[dpdk-dev] [PATCH v2 16/26] net/ngbe: support RSS hash

[Jiawen Wu <jiawenwu@trustnetic.com>]

Support RSS hashing on Rx, and configuration of RSS hash computation.

Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
---
 doc/guides/nics/features/ngbe.ini      |   3 +
 doc/guides/nics/ngbe.rst               |   2 +
 doc/guides/rel_notes/release_21_11.rst |   1 +
 drivers/net/ngbe/meson.build           |   2 +
 drivers/net/ngbe/ngbe_ethdev.c         |  99 +++++++++++
 drivers/net/ngbe/ngbe_ethdev.h         |  27 +++
 drivers/net/ngbe/ngbe_rxtx.c           | 235 +++++++++++++++++++++++++
 7 files changed, 369 insertions(+)

diff --git a/doc/guides/nics/features/ngbe.ini b/doc/guides/nics/features/ngbe.ini
index 9ff1814265..6eb510f113 100644
--- a/doc/guides/nics/features/ngbe.ini
+++ b/doc/guides/nics/features/ngbe.ini
@@ -17,6 +17,9 @@ Promiscuous mode     = Y
 Allmulticast mode    = Y
 Unicast MAC filter   = Y
 Multicast MAC filter = Y
+RSS hash             = Y
+RSS key update       = Y
+RSS reta update      = Y
 VLAN filter          = Y
 CRC offload          = Y
 VLAN offload         = Y
diff --git a/doc/guides/nics/ngbe.rst b/doc/guides/nics/ngbe.rst
index e460daf616..1ca2a70074 100644
--- a/doc/guides/nics/ngbe.rst
+++ b/doc/guides/nics/ngbe.rst
@@ -11,6 +11,8 @@ for Wangxun 1 Gigabit Ethernet NICs.
 Features
 --------
 
+- Multiple queues for Tx and Rx
+- Receiver Side Scaling (RSS)
 - MAC/VLAN filtering
 - Packet type information
 - Checksum offload
diff --git a/doc/guides/rel_notes/release_21_11.rst b/doc/guides/rel_notes/release_21_11.rst
index d6e6a41666..3250f3b954 100644
--- a/doc/guides/rel_notes/release_21_11.rst
+++ b/doc/guides/rel_notes/release_21_11.rst
@@ -150,6 +150,7 @@ New Features
   * Added offloads and packet type on RxTx.
   * Added VLAN and MAC filters.
   * Added device basic statistics and extended stats.
+  * Added multi-queue and RSS.
 
 * **Added multi-process support for testpmd.**
 
diff --git a/drivers/net/ngbe/meson.build b/drivers/net/ngbe/meson.build
index 05f94fe7d6..c55e6c20e8 100644
--- a/drivers/net/ngbe/meson.build
+++ b/drivers/net/ngbe/meson.build
@@ -16,4 +16,6 @@ sources = files(
         'ngbe_rxtx.c',
 )
 
+deps += ['hash']
+
 includes += include_directories('base')
diff --git a/drivers/net/ngbe/ngbe_ethdev.c b/drivers/net/ngbe/ngbe_ethdev.c
index c72559d137..566a2fbd27 100644
--- a/drivers/net/ngbe/ngbe_ethdev.c
+++ b/drivers/net/ngbe/ngbe_ethdev.c
@@ -807,6 +807,9 @@ ngbe_dev_configure(struct rte_eth_dev *dev)
 
 	PMD_INIT_FUNC_TRACE();
 
+	if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
+		dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
+
 	/* set flag to update link status after init */
 	intr->flags |= NGBE_FLAG_NEED_LINK_UPDATE;
 
@@ -1033,6 +1036,7 @@ static int
 ngbe_dev_stop(struct rte_eth_dev *dev)
 {
 	struct rte_eth_link link;
+	struct ngbe_adapter *adapter = ngbe_dev_adapter(dev);
 	struct ngbe_hw *hw = ngbe_dev_hw(dev);
 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
@@ -1080,6 +1084,8 @@ ngbe_dev_stop(struct rte_eth_dev *dev)
 		intr_handle->intr_vec = NULL;
 	}
 
+	adapter->rss_reta_updated = 0;
+
 	hw->adapter_stopped = true;
 	dev->data->dev_started = 0;
 
@@ -1667,6 +1673,10 @@ ngbe_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 	dev_info->rx_desc_lim = rx_desc_lim;
 	dev_info->tx_desc_lim = tx_desc_lim;
 
