drivers: advertise kmod dependencies in pmdinfo
[dpdk.git] / drivers / net / thunderx / nicvf_ethdev.c
index 33d5fba..db03fa8 100644 (file)
 
 #include "nicvf_ethdev.h"
 #include "nicvf_rxtx.h"
+#include "nicvf_svf.h"
 #include "nicvf_logs.h"
 
+static void nicvf_dev_stop(struct rte_eth_dev *dev);
+static void nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup);
+static void nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic,
+                         bool cleanup);
+
 static inline int
 nicvf_atomic_write_link_status(struct rte_eth_dev *dev,
                               struct rte_eth_link *link)
@@ -99,31 +105,40 @@ nicvf_set_eth_link_status(struct nicvf *nic, struct rte_eth_link *link)
 static void
 nicvf_interrupt(void *arg)
 {
-       struct nicvf *nic = arg;
+       struct rte_eth_dev *dev = arg;
+       struct nicvf *nic = nicvf_pmd_priv(dev);
 
        if (nicvf_reg_poll_interrupts(nic) == NIC_MBOX_MSG_BGX_LINK_CHANGE) {
-               if (nic->eth_dev->data->dev_conf.intr_conf.lsc)
-                       nicvf_set_eth_link_status(nic,
-                                       &nic->eth_dev->data->dev_link);
-               _rte_eth_dev_callback_process(nic->eth_dev,
-                               RTE_ETH_EVENT_INTR_LSC);
+               if (dev->data->dev_conf.intr_conf.lsc)
+                       nicvf_set_eth_link_status(nic, &dev->data->dev_link);
+               _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
        }
 
        rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
-                               nicvf_interrupt, nic);
+                               nicvf_interrupt, dev);
+}
+
+static void
+nicvf_vf_interrupt(void *arg)
+{
+       struct nicvf *nic = arg;
+
+       nicvf_reg_poll_interrupts(nic);
+
+       rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
+                               nicvf_vf_interrupt, nic);
 }
 
 static int
-nicvf_periodic_alarm_start(struct nicvf *nic)
+nicvf_periodic_alarm_start(void (fn)(void *), void *arg)
 {
-       return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
-                                       nicvf_interrupt, nic);
+       return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000, fn, arg);
 }
 
 static int
-nicvf_periodic_alarm_stop(struct nicvf *nic)
+nicvf_periodic_alarm_stop(void (fn)(void *), void *arg)
 {
-       return rte_eal_alarm_cancel(nicvf_interrupt, nic);
+       return rte_eal_alarm_cancel(fn, arg);
 }
 
 /*
@@ -148,6 +163,7 @@ nicvf_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 {
        struct nicvf *nic = nicvf_pmd_priv(dev);
        uint32_t buffsz, frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
+       size_t i;
 
        PMD_INIT_FUNC_TRACE();
 
@@ -183,13 +199,11 @@ nicvf_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
        /* Update max frame size */
        dev->data->dev_conf.rxmode.max_rx_pkt_len = (uint32_t)frame_size;
        nic->mtu = mtu;
-       return 0;
-}
 
-static int
-nicvf_dev_get_reg_length(struct rte_eth_dev *dev  __rte_unused)
-{
-       return nicvf_reg_get_count();
+       for (i = 0; i < nic->sqs_count; i++)
+               nic->snicvf[i]->mtu = mtu;
+
+       return 0;
 }
 
 static int
@@ -198,8 +212,11 @@ nicvf_dev_get_regs(struct rte_eth_dev *dev, struct rte_dev_reg_info *regs)
        uint64_t *data = regs->data;
        struct nicvf *nic = nicvf_pmd_priv(dev);
 
-       if (data == NULL)
-               return -EINVAL;
+       if (data == NULL) {
+               regs->length = nicvf_reg_get_count();
+               regs->width = THUNDERX_REG_BYTES;
+               return 0;
+       }
 
        /* Support only full register dump */
        if ((regs->length == 0) ||
@@ -219,9 +236,15 @@ nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
        struct nicvf_hw_tx_qstats tx_qstats;
        struct nicvf_hw_stats port_stats;
        struct nicvf *nic = nicvf_pmd_priv(dev);
+       uint16_t rx_start, rx_end;
+       uint16_t tx_start, tx_end;
+       size_t i;
+
+       /* RX queue indices for the first VF */
+       nicvf_rx_range(dev, nic, &rx_start, &rx_end);
 
        /* Reading per RX ring stats */
-       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
                if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
                        break;
 
@@ -230,8 +253,11 @@ nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
                stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
        }
 
+       /* TX queue indices for the first VF */
+       nicvf_tx_range(dev, nic, &tx_start, &tx_end);
+
        /* Reading per TX ring stats */
-       for (qidx = 0; qidx < dev->data->nb_tx_queues; qidx++) {
+       for (qidx = tx_start; qidx <= tx_end; qidx++) {
                if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
                        break;
 
@@ -240,6 +266,40 @@ nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
                stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
        }
 
+       for (i = 0; i < nic->sqs_count; i++) {
+               struct nicvf *snic = nic->snicvf[i];
+
+               if (snic == NULL)
+                       break;
+
+               /* RX queue indices for a secondary VF */
+               nicvf_rx_range(dev, snic, &rx_start, &rx_end);
+
+               /* Reading per RX ring stats */
+               for (qidx = rx_start; qidx <= rx_end; qidx++) {
+                       if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
+                               break;
+
+                       nicvf_hw_get_rx_qstats(snic, &rx_qstats,
+                                              qidx % MAX_RCV_QUEUES_PER_QS);
+                       stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
+                       stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
+               }
+
+               /* TX queue indices for a secondary VF */
+               nicvf_tx_range(dev, snic, &tx_start, &tx_end);
+               /* Reading per TX ring stats */
+               for (qidx = tx_start; qidx <= tx_end; qidx++) {
+                       if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
+                               break;
+
+                       nicvf_hw_get_tx_qstats(snic, &tx_qstats,
+                                              qidx % MAX_SND_QUEUES_PER_QS);
+                       stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
+                       stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
+               }
+       }
+
        nicvf_hw_get_stats(nic, &port_stats);
        stats->ibytes = port_stats.rx_bytes;
        stats->ipackets = port_stats.rx_ucast_frames;
@@ -266,7 +326,7 @@ nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
        size_t copied;
        static uint32_t ptypes[32];
        struct nicvf *nic = nicvf_pmd_priv(dev);
-       static const uint32_t ptypes_pass1[] = {
+       static const uint32_t ptypes_common[] = {
                RTE_PTYPE_L3_IPV4,
                RTE_PTYPE_L3_IPV4_EXT,
                RTE_PTYPE_L3_IPV6,
@@ -275,7 +335,7 @@ nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
                RTE_PTYPE_L4_UDP,
                RTE_PTYPE_L4_FRAG,
        };
-       static const uint32_t ptypes_pass2[] = {
+       static const uint32_t ptypes_tunnel[] = {
                RTE_PTYPE_TUNNEL_GRE,
                RTE_PTYPE_TUNNEL_GENEVE,
                RTE_PTYPE_TUNNEL_VXLAN,
@@ -283,12 +343,12 @@ nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
        };
        static const uint32_t ptypes_end = RTE_PTYPE_UNKNOWN;
 
