net: add macro for VLAN header length
[dpdk.git] / drivers / net / ice / ice_dcf_ethdev.c
index 23f82a4..28f7f7f 100644 (file)
@@ -4,8 +4,8 @@
 
 #include <errno.h>
 #include <stdbool.h>
+#include <sys/queue.h>
 #include <sys/types.h>
-#include <sys/ioctl.h>
 #include <unistd.h>
 
 #include <rte_interrupts.h>
@@ -14,7 +14,7 @@
 #include <rte_atomic.h>
 #include <rte_eal.h>
 #include <rte_ether.h>
-#include <rte_ethdev_pci.h>
+#include <ethdev_pci.h>
 #include <rte_kvargs.h>
 #include <rte_malloc.h>
 #include <rte_memzone.h>
 
 #include "ice_generic_flow.h"
 #include "ice_dcf_ethdev.h"
+#include "ice_rxtx.h"
+
+static int
+ice_dcf_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
+                               struct rte_eth_udp_tunnel *udp_tunnel);
+static int
+ice_dcf_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
+                               struct rte_eth_udp_tunnel *udp_tunnel);
+
+static int
+ice_dcf_dev_init(struct rte_eth_dev *eth_dev);
+
+static int
+ice_dcf_dev_uninit(struct rte_eth_dev *eth_dev);
 
 static uint16_t
 ice_dcf_recv_pkts(__rte_unused void *rx_queue,
@@ -42,49 +56,672 @@ ice_dcf_xmit_pkts(__rte_unused void *tx_queue,
 }
 
 static int
-ice_dcf_dev_start(struct rte_eth_dev *dev)
+ice_dcf_init_rxq(struct rte_eth_dev *dev, struct ice_rx_queue *rxq)
 {
-       dev->data->dev_link.link_status = ETH_LINK_UP;
+       struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
+       struct rte_eth_dev_data *dev_data = dev->data;
+       struct iavf_hw *hw = &dcf_ad->real_hw.avf;
+       uint16_t buf_size, max_pkt_len;
+
+       buf_size = 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));
+       max_pkt_len = RTE_MIN(ICE_SUPPORT_CHAIN_NUM * rxq->rx_buf_len,
+                             dev->data->mtu + ICE_ETH_OVERHEAD);
+
+       /* Check maximum packet length is set correctly.  */
+       if (max_pkt_len <= RTE_ETHER_MIN_LEN ||
+           max_pkt_len > ICE_FRAME_SIZE_MAX) {
+               PMD_DRV_LOG(ERR, "maximum packet length must be "
+                           "larger than %u and smaller than %u",
+                           (uint32_t)RTE_ETHER_MIN_LEN,
+                           (uint32_t)ICE_FRAME_SIZE_MAX);
+               return -EINVAL;
+       }
+
+       rxq->max_pkt_len = max_pkt_len;
+       if ((dev_data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_SCATTER) ||
+           (rxq->max_pkt_len + 2 * RTE_VLAN_HLEN) > buf_size) {
+               dev_data->scattered_rx = 1;
+       }
+       rxq->qrx_tail = hw->hw_addr + IAVF_QRX_TAIL1(rxq->queue_id);
+       IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
+       IAVF_WRITE_FLUSH(hw);
 
        return 0;
 }
 
-static void
-ice_dcf_dev_stop(struct rte_eth_dev *dev)
+static int
+ice_dcf_init_rx_queues(struct rte_eth_dev *dev)
 {
-       dev->data->dev_link.link_status = ETH_LINK_DOWN;
+       struct ice_rx_queue **rxq =
+               (struct ice_rx_queue **)dev->data->rx_queues;
+       int i, ret;
+
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               if (!rxq[i] || !rxq[i]->q_set)
+                       continue;
+               ret = ice_dcf_init_rxq(dev, rxq[i]);
+               if (ret)
+                       return ret;
+       }
+
+       ice_set_rx_function(dev);
+       ice_set_tx_function(dev);
+
+       return 0;
+}
+
+#define IAVF_MISC_VEC_ID                RTE_INTR_VEC_ZERO_OFFSET
+#define IAVF_RX_VEC_START               RTE_INTR_VEC_RXTX_OFFSET
+
+#define IAVF_ITR_INDEX_DEFAULT          0
+#define IAVF_QUEUE_ITR_INTERVAL_DEFAULT 32 /* 32 us */
+#define IAVF_QUEUE_ITR_INTERVAL_MAX     8160 /* 8160 us */
+
+static inline uint16_t
+iavf_calc_itr_interval(int16_t interval)
+{
+       if (interval < 0 || interval > IAVF_QUEUE_ITR_INTERVAL_MAX)
+               interval = IAVF_QUEUE_ITR_INTERVAL_DEFAULT;
+
+       /* Convert to hardware count, as writing each 1 represents 2 us */
+       return interval / 2;
 }
 
