diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x.h b/drivers/net/ethernet/broadcom/bnx2x/bnx2x.h
index 7d184fbffaf6..d60b5f073b6f 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x.h
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x.h
@@ -540,6 +540,7 @@ struct bnx2x_fastpath {
 	struct ustorm_per_queue_stats old_uclient;
 	struct xstorm_per_queue_stats old_xclient;
 	struct bnx2x_eth_q_stats eth_q_stats;
+	struct bnx2x_eth_q_stats_old eth_q_stats_old;
 
 	/* The size is calculated using the following:
 	     sizeof name field from netdev structure +
@@ -1046,7 +1047,6 @@ struct bnx2x_slowpath {
 	struct nig_stats		nig_stats;
 	struct host_port_stats		port_stats;
 	struct host_func_stats		func_stats;
-	struct host_func_stats		func_stats_base;
 
 	u32				wb_comp;
 	u32				wb_data[4];
@@ -1462,6 +1462,10 @@ struct bnx2x {
 
 	u16			stats_counter;
 	struct bnx2x_eth_stats	eth_stats;
+	struct bnx2x_eth_stats_old	eth_stats_old;
+	struct bnx2x_net_stats_old	net_stats_old;
+	struct bnx2x_fw_port_stats_old	fw_stats_old;
+	bool			stats_init;
 
 	struct z_stream_s	*strm;
 	void			*gunzip_buf;
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
index 518ec5c6872d..7f6a1b1be74d 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
@@ -31,46 +31,6 @@
 
 
 
-/**
- * bnx2x_bz_fp - zero content of the fastpath structure.
- *
- * @bp:		driver handle
- * @index:	fastpath index to be zeroed
- *
- * Makes sure the contents of the bp->fp[index].napi is kept
- * intact.
- */
-static inline void bnx2x_bz_fp(struct bnx2x *bp, int index)
-{
-	struct bnx2x_fastpath *fp = &bp->fp[index];
-	struct napi_struct orig_napi = fp->napi;
-	/* bzero bnx2x_fastpath contents */
-	memset(fp, 0, sizeof(*fp));
-
-	/* Restore the NAPI object as it has been already initialized */
-	fp->napi = orig_napi;
-
-	fp->bp = bp;
-	fp->index = index;
-	if (IS_ETH_FP(fp))
-		fp->max_cos = bp->max_cos;
-	else
-		/* Special queues support only one CoS */
-		fp->max_cos = 1;
-
-	/*
-	 * set the tpa flag for each queue. The tpa flag determines the queue
-	 * minimal size so it must be set prior to queue memory allocation
-	 */
-	fp->disable_tpa = ((bp->flags & TPA_ENABLE_FLAG) == 0);
-
-#ifdef BCM_CNIC
-	/* We don't want TPA on an FCoE L2 ring */
-	if (IS_FCOE_FP(fp))
-		fp->disable_tpa = 1;
-#endif
-}
-
 /**
  * bnx2x_move_fp - move content of the fastpath structure.
  *
@@ -2084,6 +2044,7 @@ int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
 	bnx2x_drv_pulse(bp);
 
 	bnx2x_stats_handle(bp, STATS_EVENT_STOP);
+	bnx2x_save_statistics(bp);
 
 	/* Cleanup the chip if needed */
 	if (unload_mode != UNLOAD_RECOVERY)
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h
index 67e97b709edd..f978c6a1aef1 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h
@@ -1491,6 +1491,79 @@ static inline u16 bnx2x_extract_max_cfg(struct bnx2x *bp, u32 mf_cfg)
 	return max_cfg;
 }
 
+/**
+ * bnx2x_bz_fp - zero content of the fastpath structure.
+ *
+ * @bp:		driver handle
+ * @index:	fastpath index to be zeroed
+ *
+ * Makes sure the contents of the bp->fp[index].napi is kept
+ * intact.
+ */
+static inline void bnx2x_bz_fp(struct bnx2x *bp, int index)
+{
+	struct bnx2x_fastpath *fp = &bp->fp[index];
+	struct napi_struct orig_napi = fp->napi;
+	/* bzero bnx2x_fastpath contents */
+	if (bp->stats_init)
+		memset(fp, 0, sizeof(*fp));
+	else {
+		/* Keep Queue statistics */
+		struct bnx2x_eth_q_stats *tmp_eth_q_stats;
+		struct bnx2x_eth_q_stats_old *tmp_eth_q_stats_old;
+
+		tmp_eth_q_stats = kzalloc(sizeof(struct bnx2x_eth_q_stats),
+					  GFP_KERNEL);
+		if (tmp_eth_q_stats)
+			memcpy(tmp_eth_q_stats, &fp->eth_q_stats,
+			       sizeof(struct bnx2x_eth_q_stats));
+
+		tmp_eth_q_stats_old =
+			kzalloc(sizeof(struct bnx2x_eth_q_stats_old),
+				GFP_KERNEL);
+		if (tmp_eth_q_stats_old)
+			memcpy(tmp_eth_q_stats_old, &fp->eth_q_stats_old,
+			       sizeof(struct bnx2x_eth_q_stats_old));
+
+		memset(fp, 0, sizeof(*fp));
+
+		if (tmp_eth_q_stats) {
+			memcpy(&fp->eth_q_stats, tmp_eth_q_stats,
+				   sizeof(struct bnx2x_eth_q_stats));
+			kfree(tmp_eth_q_stats);
+		}
+
+		if (tmp_eth_q_stats_old) {
+			memcpy(&fp->eth_q_stats_old, tmp_eth_q_stats_old,
+			       sizeof(struct bnx2x_eth_q_stats_old));
+			kfree(tmp_eth_q_stats_old);
+		}
+
+	}
+
+	/* Restore the NAPI object as it has been already initialized */
+	fp->napi = orig_napi;
+
+	fp->bp = bp;
+	fp->index = index;
+	if (IS_ETH_FP(fp))
+		fp->max_cos = bp->max_cos;
+	else
+		/* Special queues support only one CoS */
+		fp->max_cos = 1;
+
+	/*
+	 * set the tpa flag for each queue. The tpa flag determines the queue
+	 * minimal size so it must be set prior to queue memory allocation
+	 */
+	fp->disable_tpa = (bp->flags & TPA_ENABLE_FLAG) == 0;
+#ifdef BCM_CNIC
+	/* We don't want TPA on an FCoE L2 ring */
+	if (IS_FCOE_FP(fp))
+		fp->disable_tpa = 1;
+#endif
+}
+
 /**
  * bnx2x_get_iscsi_info - update iSCSI params according to licensing info.
  *
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
index ff19c3cf4409..e0d5f15784fd 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c
@@ -10235,6 +10235,8 @@ static int bnx2x_open(struct net_device *dev)
 	int other_engine = BP_PATH(bp) ? 0 : 1;
 	bool other_load_status, load_status;
 
