Merge branch 'qed_hw_gro'

Manish Chopra says:

====================
qed/qede: Add hardware GRO support

This patch series enables hardware GRO and add support for handling
HW aggregated TCP packets in driver receive flow by skipping
software GRO handling in stack.

Please consider applying this series to net-next.
====================

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

5 files changed, 476 insertions, 37 deletions

diff --git a/drivers/net/ethernet/qlogic/qed/qed_hsi.h b/drivers/net/ethernet/qlogic/qed/qed_hsi.h
index 592e0e6d9b42..236db8a99ec3 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_hsi.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_hsi.h
@@ -2919,7 +2919,19 @@ struct eth_vport_rx_mode {
 };
 
 struct eth_vport_tpa_param {
-	u64     reserved[2];
+	u8	tpa_ipv4_en_flg;
+	u8	tpa_ipv6_en_flg;
+	u8	tpa_ipv4_tunn_en_flg;
+	u8	tpa_ipv6_tunn_en_flg;
+	u8	tpa_pkt_split_flg;
+	u8	tpa_hdr_data_split_flg;
+	u8	tpa_gro_consistent_flg;
+	u8	tpa_max_aggs_num;
+	u16	tpa_max_size;
+	u16	tpa_min_size_to_start;
+	u16	tpa_min_size_to_cont;
+	u8	max_buff_num;
+	u8	reserved;
 };
 
 struct eth_vport_tx_mode {
diff --git a/drivers/net/ethernet/qlogic/qed/qed_l2.c b/drivers/net/ethernet/qlogic/qed/qed_l2.c
index 102ddc73b841..3f35c6ca9252 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_l2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_l2.c
@@ -132,16 +132,29 @@ struct qed_sp_vport_update_params {
 	struct qed_filter_accept_flags	accept_flags;
 };
 
+enum qed_tpa_mode {
+	QED_TPA_MODE_NONE,
+	QED_TPA_MODE_UNUSED,
+	QED_TPA_MODE_GRO,
+	QED_TPA_MODE_MAX
+};
+
+struct qed_sp_vport_start_params {
+	enum qed_tpa_mode	tpa_mode;
+	bool			remove_inner_vlan;
+	bool			drop_ttl0;
+	u8			max_buffers_per_cqe;
+	u32			concrete_fid;
+	u16			opaque_fid;
+	u8			vport_id;
+	u16			mtu;
+};
+
 #define QED_MAX_SGES_NUM 16
 #define CRC32_POLY 0x1edc6f41
 
 static int qed_sp_vport_start(struct qed_hwfn *p_hwfn,
-			      u32 concrete_fid,
-			      u16 opaque_fid,
-			      u8 vport_id,
-			      u16 mtu,
-			      u8 drop_ttl0_flg,
-			      u8 inner_vlan_removal_en_flg)
+			      struct qed_sp_vport_start_params *p_params)
 {
 	struct vport_start_ramrod_data *p_ramrod = NULL;
 	struct qed_spq_entry *p_ent =  NULL;
@@ -150,13 +163,13 @@ static int qed_sp_vport_start(struct qed_hwfn *p_hwfn,
 	u16 rx_mode = 0;
 	u8 abs_vport_id = 0;
 
-	rc = qed_fw_vport(p_hwfn, vport_id, &abs_vport_id);
+	rc = qed_fw_vport(p_hwfn, p_params->vport_id, &abs_vport_id);
 	if (rc != 0)
 		return rc;
 
 	memset(&init_data, 0, sizeof(init_data));
 	init_data.cid = qed_spq_get_cid(p_hwfn);
-	init_data.opaque_fid = opaque_fid;
+	init_data.opaque_fid = p_params->opaque_fid;
 	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
 
 	rc = qed_sp_init_request(p_hwfn, &p_ent,
@@ -168,9 +181,9 @@ static int qed_sp_vport_start(struct qed_hwfn *p_hwfn,
 	p_ramrod		= &p_ent->ramrod.vport_start;
 	p_ramrod->vport_id	= abs_vport_id;
 
