[PATCH v3 3/6] common/idpf: support single q scatter RX datapath

[Mingxia Liu]

This patch add single q recv scatter rx function.

Signed-off-by: Mingxia Liu <mingxia.liu at intel.com>
Signed-off-by: Wenjun Wu <wenjun1.wu at intel.com>
---
 drivers/common/idpf/idpf_common_rxtx.c | 134 +++++++++++++++++++++++++
 drivers/common/idpf/idpf_common_rxtx.h |   3 +
 drivers/common/idpf/version.map        |   1 +
 drivers/net/idpf/idpf_ethdev.c         |   3 +-
 drivers/net/idpf/idpf_rxtx.c           |  26 ++++-
 drivers/net/idpf/idpf_rxtx.h           |   2 +
 6 files changed, 166 insertions(+), 3 deletions(-)

diff --git a/drivers/common/idpf/idpf_common_rxtx.c b/drivers/common/idpf/idpf_common_rxtx.c
index 7a5dc3f04c..9dbf0f4764 100644
--- a/drivers/common/idpf/idpf_common_rxtx.c
+++ b/drivers/common/idpf/idpf_common_rxtx.c
@@ -1146,6 +1146,140 @@ idpf_singleq_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 	return nb_rx;
 }
 
+uint16_t
+idpf_singleq_recv_scatter_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+			       uint16_t nb_pkts)
+{
+	struct idpf_rx_queue *rxq = rx_queue;
+	volatile union virtchnl2_rx_desc *rx_ring = rxq->rx_ring;
+	volatile union virtchnl2_rx_desc *rxdp;
+	union virtchnl2_rx_desc rxd;
+	struct idpf_adapter *ad;
+	struct rte_mbuf *first_seg = rxq->pkt_first_seg;
+	struct rte_mbuf *last_seg = rxq->pkt_last_seg;
+	struct rte_mbuf *rxm;
+	struct rte_mbuf *nmb;
+	struct rte_eth_dev *dev;
+	const uint32_t *ptype_tbl = rxq->adapter->ptype_tbl;
+	uint16_t nb_hold = 0, nb_rx = 0;
+	uint16_t rx_id = rxq->rx_tail;
+	uint16_t rx_packet_len;
+	uint16_t rx_status0;
+	uint64_t pkt_flags;
+	uint64_t dma_addr;
+	uint64_t ts_ns;
+
+	ad = rxq->adapter;
+
+	if (unlikely(!rxq) || unlikely(!rxq->q_started))
+		return nb_rx;
+
+	while (nb_rx < nb_pkts) {
+		rxdp = &rx_ring[rx_id];
+		rx_status0 = rte_le_to_cpu_16(rxdp->flex_nic_wb.status_error0);
+
+		/* Check the DD bit first */
+		if (!(rx_status0 & (1 << VIRTCHNL2_RX_FLEX_DESC_STATUS0_DD_S)))
+			break;
+
+		nmb = rte_mbuf_raw_alloc(rxq->mp);
+		if (unlikely(!nmb)) {
+			rte_atomic64_inc(&(rxq->rx_stats.mbuf_alloc_failed));
+			RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
+			       "queue_id=%u", rxq->port_id, rxq->queue_id);
+			break;
+		}
+
+		rxd = *rxdp;
+
+		nb_hold++;
+		rxm = rxq->sw_ring[rx_id];
+		rxq->sw_ring[rx_id] = nmb;
+		rx_id++;
+		if (unlikely(rx_id == rxq->nb_rx_desc))
+			rx_id = 0;
+
+		/* Prefetch next mbuf */
+		rte_prefetch0(rxq->sw_ring[rx_id]);
+
+		/* When next RX descriptor is on a cache line boundary,
+		 * prefetch the next 4 RX descriptors and next 8 pointers
+		 * to mbufs.
+		 */
+		if ((rx_id & 0x3) == 0) {
+			rte_prefetch0(&rx_ring[rx_id]);
+			rte_prefetch0(rxq->sw_ring[rx_id]);
+		}
+		dma_addr =
+			rte_cpu_to_le_64(rte_mbuf_data_iova_default(nmb));
+		rxdp->read.hdr_addr = 0;
+		rxdp->read.pkt_addr = dma_addr;
+		rx_packet_len = (rte_cpu_to_le_16(rxd.flex_nic_wb.pkt_len) &
+				 VIRTCHNL2_RX_FLEX_DESC_PKT_LEN_M);
+		rxm->data_len = rx_packet_len;
+		rxm->data_off = RTE_PKTMBUF_HEADROOM;
+
+		/**
+		 * If this is the first buffer of the received packet, set the
+		 * pointer to the first mbuf of the packet and initialize its
+		 * context. Otherwise, update the total length and the number