+	dev_info->hash_key_size = NGBE_HKEY_MAX_INDEX * sizeof(uint32_t);
+	dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
+	dev_info->flow_type_rss_offloads = NGBE_RSS_OFFLOAD_ALL;
+
 	dev_info->speed_capa = ETH_LINK_SPEED_1G | ETH_LINK_SPEED_100M |
 				ETH_LINK_SPEED_10M;
 
@@ -2133,6 +2143,91 @@ ngbe_dev_interrupt_handler(void *param)
 	ngbe_dev_interrupt_action(dev);
 }
 
+int
+ngbe_dev_rss_reta_update(struct rte_eth_dev *dev,
+			  struct rte_eth_rss_reta_entry64 *reta_conf,
+			  uint16_t reta_size)
+{
+	uint8_t i, j, mask;
+	uint32_t reta;
+	uint16_t idx, shift;
+	struct ngbe_adapter *adapter = ngbe_dev_adapter(dev);
+	struct ngbe_hw *hw = ngbe_dev_hw(dev);
+
+	PMD_INIT_FUNC_TRACE();
+
+	if (!hw->is_pf) {
+		PMD_DRV_LOG(ERR, "RSS reta update is not supported on this "
+			"NIC.");
+		return -ENOTSUP;
+	}
+
+	if (reta_size != ETH_RSS_RETA_SIZE_128) {
+		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
+			"(%d) doesn't match the number hardware can supported "
+			"(%d)", reta_size, ETH_RSS_RETA_SIZE_128);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < reta_size; i += 4) {
+		idx = i / RTE_RETA_GROUP_SIZE;
+		shift = i % RTE_RETA_GROUP_SIZE;
+		mask = (uint8_t)RS64(reta_conf[idx].mask, shift, 0xF);
+		if (!mask)
+			continue;
+
+		reta = rd32a(hw, NGBE_REG_RSSTBL, i >> 2);
+		for (j = 0; j < 4; j++) {
+			if (RS8(mask, j, 0x1)) {
+				reta  &= ~(MS32(8 * j, 0xFF));
+				reta |= LS32(reta_conf[idx].reta[shift + j],
+						8 * j, 0xFF);
+			}
+		}
+		wr32a(hw, NGBE_REG_RSSTBL, i >> 2, reta);
+	}
+	adapter->rss_reta_updated = 1;
+
+	return 0;
+}
+
+int
+ngbe_dev_rss_reta_query(struct rte_eth_dev *dev,
+			 struct rte_eth_rss_reta_entry64 *reta_conf,
+			 uint16_t reta_size)
+{
+	struct ngbe_hw *hw = ngbe_dev_hw(dev);
+	uint8_t i, j, mask;
+	uint32_t reta;
+	uint16_t idx, shift;
+
+	PMD_INIT_FUNC_TRACE();
+
+	if (reta_size != ETH_RSS_RETA_SIZE_128) {
+		PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
+			"(%d) doesn't match the number hardware can supported "
+			"(%d)", reta_size, ETH_RSS_RETA_SIZE_128);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < reta_size; i += 4) {
+		idx = i / RTE_RETA_GROUP_SIZE;
+		shift = i % RTE_RETA_GROUP_SIZE;
+		mask = (uint8_t)RS64(reta_conf[idx].mask, shift, 0xF);
+		if (!mask)
+			continue;
+
+		reta = rd32a(hw, NGBE_REG_RSSTBL, i >> 2);
+		for (j = 0; j < 4; j++) {
+			if (RS8(mask, j, 0x1))
+				reta_conf[idx].reta[shift + j] =
+					(uint16_t)RS32(reta, 8 * j, 0xFF);
+		}
+	}
+
+	return 0;
+}
+
 static int
 ngbe_add_rar(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
 				uint32_t index, uint32_t pool)
@@ -2456,6 +2551,10 @@ static const struct eth_dev_ops ngbe_eth_dev_ops = {
 	.mac_addr_set               = ngbe_set_default_mac_addr,
 	.uc_hash_table_set          = ngbe_uc_hash_table_set,
 	.uc_all_hash_table_set      = ngbe_uc_all_hash_table_set,
+	.reta_update                = ngbe_dev_rss_reta_update,
+	.reta_query                 = ngbe_dev_rss_reta_query,
+	.rss_hash_update            = ngbe_dev_rss_hash_update,
+	.rss_hash_conf_get          = ngbe_dev_rss_hash_conf_get,
 	.set_mc_addr_list           = ngbe_dev_set_mc_addr_list,
 	.rx_burst_mode_get          = ngbe_rx_burst_mode_get,
 	.tx_burst_mode_get          = ngbe_tx_burst_mode_get,
diff --git a/drivers/net/ngbe/ngbe_ethdev.h b/drivers/net/ngbe/ngbe_ethdev.h
index e08f2e4029..f1bc2fee3b 100644
--- a/drivers/net/ngbe/ngbe_ethdev.h
+++ b/drivers/net/ngbe/ngbe_ethdev.h
@@ -17,6 +17,7 @@
 