 static int
-ice_dcf_dev_configure(__rte_unused struct rte_eth_dev *dev)
+ice_dcf_config_rx_queues_irqs(struct rte_eth_dev *dev,
+                                    struct rte_intr_handle *intr_handle)
 {
+       struct ice_dcf_adapter *adapter = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &adapter->real_hw;
+       uint16_t interval, i;
+       int vec;
+
+       if (rte_intr_cap_multiple(intr_handle) &&
+           dev->data->dev_conf.intr_conf.rxq) {
+               if (rte_intr_efd_enable(intr_handle, dev->data->nb_rx_queues))
+                       return -1;
+       }
+
+       if (rte_intr_dp_is_en(intr_handle)) {
+               if (rte_intr_vec_list_alloc(intr_handle, "intr_vec",
+                                                  dev->data->nb_rx_queues)) {
+                       PMD_DRV_LOG(ERR, "Failed to allocate %d rx intr_vec",
+                                   dev->data->nb_rx_queues);
+                       return -1;
+               }
+       }
+
+       if (!dev->data->dev_conf.intr_conf.rxq ||
+           !rte_intr_dp_is_en(intr_handle)) {
+               /* Rx interrupt disabled, Map interrupt only for writeback */
+               hw->nb_msix = 1;
+               if (hw->vf_res->vf_cap_flags &
+                   VIRTCHNL_VF_OFFLOAD_WB_ON_ITR) {
+                       /* If WB_ON_ITR supports, enable it */
+                       hw->msix_base = IAVF_RX_VEC_START;
+                       /* Set the ITR for index zero, to 2us to make sure that
+                        * we leave time for aggregation to occur, but don't
+                        * increase latency dramatically.
+                        */
+                       IAVF_WRITE_REG(&hw->avf,
+                                      IAVF_VFINT_DYN_CTLN1(hw->msix_base - 1),
+                                      (0 << IAVF_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
+                                      IAVF_VFINT_DYN_CTLN1_WB_ON_ITR_MASK |
+                                      (2UL << IAVF_VFINT_DYN_CTLN1_INTERVAL_SHIFT));
+               } else {
+                       /* If no WB_ON_ITR offload flags, need to set
+                        * interrupt for descriptor write back.
+                        */
+                       hw->msix_base = IAVF_MISC_VEC_ID;
+
+                       /* set ITR to max */
+                       interval =
+                       iavf_calc_itr_interval(IAVF_QUEUE_ITR_INTERVAL_MAX);
+                       IAVF_WRITE_REG(&hw->avf, IAVF_VFINT_DYN_CTL01,
+                                      IAVF_VFINT_DYN_CTL01_INTENA_MASK |
+                                      (IAVF_ITR_INDEX_DEFAULT <<
+                                       IAVF_VFINT_DYN_CTL01_ITR_INDX_SHIFT) |
+                                      (interval <<
+                                       IAVF_VFINT_DYN_CTL01_INTERVAL_SHIFT));
+               }
+               IAVF_WRITE_FLUSH(&hw->avf);
+               /* map all queues to the same interrupt */
+               for (i = 0; i < dev->data->nb_rx_queues; i++)
+                       hw->rxq_map[hw->msix_base] |= 1 << i;
+       } else {
+               if (!rte_intr_allow_others(intr_handle)) {
+                       hw->nb_msix = 1;
+                       hw->msix_base = IAVF_MISC_VEC_ID;
+                       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+                               hw->rxq_map[hw->msix_base] |= 1 << i;
+                               rte_intr_vec_list_index_set(intr_handle,
+                                                       i, IAVF_MISC_VEC_ID);
+                       }
+                       PMD_DRV_LOG(DEBUG,
+                                   "vector %u are mapping to all Rx queues",
+                                   hw->msix_base);
+               } else {
+                       /* If Rx interrupt is reuquired, and we can use
+                        * multi interrupts, then the vec is from 1
+                        */
+                       hw->nb_msix = RTE_MIN(hw->vf_res->max_vectors,
+                                     rte_intr_nb_efd_get(intr_handle));
+                       hw->msix_base = IAVF_MISC_VEC_ID;
+                       vec = IAVF_MISC_VEC_ID;
+                       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+                               hw->rxq_map[vec] |= 1 << i;
+                               rte_intr_vec_list_index_set(intr_handle,
+                                                                  i, vec++);
+                               if (vec >= hw->nb_msix)
+                                       vec = IAVF_RX_VEC_START;
+                       }
+                       PMD_DRV_LOG(DEBUG,
+                                   "%u vectors are mapping to %u Rx queues",
+                                   hw->nb_msix, dev->data->nb_rx_queues);
+               }
+       }
+
+       if (ice_dcf_config_irq_map(hw)) {
+               PMD_DRV_LOG(ERR, "config interrupt mapping failed");
+               return -1;
+       }
        return 0;
 }
 
 static int
-ice_dcf_dev_info_get(struct rte_eth_dev *dev,
-                    struct rte_eth_dev_info *dev_info)
+alloc_rxq_mbufs(struct ice_rx_queue *rxq)
 {
-       struct ice_dcf_adapter *adapter = dev->data->dev_private;
+       volatile union ice_rx_flex_desc *rxd;
+       struct rte_mbuf *mbuf = NULL;
+       uint64_t dma_addr;
+       uint16_t i;
 
