vhost: introduce API to start a specific driver
[dpdk.git] / examples / vhost / main.c
index 32a79be..64b3eea 100644 (file)
@@ -90,9 +90,6 @@
 /* Size of buffers used for snprintfs. */
 #define MAX_PRINT_BUFF 6072
 
-/* Maximum character device basename size. */
-#define MAX_BASENAME_SZ 10
-
 /* Maximum long option length for option parsing. */
 #define MAX_LONG_OPT_SZ 64
 
@@ -109,9 +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,
@@ -132,13 +126,17 @@ static uint32_t enable_tx_csum;
 /* Disable TSO offload */
 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;
 
-/* Character device basename. Can be set by user. */
-static char dev_basename[MAX_BASENAME_SZ] = "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 = {
@@ -297,6 +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);
@@ -319,19 +328,14 @@ 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);
-       }
-
        rx_rings = (uint16_t)dev_info.max_rx_queues;
        /* Configure ethernet device. */
        retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
-       if (retval != 0)
+       if (retval != 0) {
+               RTE_LOG(ERR, VHOST_PORT, "Failed to configure port %u: %s.\n",
+                       port, strerror(-retval));
                return retval;
+       }
 
        /* Setup the queues. */
        for (q = 0; q < rx_rings; q ++) {
@@ -339,21 +343,30 @@ port_init(uint8_t port)
                                                rte_eth_dev_socket_id(port),
                                                rxconf,
                                                mbuf_pool);
-               if (retval < 0)
+               if (retval < 0) {
+                       RTE_LOG(ERR, VHOST_PORT,
+                               "Failed to setup rx queue %u of port %u: %s.\n",
+                               q, port, strerror(-retval));
                        return retval;
+               }
        }
        for (q = 0; q < tx_rings; q ++) {
                retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
                                                rte_eth_dev_socket_id(port),
                                                txconf);
-               if (retval < 0)
+               if (retval < 0) {
+                       RTE_LOG(ERR, VHOST_PORT,
+                               "Failed to setup tx queue %u of port %u: %s.\n",
+                               q, port, strerror(-retval));
                        return retval;
+               }
        }
 
        /* Start the device. */
        retval  = rte_eth_dev_start(port);
        if (retval < 0) {
-               RTE_LOG(ERR, VHOST_DATA, "Failed to start the device.\n");
+               RTE_LOG(ERR, VHOST_PORT, "Failed to start port %u: %s\n",
+                       port, strerror(-retval));
                return retval;
        }
 
@@ -376,17 +389,18 @@ port_init(uint8_t port)
 }
 
 /*
- * Set character device basename.
+ * Set socket file path.
  */
 static int
-us_vhost_parse_basename(const char *q_arg)
+us_vhost_parse_socket_path(const char *q_arg)
 {
        /* parse number string */
-
-       if (strnlen(q_arg, MAX_BASENAME_SZ) > MAX_BASENAME_SZ)
+       if (strnlen(q_arg, PATH_MAX) > PATH_MAX)
                return -1;
-       else
-               snprintf((char*)&dev_basename, MAX_BASENAME_SZ, "%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;
 }
@@ -446,7 +460,7 @@ us_vhost_usage(const char *prgname)
        RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
        "               --vm2vm [0|1|2]\n"
        "               --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
-       "               --dev-basename <name>\n"
+       "               --socket-file <path>\n"
        "               --nb-devices ND\n"
        "               -p PORTMASK: Set mask for ports to be used by application\n"
        "               --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n"
@@ -454,11 +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"
-       "               --dev-basename: The basename to be used for the character device.\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",
+       "               --tso [0|1] disable/enable TCP segment offload.\n"
+       "               --client register a vhost-user socket as client mode.\n"
+       "               --dequeue-zero-copy enables dequeue zero copy\n",
               prgname);
 }
 
@@ -478,11 +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},
-               {"dev-basename", 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},
        };
 
@@ -505,7 +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;
 
@@ -600,27 +615,12 @@ 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);
                                if (ret == -1) {
-                                       RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for stats [0..N]\n");
+                                       RTE_LOG(INFO, VHOST_CONFIG,
+                                               "Invalid argument for stats [0..N]\n");
                                        us_vhost_usage(prgname);
                                        return -1;
                                } else {
@@ -628,10 +628,13 @@ us_vhost_parse_args(int argc, char **argv)
                                }
                        }
 