 #define NGBE_VFTA_SIZE 128
 #define NGBE_VLAN_TAG_SIZE 4
+#define NGBE_HKEY_MAX_INDEX 10
 /*Default value of Max Rx Queue*/
 #define NGBE_MAX_RX_QUEUE_NUM	8
 
@@ -31,6 +32,17 @@
 /* The overhead from MTU to max frame size. */
 #define NGBE_ETH_OVERHEAD (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
 
+#define NGBE_RSS_OFFLOAD_ALL ( \
+	ETH_RSS_IPV4 | \
+	ETH_RSS_NONFRAG_IPV4_TCP | \
+	ETH_RSS_NONFRAG_IPV4_UDP | \
+	ETH_RSS_IPV6 | \
+	ETH_RSS_NONFRAG_IPV6_TCP | \
+	ETH_RSS_NONFRAG_IPV6_UDP | \
+	ETH_RSS_IPV6_EX | \
+	ETH_RSS_IPV6_TCP_EX | \
+	ETH_RSS_IPV6_UDP_EX)
+
 #define NGBE_MISC_VEC_ID               RTE_INTR_VEC_ZERO_OFFSET
 #define NGBE_RX_VEC_START              RTE_INTR_VEC_RXTX_OFFSET
 
@@ -77,6 +89,9 @@ struct ngbe_adapter {
 	struct ngbe_hwstrip        hwstrip;
 	struct ngbe_uta_info       uta_info;
 	bool                       rx_bulk_alloc_allowed;
+
+	/* For RSS reta table update */
+	uint8_t rss_reta_updated;
 };
 
 static inline struct ngbe_adapter *
@@ -185,6 +200,12 @@ uint16_t ngbe_xmit_pkts_simple(void *tx_queue, struct rte_mbuf **tx_pkts,
 uint16_t ngbe_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 		uint16_t nb_pkts);
 
+int ngbe_dev_rss_hash_update(struct rte_eth_dev *dev,
+			      struct rte_eth_rss_conf *rss_conf);
+
+int ngbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
+				struct rte_eth_rss_conf *rss_conf);
+
 void ngbe_set_ivar_map(struct ngbe_hw *hw, int8_t direction,
 			       uint8_t queue, uint8_t msix_vector);
 
@@ -230,6 +251,12 @@ const uint32_t *ngbe_dev_supported_ptypes_get(struct rte_eth_dev *dev);
 int ngbe_dev_set_mc_addr_list(struct rte_eth_dev *dev,
 				      struct rte_ether_addr *mc_addr_set,
 				      uint32_t nb_mc_addr);
+int ngbe_dev_rss_reta_update(struct rte_eth_dev *dev,
+			struct rte_eth_rss_reta_entry64 *reta_conf,
+			uint16_t reta_size);
+int ngbe_dev_rss_reta_query(struct rte_eth_dev *dev,
+			struct rte_eth_rss_reta_entry64 *reta_conf,
+			uint16_t reta_size);
 void ngbe_vlan_hw_strip_bitmap_set(struct rte_eth_dev *dev,
 		uint16_t queue, bool on);
 void ngbe_config_vlan_strip_on_all_queues(struct rte_eth_dev *dev,
diff --git a/drivers/net/ngbe/ngbe_rxtx.c b/drivers/net/ngbe/ngbe_rxtx.c
index f2bc90b7e7..834c16f173 100644
--- a/drivers/net/ngbe/ngbe_rxtx.c
+++ b/drivers/net/ngbe/ngbe_rxtx.c
@@ -898,6 +898,18 @@ ngbe_rxd_pkt_info_to_pkt_type(uint32_t pkt_info, uint16_t ptid_mask)
 	return ngbe_decode_ptype(ptid);
 }
 
