
 #define INC_PERF_COUNTER(cnt)		do {} while (0)
 #define ADD_PERF_COUNTER(cnt, add)	do {} while (0)
 #define AVG_PERF_COUNTER(cnt, sample)	do {} while (0)
@@ -435,37 +434,6 @@ struct mlx4_en_port_state {
 	u32 flags;
 };
 
-struct mlx4_en_pkt_stats {
-	unsigned long broadcast;
-	unsigned long rx_prio[8];
-	unsigned long tx_prio[8];
-#define NUM_PKT_STATS		17
-};
-
-struct mlx4_en_port_stats {
-	unsigned long tso_packets;
-	unsigned long xmit_more;
-	unsigned long queue_stopped;
-	unsigned long wake_queue;
-	unsigned long tx_timeout;
-	unsigned long rx_alloc_failed;
-	unsigned long rx_chksum_good;
-	unsigned long rx_chksum_none;
-	unsigned long rx_chksum_complete;
-	unsigned long tx_chksum_offload;
-#define NUM_PORT_STATS		10
-};
-
-struct mlx4_en_perf_stats {
-	u32 tx_poll;
-	u64 tx_pktsz_avg;
-	u32 inflight_avg;
-	u16 tx_coal_avg;
-	u16 rx_coal_avg;
-	u32 napi_quota;
-#define NUM_PERF_COUNTERS		6
-};
-
 enum mlx4_en_mclist_act {
 	MCLIST_NONE,
 	MCLIST_REM,
@@ -517,6 +485,11 @@ enum {
 #define MLX4_EN_MAC_HASH_SIZE (1 << BITS_PER_BYTE)
 #define MLX4_EN_MAC_HASH_IDX 5
 
+struct mlx4_en_stats_bitmap {
+	DECLARE_BITMAP(bitmap, NUM_ALL_STATS);
+	struct mutex mutex; /* for mutual access to stats bitmap */
+};
+
 struct mlx4_en_priv {
 	struct mlx4_en_dev *mdev;
 	struct mlx4_en_port_profile *prof;
@@ -592,8 +565,12 @@ struct mlx4_en_priv {
 #endif
 	struct mlx4_en_perf_stats pstats;
 	struct mlx4_en_pkt_stats pkstats;
+	struct mlx4_en_flow_stats_rx rx_priority_flowstats[MLX4_NUM_PRIORITIES];
+	struct mlx4_en_flow_stats_tx tx_priority_flowstats[MLX4_NUM_PRIORITIES];
+	struct mlx4_en_flow_stats_rx rx_flowstats;
+	struct mlx4_en_flow_stats_tx tx_flowstats;
 	struct mlx4_en_port_stats port_stats;
-	u64 stats_bitmap;
+	struct mlx4_en_stats_bitmap stats_bitmap;
 	struct list_head mc_list;
 	struct list_head curr_list;
 	u64 broadcast_id;
@@ -762,6 +739,11 @@ int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
 int mlx4_en_start_port(struct net_device *dev);
 void mlx4_en_stop_port(struct net_device *dev, int detach);
 
+void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
+			      struct mlx4_en_stats_bitmap *stats_bitmap,
+			      u8 rx_ppp, u8 rx_pause,
+			      u8 tx_ppp, u8 tx_pause);
+
 void mlx4_en_free_resources(struct mlx4_en_priv *priv);
 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv);
 
@@ -847,7 +829,10 @@ void mlx4_en_ptp_overflow_check(struct mlx4_en_dev *mdev);
 int mlx4_en_reset_config(struct net_device *dev,
 			 struct hwtstamp_config ts_config,
 			 netdev_features_t new_features);
-
+void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
+				     struct mlx4_en_stats_bitmap *stats_bitmap,
+				     u8 rx_ppp, u8 rx_pause,
+				     u8 tx_ppp, u8 tx_pause);
 int mlx4_en_netdev_event(struct notifier_block *this,
 			 unsigned long event, void *ptr);
 
diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4_stats.h b/drivers/net/ethernet/mellanox/mlx4/mlx4_stats.h
new file mode 100644
index 000000000000..00555832a4ae
--- /dev/null
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4_stats.h
@@ -0,0 +1,107 @@
+#ifndef _MLX4_STATS_
+#define _MLX4_STATS_
+
+#ifdef MLX4_EN_PERF_STAT
+#define NUM_PERF_STATS			NUM_PERF_COUNTERS
+#else
+#define NUM_PERF_STATS			0
+#endif
+
+#define NUM_PRIORITIES	9
+#define NUM_PRIORITY_STATS 2
+
+struct mlx4_en_pkt_stats {
+	unsigned long rx_multicast_packets;
+	unsigned long rx_broadcast_packets;
+	unsigned long rx_jabbers;
+	unsigned long rx_in_range_length_error;
+	unsigned long rx_out_range_length_error;
+	unsigned long tx_multicast_packets;
+	unsigned long tx_broadcast_packets;
+	unsigned long rx_prio[NUM_PRIORITIES][NUM_PRIORITY_STATS];
+	unsigned long tx_prio[NUM_PRIORITIES][NUM_PRIORITY_STATS];
+#define NUM_PKT_STATS		43
+};
+
+struct mlx4_en_port_stats {
+	unsigned long tso_packets;
+	unsigned long xmit_more;
+	unsigned long queue_stopped;
+	unsigned long wake_queue;
+	unsigned long tx_timeout;
+	unsigned long rx_alloc_failed;
+	unsigned long rx_chksum_good;
+	unsigned long rx_chksum_none;
+	unsigned long rx_chksum_complete;
+	unsigned long tx_chksum_offload;
+#define NUM_PORT_STATS		10
+};
+
+struct mlx4_en_perf_stats {
+	u32 tx_poll;
+	u64 tx_pktsz_avg;
+	u32 inflight_avg;
+	u16 tx_coal_avg;
+	u16 rx_coal_avg;
+	u32 napi_quota;
+#define NUM_PERF_COUNTERS		6
+};
+
+#define NUM_MAIN_STATS	21
+
+#define MLX4_NUM_PRIORITIES	8
+
+struct mlx4_en_flow_stats_rx {
+	u64 rx_pause;
+	u64 rx_pause_duration;
+	u64 rx_pause_transition;
+#define NUM_FLOW_STATS_RX	3
+#define NUM_FLOW_PRIORITY_STATS_RX	(NUM_FLOW_STATS_RX * \
+					 MLX4_NUM_PRIORITIES)
+};
+
+struct mlx4_en_flow_stats_tx {
+	u64 tx_pause;
+	u64 tx_pause_duration;
+	u64 tx_pause_transition;
+#define NUM_FLOW_STATS_TX	3
+#define NUM_FLOW_PRIORITY_STATS_TX	(NUM_FLOW_STATS_TX * \
+					 MLX4_NUM_PRIORITIES)
+};
+
+#define NUM_FLOW_STATS (NUM_FLOW_STATS_RX + NUM_FLOW_STATS_TX + \
+			NUM_FLOW_PRIORITY_STATS_TX + \
+			NUM_FLOW_PRIORITY_STATS_RX)
+
+struct mlx4_en_stat_out_flow_control_mbox {
+	/* Total number of PAUSE frames received from the far-end port */
+	__be64 rx_pause;
+	/* Total number of microseconds that far-end port requested to pause
+	* transmission of packets
+	*/
+	__be64 rx_pause_duration;
+	/* Number of received transmission from XOFF state to XON state */
+	__be64 rx_pause_transition;
+	/* Total number of PAUSE frames sent from the far-end port */
+	__be64 tx_pause;
+	/* Total time in microseconds that transmission of packets has been
+	* paused
+	*/
+	__be64 tx_pause_duration;
+	/* Number of transmitter transitions from XOFF state to XON state */
+	__be64 tx_pause_transition;
+	/* Reserverd */
+	__be64 reserved[2];
+};
+
+enum {
+	MLX4_DUMP_ETH_STATS_FLOW_CONTROL = 1 << 12
+};
+
+#define NUM_ALL_STATS	(NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PKT_STATS + \
+			 NUM_FLOW_STATS + NUM_PERF_STATS)
+
+#define MLX4_FIND_NETDEV_STAT(n) (offsetof(struct net_device_stats, n) / \
+				  sizeof(((struct net_device_stats *)0)->n))
+
+#endif
diff --git a/drivers/net/ethernet/mellanox/mlx4/port.c b/drivers/net/ethernet/mellanox/mlx4/port.c
index 9f268f05290a..b97f173ab062 100644
--- a/drivers/net/ethernet/mellanox/mlx4/port.c
+++ b/drivers/net/ethernet/mellanox/mlx4/port.c
@@ -38,17 +38,13 @@
 #include <linux/mlx4/cmd.h>
 