-       copied = sizeof(ptypes_pass1);
-       memcpy(ptypes, ptypes_pass1, copied);
-       if (nicvf_hw_version(nic) == NICVF_PASS2) {
-               memcpy((char *)ptypes + copied, ptypes_pass2,
-                       sizeof(ptypes_pass2));
-               copied += sizeof(ptypes_pass2);
+       copied = sizeof(ptypes_common);
+       memcpy(ptypes, ptypes_common, copied);
+       if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
+               memcpy((char *)ptypes + copied, ptypes_tunnel,
+                       sizeof(ptypes_tunnel));
+               copied += sizeof(ptypes_tunnel);
        }
 
        memcpy((char *)ptypes + copied, &ptypes_end, sizeof(ptypes_end));
@@ -305,13 +365,36 @@ nicvf_dev_stats_reset(struct rte_eth_dev *dev)
        int i;
        uint16_t rxqs = 0, txqs = 0;
        struct nicvf *nic = nicvf_pmd_priv(dev);
+       uint16_t rx_start, rx_end;
+       uint16_t tx_start, tx_end;
 
-       for (i = 0; i < dev->data->nb_rx_queues; i++)
+       /* Reset all primary nic counters */
+       nicvf_rx_range(dev, nic, &rx_start, &rx_end);
+       for (i = rx_start; i <= rx_end; i++)
                rxqs |= (0x3 << (i * 2));
-       for (i = 0; i < dev->data->nb_tx_queues; i++)
+
+       nicvf_tx_range(dev, nic, &tx_start, &tx_end);
+       for (i = tx_start; i <= tx_end; i++)
                txqs |= (0x3 << (i * 2));
 
        nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
+
+       /* Reset secondary nic queue counters */
+       for (i = 0; i < nic->sqs_count; i++) {
+               struct nicvf *snic = nic->snicvf[i];
+               if (snic == NULL)
+                       break;
+
+               nicvf_rx_range(dev, snic, &rx_start, &rx_end);
+               for (i = rx_start; i <= rx_end; i++)
+                       rxqs |= (0x3 << ((i % MAX_CMP_QUEUES_PER_QS) * 2));
+
+               nicvf_tx_range(dev, snic, &tx_start, &tx_end);
+               for (i = tx_start; i <= tx_end; i++)
+                       txqs |= (0x3 << ((i % MAX_SND_QUEUES_PER_QS) * 2));
+
+               nicvf_mbox_reset_stat_counters(snic, 0, 0, rxqs, txqs);
+       }
 }
 
 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
@@ -489,14 +572,15 @@ nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
 }
 
 static int
-nicvf_qset_cq_alloc(struct nicvf *nic, struct nicvf_rxq *rxq, uint16_t qidx,
-                   uint32_t desc_cnt)
+nicvf_qset_cq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
+                   struct nicvf_rxq *rxq, uint16_t qidx, uint32_t desc_cnt)
 {
        const struct rte_memzone *rz;
-       uint32_t ring_size = desc_cnt * sizeof(union cq_entry_t);
+       uint32_t ring_size = CMP_QUEUE_SZ_MAX * sizeof(union cq_entry_t);
 
-       rz = rte_eth_dma_zone_reserve(nic->eth_dev, "cq_ring", qidx, ring_size,
-                                       NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
+       rz = rte_eth_dma_zone_reserve(dev, "cq_ring",
+                                     nicvf_netdev_qidx(nic, qidx), ring_size,
+                                     NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
        if (rz == NULL) {
                PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
                return -ENOMEM;
@@ -512,14 +596,15 @@ nicvf_qset_cq_alloc(struct nicvf *nic, struct nicvf_rxq *rxq, uint16_t qidx,
 }
 
 static int
-nicvf_qset_sq_alloc(struct nicvf *nic,  struct nicvf_txq *sq, uint16_t qidx,
-                   uint32_t desc_cnt)
+nicvf_qset_sq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
+                   struct nicvf_txq *sq, uint16_t qidx, uint32_t desc_cnt)
 {
        const struct rte_memzone *rz;
-       uint32_t ring_size = desc_cnt * sizeof(union sq_entry_t);
+       uint32_t ring_size = SND_QUEUE_SZ_MAX * sizeof(union sq_entry_t);
 
-       rz = rte_eth_dma_zone_reserve(nic->eth_dev, "sq", qidx, ring_size,
-                               NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
+       rz = rte_eth_dma_zone_reserve(dev, "sq",
+                                     nicvf_netdev_qidx(nic, qidx), ring_size,
+                                     NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
        if (rz == NULL) {
                PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
                return -ENOMEM;
@@ -534,6 +619,89 @@ nicvf_qset_sq_alloc(struct nicvf *nic,  struct nicvf_txq *sq, uint16_t qidx,
        return 0;
 }
 
+static int
+nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
+                     uint32_t desc_cnt, uint32_t buffsz)
+{
+       struct nicvf_rbdr *rbdr;
+       const struct rte_memzone *rz;
+       uint32_t ring_size;
+
+       assert(nic->rbdr == NULL);
+       rbdr = rte_zmalloc_socket("rbdr", sizeof(struct nicvf_rbdr),
+                                 RTE_CACHE_LINE_SIZE, nic->node);
+       if (rbdr == NULL) {
+               PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr");
+               return -ENOMEM;
+       }
+
+       ring_size = sizeof(struct rbdr_entry_t) * RBDR_QUEUE_SZ_MAX;
+       rz = rte_eth_dma_zone_reserve(dev, "rbdr",
+                                     nicvf_netdev_qidx(nic, 0), ring_size,
+                                     NICVF_RBDR_BASE_ALIGN_BYTES, nic->node);
+       if (rz == NULL) {
+               PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr desc ring");
+               return -ENOMEM;
+       }
+
+       memset(rz->addr, 0, ring_size);
+
+       rbdr->phys = rz->phys_addr;
+       rbdr->tail = 0;
+       rbdr->next_tail = 0;
+       rbdr->desc = rz->addr;
+       rbdr->buffsz = buffsz;
+       rbdr->qlen_mask = desc_cnt - 1;
+       rbdr->rbdr_status =
+               nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_STATUS0;
+       rbdr->rbdr_door =
+               nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_DOOR;
+
+       nic->rbdr = rbdr;
+       return 0;
+}
+
+static void
+nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic,
+                       nicvf_phys_addr_t phy)
+{
+       uint16_t qidx;
+       void *obj;
+       struct nicvf_rxq *rxq;
+       uint16_t rx_start, rx_end;
+
+       /* Get queue ranges for this VF */
+       nicvf_rx_range(dev, nic, &rx_start, &rx_end);
+
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
+               rxq = dev->data->rx_queues[qidx];
+               if (rxq->precharge_cnt) {
+                       obj = (void *)nicvf_mbuff_phy2virt(phy,
+                                                          rxq->mbuf_phys_off);
+                       rte_mempool_put(rxq->pool, obj);
+                       rxq->precharge_cnt--;
+                       break;
+               }
+       }
+}
+
+static inline void
+nicvf_rbdr_release_mbufs(struct rte_eth_dev *dev, struct nicvf *nic)
+{
+       uint32_t qlen_mask, head;
+       struct rbdr_entry_t *entry;
+       struct nicvf_rbdr *rbdr = nic->rbdr;
+
+       qlen_mask = rbdr->qlen_mask;
+       head = rbdr->head;
+       while (head != rbdr->tail) {
+               entry = rbdr->desc + head;
+               nicvf_rbdr_release_mbuf(dev, nic, entry->full_addr);
+               head++;
+               head = head & qlen_mask;
+       }
+}
+
 static inline void
 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
 {
@@ -562,6 +730,135 @@ nicvf_tx_queue_reset(struct nicvf_txq *txq)
        txq->xmit_bufs = 0;
 }
 