+static inline uint64_t
+ngbe_rxd_pkt_info_to_pkt_flags(uint32_t pkt_info)
+{
+	static uint64_t ip_rss_types_map[16] __rte_cache_aligned = {
+		0, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH, PKT_RX_RSS_HASH,
+		0, PKT_RX_RSS_HASH, 0, PKT_RX_RSS_HASH,
+		PKT_RX_RSS_HASH, 0, 0, 0,
+		0, 0, 0,  PKT_RX_FDIR,
+	};
+	return ip_rss_types_map[NGBE_RXD_RSSTYPE(pkt_info)];
+}
+
 static inline uint64_t
 rx_desc_status_to_pkt_flags(uint32_t rx_status, uint64_t vlan_flags)
 {
@@ -1005,10 +1017,16 @@ ngbe_rx_scan_hw_ring(struct ngbe_rx_queue *rxq)
 			pkt_flags = rx_desc_status_to_pkt_flags(s[j],
 					rxq->vlan_flags);
 			pkt_flags |= rx_desc_error_to_pkt_flags(s[j]);
+			pkt_flags |=
+				ngbe_rxd_pkt_info_to_pkt_flags(pkt_info[j]);
 			mb->ol_flags = pkt_flags;
 			mb->packet_type =
 				ngbe_rxd_pkt_info_to_pkt_type(pkt_info[j],
 				NGBE_PTID_MASK);
+
+			if (likely(pkt_flags & PKT_RX_RSS_HASH))
+				mb->hash.rss =
+					rte_le_to_cpu_32(rxdp[j].qw0.dw1);
 		}
 
 		/* Move mbuf pointers from the S/W ring to the stage */
@@ -1299,6 +1317,7 @@ ngbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 		 *    - packet length,
 		 *    - Rx port identifier.
 		 * 2) integrate hardware offload data, if any:
+		 *    - RSS flag & hash,
 		 *    - IP checksum flag,
 		 *    - VLAN TCI, if any,
 		 *    - error flags.
@@ -1320,10 +1339,14 @@ ngbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 		pkt_flags = rx_desc_status_to_pkt_flags(staterr,
 					rxq->vlan_flags);
 		pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
+		pkt_flags |= ngbe_rxd_pkt_info_to_pkt_flags(pkt_info);
 		rxm->ol_flags = pkt_flags;
 		rxm->packet_type = ngbe_rxd_pkt_info_to_pkt_type(pkt_info,
 						       NGBE_PTID_MASK);
 
+		if (likely(pkt_flags & PKT_RX_RSS_HASH))
+			rxm->hash.rss = rte_le_to_cpu_32(rxd.qw0.dw1);
+
 		/*
 		 * Store the mbuf address into the next entry of the array
 		 * of returned packets.
@@ -1363,6 +1386,7 @@ ngbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
  * Fill the following info in the HEAD buffer of the Rx cluster:
  *    - RX port identifier
  *    - hardware offload data, if any:
+ *      - RSS flag & hash
  *      - IP checksum flag
  *      - VLAN TCI, if any
  *      - error flags
@@ -1386,9 +1410,13 @@ ngbe_fill_cluster_head_buf(struct rte_mbuf *head, struct ngbe_rx_desc *desc,
 	pkt_info = rte_le_to_cpu_32(desc->qw0.dw0);
 	pkt_flags = rx_desc_status_to_pkt_flags(staterr, rxq->vlan_flags);
 	pkt_flags |= rx_desc_error_to_pkt_flags(staterr);
+	pkt_flags |= ngbe_rxd_pkt_info_to_pkt_flags(pkt_info);
 	head->ol_flags = pkt_flags;
 	head->packet_type = ngbe_rxd_pkt_info_to_pkt_type(pkt_info,
 						NGBE_PTID_MASK);
+
+	if (likely(pkt_flags & PKT_RX_RSS_HASH))
+		head->hash.rss = rte_le_to_cpu_32(desc->qw0.dw1);
 }
 
 /**
@@ -2272,6 +2300,188 @@ ngbe_dev_free_queues(struct rte_eth_dev *dev)
 	dev->data->nb_tx_queues = 0;
 }
 
