Our patches to mlx5 in dpdk1702

Signed-off-by: Hanoh Haim <hhaim@cisco.com>

6 files changed, 381 insertions, 93 deletions

diff --git a/src/dpdk/drivers/net/mlx5/mlx5.c b/src/dpdk/drivers/net/mlx5/mlx5.c
index d4bd4696..34100a66 100644
--- a/src/dpdk/drivers/net/mlx5/mlx5.c
+++ b/src/dpdk/drivers/net/mlx5/mlx5.c
@@ -181,6 +181,9 @@ mlx5_dev_close(struct rte_eth_dev *dev)
 	}
 	if (priv->reta_idx != NULL)
 		rte_free(priv->reta_idx);
+
+    mlx5_stats_free(dev);
+
 	priv_unlock(priv);
 	memset(priv, 0, sizeof(*priv));
 }
@@ -372,6 +375,15 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 	int idx;
 	int i;
 
+    /* TREX PATCH, lazy ibv_init */
+    static int ibv_was_init=0;
+
+    if (ibv_was_init==0) {
+        ibv_fork_init();
+        ibv_was_init=1;
+    }
+
+
 	(void)pci_drv;
 	assert(pci_drv == &mlx5_driver.pci_drv);
 	/* Get mlx5_dev[] index. */
@@ -533,6 +545,12 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
 		priv->mps = mps; /* Enable MPW by default if supported. */
 		priv->cqe_comp = 1; /* Enable compression by default. */
 		err = mlx5_args(priv, pci_dev->device.devargs);
+
+        /* TREX PATCH */
+        /* set for maximum performance default */
+        priv->txq_inline  =64;
+        priv->txqs_inline =4;
+
 		if (err) {
 			ERROR("failed to process device arguments: %s",
 			      strerror(err));
@@ -787,7 +805,6 @@ rte_mlx5_pmd_init(void)
 	 * using this PMD, which is not supported in forked processes.
 	 */
 	setenv("RDMAV_HUGEPAGES_SAFE", "1", 1);
-	ibv_fork_init();
 	rte_eal_pci_register(&mlx5_driver.pci_drv);
 }
 
diff --git a/src/dpdk/drivers/net/mlx5/mlx5.h b/src/dpdk/drivers/net/mlx5/mlx5.h
index 879da5ef..5970f31a 100644
--- a/src/dpdk/drivers/net/mlx5/mlx5.h
+++ b/src/dpdk/drivers/net/mlx5/mlx5.h
@@ -89,6 +89,34 @@ enum {
 	PCI_DEVICE_ID_MELLANOX_CONNECTX5EXVF = 0x101a,
 };
 
+struct mlx5_stats_priv {
+
+    struct rte_eth_stats m_shadow;
+    uint32_t      n_stats; /* number of counters */
+
+    void    *  et_stats  ;/* point to ethtool counter struct ethtool_stats*/
+
+    /* index into ethtool */
+    uint16_t inx_rx_vport_unicast_bytes;
+    uint16_t inx_rx_vport_multicast_bytes;
+    uint16_t inx_rx_vport_broadcast_bytes;
+    uint16_t inx_rx_vport_unicast_packets;
+    uint16_t inx_rx_vport_multicast_packets;
+    uint16_t inx_rx_vport_broadcast_packets;
+    uint16_t inx_tx_vport_unicast_bytes;
+    uint16_t inx_tx_vport_multicast_bytes;
+    uint16_t inx_tx_vport_broadcast_bytes;
+    uint16_t inx_tx_vport_unicast_packets;
+    uint16_t inx_tx_vport_multicast_packets;
+    uint16_t inx_tx_vport_broadcast_packets;
+    uint16_t inx_rx_wqe_err;
+    uint16_t inx_rx_crc_errors_phy;
+    uint16_t inx_rx_in_range_len_errors_phy;
+    uint16_t inx_rx_symbol_err_phy;
+    uint16_t inx_tx_errors_phy;
+};
+
+
 struct mlx5_xstats_ctrl {
 	/* Number of device stats. */
 	uint16_t stats_n;
@@ -152,6 +180,7 @@ struct priv {
 	uint32_t link_speed_capa; /* Link speed capabilities. */
 	struct mlx5_xstats_ctrl xstats_ctrl; /* Extended stats control. */
 	rte_spinlock_t lock; /* Lock for control functions. */
+    struct mlx5_stats_priv m_stats;
 };
 
