net: add rte prefix to ether defines
[dpdk.git] / drivers / net / mlx5 / mlx5_trigger.c
index 30addd2..d5ce74d 100644 (file)
-/*-
- *   BSD LICENSE
- *
- *   Copyright 2015 6WIND S.A.
- *   Copyright 2015 Mellanox.
- *
- *   Redistribution and use in source and binary forms, with or without
- *   modification, are permitted provided that the following conditions
- *   are met:
- *
- *     * Redistributions of source code must retain the above copyright
- *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright
- *       notice, this list of conditions and the following disclaimer in
- *       the documentation and/or other materials provided with the
- *       distribution.
- *     * Neither the name of 6WIND S.A. nor the names of its
- *       contributors may be used to endorse or promote products derived
- *       from this software without specific prior written permission.
- *
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright 2015 6WIND S.A.
+ * Copyright 2015 Mellanox Technologies, Ltd
  */
 
-/* DPDK headers don't like -pedantic. */
-#ifdef PEDANTIC
-#pragma GCC diagnostic ignored "-Wpedantic"
-#endif
+#include <unistd.h>
+
 #include <rte_ether.h>
-#include <rte_ethdev.h>
+#include <rte_ethdev_driver.h>
 #include <rte_interrupts.h>
 #include <rte_alarm.h>
-#ifdef PEDANTIC
-#pragma GCC diagnostic error "-Wpedantic"
-#endif
 
 #include "mlx5.h"
 #include "mlx5_rxtx.h"
 #include "mlx5_utils.h"
 
+/**
+ * Stop traffic on Tx queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+static void
+mlx5_txq_stop(struct rte_eth_dev *dev)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       unsigned int i;
+
+       for (i = 0; i != priv->txqs_n; ++i)
+               mlx5_txq_release(dev, i);
+}
+
+/**
+ * Start traffic on Tx queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_txq_start(struct rte_eth_dev *dev)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       unsigned int i;
+       int ret;
+
+       for (i = 0; i != priv->txqs_n; ++i) {
+               struct mlx5_txq_ctrl *txq_ctrl = mlx5_txq_get(dev, i);
+
+               if (!txq_ctrl)
+                       continue;
+               txq_alloc_elts(txq_ctrl);
+               txq_ctrl->ibv = mlx5_txq_ibv_new(dev, i);
+               if (!txq_ctrl->ibv) {
+                       rte_errno = ENOMEM;
+                       goto error;
+               }
+       }
+       return 0;
+error:
+       ret = rte_errno; /* Save rte_errno before cleanup. */
+       do {
+               mlx5_txq_release(dev, i);
+       } while (i-- != 0);
+       rte_errno = ret; /* Restore rte_errno. */
+       return -rte_errno;
+}
+
+/**
+ * Stop traffic on Rx queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ */
+static void
+mlx5_rxq_stop(struct rte_eth_dev *dev)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       unsigned int i;
+
+       for (i = 0; i != priv->rxqs_n; ++i)
+               mlx5_rxq_release(dev, i);
+}
+
+/**
+ * Start traffic on Rx queues.
+ *
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_rxq_start(struct rte_eth_dev *dev)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       unsigned int i;
+       int ret = 0;
+
+       /* Allocate/reuse/resize mempool for Multi-Packet RQ. */
+       if (mlx5_mprq_alloc_mp(dev)) {
+               /* Should not release Rx queues but return immediately. */
+               return -rte_errno;
+       }
+       for (i = 0; i != priv->rxqs_n; ++i) {
+               struct mlx5_rxq_ctrl *rxq_ctrl = mlx5_rxq_get(dev, i);
+               struct rte_mempool *mp;
+
+               if (!rxq_ctrl)
+                       continue;
+               /* Pre-register Rx mempool. */
+               mp = mlx5_rxq_mprq_enabled(&rxq_ctrl->rxq) ?
+                    rxq_ctrl->rxq.mprq_mp : rxq_ctrl->rxq.mp;
+               DRV_LOG(DEBUG,
+                       "port %u Rx queue %u registering"
+                       " mp %s having %u chunks",
+                       dev->data->port_id, rxq_ctrl->rxq.idx,
+                       mp->name, mp->nb_mem_chunks);
+               mlx5_mr_update_mp(dev, &rxq_ctrl->rxq.mr_ctrl, mp);
+               ret = rxq_alloc_elts(rxq_ctrl);
+               if (ret)
+                       goto error;
+               rxq_ctrl->ibv = mlx5_rxq_ibv_new(dev, i);
+               if (!rxq_ctrl->ibv)
+                       goto error;
+       }
+       return 0;
+error:
+       ret = rte_errno; /* Save rte_errno before cleanup. */
+       do {
+               mlx5_rxq_release(dev, i);
+       } while (i-- != 0);
+       rte_errno = ret; /* Restore rte_errno. */
+       return -rte_errno;
+}
+
 /**
  * DPDK callback to start the device.
  *
  *   Pointer to Ethernet device structure.
  *
  * @return
- *   0 on success, negative errno value on failure.
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
  */
 int
 mlx5_dev_start(struct rte_eth_dev *dev)
 {
-       struct priv *priv = dev->data->dev_private;
-       int err;
-
-       if (mlx5_is_secondary())
-               return -E_RTE_SECONDARY;
+       struct mlx5_priv *priv = dev->data->dev_private;
+       int ret;
 
