net/enic: reset VXLAN port during initialization
[dpdk.git] / drivers / net / enic / enic_main.c
index b3b4e96..fd940c5 100644 (file)
@@ -69,12 +69,12 @@ enic_rxmbuf_queue_release(__rte_unused struct enic *enic, struct vnic_rq *rq)
        }
 }
 
-static void enic_free_wq_buf(struct vnic_wq_buf *buf)
+static void enic_free_wq_buf(struct rte_mbuf **buf)
 {
-       struct rte_mbuf *mbuf = (struct rte_mbuf *)buf->mb;
+       struct rte_mbuf *mbuf = *buf;
 
        rte_pktmbuf_free_seg(mbuf);
-       buf->mb = NULL;
+       *buf = NULL;
 }
 
 static void enic_log_q_error(struct enic *enic)
@@ -320,6 +320,8 @@ enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
         * enic_start_rq().
         */
        rq->need_initial_post = true;
+       /* Initialize fetch index while RQ is disabled */
+       iowrite32(0, &rq->ctrl->fetch_index);
        return 0;
 }
 
@@ -345,7 +347,6 @@ enic_initial_post_rx(struct enic *enic, struct vnic_rq *rq)
        dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
                enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
        iowrite32(rq->posted_index, &rq->ctrl->posted_index);
-       iowrite32(0, &rq->ctrl->fetch_index);
        rte_rmb();
        rq->need_initial_post = false;
 }
@@ -492,6 +493,42 @@ static void enic_rxq_intr_deinit(struct enic *enic)
        }
 }
 
+static void enic_prep_wq_for_simple_tx(struct enic *enic, uint16_t queue_idx)
+{
+       struct wq_enet_desc *desc;
+       struct vnic_wq *wq;
+       unsigned int i;
+
+       /*
+        * Fill WQ descriptor fields that never change. Every descriptor is
+        * one packet, so set EOP. Also set CQ_ENTRY every ENIC_WQ_CQ_THRESH
+        * descriptors (i.e. request one completion update every 32 packets).
+        */
+       wq = &enic->wq[queue_idx];
+       desc = (struct wq_enet_desc *)wq->ring.descs;
+       for (i = 0; i < wq->ring.desc_count; i++, desc++) {
+               desc->header_length_flags = 1 << WQ_ENET_FLAGS_EOP_SHIFT;
+               if (i % ENIC_WQ_CQ_THRESH == ENIC_WQ_CQ_THRESH - 1)
+                       desc->header_length_flags |=
+                               (1 << WQ_ENET_FLAGS_CQ_ENTRY_SHIFT);
+       }
+}
+
+static void pick_rx_handler(struct enic *enic)
+{
+       struct rte_eth_dev *eth_dev;
+
+       /* Use the non-scatter, simplified RX handler if possible. */
+       eth_dev = enic->rte_dev;
+       if (enic->rq_count > 0 && enic->rq[0].data_queue_enable == 0) {
+               PMD_INIT_LOG(DEBUG, " use the non-scatter Rx handler");
+               eth_dev->rx_pkt_burst = &enic_noscatter_recv_pkts;
+       } else {
+               PMD_INIT_LOG(DEBUG, " use the normal Rx handler");
+               eth_dev->rx_pkt_burst = &enic_recv_pkts;
+       }
+}
+
 int enic_enable(struct enic *enic)
 {
        unsigned int index;
@@ -534,6 +571,22 @@ int enic_enable(struct enic *enic)
                }
        }
 
+       /*
+        * Use the simple TX handler if possible. All offloads must be
+        * disabled.
+        */
+       if (eth_dev->data->dev_conf.txmode.offloads == 0) {
+               PMD_INIT_LOG(DEBUG, " use the simple tx handler");
+               eth_dev->tx_pkt_burst = &enic_simple_xmit_pkts;
+               for (index = 0; index < enic->wq_count; index++)
+                       enic_prep_wq_for_simple_tx(enic, index);
+       } else {
+               PMD_INIT_LOG(DEBUG, " use the default tx handler");
+               eth_dev->tx_pkt_burst = &enic_xmit_pkts;
+       }
+
+       pick_rx_handler(enic);
+
        for (index = 0; index < enic->wq_count; index++)
                enic_start_wq(enic, index);
        for (index = 0; index < enic->rq_count; index++)
@@ -586,6 +639,19 @@ void enic_free_rq(void *rxq)
        enic = vnic_dev_priv(rq_sop->vdev);
        rq_data = &enic->rq[rq_sop->data_queue_idx];
 