-       dev_info->max_mac_addrs = 1;
-       dev_info->max_rx_pktlen = (uint32_t)-1;
-       dev_info->max_rx_queues = RTE_DIM(adapter->rxqs);
-       dev_info->max_tx_queues = RTE_DIM(adapter->txqs);
+       for (i = 0; i < rxq->nb_rx_desc; i++) {
+               mbuf = rte_mbuf_raw_alloc(rxq->mp);
+               if (unlikely(!mbuf)) {
+                       PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
+                       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;
+
+               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;
+#ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
+               rxd->read.rsvd1 = 0;
+               rxd->read.rsvd2 = 0;
+#endif
+
+               rxq->sw_ring[i].mbuf = (void *)mbuf;
+       }
+
+       return 0;
+}
+
+static int
+ice_dcf_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct iavf_hw *hw = &ad->real_hw.avf;
+       struct ice_rx_queue *rxq;
+       int err = 0;
+
+       if (rx_queue_id >= dev->data->nb_rx_queues)
+               return -EINVAL;
+
+       rxq = dev->data->rx_queues[rx_queue_id];
+
+       err = alloc_rxq_mbufs(rxq);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
+               return err;
+       }
+
+       rte_wmb();
+
+       /* Init the RX tail register. */
+       IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
+       IAVF_WRITE_FLUSH(hw);
+
+       /* Ready to switch the queue on */
+       err = ice_dcf_switch_queue(&ad->real_hw, rx_queue_id, true, true);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
+                           rx_queue_id);
+               return err;
+       }
+
+       dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+
+       return 0;
+}
+
+static inline void
+reset_rx_queue(struct ice_rx_queue *rxq)
+{
+       uint16_t len;
+       uint32_t i;
+
+       if (!rxq)
+               return;
+
+       len = rxq->nb_rx_desc + ICE_RX_MAX_BURST;
+
+       for (i = 0; i < len * sizeof(union ice_rx_flex_desc); i++)
+               ((volatile char *)rxq->rx_ring)[i] = 0;
+
+       memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
+
+       for (i = 0; i < ICE_RX_MAX_BURST; i++)
+               rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
+
+       /* for rx bulk */
+       rxq->rx_nb_avail = 0;
+       rxq->rx_next_avail = 0;
+       rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
+
+       rxq->rx_tail = 0;
+       rxq->nb_rx_hold = 0;
+       rxq->pkt_first_seg = NULL;
+       rxq->pkt_last_seg = NULL;
+}
+
+static inline void
+reset_tx_queue(struct ice_tx_queue *txq)
+{
+       struct ice_tx_entry *txe;
+       uint32_t i, size;
+       uint16_t prev;
+
+       if (!txq) {
+               PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
+               return;
+       }
+
+       txe = txq->sw_ring;
+       size = sizeof(struct ice_tx_desc) * txq->nb_tx_desc;
+       for (i = 0; i < size; i++)
+               ((volatile char *)txq->tx_ring)[i] = 0;
+
+       prev = (uint16_t)(txq->nb_tx_desc - 1);
+       for (i = 0; i < txq->nb_tx_desc; i++) {
+               txq->tx_ring[i].cmd_type_offset_bsz =
+                       rte_cpu_to_le_64(IAVF_TX_DESC_DTYPE_DESC_DONE);
+               txe[i].mbuf =  NULL;
+               txe[i].last_id = i;
+               txe[prev].next_id = i;
+               prev = i;
+       }
+
+       txq->tx_tail = 0;
+       txq->nb_tx_used = 0;
+
+       txq->last_desc_cleaned = txq->nb_tx_desc - 1;
+       txq->nb_tx_free = txq->nb_tx_desc - 1;
+
+       txq->tx_next_dd = txq->tx_rs_thresh - 1;
+       txq->tx_next_rs = txq->tx_rs_thresh - 1;
+}
+
+static int
+ice_dcf_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &ad->real_hw;
+       struct ice_rx_queue *rxq;
+       int err;
+
+       if (rx_queue_id >= dev->data->nb_rx_queues)
+               return -EINVAL;
+
+       err = ice_dcf_switch_queue(hw, rx_queue_id, true, false);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
+                           rx_queue_id);
+               return err;
+       }
+
+       rxq = dev->data->rx_queues[rx_queue_id];
+       rxq->rx_rel_mbufs(rxq);
+       reset_rx_queue(rxq);
+       dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+
+       return 0;
+}
+
+static int
+ice_dcf_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct iavf_hw *hw = &ad->real_hw.avf;
+       struct ice_tx_queue *txq;
+       int err = 0;
+
+       if (tx_queue_id >= dev->data->nb_tx_queues)
+               return -EINVAL;
+
+       txq = dev->data->tx_queues[tx_queue_id];
+
+       /* Init the RX tail register. */
+       txq->qtx_tail = hw->hw_addr + IAVF_QTX_TAIL1(tx_queue_id);
+       IAVF_PCI_REG_WRITE(txq->qtx_tail, 0);
+       IAVF_WRITE_FLUSH(hw);
+
+       /* Ready to switch the queue on */
+       err = ice_dcf_switch_queue(&ad->real_hw, tx_queue_id, false, true);
+
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
+                           tx_queue_id);
+               return err;
+       }
+
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+
+       return 0;
+}
+
+static int
+ice_dcf_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &ad->real_hw;
+       struct ice_tx_queue *txq;
+       int err;
+
+       if (tx_queue_id >= dev->data->nb_tx_queues)
+               return -EINVAL;
+
+       err = ice_dcf_switch_queue(hw, tx_queue_id, false, false);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch TX queue %u off",
+                           tx_queue_id);
+               return err;
+       }
+
+       txq = dev->data->tx_queues[tx_queue_id];