-       priv_lock(priv);
-       if (priv->started) {
-               priv_unlock(priv);
-               return 0;
+       DRV_LOG(DEBUG, "port %u starting device", dev->data->port_id);
+       ret = mlx5_txq_start(dev);
+       if (ret) {
+               DRV_LOG(ERR, "port %u Tx queue allocation failed: %s",
+                       dev->data->port_id, strerror(rte_errno));
+               return -rte_errno;
+       }
+       ret = mlx5_rxq_start(dev);
+       if (ret) {
+               DRV_LOG(ERR, "port %u Rx queue allocation failed: %s",
+                       dev->data->port_id, strerror(rte_errno));
+               mlx5_txq_stop(dev);
+               return -rte_errno;
+       }
+       dev->data->dev_started = 1;
+       ret = mlx5_rx_intr_vec_enable(dev);
+       if (ret) {
+               DRV_LOG(ERR, "port %u Rx interrupt vector creation failed",
+                       dev->data->port_id);
+               goto error;
        }
-       DEBUG("%p: allocating and configuring hash RX queues", (void *)dev);
-       err = priv_create_hash_rxqs(priv);
-       if (!err)
-               err = priv_rehash_flows(priv);
-       if (!err)
-               priv->started = 1;
-       else {
-               ERROR("%p: an error occurred while configuring hash RX queues:"
-                     " %s",
-                     (void *)priv, strerror(err));
-               /* Rollback. */
-               priv_special_flow_disable_all(priv);
-               priv_mac_addrs_disable(priv);
-               priv_destroy_hash_rxqs(priv);
+       mlx5_stats_init(dev);
+       ret = mlx5_traffic_enable(dev);
+       if (ret) {
+               DRV_LOG(DEBUG, "port %u failed to set defaults flows",
+                       dev->data->port_id);
+               goto error;
        }
-       if (dev->data->dev_conf.fdir_conf.mode != RTE_FDIR_MODE_NONE)
-               priv_fdir_enable(priv);
-       priv_dev_interrupt_handler_install(priv, dev);
-       err = priv_flow_start(priv);
-       priv_xstats_init(priv);
-       priv_unlock(priv);
-       return -err;
+       ret = mlx5_flow_start(dev, &priv->flows);
+       if (ret) {
+               DRV_LOG(DEBUG, "port %u failed to set flows",
+                       dev->data->port_id);
+               goto error;
+       }
+       rte_wmb();
+       dev->tx_pkt_burst = mlx5_select_tx_function(dev);
+       dev->rx_pkt_burst = mlx5_select_rx_function(dev);
+       /* Enable datapath on secondary process. */
+       mlx5_mp_req_start_rxtx(dev);
+       mlx5_dev_interrupt_handler_install(dev);
+       return 0;
+error:
+       ret = rte_errno; /* Save rte_errno before cleanup. */
+       /* Rollback. */
+       dev->data->dev_started = 0;
+       mlx5_flow_stop(dev, &priv->flows);
+       mlx5_traffic_disable(dev);
+       mlx5_txq_stop(dev);
+       mlx5_rxq_stop(dev);
+       rte_errno = ret; /* Restore rte_errno. */
+       return -rte_errno;
 }
 
