net/dpaa2: add support for VLAN filter and offload
[dpdk.git] / drivers / net / dpaa2 / dpaa2_ethdev.c
index 0b3dd67..46532b2 100644 (file)
 
 #include <fslmc_logs.h>
 #include <fslmc_vfio.h>
+#include <dpaa2_hw_pvt.h>
+#include <dpaa2_hw_mempool.h>
+#include <dpaa2_hw_dpio.h>
+
 #include "dpaa2_ethdev.h"
 
 static struct rte_dpaa2_driver rte_dpaa2_pmd;
+static int dpaa2_dev_uninit(struct rte_eth_dev *eth_dev);
+
+/**
+ * Atomically reads the link status information from global
+ * structure rte_eth_dev.
+ *
+ * @param dev
+ *   - Pointer to the structure rte_eth_dev to read from.
+ *   - Pointer to the buffer to be saved with the link status.
+ *
+ * @return
+ *   - On success, zero.
+ *   - On failure, negative value.
+ */
+static inline int
+dpaa2_dev_atomic_read_link_status(struct rte_eth_dev *dev,
+                                 struct rte_eth_link *link)
+{
+       struct rte_eth_link *dst = link;
+       struct rte_eth_link *src = &dev->data->dev_link;
+
+       if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
+                               *(uint64_t *)src) == 0)
+               return -1;
+
+       return 0;
+}
+
+/**
+ * Atomically writes the link status information into global
+ * structure rte_eth_dev.
+ *
+ * @param dev
+ *   - Pointer to the structure rte_eth_dev to read from.
+ *   - Pointer to the buffer to be saved with the link status.
+ *
+ * @return
+ *   - On success, zero.
+ *   - On failure, negative value.
+ */
+static inline int
+dpaa2_dev_atomic_write_link_status(struct rte_eth_dev *dev,
+                                  struct rte_eth_link *link)
+{
+       struct rte_eth_link *dst = &dev->data->dev_link;
+       struct rte_eth_link *src = link;
+
+       if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
+                               *(uint64_t *)src) == 0)
+               return -1;
+
+       return 0;
+}
+
+static int
+dpaa2_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = priv->hw;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return -1;
+       }
+
+       if (on)
+               ret = dpni_add_vlan_id(dpni, CMD_PRI_LOW,
+                                      priv->token, vlan_id);
+       else
+               ret = dpni_remove_vlan_id(dpni, CMD_PRI_LOW,
+                                         priv->token, vlan_id);
+
+       if (ret < 0)
+               PMD_DRV_LOG(ERR, "ret = %d Unable to add/rem vlan %d hwid =%d",
+                           ret, vlan_id, priv->hw_id);
+
+       return ret;
+}
+
+static void
+dpaa2_vlan_offload_set(struct rte_eth_dev *dev, int mask)
+{
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = priv->hw;
+       int ret;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (mask & ETH_VLAN_FILTER_MASK) {
+               if (dev->data->dev_conf.rxmode.hw_vlan_filter)
+                       ret = dpni_enable_vlan_filter(dpni, CMD_PRI_LOW,
+                                                     priv->token, true);
+               else
+                       ret = dpni_enable_vlan_filter(dpni, CMD_PRI_LOW,
+                                                     priv->token, false);
+               if (ret < 0)
+                       RTE_LOG(ERR, PMD, "Unable to set vlan filter ret = %d",
+                               ret);
+       }
+}
+
+static void
+dpaa2_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
+{
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+
+       PMD_INIT_FUNC_TRACE();
+
+       dev_info->if_index = priv->hw_id;
+
+       dev_info->max_mac_addrs = priv->max_mac_filters;
+       dev_info->max_rx_pktlen = DPAA2_MAX_RX_PKT_LEN;
+       dev_info->min_rx_bufsize = DPAA2_MIN_RX_BUF_SIZE;
+       dev_info->max_rx_queues = (uint16_t)priv->nb_rx_queues;
+       dev_info->max_tx_queues = (uint16_t)priv->nb_tx_queues;
+       dev_info->rx_offload_capa =
+               DEV_RX_OFFLOAD_IPV4_CKSUM |
+               DEV_RX_OFFLOAD_UDP_CKSUM |
+               DEV_RX_OFFLOAD_TCP_CKSUM |
+               DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM;
+       dev_info->tx_offload_capa =
+               DEV_TX_OFFLOAD_IPV4_CKSUM |
+               DEV_TX_OFFLOAD_UDP_CKSUM |
+               DEV_TX_OFFLOAD_TCP_CKSUM |