+       txq->tx_rel_mbufs(txq);
+       reset_tx_queue(txq);
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+
+       return 0;
+}
+
+static int
+ice_dcf_start_queues(struct rte_eth_dev *dev)
+{
+       struct ice_rx_queue *rxq;
+       struct ice_tx_queue *txq;
+       int nb_rxq = 0;
+       int nb_txq, i;
+
+       for (nb_txq = 0; nb_txq < dev->data->nb_tx_queues; nb_txq++) {
+               txq = dev->data->tx_queues[nb_txq];
+               if (txq->tx_deferred_start)
+                       continue;
+               if (ice_dcf_tx_queue_start(dev, nb_txq) != 0) {
+                       PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_txq);
+                       goto tx_err;
+               }
+       }
+
+       for (nb_rxq = 0; nb_rxq < dev->data->nb_rx_queues; nb_rxq++) {
+               rxq = dev->data->rx_queues[nb_rxq];
+               if (rxq->rx_deferred_start)
+                       continue;
+               if (ice_dcf_rx_queue_start(dev, nb_rxq) != 0) {
+                       PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_rxq);
+                       goto rx_err;
+               }
+       }
+
+       return 0;
+
+       /* stop the started queues if failed to start all queues */
+rx_err:
+       for (i = 0; i < nb_rxq; i++)
+               ice_dcf_rx_queue_stop(dev, i);
+tx_err:
+       for (i = 0; i < nb_txq; i++)
+               ice_dcf_tx_queue_stop(dev, i);
+
+       return -1;
+}
+
+static int
+ice_dcf_dev_start(struct rte_eth_dev *dev)
+{
+       struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
+       struct rte_intr_handle *intr_handle = dev->intr_handle;
+       struct ice_adapter *ad = &dcf_ad->parent;
+       struct ice_dcf_hw *hw = &dcf_ad->real_hw;
+       int ret;
+
+       if (hw->resetting) {
+               PMD_DRV_LOG(ERR,
+                           "The DCF has been reset by PF, please reinit first");
+               return -EIO;
+       }
+
+       if (hw->tm_conf.root && !hw->tm_conf.committed) {
+               PMD_DRV_LOG(ERR,
+                       "please call hierarchy_commit() before starting the port");
+               return -EIO;
+       }
+
+       ad->pf.adapter_stopped = 0;
+
+       hw->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
+                                     dev->data->nb_tx_queues);
+
+       ret = ice_dcf_init_rx_queues(dev);
+       if (ret) {
+               PMD_DRV_LOG(ERR, "Fail to init queues");
+               return ret;
+       }
+
+       if (hw->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RSS_PF) {
+               ret = ice_dcf_init_rss(hw);
+               if (ret) {
+                       PMD_DRV_LOG(ERR, "Failed to configure RSS");
+                       return ret;
+               }
+       }
+
+       ret = ice_dcf_configure_queues(hw);
+       if (ret) {
+               PMD_DRV_LOG(ERR, "Fail to config queues");
+               return ret;
+       }
+
+       ret = ice_dcf_config_rx_queues_irqs(dev, intr_handle);
+       if (ret) {
+               PMD_DRV_LOG(ERR, "Fail to config rx queues' irqs");
+               return ret;
+       }
+
+       if (dev->data->dev_conf.intr_conf.rxq != 0) {
+               rte_intr_disable(intr_handle);
+               rte_intr_enable(intr_handle);
+       }
+
+       ret = ice_dcf_start_queues(dev);
+       if (ret) {
+               PMD_DRV_LOG(ERR, "Failed to enable queues");
+               return ret;
+       }
+
+       ret = ice_dcf_add_del_all_mac_addr(hw, true);
+       if (ret) {
+               PMD_DRV_LOG(ERR, "Failed to add mac addr");
+               return ret;
+       }
+
+       dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
+
+       return 0;
+}
+
+static void
+ice_dcf_stop_queues(struct rte_eth_dev *dev)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &ad->real_hw;
+       struct ice_rx_queue *rxq;
+       struct ice_tx_queue *txq;
+       int ret, i;
+
+       /* Stop All queues */
+       ret = ice_dcf_disable_queues(hw);
+       if (ret)
+               PMD_DRV_LOG(WARNING, "Fail to stop queues");
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               txq = dev->data->tx_queues[i];
+               if (!txq)
+                       continue;
+               txq->tx_rel_mbufs(txq);
+               reset_tx_queue(txq);
+               dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
+       }
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               rxq = dev->data->rx_queues[i];
+               if (!rxq)
+                       continue;
+               rxq->rx_rel_mbufs(rxq);
+               reset_rx_queue(rxq);
+               dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
+       }
+}
+
+static int
+ice_dcf_dev_stop(struct rte_eth_dev *dev)
+{
+       struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
+       struct rte_intr_handle *intr_handle = dev->intr_handle;
+       struct ice_adapter *ad = &dcf_ad->parent;
+       struct ice_dcf_hw *hw = &dcf_ad->real_hw;
+
+       if (ad->pf.adapter_stopped == 1) {
+               PMD_DRV_LOG(DEBUG, "Port is already stopped");
+               return 0;
+       }
+
+       /* Stop the VF representors for this device */
+       ice_dcf_vf_repr_stop_all(dcf_ad);
+
+       ice_dcf_stop_queues(dev);
+
+       rte_intr_efd_disable(intr_handle);
+       rte_intr_vec_list_free(intr_handle);
+
+       ice_dcf_add_del_all_mac_addr(&dcf_ad->real_hw, false);
+       dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
+       ad->pf.adapter_stopped = 1;
+       hw->tm_conf.committed = false;
 
