[PATCH v7 4/4] net/ice: support buffer split in Rx path

[Yuan Wang <yuanx.wang@intel.com>]

This patch adds support for protocol based buffer split in normal Rx
data paths. When the Rx queue is configured with specific protocol type,
packets received will be directly split into protocol header and
payload parts limitation of pmd. And the two parts will be
put into different mempools.

Currently, protocol based buffer split is not supported in vectorized
paths.

A new api ice_buffer_split_supported_hdr_ptypes_get() has been
introduced, it will return the supported header protocols of ice PMD
to app for splitting.

Signed-off-by: Yuan Wang <yuanx.wang@intel.com>
Signed-off-by: Xuan Ding <xuan.ding@intel.com>
Signed-off-by: Wenxuan Wu <wenxuanx.wu@intel.com>
Tested-by: Yaqi Tang <yaqi.tang@intel.com>
---
 doc/guides/rel_notes/release_22_11.rst |   4 +
 drivers/net/ice/ice_ethdev.c           |  55 +++++-
 drivers/net/ice/ice_rxtx.c             | 259 ++++++++++++++++++++++---
 drivers/net/ice/ice_rxtx.h             |  16 ++
 drivers/net/ice/ice_rxtx_vec_common.h  |   3 +
 5 files changed, 305 insertions(+), 32 deletions(-)

diff --git a/doc/guides/rel_notes/release_22_11.rst b/doc/guides/rel_notes/release_22_11.rst
index 510869c73a..3fa5377d96 100644
--- a/doc/guides/rel_notes/release_22_11.rst
+++ b/doc/guides/rel_notes/release_22_11.rst
@@ -108,6 +108,10 @@ New Features
     User can choose length or protocol header to configure buffer split
     according to NIC's capability.
 
+* **Updated Intel ice driver.**
+
+  * Added protocol based buffer split support in scalar path.
+
 
 Removed Items
 -------------
diff --git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
index 71302b03d8..709e7df408 100644
--- a/drivers/net/ice/ice_ethdev.c
+++ b/drivers/net/ice/ice_ethdev.c
@@ -159,6 +159,7 @@ static int ice_timesync_read_time(struct rte_eth_dev *dev,
 static int ice_timesync_write_time(struct rte_eth_dev *dev,
 				   const struct timespec *timestamp);
 static int ice_timesync_disable(struct rte_eth_dev *dev);
+static const uint32_t *ice_buffer_split_supported_hdr_ptypes_get(struct rte_eth_dev *dev);
 
 static const struct rte_pci_id pci_id_ice_map[] = {
 	{ RTE_PCI_DEVICE(ICE_INTEL_VENDOR_ID, ICE_DEV_ID_E823L_BACKPLANE) },
@@ -270,6 +271,7 @@ static const struct eth_dev_ops ice_eth_dev_ops = {
 	.timesync_write_time          = ice_timesync_write_time,
 	.timesync_disable             = ice_timesync_disable,
 	.tm_ops_get                   = ice_tm_ops_get,
+	.buffer_split_supported_hdr_ptypes_get = ice_buffer_split_supported_hdr_ptypes_get,
 };
 
 /* store statistics names and its offset in stats structure */
@@ -3797,7 +3799,8 @@ ice_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 			RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
 			RTE_ETH_RX_OFFLOAD_VLAN_EXTEND |
 			RTE_ETH_RX_OFFLOAD_RSS_HASH |
-			RTE_ETH_RX_OFFLOAD_TIMESTAMP;
+			RTE_ETH_RX_OFFLOAD_TIMESTAMP |
+			RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT;
 		dev_info->tx_offload_capa |=
 			RTE_ETH_TX_OFFLOAD_QINQ_INSERT |
 			RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
@@ -3809,7 +3812,7 @@ ice_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 		dev_info->flow_type_rss_offloads |= ICE_RSS_OFFLOAD_ALL;
 	}
 
-	dev_info->rx_queue_offload_capa = 0;
+	dev_info->rx_queue_offload_capa = RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT;
 	dev_info->tx_queue_offload_capa = RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
 
 	dev_info->reta_size = pf->hash_lut_size;
@@ -3878,6 +3881,11 @@ ice_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 	dev_info->default_rxportconf.ring_size = ICE_BUF_SIZE_MIN;
 	dev_info->default_txportconf.ring_size = ICE_BUF_SIZE_MIN;
 