+               DEV_TX_OFFLOAD_SCTP_CKSUM |
+               DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
+       dev_info->speed_capa = ETH_LINK_SPEED_1G |
+                       ETH_LINK_SPEED_2_5G |
+                       ETH_LINK_SPEED_10G;
+}
+
+static int
+dpaa2_alloc_rx_tx_queues(struct rte_eth_dev *dev)
+{
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       uint16_t dist_idx;
+       uint32_t vq_id;
+       struct dpaa2_queue *mc_q, *mcq;
+       uint32_t tot_queues;
+       int i;
+       struct dpaa2_queue *dpaa2_q;
+
+       PMD_INIT_FUNC_TRACE();
+
+       tot_queues = priv->nb_rx_queues + priv->nb_tx_queues;
+       mc_q = rte_malloc(NULL, sizeof(struct dpaa2_queue) * tot_queues,
+                         RTE_CACHE_LINE_SIZE);
+       if (!mc_q) {
+               PMD_INIT_LOG(ERR, "malloc failed for rx/tx queues\n");
+               return -1;
+       }
+
+       for (i = 0; i < priv->nb_rx_queues; i++) {
+               mc_q->dev = dev;
+               priv->rx_vq[i] = mc_q++;
+               dpaa2_q = (struct dpaa2_queue *)priv->rx_vq[i];
+               dpaa2_q->q_storage = rte_malloc("dq_storage",
+                                       sizeof(struct queue_storage_info_t),
+                                       RTE_CACHE_LINE_SIZE);
+               if (!dpaa2_q->q_storage)
+                       goto fail;
+
+               memset(dpaa2_q->q_storage, 0,
+                      sizeof(struct queue_storage_info_t));
+               if (dpaa2_alloc_dq_storage(dpaa2_q->q_storage))
+                       goto fail;
+       }
+
+       for (i = 0; i < priv->nb_tx_queues; i++) {
+               mc_q->dev = dev;
+               mc_q->flow_id = 0xffff;
+               priv->tx_vq[i] = mc_q++;
+               dpaa2_q = (struct dpaa2_queue *)priv->tx_vq[i];
+               dpaa2_q->cscn = rte_malloc(NULL,
+                                          sizeof(struct qbman_result), 16);
+               if (!dpaa2_q->cscn)
+                       goto fail_tx;
+       }
+
+       vq_id = 0;
+       for (dist_idx = 0; dist_idx < priv->num_dist_per_tc[DPAA2_DEF_TC];
+            dist_idx++) {
+               mcq = (struct dpaa2_queue *)priv->rx_vq[vq_id];
+               mcq->tc_index = DPAA2_DEF_TC;
+               mcq->flow_id = dist_idx;
+               vq_id++;
+       }
+
+       return 0;
+fail_tx:
+       i -= 1;
+       while (i >= 0) {
+               dpaa2_q = (struct dpaa2_queue *)priv->tx_vq[i];
+               rte_free(dpaa2_q->cscn);
+               priv->tx_vq[i--] = NULL;
+       }
+       i = priv->nb_rx_queues;
+fail:
+       i -= 1;
+       mc_q = priv->rx_vq[0];
+       while (i >= 0) {
+               dpaa2_q = (struct dpaa2_queue *)priv->rx_vq[i];
+               dpaa2_free_dq_storage(dpaa2_q->q_storage);
+               rte_free(dpaa2_q->q_storage);
+               priv->rx_vq[i--] = NULL;
+       }
+       rte_free(mc_q);
+       return -1;
+}
+
+static int
+dpaa2_eth_dev_configure(struct rte_eth_dev *dev)
+{
+       struct rte_eth_dev_data *data = dev->data;
+       struct rte_eth_conf *eth_conf = &data->dev_conf;
+       int ret;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* Check for correct configuration */
+       if (eth_conf->rxmode.mq_mode != ETH_MQ_RX_RSS &&
+           data->nb_rx_queues > 1) {
+               PMD_INIT_LOG(ERR, "Distribution is not enabled, "
+                           "but Rx queues more than 1\n");
+               return -1;
+       }
+
+       if (eth_conf->rxmode.mq_mode == ETH_MQ_RX_RSS) {
+               /* Return in case number of Rx queues is 1 */
+               if (data->nb_rx_queues == 1)
+                       return 0;
+               ret = dpaa2_setup_flow_dist(dev,
+                               eth_conf->rx_adv_conf.rss_conf.rss_hf);
+               if (ret) {
+                       PMD_INIT_LOG(ERR, "unable to set flow distribution."
+                                    "please check queue config\n");
+                       return ret;
+               }
+       }
+       return 0;
+}
+
+/* Function to setup RX flow information. It contains traffic class ID,
+ * flow ID, destination configuration etc.