        return 0;
 }
 
 static int
-ice_dcf_stats_get(__rte_unused struct rte_eth_dev *dev,
-                 __rte_unused struct rte_eth_stats *igb_stats)
+ice_dcf_dev_configure(struct rte_eth_dev *dev)
 {
+       struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
+       struct ice_adapter *ad = &dcf_ad->parent;
+
+       ad->rx_bulk_alloc_allowed = true;
+       ad->tx_simple_allowed = true;
+
+       if (dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
+               dev->data->dev_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
+
        return 0;
 }
 
 static int
-ice_dcf_stats_reset(__rte_unused struct rte_eth_dev *dev)
+ice_dcf_dev_info_get(struct rte_eth_dev *dev,
+                    struct rte_eth_dev_info *dev_info)
 {
+       struct ice_dcf_adapter *adapter = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &adapter->real_hw;
+
+       dev_info->max_mac_addrs = 1;
+       dev_info->max_rx_queues = hw->vsi_res->num_queue_pairs;
+       dev_info->max_tx_queues = hw->vsi_res->num_queue_pairs;
+       dev_info->min_rx_bufsize = ICE_BUF_SIZE_MIN;
+       dev_info->max_rx_pktlen = ICE_FRAME_SIZE_MAX;
+       dev_info->hash_key_size = hw->vf_res->rss_key_size;
+       dev_info->reta_size = hw->vf_res->rss_lut_size;
+       dev_info->flow_type_rss_offloads = ICE_RSS_OFFLOAD_ALL;
+       dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
+
+       dev_info->rx_offload_capa =
+               RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
+               RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
+               RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
+               RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
+               RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
+               RTE_ETH_RX_OFFLOAD_SCATTER |
+               RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
+               RTE_ETH_RX_OFFLOAD_RSS_HASH;
+       dev_info->tx_offload_capa =
+               RTE_ETH_TX_OFFLOAD_VLAN_INSERT |
+               RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
+               RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
+               RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
+               RTE_ETH_TX_OFFLOAD_SCTP_CKSUM |
+               RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM |
+               RTE_ETH_TX_OFFLOAD_TCP_TSO |
+               RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO |
+               RTE_ETH_TX_OFFLOAD_GRE_TNL_TSO |
+               RTE_ETH_TX_OFFLOAD_IPIP_TNL_TSO |
+               RTE_ETH_TX_OFFLOAD_GENEVE_TNL_TSO |
+               RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
+
+       dev_info->default_rxconf = (struct rte_eth_rxconf) {
+               .rx_thresh = {
+                       .pthresh = ICE_DEFAULT_RX_PTHRESH,
+                       .hthresh = ICE_DEFAULT_RX_HTHRESH,
+                       .wthresh = ICE_DEFAULT_RX_WTHRESH,
+               },
+               .rx_free_thresh = ICE_DEFAULT_RX_FREE_THRESH,
+               .rx_drop_en = 0,
+               .offloads = 0,
+       };
+
+       dev_info->default_txconf = (struct rte_eth_txconf) {
+               .tx_thresh = {
+                       .pthresh = ICE_DEFAULT_TX_PTHRESH,
+                       .hthresh = ICE_DEFAULT_TX_HTHRESH,
+                       .wthresh = ICE_DEFAULT_TX_WTHRESH,
+               },
+               .tx_free_thresh = ICE_DEFAULT_TX_FREE_THRESH,
+               .tx_rs_thresh = ICE_DEFAULT_TX_RSBIT_THRESH,
+               .offloads = 0,
+       };
+
+       dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
+               .nb_max = ICE_MAX_RING_DESC,
+               .nb_min = ICE_MIN_RING_DESC,
+               .nb_align = ICE_ALIGN_RING_DESC,
+       };
+
+       dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
+               .nb_max = ICE_MAX_RING_DESC,
+               .nb_min = ICE_MIN_RING_DESC,
+               .nb_align = ICE_ALIGN_RING_DESC,
+       };
+
        return 0;
 }
 
@@ -113,94 +750,344 @@ ice_dcf_dev_allmulticast_disable(__rte_unused struct rte_eth_dev *dev)
 }
 
 static int
-ice_dcf_dev_filter_ctrl(struct rte_eth_dev *dev,
-                       enum rte_filter_type filter_type,
-                       __rte_unused enum rte_filter_op filter_op,
-                       __rte_unused void *arg)
+ice_dcf_dev_flow_ops_get(struct rte_eth_dev *dev,
+                        const struct rte_flow_ops **ops)
 {
-       int ret = 0;
-
        if (!dev)
                return -EINVAL;
 
-       switch (filter_type) {
-       default:
-               PMD_DRV_LOG(WARNING, "Filter type (%d) not supported",
-                           filter_type);
-               ret = -EINVAL;
-               break;
+       *ops = &ice_flow_ops;
+       return 0;
+}
+
+#define ICE_DCF_32_BIT_WIDTH (CHAR_BIT * 4)
+#define ICE_DCF_48_BIT_WIDTH (CHAR_BIT * 6)
+#define ICE_DCF_48_BIT_MASK  RTE_LEN2MASK(ICE_DCF_48_BIT_WIDTH, uint64_t)
+
+static void
+ice_dcf_stat_update_48(uint64_t *offset, uint64_t *stat)
+{
+       if (*stat >= *offset)
+               *stat = *stat - *offset;
+       else
+               *stat = (uint64_t)((*stat +
+                       ((uint64_t)1 << ICE_DCF_48_BIT_WIDTH)) - *offset);
+
+       *stat &= ICE_DCF_48_BIT_MASK;
+}
+
+static void
+ice_dcf_stat_update_32(uint64_t *offset, uint64_t *stat)
+{
+       if (*stat >= *offset)
+               *stat = (uint64_t)(*stat - *offset);
+       else
+               *stat = (uint64_t)((*stat +
+                       ((uint64_t)1 << ICE_DCF_32_BIT_WIDTH)) - *offset);
+}
+
+static void
+ice_dcf_update_stats(struct virtchnl_eth_stats *oes,
+                    struct virtchnl_eth_stats *nes)
+{
+       ice_dcf_stat_update_48(&oes->rx_bytes, &nes->rx_bytes);
+       ice_dcf_stat_update_48(&oes->rx_unicast, &nes->rx_unicast);
+       ice_dcf_stat_update_48(&oes->rx_multicast, &nes->rx_multicast);
+       ice_dcf_stat_update_48(&oes->rx_broadcast, &nes->rx_broadcast);
+       ice_dcf_stat_update_32(&oes->rx_discards, &nes->rx_discards);
+       ice_dcf_stat_update_48(&oes->tx_bytes, &nes->tx_bytes);
+       ice_dcf_stat_update_48(&oes->tx_unicast, &nes->tx_unicast);
+       ice_dcf_stat_update_48(&oes->tx_multicast, &nes->tx_multicast);
+       ice_dcf_stat_update_48(&oes->tx_broadcast, &nes->tx_broadcast);
+       ice_dcf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
+       ice_dcf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
+}
+
+
+static int
+ice_dcf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &ad->real_hw;
+       struct virtchnl_eth_stats pstats;
+       int ret;
+
+       if (hw->resetting) {
+               PMD_DRV_LOG(ERR,
+                           "The DCF has been reset by PF, please reinit first");
+               return -EIO;
        }
 