-                       /* Set character device basename. */
-                       if (!strncmp(long_option[option_index].name, "dev-basename", MAX_LONG_OPT_SZ)) {
-                               if (us_vhost_parse_basename(optarg) == -1) {
-                                       RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for character device basename (Max %d characters)\n", MAX_BASENAME_SZ);
+                       /* Set socket file path. */
+                       if (!strncmp(long_option[option_index].name,
+                                               "socket-file", MAX_LONG_OPT_SZ)) {
+                               if (us_vhost_parse_socket_path(optarg) == -1) {
+                                       RTE_LOG(INFO, VHOST_CONFIG,
+                                       "Invalid argument for socket name (Max %d characters)\n",
+                                       PATH_MAX);
                                        us_vhost_usage(prgname);
                                        return -1;
                                }
@@ -691,7 +694,7 @@ find_vhost_dev(struct ether_addr *mac)
 {
        struct vhost_dev *vdev;
 
-       TAILQ_FOREACH(vdev, &vhost_dev_list, next) {
+       TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
                if (vdev->ready == DEVICE_RX &&
                    is_same_ether_addr(mac, &vdev->mac_address))
                        return vdev;
@@ -708,7 +711,6 @@ static int
 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
 {
        struct ether_hdr *pkt_hdr;
-       struct virtio_net *dev = vdev->dev;
        int i, ret;
 
        /* Learn MAC address of guest device from packet */
@@ -716,8 +718,8 @@ link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
 
        if (find_vhost_dev(&pkt_hdr->s_addr)) {
                RTE_LOG(ERR, VHOST_DATA,
-                       "Device (%" PRIu64 ") is using a registered MAC!\n",
-                       dev->device_fh);
+                       "(%d) device is using a registered MAC!\n",
+                       vdev->vid);
                return -1;
        }
 
@@ -725,11 +727,12 @@ link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
                vdev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i];
 
        /* vlan_tag currently uses the device_id. */
-       vdev->vlan_tag = vlan_tags[dev->device_fh];
+       vdev->vlan_tag = vlan_tags[vdev->vid];
 
        /* Print out VMDQ registration info. */
-       RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") MAC_ADDRESS %02x:%02x:%02x:%02x:%02x:%02x and VLAN_TAG %d registered\n",
-               dev->device_fh,
+       RTE_LOG(INFO, VHOST_DATA,
+               "(%d) mac %02x:%02x:%02x:%02x:%02x:%02x and vlan %d registered\n",
+               vdev->vid,
                vdev->mac_address.addr_bytes[0], vdev->mac_address.addr_bytes[1],
                vdev->mac_address.addr_bytes[2], vdev->mac_address.addr_bytes[3],
                vdev->mac_address.addr_bytes[4], vdev->mac_address.addr_bytes[5],
@@ -737,15 +740,13 @@ link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
 
        /* Register the MAC address. */
        ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
-                               (uint32_t)dev->device_fh + vmdq_pool_base);
+                               (uint32_t)vdev->vid + vmdq_pool_base);
        if (ret)
-               RTE_LOG(ERR, VHOST_DATA, "(%"PRIu64") Failed to add device MAC address to VMDQ\n",
-                                       dev->device_fh);
+               RTE_LOG(ERR, VHOST_DATA,
+                       "(%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;
@@ -794,7 +795,7 @@ virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
 {
        uint16_t ret;
 
-       ret = rte_vhost_enqueue_burst(dst_vdev->dev, VIRTIO_RXQ, &m, 1);
+       ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1);
        if (enable_stats) {
                rte_atomic64_inc(&dst_vdev->stats.rx_total_atomic);
                rte_atomic64_add(&dst_vdev->stats.rx_atomic, ret);
@@ -812,7 +813,6 @@ virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
 {
        struct ether_hdr *pkt_hdr;
        struct vhost_dev *dst_vdev;
-       uint64_t fh;
 
        pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 
@@ -820,20 +820,19 @@ virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
        if (!dst_vdev)
                return -1;
 