 /**
@@ -107,23 +218,186 @@ mlx5_dev_start(struct rte_eth_dev *dev)
 void
 mlx5_dev_stop(struct rte_eth_dev *dev)
 {
-       struct priv *priv = dev->data->dev_private;
+       struct mlx5_priv *priv = dev->data->dev_private;
+
+       dev->data->dev_started = 0;
+       /* Prevent crashes when queues are still in use. */
+       dev->rx_pkt_burst = removed_rx_burst;
+       dev->tx_pkt_burst = removed_tx_burst;
+       rte_wmb();
+       /* Disable datapath on secondary process. */
+       mlx5_mp_req_stop_rxtx(dev);
+       usleep(1000 * priv->rxqs_n);
+       DRV_LOG(DEBUG, "port %u stopping device", dev->data->port_id);
+       mlx5_flow_stop(dev, &priv->flows);
+       mlx5_traffic_disable(dev);
+       mlx5_rx_intr_vec_disable(dev);
+       mlx5_dev_interrupt_handler_uninstall(dev);
+       mlx5_txq_stop(dev);
+       mlx5_rxq_stop(dev);
+}
+
+/**
+ * Enable traffic flows configured by control plane
+ *
+ * @param dev
+ *   Pointer to Ethernet device private data.
+ * @param dev
+ *   Pointer to Ethernet device structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_traffic_enable(struct rte_eth_dev *dev)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct rte_flow_item_eth bcast = {
+               .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
+       };
+       struct rte_flow_item_eth ipv6_multi_spec = {
+               .dst.addr_bytes = "\x33\x33\x00\x00\x00\x00",
+       };
+       struct rte_flow_item_eth ipv6_multi_mask = {
+               .dst.addr_bytes = "\xff\xff\x00\x00\x00\x00",
+       };
+       struct rte_flow_item_eth unicast = {
+               .src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
+       };
+       struct rte_flow_item_eth unicast_mask = {
+               .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
+       };
+       const unsigned int vlan_filter_n = priv->vlan_filter_n;
+       const struct rte_ether_addr cmp = {
+               .addr_bytes = "\x00\x00\x00\x00\x00\x00",
+       };
+       unsigned int i;
+       unsigned int j;
+       int ret;
+
+       if (priv->isolated)
+               return 0;
+       if (dev->data->promiscuous) {
+               struct rte_flow_item_eth promisc = {
+                       .dst.addr_bytes = "\x00\x00\x00\x00\x00\x00",
+                       .src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
+                       .type = 0,
+               };
+
+               ret = mlx5_ctrl_flow(dev, &promisc, &promisc);
+               if (ret)
+                       goto error;
+       }
+       if (dev->data->all_multicast) {
+               struct rte_flow_item_eth multicast = {
+                       .dst.addr_bytes = "\x01\x00\x00\x00\x00\x00",
+                       .src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
+                       .type = 0,
+               };
 