 /* Local storage for secondary process data. */
@@ -261,6 +290,8 @@ void mlx5_allmulticast_disable(struct rte_eth_dev *);
 void priv_xstats_init(struct priv *);
 void mlx5_stats_get(struct rte_eth_dev *, struct rte_eth_stats *);
 void mlx5_stats_reset(struct rte_eth_dev *);
+void mlx5_stats_free(struct rte_eth_dev *dev);
+
 int mlx5_xstats_get(struct rte_eth_dev *,
 		    struct rte_eth_xstat *, unsigned int);
 void mlx5_xstats_reset(struct rte_eth_dev *);
diff --git a/src/dpdk/drivers/net/mlx5/mlx5_fdir.c b/src/dpdk/drivers/net/mlx5/mlx5_fdir.c
index f80c58b4..92a923d4 100644
--- a/src/dpdk/drivers/net/mlx5/mlx5_fdir.c
+++ b/src/dpdk/drivers/net/mlx5/mlx5_fdir.c
@@ -37,12 +37,14 @@
 #include <string.h>
 #include <errno.h>
 
+#define TREX_PATCH
+         
 /* Verbs header. */
 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
 #ifdef PEDANTIC
 #pragma GCC diagnostic ignored "-Wpedantic"
 #endif
-#include <infiniband/verbs.h>
+#include <infiniband/verbs_exp.h>
 #ifdef PEDANTIC
 #pragma GCC diagnostic error "-Wpedantic"
 #endif
@@ -69,6 +71,9 @@ struct fdir_flow_desc {
 	uint16_t src_port;
 	uint32_t src_ip[4];
 	uint32_t dst_ip[4];
+    uint8_t  tos;
+    uint8_t  ip_id;
+    uint8_t  proto;
 	uint8_t	mac[6];
 	uint16_t vlan_tag;
 	enum hash_rxq_type type;
@@ -104,6 +109,7 @@ fdir_filter_to_flow_desc(const struct rte_eth_fdir_filter *fdir_filter,
 	/* Set VLAN ID. */
 	desc->vlan_tag = fdir_filter->input.flow_ext.vlan_tci;
 
+#ifndef TREX_PATCH
 	/* Set MAC address. */
 	if (mode == RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
 		rte_memcpy(desc->mac,
@@ -113,6 +119,14 @@ fdir_filter_to_flow_desc(const struct rte_eth_fdir_filter *fdir_filter,
 		desc->type = HASH_RXQ_ETH;
 		return;
 	}
+#else
+    if (fdir_filter->input.flow.ip4_flow.ip_id == 2) {
+        desc->type = HASH_RXQ_ETH;
+        desc->ip_id = fdir_filter->input.flow.ip4_flow.ip_id;
+        return;
+    }
+#endif
+
 