+	bp->stats_init = true;
+
 	netif_carrier_off(dev);
 
 	bnx2x_set_power_state(bp, PCI_D0);
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_stats.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_stats.c
index 7b9b304b9107..abd310d71512 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_stats.c
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_stats.c
@@ -161,7 +161,7 @@ static void bnx2x_stats_pmf_update(struct bnx2x *bp)
 	u32 *stats_comp = bnx2x_sp(bp, stats_comp);
 
 	/* sanity */
-	if (!IS_MF(bp) || !bp->port.pmf || !bp->port.port_stx) {
+	if (!bp->port.pmf || !bp->port.port_stx) {
 		BNX2X_ERR("BUG!\n");
 		return;
 	}
@@ -638,31 +638,30 @@ static void bnx2x_mstat_stats_update(struct bnx2x *bp)
 			tx_stat_dot3statsinternalmactransmiterrors);
 	ADD_STAT64(stats_tx.tx_gtufl, tx_stat_mac_ufl);
 
-	ADD_64(estats->etherstatspkts1024octetsto1522octets_hi,
-	       new->stats_tx.tx_gt1518_hi,
-	       estats->etherstatspkts1024octetsto1522octets_lo,
-	       new->stats_tx.tx_gt1518_lo);
+	estats->etherstatspkts1024octetsto1522octets_hi =
+	    pstats->mac_stx[1].tx_stat_etherstatspkts1024octetsto1522octets_hi;
+	estats->etherstatspkts1024octetsto1522octets_lo =
+	    pstats->mac_stx[1].tx_stat_etherstatspkts1024octetsto1522octets_lo;
+
+	estats->etherstatspktsover1522octets_hi =
+	    pstats->mac_stx[1].tx_stat_mac_2047_hi;
+	estats->etherstatspktsover1522octets_lo =
+	    pstats->mac_stx[1].tx_stat_mac_2047_lo;
 
 	ADD_64(estats->etherstatspktsover1522octets_hi,
-	       new->stats_tx.tx_gt2047_hi,
+	       pstats->mac_stx[1].tx_stat_mac_4095_hi,
 	       estats->etherstatspktsover1522octets_lo,
-	       new->stats_tx.tx_gt2047_lo);
+	       pstats->mac_stx[1].tx_stat_mac_4095_lo);
 
 	ADD_64(estats->etherstatspktsover1522octets_hi,
-	       new->stats_tx.tx_gt4095_hi,
+	       pstats->mac_stx[1].tx_stat_mac_9216_hi,
 	       estats->etherstatspktsover1522octets_lo,
-	       new->stats_tx.tx_gt4095_lo);
+	       pstats->mac_stx[1].tx_stat_mac_9216_lo);
 
 	ADD_64(estats->etherstatspktsover1522octets_hi,
-	       new->stats_tx.tx_gt9216_hi,
+	       pstats->mac_stx[1].tx_stat_mac_16383_hi,
 	       estats->etherstatspktsover1522octets_lo,
-	       new->stats_tx.tx_gt9216_lo);
-
-
-	ADD_64(estats->etherstatspktsover1522octets_hi,
-	       new->stats_tx.tx_gt16383_hi,
-	       estats->etherstatspktsover1522octets_lo,
-	       new->stats_tx.tx_gt16383_lo);
+	       pstats->mac_stx[1].tx_stat_mac_16383_lo);
 
 	estats->pause_frames_received_hi =
 				pstats->mac_stx[1].rx_stat_mac_xpf_hi;
@@ -817,6 +816,7 @@ static int bnx2x_storm_stats_update(struct bnx2x *bp)
 				&bp->fw_stats_data->pf.tstorm_pf_statistics;
 	struct host_func_stats *fstats = bnx2x_sp(bp, func_stats);
 	struct bnx2x_eth_stats *estats = &bp->eth_stats;
+	struct bnx2x_eth_stats_old *estats_old = &bp->eth_stats_old;
 	struct stats_counter *counters = &bp->fw_stats_data->storm_counters;
 	int i;
 	u16 cur_stats_counter;
@@ -857,21 +857,8 @@ static int bnx2x_storm_stats_update(struct bnx2x *bp)
 		return -EAGAIN;
 	}
 