+static inline int
+nicvf_vf_start_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
+                       uint16_t qidx)
+{
+       struct nicvf_txq *txq;
+       int ret;
+
+       assert(qidx < MAX_SND_QUEUES_PER_QS);
+
+       if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
+               RTE_ETH_QUEUE_STATE_STARTED)
+               return 0;
+
+       txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
+       txq->pool = NULL;
+       ret = nicvf_qset_sq_config(nic, qidx, txq);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to configure sq VF%d %d %d",
+                            nic->vf_id, qidx, ret);
+               goto config_sq_error;
+       }
+
+       dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
+               RTE_ETH_QUEUE_STATE_STARTED;
+       return ret;
+
+config_sq_error:
+       nicvf_qset_sq_reclaim(nic, qidx);
+       return ret;
+}
+
+static inline int
+nicvf_vf_stop_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
+                      uint16_t qidx)
+{
+       struct nicvf_txq *txq;
+       int ret;
+
+       assert(qidx < MAX_SND_QUEUES_PER_QS);
+
+       if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
+               RTE_ETH_QUEUE_STATE_STOPPED)
+               return 0;
+
+       ret = nicvf_qset_sq_reclaim(nic, qidx);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to reclaim sq VF%d %d %d",
+                            nic->vf_id, qidx, ret);
+
+       txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
+       nicvf_tx_queue_release_mbufs(txq);
+       nicvf_tx_queue_reset(txq);
+
+       dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
+               RTE_ETH_QUEUE_STATE_STOPPED;
+       return ret;
+}
+
+static inline int
+nicvf_configure_cpi(struct rte_eth_dev *dev)
+{
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+       uint16_t qidx, qcnt;
+       int ret;
+
+       /* Count started rx queues */
+       for (qidx = qcnt = 0; qidx < dev->data->nb_rx_queues; qidx++)
+               if (dev->data->rx_queue_state[qidx] ==
+                   RTE_ETH_QUEUE_STATE_STARTED)
+                       qcnt++;
+
+       nic->cpi_alg = CPI_ALG_NONE;
+       ret = nicvf_mbox_config_cpi(nic, qcnt);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to configure CPI %d", ret);
+
+       return ret;
+}
+
+static inline int
+nicvf_configure_rss(struct rte_eth_dev *dev)
+{
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+       uint64_t rsshf;
+       int ret = -EINVAL;
+
+       rsshf = nicvf_rss_ethdev_to_nic(nic,
+                       dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf);
+       PMD_DRV_LOG(INFO, "mode=%d rx_queues=%d loopback=%d rsshf=0x%" PRIx64,
+                   dev->data->dev_conf.rxmode.mq_mode,
+                   dev->data->nb_rx_queues,
+                   dev->data->dev_conf.lpbk_mode, rsshf);
+
+       if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_NONE)
+               ret = nicvf_rss_term(nic);
+       else if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
+               ret = nicvf_rss_config(nic, dev->data->nb_rx_queues, rsshf);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to configure RSS %d", ret);
+
+       return ret;
+}
+
+static int
+nicvf_configure_rss_reta(struct rte_eth_dev *dev)
+{
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+       unsigned int idx, qmap_size;
+       uint8_t qmap[RTE_MAX_QUEUES_PER_PORT];
+       uint8_t default_reta[NIC_MAX_RSS_IDR_TBL_SIZE];
+
+       if (nic->cpi_alg != CPI_ALG_NONE)
+               return -EINVAL;
+
+       /* Prepare queue map */
+       for (idx = 0, qmap_size = 0; idx < dev->data->nb_rx_queues; idx++) {
+               if (dev->data->rx_queue_state[idx] ==
+                               RTE_ETH_QUEUE_STATE_STARTED)
+                       qmap[qmap_size++] = idx;
+       }
+
+       /* Update default RSS RETA */
+       for (idx = 0; idx < NIC_MAX_RSS_IDR_TBL_SIZE; idx++)
+               default_reta[idx] = qmap[idx % qmap_size];
+
+       return nicvf_rss_reta_update(nic, default_reta,
+                                    NIC_MAX_RSS_IDR_TBL_SIZE);
+}
+
 static void
 nicvf_dev_tx_queue_release(void *sq)
 {
@@ -580,6 +877,48 @@ nicvf_dev_tx_queue_release(void *sq)
        }
 }
 
+static void
+nicvf_set_tx_function(struct rte_eth_dev *dev)
+{
+       struct nicvf_txq *txq;
+       size_t i;
+       bool multiseg = false;
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               txq = dev->data->tx_queues[i];
+               if ((txq->txq_flags & ETH_TXQ_FLAGS_NOMULTSEGS) == 0) {
+                       multiseg = true;
+                       break;
+               }
+       }
+
+       /* Use a simple Tx queue (no offloads, no multi segs) if possible */
+       if (multiseg) {
+               PMD_DRV_LOG(DEBUG, "Using multi-segment tx callback");
+               dev->tx_pkt_burst = nicvf_xmit_pkts_multiseg;
+       } else {
+               PMD_DRV_LOG(DEBUG, "Using single-segment tx callback");
+               dev->tx_pkt_burst = nicvf_xmit_pkts;
+       }
+
+       if (txq->pool_free == nicvf_single_pool_free_xmited_buffers)
+               PMD_DRV_LOG(DEBUG, "Using single-mempool tx free method");
+       else
+               PMD_DRV_LOG(DEBUG, "Using multi-mempool tx free method");
+}
+
+static void
+nicvf_set_rx_function(struct rte_eth_dev *dev)
+{
+       if (dev->data->scattered_rx) {
+               PMD_DRV_LOG(DEBUG, "Using multi-segment rx callback");
+               dev->rx_pkt_burst = nicvf_recv_pkts_multiseg;
+       } else {
+               PMD_DRV_LOG(DEBUG, "Using single-segment rx callback");
+               dev->rx_pkt_burst = nicvf_recv_pkts;
+       }
+}
+
 static int
 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
                         uint16_t nb_desc, unsigned int socket_id,
@@ -592,6 +931,11 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
 
        PMD_INIT_FUNC_TRACE();
 
+       if (qidx >= MAX_SND_QUEUES_PER_QS)
+               nic = nic->snicvf[qidx / MAX_SND_QUEUES_PER_QS - 1];
+
+       qidx = qidx % MAX_SND_QUEUES_PER_QS;
+
        /* Socket id check */
        if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
                PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
@@ -626,18 +970,20 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        }
 