 	/* Set mode */
 	switch (fdir_filter->input.flow_type) {
@@ -147,6 +161,9 @@ fdir_filter_to_flow_desc(const struct rte_eth_fdir_filter *fdir_filter,
 	case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
 		desc->src_ip[0] = fdir_filter->input.flow.ip4_flow.src_ip;
 		desc->dst_ip[0] = fdir_filter->input.flow.ip4_flow.dst_ip;
+        desc->tos       = fdir_filter->input.flow.ip4_flow.ttl; /* TTL is mapped to TOS TREX_PATCH */
+        desc->ip_id     = fdir_filter->input.flow.ip4_flow.ip_id;
+        desc->proto     = fdir_filter->input.flow.ip4_flow.proto;
 		break;
 	case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
 	case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
@@ -160,6 +177,9 @@ fdir_filter_to_flow_desc(const struct rte_eth_fdir_filter *fdir_filter,
 		rte_memcpy(desc->dst_ip,
 			   fdir_filter->input.flow.ipv6_flow.dst_ip,
 			   sizeof(desc->dst_ip));
+        desc->tos       = (uint8_t)fdir_filter->input.flow.ipv6_flow.hop_limits;  /* TTL is mapped to TOS - TREX_PATCH */
+        desc->ip_id     = (uint8_t)fdir_filter->input.flow.ipv6_flow.flow_label;
+        desc->proto     = fdir_filter->input.flow.ipv6_flow.proto;
 		break;
 	default:
 		break;
@@ -200,6 +220,11 @@ priv_fdir_overlap(const struct priv *priv,
 	    ((desc1->dst_port & mask->dst_port_mask) !=
 	     (desc2->dst_port & mask->dst_port_mask)))
 		return 0;
+    if  ( (desc1->tos    != desc2->tos)  ||
+          (desc1->ip_id  != desc2->ip_id) ||
+          (desc1->proto  != desc2->proto) ) 
+        return 0;
+
 	switch (desc1->type) {
 	case HASH_RXQ_IPV4:
 	case HASH_RXQ_UDPV4:
@@ -254,8 +279,8 @@ priv_fdir_flow_add(struct priv *priv,
 	struct ibv_exp_flow_attr *attr = &data->attr;
 	uintptr_t spec_offset = (uintptr_t)&data->spec;
 	struct ibv_exp_flow_spec_eth *spec_eth;
-	struct ibv_exp_flow_spec_ipv4 *spec_ipv4;
-	struct ibv_exp_flow_spec_ipv6 *spec_ipv6;
+	struct ibv_exp_flow_spec_ipv4_ext *spec_ipv4;
+	struct ibv_exp_flow_spec_ipv6_ext *spec_ipv6;
 	struct ibv_exp_flow_spec_tcp_udp *spec_tcp_udp;
 	struct mlx5_fdir_filter *iter_fdir_filter;
 	unsigned int i;
@@ -267,8 +292,10 @@ priv_fdir_flow_add(struct priv *priv,
 		    (iter_fdir_filter->flow != NULL) &&
 		    (priv_fdir_overlap(priv,
 				       &mlx5_fdir_filter->desc,
-				       &iter_fdir_filter->desc)))
-			return EEXIST;
+				       &iter_fdir_filter->desc))){
+            ERROR("overlap rules, please check your rules");
+            return EEXIST;
+        }
 
 	/*
 	 * No padding must be inserted by the compiler between attr and spec.
@@ -291,6 +318,8 @@ priv_fdir_flow_add(struct priv *priv,
 	/* Update priority */
 	attr->priority = 2;
 
+#ifndef TREX_PATCH
+
 	if (fdir_mode == RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
 		/* MAC Address */
 		for (i = 0; i != RTE_DIM(spec_eth->mask.dst_mac); ++i) {
@@ -300,6 +329,15 @@ priv_fdir_flow_add(struct priv *priv,
 		}
 		goto create_flow;
 	}
+#else
+    // empty mask means "match everything". This rule will match all packets, no matter what is the ether type
+    if (desc->ip_id == 2) {
+        spec_eth->val.ether_type = 0x0806;
+        spec_eth->mask.ether_type = 0x0000;
+        goto create_flow;
+    }
+#endif    
+
 
 	switch (desc->type) {
 	case HASH_RXQ_IPV4:
@@ -308,10 +346,10 @@ priv_fdir_flow_add(struct priv *priv,
 		spec_offset += spec_eth->size;
 
 		/* Set IP spec */
-		spec_ipv4 = (struct ibv_exp_flow_spec_ipv4 *)spec_offset;
+		spec_ipv4 = (struct ibv_exp_flow_spec_ipv4_ext *)spec_offset;
 
 		/* The second specification must be IP. */
-		assert(spec_ipv4->type == IBV_EXP_FLOW_SPEC_IPV4);
+		assert(spec_ipv4->type == IBV_EXP_FLOW_SPEC_IPV4_EXT);
 		assert(spec_ipv4->size == sizeof(*spec_ipv4));
 
 		spec_ipv4->val.src_ip =
@@ -324,6 +362,21 @@ priv_fdir_flow_add(struct priv *priv,
 		/* Update priority */
 		attr->priority = 1;
 