-	memcpy(&(fstats->total_bytes_received_hi),
-	       &(bnx2x_sp(bp, func_stats_base)->total_bytes_received_hi),
-	       sizeof(struct host_func_stats) - 2*sizeof(u32));
 	estats->error_bytes_received_hi = 0;
 	estats->error_bytes_received_lo = 0;
-	estats->etherstatsoverrsizepkts_hi = 0;
-	estats->etherstatsoverrsizepkts_lo = 0;
-	estats->no_buff_discard_hi = 0;
-	estats->no_buff_discard_lo = 0;
-	estats->total_tpa_aggregations_hi = 0;
-	estats->total_tpa_aggregations_lo = 0;
-	estats->total_tpa_aggregated_frames_hi = 0;
-	estats->total_tpa_aggregated_frames_lo = 0;
-	estats->total_tpa_bytes_hi = 0;
-	estats->total_tpa_bytes_lo = 0;
 
 	for_each_eth_queue(bp, i) {
 		struct bnx2x_fastpath *fp = &bp->fp[i];
@@ -888,6 +875,8 @@ static int bnx2x_storm_stats_update(struct bnx2x *bp)
 			xstorm_queue_statistics;
 		struct xstorm_per_queue_stats *old_xclient = &fp->old_xclient;
 		struct bnx2x_eth_q_stats *qstats = &fp->eth_q_stats;
+		struct bnx2x_eth_q_stats_old *qstats_old = &fp->eth_q_stats_old;
+
 		u32 diff;
 
 		DP(BNX2X_MSG_STATS, "queue[%d]: ucast_sent 0x%x, "
@@ -897,20 +886,12 @@ static int bnx2x_storm_stats_update(struct bnx2x *bp)
 
 		DP(BNX2X_MSG_STATS, "---------------\n");
 
-		qstats->total_broadcast_bytes_received_hi =
-			le32_to_cpu(tclient->rcv_bcast_bytes.hi);
-		qstats->total_broadcast_bytes_received_lo =
-			le32_to_cpu(tclient->rcv_bcast_bytes.lo);
-
-		qstats->total_multicast_bytes_received_hi =
-			le32_to_cpu(tclient->rcv_mcast_bytes.hi);
-		qstats->total_multicast_bytes_received_lo =
-			le32_to_cpu(tclient->rcv_mcast_bytes.lo);
-
-		qstats->total_unicast_bytes_received_hi =
-			le32_to_cpu(tclient->rcv_ucast_bytes.hi);
-		qstats->total_unicast_bytes_received_lo =
-			le32_to_cpu(tclient->rcv_ucast_bytes.lo);
+		UPDATE_QSTAT(tclient->rcv_bcast_bytes,
+			     total_broadcast_bytes_received);
+		UPDATE_QSTAT(tclient->rcv_mcast_bytes,
+			     total_multicast_bytes_received);
+		UPDATE_QSTAT(tclient->rcv_ucast_bytes,
+			     total_unicast_bytes_received);
 
 		/*
 		 * sum to total_bytes_received all
@@ -943,9 +924,9 @@ static int bnx2x_storm_stats_update(struct bnx2x *bp)
 					total_multicast_packets_received);
 		UPDATE_EXTEND_TSTAT(rcv_bcast_pkts,
 					total_broadcast_packets_received);
-		UPDATE_EXTEND_TSTAT(pkts_too_big_discard,
-					etherstatsoverrsizepkts);
-		UPDATE_EXTEND_TSTAT(no_buff_discard, no_buff_discard);
+		UPDATE_EXTEND_E_TSTAT(pkts_too_big_discard,
+				      etherstatsoverrsizepkts);
+		UPDATE_EXTEND_E_TSTAT(no_buff_discard, no_buff_discard);
 
 		SUB_EXTEND_USTAT(ucast_no_buff_pkts,
 					total_unicast_packets_received);
@@ -953,24 +934,17 @@ static int bnx2x_storm_stats_update(struct bnx2x *bp)
 					total_multicast_packets_received);
 		SUB_EXTEND_USTAT(bcast_no_buff_pkts,
 					total_broadcast_packets_received);
-		UPDATE_EXTEND_USTAT(ucast_no_buff_pkts, no_buff_discard);
-		UPDATE_EXTEND_USTAT(mcast_no_buff_pkts, no_buff_discard);
-		UPDATE_EXTEND_USTAT(bcast_no_buff_pkts, no_buff_discard);
+		UPDATE_EXTEND_E_USTAT(ucast_no_buff_pkts, no_buff_discard);
+		UPDATE_EXTEND_E_USTAT(mcast_no_buff_pkts, no_buff_discard);
+		UPDATE_EXTEND_E_USTAT(bcast_no_buff_pkts, no_buff_discard);
 