+ */
+static int
+dpaa2_dev_rx_queue_setup(struct rte_eth_dev *dev,
+                        uint16_t rx_queue_id,
+                        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)
+{
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       struct dpaa2_queue *dpaa2_q;
+       struct dpni_queue cfg;
+       uint8_t options = 0;
+       uint8_t flow_id;
+       uint32_t bpid;
+       int ret;
+
+       PMD_INIT_FUNC_TRACE();
+
+       PMD_INIT_LOG(DEBUG, "dev =%p, queue =%d, pool = %p, conf =%p",
+                    dev, rx_queue_id, mb_pool, rx_conf);
+
+       if (!priv->bp_list || priv->bp_list->mp != mb_pool) {
+               bpid = mempool_to_bpid(mb_pool);
+               ret = dpaa2_attach_bp_list(priv,
+                                          rte_dpaa2_bpid_info[bpid].bp_list);
+               if (ret)
+                       return ret;
+       }
+       dpaa2_q = (struct dpaa2_queue *)priv->rx_vq[rx_queue_id];
+       dpaa2_q->mb_pool = mb_pool; /**< mbuf pool to populate RX ring. */
+
+       /*Get the tc id and flow id from given VQ id*/
+       flow_id = rx_queue_id % priv->num_dist_per_tc[dpaa2_q->tc_index];
+       memset(&cfg, 0, sizeof(struct dpni_queue));
+
+       options = options | DPNI_QUEUE_OPT_USER_CTX;
+       cfg.user_context = (uint64_t)(dpaa2_q);
+
+       /*if ls2088 or rev2 device, enable the stashing */
+       if ((qbman_get_version() & 0xFFFF0000) > QMAN_REV_4000) {
+               options |= DPNI_QUEUE_OPT_FLC;
+               cfg.flc.stash_control = true;
+               cfg.flc.value &= 0xFFFFFFFFFFFFFFC0;
+               /* 00 00 00 - last 6 bit represent annotation, context stashing,
+                * data stashing setting 01 01 00 (0x14) to enable
+                * 1 line data, 1 line annotation
+                */
+               cfg.flc.value |= 0x14;
+       }
+       ret = dpni_set_queue(dpni, CMD_PRI_LOW, priv->token, DPNI_QUEUE_RX,
+                            dpaa2_q->tc_index, flow_id, options, &cfg);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error in setting the rx flow: = %d\n", ret);
+               return -1;
+       }
+
+       if (!(priv->flags & DPAA2_RX_TAILDROP_OFF)) {
+               struct dpni_taildrop taildrop;
+
+               taildrop.enable = 1;
+               /*enabling per rx queue congestion control */
+               taildrop.threshold = CONG_THRESHOLD_RX_Q;
+               taildrop.units = DPNI_CONGESTION_UNIT_BYTES;
+               PMD_INIT_LOG(DEBUG, "Enabling Early Drop on queue = %d",
+                            rx_queue_id);
+               ret = dpni_set_taildrop(dpni, CMD_PRI_LOW, priv->token,
+                                       DPNI_CP_QUEUE, DPNI_QUEUE_RX,
+                                       dpaa2_q->tc_index, flow_id, &taildrop);
+               if (ret) {
+                       PMD_INIT_LOG(ERR, "Error in setting the rx flow"
+                                    " err : = %d\n", ret);
+                       return -1;
+               }
+       }
+
+       dev->data->rx_queues[rx_queue_id] = dpaa2_q;
+       return 0;
+}
+
+static int
+dpaa2_dev_tx_queue_setup(struct rte_eth_dev *dev,
+                        uint16_t tx_queue_id,
+                        uint16_t nb_tx_desc __rte_unused,
+                        unsigned int socket_id __rte_unused,
+                        const struct rte_eth_txconf *tx_conf __rte_unused)
+{
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct dpaa2_queue *dpaa2_q = (struct dpaa2_queue *)
+               priv->tx_vq[tx_queue_id];
+       struct fsl_mc_io *dpni = priv->hw;
+       struct dpni_queue tx_conf_cfg;
+       struct dpni_queue tx_flow_cfg;
+       uint8_t options = 0, flow_id;
+       uint32_t tc_id;
+       int ret;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* Return if queue already configured */
+       if (dpaa2_q->flow_id != 0xffff)
+               return 0;
+
+       memset(&tx_conf_cfg, 0, sizeof(struct dpni_queue));
+       memset(&tx_flow_cfg, 0, sizeof(struct dpni_queue));
+
+       if (priv->num_tc == 1) {
+               tc_id = 0;
+               flow_id = tx_queue_id % priv->num_dist_per_tc[tc_id];
+       } else {
+               tc_id = tx_queue_id;
+               flow_id = 0;
+       }
+
+       ret = dpni_set_queue(dpni, CMD_PRI_LOW, priv->token, DPNI_QUEUE_TX,
+                            tc_id, flow_id, options, &tx_flow_cfg);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error in setting the tx flow: "
+                            "tc_id=%d, flow =%d ErrorCode = %x\n",
+                            tc_id, flow_id, -ret);
+                       return -1;
+       }
+
+       dpaa2_q->flow_id = flow_id;
+
+       if (tx_queue_id == 0) {
+               /*Set tx-conf and error configuration*/
+               ret = dpni_set_tx_confirmation_mode(dpni, CMD_PRI_LOW,
+                                                   priv->token,
+                                                   DPNI_CONF_DISABLE);
+               if (ret) {
+                       PMD_INIT_LOG(ERR, "Error in set tx conf mode settings"
+                                    " ErrorCode = %x", ret);
+                       return -1;
+               }
+       }
+       dpaa2_q->tc_index = tc_id;
+
+       if (priv->flags & DPAA2_TX_CGR_SUPPORT) {
+               struct dpni_congestion_notification_cfg cong_notif_cfg;
+
+               cong_notif_cfg.units = DPNI_CONGESTION_UNIT_BYTES;
+               /* Notify about congestion when the queue size is 32 KB */
+               cong_notif_cfg.threshold_entry = CONG_ENTER_TX_THRESHOLD;
+               /* Notify that the queue is not congested when the data in
+                * the queue is below this thershold.