+       if (rq_sop->free_mbufs) {
+               struct rte_mbuf **mb;
+               int i;
+
+               mb = rq_sop->free_mbufs;
+               for (i = ENIC_RX_BURST_MAX - rq_sop->num_free_mbufs;
+                    i < ENIC_RX_BURST_MAX; i++)
+                       rte_pktmbuf_free(mb[i]);
+               rte_free(rq_sop->free_mbufs);
+               rq_sop->free_mbufs = NULL;
+               rq_sop->num_free_mbufs = 0;
+       }
+
        enic_rxmbuf_queue_release(enic, rq_sop);
        if (rq_data->in_use)
                enic_rxmbuf_queue_release(enic, rq_data);
@@ -742,20 +808,20 @@ int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
        }
 
        /* number of descriptors have to be a multiple of 32 */
-       nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
-       nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
+       nb_sop_desc = (nb_desc / mbufs_per_pkt) & ENIC_ALIGN_DESCS_MASK;
+       nb_data_desc = (nb_desc - nb_sop_desc) & ENIC_ALIGN_DESCS_MASK;
 
        rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
        rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
 
        if (mbufs_per_pkt > 1) {
-               min_sop = 64;
+               min_sop = ENIC_RX_BURST_MAX;
                max_sop = ((enic->config.rq_desc_count /
-                           (mbufs_per_pkt - 1)) & ~0x1F);
+                           (mbufs_per_pkt - 1)) & ENIC_ALIGN_DESCS_MASK);
                min_data = min_sop * (mbufs_per_pkt - 1);
                max_data = enic->config.rq_desc_count;
        } else {
-               min_sop = 64;
+               min_sop = ENIC_RX_BURST_MAX;
                max_sop = enic->config.rq_desc_count;
                min_data = 0;
                max_data = 0;
@@ -826,10 +892,21 @@ int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
                        goto err_free_sop_mbuf;
        }
 
+       rq_sop->free_mbufs = (struct rte_mbuf **)
+               rte_zmalloc_socket("rq->free_mbufs",
+                                  sizeof(struct rte_mbuf *) *
+                                  ENIC_RX_BURST_MAX,
+                                  RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
+       if (rq_sop->free_mbufs == NULL)
+               goto err_free_data_mbuf;
+       rq_sop->num_free_mbufs = 0;
+
        rq_sop->tot_nb_desc = nb_desc; /* squirl away for MTU update function */
 
        return 0;
 
+err_free_data_mbuf:
+       rte_free(rq_data->mbuf_ring);
 err_free_sop_mbuf:
        rte_free(rq_sop->mbuf_ring);
 err_free_cq:
@@ -869,25 +946,15 @@ int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
        static int instance;
 
        wq->socket_id = socket_id;
-       if (nb_desc) {
-               if (nb_desc > enic->config.wq_desc_count) {
-                       dev_warning(enic,
-                               "WQ %d - number of tx desc in cmd line (%d)"\
-                               "is greater than that in the UCSM/CIMC adapter"\
-                               "policy.  Applying the value in the adapter "\
-                               "policy (%d)\n",
-                               queue_idx, nb_desc, enic->config.wq_desc_count);
-               } else if (nb_desc != enic->config.wq_desc_count) {
-                       enic->config.wq_desc_count = nb_desc;
-                       dev_info(enic,
-                               "TX Queues - effective number of descs:%d\n",
-                               nb_desc);
-               }
-       }
+       /*
+        * rte_eth_tx_queue_setup() checks min, max, and alignment. So just
+        * print an info message for diagnostics.
+        */
+       dev_info(enic, "TX Queues - effective number of descs:%d\n", nb_desc);
 