-		qstats->total_broadcast_bytes_transmitted_hi =
-			le32_to_cpu(xclient->bcast_bytes_sent.hi);
-		qstats->total_broadcast_bytes_transmitted_lo =
-			le32_to_cpu(xclient->bcast_bytes_sent.lo);
+		UPDATE_QSTAT(xclient->bcast_bytes_sent,
+			     total_broadcast_bytes_transmitted);
+		UPDATE_QSTAT(xclient->mcast_bytes_sent,
+			     total_multicast_bytes_transmitted);
+		UPDATE_QSTAT(xclient->ucast_bytes_sent,
+			     total_unicast_bytes_transmitted);
 
-		qstats->total_multicast_bytes_transmitted_hi =
-			le32_to_cpu(xclient->mcast_bytes_sent.hi);
-		qstats->total_multicast_bytes_transmitted_lo =
-			le32_to_cpu(xclient->mcast_bytes_sent.lo);
-
-		qstats->total_unicast_bytes_transmitted_hi =
-			le32_to_cpu(xclient->ucast_bytes_sent.hi);
-		qstats->total_unicast_bytes_transmitted_lo =
-			le32_to_cpu(xclient->ucast_bytes_sent.lo);
 		/*
 		 * sum to total_bytes_transmitted all
 		 * unicast/multicast/broadcast
@@ -1006,110 +980,54 @@ static int bnx2x_storm_stats_update(struct bnx2x *bp)
 				    total_transmitted_dropped_packets_error);
 
 		/* TPA aggregations completed */
-		UPDATE_EXTEND_USTAT(coalesced_events, total_tpa_aggregations);
+		UPDATE_EXTEND_E_USTAT(coalesced_events, total_tpa_aggregations);
 		/* Number of network frames aggregated by TPA */
-		UPDATE_EXTEND_USTAT(coalesced_pkts,
-				    total_tpa_aggregated_frames);
+		UPDATE_EXTEND_E_USTAT(coalesced_pkts,
+				      total_tpa_aggregated_frames);
 		/* Total number of bytes in completed TPA aggregations */
-		qstats->total_tpa_bytes_lo =
-			le32_to_cpu(uclient->coalesced_bytes.lo);
-		qstats->total_tpa_bytes_hi =
-			le32_to_cpu(uclient->coalesced_bytes.hi);
+		UPDATE_QSTAT(uclient->coalesced_bytes, total_tpa_bytes);
 
-		/* TPA stats per-function */
-		ADD_64(estats->total_tpa_aggregations_hi,
-		       qstats->total_tpa_aggregations_hi,
-		       estats->total_tpa_aggregations_lo,
-		       qstats->total_tpa_aggregations_lo);
-		ADD_64(estats->total_tpa_aggregated_frames_hi,
-		       qstats->total_tpa_aggregated_frames_hi,
-		       estats->total_tpa_aggregated_frames_lo,
-		       qstats->total_tpa_aggregated_frames_lo);
-		ADD_64(estats->total_tpa_bytes_hi,
-		       qstats->total_tpa_bytes_hi,
-		       estats->total_tpa_bytes_lo,
-		       qstats->total_tpa_bytes_lo);
+		UPDATE_ESTAT_QSTAT_64(total_tpa_bytes);
 
-		ADD_64(fstats->total_bytes_received_hi,
-		       qstats->total_bytes_received_hi,
-		       fstats->total_bytes_received_lo,
-		       qstats->total_bytes_received_lo);
-		ADD_64(fstats->total_bytes_transmitted_hi,
-		       qstats->total_bytes_transmitted_hi,
-		       fstats->total_bytes_transmitted_lo,
-		       qstats->total_bytes_transmitted_lo);
-		ADD_64(fstats->total_unicast_packets_received_hi,
-		       qstats->total_unicast_packets_received_hi,
-		       fstats->total_unicast_packets_received_lo,
-		       qstats->total_unicast_packets_received_lo);
-		ADD_64(fstats->total_multicast_packets_received_hi,
-		       qstats->total_multicast_packets_received_hi,
-		       fstats->total_multicast_packets_received_lo,
-		       qstats->total_multicast_packets_received_lo);
-		ADD_64(fstats->total_broadcast_packets_received_hi,
-		       qstats->total_broadcast_packets_received_hi,
-		       fstats->total_broadcast_packets_received_lo,
-		       qstats->total_broadcast_packets_received_lo);
-		ADD_64(fstats->total_unicast_packets_transmitted_hi,
-		       qstats->total_unicast_packets_transmitted_hi,
-		       fstats->total_unicast_packets_transmitted_lo,
-		       qstats->total_unicast_packets_transmitted_lo);
-		ADD_64(fstats->total_multicast_packets_transmitted_hi,
-		       qstats->total_multicast_packets_transmitted_hi,
-		       fstats->total_multicast_packets_transmitted_lo,
-		       qstats->total_multicast_packets_transmitted_lo);
-		ADD_64(fstats->total_broadcast_packets_transmitted_hi,
-		       qstats->total_broadcast_packets_transmitted_hi,
-		       fstats->total_broadcast_packets_transmitted_lo,
-		       qstats->total_broadcast_packets_transmitted_lo);
-		ADD_64(fstats->valid_bytes_received_hi,
-		       qstats->valid_bytes_received_hi,
-		       fstats->valid_bytes_received_lo,
-		       qstats->valid_bytes_received_lo);
-
-		ADD_64(estats->etherstatsoverrsizepkts_hi,
-		       qstats->etherstatsoverrsizepkts_hi,
-		       estats->etherstatsoverrsizepkts_lo,
-		       qstats->etherstatsoverrsizepkts_lo);
-		ADD_64(estats->no_buff_discard_hi, qstats->no_buff_discard_hi,
-		       estats->no_buff_discard_lo, qstats->no_buff_discard_lo);
+		UPDATE_FSTAT_QSTAT(total_bytes_received);
+		UPDATE_FSTAT_QSTAT(total_bytes_transmitted);
+		UPDATE_FSTAT_QSTAT(total_unicast_packets_received);
+		UPDATE_FSTAT_QSTAT(total_multicast_packets_received);
+		UPDATE_FSTAT_QSTAT(total_broadcast_packets_received);
+		UPDATE_FSTAT_QSTAT(total_unicast_packets_transmitted);
+		UPDATE_FSTAT_QSTAT(total_multicast_packets_transmitted);
+		UPDATE_FSTAT_QSTAT(total_broadcast_packets_transmitted);
+		UPDATE_FSTAT_QSTAT(valid_bytes_received);
 	}
 