-	p_ramrod->mtu			= cpu_to_le16(mtu);
-	p_ramrod->inner_vlan_removal_en = inner_vlan_removal_en_flg;
-	p_ramrod->drop_ttl0_en		= drop_ttl0_flg;
+	p_ramrod->mtu			= cpu_to_le16(p_params->mtu);
+	p_ramrod->inner_vlan_removal_en	= p_params->remove_inner_vlan;
+	p_ramrod->drop_ttl0_en		= p_params->drop_ttl0;
 
 	SET_FIELD(rx_mode, ETH_VPORT_RX_MODE_UCAST_DROP_ALL, 1);
 	SET_FIELD(rx_mode, ETH_VPORT_RX_MODE_MCAST_DROP_ALL, 1);
@@ -181,9 +194,26 @@ static int qed_sp_vport_start(struct qed_hwfn *p_hwfn,
 	memset(&p_ramrod->tpa_param, 0,
 	       sizeof(struct eth_vport_tpa_param));
 
+	p_ramrod->tpa_param.max_buff_num = p_params->max_buffers_per_cqe;
+
+	switch (p_params->tpa_mode) {
+	case QED_TPA_MODE_GRO:
+		p_ramrod->tpa_param.tpa_max_aggs_num = ETH_TPA_MAX_AGGS_NUM;
+		p_ramrod->tpa_param.tpa_max_size = (u16)-1;
+		p_ramrod->tpa_param.tpa_min_size_to_cont = p_params->mtu / 2;
+		p_ramrod->tpa_param.tpa_min_size_to_start = p_params->mtu / 2;
+		p_ramrod->tpa_param.tpa_ipv4_en_flg = 1;
+		p_ramrod->tpa_param.tpa_ipv6_en_flg = 1;
+		p_ramrod->tpa_param.tpa_pkt_split_flg = 1;
+		p_ramrod->tpa_param.tpa_gro_consistent_flg = 1;
+		break;
+	default:
+		break;
+	}
+
 	/* Software Function ID in hwfn (PFs are 0 - 15, VFs are 16 - 135) */
 	p_ramrod->sw_fid = qed_concrete_to_sw_fid(p_hwfn->cdev,
-						  concrete_fid);
+						  p_params->concrete_fid);
 
 	return qed_spq_post(p_hwfn, p_ent, NULL);
 }
@@ -1592,24 +1622,25 @@ static void qed_register_eth_ops(struct qed_dev *cdev,
 }
 
 static int qed_start_vport(struct qed_dev *cdev,
-			   u8 vport_id,
-			   u16 mtu,
-			   u8 drop_ttl0_flg,
-			   u8 inner_vlan_removal_en_flg)
+			   struct qed_start_vport_params *params)
 {
 	int rc, i;
 
 	for_each_hwfn(cdev, i) {
+		struct qed_sp_vport_start_params start = { 0 };
 		struct qed_hwfn *p_hwfn = &cdev->hwfns[i];
 
-		rc = qed_sp_vport_start(p_hwfn,
-					p_hwfn->hw_info.concrete_fid,
-					p_hwfn->hw_info.opaque_fid,
-					vport_id,
-					mtu,
-					drop_ttl0_flg,
-					inner_vlan_removal_en_flg);
-
+		start.tpa_mode = params->gro_enable ? QED_TPA_MODE_GRO :
+							QED_TPA_MODE_NONE;
+		start.remove_inner_vlan = params->remove_inner_vlan;
+		start.drop_ttl0 = params->drop_ttl0;
+		start.opaque_fid = p_hwfn->hw_info.opaque_fid;
+		start.concrete_fid = p_hwfn->hw_info.concrete_fid;
+		start.vport_id = params->vport_id;
+		start.max_buffers_per_cqe = 16;
+		start.mtu = params->mtu;
+
+		rc = qed_sp_vport_start(p_hwfn, &start);
 		if (rc) {
 			DP_ERR(cdev, "Failed to start VPORT\n");
 			return rc;
@@ -1619,7 +1650,7 @@ static int qed_start_vport(struct qed_dev *cdev,
 