+        spec_ipv4->val.proto  = desc->proto & mask->ipv4_mask.proto;
+        spec_ipv4->mask.proto = mask->ipv4_mask.proto;
+
+#ifdef TREX_PATCH
+        /* TOS */
+        if (desc->ip_id == 1) {
+            spec_ipv4->mask.tos = 0x1;
+            spec_ipv4->val.tos = 0x1;
+        } else {
+            spec_ipv4->mask.tos = 0x0;
+            spec_ipv4->val.tos = 0x0;
+        }
+#endif
+
+
 		if (desc->type == HASH_RXQ_IPV4)
 			goto create_flow;
 
@@ -335,10 +388,10 @@ priv_fdir_flow_add(struct priv *priv,
 		spec_offset += spec_eth->size;
 
 		/* Set IP spec */
-		spec_ipv6 = (struct ibv_exp_flow_spec_ipv6 *)spec_offset;
+		spec_ipv6 = (struct ibv_exp_flow_spec_ipv6_ext *)spec_offset;
 
 		/* The second specification must be IP. */
-		assert(spec_ipv6->type == IBV_EXP_FLOW_SPEC_IPV6);
+		assert(spec_ipv6->type == IBV_EXP_FLOW_SPEC_IPV6_EXT);
 		assert(spec_ipv6->size == sizeof(*spec_ipv6));
 
 		for (i = 0; i != RTE_DIM(desc->src_ip); ++i) {
@@ -354,6 +407,20 @@ priv_fdir_flow_add(struct priv *priv,
 			   mask->ipv6_mask.dst_ip,
 			   sizeof(spec_ipv6->mask.dst_ip));
 
+        spec_ipv6->val.next_hdr  = desc->proto & mask->ipv6_mask.proto;
+        spec_ipv6->mask.next_hdr = mask->ipv6_mask.proto;
+
+#ifdef TREX_PATCH
+        /* TOS */
+        if (desc->ip_id == 1) {
+            spec_ipv6->mask.traffic_class = 0x1;
+            spec_ipv6->val.traffic_class = 0x1;
+        } else {
+            spec_ipv6->mask.traffic_class = 0;
+            spec_ipv6->val.traffic_class = 0;
+        }
+#endif
+
 		/* Update priority */
 		attr->priority = 1;
 
@@ -831,8 +898,10 @@ priv_fdir_filter_add(struct priv *priv,
 	/* Duplicate filters are currently unsupported. */
 	mlx5_fdir_filter = priv_find_filter_in_list(priv, fdir_filter);
 	if (mlx5_fdir_filter != NULL) {
+#ifndef TREX_PATCH
 		ERROR("filter already exists");
-		return EINVAL;
+#endif
+		return EEXIST;
 	}
 
 	/* Create new flow director filter. */
@@ -957,9 +1026,11 @@ priv_fdir_filter_delete(struct priv *priv,
 		return 0;
 	}
 
+#ifndef TREX_PATCH
 	ERROR("%p: flow director delete failed, cannot find filter",
 	      (void *)priv);
-	return EINVAL;
+#endif
+	 return ENOENT;
 }
 