+	dev_info->rx_seg_capa.max_nseg = ICE_RX_MAX_NSEG;
+	dev_info->rx_seg_capa.multi_pools = 1;
+	dev_info->rx_seg_capa.offset_allowed = 0;
+	dev_info->rx_seg_capa.offset_align_log2 = 0;
+
 	return 0;
 }
 
@@ -5932,6 +5940,49 @@ ice_timesync_disable(struct rte_eth_dev *dev)
 	return 0;
 }
 
+static const uint32_t *
+ice_buffer_split_supported_hdr_ptypes_get(struct rte_eth_dev *dev __rte_unused)
+{
+	/* Buffer split protocol header capability. */
+	static const uint32_t ptypes[] = {
+		/* Non tunneled */
+		RTE_PTYPE_L2_ETHER,
+		RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
+		RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_UDP,
+		RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_TCP,
+		RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_L4_SCTP,
+		RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
+		RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_UDP,
+		RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_TCP,
+		RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_L4_SCTP,
+
+		/* Tunneled */
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER,
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_INNER_L4_UDP,
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_INNER_L4_TCP,
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN | RTE_PTYPE_INNER_L4_SCTP,
+
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER,
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_INNER_L4_UDP,
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_INNER_L4_TCP,
+		RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN | RTE_PTYPE_INNER_L4_SCTP,
+
+		RTE_PTYPE_UNKNOWN
+	};
+
+	return ptypes;
+}
+
 static int
 ice_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 	      struct rte_pci_device *pci_dev)
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
index 5af7c0c8f6..36d1193cec 100644
--- a/drivers/net/ice/ice_rxtx.c
+++ b/drivers/net/ice/ice_rxtx.c
@@ -259,7 +259,6 @@ ice_program_hw_rx_queue(struct ice_rx_queue *rxq)
 	/* Set buffer size as the head split is disabled. */
 	buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
 			      RTE_PKTMBUF_HEADROOM);
-	rxq->rx_hdr_len = 0;
 	rxq->rx_buf_len = RTE_ALIGN(buf_size, (1 << ICE_RLAN_CTX_DBUF_S));
 	rxq->max_pkt_len =
 		RTE_MIN((uint32_t)ICE_SUPPORT_CHAIN_NUM * rxq->rx_buf_len,
@@ -288,11 +287,84 @@ ice_program_hw_rx_queue(struct ice_rx_queue *rxq)
 
 	memset(&rx_ctx, 0, sizeof(rx_ctx));
 