 		DP_VERBOSE(cdev, (QED_MSG_SPQ | NETIF_MSG_IFUP),
 			   "Started V-PORT %d with MTU %d\n",
-			   vport_id, mtu);
+			   start.vport_id, start.mtu);
 	}
 
 	qed_reset_vport_stats(cdev);
diff --git a/drivers/net/ethernet/qlogic/qede/qede.h b/drivers/net/ethernet/qlogic/qede/qede.h
index 02e17d331f22..d023251544d9 100644
--- a/drivers/net/ethernet/qlogic/qede/qede.h
+++ b/drivers/net/ethernet/qlogic/qede/qede.h
@@ -160,6 +160,7 @@ struct qede_dev {
 	u16			q_num_rx_buffers; /* Must be a power of two */
 	u16			q_num_tx_buffers; /* Must be a power of two */
 
+	bool gro_disable;
 	struct list_head vlan_list;
 	u16 configured_vlans;
 	u16 non_configured_vlans;
@@ -188,6 +189,24 @@ struct sw_rx_data {
 	unsigned int page_offset;
 };
 
+enum qede_agg_state {
+	QEDE_AGG_STATE_NONE  = 0,
+	QEDE_AGG_STATE_START = 1,
+	QEDE_AGG_STATE_ERROR = 2
+};
+
+struct qede_agg_info {
+	struct sw_rx_data replace_buf;
+	dma_addr_t replace_buf_mapping;
+	struct sw_rx_data start_buf;
+	dma_addr_t start_buf_mapping;
+	struct eth_fast_path_rx_tpa_start_cqe start_cqe;
+	enum qede_agg_state agg_state;
+	struct sk_buff *skb;
+	int frag_id;
+	u16 vlan_tag;
+};
+
 struct qede_rx_queue {
 	__le16			*hw_cons_ptr;
 	struct sw_rx_data	*sw_rx_ring;
@@ -197,6 +216,9 @@ struct qede_rx_queue {
 	struct qed_chain	rx_comp_ring;
 	void __iomem		*hw_rxq_prod_addr;
 
+	/* GRO */
+	struct qede_agg_info	tpa_info[ETH_TPA_MAX_AGGS_NUM];
+
 	int			rx_buf_size;
 	unsigned int		rx_buf_seg_size;
 
diff --git a/drivers/net/ethernet/qlogic/qede/qede_main.c b/drivers/net/ethernet/qlogic/qede/qede_main.c
index ddd9e4aaa500..572862564ab6 100644
--- a/drivers/net/ethernet/qlogic/qede/qede_main.c
+++ b/drivers/net/ethernet/qlogic/qede/qede_main.c
@@ -866,6 +866,278 @@ static inline void qede_skb_receive(struct qede_dev *edev,
 	napi_gro_receive(&fp->napi, skb);
 }
 