+                */
+               cong_notif_cfg.threshold_exit = CONG_EXIT_TX_THRESHOLD;
+               cong_notif_cfg.message_ctx = 0;
+               cong_notif_cfg.message_iova = (uint64_t)dpaa2_q->cscn;
+               cong_notif_cfg.dest_cfg.dest_type = DPNI_DEST_NONE;
+               cong_notif_cfg.notification_mode =
+                                        DPNI_CONG_OPT_WRITE_MEM_ON_ENTER |
+                                        DPNI_CONG_OPT_WRITE_MEM_ON_EXIT |
+                                        DPNI_CONG_OPT_COHERENT_WRITE;
+
+               ret = dpni_set_congestion_notification(dpni, CMD_PRI_LOW,
+                                                      priv->token,
+                                                      DPNI_QUEUE_TX,
+                                                      tc_id,
+                                                      &cong_notif_cfg);
+               if (ret) {
+                       PMD_INIT_LOG(ERR,
+                          "Error in setting tx congestion notification: = %d",
+                          -ret);
+                       return -ret;
+               }
+       }
+       dev->data->tx_queues[tx_queue_id] = dpaa2_q;
+       return 0;
+}
+
+static void
+dpaa2_dev_rx_queue_release(void *q __rte_unused)
+{
+       PMD_INIT_FUNC_TRACE();
+}
+
+static void
+dpaa2_dev_tx_queue_release(void *q __rte_unused)
+{
+       PMD_INIT_FUNC_TRACE();
+}
+
+static const uint32_t *
+dpaa2_supported_ptypes_get(struct rte_eth_dev *dev)
+{
+       static const uint32_t ptypes[] = {
+               /*todo -= add more types */
+               RTE_PTYPE_L2_ETHER,
+               RTE_PTYPE_L3_IPV4,
+               RTE_PTYPE_L3_IPV4_EXT,
+               RTE_PTYPE_L3_IPV6,
+               RTE_PTYPE_L3_IPV6_EXT,
+               RTE_PTYPE_L4_TCP,
+               RTE_PTYPE_L4_UDP,
+               RTE_PTYPE_L4_SCTP,
+               RTE_PTYPE_L4_ICMP,
+               RTE_PTYPE_UNKNOWN
+       };
+
+       if (dev->rx_pkt_burst == dpaa2_dev_prefetch_rx)
+               return ptypes;
+       return NULL;
+}
+
+static int
+dpaa2_dev_start(struct rte_eth_dev *dev)
+{
+       struct rte_eth_dev_data *data = dev->data;
+       struct dpaa2_dev_priv *priv = data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       struct dpni_queue cfg;
+       struct dpni_error_cfg   err_cfg;
+       uint16_t qdid;
+       struct dpni_queue_id qid;
+       struct dpaa2_queue *dpaa2_q;
+       int ret, i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       ret = dpni_enable(dpni, CMD_PRI_LOW, priv->token);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failure %d in enabling dpni %d device\n",
+                            ret, priv->hw_id);
+               return ret;
+       }
+
+       ret = dpni_get_qdid(dpni, CMD_PRI_LOW, priv->token,
+                           DPNI_QUEUE_TX, &qdid);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error to get qdid:ErrorCode = %d\n", ret);
+               return ret;
+       }
+       priv->qdid = qdid;
+
+       for (i = 0; i < data->nb_rx_queues; i++) {
+               dpaa2_q = (struct dpaa2_queue *)data->rx_queues[i];
+               ret = dpni_get_queue(dpni, CMD_PRI_LOW, priv->token,
+                                    DPNI_QUEUE_RX, dpaa2_q->tc_index,
+                                      dpaa2_q->flow_id, &cfg, &qid);
+               if (ret) {
+                       PMD_INIT_LOG(ERR, "Error to get flow "
+                                    "information Error code = %d\n", ret);
+                       return ret;
+               }
+               dpaa2_q->fqid = qid.fqid;
+       }
+
+       ret = dpni_set_offload(dpni, CMD_PRI_LOW, priv->token,
+                              DPNI_OFF_RX_L3_CSUM, true);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error to set RX l3 csum:Error = %d\n", ret);
+               return ret;
+       }
+
+       ret = dpni_set_offload(dpni, CMD_PRI_LOW, priv->token,
+                              DPNI_OFF_RX_L4_CSUM, true);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error to get RX l4 csum:Error = %d\n", ret);
+               return ret;
+       }
+
+       ret = dpni_set_offload(dpni, CMD_PRI_LOW, priv->token,
+                              DPNI_OFF_TX_L3_CSUM, true);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error to set TX l3 csum:Error = %d\n", ret);
+               return ret;
+       }
+
+       ret = dpni_set_offload(dpni, CMD_PRI_LOW, priv->token,
+                              DPNI_OFF_TX_L4_CSUM, true);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error to get TX l4 csum:Error = %d\n", ret);
+               return ret;
+       }
+
+       /*checksum errors, send them to normal path and set it in annotation */
+       err_cfg.errors = DPNI_ERROR_L3CE | DPNI_ERROR_L4CE;
+
+       err_cfg.error_action = DPNI_ERROR_ACTION_CONTINUE;
+       err_cfg.set_frame_annotation = true;
+
+       ret = dpni_set_errors_behavior(dpni, CMD_PRI_LOW,
+                                      priv->token, &err_cfg);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error to dpni_set_errors_behavior:"
+                            "code = %d\n", ret);
+               return ret;
+       }
+       /* VLAN Offload Settings */
+       if (priv->max_vlan_filters)
+               dpaa2_vlan_offload_set(dev, ETH_VLAN_FILTER_MASK);
+
+       return 0;
+}
+
+/**
+ *  This routine disables all traffic on the adapter by issuing a
+ *  global reset on the MAC.