+	if (rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) {
+		uint32_t proto_hdr;
+		proto_hdr = rxq->rxseg[0].proto_hdr;
+
+		if (proto_hdr == RTE_PTYPE_UNKNOWN) {
+			PMD_DRV_LOG(ERR, "Buffer split protocol must be configured");
+			return -EINVAL;
+		}
+
+		switch (proto_hdr & RTE_PTYPE_L4_MASK) {
+		case RTE_PTYPE_L4_TCP:
+		case RTE_PTYPE_L4_UDP:
+			rx_ctx.dtype = ICE_RX_DTYPE_HEADER_SPLIT;
+			rx_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_SPLIT_TCP_UDP;
+			goto set_hsplit_finish;
+		case RTE_PTYPE_L4_SCTP:
+			rx_ctx.dtype = ICE_RX_DTYPE_HEADER_SPLIT;
+			rx_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_SPLIT_SCTP;
+			goto set_hsplit_finish;
+		}
+
+		switch (proto_hdr & RTE_PTYPE_L3_MASK) {
+		case RTE_PTYPE_L3_IPV4_EXT_UNKNOWN:
+		case RTE_PTYPE_L3_IPV6_EXT_UNKNOWN:
+			rx_ctx.dtype = ICE_RX_DTYPE_HEADER_SPLIT;
+			rx_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_SPLIT_IP;
+			goto set_hsplit_finish;
+		}
+
+		switch (proto_hdr & RTE_PTYPE_L2_MASK) {
+		case RTE_PTYPE_L2_ETHER:
+			rx_ctx.dtype = ICE_RX_DTYPE_HEADER_SPLIT;
+			rx_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_SPLIT_L2;
+			rx_ctx.hsplit_1 = ICE_RLAN_RX_HSPLIT_1_SPLIT_L2;
+			goto set_hsplit_finish;
+		}
+
+		switch (proto_hdr & RTE_PTYPE_INNER_L4_MASK) {
+		case RTE_PTYPE_INNER_L4_TCP:
+		case RTE_PTYPE_INNER_L4_UDP:
+			rx_ctx.dtype = ICE_RX_DTYPE_HEADER_SPLIT;
+			rx_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_SPLIT_TCP_UDP;
+			goto set_hsplit_finish;
+		case RTE_PTYPE_INNER_L4_SCTP:
+			rx_ctx.dtype = ICE_RX_DTYPE_HEADER_SPLIT;
+			rx_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_SPLIT_SCTP;
+			goto set_hsplit_finish;
+		}
+
+		switch (proto_hdr & RTE_PTYPE_INNER_L3_MASK) {
+		case RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN:
+		case RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN:
+			rx_ctx.dtype = ICE_RX_DTYPE_HEADER_SPLIT;
+			rx_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_SPLIT_IP;
+			goto set_hsplit_finish;
+		}
+
+		switch (proto_hdr & RTE_PTYPE_INNER_L2_MASK) {
+		case RTE_PTYPE_INNER_L2_ETHER:
+			rx_ctx.dtype = ICE_RX_DTYPE_HEADER_SPLIT;
+			rx_ctx.hsplit_0 = ICE_RLAN_RX_HSPLIT_0_SPLIT_L2;
+			goto set_hsplit_finish;
+		}
+
+		PMD_DRV_LOG(ERR, "Buffer split protocol is not supported");
+		return -EINVAL;
+
+set_hsplit_finish:
+		rxq->rx_hdr_len = ICE_RX_HDR_BUF_SIZE;
+	} else {
+		rxq->rx_hdr_len = 0;
+		rx_ctx.dtype = 0; /* No Protocol Based Buffer Split mode */
+	}
+
 	rx_ctx.base = rxq->rx_ring_dma / ICE_QUEUE_BASE_ADDR_UNIT;
 	rx_ctx.qlen = rxq->nb_rx_desc;
 	rx_ctx.dbuf = rxq->rx_buf_len >> ICE_RLAN_CTX_DBUF_S;
 	rx_ctx.hbuf = rxq->rx_hdr_len >> ICE_RLAN_CTX_HBUF_S;
-	rx_ctx.dtype = 0; /* No Header Split mode */
 #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
 	rx_ctx.dsize = 1; /* 32B descriptors */
 #endif
@@ -378,6 +450,7 @@ ice_alloc_rx_queue_mbufs(struct ice_rx_queue *rxq)
 
 	for (i = 0; i < rxq->nb_rx_desc; i++) {
 		volatile union ice_rx_flex_desc *rxd;
+		rxd = &rxq->rx_ring[i];
 		struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mp);
 
 		if (unlikely(!mbuf)) {
@@ -385,8 +458,6 @@ ice_alloc_rx_queue_mbufs(struct ice_rx_queue *rxq)
 			return -ENOMEM;
 		}
 
-		rte_mbuf_refcnt_set(mbuf, 1);
-		mbuf->next = NULL;
 		mbuf->data_off = RTE_PKTMBUF_HEADROOM;
 		mbuf->nb_segs = 1;
 		mbuf->port = rxq->port_id;
@@ -394,9 +465,32 @@ ice_alloc_rx_queue_mbufs(struct ice_rx_queue *rxq)
 		dma_addr =
 			rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
 
-		rxd = &rxq->rx_ring[i];
-		rxd->read.pkt_addr = dma_addr;
-		rxd->read.hdr_addr = 0;
+		if (!(rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT)) {
+			rte_mbuf_refcnt_set(mbuf, 1);
+			mbuf->next = NULL;
+			rxd->read.hdr_addr = 0;
+			rxd->read.pkt_addr = dma_addr;
+		} else {
+			struct rte_mbuf *mbuf_pay;
+			mbuf_pay = rte_mbuf_raw_alloc(rxq->rxseg[1].mp);
+			if (unlikely(!mbuf_pay)) {
+				PMD_DRV_LOG(ERR, "Failed to allocate payload mbuf for RX");
+				return -ENOMEM;
+			}
+
+			mbuf_pay->next = NULL;
+			mbuf_pay->data_off = RTE_PKTMBUF_HEADROOM;
+			mbuf_pay->nb_segs = 1;
+			mbuf_pay->port = rxq->port_id;
+			mbuf->next = mbuf_pay;
+
+			rxd->read.hdr_addr = dma_addr;
+			/* The LS bit should be set to zero regardless of
+			 * buffer split enablement.
+			 */
+			rxd->read.pkt_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf_pay));
+		}
+
 #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
 		rxd->read.rsvd1 = 0;
 		rxd->read.rsvd2 = 0;
@@ -420,14 +514,14 @@ _ice_rx_queue_release_mbufs(struct ice_rx_queue *rxq)
 