 /**
diff --git a/src/dpdk/drivers/net/mlx5/mlx5_rxq.c b/src/dpdk/drivers/net/mlx5/mlx5_rxq.c
index 28e93d3e..c5746fa0 100644
--- a/src/dpdk/drivers/net/mlx5/mlx5_rxq.c
+++ b/src/dpdk/drivers/net/mlx5/mlx5_rxq.c
@@ -102,7 +102,7 @@ const struct hash_rxq_init hash_rxq_init[] = {
 				ETH_RSS_FRAG_IPV4),
 		.flow_priority = 1,
 		.flow_spec.ipv4 = {
-			.type = IBV_EXP_FLOW_SPEC_IPV4,
+			.type = IBV_EXP_FLOW_SPEC_IPV4_EXT,
 			.size = sizeof(hash_rxq_init[0].flow_spec.ipv4),
 		},
 		.underlayer = &hash_rxq_init[HASH_RXQ_ETH],
@@ -140,7 +140,7 @@ const struct hash_rxq_init hash_rxq_init[] = {
 				ETH_RSS_FRAG_IPV6),
 		.flow_priority = 1,
 		.flow_spec.ipv6 = {
-			.type = IBV_EXP_FLOW_SPEC_IPV6,
+			.type = IBV_EXP_FLOW_SPEC_IPV6_EXT,
 			.size = sizeof(hash_rxq_init[0].flow_spec.ipv6),
 		},
 		.underlayer = &hash_rxq_init[HASH_RXQ_ETH],
diff --git a/src/dpdk/drivers/net/mlx5/mlx5_rxtx.h b/src/dpdk/drivers/net/mlx5/mlx5_rxtx.h
index 41a34d7f..5cf6eaca 100644
--- a/src/dpdk/drivers/net/mlx5/mlx5_rxtx.h
+++ b/src/dpdk/drivers/net/mlx5/mlx5_rxtx.h
@@ -179,8 +179,8 @@ struct hash_rxq_init {
 			uint16_t size;
 		} hdr;
 		struct ibv_exp_flow_spec_tcp_udp tcp_udp;
-		struct ibv_exp_flow_spec_ipv4 ipv4;
-		struct ibv_exp_flow_spec_ipv6 ipv6;
+		struct ibv_exp_flow_spec_ipv4_ext ipv4;
+		struct ibv_exp_flow_spec_ipv6_ext ipv6;
 		struct ibv_exp_flow_spec_eth eth;
 	} flow_spec; /* Flow specification template. */
 	const struct hash_rxq_init *underlayer; /* Pointer to underlayer. */
diff --git a/src/dpdk/drivers/net/mlx5/mlx5_stats.c b/src/dpdk/drivers/net/mlx5/mlx5_stats.c
index 20c957e8..99f2d2d2 100644
--- a/src/dpdk/drivers/net/mlx5/mlx5_stats.c
+++ b/src/dpdk/drivers/net/mlx5/mlx5_stats.c
@@ -246,6 +246,223 @@ free:
 	rte_free(strings);
 }
 