+       ret = ice_dcf_query_stats(hw, &pstats);
+       if (ret == 0) {
+               ice_dcf_update_stats(&hw->eth_stats_offset, &pstats);
+               stats->ipackets = pstats.rx_unicast + pstats.rx_multicast +
+                               pstats.rx_broadcast - pstats.rx_discards;
+               stats->opackets = pstats.tx_broadcast + pstats.tx_multicast +
+                                               pstats.tx_unicast;
+               stats->imissed = pstats.rx_discards;
+               stats->oerrors = pstats.tx_errors + pstats.tx_discards;
+               stats->ibytes = pstats.rx_bytes;
+               stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
+               stats->obytes = pstats.tx_bytes;
+       } else {
+               PMD_DRV_LOG(ERR, "Get statistics failed");
+       }
        return ret;
 }
 
+static int
+ice_dcf_stats_reset(struct rte_eth_dev *dev)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &ad->real_hw;
+       struct virtchnl_eth_stats pstats;
+       int ret;
+
+       if (hw->resetting)
+               return 0;
+
+       /* read stat values to clear hardware registers */
+       ret = ice_dcf_query_stats(hw, &pstats);
+       if (ret != 0)
+               return ret;
+
+       /* set stats offset base on current values */
+       hw->eth_stats_offset = pstats;
+
+       return 0;
+}
+
 static void
+ice_dcf_free_repr_info(struct ice_dcf_adapter *dcf_adapter)
+{
+       if (dcf_adapter->repr_infos) {
+               rte_free(dcf_adapter->repr_infos);
+               dcf_adapter->repr_infos = NULL;
+       }
+}
+
+static int
+ice_dcf_init_repr_info(struct ice_dcf_adapter *dcf_adapter)
+{
+       dcf_adapter->repr_infos =
+                       rte_calloc("ice_dcf_rep_info",
+                                  dcf_adapter->real_hw.num_vfs,
+                                  sizeof(dcf_adapter->repr_infos[0]), 0);
+       if (!dcf_adapter->repr_infos) {
+               PMD_DRV_LOG(ERR, "Failed to alloc memory for VF representors\n");
+               return -ENOMEM;
+       }
+
+       return 0;
+}
+
+static int
 ice_dcf_dev_close(struct rte_eth_dev *dev)
 {
        struct ice_dcf_adapter *adapter = dev->data->dev_private;
 
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
-               return;
+               return 0;
 
-       dev->dev_ops = NULL;
-       dev->rx_pkt_burst = NULL;
-       dev->tx_pkt_burst = NULL;
-       dev->data->mac_addrs = NULL;
+       (void)ice_dcf_dev_stop(dev);
 
+       ice_free_queues(dev);
+
+       ice_dcf_free_repr_info(adapter);
+       ice_dcf_uninit_parent_adapter(dev);
        ice_dcf_uninit_hw(dev, &adapter->real_hw);
+
+       return 0;
 }
 
-static void
-ice_dcf_queue_release(__rte_unused void *q)
+int
+ice_dcf_link_update(struct rte_eth_dev *dev,
+                   __rte_unused int wait_to_complete)
 {
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &ad->real_hw;
+       struct rte_eth_link new_link;
+
+       memset(&new_link, 0, sizeof(new_link));
+
+       /* Only read status info stored in VF, and the info is updated
+        * when receive LINK_CHANGE event from PF by virtchnl.
+        */
+       switch (hw->link_speed) {
+       case 10:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_10M;
+               break;
+       case 100:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_100M;
+               break;
+       case 1000:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_1G;
+               break;
+       case 10000:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_10G;
+               break;
+       case 20000:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_20G;
+               break;
+       case 25000:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_25G;
+               break;
+       case 40000:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_40G;
+               break;
+       case 50000:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_50G;
+               break;
+       case 100000:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_100G;
+               break;
+       default:
+               new_link.link_speed = RTE_ETH_SPEED_NUM_NONE;
+               break;
+       }
+
+       new_link.link_duplex = RTE_ETH_LINK_FULL_DUPLEX;
+       new_link.link_status = hw->link_up ? RTE_ETH_LINK_UP :
+                                            RTE_ETH_LINK_DOWN;
+       new_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
+                               RTE_ETH_LINK_SPEED_FIXED);
+
+       return rte_eth_linkstatus_set(dev, &new_link);
 }
 
+/* Add UDP tunneling port */
 static int
-ice_dcf_link_update(__rte_unused struct rte_eth_dev *dev,
-                   __rte_unused int wait_to_complete)
+ice_dcf_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
+                               struct rte_eth_udp_tunnel *udp_tunnel)
 {
-       return 0;
+       struct ice_dcf_adapter *adapter = dev->data->dev_private;
+       struct ice_adapter *parent_adapter = &adapter->parent;
+       struct ice_hw *parent_hw = &parent_adapter->hw;
+       int ret = 0;
+
+       if (!udp_tunnel)
+               return -EINVAL;
+
+       switch (udp_tunnel->prot_type) {
+       case RTE_ETH_TUNNEL_TYPE_VXLAN:
+               ret = ice_create_tunnel(parent_hw, TNL_VXLAN,
+                                       udp_tunnel->udp_port);
+               break;
+       case RTE_ETH_TUNNEL_TYPE_ECPRI:
+               ret = ice_create_tunnel(parent_hw, TNL_ECPRI,
+                                       udp_tunnel->udp_port);
+               break;
+       default:
+               PMD_DRV_LOG(ERR, "Invalid tunnel type");
+               ret = -EINVAL;
+               break;
+       }
+
+       return ret;
 }
 
+/* Delete UDP tunneling port */
 static int
-ice_dcf_rx_queue_setup(struct rte_eth_dev *dev,
-                      uint16_t rx_queue_id,
-                      __rte_unused uint16_t nb_rx_desc,
-                      __rte_unused unsigned int socket_id,
-                      __rte_unused const struct rte_eth_rxconf *rx_conf,
-                      __rte_unused struct rte_mempool *mb_pool)
+ice_dcf_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
+                               struct rte_eth_udp_tunnel *udp_tunnel)
 {
        struct ice_dcf_adapter *adapter = dev->data->dev_private;
+       struct ice_adapter *parent_adapter = &adapter->parent;
+       struct ice_hw *parent_hw = &parent_adapter->hw;
+       int ret = 0;
 