+ */
+static void
+dpaa2_dev_stop(struct rte_eth_dev *dev)
+{
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       int ret;
+       struct rte_eth_link link;
+
+       PMD_INIT_FUNC_TRACE();
+
+       ret = dpni_disable(dpni, CMD_PRI_LOW, priv->token);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failure (ret %d) in disabling dpni %d dev\n",
+                            ret, priv->hw_id);
+               return;
+       }
+
+       /* clear the recorded link status */
+       memset(&link, 0, sizeof(link));
+       dpaa2_dev_atomic_write_link_status(dev, &link);
+}
+
+static void
+dpaa2_dev_close(struct rte_eth_dev *dev)
+{
+       struct rte_eth_dev_data *data = dev->data;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       int i, ret;
+       struct dpaa2_queue *dpaa2_q;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < data->nb_tx_queues; i++) {
+               dpaa2_q = (struct dpaa2_queue *)data->tx_queues[i];
+               if (!dpaa2_q->cscn) {
+                       rte_free(dpaa2_q->cscn);
+                       dpaa2_q->cscn = NULL;
+               }
+       }
+
+       /* Clean the device first */
+       ret = dpni_reset(dpni, CMD_PRI_LOW, priv->token);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Failure cleaning dpni device with"
+                            " error code %d\n", ret);
+               return;
+       }
+}
+
+static void
+dpaa2_dev_promiscuous_enable(
+               struct rte_eth_dev *dev)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return;
+       }
+
+       ret = dpni_set_unicast_promisc(dpni, CMD_PRI_LOW, priv->token, true);
+       if (ret < 0)
+               RTE_LOG(ERR, PMD, "Unable to enable U promisc mode %d", ret);
+
+       ret = dpni_set_multicast_promisc(dpni, CMD_PRI_LOW, priv->token, true);
+       if (ret < 0)
+               RTE_LOG(ERR, PMD, "Unable to enable M promisc mode %d", ret);
+}
+
+static void
+dpaa2_dev_promiscuous_disable(
+               struct rte_eth_dev *dev)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return;
+       }
+
+       ret = dpni_set_unicast_promisc(dpni, CMD_PRI_LOW, priv->token, false);
+       if (ret < 0)
+               RTE_LOG(ERR, PMD, "Unable to disable U promisc mode %d", ret);
+
+       if (dev->data->all_multicast == 0) {
+               ret = dpni_set_multicast_promisc(dpni, CMD_PRI_LOW,
+                                                priv->token, false);
+               if (ret < 0)
+                       RTE_LOG(ERR, PMD, "Unable to disable M promisc mode %d",
+                               ret);
+       }
+}
+
+static void
+dpaa2_dev_allmulticast_enable(
+               struct rte_eth_dev *dev)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return;
+       }
+
+       ret = dpni_set_multicast_promisc(dpni, CMD_PRI_LOW, priv->token, true);
+       if (ret < 0)
+               RTE_LOG(ERR, PMD, "Unable to enable multicast mode %d", ret);
+}
+
+static void
+dpaa2_dev_allmulticast_disable(struct rte_eth_dev *dev)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return;
+       }
+
+       /* must remain on for all promiscuous */
+       if (dev->data->promiscuous == 1)
+               return;
+
+       ret = dpni_set_multicast_promisc(dpni, CMD_PRI_LOW, priv->token, false);
+       if (ret < 0)
+               RTE_LOG(ERR, PMD, "Unable to disable multicast mode %d", ret);
+}
+
+static int
+dpaa2_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       uint32_t frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return -EINVAL;
+       }
+
+       /* check that mtu is within the allowed range */
+       if ((mtu < ETHER_MIN_MTU) || (frame_size > DPAA2_MAX_RX_PKT_LEN))
+               return -EINVAL;
+
+       /* Set the Max Rx frame length as 'mtu' +
+        * Maximum Ethernet header length
+        */
+       ret = dpni_set_max_frame_length(dpni, CMD_PRI_LOW, priv->token,
+                                       mtu + ETH_VLAN_HLEN);
+       if (ret) {
+               PMD_DRV_LOG(ERR, "setting the max frame length failed");
+               return -1;
+       }
+       PMD_DRV_LOG(INFO, "MTU is configured %d for the device\n", mtu);
+       return 0;
+}
+
+static int
+dpaa2_dev_add_mac_addr(struct rte_eth_dev *dev,
+                      struct ether_addr *addr,
+                      __rte_unused uint32_t index,
+                      __rte_unused uint32_t pool)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return -1;
+       }
+
+       ret = dpni_add_mac_addr(dpni, CMD_PRI_LOW,
+                               priv->token, addr->addr_bytes);