        /* Free memory prior to re-allocation if needed. */
-       if (dev->data->tx_queues[qidx] != NULL) {
+       if (dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
                PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
-                               qidx);
-               nicvf_dev_tx_queue_release(dev->data->tx_queues[qidx]);
-               dev->data->tx_queues[qidx] = NULL;
+                               nicvf_netdev_qidx(nic, qidx));
+               nicvf_dev_tx_queue_release(
+                       dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)]);
+               dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
        }
 
        /* Allocating tx queue data structure */
        txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
                                        RTE_CACHE_LINE_SIZE, nic->node);
        if (txq == NULL) {
-               PMD_INIT_LOG(ERR, "Failed to allocate txq=%d", qidx);
+               PMD_INIT_LOG(ERR, "Failed to allocate txq=%d",
+                            nicvf_netdev_qidx(nic, qidx));
                return -ENOMEM;
        }
 
@@ -671,7 +1017,7 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
                return -ENOMEM;
        }
 
-       if (nicvf_qset_sq_alloc(nic, txq, qidx, nb_desc)) {
+       if (nicvf_qset_sq_alloc(dev, nic, txq, qidx, nb_desc)) {
                PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
                nicvf_dev_tx_queue_release(txq);
                return -ENOMEM;
@@ -680,13 +1026,45 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        nicvf_tx_queue_reset(txq);
 
        PMD_TX_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p phys=0x%" PRIx64,
-                       qidx, txq, nb_desc, txq->desc, txq->phys);
+                       nicvf_netdev_qidx(nic, qidx), txq, nb_desc, txq->desc,
+                       txq->phys);
 
-       dev->data->tx_queues[qidx] = txq;
-       dev->data->tx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
+       dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = txq;
+       dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
+               RTE_ETH_QUEUE_STATE_STOPPED;
        return 0;
 }
 
+static inline void
+nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
+{
+       uint32_t rxq_cnt;
+       uint32_t nb_pkts, released_pkts = 0;
+       uint32_t refill_cnt = 0;
+       struct rte_mbuf *rx_pkts[NICVF_MAX_RX_FREE_THRESH];
+
+       if (dev->rx_pkt_burst == NULL)
+               return;
+
+       while ((rxq_cnt = nicvf_dev_rx_queue_count(dev,
+                               nicvf_netdev_qidx(rxq->nic, rxq->queue_id)))) {
+               nb_pkts = dev->rx_pkt_burst(rxq, rx_pkts,
+                                       NICVF_MAX_RX_FREE_THRESH);
+               PMD_DRV_LOG(INFO, "nb_pkts=%d  rxq_cnt=%d", nb_pkts, rxq_cnt);
+               while (nb_pkts) {
+                       rte_pktmbuf_free_seg(rx_pkts[--nb_pkts]);
+                       released_pkts++;
+               }
+       }
+
+
+       refill_cnt += nicvf_dev_rbdr_refill(dev,
+                       nicvf_netdev_qidx(rxq->nic, rxq->queue_id));
+
+       PMD_DRV_LOG(INFO, "free_cnt=%d  refill_cnt=%d",
+                   released_pkts, refill_cnt);
+}
+
 static void
 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
 {
@@ -695,17 +1073,153 @@ nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
        rxq->recv_buffers = 0;
 }
 
+static inline int
+nicvf_vf_start_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
+                       uint16_t qidx)
+{
+       struct nicvf_rxq *rxq;
+       int ret;
+
+       assert(qidx < MAX_RCV_QUEUES_PER_QS);
+
+       if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
+               RTE_ETH_QUEUE_STATE_STARTED)
+               return 0;
+
+       /* Update rbdr pointer to all rxq */
+       rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
+       rxq->shared_rbdr = nic->rbdr;
+
+       ret = nicvf_qset_rq_config(nic, qidx, rxq);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to configure rq VF%d %d %d",
+                            nic->vf_id, qidx, ret);
+               goto config_rq_error;
+       }
+       ret = nicvf_qset_cq_config(nic, qidx, rxq);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to configure cq VF%d %d %d",
+                            nic->vf_id, qidx, ret);
+               goto config_cq_error;
+       }
+
+       dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
+               RTE_ETH_QUEUE_STATE_STARTED;
+       return 0;
+
+config_cq_error:
+       nicvf_qset_cq_reclaim(nic, qidx);
+config_rq_error:
+       nicvf_qset_rq_reclaim(nic, qidx);
+       return ret;
+}
+
+static inline int
+nicvf_vf_stop_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
+                      uint16_t qidx)
+{
+       struct nicvf_rxq *rxq;
+       int ret, other_error;
+
+       if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
+               RTE_ETH_QUEUE_STATE_STOPPED)
+               return 0;
+
+       ret = nicvf_qset_rq_reclaim(nic, qidx);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to reclaim rq VF%d %d %d",
+                            nic->vf_id, qidx, ret);
+
+       other_error = ret;
+       rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
+       nicvf_rx_queue_release_mbufs(dev, rxq);
+       nicvf_rx_queue_reset(rxq);
+
+       ret = nicvf_qset_cq_reclaim(nic, qidx);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to reclaim cq VF%d %d %d",
+                            nic->vf_id, qidx, ret);
+
+       other_error |= ret;
+       dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
+               RTE_ETH_QUEUE_STATE_STOPPED;
+       return other_error;
+}
+
 static void
 nicvf_dev_rx_queue_release(void *rx_queue)
 {
-       struct nicvf_rxq *rxq = rx_queue;
-
        PMD_INIT_FUNC_TRACE();
 
-       if (rxq)
-               rte_free(rxq);
+       rte_free(rx_queue);
+}
+
+static int
+nicvf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
+{
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+       int ret;
+
+       if (qidx >= MAX_RCV_QUEUES_PER_QS)
+               nic = nic->snicvf[(qidx / MAX_RCV_QUEUES_PER_QS - 1)];
+
+       qidx = qidx % MAX_RCV_QUEUES_PER_QS;
+
+       ret = nicvf_vf_start_rx_queue(dev, nic, qidx);
+       if (ret)
+               return ret;
+
+       ret = nicvf_configure_cpi(dev);
+       if (ret)
+               return ret;
+
+       return nicvf_configure_rss_reta(dev);
+}
+
+static int
+nicvf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
+{
+       int ret;
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+
+       if (qidx >= MAX_SND_QUEUES_PER_QS)
+               nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
+
+       qidx = qidx % MAX_RCV_QUEUES_PER_QS;
+
+       ret = nicvf_vf_stop_rx_queue(dev, nic, qidx);
+       ret |= nicvf_configure_cpi(dev);
+       ret |= nicvf_configure_rss_reta(dev);
+       return ret;
+}
+
+static int
+nicvf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
+{
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+
+       if (qidx >= MAX_SND_QUEUES_PER_QS)
+               nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
+
+       qidx = qidx % MAX_SND_QUEUES_PER_QS;
+
+       return nicvf_vf_start_tx_queue(dev, nic, qidx);
 }
 
+static int
+nicvf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
+{
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+
+       if (qidx >= MAX_SND_QUEUES_PER_QS)
+               nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
+
+       qidx = qidx % MAX_SND_QUEUES_PER_QS;
+
+       return nicvf_vf_stop_tx_queue(dev, nic, qidx);
+}
+
+
 static int
 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
                         uint16_t nb_desc, unsigned int socket_id,
@@ -718,14 +1232,25 @@ nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
 