+
+
+static void
+mlx5_stats_read_hw(struct rte_eth_dev *dev,
+                struct rte_eth_stats *stats){
+    struct priv *priv = mlx5_get_priv(dev);
+    struct mlx5_stats_priv * lps = &priv->m_stats;
+    unsigned int i;
+
+    struct rte_eth_stats tmp = {0};
+    struct ethtool_stats    *et_stats   = (struct ethtool_stats    *)lps->et_stats;
+    struct ifreq ifr;
+
+    et_stats->cmd = ETHTOOL_GSTATS;
+    et_stats->n_stats = lps->n_stats;
+
+    ifr.ifr_data = (caddr_t) et_stats;
+
+    if (priv_ifreq(priv, SIOCETHTOOL, &ifr) != 0) { 
+        WARN("unable to get statistic values for mlnx5 "); 
+    }
+
+    tmp.ibytes += et_stats->data[lps->inx_rx_vport_unicast_bytes] +
+                  et_stats->data[lps->inx_rx_vport_multicast_bytes] +
+                  et_stats->data[lps->inx_rx_vport_broadcast_bytes];
+
+    tmp.ipackets += et_stats->data[lps->inx_rx_vport_unicast_packets] +
+                et_stats->data[lps->inx_rx_vport_multicast_packets] +
+                et_stats->data[lps->inx_rx_vport_broadcast_packets];
+
+    tmp.ierrors += 	(et_stats->data[lps->inx_rx_wqe_err] +
+                    et_stats->data[lps->inx_rx_crc_errors_phy] +
+                    et_stats->data[lps->inx_rx_in_range_len_errors_phy] +
+                    et_stats->data[lps->inx_rx_symbol_err_phy]);
+
+    tmp.obytes += et_stats->data[lps->inx_tx_vport_unicast_bytes] +
+                  et_stats->data[lps->inx_tx_vport_multicast_bytes] +
+                  et_stats->data[lps->inx_tx_vport_broadcast_bytes];
+
+    tmp.opackets += (et_stats->data[lps->inx_tx_vport_unicast_packets] +
+                     et_stats->data[lps->inx_tx_vport_multicast_packets] +
+                     et_stats->data[lps->inx_tx_vport_broadcast_packets]);
+
+    tmp.oerrors += et_stats->data[lps->inx_tx_errors_phy];
+
+    /* SW Rx */
+    for (i = 0; (i != priv->rxqs_n); ++i) {
+        struct rxq *rxq = (*priv->rxqs)[i];
+        if (rxq) {
+            tmp.imissed += rxq->stats.idropped;
+            tmp.rx_nombuf += rxq->stats.rx_nombuf;
+        }
+    }
+
+    /*SW Tx */
+    for (i = 0; (i != priv->txqs_n); ++i) {
+        struct txq *txq = (*priv->txqs)[i];
+        if (txq) {
+            tmp.oerrors += txq->stats.odropped;
+        }
+    }
+
+    *stats =tmp;
+}
+
+void
+mlx5_stats_free(struct rte_eth_dev *dev)
+{
+    struct priv *priv = mlx5_get_priv(dev);
+    struct mlx5_stats_priv * lps = &priv->m_stats;
+
+    if ( lps->et_stats ){
+        free(lps->et_stats);
+        lps->et_stats=0;
+    }
+}
+
+
+static void
+mlx5_stats_init(struct rte_eth_dev *dev)
+{
+    struct priv *priv = mlx5_get_priv(dev);
+    struct mlx5_stats_priv * lps = &priv->m_stats;
+    struct rte_eth_stats tmp = {0};
+
+    unsigned int i;
+    unsigned int idx;
+    char ifname[IF_NAMESIZE];
+    struct ifreq ifr;
+
+    struct ethtool_stats    *et_stats   = NULL;
+    struct ethtool_drvinfo drvinfo;
+    struct ethtool_gstrings *strings = NULL;
+    unsigned int n_stats, sz_str, sz_stats;
+
+    if (priv_get_ifname(priv, &ifname)) {
+            WARN("unable to get interface name");
+            return;
+    }
+    /* How many statistics are available ? */
+    drvinfo.cmd = ETHTOOL_GDRVINFO;
+    ifr.ifr_data = (caddr_t) &drvinfo;
+    if (priv_ifreq(priv, SIOCETHTOOL, &ifr) != 0) {
+            WARN("unable to get driver info for %s", ifname);
+            return;
+    }
+
+    n_stats = drvinfo.n_stats;
+    if (n_stats < 1) {
+            WARN("no statistics available for %s", ifname);
+            return;
+    }
+    lps->n_stats = n_stats;
+
+    /* Allocate memory to grab stat names and values */ 
+    sz_str = n_stats * ETH_GSTRING_LEN; 
+    sz_stats = n_stats * sizeof(uint64_t); 
+    strings = calloc(1, sz_str + sizeof(struct ethtool_gstrings)); 
+    if (!strings) { 
+        WARN("unable to allocate memory for strings"); 
+        return;