+		 * of segments of the current scattered packet, and update the
+		 * pointer to the last mbuf of the current packet.
+		 */
+		if (!first_seg) {
+			first_seg = rxm;
+			first_seg->nb_segs = 1;
+			first_seg->pkt_len = rx_packet_len;
+		} else {
+			first_seg->pkt_len =
+				(uint16_t)(first_seg->pkt_len +
+					   rx_packet_len);
+			first_seg->nb_segs++;
+			last_seg->next = rxm;
+		}
+
+		if (!(rx_status0 & (1 << VIRTCHNL2_RX_FLEX_DESC_STATUS0_EOF_S))) {
+			last_seg = rxm;
+			continue;
+		}
+
+		rxm->next = NULL;
+
+		first_seg->port = rxq->port_id;
+		first_seg->ol_flags = 0;
+		pkt_flags = idpf_rxd_to_pkt_flags(rx_status0);
+		first_seg->packet_type =
+			ptype_tbl[(uint8_t)(rte_cpu_to_le_16(rxd.flex_nic_wb.ptype_flex_flags0) &
+				VIRTCHNL2_RX_FLEX_DESC_PTYPE_M)];
+
+		if (idpf_timestamp_dynflag > 0 &&
+		    (rxq->offloads & IDPF_RX_OFFLOAD_TIMESTAMP) != 0) {
+			/* timestamp */
+			ts_ns = idpf_tstamp_convert_32b_64b(ad,
+				rxq->hw_register_set,
+				rte_le_to_cpu_32(rxd.flex_nic_wb.flex_ts.ts_high));
+			rxq->hw_register_set = 0;
+			*RTE_MBUF_DYNFIELD(rxm,
+					   idpf_timestamp_dynfield_offset,
+					   rte_mbuf_timestamp_t *) = ts_ns;
+			first_seg->ol_flags |= idpf_timestamp_dynflag;
+		}
+
+		first_seg->ol_flags |= pkt_flags;
+		rte_prefetch0(RTE_PTR_ADD(first_seg->buf_addr,
+					  first_seg->data_off));
+		rx_pkts[nb_rx++] = first_seg;
+		first_seg = NULL;
+	}
+	rxq->rx_tail = rx_id;
+	rxq->pkt_first_seg = first_seg;
+	rxq->pkt_last_seg = last_seg;
+
+	idpf_update_rx_tail(rxq, nb_hold, rx_id);
+
+	return nb_rx;
+}
+
 static inline int
 idpf_xmit_cleanup(struct idpf_tx_queue *txq)
 {
diff --git a/drivers/common/idpf/idpf_common_rxtx.h b/drivers/common/idpf/idpf_common_rxtx.h
index 98f492a8c1..aac61ea2cb 100644
--- a/drivers/common/idpf/idpf_common_rxtx.h
+++ b/drivers/common/idpf/idpf_common_rxtx.h
@@ -260,6 +260,9 @@ __rte_internal
 uint16_t idpf_splitq_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 			       uint16_t nb_pkts);
 __rte_internal
+uint16_t idpf_singleq_recv_scatter_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
+			  uint16_t nb_pkts);
+__rte_internal
 uint16_t idpf_splitq_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 			       uint16_t nb_pkts);
 __rte_internal
diff --git a/drivers/common/idpf/version.map b/drivers/common/idpf/version.map
index 36a3a90d39..591af6b046 100644
--- a/drivers/common/idpf/version.map
+++ b/drivers/common/idpf/version.map
@@ -32,6 +32,7 @@ INTERNAL {
 	idpf_reset_split_tx_descq;
 	idpf_rx_queue_release;
 	idpf_singleq_recv_pkts;
+	idpf_singleq_recv_scatter_pkts;
 	idpf_singleq_recv_pkts_avx512;
 	idpf_singleq_rx_vec_setup;
 	idpf_splitq_rx_vec_setup;
diff --git a/drivers/net/idpf/idpf_ethdev.c b/drivers/net/idpf/idpf_ethdev.c
index 037cabb04e..2ab31792ba 100644
--- a/drivers/net/idpf/idpf_ethdev.c
+++ b/drivers/net/idpf/idpf_ethdev.c
@@ -119,7 +119,8 @@ idpf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 		RTE_ETH_RX_OFFLOAD_UDP_CKSUM            |
 		RTE_ETH_RX_OFFLOAD_TCP_CKSUM            |
 		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM     |
-		RTE_ETH_RX_OFFLOAD_TIMESTAMP;
+		RTE_ETH_RX_OFFLOAD_TIMESTAMP		|
+		RTE_ETH_RX_OFFLOAD_SCATTER;
 