        /* Allocate queue resources */
        err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
-               enic->config.wq_desc_count,
+               nb_desc,
                sizeof(struct wq_enet_desc));
        if (err) {
                dev_err(enic, "error in allocation of wq\n");
@@ -895,7 +962,7 @@ int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
        }
 
        err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
-               socket_id, enic->config.wq_desc_count,
+               socket_id, nb_desc,
                sizeof(struct cq_enet_wq_desc));
        if (err) {
                vnic_wq_free(wq);
@@ -1190,30 +1257,32 @@ int enic_set_rss_conf(struct enic *enic, struct rte_eth_rss_conf *rss_conf)
            (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) &&
            rss_hf != 0) {
                rss_enable = 1;
-               if (rss_hf & ETH_RSS_IPV4)
+               if (rss_hf & (ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
+                             ETH_RSS_NONFRAG_IPV4_OTHER))
                        rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_IPV4;
                if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
                        rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_TCP_IPV4;
                if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
-                       /*
-                        * 'TCP' is not a typo. HW does not have a separate
-                        * enable bit for UDP RSS. The TCP bit enables both TCP
-                        * and UDP RSS..
-                        */
-                       rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_TCP_IPV4;
+                       rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_UDP_IPV4;
+                       if (enic->udp_rss_weak) {
+                               /*
+                                * 'TCP' is not a typo. The "weak" version of
+                                * UDP RSS requires both the TCP and UDP bits
+                                * be set. It does enable TCP RSS as well.
+                                */
+                               rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_TCP_IPV4;
+                       }
                }
-               if (rss_hf & ETH_RSS_IPV6)
+               if (rss_hf & (ETH_RSS_IPV6 | ETH_RSS_IPV6_EX |
+                             ETH_RSS_FRAG_IPV6 | ETH_RSS_NONFRAG_IPV6_OTHER))
                        rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_IPV6;
-               if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
-                       rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
-               if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP) {
-                       /* Again, 'TCP' is not a typo. */
+               if (rss_hf & (ETH_RSS_NONFRAG_IPV6_TCP | ETH_RSS_IPV6_TCP_EX))
                        rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
+               if (rss_hf & (ETH_RSS_NONFRAG_IPV6_UDP | ETH_RSS_IPV6_UDP_EX)) {
+                       rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_UDP_IPV6;
+                       if (enic->udp_rss_weak)
+                               rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
                }
-               if (rss_hf & ETH_RSS_IPV6_EX)
-                       rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_IPV6_EX;
-               if (rss_hf & ETH_RSS_IPV6_TCP_EX)
-                       rss_hash_type |= NIC_CFG_RSS_HASH_TYPE_TCP_IPV6_EX;
        } else {
                rss_enable = 0;
                rss_hf = 0;
@@ -1235,8 +1304,11 @@ int enic_set_rss_conf(struct enic *enic, struct rte_eth_rss_conf *rss_conf)
                enic->rss_hf = rss_hf;
                enic->rss_hash_type = rss_hash_type;
                enic->rss_enable = rss_enable;
+       } else {
+               dev_err(enic, "Failed to update RSS configurations."
+                       " hash=0x%x\n", rss_hash_type);
        }
-       return 0;
+       return ret;
 }
 
 int enic_set_vlan_strip(struct enic *enic)
@@ -1486,7 +1558,7 @@ int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
 
        /* put back the real receive function */
        rte_mb();
-       eth_dev->rx_pkt_burst = enic_recv_pkts;
+       pick_rx_handler(enic);
        rte_mb();
 
        /* restart Rx traffic */
@@ -1589,7 +1661,18 @@ static int enic_dev_init(struct enic *enic)
                        PKT_TX_OUTER_IP_CKSUM |
                        PKT_TX_TUNNEL_MASK;
                enic->overlay_offload = true;
+               enic->vxlan_port = ENIC_DEFAULT_VXLAN_PORT;
                dev_info(enic, "Overlay offload is enabled\n");
+               /*
+                * Reset the vxlan port to the default, as the NIC firmware
+                * does not reset it automatically and keeps the old setting.
+                */
+               if (vnic_dev_overlay_offload_cfg(enic->vdev,
+                                                OVERLAY_CFG_VXLAN_PORT_UPDATE,
+                                                ENIC_DEFAULT_VXLAN_PORT)) {
+                       dev_err(enic, "failed to update vxlan port\n");
+                       return -EINVAL;
+               }
        }
 
        return 0;
@@ -1641,8 +1724,10 @@ int enic_probe(struct enic *enic)
        }
 
        /* Set ingress vlan rewrite mode before vnic initialization */
+       dev_debug(enic, "Set ig_vlan_rewrite_mode=%u\n",
+                 enic->ig_vlan_rewrite_mode);
        err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
-               IG_VLAN_REWRITE_MODE_PASS_THRU);
+               enic->ig_vlan_rewrite_mode);
        if (err) {
                dev_err(enic,
                        "Failed to set ingress vlan rewrite mode, aborting.\n");