 	for (i = 0; i < rxq->nb_rx_desc; i++) {
 		if (rxq->sw_ring[i].mbuf) {
-			rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
+			rte_pktmbuf_free(rxq->sw_ring[i].mbuf);
 			rxq->sw_ring[i].mbuf = NULL;
 		}
 	}
 	if (rxq->rx_nb_avail == 0)
 		return;
 	for (i = 0; i < rxq->rx_nb_avail; i++)
-		rte_pktmbuf_free_seg(rxq->rx_stage[rxq->rx_next_avail + i]);
+		rte_pktmbuf_free(rxq->rx_stage[rxq->rx_next_avail + i]);
 
 	rxq->rx_nb_avail = 0;
 }
@@ -719,7 +813,7 @@ ice_fdir_program_hw_rx_queue(struct ice_rx_queue *rxq)
 	rx_ctx.qlen = rxq->nb_rx_desc;
 	rx_ctx.dbuf = rxq->rx_buf_len >> ICE_RLAN_CTX_DBUF_S;
 	rx_ctx.hbuf = rxq->rx_hdr_len >> ICE_RLAN_CTX_HBUF_S;
-	rx_ctx.dtype = 0; /* No Header Split mode */
+	rx_ctx.dtype = 0; /* No Buffer Split mode */
 	rx_ctx.dsize = 1; /* 32B descriptors */
 	rx_ctx.rxmax = ICE_ETH_MAX_LEN;
 	/* TPH: Transaction Layer Packet (TLP) processing hints */
@@ -1053,6 +1147,8 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
 	uint16_t len;
 	int use_def_burst_func = 1;
 	uint64_t offloads;
+	uint16_t n_seg = rx_conf->rx_nseg;
+	uint16_t i;
 
 	if (nb_desc % ICE_ALIGN_RING_DESC != 0 ||
 	    nb_desc > ICE_MAX_RING_DESC ||
@@ -1064,6 +1160,15 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
 
 	offloads = rx_conf->offloads | dev->data->dev_conf.rxmode.offloads;
 