-	ADD_64(fstats->total_bytes_received_hi,
+	ADD_64(estats->total_bytes_received_hi,
 	       estats->rx_stat_ifhcinbadoctets_hi,
-	       fstats->total_bytes_received_lo,
+	       estats->total_bytes_received_lo,
 	       estats->rx_stat_ifhcinbadoctets_lo);
 
-	ADD_64(fstats->total_bytes_received_hi,
+	ADD_64(estats->total_bytes_received_hi,
 	       tfunc->rcv_error_bytes.hi,
-	       fstats->total_bytes_received_lo,
+	       estats->total_bytes_received_lo,
 	       tfunc->rcv_error_bytes.lo);
 
-	memcpy(estats, &(fstats->total_bytes_received_hi),
-	       sizeof(struct host_func_stats) - 2*sizeof(u32));
-
 	ADD_64(estats->error_bytes_received_hi,
 	       tfunc->rcv_error_bytes.hi,
 	       estats->error_bytes_received_lo,
 	       tfunc->rcv_error_bytes.lo);
 
-	ADD_64(estats->etherstatsoverrsizepkts_hi,
-	       estats->rx_stat_dot3statsframestoolong_hi,
-	       estats->etherstatsoverrsizepkts_lo,
-	       estats->rx_stat_dot3statsframestoolong_lo);
+	UPDATE_ESTAT(etherstatsoverrsizepkts, rx_stat_dot3statsframestoolong);
+
 	ADD_64(estats->error_bytes_received_hi,
 	       estats->rx_stat_ifhcinbadoctets_hi,
 	       estats->error_bytes_received_lo,
 	       estats->rx_stat_ifhcinbadoctets_lo);
 
 	if (bp->port.pmf) {
-		estats->mac_filter_discard =
-				le32_to_cpu(tport->mac_filter_discard);
-		estats->mf_tag_discard =
-				le32_to_cpu(tport->mf_tag_discard);
-		estats->brb_truncate_discard =
-				le32_to_cpu(tport->brb_truncate_discard);
-		estats->mac_discard = le32_to_cpu(tport->mac_discard);
+		struct bnx2x_fw_port_stats_old *fwstats = &bp->fw_stats_old;
+		UPDATE_FW_STAT(mac_filter_discard);
+		UPDATE_FW_STAT(mf_tag_discard);
+		UPDATE_FW_STAT(brb_truncate_discard);
+		UPDATE_FW_STAT(mac_discard);
 	}
 
 	fstats->host_func_stats_start = ++fstats->host_func_stats_end;
@@ -1143,7 +1061,7 @@ static void bnx2x_net_stats_update(struct bnx2x *bp)
 	tmp = estats->mac_discard;
 	for_each_rx_queue(bp, i)
 		tmp += le32_to_cpu(bp->fp[i].old_tclient.checksum_discard);
-	nstats->rx_dropped = tmp;
+	nstats->rx_dropped = tmp + bp->net_stats_old.rx_dropped;
 
 	nstats->tx_dropped = 0;
 
@@ -1191,17 +1109,15 @@ static void bnx2x_drv_stats_update(struct bnx2x *bp)
 	struct bnx2x_eth_stats *estats = &bp->eth_stats;
 	int i;
 
-	estats->driver_xoff = 0;
-	estats->rx_err_discard_pkt = 0;
-	estats->rx_skb_alloc_failed = 0;
-	estats->hw_csum_err = 0;
 	for_each_queue(bp, i) {
 		struct bnx2x_eth_q_stats *qstats = &bp->fp[i].eth_q_stats;
+		struct bnx2x_eth_q_stats_old *qstats_old =
+						&bp->fp[i].eth_q_stats_old;
 
-		estats->driver_xoff += qstats->driver_xoff;
-		estats->rx_err_discard_pkt += qstats->rx_err_discard_pkt;
-		estats->rx_skb_alloc_failed += qstats->rx_skb_alloc_failed;
-		estats->hw_csum_err += qstats->hw_csum_err;
+		UPDATE_ESTAT_QSTAT(driver_xoff);
+		UPDATE_ESTAT_QSTAT(rx_err_discard_pkt);
+		UPDATE_ESTAT_QSTAT(rx_skb_alloc_failed);
+		UPDATE_ESTAT_QSTAT(hw_csum_err);
 	}
 }
 