+static void qede_set_gro_params(struct qede_dev *edev,
+				struct sk_buff *skb,
+				struct eth_fast_path_rx_tpa_start_cqe *cqe)
+{
+	u16 parsing_flags = le16_to_cpu(cqe->pars_flags.flags);
+
+	if (((parsing_flags >> PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) &
+	    PARSING_AND_ERR_FLAGS_L3TYPE_MASK) == 2)
+		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
+	else
+		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
+
+	skb_shinfo(skb)->gso_size = __le16_to_cpu(cqe->len_on_first_bd) -
+					cqe->header_len;
+}
+
+static int qede_fill_frag_skb(struct qede_dev *edev,
+			      struct qede_rx_queue *rxq,
+			      u8 tpa_agg_index,
+			      u16 len_on_bd)
+{
+	struct sw_rx_data *current_bd = &rxq->sw_rx_ring[rxq->sw_rx_cons &
+							 NUM_RX_BDS_MAX];
+	struct qede_agg_info *tpa_info = &rxq->tpa_info[tpa_agg_index];
+	struct sk_buff *skb = tpa_info->skb;
+
+	if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
+		goto out;
+
+	/* Add one frag and update the appropriate fields in the skb */
+	skb_fill_page_desc(skb, tpa_info->frag_id++,
+			   current_bd->data, current_bd->page_offset,
+			   len_on_bd);
+
+	if (unlikely(qede_realloc_rx_buffer(edev, rxq, current_bd))) {
+		tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
+		goto out;
+	}
+
+	qed_chain_consume(&rxq->rx_bd_ring);
+	rxq->sw_rx_cons++;
+
+	skb->data_len += len_on_bd;
+	skb->truesize += rxq->rx_buf_seg_size;
+	skb->len += len_on_bd;
+
+	return 0;
+
+out:
+	return -ENOMEM;
+}
+
+static void qede_tpa_start(struct qede_dev *edev,
+			   struct qede_rx_queue *rxq,
+			   struct eth_fast_path_rx_tpa_start_cqe *cqe)
+{
+	struct qede_agg_info *tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
+	struct eth_rx_bd *rx_bd_cons = qed_chain_consume(&rxq->rx_bd_ring);
+	struct eth_rx_bd *rx_bd_prod = qed_chain_produce(&rxq->rx_bd_ring);
+	struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
+	dma_addr_t mapping = tpa_info->replace_buf_mapping;
+	struct sw_rx_data *sw_rx_data_cons;
+	struct sw_rx_data *sw_rx_data_prod;
+	enum pkt_hash_types rxhash_type;
+	u32 rxhash;
+
+	sw_rx_data_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS_MAX];
+	sw_rx_data_prod = &rxq->sw_rx_ring[rxq->sw_rx_prod & NUM_RX_BDS_MAX];
+
+	/* Use pre-allocated replacement buffer - we can't release the agg.
+	 * start until its over and we don't want to risk allocation failing
+	 * here, so re-allocate when aggregation will be over.
+	 */
+	dma_unmap_addr_set(sw_rx_data_prod, mapping,
+			   dma_unmap_addr(replace_buf, mapping));
+
+	sw_rx_data_prod->data = replace_buf->data;
+	rx_bd_prod->addr.hi = cpu_to_le32(upper_32_bits(mapping));
+	rx_bd_prod->addr.lo = cpu_to_le32(lower_32_bits(mapping));
+	sw_rx_data_prod->page_offset = replace_buf->page_offset;
+
+	rxq->sw_rx_prod++;
+
+	/* move partial skb from cons to pool (don't unmap yet)
+	 * save mapping, incase we drop the packet later on.
+	 */
+	tpa_info->start_buf = *sw_rx_data_cons;
+	mapping = HILO_U64(le32_to_cpu(rx_bd_cons->addr.hi),
+			   le32_to_cpu(rx_bd_cons->addr.lo));
+
+	tpa_info->start_buf_mapping = mapping;
+	rxq->sw_rx_cons++;
+
+	/* set tpa state to start only if we are able to allocate skb
+	 * for this aggregation, otherwise mark as error and aggregation will
+	 * be dropped
+	 */
+	tpa_info->skb = netdev_alloc_skb(edev->ndev,
+					 le16_to_cpu(cqe->len_on_first_bd));
+	if (unlikely(!tpa_info->skb)) {
+		tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
+		return;
+	}
+
+	skb_put(tpa_info->skb, le16_to_cpu(cqe->len_on_first_bd));
+	memcpy(&tpa_info->start_cqe, cqe, sizeof(tpa_info->start_cqe));
+
+	/* Start filling in the aggregation info */
+	tpa_info->frag_id = 0;
+	tpa_info->agg_state = QEDE_AGG_STATE_START;
+
+	rxhash = qede_get_rxhash(edev, cqe->bitfields,
+				 cqe->rss_hash, &rxhash_type);
+	skb_set_hash(tpa_info->skb, rxhash, rxhash_type);
+	if ((le16_to_cpu(cqe->pars_flags.flags) >>
+	     PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT) &
+		    PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK)
+		tpa_info->vlan_tag = le16_to_cpu(cqe->vlan_tag);
+	else
+		tpa_info->vlan_tag = 0;
+
+	/* This is needed in order to enable forwarding support */
+	qede_set_gro_params(edev, tpa_info->skb, cqe);
+
+	if (likely(cqe->ext_bd_len_list[0]))
+		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
+				   le16_to_cpu(cqe->ext_bd_len_list[0]));