        PMD_INIT_FUNC_TRACE();
 
+       if (qidx >= MAX_RCV_QUEUES_PER_QS)
+               nic = nic->snicvf[qidx / MAX_RCV_QUEUES_PER_QS - 1];
+
+       qidx = qidx % MAX_RCV_QUEUES_PER_QS;
+
        /* Socket id check */
        if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
                PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
                socket_id, nic->node);
 
-       /* Mempool memory should be contiguous */
+       /* Mempool memory must be contiguous, so must be one memory segment*/
        if (mp->nb_mem_chunks != 1) {
-               PMD_INIT_LOG(ERR, "Non contiguous mempool, check huge page sz");
+               PMD_INIT_LOG(ERR, "Non-contiguous mempool, add more huge pages");
+               return -EINVAL;
+       }
+
+       /* Mempool memory must be physically contiguous */
+       if (mp->flags & MEMPOOL_F_NO_PHYS_CONTIG) {
+               PMD_INIT_LOG(ERR, "Mempool memory must be physically contiguous");
                return -EINVAL;
        }
 
@@ -754,18 +1279,20 @@ nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        }
 
        /* Free memory prior to re-allocation if needed */
-       if (dev->data->rx_queues[qidx] != NULL) {
+       if (dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
                PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
-                               qidx);
-               nicvf_dev_rx_queue_release(dev->data->rx_queues[qidx]);
-               dev->data->rx_queues[qidx] = NULL;
+                               nicvf_netdev_qidx(nic, qidx));
+               nicvf_dev_rx_queue_release(
+                       dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)]);
+               dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
        }
 
        /* Allocate rxq memory */
        rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
                                        RTE_CACHE_LINE_SIZE, nic->node);
        if (rxq == NULL) {
-               PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d", qidx);
+               PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d",
+                            nicvf_netdev_qidx(nic, qidx));
                return -ENOMEM;
        }
 
@@ -778,10 +1305,15 @@ nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
        rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
        rxq->precharge_cnt = 0;
-       rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
+
+       if (nicvf_hw_cap(nic) & NICVF_CAP_CQE_RX2)
+               rxq->rbptr_offset = NICVF_CQE_RX2_RBPTR_WORD;
+       else
+               rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
+
 
        /* Alloc completion queue */
-       if (nicvf_qset_cq_alloc(nic, rxq, rxq->queue_id, nb_desc)) {
+       if (nicvf_qset_cq_alloc(dev, nic, rxq, rxq->queue_id, nb_desc)) {
                PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
                nicvf_dev_rx_queue_release(rxq);
                return -ENOMEM;
@@ -790,11 +1322,12 @@ nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
        nicvf_rx_queue_reset(rxq);
 
        PMD_RX_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d) phy=%" PRIx64,
-                       qidx, rxq, mp->name, nb_desc,
-                       rte_mempool_count(mp), rxq->phys);
+                       nicvf_netdev_qidx(nic, qidx), rxq, mp->name, nb_desc,
+                       rte_mempool_avail_count(mp), rxq->phys);
 
-       dev->data->rx_queues[qidx] = rxq;
-       dev->data->rx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
+       dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = rxq;
+       dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
+               RTE_ETH_QUEUE_STATE_STOPPED;
        return 0;
 }
 
@@ -807,8 +1340,10 @@ nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 
        dev_info->min_rx_bufsize = ETHER_MIN_MTU;
        dev_info->max_rx_pktlen = NIC_HW_MAX_FRS;
-       dev_info->max_rx_queues = (uint16_t)MAX_RCV_QUEUES_PER_QS;
-       dev_info->max_tx_queues = (uint16_t)MAX_SND_QUEUES_PER_QS;
+       dev_info->max_rx_queues =
+                       (uint16_t)MAX_RCV_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
+       dev_info->max_tx_queues =
+                       (uint16_t)MAX_SND_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
        dev_info->max_mac_addrs = 1;
        dev_info->max_vfs = dev->pci_dev->max_vfs;
 
@@ -842,13 +1377,460 @@ nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
        };
 }
 