@@ -1446,33 +1362,6 @@ static void bnx2x_port_stats_base_init(struct bnx2x *bp)
 	bnx2x_stats_comp(bp);
 }
 
-static void bnx2x_func_stats_base_init(struct bnx2x *bp)
-{
-	int vn, vn_max = IS_MF(bp) ? BP_MAX_VN_NUM(bp) : E1VN_MAX;
-	u32 func_stx;
-
-	/* sanity */
-	if (!bp->port.pmf || !bp->func_stx) {
-		BNX2X_ERR("BUG!\n");
-		return;
-	}
-
-	/* save our func_stx */
-	func_stx = bp->func_stx;
-
-	for (vn = VN_0; vn < vn_max; vn++) {
-		int mb_idx = BP_FW_MB_IDX_VN(bp, vn);
-
-		bp->func_stx = SHMEM_RD(bp, func_mb[mb_idx].fw_mb_param);
-		bnx2x_func_stats_init(bp);
-		bnx2x_hw_stats_post(bp);
-		bnx2x_stats_comp(bp);
-	}
-
-	/* restore our func_stx */
-	bp->func_stx = func_stx;
-}
-
 static void bnx2x_func_stats_base_update(struct bnx2x *bp)
 {
 	struct dmae_command *dmae = &bp->stats_dmae;
@@ -1491,8 +1380,8 @@ static void bnx2x_func_stats_base_update(struct bnx2x *bp)
 					 true, DMAE_COMP_PCI);
 	dmae->src_addr_lo = bp->func_stx >> 2;
 	dmae->src_addr_hi = 0;
-	dmae->dst_addr_lo = U64_LO(bnx2x_sp_mapping(bp, func_stats_base));
-	dmae->dst_addr_hi = U64_HI(bnx2x_sp_mapping(bp, func_stats_base));
+	dmae->dst_addr_lo = U64_LO(bnx2x_sp_mapping(bp, func_stats));
+	dmae->dst_addr_hi = U64_HI(bnx2x_sp_mapping(bp, func_stats));
 	dmae->len = sizeof(struct host_func_stats) >> 2;
 	dmae->comp_addr_lo = U64_LO(bnx2x_sp_mapping(bp, stats_comp));
 	dmae->comp_addr_hi = U64_HI(bnx2x_sp_mapping(bp, stats_comp));
@@ -1653,6 +1542,10 @@ void bnx2x_stats_init(struct bnx2x *bp)
 	DP(BNX2X_MSG_STATS, "port_stx 0x%x  func_stx 0x%x\n",
 	   bp->port.port_stx, bp->func_stx);
 
+	/* pmf should retrieve port statistics from SP on a non-init*/
+	if (!bp->stats_init && bp->port.pmf && bp->port.port_stx)
+		bnx2x_stats_handle(bp, STATS_EVENT_PMF);
+
 	port = BP_PORT(bp);
 	/* port stats */
 	memset(&(bp->port.old_nig_stats), 0, sizeof(struct nig_stats));
@@ -1674,24 +1567,83 @@ void bnx2x_stats_init(struct bnx2x *bp)
 		memset(&fp->old_tclient, 0, sizeof(fp->old_tclient));
 		memset(&fp->old_uclient, 0, sizeof(fp->old_uclient));
 		memset(&fp->old_xclient, 0, sizeof(fp->old_xclient));
-		memset(&fp->eth_q_stats, 0, sizeof(fp->eth_q_stats));
+		if (bp->stats_init) {
+			memset(&fp->eth_q_stats, 0, sizeof(fp->eth_q_stats));
+			memset(&fp->eth_q_stats_old, 0,
+			       sizeof(fp->eth_q_stats_old));
+		}
 	}
 
 	/* Prepare statistics ramrod data */
 	bnx2x_prep_fw_stats_req(bp);
 
 	memset(&bp->dev->stats, 0, sizeof(bp->dev->stats));
-	memset(&bp->eth_stats, 0, sizeof(bp->eth_stats));
+	if (bp->stats_init) {
+		memset(&bp->net_stats_old, 0, sizeof(bp->net_stats_old));
+		memset(&bp->fw_stats_old, 0, sizeof(bp->fw_stats_old));
+		memset(&bp->eth_stats_old, 0, sizeof(bp->eth_stats_old));
+		memset(&bp->eth_stats, 0, sizeof(bp->eth_stats));
+
+		/* Clean SP from previous statistics */
+		if (bp->func_stx) {
+			memset(bnx2x_sp(bp, func_stats), 0,
+			       sizeof(struct host_func_stats));
+			bnx2x_func_stats_init(bp);
+			bnx2x_hw_stats_post(bp);
+			bnx2x_stats_comp(bp);
+		}
+	}
 
 	bp->stats_state = STATS_STATE_DISABLED;
 
-	if (bp->port.pmf) {
-		if (bp->port.port_stx)
-			bnx2x_port_stats_base_init(bp);
+	if (bp->port.pmf && bp->port.port_stx)
+		bnx2x_port_stats_base_init(bp);
 
