vhost: introduce API to start a specific driver
[dpdk.git] / examples / vhost / main.c
index a718577..64b3eea 100644 (file)
@@ -106,9 +106,6 @@ static uint32_t num_devices;
 static struct rte_mempool *mbuf_pool;
 static int mergeable;
 
-/* Do vlan strip on host, enabled on default */
-static uint32_t vlan_strip = 1;
-
 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */
 typedef enum {
        VM2VM_DISABLED = 0,
@@ -130,14 +127,16 @@ static uint32_t enable_tx_csum;
 static uint32_t enable_tso;
 
 static int client_mode;
+static int dequeue_zero_copy;
 
 /* Specify timeout (in useconds) between retries on RX. */
 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
 /* Specify the number of retries on RX. */
 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
 
-/* Socket file path. Can be set by user */
-static char socket_file[PATH_MAX] = "vhost-net";
+/* Socket file paths. Can be set by user */
+static char *socket_files;
+static int nb_sockets;
 
 /* empty vmdq configuration structure. Filled in programatically */
 static struct rte_eth_conf vmdq_conf_default = {
@@ -296,6 +295,17 @@ port_init(uint8_t port)
 
        rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
        tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
+
+       /*
+        * When dequeue zero copy is enabled, guest Tx used vring will be
+        * updated only when corresponding mbuf is freed. Thus, the nb_tx_desc
+        * (tx_ring_size here) must be small enough so that the driver will
+        * hit the free threshold easily and free mbufs timely. Otherwise,
+        * guest Tx vring would be starved.
+        */
+       if (dequeue_zero_copy)
+               tx_ring_size = 64;
+
        tx_rings = (uint16_t)rte_lcore_count();
 
        retval = validate_num_devices(MAX_DEVICES);
@@ -318,16 +328,6 @@ port_init(uint8_t port)
 
        if (port >= rte_eth_dev_count()) return -1;
 
-       if (enable_tx_csum == 0)
-               rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_CSUM);
-
-       if (enable_tso == 0) {
-               rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4);
-               rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO6);
-               rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_GUEST_TSO4);
-               rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_GUEST_TSO6);
-       }
-
        rx_rings = (uint16_t)dev_info.max_rx_queues;
        /* Configure ethernet device. */
        retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
@@ -395,11 +395,12 @@ static int
 us_vhost_parse_socket_path(const char *q_arg)
 {
        /* parse number string */
-
        if (strnlen(q_arg, PATH_MAX) > PATH_MAX)
                return -1;
-       else
-               snprintf((char *)&socket_file, PATH_MAX, "%s", q_arg);
+
+       socket_files = realloc(socket_files, PATH_MAX * (nb_sockets + 1));
+       snprintf(socket_files + nb_sockets * PATH_MAX, PATH_MAX, "%s", q_arg);
+       nb_sockets++;
 
        return 0;
 }
@@ -467,12 +468,12 @@ us_vhost_usage(const char *prgname)
        "               --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n"
        "               --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n"
        "               --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n"
-       "               --vlan-strip [0|1]: disable/enable(default) RX VLAN strip on host\n"
        "               --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
        "               --socket-file: The path of the socket file.\n"
        "               --tx-csum [0|1] disable/enable TX checksum offload.\n"
        "               --tso [0|1] disable/enable TCP segment offload.\n"
-       "               --client register a vhost-user socket as client mode.\n",
+       "               --client register a vhost-user socket as client mode.\n"
+       "               --dequeue-zero-copy enables dequeue zero copy\n",
               prgname);
 }
 
@@ -492,12 +493,12 @@ us_vhost_parse_args(int argc, char **argv)
                {"rx-retry-delay", required_argument, NULL, 0},
                {"rx-retry-num", required_argument, NULL, 0},
                {"mergeable", required_argument, NULL, 0},
-               {"vlan-strip", required_argument, NULL, 0},
                {"stats", required_argument, NULL, 0},
                {"socket-file", required_argument, NULL, 0},
                {"tx-csum", required_argument, NULL, 0},
                {"tso", required_argument, NULL, 0},
                {"client", no_argument, &client_mode, 1},
+               {"dequeue-zero-copy", no_argument, &dequeue_zero_copy, 1},
                {NULL, 0, 0, 0},
        };
 
@@ -520,7 +521,6 @@ us_vhost_parse_args(int argc, char **argv)
                        vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
                                ETH_VMDQ_ACCEPT_BROADCAST |
                                ETH_VMDQ_ACCEPT_MULTICAST;
-                       rte_vhost_feature_enable(1ULL << VIRTIO_NET_F_CTRL_RX);
 
                        break;
 
@@ -615,22 +615,6 @@ us_vhost_parse_args(int argc, char **argv)
                                }
                        }
 