+    } 
+
+    et_stats = calloc(1, sz_stats + sizeof(struct ethtool_stats)); 
+    if (!et_stats) { 
+        free(strings);
+        WARN("unable to allocate memory for stats"); 
+    } 
+
+    strings->cmd = ETHTOOL_GSTRINGS; 
+    strings->string_set = ETH_SS_STATS; 
+    strings->len = n_stats; 
+    ifr.ifr_data = (caddr_t) strings; 
+    if (priv_ifreq(priv, SIOCETHTOOL, &ifr) != 0) { 
+        WARN("unable to get statistic names for %s", ifname); 
+        free(strings);
+        free(et_stats);
+        return;
+    } 
+
+    for (i = 0; (i != n_stats); ++i) {
+
+            const char * curr_string = (const char*) &(strings->data[i * ETH_GSTRING_LEN]);
+
+            if (!strcmp("rx_vport_unicast_bytes", curr_string)) lps->inx_rx_vport_unicast_bytes = i;
+            if (!strcmp("rx_vport_multicast_bytes", curr_string)) lps->inx_rx_vport_multicast_bytes = i;
+            if (!strcmp("rx_vport_broadcast_bytes", curr_string)) lps->inx_rx_vport_broadcast_bytes = i;
+
+            if (!strcmp("rx_vport_unicast_packets", curr_string)) lps->inx_rx_vport_unicast_packets = i;
+            if (!strcmp("rx_vport_multicast_packets", curr_string)) lps->inx_rx_vport_multicast_packets = i;
+            if (!strcmp("rx_vport_broadcast_packets", curr_string)) lps->inx_rx_vport_broadcast_packets = i;
+
+            if (!strcmp("tx_vport_unicast_bytes", curr_string)) lps->inx_tx_vport_unicast_bytes = i;
+            if (!strcmp("tx_vport_multicast_bytes", curr_string)) lps->inx_tx_vport_multicast_bytes = i;
+            if (!strcmp("tx_vport_broadcast_bytes", curr_string)) lps->inx_tx_vport_broadcast_bytes = i;
+
+            if (!strcmp("tx_vport_unicast_packets", curr_string)) lps->inx_tx_vport_unicast_packets = i;
+            if (!strcmp("tx_vport_multicast_packets", curr_string)) lps->inx_tx_vport_multicast_packets = i;
+            if (!strcmp("tx_vport_broadcast_packets", curr_string)) lps->inx_tx_vport_broadcast_packets = i;
+
+            if (!strcmp("rx_wqe_err", curr_string)) lps->inx_rx_wqe_err = i;
+            if (!strcmp("rx_crc_errors_phy", curr_string)) lps->inx_rx_crc_errors_phy = i;
+            if (!strcmp("rx_in_range_len_errors_phy", curr_string)) lps->inx_rx_in_range_len_errors_phy = i;
+            if (!strcmp("rx_symbol_err_phy", curr_string)) lps->inx_rx_symbol_err_phy = i;
+
+            if (!strcmp("tx_errors_phy", curr_string)) lps->inx_tx_errors_phy = i;
+    }
+
+    lps->et_stats =(void *)et_stats;
+
+    if (!lps->inx_rx_vport_unicast_bytes ||
+    !lps->inx_rx_vport_multicast_bytes ||
+    !lps->inx_rx_vport_broadcast_bytes || 
+    !lps->inx_rx_vport_unicast_packets ||
+    !lps->inx_rx_vport_multicast_packets ||
+    !lps->inx_rx_vport_broadcast_packets ||
+    !lps->inx_tx_vport_unicast_bytes || 
+    !lps->inx_tx_vport_multicast_bytes ||
+    !lps->inx_tx_vport_broadcast_bytes ||
+    !lps->inx_tx_vport_unicast_packets ||
+    !lps->inx_tx_vport_multicast_packets ||
+    !lps->inx_tx_vport_broadcast_packets ||
+    !lps->inx_rx_wqe_err ||
+    !lps->inx_rx_crc_errors_phy ||
+    !lps->inx_rx_in_range_len_errors_phy) {
+        WARN("Counters are not recognized %s", ifname);
+        return;
+    }
+
+    mlx5_stats_read_hw(dev,&tmp);
+
+    /* copy yo shadow at first time */
+    lps->m_shadow = tmp;
+
+    free(strings);
+}
+
+
+static void
+mlx5_stats_diff(struct rte_eth_stats *a,
+                struct rte_eth_stats *b,
+                struct rte_eth_stats *c){
+    #define MLX5_DIFF(cnt) { a->cnt = (b->cnt - c->cnt);  }
+
+    MLX5_DIFF(ipackets);
+    MLX5_DIFF(opackets); 
+    MLX5_DIFF(ibytes); 
+    MLX5_DIFF(obytes);
+    MLX5_DIFF(imissed);
+
+    MLX5_DIFF(ierrors); 
+    MLX5_DIFF(oerrors); 
+    MLX5_DIFF(rx_nombuf);
+}
+
+
+
 /**
  * Get device extended statistics.
  *
@@ -295,68 +512,25 @@ priv_xstats_reset(struct priv *priv)
 		xstats_ctrl->base[i] = counters[i];
 }
 