-		if (bp->func_stx)
-			bnx2x_func_stats_base_init(bp);
-
-	} else if (bp->func_stx)
+	/* On a non-init, retrieve previous statistics from SP */
+	if (!bp->stats_init && bp->func_stx)
 		bnx2x_func_stats_base_update(bp);
+
+	/* mark the end of statistics initializiation */
+	bp->stats_init = false;
+}
+
+void bnx2x_save_statistics(struct bnx2x *bp)
+{
+	int i;
+	struct net_device_stats *nstats = &bp->dev->stats;
+
+	/* save queue statistics */
+	for_each_eth_queue(bp, i) {
+		struct bnx2x_fastpath *fp = &bp->fp[i];
+		struct bnx2x_eth_q_stats *qstats = &fp->eth_q_stats;
+		struct bnx2x_eth_q_stats_old *qstats_old = &fp->eth_q_stats_old;
+
+		UPDATE_QSTAT_OLD(total_unicast_bytes_received_hi);
+		UPDATE_QSTAT_OLD(total_unicast_bytes_received_lo);
+		UPDATE_QSTAT_OLD(total_broadcast_bytes_received_hi);
+		UPDATE_QSTAT_OLD(total_broadcast_bytes_received_lo);
+		UPDATE_QSTAT_OLD(total_multicast_bytes_received_hi);
+		UPDATE_QSTAT_OLD(total_multicast_bytes_received_lo);
+		UPDATE_QSTAT_OLD(total_unicast_bytes_transmitted_hi);
+		UPDATE_QSTAT_OLD(total_unicast_bytes_transmitted_lo);
+		UPDATE_QSTAT_OLD(total_broadcast_bytes_transmitted_hi);
+		UPDATE_QSTAT_OLD(total_broadcast_bytes_transmitted_lo);
+		UPDATE_QSTAT_OLD(total_multicast_bytes_transmitted_hi);
+		UPDATE_QSTAT_OLD(total_multicast_bytes_transmitted_lo);
+		UPDATE_QSTAT_OLD(total_tpa_bytes_hi);
+		UPDATE_QSTAT_OLD(total_tpa_bytes_lo);
+	}
+
+	/* save net_device_stats statistics */
+	bp->net_stats_old.rx_dropped = nstats->rx_dropped;
+
+	/* store port firmware statistics */
+	if (bp->port.pmf && IS_MF(bp)) {
+		struct bnx2x_eth_stats *estats = &bp->eth_stats;
+		struct bnx2x_fw_port_stats_old *fwstats = &bp->fw_stats_old;
+		UPDATE_FW_STAT_OLD(mac_filter_discard);
+		UPDATE_FW_STAT_OLD(mf_tag_discard);
+		UPDATE_FW_STAT_OLD(brb_truncate_discard);
+		UPDATE_FW_STAT_OLD(mac_discard);
+	}
 }
diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_stats.h b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_stats.h
index 7e979686cd68..39ffd6dcdf1a 100644
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_stats.h
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_stats.h
@@ -264,6 +264,69 @@ struct bnx2x_eth_q_stats {
 	u32 total_tpa_bytes_lo;
 };
 
+struct bnx2x_eth_stats_old {
+	u32 rx_stat_dot3statsframestoolong_hi;
+	u32 rx_stat_dot3statsframestoolong_lo;
+};
+
+struct bnx2x_eth_q_stats_old {
+	/* Fields to perserve over fw reset*/
+	u32 total_unicast_bytes_received_hi;
+	u32 total_unicast_bytes_received_lo;
+	u32 total_broadcast_bytes_received_hi;
+	u32 total_broadcast_bytes_received_lo;
+	u32 total_multicast_bytes_received_hi;
+	u32 total_multicast_bytes_received_lo;
+	u32 total_unicast_bytes_transmitted_hi;
+	u32 total_unicast_bytes_transmitted_lo;
+	u32 total_broadcast_bytes_transmitted_hi;
+	u32 total_broadcast_bytes_transmitted_lo;
+	u32 total_multicast_bytes_transmitted_hi;
+	u32 total_multicast_bytes_transmitted_lo;
+	u32 total_tpa_bytes_hi;
+	u32 total_tpa_bytes_lo;
+
+	/* Fields to perserve last of */
+	u32 total_bytes_received_hi;
+	u32 total_bytes_received_lo;
+	u32 total_bytes_transmitted_hi;
+	u32 total_bytes_transmitted_lo;
+	u32 total_unicast_packets_received_hi;
+	u32 total_unicast_packets_received_lo;
+	u32 total_multicast_packets_received_hi;
+	u32 total_multicast_packets_received_lo;
+	u32 total_broadcast_packets_received_hi;
+	u32 total_broadcast_packets_received_lo;
+	u32 total_unicast_packets_transmitted_hi;
+	u32 total_unicast_packets_transmitted_lo;
+	u32 total_multicast_packets_transmitted_hi;
+	u32 total_multicast_packets_transmitted_lo;
+	u32 total_broadcast_packets_transmitted_hi;
+	u32 total_broadcast_packets_transmitted_lo;
+	u32 valid_bytes_received_hi;
+	u32 valid_bytes_received_lo;
+
+	u32 total_tpa_bytes_hi_old;
+	u32 total_tpa_bytes_lo_old;
+
+	u32 driver_xoff_old;
+	u32 rx_err_discard_pkt_old;
+	u32 rx_skb_alloc_failed_old;
+	u32 hw_csum_err_old;
+};
+
+struct bnx2x_net_stats_old {
+	 u32 rx_dropped;
+};
+
+struct bnx2x_fw_port_stats_old {
+	 u32 mac_filter_discard;
+	 u32 mf_tag_discard;
+	 u32 brb_truncate_discard;
+	 u32 mac_discard;
+};
+
+
 /****************************************************************************
 * Macros
 ****************************************************************************/
@@ -348,11 +411,28 @@ struct bnx2x_eth_q_stats {
 		ADD_EXTEND_64(qstats->t##_hi, qstats->t##_lo, diff); \
 	} while (0)
 