+/**
+ * Receive Side Scaling (RSS)
+ *
+ * Principles:
+ * The source and destination IP addresses of the IP header and the source
+ * and destination ports of TCP/UDP headers, if any, of received packets are
+ * hashed against a configurable random key to compute a 32-bit RSS hash result.
+ * The seven (7) LSBs of the 32-bit hash result are used as an index into a
+ * 128-entry redirection table (RETA).  Each entry of the RETA provides a 3-bit
+ * RSS output index which is used as the Rx queue index where to store the
+ * received packets.
+ * The following output is supplied in the Rx write-back descriptor:
+ *     - 32-bit result of the Microsoft RSS hash function,
+ *     - 4-bit RSS type field.
+ */
+
+/*
+ * Used as the default key.
+ */
+static uint8_t rss_intel_key[40] = {
+	0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
+	0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
+	0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
+	0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
+	0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
+};
+
+static void
+ngbe_rss_disable(struct rte_eth_dev *dev)
+{
+	struct ngbe_hw *hw = ngbe_dev_hw(dev);
+
+	wr32m(hw, NGBE_RACTL, NGBE_RACTL_RSSENA, 0);
+}
+
+int
+ngbe_dev_rss_hash_update(struct rte_eth_dev *dev,
+			  struct rte_eth_rss_conf *rss_conf)
+{
+	struct ngbe_hw *hw = ngbe_dev_hw(dev);
+	uint8_t  *hash_key;
+	uint32_t mrqc;
+	uint32_t rss_key;
+	uint64_t rss_hf;
+	uint16_t i;
+
+	if (!hw->is_pf) {
+		PMD_DRV_LOG(ERR, "RSS hash update is not supported on this "
+			"NIC.");
+		return -ENOTSUP;
+	}
+
+	hash_key = rss_conf->rss_key;
+	if (hash_key) {
+		/* Fill in RSS hash key */
+		for (i = 0; i < 10; i++) {
+			rss_key  = LS32(hash_key[(i * 4) + 0], 0, 0xFF);
+			rss_key |= LS32(hash_key[(i * 4) + 1], 8, 0xFF);
+			rss_key |= LS32(hash_key[(i * 4) + 2], 16, 0xFF);
+			rss_key |= LS32(hash_key[(i * 4) + 3], 24, 0xFF);
+			wr32a(hw, NGBE_REG_RSSKEY, i, rss_key);
+		}
+	}
+
+	/* Set configured hashing protocols */
+	rss_hf = rss_conf->rss_hf & NGBE_RSS_OFFLOAD_ALL;
+
+	mrqc = rd32(hw, NGBE_RACTL);
+	mrqc &= ~NGBE_RACTL_RSSMASK;
+	if (rss_hf & ETH_RSS_IPV4)
+		mrqc |= NGBE_RACTL_RSSIPV4;
+	if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
+		mrqc |= NGBE_RACTL_RSSIPV4TCP;
+	if (rss_hf & ETH_RSS_IPV6 ||
+	    rss_hf & ETH_RSS_IPV6_EX)
+		mrqc |= NGBE_RACTL_RSSIPV6;
+	if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP ||
+	    rss_hf & ETH_RSS_IPV6_TCP_EX)
+		mrqc |= NGBE_RACTL_RSSIPV6TCP;
+	if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
+		mrqc |= NGBE_RACTL_RSSIPV4UDP;
+	if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP ||
+	    rss_hf & ETH_RSS_IPV6_UDP_EX)
+		mrqc |= NGBE_RACTL_RSSIPV6UDP;
+
+	if (rss_hf)
+		mrqc |= NGBE_RACTL_RSSENA;
+	else
+		mrqc &= ~NGBE_RACTL_RSSENA;
+
+	wr32(hw, NGBE_RACTL, mrqc);
+
+	return 0;
+}
+
+int
+ngbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
+			    struct rte_eth_rss_conf *rss_conf)
+{
+	struct ngbe_hw *hw = ngbe_dev_hw(dev);
+	uint8_t *hash_key;
+	uint32_t mrqc;
+	uint32_t rss_key;
+	uint64_t rss_hf;
+	uint16_t i;
+
+	hash_key = rss_conf->rss_key;