+
+	if (unlikely(cqe->ext_bd_len_list[1])) {
+		DP_ERR(edev,
+		       "Unlikely - got a TPA aggregation with more than one ext_bd_len_list entry in the TPA start\n");
+		tpa_info->agg_state = QEDE_AGG_STATE_ERROR;
+	}
+}
+
+static void qede_gro_ip_csum(struct sk_buff *skb)
+{
+	const struct iphdr *iph = ip_hdr(skb);
+	struct tcphdr *th;
+
+	skb_set_network_header(skb, 0);
+	skb_set_transport_header(skb, sizeof(struct iphdr));
+	th = tcp_hdr(skb);
+
+	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
+				  iph->saddr, iph->daddr, 0);
+
+	tcp_gro_complete(skb);
+}
+
+static void qede_gro_ipv6_csum(struct sk_buff *skb)
+{
+	struct ipv6hdr *iph = ipv6_hdr(skb);
+	struct tcphdr *th;
+
+	skb_set_network_header(skb, 0);
+	skb_set_transport_header(skb, sizeof(struct ipv6hdr));
+	th = tcp_hdr(skb);
+
+	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
+				  &iph->saddr, &iph->daddr, 0);
+	tcp_gro_complete(skb);
+}
+
+static void qede_gro_receive(struct qede_dev *edev,
+			     struct qede_fastpath *fp,
+			     struct sk_buff *skb,
+			     u16 vlan_tag)
+{
+	if (skb_shinfo(skb)->gso_size) {
+		switch (skb->protocol) {
+		case htons(ETH_P_IP):
+			qede_gro_ip_csum(skb);
+			break;
+		case htons(ETH_P_IPV6):
+			qede_gro_ipv6_csum(skb);
+			break;
+		default:
+			DP_ERR(edev,
+			       "Error: FW GRO supports only IPv4/IPv6, not 0x%04x\n",
+			       ntohs(skb->protocol));
+		}
+	}
+
+	skb_record_rx_queue(skb, fp->rss_id);
+	qede_skb_receive(edev, fp, skb, vlan_tag);
+}
+
+static inline void qede_tpa_cont(struct qede_dev *edev,
+				 struct qede_rx_queue *rxq,
+				 struct eth_fast_path_rx_tpa_cont_cqe *cqe)
+{
+	int i;
+
+	for (i = 0; cqe->len_list[i]; i++)
+		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
+				   le16_to_cpu(cqe->len_list[i]));
+
+	if (unlikely(i > 1))
+		DP_ERR(edev,
+		       "Strange - TPA cont with more than a single len_list entry\n");
+}
+
+static void qede_tpa_end(struct qede_dev *edev,
+			 struct qede_fastpath *fp,
+			 struct eth_fast_path_rx_tpa_end_cqe *cqe)
+{
+	struct qede_rx_queue *rxq = fp->rxq;
+	struct qede_agg_info *tpa_info;
+	struct sk_buff *skb;
+	int i;
+
+	tpa_info = &rxq->tpa_info[cqe->tpa_agg_index];
+	skb = tpa_info->skb;
+
+	for (i = 0; cqe->len_list[i]; i++)
+		qede_fill_frag_skb(edev, rxq, cqe->tpa_agg_index,
+				   le16_to_cpu(cqe->len_list[i]));
+	if (unlikely(i > 1))
+		DP_ERR(edev,
+		       "Strange - TPA emd with more than a single len_list entry\n");
+
+	if (unlikely(tpa_info->agg_state != QEDE_AGG_STATE_START))
+		goto err;
+
+	/* Sanity */
+	if (unlikely(cqe->num_of_bds != tpa_info->frag_id + 1))
+		DP_ERR(edev,
+		       "Strange - TPA had %02x BDs, but SKB has only %d frags\n",
+		       cqe->num_of_bds, tpa_info->frag_id);
+	if (unlikely(skb->len != le16_to_cpu(cqe->total_packet_len)))
+		DP_ERR(edev,
+		       "Strange - total packet len [cqe] is %4x but SKB has len %04x\n",
+		       le16_to_cpu(cqe->total_packet_len), skb->len);
+
+	memcpy(skb->data,
+	       page_address(tpa_info->start_buf.data) +
+		tpa_info->start_cqe.placement_offset +
+		tpa_info->start_buf.page_offset,
+	       le16_to_cpu(tpa_info->start_cqe.len_on_first_bd));
+
+	/* Recycle [mapped] start buffer for the next replacement */
+	tpa_info->replace_buf = tpa_info->start_buf;
+	tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
+
+	/* Finalize the SKB */
+	skb->protocol = eth_type_trans(skb, edev->ndev);
+	skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+	/* tcp_gro_complete() will copy NAPI_GRO_CB(skb)->count
+	 * to skb_shinfo(skb)->gso_segs
+	 */
+	NAPI_GRO_CB(skb)->count = le16_to_cpu(cqe->num_of_coalesced_segs);
+
+	qede_gro_receive(edev, fp, skb, tpa_info->vlan_tag);
+
+	tpa_info->agg_state = QEDE_AGG_STATE_NONE;
+
+	return;
+err:
+	/* The BD starting the aggregation is still mapped; Re-use it for
+	 * future aggregations [as replacement buffer]
+	 */
+	memcpy(&tpa_info->replace_buf, &tpa_info->start_buf,
+	       sizeof(struct sw_rx_data));
+	tpa_info->replace_buf_mapping = tpa_info->start_buf_mapping;
+	tpa_info->start_buf.data = NULL;
+	tpa_info->agg_state = QEDE_AGG_STATE_NONE;
+	dev_kfree_skb_any(tpa_info->skb);
+	tpa_info->skb = NULL;
+}
+
 static u8 qede_check_csum(u16 flag)
 {
 	u16 csum_flag = 0;
@@ -931,6 +1203,25 @@ static int qede_rx_int(struct qede_fastpath *fp, int budget)
 			goto next_cqe;
 		}
 