+static nicvf_phys_addr_t
+rbdr_rte_mempool_get(void *dev, void *opaque)
+{
+       uint16_t qidx;
+       uintptr_t mbuf;
+       struct nicvf_rxq *rxq;
+       struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
+       struct nicvf *nic = (struct nicvf *)opaque;
+       uint16_t rx_start, rx_end;
+
+       /* Get queue ranges for this VF */
+       nicvf_rx_range(eth_dev, nic, &rx_start, &rx_end);
+
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
+               rxq = eth_dev->data->rx_queues[qidx];
+               /* Maintain equal buffer count across all pools */
+               if (rxq->precharge_cnt >= rxq->qlen_mask)
+                       continue;
+               rxq->precharge_cnt++;
+               mbuf = (uintptr_t)rte_pktmbuf_alloc(rxq->pool);
+               if (mbuf)
+                       return nicvf_mbuff_virt2phy(mbuf, rxq->mbuf_phys_off);
+       }
+       return 0;
+}
+
+static int
+nicvf_vf_start(struct rte_eth_dev *dev, struct nicvf *nic, uint32_t rbdrsz)
+{
+       int ret;
+       uint16_t qidx;
+       uint32_t total_rxq_desc, nb_rbdr_desc, exp_buffs;
+       uint64_t mbuf_phys_off = 0;
+       struct nicvf_rxq *rxq;
+       struct rte_mbuf *mbuf;
+       uint16_t rx_start, rx_end;
+       uint16_t tx_start, tx_end;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* Userspace process exited without proper shutdown in last run */
+       if (nicvf_qset_rbdr_active(nic, 0))
+               nicvf_vf_stop(dev, nic, false);
+
+       /* Get queue ranges for this VF */
+       nicvf_rx_range(dev, nic, &rx_start, &rx_end);
+
+       /*
+        * Thunderx nicvf PMD can support more than one pool per port only when
+        * 1) Data payload size is same across all the pools in given port
+        * AND
+        * 2) All mbuffs in the pools are from the same hugepage
+        * AND
+        * 3) Mbuff metadata size is same across all the pools in given port
+        *
+        * This is to support existing application that uses multiple pool/port.
+        * But, the purpose of using multipool for QoS will not be addressed.
+        *
+        */
+
+       /* Validate mempool attributes */
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
+               rxq = dev->data->rx_queues[qidx];
+               rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
+               mbuf = rte_pktmbuf_alloc(rxq->pool);
+               if (mbuf == NULL) {
+                       PMD_INIT_LOG(ERR, "Failed allocate mbuf VF%d qid=%d "
+                                    "pool=%s",
+                                    nic->vf_id, qidx, rxq->pool->name);
+                       return -ENOMEM;
+               }
+               rxq->mbuf_phys_off -= nicvf_mbuff_meta_length(mbuf);
+               rxq->mbuf_phys_off -= RTE_PKTMBUF_HEADROOM;
+               rte_pktmbuf_free(mbuf);
+
+               if (mbuf_phys_off == 0)
+                       mbuf_phys_off = rxq->mbuf_phys_off;
+               if (mbuf_phys_off != rxq->mbuf_phys_off) {
+                       PMD_INIT_LOG(ERR, "pool params not same,%s VF%d %"
+                                    PRIx64, rxq->pool->name, nic->vf_id,
+                                    mbuf_phys_off);
+                       return -EINVAL;
+               }
+       }
+
+       /* Check the level of buffers in the pool */
+       total_rxq_desc = 0;
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
+               rxq = dev->data->rx_queues[qidx];
+               /* Count total numbers of rxq descs */
+               total_rxq_desc += rxq->qlen_mask + 1;
+               exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
+               exp_buffs *= dev->data->nb_rx_queues;
+               if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
+                       PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
+                                    rxq->pool->name,
+                                    rte_mempool_avail_count(rxq->pool),
+                                    exp_buffs);
+                       return -ENOENT;
+               }
+       }
+
+       /* Check RBDR desc overflow */
+       ret = nicvf_qsize_rbdr_roundup(total_rxq_desc);
+       if (ret == 0) {
+               PMD_INIT_LOG(ERR, "Reached RBDR desc limit, reduce nr desc "
+                            "VF%d", nic->vf_id);
+               return -ENOMEM;
+       }
+
+       /* Enable qset */
+       ret = nicvf_qset_config(nic);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to enable qset %d VF%d", ret,
+                            nic->vf_id);
+               return ret;
+       }
+
+       /* Allocate RBDR and RBDR ring desc */
+       nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
+       ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc "
+                            "VF%d", nic->vf_id);
+               goto qset_reclaim;
+       }
+
+       /* Enable and configure RBDR registers */
+       ret = nicvf_qset_rbdr_config(nic, 0);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to configure rbdr %d VF%d", ret,
+                            nic->vf_id);
+               goto qset_rbdr_free;
+       }
+
+       /* Fill rte_mempool buffers in RBDR pool and precharge it */
+       ret = nicvf_qset_rbdr_precharge(dev, nic, 0, rbdr_rte_mempool_get,
+                                       total_rxq_desc);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to fill rbdr %d VF%d", ret,
+                            nic->vf_id);
+               goto qset_rbdr_reclaim;
+       }
+
+       PMD_DRV_LOG(INFO, "Filled %d out of %d entries in RBDR VF%d",
+                    nic->rbdr->tail, nb_rbdr_desc, nic->vf_id);
+
+       /* Configure VLAN Strip */
+       nicvf_vlan_hw_strip(nic, dev->data->dev_conf.rxmode.hw_vlan_strip);
+
+       /* Based on the packet type(IPv4 or IPv6), the nicvf HW aligns L3 data
+        * to the 64bit memory address.
+        * The alignment creates a hole in mbuf(between the end of headroom and
+        * packet data start). The new revision of the HW provides an option to
+        * disable the L3 alignment feature and make mbuf layout looks
+        * more like other NICs. For better application compatibility, disabling
+        * l3 alignment feature on the hardware revisions it supports
+        */
+       nicvf_apad_config(nic, false);
+
+       /* Get queue ranges for this VF */
+       nicvf_tx_range(dev, nic, &tx_start, &tx_end);
+
+       /* Configure TX queues */
+       for (qidx = tx_start; qidx <= tx_end; qidx++) {
+               ret = nicvf_vf_start_tx_queue(dev, nic,
+                       qidx % MAX_SND_QUEUES_PER_QS);
+               if (ret)
+                       goto start_txq_error;
+       }
+
+       /* Configure RX queues */
+       for (qidx = rx_start; qidx <= rx_end; qidx++) {
+               ret = nicvf_vf_start_rx_queue(dev, nic,
+                       qidx % MAX_RCV_QUEUES_PER_QS);
+               if (ret)
+                       goto start_rxq_error;
+       }
+
+       if (!nic->sqs_mode) {
+               /* Configure CPI algorithm */
+               ret = nicvf_configure_cpi(dev);
+               if (ret)
+                       goto start_txq_error;
+
+               ret = nicvf_mbox_get_rss_size(nic);
+               if (ret) {
+                       PMD_INIT_LOG(ERR, "Failed to get rss table size");
+                       goto qset_rss_error;
+               }
+
+               /* Configure RSS */
+               ret = nicvf_configure_rss(dev);
+               if (ret)
+                       goto qset_rss_error;
+       }
+
+       /* Done; Let PF make the BGX's RX and TX switches to ON position */
+       nicvf_mbox_cfg_done(nic);
+       return 0;
+
+qset_rss_error:
+       nicvf_rss_term(nic);
+start_rxq_error:
+       for (qidx = rx_start; qidx <= rx_end; qidx++)
+               nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
+start_txq_error:
+       for (qidx = tx_start; qidx <= tx_end; qidx++)
+               nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
+qset_rbdr_reclaim:
+       nicvf_qset_rbdr_reclaim(nic, 0);
+       nicvf_rbdr_release_mbufs(dev, nic);
+qset_rbdr_free:
+       if (nic->rbdr) {
+               rte_free(nic->rbdr);
+               nic->rbdr = NULL;
+       }
+qset_reclaim:
+       nicvf_qset_reclaim(nic);
+       return ret;
+}
+
+static int
+nicvf_dev_start(struct rte_eth_dev *dev)
+{
+       uint16_t qidx;
+       int ret;
+       size_t i;
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+       struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
+       uint16_t mtu;
+       uint32_t buffsz = 0, rbdrsz = 0;
+       struct rte_pktmbuf_pool_private *mbp_priv;
+       struct nicvf_rxq *rxq;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* This function must be called for a primary device */
+       assert_primary(nic);
+
+       /* Validate RBDR buff size */
+       for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
+               rxq = dev->data->rx_queues[qidx];
+               mbp_priv = rte_mempool_get_priv(rxq->pool);
+               buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
+               if (buffsz % 128) {
+                       PMD_INIT_LOG(ERR, "rxbuf size must be multiply of 128");
+                       return -EINVAL;
+               }
+               if (rbdrsz == 0)
+                       rbdrsz = buffsz;
+               if (rbdrsz != buffsz) {
+                       PMD_INIT_LOG(ERR, "buffsz not same, qidx=%d (%d/%d)",
+                                    qidx, rbdrsz, buffsz);
+                       return -EINVAL;
+               }
+       }
+
+       /* Configure loopback */
+       ret = nicvf_loopback_config(nic, dev->data->dev_conf.lpbk_mode);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to configure loopback %d", ret);
+               return ret;
+       }
+
+       /* Reset all statistics counters attached to this port */
+       ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, 0xFFFF, 0xFFFF);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to reset stat counters %d", ret);
+               return ret;
+       }
+
+       /* Setup scatter mode if needed by jumbo */
+       if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
+                                           2 * VLAN_TAG_SIZE > buffsz)
+               dev->data->scattered_rx = 1;
+       if (rx_conf->enable_scatter)
+               dev->data->scattered_rx = 1;
+
+       /* Setup MTU based on max_rx_pkt_len or default */
+       mtu = dev->data->dev_conf.rxmode.jumbo_frame ?
+               dev->data->dev_conf.rxmode.max_rx_pkt_len
+                       -  ETHER_HDR_LEN - ETHER_CRC_LEN
+               : ETHER_MTU;
+
+       if (nicvf_dev_set_mtu(dev, mtu)) {
+               PMD_INIT_LOG(ERR, "Failed to set default mtu size");
+               return -EBUSY;
+       }
+
+       ret = nicvf_vf_start(dev, nic, rbdrsz);
+       if (ret != 0)
+               return ret;
+
+       for (i = 0; i < nic->sqs_count; i++) {
+               assert(nic->snicvf[i]);
+
+               ret = nicvf_vf_start(dev, nic->snicvf[i], rbdrsz);
+               if (ret != 0)
+                       return ret;
+       }
+
+       /* Configure callbacks based on scatter mode */
+       nicvf_set_tx_function(dev);
+       nicvf_set_rx_function(dev);
+
+       return 0;
+}
+
+static void
+nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup)
+{
+       size_t i;
+       int ret;
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* Teardown secondary vf first */
+       for (i = 0; i < nic->sqs_count; i++) {
+               if (!nic->snicvf[i])
+                       continue;
+
+               nicvf_vf_stop(dev, nic->snicvf[i], cleanup);
+       }
+
+       /* Stop the primary VF now */
+       nicvf_vf_stop(dev, nic, cleanup);
+
+       /* Disable loopback */
+       ret = nicvf_loopback_config(nic, 0);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to disable loopback %d", ret);
+
+       /* Reclaim CPI configuration */
+       ret = nicvf_mbox_config_cpi(nic, 0);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to reclaim CPI config %d", ret);
+}
+
+static void
+nicvf_dev_stop(struct rte_eth_dev *dev)
+{
+       PMD_INIT_FUNC_TRACE();
+
+       nicvf_dev_stop_cleanup(dev, false);
+}
+
+static void
+nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic, bool cleanup)
+{
+       int ret;
+       uint16_t qidx;
+       uint16_t tx_start, tx_end;
+       uint16_t rx_start, rx_end;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (cleanup) {
+               /* Let PF make the BGX's RX and TX switches to OFF position */
+               nicvf_mbox_shutdown(nic);
+       }
+
+       /* Disable VLAN Strip */
+       nicvf_vlan_hw_strip(nic, 0);
+
+       /* Get queue ranges for this VF */
+       nicvf_tx_range(dev, nic, &tx_start, &tx_end);
+
+       for (qidx = tx_start; qidx <= tx_end; qidx++)
+               nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
+
+       /* Get queue ranges for this VF */
+       nicvf_rx_range(dev, nic, &rx_start, &rx_end);
+
+       /* Reclaim rq */
+       for (qidx = rx_start; qidx <= rx_end; qidx++)
+               nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
+
+       /* Reclaim RBDR */
+       ret = nicvf_qset_rbdr_reclaim(nic, 0);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to reclaim RBDR %d", ret);
+
+       /* Move all charged buffers in RBDR back to pool */
+       if (nic->rbdr != NULL)
+               nicvf_rbdr_release_mbufs(dev, nic);
+
+       /* Disable qset */
+       ret = nicvf_qset_reclaim(nic);
+       if (ret)
+               PMD_INIT_LOG(ERR, "Failed to disable qset %d", ret);
+
+       /* Disable all interrupts */
+       nicvf_disable_all_interrupts(nic);
+
+       /* Free RBDR SW structure */
+       if (nic->rbdr) {
+               rte_free(nic->rbdr);
+               nic->rbdr = NULL;
+       }
+}
+
+static void
+nicvf_dev_close(struct rte_eth_dev *dev)
+{
+       size_t i;
+       struct nicvf *nic = nicvf_pmd_priv(dev);
+
+       PMD_INIT_FUNC_TRACE();
+
+       nicvf_dev_stop_cleanup(dev, true);
+       nicvf_periodic_alarm_stop(nicvf_interrupt, dev);
+
+       for (i = 0; i < nic->sqs_count; i++) {
+               if (!nic->snicvf[i])
+                       continue;
+
+               nicvf_periodic_alarm_stop(nicvf_vf_interrupt, nic->snicvf[i]);
+       }
+}
+
+static int
+nicvf_request_sqs(struct nicvf *nic)
+{
+       size_t i;
+
+       assert_primary(nic);
+       assert(nic->sqs_count > 0);
+       assert(nic->sqs_count <= MAX_SQS_PER_VF);
+
+       /* Set no of Rx/Tx queues in each of the SQsets */
+       for (i = 0; i < nic->sqs_count; i++) {
+               if (nicvf_svf_empty())
+                       rte_panic("Cannot assign sufficient number of "
+                                 "secondary queues to primary VF%" PRIu8 "\n",
+                                 nic->vf_id);
+
+               nic->snicvf[i] = nicvf_svf_pop();
+               nic->snicvf[i]->sqs_id = i;
+       }
+
+       return nicvf_mbox_request_sqs(nic);
+}
+
 static int
 nicvf_dev_configure(struct rte_eth_dev *dev)
 {
-       struct rte_eth_conf *conf = &dev->data->dev_conf;
+       struct rte_eth_dev_data *data = dev->data;
+       struct rte_eth_conf *conf = &data->dev_conf;
        struct rte_eth_rxmode *rxmode = &conf->rxmode;
        struct rte_eth_txmode *txmode = &conf->txmode;
        struct nicvf *nic = nicvf_pmd_priv(dev);
+       uint8_t cqcount;
 