-       fh = dst_vdev->dev->device_fh;
-       if (fh == vdev->dev->device_fh) {
-               RTE_LOG(DEBUG, VHOST_DATA,
-                       "(%" PRIu64 ") TX: src and dst MAC is same. "
-                       "Dropping packet.\n", fh);
+       if (vdev->vid == dst_vdev->vid) {
+               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,
-               "(%" PRIu64 ") TX: MAC address is local\n", fh);
+       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, "(%" PRIu64 ") "
-                       "Device is marked for removal\n", fh);
+               RTE_LOG_DP(DEBUG, VHOST_DATA,
+                       "(%d) device is marked for removal\n", dst_vdev->vid);
                return 0;
        }
 
@@ -846,7 +845,7 @@ virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
  * and get its vlan tag, and offset if it is.
  */
 static inline int __attribute__((always_inline))
-find_local_dest(struct virtio_net *dev, struct rte_mbuf *m,
+find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
        uint32_t *offset, uint16_t *vlan_tag)
 {
        struct vhost_dev *dst_vdev;
@@ -856,10 +855,10 @@ find_local_dest(struct virtio_net *dev, struct rte_mbuf *m,
        if (!dst_vdev)
                return 0;
 
-       if (dst_vdev->dev->device_fh == dev->device_fh) {
-               RTE_LOG(DEBUG, VHOST_DATA,
-                       "(%" PRIu64 ") TX: src and dst MAC is same. "
-                       " Dropping packet.\n", dst_vdev->dev->device_fh);
+       if (vdev->vid == dst_vdev->vid) {
+               RTE_LOG_DP(DEBUG, VHOST_DATA,
+                       "(%d) TX: src and dst MAC is same. Dropping packet.\n",
+                       vdev->vid);
                return -1;
        }
 
@@ -869,12 +868,11 @@ find_local_dest(struct virtio_net *dev, struct rte_mbuf *m,
         * the packet length by plus it.
         */
        *offset  = VLAN_HLEN;
-       *vlan_tag = vlan_tags[(uint16_t)dst_vdev->dev->device_fh];
+       *vlan_tag = vlan_tags[vdev->vid];
 
-       RTE_LOG(DEBUG, VHOST_DATA,
-               "(%" PRIu64 ") TX: pkt to local VM device id: (%" PRIu64 ") "
-               "vlan tag: %u.\n",
-               dev->device_fh, dst_vdev->dev->device_fh, *vlan_tag);
+       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);
 
        return 0;
 }
@@ -937,7 +935,6 @@ virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
        struct mbuf_table *tx_q;
        unsigned offset = 0;
        const uint16_t lcore_id = rte_lcore_id();
-       struct virtio_net *dev = vdev->dev;
        struct ether_hdr *nh;
 
 
@@ -945,7 +942,7 @@ virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
        if (unlikely(is_broadcast_ether_addr(&nh->d_addr))) {
                struct vhost_dev *vdev2;
 
-               TAILQ_FOREACH(vdev2, &vhost_dev_list, next) {
+               TAILQ_FOREACH(vdev2, &vhost_dev_list, global_vdev_entry) {
                        virtio_xmit(vdev2, vdev, m);
                }
                goto queue2nic;
@@ -958,14 +955,15 @@ virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
        }
 
        if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
-               if (unlikely(find_local_dest(dev, m, &offset, &vlan_tag) != 0)) {
+               if (unlikely(find_local_dest(vdev, m, &offset,
+                                            &vlan_tag) != 0)) {
                        rte_pktmbuf_free(m);
                        return;
                }
        }
 
-       RTE_LOG(DEBUG, VHOST_DATA, "(%" PRIu64 ") TX: "
-               "MAC address is external\n", dev->device_fh);
+       RTE_LOG_DP(DEBUG, VHOST_DATA,
+               "(%d) TX: MAC address is external\n", vdev->vid);
 
 queue2nic:
 