+#define UPDATE_EXTEND_E_TSTAT(s, t) \
+	do { \
+		UPDATE_EXTEND_TSTAT(s, t); \
+		ADD_EXTEND_64(estats->t##_hi, estats->t##_lo, diff); \
+	} while (0)
+
 #define UPDATE_EXTEND_USTAT(s, t) \
 	do { \
 		diff = le32_to_cpu(uclient->s) - le32_to_cpu(old_uclient->s); \
 		old_uclient->s = uclient->s; \
-		ADD_EXTEND_64(qstats->t##_hi, qstats->t##_lo, diff); \
+	} while (0)
+
+#define UPDATE_EXTEND_E_USTAT(s, t) \
+	do { \
+		UPDATE_EXTEND_USTAT(s, t); \
+		ADD_EXTEND_64(estats->t##_hi, estats->t##_lo, diff); \
+	} while (0)
+
+#define UPDATE_EXTEND_E_USTAT(s, t) \
+	do { \
+		UPDATE_EXTEND_USTAT(s, t); \
+		ADD_EXTEND_64(estats->t##_hi, estats->t##_lo, diff); \
 	} while (0)
 
 #define UPDATE_EXTEND_XSTAT(s, t) \
@@ -362,6 +442,66 @@ struct bnx2x_eth_q_stats {
 		ADD_EXTEND_64(qstats->t##_hi, qstats->t##_lo, diff); \
 	} while (0)
 
+#define UPDATE_QSTAT(s, t) \
+	do { \
+		qstats->t##_hi = qstats_old->t##_hi + le32_to_cpu(s.hi); \
+		qstats->t##_lo = qstats_old->t##_lo + le32_to_cpu(s.lo); \
+	} while (0)
+
+#define UPDATE_QSTAT_OLD(f) \
+	do { \
+		qstats_old->f = qstats->f; \
+	} while (0)
+
+#define UPDATE_ESTAT_QSTAT_64(s) \
+	do { \
+		ADD_64(estats->s##_hi, qstats->s##_hi, \
+		       estats->s##_lo, qstats->s##_lo); \
+		SUB_64(estats->s##_hi, qstats_old->s##_hi_old, \
+		       estats->s##_lo, qstats_old->s##_lo_old); \
+		qstats_old->s##_hi_old = qstats->s##_hi; \
+		qstats_old->s##_lo_old = qstats->s##_lo; \
+	} while (0)
+
+#define UPDATE_ESTAT_QSTAT(s) \
+	do { \
+		estats->s += qstats->s; \
+		estats->s -= qstats_old->s##_old; \
+		qstats_old->s##_old = qstats->s; \
+	} while (0)
+
+#define UPDATE_FSTAT_QSTAT(s) \
+	do { \
+		ADD_64(fstats->s##_hi, qstats->s##_hi, \
+		       fstats->s##_lo, qstats->s##_lo); \
+		SUB_64(fstats->s##_hi, qstats_old->s##_hi, \
+		       fstats->s##_lo, qstats_old->s##_lo); \
+		estats->s##_hi = fstats->s##_hi; \
+		estats->s##_lo = fstats->s##_lo; \
+		qstats_old->s##_hi = qstats->s##_hi; \
+		qstats_old->s##_lo = qstats->s##_lo; \
+	} while (0)
+
+#define UPDATE_FW_STAT(s) \
+	do { \
+		estats->s = le32_to_cpu(tport->s) + fwstats->s; \
+	} while (0)
+
+#define UPDATE_FW_STAT_OLD(f) \
+	do { \
+		fwstats->f = estats->f; \
+	} while (0)
+
+#define UPDATE_ESTAT(s, t) \
+	do { \
+		SUB_64(estats->s##_hi, estats_old->t##_hi, \
+		       estats->s##_lo, estats_old->t##_lo); \
+		ADD_64(estats->s##_hi, estats->t##_hi, \
+		       estats->s##_lo, estats->t##_lo); \
+		estats_old->t##_hi = estats->t##_hi; \
+		estats_old->t##_lo = estats->t##_lo; \
+	} while (0)
+
 /* minuend -= subtrahend */
 #define SUB_64(m_hi, s_hi, m_lo, s_lo) \
 	do { \
@@ -388,4 +528,10 @@ void bnx2x_stats_init(struct bnx2x *bp);
 
 void bnx2x_stats_handle(struct bnx2x *bp, enum bnx2x_stats_event event);
 
+/**
+ * bnx2x_save_statistics - save statistics when unloading.
+ *
+ * @bp:		driver handle
+ */
+void bnx2x_save_statistics(struct bnx2x *bp);
 #endif /* BNX2X_STATS_H */