* Observance of NAPI budget is not our concern, leaving that to the caller.
  */
 static int consume_frames(struct dpaa2_eth_channel *ch,
-                         enum dpaa2_eth_fq_type *type)
+                         struct dpaa2_eth_fq **src)
 {
        struct dpaa2_eth_priv *priv = ch->priv;
        struct dpaa2_eth_fq *fq = NULL;
        ch->stats.frames += cleaned;
 
        /* A dequeue operation only pulls frames from a single queue
-        * into the store. Return the frame queue type as an out param.
+        * into the store. Return the frame queue as an out param.
         */
-       if (type)
-               *type = fq->type;
+       if (src)
+               *src = fq;
 
        return cleaned;
 }
        struct rtnl_link_stats64 *percpu_stats;
        struct dpaa2_eth_drv_stats *percpu_extras;
        struct dpaa2_eth_fq *fq;
+       struct netdev_queue *nq;
        u16 queue_mapping;
        unsigned int needed_headroom;
+       u32 fd_len;
        int err, i;
 
        percpu_stats = this_cpu_ptr(priv->percpu_stats);
                /* Clean up everything, including freeing the skb */
                free_tx_fd(priv, &fd);
        } else {
+               fd_len = dpaa2_fd_get_len(&fd);
                percpu_stats->tx_packets++;
-               percpu_stats->tx_bytes += dpaa2_fd_get_len(&fd);
+               percpu_stats->tx_bytes += fd_len;
+
+               nq = netdev_get_tx_queue(net_dev, queue_mapping);
+               netdev_tx_sent_queue(nq, fd_len);
        }
 
        return NETDEV_TX_OK;
 static void dpaa2_eth_tx_conf(struct dpaa2_eth_priv *priv,
                              struct dpaa2_eth_channel *ch __always_unused,
                              const struct dpaa2_fd *fd,
-                             struct dpaa2_eth_fq *fq __always_unused)
+                             struct dpaa2_eth_fq *fq)
 {
        struct rtnl_link_stats64 *percpu_stats;
        struct dpaa2_eth_drv_stats *percpu_extras;
+       u32 fd_len = dpaa2_fd_get_len(fd);
        u32 fd_errors;
 
        /* Tracing point */
 
        percpu_extras = this_cpu_ptr(priv->percpu_extras);
        percpu_extras->tx_conf_frames++;
-       percpu_extras->tx_conf_bytes += dpaa2_fd_get_len(fd);
+       percpu_extras->tx_conf_bytes += fd_len;
+
+       fq->dq_frames++;
+       fq->dq_bytes += fd_len;
 
        /* Check frame errors in the FD field */
        fd_errors = dpaa2_fd_get_ctrl(fd) & DPAA2_FD_TX_ERR_MASK;
        struct dpaa2_eth_channel *ch;
        struct dpaa2_eth_priv *priv;
        int rx_cleaned = 0, txconf_cleaned = 0;
-       enum dpaa2_eth_fq_type type = 0;
-       int store_cleaned;
+       struct dpaa2_eth_fq *fq, *txc_fq = NULL;
+       struct netdev_queue *nq;
+       int store_cleaned, work_done;
        int err;
 
        ch = container_of(napi, struct dpaa2_eth_channel, napi);
                /* Refill pool if appropriate */
                refill_pool(priv, ch, priv->bpid);
 
-               store_cleaned = consume_frames(ch, &type);
-               if (type == DPAA2_RX_FQ)
+               store_cleaned = consume_frames(ch, &fq);
+               if (!store_cleaned)
+                       break;
+               if (fq->type == DPAA2_RX_FQ) {
                        rx_cleaned += store_cleaned;
-               else
+               } else {
                        txconf_cleaned += store_cleaned;
+                       /* We have a single Tx conf FQ on this channel */
+                       txc_fq = fq;
+               }
 
                /* If we either consumed the whole NAPI budget with Rx frames
                 * or we reached the Tx confirmations threshold, we're done.
                 */
                if (rx_cleaned >= budget ||
-                   txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI)
-                       return budget;
+                   txconf_cleaned >= DPAA2_ETH_TXCONF_PER_NAPI) {
+                       work_done = budget;
+                       goto out;
+               }
        } while (store_cleaned);
 
        /* We didn't consume the entire budget, so finish napi and
        WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
                  ch->nctx.desired_cpu);
 
-       return max(rx_cleaned, 1);
+       work_done = max(rx_cleaned, 1);
+
+out:
+       if (txc_fq) {
+               nq = netdev_get_tx_queue(priv->net_dev, txc_fq->flowid);
+               netdev_tx_completed_queue(nq, txc_fq->dq_frames,
+                                         txc_fq->dq_bytes);
+               txc_fq->dq_frames = 0;
+               txc_fq->dq_bytes = 0;
+       }
+
+       return work_done;
 }
 
 static void enable_ch_napi(struct dpaa2_eth_priv *priv)