+		if (cqe_type != ETH_RX_CQE_TYPE_REGULAR) {
+			switch (cqe_type) {
+			case ETH_RX_CQE_TYPE_TPA_START:
+				qede_tpa_start(edev, rxq,
+					       &cqe->fast_path_tpa_start);
+				goto next_cqe;
+			case ETH_RX_CQE_TYPE_TPA_CONT:
+				qede_tpa_cont(edev, rxq,
+					      &cqe->fast_path_tpa_cont);
+				goto next_cqe;
+			case ETH_RX_CQE_TYPE_TPA_END:
+				qede_tpa_end(edev, fp,
+					     &cqe->fast_path_tpa_end);
+				goto next_rx_only;
+			default:
+				break;
+			}
+		}
+
 		/* Get the data from the SW ring */
 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS_MAX;
 		sw_rx_data = &rxq->sw_rx_ring[sw_rx_index];
@@ -1057,9 +1348,9 @@ static int qede_rx_int(struct qede_fastpath *fp, int budget)
 		qede_skb_receive(edev, fp, skb, le16_to_cpu(fp_cqe->vlan_tag));
 
 		qed_chain_consume(&rxq->rx_bd_ring);
-
 next_rx:
 		rxq->sw_rx_cons++;
+next_rx_only:
 		rx_pkt++;
 
