Beispiel #1
0
/* Init eth TX queue */
struct vr_dpdk_queue *
vr_dpdk_ethdev_tx_queue_init(unsigned lcore_id, struct vr_interface *vif,
    unsigned queue_or_lcore_id)
{
    uint8_t port_id;
    uint16_t tx_queue_id = queue_or_lcore_id;
    unsigned int vif_idx = vif->vif_idx, dpdk_queue_index;
    const unsigned int socket_id = rte_lcore_to_socket_id(lcore_id);

    struct vr_dpdk_ethdev *ethdev;
    struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
    struct vr_dpdk_queue *tx_queue;
    struct vr_dpdk_queue_params *tx_queue_params;

    ethdev = (struct vr_dpdk_ethdev *)vif->vif_os;
    port_id = ethdev->ethdev_port_id;

    if (lcore->lcore_hw_queue_to_dpdk_index[vif->vif_idx]) {
        dpdk_queue_index =
            lcore->lcore_hw_queue_to_dpdk_index[vif->vif_idx][tx_queue_id];
    } else {
        dpdk_queue_index = 0;
    }

    tx_queue = &lcore->lcore_tx_queues[vif_idx][dpdk_queue_index];
    tx_queue_params = &lcore->lcore_tx_queue_params[vif_idx][dpdk_queue_index];

    /* init queue */
    tx_queue->txq_ops = rte_port_ethdev_writer_ops;
    tx_queue->q_queue_h = NULL;
    tx_queue->q_vif = vrouter_get_interface(vif->vif_rid, vif_idx);

    /* create the queue */
    struct rte_port_ethdev_writer_params writer_params = {
        .port_id = port_id,
        .queue_id = tx_queue_id,
        .tx_burst_sz = VR_DPDK_TX_BURST_SZ,
    };
    tx_queue->q_queue_h = tx_queue->txq_ops.f_create(&writer_params, socket_id);
    if (tx_queue->q_queue_h == NULL) {
        RTE_LOG(ERR, VROUTER, "    error creating eth device %" PRIu8
                " TX queue %" PRIu16 "\n", port_id, tx_queue_id);
        return NULL;
    }

    /* store queue params */
    tx_queue_params->qp_release_op = &dpdk_ethdev_tx_queue_release;
    tx_queue_params->qp_ethdev.queue_id = tx_queue_id;
    tx_queue_params->qp_ethdev.port_id = port_id;

    /* for the queue 0 add queue params to the list of bonds to TX */
    if (ethdev->ethdev_nb_slaves > 0 && tx_queue_id == 0) {
        /* make sure queue params have been stored */
        rte_wmb();
        lcore->lcore_bonds_to_tx[lcore->lcore_nb_bonds_to_tx++] = tx_queue_params;
        RTE_VERIFY(lcore->lcore_nb_bonds_to_tx <= VR_DPDK_MAX_BONDS);
    }

    return tx_queue;
}
/*
 * vr_dpdk_virtio_rx_queue_init - initializes a virtio RX queue.
 *
 * Returns a pointer to the RX queue on success, NULL otherwise.
 */
struct vr_dpdk_queue *
vr_dpdk_virtio_rx_queue_init(unsigned int lcore_id, struct vr_interface *vif,
                             unsigned int queue_or_lcore_id)
{
    uint16_t queue_id = queue_or_lcore_id;
    struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
    const unsigned int socket_id = rte_lcore_to_socket_id(lcore_id);
    unsigned int vif_idx = vif->vif_idx;
    struct vr_dpdk_queue *rx_queue = &lcore->lcore_rx_queues[vif_idx];
    struct vr_dpdk_queue_params *rx_queue_params =
        &lcore->lcore_rx_queue_params[vif_idx];

    /* Check input parameters */
    if (queue_id >= vr_dpdk_virtio_nrxqs(vif)) {
        RTE_LOG(ERR, VROUTER, "    error creating virtio device %s RX queue %"
            PRIu16 "\n", vif->vif_name, queue_id);
        return NULL;
    }

    /* init queue */
    rx_queue->rxq_ops = vr_dpdk_virtio_reader_ops;
    rx_queue->q_vif = vrouter_get_interface(vif->vif_rid, vif_idx);