-                       /* Enable/disable RX VLAN strip on host. */
-                       if (!strncmp(long_option[option_index].name,
-                               "vlan-strip", MAX_LONG_OPT_SZ)) {
-                               ret = parse_num_opt(optarg, 1);
-                               if (ret == -1) {
-                                       RTE_LOG(INFO, VHOST_CONFIG,
-                                               "Invalid argument for VLAN strip [0|1]\n");
-                                       us_vhost_usage(prgname);
-                                       return -1;
-                               } else {
-                                       vlan_strip = !!ret;
-                                       vmdq_conf_default.rxmode.hw_vlan_strip =
-                                               vlan_strip;
-                               }
-                       }
-
                        /* Enable/disable stats. */
                        if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) {
                                ret = parse_num_opt(optarg, INT32_MAX);
@@ -762,10 +746,7 @@ link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
                        "(%d) failed to add device MAC address to VMDQ\n",
                        vdev->vid);
 
-       /* Enable stripping of the vlan tag as we handle routing. */
-       if (vlan_strip)
-               rte_eth_dev_set_vlan_strip_on_queue(ports[0],
-                       (uint16_t)vdev->vmdq_rx_q, 1);
+       rte_eth_dev_set_vlan_strip_on_queue(ports[0], vdev->vmdq_rx_q, 1);
 
        /* Set device as ready for RX. */
        vdev->ready = DEVICE_RX;
@@ -840,17 +821,17 @@ virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
                return -1;
 
        if (vdev->vid == dst_vdev->vid) {
-               RTE_LOG(DEBUG, VHOST_DATA,
+               RTE_LOG_DP(DEBUG, VHOST_DATA,
                        "(%d) TX: src and dst MAC is same. Dropping packet.\n",
                        vdev->vid);
                return 0;
        }
 
-       RTE_LOG(DEBUG, VHOST_DATA,
+       RTE_LOG_DP(DEBUG, VHOST_DATA,
                "(%d) TX: MAC address is local\n", dst_vdev->vid);
 
        if (unlikely(dst_vdev->remove)) {
-               RTE_LOG(DEBUG, VHOST_DATA,
+               RTE_LOG_DP(DEBUG, VHOST_DATA,
                        "(%d) device is marked for removal\n", dst_vdev->vid);
                return 0;
        }
@@ -875,7 +856,7 @@ find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
                return 0;
 
        if (vdev->vid == dst_vdev->vid) {
-               RTE_LOG(DEBUG, VHOST_DATA,
+               RTE_LOG_DP(DEBUG, VHOST_DATA,
                        "(%d) TX: src and dst MAC is same. Dropping packet.\n",
                        vdev->vid);
                return -1;
@@ -889,7 +870,7 @@ find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
        *offset  = VLAN_HLEN;
        *vlan_tag = vlan_tags[vdev->vid];
 
-       RTE_LOG(DEBUG, VHOST_DATA,
+       RTE_LOG_DP(DEBUG, VHOST_DATA,
                "(%d) TX: pkt to local VM device id: (%d), vlan tag: %u.\n",
                vdev->vid, dst_vdev->vid, *vlan_tag);
 
@@ -981,7 +962,7 @@ virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
                }
        }
 
-       RTE_LOG(DEBUG, VHOST_DATA,
+       RTE_LOG_DP(DEBUG, VHOST_DATA,
                "(%d) TX: MAC address is external\n", vdev->vid);
 
 queue2nic:
@@ -1049,7 +1030,7 @@ drain_mbuf_table(struct mbuf_table *tx_q)
        if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) {
                prev_tsc = cur_tsc;
 
-               RTE_LOG(DEBUG, VHOST_DATA,
+               RTE_LOG_DP(DEBUG, VHOST_DATA,
                        "TX queue drained after timeout with burst size %u\n",
                        tx_q->len);
                do_drain_mbuf_table(tx_q);
@@ -1289,7 +1270,7 @@ new_device(int vid)
  * These callback allow devices to be added to the data core when configuration
  * has been fully complete.
  */
-static const struct virtio_net_device_ops virtio_net_device_ops =
+static const struct vhost_device_ops virtio_net_device_ops =
 {
        .new_device =  new_device,
        .destroy_device = destroy_device,
@@ -1341,14 +1322,27 @@ print_stats(void)
        }
 }
 
+static void
+unregister_drivers(int socket_num)
+{
+       int i, ret;
+
+       for (i = 0; i < socket_num; i++) {
+               ret = rte_vhost_driver_unregister(socket_files + i * PATH_MAX);
+               if (ret != 0)
+                       RTE_LOG(ERR, VHOST_CONFIG,
+                               "Fail to unregister vhost driver for %s.\n",
+                               socket_files + i * PATH_MAX);
+       }
+}
+
 /* When we receive a INT signal, unregister vhost driver */
 static void
 sigint_handler(__rte_unused int signum)
 {
        /* Unregister vhost driver. */
-       int ret = rte_vhost_driver_unregister((char *)&socket_file);
-       if (ret != 0)
-               rte_exit(EXIT_FAILURE, "vhost driver unregister failure.\n");
+       unregister_drivers(nb_sockets);
+
        exit(0);
 }
 