+       if (ret)
+               RTE_LOG(ERR, PMD, "error: Adding the MAC ADDR failed:"
+                       " err = %d", ret);
+       return 0;
+}
+
+static void
+dpaa2_dev_remove_mac_addr(struct rte_eth_dev *dev,
+                         uint32_t index)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       struct rte_eth_dev_data *data = dev->data;
+       struct ether_addr *macaddr;
+
+       PMD_INIT_FUNC_TRACE();
+
+       macaddr = &data->mac_addrs[index];
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return;
+       }
+
+       ret = dpni_remove_mac_addr(dpni, CMD_PRI_LOW,
+                                  priv->token, macaddr->addr_bytes);
+       if (ret)
+               RTE_LOG(ERR, PMD, "error: Removing the MAC ADDR failed:"
+                       " err = %d", ret);
+}
+
+static void
+dpaa2_dev_set_mac_addr(struct rte_eth_dev *dev,
+                      struct ether_addr *addr)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return;
+       }
+
+       ret = dpni_set_primary_mac_addr(dpni, CMD_PRI_LOW,
+                                       priv->token, addr->addr_bytes);
+
+       if (ret)
+               RTE_LOG(ERR, PMD, "error: Setting the MAC ADDR failed %d", ret);
+}
+static
+void dpaa2_dev_stats_get(struct rte_eth_dev *dev,
+                        struct rte_eth_stats *stats)
+{
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       int32_t  retcode;
+       uint8_t page0 = 0, page1 = 1, page2 = 2;
+       union dpni_statistics value;
+
+       memset(&value, 0, sizeof(union dpni_statistics));
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (!dpni) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return;
+       }
+
+       if (!stats) {
+               RTE_LOG(ERR, PMD, "stats is NULL");
+               return;
+       }
+
+       /*Get Counters from page_0*/
+       retcode = dpni_get_statistics(dpni, CMD_PRI_LOW, priv->token,
+                                     page0, &value);
+       if (retcode)
+               goto err;
+
+       stats->ipackets = value.page_0.ingress_all_frames;
+       stats->ibytes = value.page_0.ingress_all_bytes;
+
+       /*Get Counters from page_1*/
+       retcode = dpni_get_statistics(dpni, CMD_PRI_LOW, priv->token,
+                                     page1, &value);
+       if (retcode)
+               goto err;
+
+       stats->opackets = value.page_1.egress_all_frames;
+       stats->obytes = value.page_1.egress_all_bytes;
+
+       /*Get Counters from page_2*/
+       retcode = dpni_get_statistics(dpni, CMD_PRI_LOW, priv->token,
+                                     page2, &value);
+       if (retcode)
+               goto err;
+
+       /* Ingress drop frame count due to configured rules */
+       stats->ierrors = value.page_2.ingress_filtered_frames;
+       /* Ingress drop frame count due to error */
+       stats->ierrors += value.page_2.ingress_discarded_frames;
+
+       stats->oerrors = value.page_2.egress_discarded_frames;
+       stats->imissed = value.page_2.ingress_nobuffer_discards;
+
+       return;
+
+err:
+       RTE_LOG(ERR, PMD, "Operation not completed:Error Code = %d\n", retcode);
+       return;
+};
+
+static
+void dpaa2_dev_stats_reset(struct rte_eth_dev *dev)
+{
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       int32_t  retcode;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "dpni is NULL");
+               return;
+       }
+
+       retcode =  dpni_reset_statistics(dpni, CMD_PRI_LOW, priv->token);
+       if (retcode)
+               goto error;
+
+       return;
+
+error:
+       RTE_LOG(ERR, PMD, "Operation not completed:Error Code = %d\n", retcode);
+       return;
+};
+
+/* return 0 means link status changed, -1 means not changed */
+static int
+dpaa2_dev_link_update(struct rte_eth_dev *dev,
+                       int wait_to_complete __rte_unused)
+{
+       int ret;
+       struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       struct rte_eth_link link, old;
+       struct dpni_link_state state = {0};
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (dpni == NULL) {
+               RTE_LOG(ERR, PMD, "error : dpni is NULL");
+               return 0;
+       }
+       memset(&old, 0, sizeof(old));
+       dpaa2_dev_atomic_read_link_status(dev, &old);
+
+       ret = dpni_get_link_state(dpni, CMD_PRI_LOW, priv->token, &state);
+       if (ret < 0) {
+               RTE_LOG(ERR, PMD, "error: dpni_get_link_state %d", ret);
+               return -1;
+       }
+
+       if ((old.link_status == state.up) && (old.link_speed == state.rate)) {