@@ -1032,7 +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);
@@ -1043,7 +1041,6 @@ static inline void __attribute__((always_inline))
 drain_eth_rx(struct vhost_dev *vdev)
 {
        uint16_t rx_count, enqueue_count;
-       struct virtio_net *dev = vdev->dev;
        struct rte_mbuf *pkts[MAX_PKT_BURST];
 
        rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q,
@@ -1057,19 +1054,19 @@ drain_eth_rx(struct vhost_dev *vdev)
         * to diminish packet loss.
         */
        if (enable_retry &&
-           unlikely(rx_count > rte_vring_available_entries(dev,
+           unlikely(rx_count > rte_vhost_avail_entries(vdev->vid,
                        VIRTIO_RXQ))) {
                uint32_t retry;
 
                for (retry = 0; retry < burst_rx_retry_num; retry++) {
                        rte_delay_us(burst_rx_delay_time);
-                       if (rx_count <= rte_vring_available_entries(dev,
+                       if (rx_count <= rte_vhost_avail_entries(vdev->vid,
                                        VIRTIO_RXQ))
                                break;
                }
        }
 
-       enqueue_count = rte_vhost_enqueue_burst(dev, VIRTIO_RXQ,
+       enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
                                                pkts, rx_count);
        if (enable_stats) {
                rte_atomic64_add(&vdev->stats.rx_total_atomic, rx_count);
@@ -1086,7 +1083,7 @@ drain_virtio_tx(struct vhost_dev *vdev)
        uint16_t count;
        uint16_t i;
 
-       count = rte_vhost_dequeue_burst(vdev->dev, VIRTIO_TXQ, mbuf_pool,
+       count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ, mbuf_pool,
                                        pkts, MAX_PKT_BURST);
 
        /* setup VMDq for the first packet */
@@ -1095,10 +1092,8 @@ drain_virtio_tx(struct vhost_dev *vdev)
                        free_pkts(pkts, count);
        }
 
-       for (i = 0; i < count; ++i) {
-               virtio_tx_route(vdev, pkts[i],
-                       vlan_tags[(uint16_t)vdev->dev->device_fh]);
-       }
+       for (i = 0; i < count; ++i)
+               virtio_tx_route(vdev, pkts[i], vlan_tags[vdev->vid]);
 }
 
 /*
@@ -1149,7 +1144,8 @@ switch_worker(void *arg __rte_unused)
                /*
                 * Process vhost devices
                 */
-               TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list, next) {
+               TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list,
+                             lcore_vdev_entry) {
                        if (unlikely(vdev->remove)) {
                                unlink_vmdq(vdev);
                                vdev->ready = DEVICE_SAFE_REMOVE;
@@ -1174,22 +1170,27 @@ switch_worker(void *arg __rte_unused)
  * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
  */
 static void
-destroy_device (volatile struct virtio_net *dev)
+destroy_device(int vid)
 {
-       struct vhost_dev *vdev;
+       struct vhost_dev *vdev = NULL;
        int lcore;
 
-       dev->flags &= ~VIRTIO_DEV_RUNNING;
-
-       vdev = (struct vhost_dev *)dev->priv;
+       TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
+               if (vdev->vid == vid)
+                       break;
+       }
+       if (!vdev)
+               return;
        /*set the remove flag. */
        vdev->remove = 1;
        while(vdev->ready != DEVICE_SAFE_REMOVE) {
                rte_pause();
        }
 
-       TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev, next);
-       TAILQ_REMOVE(&vhost_dev_list, vdev, next);
+       TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev,
+                    lcore_vdev_entry);
+       TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry);
+
 
        /* Set the dev_removal_flag on each lcore. */
        RTE_LCORE_FOREACH_SLAVE(lcore)
@@ -1208,8 +1209,8 @@ destroy_device (volatile struct virtio_net *dev)
        lcore_info[vdev->coreid].device_num--;
 
        RTE_LOG(INFO, VHOST_DATA,
-               "(%" PRIu64 ") Device has been removed from data core\n",
-               dev->device_fh);
+               "(%d) device has been removed from data core\n",
+               vdev->vid);
 
        rte_free(vdev);
 }
@@ -1219,7 +1220,7 @@ destroy_device (volatile struct virtio_net *dev)
  * and the allocated to a specific data core.
  */
 static int
-new_device (struct virtio_net *dev)
+new_device(int vid)
 {
        int lcore, core_add = 0;
        uint32_t device_num_min = num_devices;
@@ -1227,16 +1228,15 @@ new_device (struct virtio_net *dev)
 
        vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
        if (vdev == NULL) {
-               RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Couldn't allocate memory for vhost dev\n",
-                       dev->device_fh);
+               RTE_LOG(INFO, VHOST_DATA,
+                       "(%d) couldn't allocate memory for vhost dev\n",
+                       vid);
                return -1;
        }
-       vdev->dev = dev;
-       dev->priv = vdev;
+       vdev->vid = vid;
 