    /* init virtio queue */
    vr_dpdk_virtio_rxqs[vif_idx][queue_id].vdv_ready_state = VQ_NOT_READY;
    vr_dpdk_virtio_rxqs[vif_idx][queue_id].vdv_last_used_idx = 0;
    vr_dpdk_virtio_rxqs[vif_idx][queue_id].vdv_last_used_idx_res = 0;
    vr_dpdk_virtio_rxqs[vif_idx][queue_id].vdv_vif_idx = vif->vif_idx;

    /* create the queue */
    struct dpdk_virtio_reader_params reader_params = {
        .rx_virtioq = &vr_dpdk_virtio_rxqs[vif_idx][queue_id],
    };
    rx_queue->q_queue_h = rx_queue->rxq_ops.f_create(&reader_params, socket_id);
    if (rx_queue->q_queue_h == NULL) {
        RTE_LOG(ERR, VROUTER, "    error creating virtio device %s RX queue %"
            PRIu16 "\n", vif->vif_name, queue_id);
        return NULL;
    }

    /* store queue params */
    rx_queue_params->qp_release_op = &dpdk_virtio_rx_queue_release;

    return rx_queue;
}
/* Init eth RX queue */
struct vr_dpdk_queue *
vr_dpdk_ethdev_rx_queue_init(unsigned lcore_id, struct vr_interface *vif,
    unsigned queue_or_lcore_id)
{
    uint16_t rx_queue_id = queue_or_lcore_id;
    uint8_t port_id;
    unsigned int vif_idx = vif->vif_idx;
    const unsigned int socket_id = rte_lcore_to_socket_id(lcore_id);

    struct vr_dpdk_ethdev *ethdev;
    struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
    struct vr_dpdk_queue *rx_queue = &lcore->lcore_rx_queues[vif_idx];
    struct vr_dpdk_queue_params *rx_queue_params
                    = &lcore->lcore_rx_queue_params[vif_idx];

    ethdev = (struct vr_dpdk_ethdev *)vif->vif_os;
    port_id = ethdev->ethdev_port_id;

    /* init queue */
    rx_queue->rxq_ops = rte_port_ethdev_reader_ops;
    rx_queue->q_queue_h = NULL;
    rx_queue->q_vif = vrouter_get_interface(vif->vif_rid, vif_idx);

    /* create the queue */
    struct rte_port_ethdev_reader_params reader_params = {
        .port_id = port_id,
        .queue_id = rx_queue_id,
    };
    rx_queue->q_queue_h = rx_queue->rxq_ops.f_create(&reader_params, socket_id);
    if (rx_queue->q_queue_h == NULL) {
        RTE_LOG(ERR, VROUTER, "    error creating eth device %" PRIu8
                " RX queue %" PRIu16 "\n", port_id, rx_queue_id);
        return NULL;
    }

    /* store queue params */
    rx_queue_params->qp_release_op = &dpdk_ethdev_rx_queue_release;
    rx_queue_params->qp_ethdev.queue_id = rx_queue_id;
    rx_queue_params->qp_ethdev.port_id = port_id;

    return rx_queue;
}
/*
 * vr_dpdk_virtio_tx_queue_init - initializes a virtio TX queue.
 *
 * Returns a pointer to the TX queue on success, NULL otherwise.
 */
struct vr_dpdk_queue *
vr_dpdk_virtio_tx_queue_init(unsigned int lcore_id, struct vr_interface *vif,
                             unsigned int queue_or_lcore_id)
{
    uint16_t queue_id = queue_or_lcore_id;
    struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
    const unsigned int socket_id = rte_lcore_to_socket_id(lcore_id);
    unsigned int vif_idx = vif->vif_idx;
    struct vr_dpdk_queue *tx_queue = &lcore->lcore_tx_queues[vif_idx];
    struct vr_dpdk_queue_params *tx_queue_params
                = &lcore->lcore_tx_queue_params[vif_idx];

    /* Check input parameters */
    /* virtio TX is thread safe, so just use one of the rings */
    queue_id = queue_id % vr_dpdk_virtio_ntxqs(vif);