+	if (hash_key) {
+		/* Return RSS hash key */
+		for (i = 0; i < 10; i++) {
+			rss_key = rd32a(hw, NGBE_REG_RSSKEY, i);
+			hash_key[(i * 4) + 0] = RS32(rss_key, 0, 0xFF);
+			hash_key[(i * 4) + 1] = RS32(rss_key, 8, 0xFF);
+			hash_key[(i * 4) + 2] = RS32(rss_key, 16, 0xFF);
+			hash_key[(i * 4) + 3] = RS32(rss_key, 24, 0xFF);
+		}
+	}
+
+	rss_hf = 0;
+
+	mrqc = rd32(hw, NGBE_RACTL);
+	if (mrqc & NGBE_RACTL_RSSIPV4)
+		rss_hf |= ETH_RSS_IPV4;
+	if (mrqc & NGBE_RACTL_RSSIPV4TCP)
+		rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
+	if (mrqc & NGBE_RACTL_RSSIPV6)
+		rss_hf |= ETH_RSS_IPV6 |
+			  ETH_RSS_IPV6_EX;
+	if (mrqc & NGBE_RACTL_RSSIPV6TCP)
+		rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP |
+			  ETH_RSS_IPV6_TCP_EX;
+	if (mrqc & NGBE_RACTL_RSSIPV4UDP)
+		rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
+	if (mrqc & NGBE_RACTL_RSSIPV6UDP)
+		rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP |
+			  ETH_RSS_IPV6_UDP_EX;
+	if (!(mrqc & NGBE_RACTL_RSSENA))
+		rss_hf = 0;
+
+	rss_hf &= NGBE_RSS_OFFLOAD_ALL;
+
+	rss_conf->rss_hf = rss_hf;
+	return 0;
+}
+
+static void
+ngbe_rss_configure(struct rte_eth_dev *dev)
+{
+	struct rte_eth_rss_conf rss_conf;
+	struct ngbe_adapter *adapter = ngbe_dev_adapter(dev);
+	struct ngbe_hw *hw = ngbe_dev_hw(dev);
+	uint32_t reta;
+	uint16_t i;
+	uint16_t j;
+
+	PMD_INIT_FUNC_TRACE();
+
+	/*
+	 * Fill in redirection table
+	 * The byte-swap is needed because NIC registers are in
+	 * little-endian order.
+	 */
+	if (adapter->rss_reta_updated == 0) {
+		reta = 0;
+		for (i = 0, j = 0; i < ETH_RSS_RETA_SIZE_128; i++, j++) {
+			if (j == dev->data->nb_rx_queues)
+				j = 0;
+			reta = (reta >> 8) | LS32(j, 24, 0xFF);
+			if ((i & 3) == 3)
+				wr32a(hw, NGBE_REG_RSSTBL, i >> 2, reta);
+		}
+	}
+	/*
+	 * Configure the RSS key and the RSS protocols used to compute
+	 * the RSS hash of input packets.
+	 */
+	rss_conf = dev->data->dev_conf.rx_adv_conf.rss_conf;
+	if (rss_conf.rss_key == NULL)
+		rss_conf.rss_key = rss_intel_key; /* Default hash key */
+	ngbe_dev_rss_hash_update(dev, &rss_conf);
+}
+
 void ngbe_configure_port(struct rte_eth_dev *dev)
 {
 	struct ngbe_hw *hw = ngbe_dev_hw(dev);
@@ -2340,6 +2550,24 @@ ngbe_alloc_rx_queue_mbufs(struct ngbe_rx_queue *rxq)
 	return 0;
 }
 
+static int
+ngbe_dev_mq_rx_configure(struct rte_eth_dev *dev)
+{
+	switch (dev->data->dev_conf.rxmode.mq_mode) {
+	case ETH_MQ_RX_RSS:
+		ngbe_rss_configure(dev);
+		break;
+
+	case ETH_MQ_RX_NONE:
+	default:
+		/* if mq_mode is none, disable rss mode.*/
+		ngbe_rss_disable(dev);
+		break;
+	}
+
+	return 0;
+}
+
 void
 ngbe_set_rx_function(struct rte_eth_dev *dev)
 {
@@ -2536,8 +2764,15 @@ ngbe_dev_rx_init(struct rte_eth_dev *dev)
 
 	if (rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER)
 		dev->data->scattered_rx = 1;
+
+	/*
+	 * Device configured with multiple RX queues.
+	 */
+	ngbe_dev_mq_rx_configure(dev);
+
 	/*
 	 * Setup the Checksum Register.
+	 * Disable Full-Packet Checksum which is mutually exclusive with RSS.
 	 * Enable IP/L4 checksum computation by hardware if requested to do so.
 	 */
 	rxcsum = rd32(hw, NGBE_PSRCTL);
-- 
2.21.0.windows.1