        PMD_INIT_FUNC_TRACE();
 
@@ -913,6 +1895,26 @@ nicvf_dev_configure(struct rte_eth_dev *dev)
                return -EINVAL;
        }
 
+       assert_primary(nic);
+       NICVF_STATIC_ASSERT(MAX_RCV_QUEUES_PER_QS == MAX_SND_QUEUES_PER_QS);
+       cqcount = RTE_MAX(data->nb_tx_queues, data->nb_rx_queues);
+       if (cqcount > MAX_RCV_QUEUES_PER_QS) {
+               nic->sqs_count = RTE_ALIGN_CEIL(cqcount, MAX_RCV_QUEUES_PER_QS);
+               nic->sqs_count = (nic->sqs_count / MAX_RCV_QUEUES_PER_QS) - 1;
+       } else {
+               nic->sqs_count = 0;
+       }
+
+       assert(nic->sqs_count <= MAX_SQS_PER_VF);
+
+       if (nic->sqs_count > 0) {
+               if (nicvf_request_sqs(nic)) {
+                       rte_panic("Cannot assign sufficient number of "
+                                 "secondary queues to PORT%d VF%" PRIu8 "\n",
+                                 dev->data->port_id, nic->vf_id);
+               }
+       }
+
        PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
                dev->data->port_id, nicvf_hw_cap(nic));
 