-/**
- * DPDK callback to get device statistics.
- *
- * @param dev
- *   Pointer to Ethernet device structure.
- * @param[out] stats
- *   Stats structure output buffer.
- */
 void
 mlx5_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 {
 	struct priv *priv = mlx5_get_priv(dev);
-	struct rte_eth_stats tmp = {0};
-	unsigned int i;
-	unsigned int idx;
 
-	priv_lock(priv);
-	/* Add software counters. */
-	for (i = 0; (i != priv->rxqs_n); ++i) {
-		struct rxq *rxq = (*priv->rxqs)[i];
-
-		if (rxq == NULL)
-			continue;
-		idx = rxq->stats.idx;
-		if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
-#ifdef MLX5_PMD_SOFT_COUNTERS
-			tmp.q_ipackets[idx] += rxq->stats.ipackets;
-			tmp.q_ibytes[idx] += rxq->stats.ibytes;
-#endif
-			tmp.q_errors[idx] += (rxq->stats.idropped +
-					      rxq->stats.rx_nombuf);
-		}
-#ifdef MLX5_PMD_SOFT_COUNTERS
-		tmp.ipackets += rxq->stats.ipackets;
-		tmp.ibytes += rxq->stats.ibytes;
-#endif
-		tmp.ierrors += rxq->stats.idropped;
-		tmp.rx_nombuf += rxq->stats.rx_nombuf;
-	}
-	for (i = 0; (i != priv->txqs_n); ++i) {
-		struct txq *txq = (*priv->txqs)[i];
-
-		if (txq == NULL)
-			continue;
-		idx = txq->stats.idx;
-		if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
-#ifdef MLX5_PMD_SOFT_COUNTERS
-			tmp.q_opackets[idx] += txq->stats.opackets;
-			tmp.q_obytes[idx] += txq->stats.obytes;
-#endif
-			tmp.q_errors[idx] += txq->stats.odropped;
-		}
-#ifdef MLX5_PMD_SOFT_COUNTERS
-		tmp.opackets += txq->stats.opackets;
-		tmp.obytes += txq->stats.obytes;
-#endif
-		tmp.oerrors += txq->stats.odropped;
-	}
-#ifndef MLX5_PMD_SOFT_COUNTERS
-	/* FIXME: retrieve and add hardware counters. */
-#endif
-	*stats = tmp;
+    struct mlx5_stats_priv * lps = &priv->m_stats;
+    priv_lock(priv);
+
+    if (lps->et_stats == NULL) {
+        mlx5_stats_init(dev);
+    }
+    struct rte_eth_stats tmp = {0};
+
+    mlx5_stats_read_hw(dev,&tmp);
+
+    mlx5_stats_diff(stats,
+                    &tmp,
+                    &lps->m_shadow);
+
 	priv_unlock(priv);
 }
 
@@ -370,30 +544,25 @@ void
 mlx5_stats_reset(struct rte_eth_dev *dev)
 {
 	struct priv *priv = dev->data->dev_private;
-	unsigned int i;
-	unsigned int idx;
+    struct mlx5_stats_priv * lps = &priv->m_stats;
+
+    priv_lock(priv);
+
+    if (lps->et_stats == NULL) {
+        mlx5_stats_init(dev);
+    }
+    struct rte_eth_stats tmp = {0};
+
+
+    mlx5_stats_read_hw(dev,&tmp);
+
+    /* copy to shadow */
+    lps->m_shadow = tmp;
 
-	priv_lock(priv);
-	for (i = 0; (i != priv->rxqs_n); ++i) {
-		if ((*priv->rxqs)[i] == NULL)
-			continue;
-		idx = (*priv->rxqs)[i]->stats.idx;
-		(*priv->rxqs)[i]->stats =
-			(struct mlx5_rxq_stats){ .idx = idx };
-	}
-	for (i = 0; (i != priv->txqs_n); ++i) {
-		if ((*priv->txqs)[i] == NULL)
-			continue;
-		idx = (*priv->txqs)[i]->stats.idx;
-		(*priv->txqs)[i]->stats =
-			(struct mlx5_txq_stats){ .idx = idx };
-	}
-#ifndef MLX5_PMD_SOFT_COUNTERS
-	/* FIXME: reset hardware counters. */
-#endif
 	priv_unlock(priv);
 }
 
+
 /**
  * DPDK callback to get extended device statistics.
  *