+	if (mp)
+		n_seg = 1;
+
+	if (n_seg > 1 && !(offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT)) {
+		PMD_INIT_LOG(ERR, "port %u queue index %u split offload not configured",
+				dev->data->port_id, queue_idx);
+		return -EINVAL;
+	}
+
 	/* Free memory if needed */
 	if (dev->data->rx_queues[queue_idx]) {
 		ice_rx_queue_release(dev->data->rx_queues[queue_idx]);
@@ -1075,12 +1180,24 @@ ice_rx_queue_setup(struct rte_eth_dev *dev,
 				 sizeof(struct ice_rx_queue),
 				 RTE_CACHE_LINE_SIZE,
 				 socket_id);
+
 	if (!rxq) {
 		PMD_INIT_LOG(ERR, "Failed to allocate memory for "
 			     "rx queue data structure");
 		return -ENOMEM;
 	}
-	rxq->mp = mp;
+
+	rxq->rxseg_nb = n_seg;
+	if (n_seg > 1) {
+		for (i = 0; i < n_seg; i++)
+			memcpy(&rxq->rxseg[i], &rx_conf->rx_seg[i].split,
+				sizeof(struct rte_eth_rxseg_split));
+
+		rxq->mp = rxq->rxseg[0].mp;
+	} else {
+		rxq->mp = mp;
+	}
+
 	rxq->nb_rx_desc = nb_desc;
 	rxq->rx_free_thresh = rx_conf->rx_free_thresh;
 	rxq->queue_id = queue_idx;
@@ -1551,7 +1668,7 @@ ice_rx_scan_hw_ring(struct ice_rx_queue *rxq)
 	struct ice_rx_entry *rxep;
 	struct rte_mbuf *mb;
 	uint16_t stat_err0;
-	uint16_t pkt_len;
+	uint16_t pkt_len, hdr_len;
 	int32_t s[ICE_LOOK_AHEAD], nb_dd;
 	int32_t i, j, nb_rx = 0;
 	uint64_t pkt_flags = 0;
@@ -1606,6 +1723,27 @@ ice_rx_scan_hw_ring(struct ice_rx_queue *rxq)
 				   ICE_RX_FLX_DESC_PKT_LEN_M) - rxq->crc_len;
 			mb->data_len = pkt_len;
 			mb->pkt_len = pkt_len;
+
+			if (!(rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT)) {
+				pkt_len = (rte_le_to_cpu_16(rxdp[j].wb.pkt_len) &
+					ICE_RX_FLX_DESC_PKT_LEN_M) - rxq->crc_len;
+				mb->data_len = pkt_len;
+				mb->pkt_len = pkt_len;
+			} else {
+				mb->nb_segs = (uint16_t)(mb->nb_segs + mb->next->nb_segs);
+				mb->next->next = NULL;
+				hdr_len = rte_le_to_cpu_16(rxdp[j].wb.hdr_len_sph_flex_flags1) &
+						ICE_RX_FLEX_DESC_HEADER_LEN_M;
+				pkt_len = (rte_le_to_cpu_16(rxdp[j].wb.pkt_len) &
+					ICE_RX_FLX_DESC_PKT_LEN_M) - rxq->crc_len;
+				mb->data_len = hdr_len;
+				mb->pkt_len = hdr_len + pkt_len;
+				mb->next->data_len = pkt_len;
+#ifdef RTE_ETHDEV_DEBUG_RX
+				rte_pktmbuf_dump(stdout, mb, rte_pktmbuf_pkt_len(mb));
+#endif
+			}
+
 			mb->ol_flags = 0;
 			stat_err0 = rte_le_to_cpu_16(rxdp[j].wb.status_error0);
 			pkt_flags = ice_rxd_error_to_pkt_flags(stat_err0);
@@ -1697,7 +1835,9 @@ ice_rx_alloc_bufs(struct ice_rx_queue *rxq)
 	struct rte_mbuf *mb;
 	uint16_t alloc_idx, i;
 	uint64_t dma_addr;
-	int diag;
+	int diag, diag_pay;
+	uint64_t pay_addr;
+	struct rte_mbuf *mbufs_pay[rxq->rx_free_thresh];
 
 	/* Allocate buffers in bulk */
 	alloc_idx = (uint16_t)(rxq->rx_free_trigger -
@@ -1710,6 +1850,15 @@ ice_rx_alloc_bufs(struct ice_rx_queue *rxq)
 		return -ENOMEM;
 	}
 
+	if (rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT) {
+		diag_pay = rte_mempool_get_bulk(rxq->rxseg[1].mp,
+				(void *)mbufs_pay, rxq->rx_free_thresh);
+		if (unlikely(diag_pay != 0)) {
+			PMD_RX_LOG(ERR, "Failed to get payload mbufs in bulk");
+			return -ENOMEM;
+		}
+	}
+
 	rxdp = &rxq->rx_ring[alloc_idx];
 	for (i = 0; i < rxq->rx_free_thresh; i++) {
 		if (likely(i < (rxq->rx_free_thresh - 1)))
@@ -1718,13 +1867,21 @@ ice_rx_alloc_bufs(struct ice_rx_queue *rxq)
 
 		mb = rxep[i].mbuf;
 		rte_mbuf_refcnt_set(mb, 1);
-		mb->next = NULL;
 		mb->data_off = RTE_PKTMBUF_HEADROOM;
 		mb->nb_segs = 1;
 		mb->port = rxq->port_id;
 		dma_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(mb));
-		rxdp[i].read.hdr_addr = 0;
-		rxdp[i].read.pkt_addr = dma_addr;
+
+		if (!(rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT)) {
+			mb->next = NULL;
+			rxdp[i].read.hdr_addr = 0;
+			rxdp[i].read.pkt_addr = dma_addr;
+		} else {
+			mb->next = mbufs_pay[i];
+			pay_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbufs_pay[i]));
+			rxdp[i].read.hdr_addr = dma_addr;
+			rxdp[i].read.pkt_addr = pay_addr;
+		}
 	}
 