@@ -922,7 +1924,10 @@ nicvf_dev_configure(struct rte_eth_dev *dev)
 /* Initialize and register driver with DPDK Application */
 static const struct eth_dev_ops nicvf_eth_dev_ops = {
        .dev_configure            = nicvf_dev_configure,
+       .dev_start                = nicvf_dev_start,
+       .dev_stop                 = nicvf_dev_stop,
        .link_update              = nicvf_dev_link_update,
+       .dev_close                = nicvf_dev_close,
        .stats_get                = nicvf_dev_stats_get,
        .stats_reset              = nicvf_dev_stats_reset,
        .promiscuous_enable       = nicvf_dev_promisc_enable,
@@ -933,12 +1938,15 @@ static const struct eth_dev_ops nicvf_eth_dev_ops = {
        .reta_query               = nicvf_dev_reta_query,
        .rss_hash_update          = nicvf_dev_rss_hash_update,
        .rss_hash_conf_get        = nicvf_dev_rss_hash_conf_get,
+       .rx_queue_start           = nicvf_dev_rx_queue_start,
+       .rx_queue_stop            = nicvf_dev_rx_queue_stop,
+       .tx_queue_start           = nicvf_dev_tx_queue_start,
+       .tx_queue_stop            = nicvf_dev_tx_queue_stop,
        .rx_queue_setup           = nicvf_dev_rx_queue_setup,
        .rx_queue_release         = nicvf_dev_rx_queue_release,
        .rx_queue_count           = nicvf_dev_rx_queue_count,
        .tx_queue_setup           = nicvf_dev_tx_queue_setup,
        .tx_queue_release         = nicvf_dev_tx_queue_release,
-       .get_reg_length           = nicvf_dev_get_reg_length,
        .get_reg                  = nicvf_dev_get_regs,
 };
 
@@ -953,6 +1961,20 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
 
        eth_dev->dev_ops = &nicvf_eth_dev_ops;
 
+       /* For secondary processes, the primary has done all the work */
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
+               if (nic) {
+                       /* Setup callbacks for secondary process */
+                       nicvf_set_tx_function(eth_dev);
+                       nicvf_set_rx_function(eth_dev);
+                       return 0;
+               } else {
+                       /* If nic == NULL than it is secondary function
+                        * so ethdev need to be released by caller */
+                       return ENOTSUP;
+               }
+       }
+
        pci_dev = eth_dev->pci_dev;
        rte_eth_copy_pci_info(eth_dev, pci_dev);
 
@@ -960,7 +1982,6 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
        nic->vendor_id = pci_dev->id.vendor_id;
        nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
        nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
-       nic->eth_dev = eth_dev;
 
        PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
                        pci_dev->id.vendor_id, pci_dev->id.device_id,
@@ -976,7 +1997,7 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
 
        nicvf_disable_all_interrupts(nic);
 
-       ret = nicvf_periodic_alarm_start(nic);
+       ret = nicvf_periodic_alarm_start(nicvf_interrupt, eth_dev);
        if (ret) {
                PMD_INIT_LOG(ERR, "Failed to start period alarm");
                goto fail;
@@ -996,11 +2017,28 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
                        );
        }
 
+       ret = nicvf_base_init(nic);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
+               goto malloc_fail;
+       }
+
        if (nic->sqs_mode) {
-               PMD_INIT_LOG(INFO, "Unsupported SQS VF detected, Detaching...");
-               /* Detach port by returning Positive error number */
-               ret = ENOTSUP;
-               goto alarm_fail;
+               /* Push nic to stack of secondary vfs */
+               nicvf_svf_push(nic);
+
+               /* Steal nic pointer from the device for further reuse */
+               eth_dev->data->dev_private = NULL;
+
+               nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
+               ret = nicvf_periodic_alarm_start(nicvf_vf_interrupt, nic);
+               if (ret) {
+                       PMD_INIT_LOG(ERR, "Failed to start period alarm");
+                       goto fail;
+               }
+
+               /* Detach port by returning postive error number */
+               return ENOTSUP;
        }
 
        eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
@@ -1021,18 +2059,6 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
                goto malloc_fail;
        }
 
-       ret = nicvf_base_init(nic);
-       if (ret) {
-               PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
-               goto malloc_fail;
-       }
-
-       ret = nicvf_mbox_get_rss_size(nic);
-       if (ret) {
-               PMD_INIT_LOG(ERR, "Failed to get rss table size");
-               goto malloc_fail;
-       }
-
        PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
                eth_dev->data->port_id, nic->vendor_id, nic->device_id,
                nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
@@ -1043,7 +2069,7 @@ nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
 malloc_fail:
        rte_free(eth_dev->data->mac_addrs);
 alarm_fail:
-       nicvf_periodic_alarm_stop(nic);
+       nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
 fail:
        return ret;
 }
@@ -1052,16 +2078,30 @@ static const struct rte_pci_id pci_id_nicvf_map[] = {
        {
                .class_id = RTE_CLASS_ANY_ID,
                .vendor_id = PCI_VENDOR_ID_CAVIUM,
-               .device_id = PCI_DEVICE_ID_THUNDERX_PASS1_NICVF,
+               .device_id = PCI_DEVICE_ID_THUNDERX_CN88XX_PASS1_NICVF,
+               .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
+               .subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS1_NICVF,
+       },
+       {
+               .class_id = RTE_CLASS_ANY_ID,
+               .vendor_id = PCI_VENDOR_ID_CAVIUM,
+               .device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
                .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
-               .subsystem_device_id = PCI_SUB_DEVICE_ID_THUNDERX_PASS1_NICVF,
+               .subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS2_NICVF,
        },
        {
                .class_id = RTE_CLASS_ANY_ID,
                .vendor_id = PCI_VENDOR_ID_CAVIUM,
-               .device_id = PCI_DEVICE_ID_THUNDERX_PASS2_NICVF,
+               .device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
                .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
-               .subsystem_device_id = PCI_SUB_DEVICE_ID_THUNDERX_PASS2_NICVF,
+               .subsystem_device_id = PCI_SUB_DEVICE_ID_CN81XX_NICVF,
+       },
+       {
+               .class_id = RTE_CLASS_ANY_ID,
+               .vendor_id = PCI_VENDOR_ID_CAVIUM,
+               .device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
+               .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
+               .subsystem_device_id = PCI_SUB_DEVICE_ID_CN83XX_NICVF,
        },
        {
                .vendor_id = 0,
@@ -1070,29 +2110,15 @@ static const struct rte_pci_id pci_id_nicvf_map[] = {
 
 static struct eth_driver rte_nicvf_pmd = {
        .pci_drv = {
-               .name = "rte_nicvf_pmd",
                .id_table = pci_id_nicvf_map,
                .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
+               .probe = rte_eth_dev_pci_probe,
+               .remove = rte_eth_dev_pci_remove,
        },
        .eth_dev_init = nicvf_eth_dev_init,
        .dev_private_size = sizeof(struct nicvf),
 };
 
-static int
-rte_nicvf_pmd_init(const char *name __rte_unused, const char *para __rte_unused)
-{
-       PMD_INIT_FUNC_TRACE();
-       PMD_INIT_LOG(INFO, "librte_pmd_thunderx nicvf version %s",
-                       THUNDERX_NICVF_PMD_VERSION);
-
-       rte_eth_driver_register(&rte_nicvf_pmd);
-       return 0;
-}
-
-static struct rte_driver rte_nicvf_driver = {
-       .name = "nicvf_driver",
-       .type = PMD_PDEV,
-       .init = rte_nicvf_pmd_init,
-};
-
-PMD_REGISTER_DRIVER(rte_nicvf_driver);
+RTE_PMD_REGISTER_PCI(net_thunderx, rte_nicvf_pmd.pci_drv);
+RTE_PMD_REGISTER_PCI_TABLE(net_thunderx, pci_id_nicvf_map);
+RTE_PMD_REGISTER_KMOD_DEP(net_thunderx, "* igb_uio | uio_pci_generic | vfio");