 #include "mlx4.h"
+#include "mlx4_stats.h"
 
 #define MLX4_MAC_VALID		(1ull << 63)
 
 #define MLX4_VLAN_VALID		(1u << 31)
 #define MLX4_VLAN_MASK		0xfff
 
-#define MLX4_STATS_TRAFFIC_COUNTERS_MASK	0xfULL
-#define MLX4_STATS_TRAFFIC_DROPS_MASK		0xc0ULL
-#define MLX4_STATS_ERROR_COUNTERS_MASK		0x1ffc30ULL
-#define MLX4_STATS_PORT_COUNTERS_MASK		0x1fe00000ULL
-
 void mlx4_init_mac_table(struct mlx4_dev *dev, struct mlx4_mac_table *table)
 {
 	int i;
@@ -1184,22 +1180,6 @@ int mlx4_DUMP_ETH_STATS_wrapper(struct mlx4_dev *dev, int slave,
 					  vhcr->in_modifier, outbox);
 }
 
-void mlx4_set_stats_bitmap(struct mlx4_dev *dev, u64 *stats_bitmap)
-{
-	if (!mlx4_is_mfunc(dev)) {
-		*stats_bitmap = 0;
-		return;
-	}
-
-	*stats_bitmap = (MLX4_STATS_TRAFFIC_COUNTERS_MASK |
-			 MLX4_STATS_TRAFFIC_DROPS_MASK |
-			 MLX4_STATS_PORT_COUNTERS_MASK);
-
-	if (mlx4_is_master(dev))
-		*stats_bitmap |= MLX4_STATS_ERROR_COUNTERS_MASK;
-}
-EXPORT_SYMBOL(mlx4_set_stats_bitmap);
-
 int mlx4_get_slave_from_roce_gid(struct mlx4_dev *dev, int port, u8 *gid,
 				 int *slave_id)
 {
diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h
index 4550c67b92e4..ab7ebec943b8 100644
--- a/include/linux/mlx4/device.h
+++ b/include/linux/mlx4/device.h
@@ -205,7 +205,8 @@ enum {
 	MLX4_DEV_CAP_FLAG2_RECOVERABLE_ERROR_EVENT = 1LL << 20,
 	MLX4_DEV_CAP_FLAG2_PORT_REMAP		= 1LL <<  21,
 	MLX4_DEV_CAP_FLAG2_QCN			= 1LL <<  22,
-	MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT	= 1LL <<  23
+	MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT	= 1LL <<  23,
+	MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN         = 1LL <<  24
 };
 
 enum {
@@ -1300,7 +1301,6 @@ int mlx4_register_mac(struct mlx4_dev *dev, u8 port, u64 mac);
 void mlx4_unregister_mac(struct mlx4_dev *dev, u8 port, u64 mac);
 int mlx4_get_base_qpn(struct mlx4_dev *dev, u8 port);
 int __mlx4_replace_mac(struct mlx4_dev *dev, u8 port, int qpn, u64 new_mac);
-void mlx4_set_stats_bitmap(struct mlx4_dev *dev, u64 *stats_bitmap);
 int mlx4_SET_PORT_general(struct mlx4_dev *dev, u8 port, int mtu,
 			  u8 pptx, u8 pfctx, u8 pprx, u8 pfcrx);
 int mlx4_SET_PORT_qpn_calc(struct mlx4_dev *dev, u8 port, u32 base_qpn,