@@ -1388,7 +1382,7 @@ create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
                mtu = 64 * 1024;
 
        nr_mbufs_per_core  = (mtu + mbuf_size) * MAX_PKT_BURST /
-                       (mbuf_size - RTE_PKTMBUF_HEADROOM) * MAX_PKT_BURST;
+                       (mbuf_size - RTE_PKTMBUF_HEADROOM);
        nr_mbufs_per_core += nr_rx_desc;
        nr_mbufs_per_core  = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache);
 
@@ -1404,15 +1398,14 @@ create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
 }
 
 /*
- * Main function, does initialisation and calls the per-lcore functions. The CUSE
- * device is also registered here to handle the IOCTLs.
+ * Main function, does initialisation and calls the per-lcore functions.
  */
 int
 main(int argc, char *argv[])
 {
        unsigned lcore_id, core_id = 0;
        unsigned nb_ports, valid_num_ports;
-       int ret;
+       int ret, i;
        uint8_t portid;
        static pthread_t tid;
        char thread_name[RTE_MAX_THREAD_NAME_LEN];
@@ -1432,11 +1425,12 @@ main(int argc, char *argv[])
        if (ret < 0)
                rte_exit(EXIT_FAILURE, "Invalid argument\n");
 
-       for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id ++)
+       for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
                TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
 
                if (rte_lcore_is_enabled(lcore_id))
-                       lcore_ids[core_id ++] = lcore_id;
+                       lcore_ids[core_id++] = lcore_id;
+       }
 
        if (rte_lcore_count() > RTE_MAX_LCORE)
                rte_exit(EXIT_FAILURE,"Not enough cores\n");
@@ -1504,21 +1498,63 @@ main(int argc, char *argv[])
        RTE_LCORE_FOREACH_SLAVE(lcore_id)
                rte_eal_remote_launch(switch_worker, NULL, lcore_id);
 
-       if (mergeable == 0)
-               rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_MRG_RXBUF);
-
        if (client_mode)
                flags |= RTE_VHOST_USER_CLIENT;
 
+       if (dequeue_zero_copy)
+               flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
+
        /* Register vhost user driver to handle vhost messages. */
-       ret = rte_vhost_driver_register(socket_file, flags);
-       if (ret != 0)
-               rte_exit(EXIT_FAILURE, "vhost driver register failure.\n");
+       for (i = 0; i < nb_sockets; i++) {
+               char *file = socket_files + i * PATH_MAX;
+               ret = rte_vhost_driver_register(file, flags);
+               if (ret != 0) {
+                       unregister_drivers(i);
+                       rte_exit(EXIT_FAILURE,
+                               "vhost driver register failure.\n");
+               }
+               if (mergeable == 0) {
+                       rte_vhost_driver_disable_features(file,
+                               1ULL << VIRTIO_NET_F_MRG_RXBUF);
+               }
+
+               if (enable_tx_csum == 0) {
+                       rte_vhost_driver_disable_features(file,
+                               1ULL << VIRTIO_NET_F_CSUM);
+               }
 
-       rte_vhost_driver_callback_register(&virtio_net_device_ops);
+               if (enable_tso == 0) {
+                       rte_vhost_driver_disable_features(file,
+                               1ULL << VIRTIO_NET_F_HOST_TSO4);
+                       rte_vhost_driver_disable_features(file,
+                               1ULL << VIRTIO_NET_F_HOST_TSO6);
+                       rte_vhost_driver_disable_features(file,
+                               1ULL << VIRTIO_NET_F_GUEST_TSO4);
+                       rte_vhost_driver_disable_features(file,
+                               1ULL << VIRTIO_NET_F_GUEST_TSO6);
+               }
+
+               if (promiscuous) {
+                       rte_vhost_driver_enable_features(file,
+                               1ULL << VIRTIO_NET_F_CTRL_RX);
+               }
+
+               ret = rte_vhost_driver_callback_register(file,
+                       &virtio_net_device_ops);
+               if (ret != 0) {
+                       rte_exit(EXIT_FAILURE,
+                               "failed to register vhost driver callbacks.\n");
+               }
+
+               if (rte_vhost_driver_start(file) < 0) {
+                       rte_exit(EXIT_FAILURE,
+                               "failed to start vhost driver.\n");
+               }
+       }
+
+       RTE_LCORE_FOREACH_SLAVE(lcore_id)
+               rte_eal_wait_lcore(lcore_id);
 
-       /* Start CUSE session. */
-       rte_vhost_driver_session_start();
        return 0;
 
 }