-       TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, next);
-       vdev->vmdq_rx_q
-               = dev->device_fh * queues_per_pool + vmdq_queue_base;
+       TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry);
+       vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base;
 
        /*reset ready flag*/
        vdev->ready = DEVICE_MAC_LEARNING;
@@ -1251,15 +1251,17 @@ new_device (struct virtio_net *dev)
        }
        vdev->coreid = core_add;
 
-       TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev, next);
+       TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev,
+                         lcore_vdev_entry);
        lcore_info[vdev->coreid].device_num++;
 
        /* Disable notifications. */
-       rte_vhost_enable_guest_notification(dev, VIRTIO_RXQ, 0);
-       rte_vhost_enable_guest_notification(dev, VIRTIO_TXQ, 0);
-       dev->flags |= VIRTIO_DEV_RUNNING;
+       rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
+       rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
 
-       RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Device has been added to data core %d\n", dev->device_fh, vdev->coreid);
+       RTE_LOG(INFO, VHOST_DATA,
+               "(%d) device has been added to data core %d\n",
+               vid, vdev->coreid);
 
        return 0;
 }
@@ -1268,7 +1270,7 @@ new_device (struct virtio_net *dev)
  * 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,
@@ -1294,7 +1296,7 @@ print_stats(void)
                printf("%s%s\n", clr, top_left);
                printf("Device statistics =================================\n");
 
-               TAILQ_FOREACH(vdev, &vhost_dev_list, next) {
+               TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
                        tx_total   = vdev->stats.tx_total;
                        tx         = vdev->stats.tx;
                        tx_dropped = tx_total - tx;
@@ -1303,7 +1305,7 @@ print_stats(void)
                        rx         = rte_atomic64_read(&vdev->stats.rx_atomic);
                        rx_dropped = rx_total - rx;
 
-                       printf("Statistics for device %" PRIu64 "\n"
+                       printf("Statistics for device %d\n"
                                "-----------------------\n"
                                "TX total:              %" PRIu64 "\n"
                                "TX dropped:            %" PRIu64 "\n"
@@ -1311,7 +1313,7 @@ print_stats(void)
                                "RX total:              %" PRIu64 "\n"
                                "RX dropped:            %" PRIu64 "\n"
                                "RX successful:         %" PRIu64 "\n",
-                               vdev->dev->device_fh,
+                               vdev->vid,
                                tx_total, tx_dropped, tx,
                                rx_total, rx_dropped, rx);
                }
@@ -1320,14 +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 *)&dev_basename);
-       if (ret != 0)
-               rte_exit(EXIT_FAILURE, "vhost driver unregister failure.\n");
+       unregister_drivers(nb_sockets);
+
        exit(0);
 }
 
@@ -1367,7 +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);
 
@@ -1383,18 +1398,18 @@ 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];
+       uint64_t flags = 0;
 
        signal(SIGINT, sigint_handler);
 
@@ -1410,19 +1425,18 @@ 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");
 
        /* Get the number of physical ports. */
        nb_ports = rte_eth_dev_count();
-       if (nb_ports > RTE_MAX_ETHPORTS)
-               nb_ports = RTE_MAX_ETHPORTS;
 
        /*
         * Update the global var NUM_PORTS and global array PORTS
@@ -1476,7 +1490,7 @@ main(int argc, char *argv[])
                snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN, "print-stats");
                ret = rte_thread_setname(tid, thread_name);
                if (ret != 0)
-                       RTE_LOG(ERR, VHOST_CONFIG,
+                       RTE_LOG(DEBUG, VHOST_CONFIG,
                                "Cannot set print-stats name\n");
        }
 
@@ -1484,18 +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(cuse or user) driver to handle vhost messages. */
-       ret = rte_vhost_driver_register((char *)&dev_basename);
-       if (ret != 0)
-               rte_exit(EXIT_FAILURE, "vhost driver register failure.\n");
+       /* Register vhost user driver to handle vhost messages. */
+       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);
+               }
+
+               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);
+               }
 
-       rte_vhost_driver_callback_register(&virtio_net_device_ops);
+               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;
 
 }