 next_cqe: /* don't consume bd rx buffer */
@@ -1952,9 +2243,31 @@ static void qede_free_rx_buffers(struct qede_dev *edev,
 	}
 }
 
+static void qede_free_sge_mem(struct qede_dev *edev,
+			      struct qede_rx_queue *rxq) {
+	int i;
+
+	if (edev->gro_disable)
+		return;
+
+	for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
+		struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
+		struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
+
+		if (replace_buf) {
+			dma_unmap_page(&edev->pdev->dev,
+				       dma_unmap_addr(replace_buf, mapping),
+				       PAGE_SIZE, DMA_FROM_DEVICE);
+			__free_page(replace_buf->data);
+		}
+	}
+}
+
 static void qede_free_mem_rxq(struct qede_dev *edev,
 			      struct qede_rx_queue *rxq)
 {
+	qede_free_sge_mem(edev, rxq);
+
 	/* Free rx buffers */
 	qede_free_rx_buffers(edev, rxq);
 
@@ -2010,6 +2323,53 @@ static int qede_alloc_rx_buffer(struct qede_dev *edev,
 	return 0;
 }
 
+static int qede_alloc_sge_mem(struct qede_dev *edev,
+			      struct qede_rx_queue *rxq)
+{
+	dma_addr_t mapping;
+	int i;
+
+	if (edev->gro_disable)
+		return 0;
+
+	if (edev->ndev->mtu > PAGE_SIZE) {
+		edev->gro_disable = 1;
+		return 0;
+	}
+
+	for (i = 0; i < ETH_TPA_MAX_AGGS_NUM; i++) {
+		struct qede_agg_info *tpa_info = &rxq->tpa_info[i];
+		struct sw_rx_data *replace_buf = &tpa_info->replace_buf;
+
+		replace_buf->data = alloc_pages(GFP_ATOMIC, 0);
+		if (unlikely(!replace_buf->data)) {
+			DP_NOTICE(edev,
+				  "Failed to allocate TPA skb pool [replacement buffer]\n");
+			goto err;
+		}
+
+		mapping = dma_map_page(&edev->pdev->dev, replace_buf->data, 0,
+				       rxq->rx_buf_size, DMA_FROM_DEVICE);
+		if (unlikely(dma_mapping_error(&edev->pdev->dev, mapping))) {
+			DP_NOTICE(edev,
+				  "Failed to map TPA replacement buffer\n");
+			goto err;
+		}
+
+		dma_unmap_addr_set(replace_buf, mapping, mapping);
+		tpa_info->replace_buf.page_offset = 0;
+
+		tpa_info->replace_buf_mapping = mapping;
+		tpa_info->agg_state = QEDE_AGG_STATE_NONE;
+	}
+
+	return 0;
+err:
+	qede_free_sge_mem(edev, rxq);
+	edev->gro_disable = 1;
+	return -ENOMEM;
+}
+
 /* This function allocates all memory needed per Rx queue */
 static int qede_alloc_mem_rxq(struct qede_dev *edev,
 			      struct qede_rx_queue *rxq)
@@ -2071,6 +2431,8 @@ static int qede_alloc_mem_rxq(struct qede_dev *edev,
 			  num_allocated);
 	}
 