 	/* Update Rx tail register */
@@ -2333,11 +2490,13 @@ ice_recv_pkts(void *rx_queue,
 	struct ice_rx_entry *sw_ring = rxq->sw_ring;
 	struct ice_rx_entry *rxe;
 	struct rte_mbuf *nmb; /* new allocated mbuf */
+	struct rte_mbuf *nmb_pay; /* new allocated payload mbuf */
 	struct rte_mbuf *rxm; /* pointer to store old mbuf in SW ring */
 	uint16_t rx_id = rxq->rx_tail;
 	uint16_t nb_rx = 0;
 	uint16_t nb_hold = 0;
 	uint16_t rx_packet_len;
+	uint16_t rx_header_len;
 	uint16_t rx_stat_err0;
 	uint64_t dma_addr;
 	uint64_t pkt_flags;
@@ -2365,12 +2524,16 @@ ice_recv_pkts(void *rx_queue,
 		if (!(rx_stat_err0 & (1 << ICE_RX_FLEX_DESC_STATUS0_DD_S)))
 			break;
 
-		/* allocate mbuf */
+		if (rx_stat_err0 & (1 << ICE_RX_FLEX_DESC_STATUS0_HBO_S))
+			break;
+
+		/* allocate header mbuf */
 		nmb = rte_mbuf_raw_alloc(rxq->mp);
 		if (unlikely(!nmb)) {
 			rxq->vsi->adapter->pf.dev_data->rx_mbuf_alloc_failed++;
 			break;
 		}
+
 		rxd = *rxdp; /* copy descriptor in ring to temp variable*/
 
 		nb_hold++;
@@ -2383,24 +2546,60 @@ ice_recv_pkts(void *rx_queue,
 		dma_addr =
 			rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
 
-		/**
-		 * fill the read format of descriptor with physic address in
-		 * new allocated mbuf: nmb
-		 */
-		rxdp->read.hdr_addr = 0;
-		rxdp->read.pkt_addr = dma_addr;
+		if (!(rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT)) {
+			/**
+			 * fill the read format of descriptor with physic address in
+			 * new allocated mbuf: nmb
+			 */
+			rxdp->read.hdr_addr = 0;
+			rxdp->read.pkt_addr = dma_addr;
+		} else {
+			/* allocate payload mbuf */
+			nmb_pay = rte_mbuf_raw_alloc(rxq->rxseg[1].mp);
+			if (unlikely(!nmb_pay)) {
+				rxq->vsi->adapter->pf.dev_data->rx_mbuf_alloc_failed++;
+				break;
+			}
 
-		/* calculate rx_packet_len of the received pkt */
-		rx_packet_len = (rte_le_to_cpu_16(rxd.wb.pkt_len) &
-				 ICE_RX_FLX_DESC_PKT_LEN_M) - rxq->crc_len;
+			nmb->next = nmb_pay;
+			nmb_pay->next = NULL;
+
+			/**
+			 * fill the read format of descriptor with physic address in
+			 * new allocated mbuf: nmb
+			 */
+			rxdp->read.hdr_addr = dma_addr;
+			rxdp->read.pkt_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb_pay));
+		}
 