    /* init queue */
    tx_queue->txq_ops = vr_dpdk_virtio_writer_ops;
    tx_queue->q_vif = vrouter_get_interface(vif->vif_rid, vif_idx);

    /* init virtio queue */
    vr_dpdk_virtio_txqs[vif_idx][queue_id].vdv_ready_state = VQ_NOT_READY;
    vr_dpdk_virtio_txqs[vif_idx][queue_id].vdv_last_used_idx = 0;
    vr_dpdk_virtio_txqs[vif_idx][queue_id].vdv_last_used_idx_res = 0;
    vr_dpdk_virtio_txqs[vif_idx][queue_id].vdv_vif_idx = vif->vif_idx;

    /* create the queue */
    struct dpdk_virtio_writer_params writer_params = {
        .tx_virtioq = &vr_dpdk_virtio_txqs[vif_idx][queue_id],
    };
    tx_queue->q_queue_h = tx_queue->txq_ops.f_create(&writer_params, socket_id);
    if (tx_queue->q_queue_h == NULL) {
        RTE_LOG(ERR, VROUTER, "    error creating virtio device %s TX queue %"
            PRIu16 "\n", vif->vif_name, queue_id);
        return NULL;
    }

    /* store queue params */
    tx_queue_params->qp_release_op = &dpdk_virtio_tx_queue_release;

    return tx_queue;
}
/* Init KNI RX queue */
struct vr_dpdk_queue *
vr_dpdk_kni_rx_queue_init(unsigned lcore_id, struct vr_interface *vif,
    unsigned host_lcore_id)
{
    struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
    const unsigned socket_id = rte_lcore_to_socket_id(lcore_id);
    uint8_t port_id = 0;
    unsigned vif_idx = vif->vif_idx;
    struct vr_dpdk_queue *rx_queue = &lcore->lcore_rx_queues[vif_idx];
    struct vr_dpdk_queue_params *rx_queue_params
                    = &lcore->lcore_rx_queue_params[vif_idx];

    if (vif->vif_type == VIF_TYPE_HOST) {
        port_id = (((struct vr_dpdk_ethdev *)(vif->vif_bridge->vif_os))->
                ethdev_port_id);
    }

    /* init queue */
    rx_queue->rxq_ops = dpdk_knidev_reader_ops;
    rx_queue->q_queue_h = NULL;
    rx_queue->q_vif = vrouter_get_interface(vif->vif_rid, vif_idx);

    /* create the queue */
    struct dpdk_knidev_reader_params reader_params = {
        .kni = vif->vif_os,
    };
    rx_queue->q_queue_h = rx_queue->rxq_ops.f_create(&reader_params, socket_id);
    if (rx_queue->q_queue_h == NULL) {
        RTE_LOG(ERR, VROUTER, "    error creating KNI device %s RX queue"
            " at eth device %" PRIu8 "\n", vif->vif_name, port_id);
        return NULL;
    }

    /* store queue params */
    rx_queue_params->qp_release_op = &dpdk_kni_rx_queue_release;

    return rx_queue;
}

/* Release KNI TX queue */
static void
dpdk_kni_tx_queue_release(unsigned lcore_id, struct vr_interface *vif)
{
    struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
    struct vr_dpdk_queue *tx_queue = &lcore->lcore_tx_queues[vif->vif_idx];
    struct vr_dpdk_queue_params *tx_queue_params
                        = &lcore->lcore_tx_queue_params[vif->vif_idx];

    tx_queue->txq_ops.f_tx = NULL;
    rte_wmb();

    /* flush and free the queue */
    if (tx_queue->txq_ops.f_free(tx_queue->q_queue_h)) {
        RTE_LOG(ERR, VROUTER, "    error freeing lcore %u KNI device TX queue\n",
                    lcore_id);
    }

    /* reset the queue */
    vrouter_put_interface(tx_queue->q_vif);
    memset(tx_queue, 0, sizeof(*tx_queue));
    memset(tx_queue_params, 0, sizeof(*tx_queue_params));
}