-       dev->data->rx_queues[rx_queue_id] = &adapter->rxqs[rx_queue_id];
+       if (!udp_tunnel)
+               return -EINVAL;
 
-       return 0;
+       switch (udp_tunnel->prot_type) {
+       case RTE_ETH_TUNNEL_TYPE_VXLAN:
+       case RTE_ETH_TUNNEL_TYPE_ECPRI:
+               ret = ice_destroy_tunnel(parent_hw, udp_tunnel->udp_port, 0);
+               break;
+       default:
+               PMD_DRV_LOG(ERR, "Invalid tunnel type");
+               ret = -EINVAL;
+               break;
+       }
+
+       return ret;
 }
 
 static int
-ice_dcf_tx_queue_setup(struct rte_eth_dev *dev,
-                      uint16_t tx_queue_id,
-                      __rte_unused uint16_t nb_tx_desc,
-                      __rte_unused unsigned int socket_id,
-                      __rte_unused const struct rte_eth_txconf *tx_conf)
+ice_dcf_tm_ops_get(struct rte_eth_dev *dev __rte_unused,
+               void *arg)
 {
-       struct ice_dcf_adapter *adapter = dev->data->dev_private;
+       if (!arg)
+               return -EINVAL;
 
-       dev->data->tx_queues[tx_queue_id] = &adapter->txqs[tx_queue_id];
+       *(const void **)arg = &ice_dcf_tm_ops;
 
        return 0;
 }
 
+static inline void
+ice_dcf_reset_hw(struct rte_eth_dev *eth_dev, struct ice_dcf_hw *hw)
+{
+       ice_dcf_uninit_hw(eth_dev, hw);
+       ice_dcf_init_hw(eth_dev, hw);
+}
+
+/* Check if reset has been triggered by PF */
+static inline bool
+ice_dcf_is_reset(struct rte_eth_dev *dev)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct iavf_hw *hw = &ad->real_hw.avf;
+
+       return !(IAVF_READ_REG(hw, IAVF_VF_ARQLEN1) &
+                IAVF_VF_ARQLEN1_ARQENABLE_MASK);
+}
+
+static int
+ice_dcf_dev_reset(struct rte_eth_dev *dev)
+{
+       struct ice_dcf_adapter *ad = dev->data->dev_private;
+       struct ice_dcf_hw *hw = &ad->real_hw;
+       int ret;
+
+       if (ice_dcf_is_reset(dev)) {
+               if (!ad->real_hw.resetting)
+                       ad->real_hw.resetting = true;
+               PMD_DRV_LOG(ERR, "The DCF has been reset by PF");
+
+               /*
+                * Simply reset hw to trigger an additional DCF enable/disable
+                * cycle which help to workaround the issue that kernel driver
+                * may not clean up resource during previous reset.
+                */
+               ice_dcf_reset_hw(dev, hw);
+       }
+
+       ret = ice_dcf_dev_uninit(dev);
+       if (ret)
+               return ret;
+
+       ret = ice_dcf_dev_init(dev);
+
+       return ret;
+}
+
 static const struct eth_dev_ops ice_dcf_eth_dev_ops = {
        .dev_start               = ice_dcf_dev_start,
        .dev_stop                = ice_dcf_dev_stop,
        .dev_close               = ice_dcf_dev_close,
+       .dev_reset               = ice_dcf_dev_reset,
        .dev_configure           = ice_dcf_dev_configure,
        .dev_infos_get           = ice_dcf_dev_info_get,
-       .rx_queue_setup          = ice_dcf_rx_queue_setup,
-       .tx_queue_setup          = ice_dcf_tx_queue_setup,
-       .rx_queue_release        = ice_dcf_queue_release,
-       .tx_queue_release        = ice_dcf_queue_release,
+       .rx_queue_setup          = ice_rx_queue_setup,
+       .tx_queue_setup          = ice_tx_queue_setup,
+       .rx_queue_release        = ice_dev_rx_queue_release,
+       .tx_queue_release        = ice_dev_tx_queue_release,
+       .rx_queue_start          = ice_dcf_rx_queue_start,
+       .tx_queue_start          = ice_dcf_tx_queue_start,
+       .rx_queue_stop           = ice_dcf_rx_queue_stop,
+       .tx_queue_stop           = ice_dcf_tx_queue_stop,
        .link_update             = ice_dcf_link_update,
        .stats_get               = ice_dcf_stats_get,
        .stats_reset             = ice_dcf_stats_reset,
@@ -208,7 +1095,10 @@ static const struct eth_dev_ops ice_dcf_eth_dev_ops = {
        .promiscuous_disable     = ice_dcf_dev_promiscuous_disable,
        .allmulticast_enable     = ice_dcf_dev_allmulticast_enable,
        .allmulticast_disable    = ice_dcf_dev_allmulticast_disable,
-       .filter_ctrl             = ice_dcf_dev_filter_ctrl,
+       .flow_ops_get            = ice_dcf_dev_flow_ops_get,
+       .udp_tunnel_port_add     = ice_dcf_dev_udp_tunnel_port_add,
+       .udp_tunnel_port_del     = ice_dcf_dev_udp_tunnel_port_del,
+       .tm_ops_get              = ice_dcf_tm_ops_get,
 };
 
 static int
@@ -223,15 +1113,17 @@ ice_dcf_dev_init(struct rte_eth_dev *eth_dev)
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return 0;
 