 		/* fill old mbuf with received descriptor: rxd */
 		rxm->data_off = RTE_PKTMBUF_HEADROOM;
 		rte_prefetch0(RTE_PTR_ADD(rxm->buf_addr, RTE_PKTMBUF_HEADROOM));
-		rxm->nb_segs = 1;
-		rxm->next = NULL;
-		rxm->pkt_len = rx_packet_len;
-		rxm->data_len = rx_packet_len;
+		if (!(rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT)) {
+			rxm->nb_segs = 1;
+			rxm->next = NULL;
+			/* calculate rx_packet_len of the received pkt */
+			rx_packet_len = (rte_le_to_cpu_16(rxd.wb.pkt_len) &
+					ICE_RX_FLX_DESC_PKT_LEN_M) - rxq->crc_len;
+			rxm->data_len = rx_packet_len;
+			rxm->pkt_len = rx_packet_len;
+		} else {
+			rxm->nb_segs = (uint16_t)(rxm->nb_segs + rxm->next->nb_segs);
+			rxm->next->next = NULL;
+			/* calculate rx_packet_len of the received pkt */
+			rx_header_len = rte_le_to_cpu_16(rxd.wb.hdr_len_sph_flex_flags1) &
+					ICE_RX_FLEX_DESC_HEADER_LEN_M;
+			rx_packet_len = (rte_le_to_cpu_16(rxd.wb.pkt_len) &
+					ICE_RX_FLX_DESC_PKT_LEN_M) - rxq->crc_len;
+			rxm->data_len = rx_header_len;
+			rxm->pkt_len = rx_header_len + rx_packet_len;
+			rxm->next->data_len = rx_packet_len;
+
+#ifdef RTE_ETHDEV_DEBUG_RX
+			rte_pktmbuf_dump(stdout, rxm, rte_pktmbuf_pkt_len(rxm));
+#endif
+		}
+
 		rxm->port = rxq->port_id;
 		rxm->packet_type = ptype_tbl[ICE_RX_FLEX_DESC_PTYPE_M &
 			rte_le_to_cpu_16(rxd.wb.ptype_flex_flags0)];
diff --git a/drivers/net/ice/ice_rxtx.h b/drivers/net/ice/ice_rxtx.h
index 6c08c175dc..0cfc3ca57d 100644
--- a/drivers/net/ice/ice_rxtx.h
+++ b/drivers/net/ice/ice_rxtx.h
@@ -16,6 +16,9 @@
 #define ICE_RX_MAX_BURST 32
 #define ICE_TX_MAX_BURST 32
 
+/* Maximal number of segments to split. */
+#define ICE_RX_MAX_NSEG 2
+
 #define ICE_CHK_Q_ENA_COUNT        100
 #define ICE_CHK_Q_ENA_INTERVAL_US  100
 
@@ -43,6 +46,11 @@
 extern uint64_t ice_timestamp_dynflag;
 extern int ice_timestamp_dynfield_offset;
 
+/* Max header size can be 2K - 64 bytes */
+#define ICE_RX_HDR_BUF_SIZE    (2048 - 64)
+
+#define ICE_HEADER_SPLIT_ENA   BIT(0)
+
 typedef void (*ice_rx_release_mbufs_t)(struct ice_rx_queue *rxq);
 typedef void (*ice_tx_release_mbufs_t)(struct ice_tx_queue *txq);
 typedef void (*ice_rxd_to_pkt_fields_t)(struct ice_rx_queue *rxq,
@@ -53,6 +61,12 @@ struct ice_rx_entry {
 	struct rte_mbuf *mbuf;
 };
 
+enum ice_rx_dtype {
+	ICE_RX_DTYPE_NO_SPLIT       = 0,
+	ICE_RX_DTYPE_HEADER_SPLIT   = 1,
+	ICE_RX_DTYPE_SPLIT_ALWAYS   = 2,
+};
+
 struct ice_rx_queue {
 	struct rte_mempool *mp; /* mbuf pool to populate RX ring */
 	volatile union ice_rx_flex_desc *rx_ring;/* RX ring virtual address */
@@ -99,6 +113,8 @@ struct ice_rx_queue {
 	uint32_t hw_time_high; /* high 32 bits of timestamp */
 	uint32_t hw_time_low; /* low 32 bits of timestamp */
 	uint64_t hw_time_update; /* SW time of HW record updating */
+	struct rte_eth_rxseg_split rxseg[ICE_RX_MAX_NSEG];
+	uint32_t rxseg_nb;
 };
 
 struct ice_tx_entry {
diff --git a/drivers/net/ice/ice_rxtx_vec_common.h b/drivers/net/ice/ice_rxtx_vec_common.h
index 2dd2d83650..eec6ea2134 100644
--- a/drivers/net/ice/ice_rxtx_vec_common.h
+++ b/drivers/net/ice/ice_rxtx_vec_common.h
@@ -291,6 +291,9 @@ ice_rx_vec_queue_default(struct ice_rx_queue *rxq)
 	if (rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP)
 		return -1;
 
+	if (rxq->offloads & RTE_ETH_RX_OFFLOAD_BUFFER_SPLIT)
+		return -1;
+
 	if (rxq->offloads & ICE_RX_VECTOR_OFFLOAD)
 		return ICE_VECTOR_OFFLOAD_PATH;
 
-- 
2.25.1