+	qede_alloc_sge_mem(edev, rxq);
+
 	return 0;
 
 err:
@@ -2233,6 +2595,8 @@ static void qede_init_fp(struct qede_dev *edev)
 		snprintf(fp->name, sizeof(fp->name), "%s-fp-%d",
 			 edev->ndev->name, rss_id);
 	}
+
+	edev->gro_disable = !(edev->ndev->features & NETIF_F_GRO);
 }
 
 static int qede_set_real_num_queues(struct qede_dev *edev)
@@ -2466,11 +2830,12 @@ static int qede_stop_queues(struct qede_dev *edev)
 static int qede_start_queues(struct qede_dev *edev)
 {
 	int rc, tc, i;
-	int vport_id = 0, drop_ttl0_flg = 1, vlan_removal_en = 1;
+	int vlan_removal_en = 1;
 	struct qed_dev *cdev = edev->cdev;
 	struct qed_update_vport_rss_params *rss_params = &edev->rss_params;
 	struct qed_update_vport_params vport_update_params;
 	struct qed_queue_start_common_params q_params;
+	struct qed_start_vport_params start = {0};
 
 	if (!edev->num_rss) {
 		DP_ERR(edev,
@@ -2478,10 +2843,13 @@ static int qede_start_queues(struct qede_dev *edev)
 		return -EINVAL;
 	}
 
-	rc = edev->ops->vport_start(cdev, vport_id,
-				    edev->ndev->mtu,
-				    drop_ttl0_flg,
-				    vlan_removal_en);
+	start.gro_enable = !edev->gro_disable;
+	start.mtu = edev->ndev->mtu;
+	start.vport_id = 0;
+	start.drop_ttl0 = true;
+	start.remove_inner_vlan = vlan_removal_en;
+
+	rc = edev->ops->vport_start(cdev, &start);
 
 	if (rc) {
 		DP_ERR(edev, "Start V-PORT failed %d\n", rc);
@@ -2490,7 +2858,7 @@ static int qede_start_queues(struct qede_dev *edev)
 
 	DP_VERBOSE(edev, NETIF_MSG_IFUP,
 		   "Start vport ramrod passed, vport_id = %d, MTU = %d, vlan_removal_en = %d\n",
-		   vport_id, edev->ndev->mtu + 0xe, vlan_removal_en);
+		   start.vport_id, edev->ndev->mtu + 0xe, vlan_removal_en);
 
 	for_each_rss(i) {
 		struct qede_fastpath *fp = &edev->fp_array[i];
@@ -2555,7 +2923,7 @@ static int qede_start_queues(struct qede_dev *edev)
 
 	/* Prepare and send the vport enable */
 	memset(&vport_update_params, 0, sizeof(vport_update_params));
-	vport_update_params.vport_id = vport_id;
+	vport_update_params.vport_id = start.vport_id;
 	vport_update_params.update_vport_active_flg = 1;
 	vport_update_params.vport_active_flg = 1;
 
diff --git a/include/linux/qed/qed_eth_if.h b/include/linux/qed/qed_eth_if.h
index e53b0ca49e41..e1d69834a11f 100644
--- a/include/linux/qed/qed_eth_if.h
+++ b/include/linux/qed/qed_eth_if.h
@@ -39,6 +39,14 @@ struct qed_update_vport_params {
 	struct qed_update_vport_rss_params rss_params;
 };
 
+struct qed_start_vport_params {
+	bool remove_inner_vlan;
+	bool gro_enable;
+	bool drop_ttl0;
+	u8 vport_id;
+	u16 mtu;
+};
+
 struct qed_stop_rxq_params {
 	u8 rss_id;
 	u8 rx_queue_id;
@@ -118,9 +126,7 @@ struct qed_eth_ops {
 			     void *cookie);
 
 	int (*vport_start)(struct qed_dev *cdev,
-			   u8 vport_id, u16 mtu,
-			   u8 drop_ttl0_flg,
-			   u8 inner_vlan_removal_en_flg);
+			   struct qed_start_vport_params *params);
 
 	int (*vport_stop)(struct qed_dev *cdev,
 			  u8 vport_id);