 	dev_info->tx_offload_capa =
 		RTE_ETH_TX_OFFLOAD_IPV4_CKSUM		|
diff --git a/drivers/net/idpf/idpf_rxtx.c b/drivers/net/idpf/idpf_rxtx.c
index 6eeaab41cc..a865d14fea 100644
--- a/drivers/net/idpf/idpf_rxtx.c
+++ b/drivers/net/idpf/idpf_rxtx.c
@@ -503,6 +503,8 @@ int
 idpf_rx_queue_init(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
 	struct idpf_rx_queue *rxq;
+	uint16_t max_pkt_len;
+	uint32_t frame_size;
 	int err;
 
 	if (rx_queue_id >= dev->data->nb_rx_queues)
@@ -516,6 +518,17 @@ idpf_rx_queue_init(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		return -EINVAL;
 	}
 
+	frame_size = dev->data->mtu + IDPF_ETH_OVERHEAD;
+
+	max_pkt_len =
+	    RTE_MIN((uint32_t)IDPF_SUPPORT_CHAIN_NUM * rxq->rx_buf_len,
+		    frame_size);
+
+	rxq->max_pkt_len = max_pkt_len;
+	if ((dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_SCATTER) ||
+	    frame_size > rxq->rx_buf_len)
+		dev->data->scattered_rx = 1;
+
 	err = idpf_register_ts_mbuf(rxq);
 	if (err != 0) {
 		PMD_DRV_LOG(ERR, "fail to regidter timestamp mbuf %u",
@@ -801,13 +814,22 @@ idpf_set_rx_function(struct rte_eth_dev *dev)
 #endif /* CC_AVX512_SUPPORT */
 		}
 
+		if (dev->data->scattered_rx) {
+			dev->rx_pkt_burst = idpf_singleq_recv_scatter_pkts;
+			return;
+		}
 		dev->rx_pkt_burst = idpf_singleq_recv_pkts;
 	}
 #else
-	if (vport->rxq_model == VIRTCHNL2_QUEUE_MODEL_SPLIT)
+	if (vport->rxq_model == VIRTCHNL2_QUEUE_MODEL_SPLIT) {
 		dev->rx_pkt_burst = idpf_splitq_recv_pkts;
-	else
+	} else {
+		if (dev->data->scattered_rx) {
+			dev->rx_pkt_burst = idpf_singleq_recv_scatter_pkts;
+			return;
+		}
 		dev->rx_pkt_burst = idpf_singleq_recv_pkts;
+	}
 #endif /* RTE_ARCH_X86 */
 }
 
diff --git a/drivers/net/idpf/idpf_rxtx.h b/drivers/net/idpf/idpf_rxtx.h
index 3a5084dfd6..41a7495083 100644
--- a/drivers/net/idpf/idpf_rxtx.h
+++ b/drivers/net/idpf/idpf_rxtx.h
@@ -23,6 +23,8 @@
 #define IDPF_DEFAULT_TX_RS_THRESH	32
 #define IDPF_DEFAULT_TX_FREE_THRESH	32
 
+#define IDPF_SUPPORT_CHAIN_NUM 5
+
 int idpf_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 			uint16_t nb_desc, unsigned int socket_id,
 			const struct rte_eth_rxconf *rx_conf,
-- 
2.25.1