+               RTE_LOG(DEBUG, PMD, "No change in status\n");
+               return -1;
+       }
+
+       memset(&link, 0, sizeof(struct rte_eth_link));
+       link.link_status = state.up;
+       link.link_speed = state.rate;
+
+       if (state.options & DPNI_LINK_OPT_HALF_DUPLEX)
+               link.link_duplex = ETH_LINK_HALF_DUPLEX;
+       else
+               link.link_duplex = ETH_LINK_FULL_DUPLEX;
+
+       dpaa2_dev_atomic_write_link_status(dev, &link);
+
+       if (link.link_status)
+               PMD_DRV_LOG(INFO, "Port %d Link is Up\n", dev->data->port_id);
+       else
+               PMD_DRV_LOG(INFO, "Port %d Link is Down\n", dev->data->port_id);
+       return 0;
+}
+
+static struct eth_dev_ops dpaa2_ethdev_ops = {
+       .dev_configure    = dpaa2_eth_dev_configure,
+       .dev_start            = dpaa2_dev_start,
+       .dev_stop             = dpaa2_dev_stop,
+       .dev_close            = dpaa2_dev_close,
+       .promiscuous_enable   = dpaa2_dev_promiscuous_enable,
+       .promiscuous_disable  = dpaa2_dev_promiscuous_disable,
+       .allmulticast_enable  = dpaa2_dev_allmulticast_enable,
+       .allmulticast_disable = dpaa2_dev_allmulticast_disable,
+       .link_update       = dpaa2_dev_link_update,
+       .stats_get             = dpaa2_dev_stats_get,
+       .stats_reset       = dpaa2_dev_stats_reset,
+       .dev_infos_get     = dpaa2_dev_info_get,
+       .dev_supported_ptypes_get = dpaa2_supported_ptypes_get,
+       .mtu_set           = dpaa2_dev_mtu_set,
+       .vlan_filter_set      = dpaa2_vlan_filter_set,
+       .vlan_offload_set     = dpaa2_vlan_offload_set,
+       .rx_queue_setup    = dpaa2_dev_rx_queue_setup,
+       .rx_queue_release  = dpaa2_dev_rx_queue_release,
+       .tx_queue_setup    = dpaa2_dev_tx_queue_setup,
+       .tx_queue_release  = dpaa2_dev_tx_queue_release,
+       .mac_addr_add         = dpaa2_dev_add_mac_addr,
+       .mac_addr_remove      = dpaa2_dev_remove_mac_addr,
+       .mac_addr_set         = dpaa2_dev_set_mac_addr,
+};
 
 static int
 dpaa2_dev_init(struct rte_eth_dev *eth_dev)
 {
+       struct rte_device *dev = eth_dev->device;
+       struct rte_dpaa2_device *dpaa2_dev;
+       struct fsl_mc_io *dpni_dev;
+       struct dpni_attr attr;
+       struct dpaa2_dev_priv *priv = eth_dev->data->dev_private;
+       struct dpni_buffer_layout layout;
+       int i, ret, hw_id;
+
        PMD_INIT_FUNC_TRACE();
 
        /* For secondary processes, the primary has done all the work */
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return 0;
 
+       dpaa2_dev = container_of(dev, struct rte_dpaa2_device, device);
+
+       hw_id = dpaa2_dev->object_id;
+
+       dpni_dev = rte_malloc(NULL, sizeof(struct fsl_mc_io), 0);
+       if (!dpni_dev) {
+               PMD_INIT_LOG(ERR, "malloc failed for dpni device\n");
+               return -1;
+       }
+
+       dpni_dev->regs = rte_mcp_ptr_list[0];
+       ret = dpni_open(dpni_dev, CMD_PRI_LOW, hw_id, &priv->token);
+       if (ret) {
+               PMD_INIT_LOG(ERR,
+                            "Failure in opening dpni@%d with err code %d\n",
+                            hw_id, ret);
+               rte_free(dpni_dev);
+               return -1;
+       }
+
+       /* Clean the device first */
+       ret = dpni_reset(dpni_dev, CMD_PRI_LOW, priv->token);
+       if (ret) {
+               PMD_INIT_LOG(ERR,
+                            "Failure cleaning dpni@%d with err code %d\n",
+                            hw_id, ret);
+               goto init_err;
+       }
+
+       ret = dpni_get_attributes(dpni_dev, CMD_PRI_LOW, priv->token, &attr);
+       if (ret) {
+               PMD_INIT_LOG(ERR,
+                            "Failure in get dpni@%d attribute, err code %d\n",
+                            hw_id, ret);
+               goto init_err;
+       }
+
+       priv->num_tc = attr.num_tcs;
+       for (i = 0; i < attr.num_tcs; i++) {
+               priv->num_dist_per_tc[i] = attr.num_queues;
+               break;
+       }
+
+       /* Distribution is per Tc only,
+        * so choosing RX queues from default TC only
+        */
+       priv->nb_rx_queues = priv->num_dist_per_tc[DPAA2_DEF_TC];
+
+       if (attr.num_tcs == 1)
+               priv->nb_tx_queues = attr.num_queues;
+       else
+               priv->nb_tx_queues = attr.num_tcs;
+
+       PMD_INIT_LOG(DEBUG, "num_tc %d", priv->num_tc);
+       PMD_INIT_LOG(DEBUG, "nb_rx_queues %d", priv->nb_rx_queues);
+
+       priv->hw = dpni_dev;
+       priv->hw_id = hw_id;
+       priv->options = attr.options;