-       eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
-
+       adapter->real_hw.vc_event_msg_cb = ice_dcf_handle_pf_event_msg;
        if (ice_dcf_init_hw(eth_dev, &adapter->real_hw) != 0) {
                PMD_INIT_LOG(ERR, "Failed to init DCF hardware");
                return -1;
        }
 
-       rte_eth_random_addr(adapter->mac_addr.addr_bytes);
-       eth_dev->data->mac_addrs = &adapter->mac_addr;
+       if (ice_dcf_init_parent_adapter(eth_dev) != 0) {
+               PMD_INIT_LOG(ERR, "Failed to init DCF parent adapter");
+               ice_dcf_uninit_hw(eth_dev, &adapter->real_hw);
+               return -1;
+       }
 
        return 0;
 }
@@ -283,20 +1175,116 @@ exit:
        return ret;
 }
 
-static int eth_ice_dcf_pci_probe(__rte_unused struct rte_pci_driver *pci_drv,
-                            struct rte_pci_device *pci_dev)
+static int
+eth_ice_dcf_pci_probe(__rte_unused struct rte_pci_driver *pci_drv,
+                     struct rte_pci_device *pci_dev)
 {
+       struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
+       struct ice_dcf_vf_repr_param repr_param;
+       char repr_name[RTE_ETH_NAME_MAX_LEN];
+       struct ice_dcf_adapter *dcf_adapter;
+       struct rte_eth_dev *dcf_ethdev;
+       uint16_t dcf_vsi_id;
+       int i, ret;
+
        if (!ice_dcf_cap_selected(pci_dev->device.devargs))
                return 1;
 
-       return rte_eth_dev_pci_generic_probe(pci_dev,
-                                            sizeof(struct ice_dcf_adapter),
-                                            ice_dcf_dev_init);
+       ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, &eth_da);
+       if (ret)
+               return ret;
+
+       ret = rte_eth_dev_pci_generic_probe(pci_dev,
+                                           sizeof(struct ice_dcf_adapter),
+                                           ice_dcf_dev_init);
+       if (ret || !eth_da.nb_representor_ports)
+               return ret;
+       if (eth_da.type != RTE_ETH_REPRESENTOR_VF)
+               return -ENOTSUP;
+
+       dcf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
+       if (dcf_ethdev == NULL)
+               return -ENODEV;
+
+       dcf_adapter = dcf_ethdev->data->dev_private;
+       ret = ice_dcf_init_repr_info(dcf_adapter);
+       if (ret)
+               return ret;
+
+       if (eth_da.nb_representor_ports > dcf_adapter->real_hw.num_vfs ||
+           eth_da.nb_representor_ports >= RTE_MAX_ETHPORTS) {
+               PMD_DRV_LOG(ERR, "the number of port representors is too large: %u",
+                           eth_da.nb_representor_ports);
+               ice_dcf_free_repr_info(dcf_adapter);
+               return -EINVAL;
+       }
+
+       dcf_vsi_id = dcf_adapter->real_hw.vsi_id | VIRTCHNL_DCF_VF_VSI_VALID;
+
+       repr_param.dcf_eth_dev = dcf_ethdev;
+       repr_param.switch_domain_id = 0;
+
+       for (i = 0; i < eth_da.nb_representor_ports; i++) {
+               uint16_t vf_id = eth_da.representor_ports[i];
+               struct rte_eth_dev *vf_rep_eth_dev;
+
+               if (vf_id >= dcf_adapter->real_hw.num_vfs) {
+                       PMD_DRV_LOG(ERR, "VF ID %u is out of range (0 ~ %u)",
+                                   vf_id, dcf_adapter->real_hw.num_vfs - 1);
+                       ret = -EINVAL;
+                       break;
+               }
+
+               if (dcf_adapter->real_hw.vf_vsi_map[vf_id] == dcf_vsi_id) {
+                       PMD_DRV_LOG(ERR, "VF ID %u is DCF's ID.\n", vf_id);
+                       ret = -EINVAL;
+                       break;
+               }
+
+               repr_param.vf_id = vf_id;
+               snprintf(repr_name, sizeof(repr_name), "net_%s_representor_%u",
+                        pci_dev->device.name, vf_id);
+               ret = rte_eth_dev_create(&pci_dev->device, repr_name,
+                                        sizeof(struct ice_dcf_vf_repr),
+                                        NULL, NULL, ice_dcf_vf_repr_init,
+                                        &repr_param);
+               if (ret) {
+                       PMD_DRV_LOG(ERR, "failed to create DCF VF representor %s",
+                                   repr_name);
+                       break;
+               }
+
+               vf_rep_eth_dev = rte_eth_dev_allocated(repr_name);
+               if (!vf_rep_eth_dev) {
+                       PMD_DRV_LOG(ERR,
+                                   "Failed to find the ethdev for DCF VF representor: %s",
+                                   repr_name);
+                       ret = -ENODEV;
+                       break;
+               }
+
+               dcf_adapter->repr_infos[vf_id].vf_rep_eth_dev = vf_rep_eth_dev;
+               dcf_adapter->num_reprs++;
+       }
+
+       return ret;
 }
 
-static int eth_ice_dcf_pci_remove(struct rte_pci_device *pci_dev)
+static int
+eth_ice_dcf_pci_remove(struct rte_pci_device *pci_dev)
 {
-       return rte_eth_dev_pci_generic_remove(pci_dev, ice_dcf_dev_uninit);
+       struct rte_eth_dev *eth_dev;
+
+       eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
+       if (!eth_dev)
+               return 0;
+
+       if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
+               return rte_eth_dev_pci_generic_remove(pci_dev,
+                                                     ice_dcf_vf_repr_uninit);
+       else
+               return rte_eth_dev_pci_generic_remove(pci_dev,
+                                                     ice_dcf_dev_uninit);
 }
 
 static const struct rte_pci_id pci_id_ice_dcf_map[] = {