-       if (mlx5_is_secondary())
-               return;
+               ret = mlx5_ctrl_flow(dev, &multicast, &multicast);
+               if (ret)
+                       goto error;
+       } else {
+               /* Add broadcast/multicast flows. */
+               for (i = 0; i != vlan_filter_n; ++i) {
+                       uint16_t vlan = priv->vlan_filter[i];
+
+                       struct rte_flow_item_vlan vlan_spec = {
+                               .tci = rte_cpu_to_be_16(vlan),
+                       };
+                       struct rte_flow_item_vlan vlan_mask =
+                               rte_flow_item_vlan_mask;
+
+                       ret = mlx5_ctrl_flow_vlan(dev, &bcast, &bcast,
+                                                 &vlan_spec, &vlan_mask);
+                       if (ret)
+                               goto error;
+                       ret = mlx5_ctrl_flow_vlan(dev, &ipv6_multi_spec,
+                                                 &ipv6_multi_mask,
+                                                 &vlan_spec, &vlan_mask);
+                       if (ret)
+                               goto error;
+               }
+               if (!vlan_filter_n) {
+                       ret = mlx5_ctrl_flow(dev, &bcast, &bcast);
+                       if (ret)
+                               goto error;
+                       ret = mlx5_ctrl_flow(dev, &ipv6_multi_spec,
+                                            &ipv6_multi_mask);
+                       if (ret)
+                               goto error;
+               }
+       }
+       /* Add MAC address flows. */
+       for (i = 0; i != MLX5_MAX_MAC_ADDRESSES; ++i) {
+               struct rte_ether_addr *mac = &dev->data->mac_addrs[i];
 
-       priv_lock(priv);
-       if (!priv->started) {
-               priv_unlock(priv);
-               return;
+               if (!memcmp(mac, &cmp, sizeof(*mac)))
+                       continue;
+               memcpy(&unicast.dst.addr_bytes,
+                      mac->addr_bytes,
+                      RTE_ETHER_ADDR_LEN);
+               for (j = 0; j != vlan_filter_n; ++j) {
+                       uint16_t vlan = priv->vlan_filter[j];
+
+                       struct rte_flow_item_vlan vlan_spec = {
+                               .tci = rte_cpu_to_be_16(vlan),
+                       };
+                       struct rte_flow_item_vlan vlan_mask =
+                               rte_flow_item_vlan_mask;
+
+                       ret = mlx5_ctrl_flow_vlan(dev, &unicast,
+                                                 &unicast_mask,
+                                                 &vlan_spec,
+                                                 &vlan_mask);
+                       if (ret)
+                               goto error;
+               }
+               if (!vlan_filter_n) {
+                       ret = mlx5_ctrl_flow(dev, &unicast, &unicast_mask);
+                       if (ret)
+                               goto error;
+               }
+       }
+       return 0;
+error:
+       ret = rte_errno; /* Save rte_errno before cleanup. */
+       mlx5_flow_list_flush(dev, &priv->ctrl_flows);
+       rte_errno = ret; /* Restore rte_errno. */
+       return -rte_errno;
+}
+
+
+/**
+ * Disable traffic flows configured by control plane
+ *
+ * @param dev
+ *   Pointer to Ethernet device private data.
+ */
+void
+mlx5_traffic_disable(struct rte_eth_dev *dev)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+
+       mlx5_flow_list_flush(dev, &priv->ctrl_flows);
+}
+
+/**
+ * Restart traffic flows configured by control plane
+ *
+ * @param dev
+ *   Pointer to Ethernet device private data.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_traffic_restart(struct rte_eth_dev *dev)
+{
+       if (dev->data->dev_started) {
+               mlx5_traffic_disable(dev);
+               return mlx5_traffic_enable(dev);
        }
-       DEBUG("%p: cleaning up and destroying hash RX queues", (void *)dev);
-       priv_special_flow_disable_all(priv);
-       priv_mac_addrs_disable(priv);
-       priv_destroy_hash_rxqs(priv);
-       priv_fdir_disable(priv);
-       priv_flow_stop(priv);
-       priv_dev_interrupt_handler_uninstall(priv, dev);
-       priv->started = 0;
-       priv_unlock(priv);
+       return 0;
 }