+       priv->max_mac_filters = attr.mac_filter_entries;
+       priv->max_vlan_filters = attr.vlan_filter_entries;
+       priv->flags = 0;
+
+       priv->flags |= DPAA2_TX_CGR_SUPPORT;
+       PMD_INIT_LOG(INFO, "Enable the tx congestion control support");
+
+       /* Allocate memory for hardware structure for queues */
+       ret = dpaa2_alloc_rx_tx_queues(eth_dev);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "dpaa2_alloc_rx_tx_queuesFailed\n");
+               goto init_err;
+       }
+
+       /* Allocate memory for storing MAC addresses */
+       eth_dev->data->mac_addrs = rte_zmalloc("dpni",
+               ETHER_ADDR_LEN * attr.mac_filter_entries, 0);
+       if (eth_dev->data->mac_addrs == NULL) {
+               PMD_INIT_LOG(ERR,
+                  "Failed to allocate %d bytes needed to store MAC addresses",
+                            ETHER_ADDR_LEN * attr.mac_filter_entries);
+               ret = -ENOMEM;
+               goto init_err;
+       }
+
+       ret = dpni_get_primary_mac_addr(dpni_dev, CMD_PRI_LOW,
+                                       priv->token,
+                       (uint8_t *)(eth_dev->data->mac_addrs[0].addr_bytes));
+       if (ret) {
+               PMD_INIT_LOG(ERR, "DPNI get mac address failed:Err Code = %d\n",
+                            ret);
+               goto init_err;
+       }
+
+       /* ... tx buffer layout ... */
+       memset(&layout, 0, sizeof(struct dpni_buffer_layout));
+       layout.options = DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
+       layout.pass_frame_status = 1;
+       ret = dpni_set_buffer_layout(dpni_dev, CMD_PRI_LOW, priv->token,
+                                    DPNI_QUEUE_TX, &layout);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error (%d) in setting tx buffer layout",
+                            ret);
+               goto init_err;
+       }
+
+       /* ... tx-conf and error buffer layout ... */
+       memset(&layout, 0, sizeof(struct dpni_buffer_layout));
+       layout.options = DPNI_BUF_LAYOUT_OPT_FRAME_STATUS;
+       layout.pass_frame_status = 1;
+       ret = dpni_set_buffer_layout(dpni_dev, CMD_PRI_LOW, priv->token,
+                                    DPNI_QUEUE_TX_CONFIRM, &layout);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Error (%d) in setting tx-conf buffer layout",
+                            ret);
+               goto init_err;
+       }
+
+       eth_dev->dev_ops = &dpaa2_ethdev_ops;
        eth_dev->data->drv_name = rte_dpaa2_pmd.driver.name;
 
+       eth_dev->rx_pkt_burst = dpaa2_dev_prefetch_rx;
+       eth_dev->tx_pkt_burst = dpaa2_dev_tx;
+       rte_fslmc_vfio_dmamap();
+
        return 0;
+init_err:
+       dpaa2_dev_uninit(eth_dev);
+       return ret;
 }
 
 static int
-dpaa2_dev_uninit(struct rte_eth_dev *eth_dev __rte_unused)
+dpaa2_dev_uninit(struct rte_eth_dev *eth_dev)
 {
+       struct dpaa2_dev_priv *priv = eth_dev->data->dev_private;
+       struct fsl_mc_io *dpni = (struct fsl_mc_io *)priv->hw;
+       int i, ret;
+       struct dpaa2_queue *dpaa2_q;
+
        PMD_INIT_FUNC_TRACE();
 
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return -EPERM;
 
+       if (!dpni) {
+               PMD_INIT_LOG(WARNING, "Already closed or not started");
+               return -1;
+       }
+
+       dpaa2_dev_close(eth_dev);
+
+       if (priv->rx_vq[0]) {
+               /* cleaning up queue storage */
+               for (i = 0; i < priv->nb_rx_queues; i++) {
+                       dpaa2_q = (struct dpaa2_queue *)priv->rx_vq[i];
+                       if (dpaa2_q->q_storage)
+                               rte_free(dpaa2_q->q_storage);
+               }
+               /*free the all queue memory */
+               rte_free(priv->rx_vq[0]);
+               priv->rx_vq[0] = NULL;
+       }
+
+       /* free memory for storing MAC addresses */
+       if (eth_dev->data->mac_addrs) {
+               rte_free(eth_dev->data->mac_addrs);
+               eth_dev->data->mac_addrs = NULL;
+       }
+
+       /* Close the device at underlying layer*/
+       ret = dpni_close(dpni, CMD_PRI_LOW, priv->token);
+       if (ret) {
+               PMD_INIT_LOG(ERR,
+                            "Failure closing dpni device with err code %d\n",
+                            ret);
+       }
+
+       /* Free the allocated memory for ethernet private data and dpni*/
+       priv->hw = NULL;
+       rte_free(dpni);
+
+       eth_dev->dev_ops = NULL;
+       eth_dev->rx_pkt_burst = NULL;
+       eth_dev->tx_pkt_burst = NULL;
+
        return 0;
 }