/* Init KNI TX queue */
struct vr_dpdk_queue *
vr_dpdk_kni_tx_queue_init(unsigned lcore_id, struct vr_interface *vif,
    unsigned host_lcore_id)
{
    struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
    const unsigned socket_id = rte_lcore_to_socket_id(lcore_id);
    uint8_t port_id = 0;
    unsigned vif_idx = vif->vif_idx;
    struct vr_dpdk_queue *tx_queue = &lcore->lcore_tx_queues[vif_idx];
    struct vr_dpdk_queue_params *tx_queue_params
                    = &lcore->lcore_tx_queue_params[vif_idx];
    struct vr_dpdk_ethdev *ethdev;

    if (vif->vif_type == VIF_TYPE_HOST) {
        ethdev = vif->vif_bridge->vif_os;
        if (ethdev == NULL) {
            RTE_LOG(ERR, VROUTER, "    error creating KNI device %s TX queue:"
                " bridge vif %u ethdev is not initialized\n",
                vif->vif_name, vif->vif_bridge->vif_idx);
            return NULL;
        }
        port_id = ethdev->ethdev_port_id;
    }

    /* init queue */
    tx_queue->txq_ops = dpdk_knidev_writer_ops;
    tx_queue->q_queue_h = NULL;
    tx_queue->q_vif = vrouter_get_interface(vif->vif_rid, vif_idx);

    /* create the queue */
    struct dpdk_knidev_writer_params writer_params = {
        .kni = vif->vif_os,
        .tx_burst_sz = VR_DPDK_TX_BURST_SZ,
    };
    tx_queue->q_queue_h = tx_queue->txq_ops.f_create(&writer_params, socket_id);
    if (tx_queue->q_queue_h == NULL) {
        RTE_LOG(ERR, VROUTER, "    error creating KNI device %s TX queue"
            " at eth device %" PRIu8 "\n", vif->vif_name, port_id);
        return NULL;
    }

    /* store queue params */
    tx_queue_params->qp_release_op = &dpdk_kni_tx_queue_release;

    return tx_queue;
}

/* Change KNI MTU size callback */
static int
dpdk_knidev_change_mtu(uint8_t port_id, unsigned new_mtu)
{
    struct vrouter *router = vrouter_get(0);
    struct vr_interface *vif;
    int i, ret;
    uint8_t ethdev_port_id, slave_port_id;
    struct vr_dpdk_ethdev *ethdev = NULL;

    RTE_LOG(INFO, VROUTER, "Changing eth device %" PRIu8 " MTU to %u\n",
                    port_id, new_mtu);
    if (port_id >= rte_eth_dev_count()) {
        RTE_LOG(ERR, VROUTER, "Error changing eth device %"PRIu8" MTU: invalid eth device\n", port_id);
        return -EINVAL;
    }

    /*
     * TODO: DPDK bond PMD does not implement mtu_set op, so we need to
     * set the MTU manually for all the slaves.
     */
    /* Bond vif uses first slave port ID. */
    if (router->vr_eth_if) {
        ethdev = (struct vr_dpdk_ethdev *)router->vr_eth_if->vif_os;
        if (ethdev && ethdev->ethdev_nb_slaves > 0) {
            for (i = 0; i < ethdev->ethdev_nb_slaves; i++) {
                if (port_id == ethdev->ethdev_slaves[i])
                    break;
            }
            /* Clear ethdev if no port match. */
            if (i >= ethdev->ethdev_nb_slaves)
                ethdev = NULL;
        }
    }
    if (ethdev && ethdev->ethdev_nb_slaves > 0) {
        for (i = 0; i < ethdev->ethdev_nb_slaves; i++) {
            slave_port_id = ethdev->ethdev_slaves[i];
            RTE_LOG(INFO, VROUTER, "    changing bond member eth device %" PRIu8
                " MTU to %u\n", slave_port_id, new_mtu);

            ret =  rte_eth_dev_set_mtu(slave_port_id, new_mtu);
            if (ret < 0) {
                RTE_LOG(ERR, VROUTER, "    error changing bond member eth device %" PRIu8
                    " MTU: %s (%d)\n", slave_port_id, rte_strerror(-ret), -ret);
                return ret;
            }
        }
    } else {
        ret =  rte_eth_dev_set_mtu(port_id, new_mtu);
        if (ret < 0) {
            RTE_LOG(ERR, VROUTER, "Error changing eth device %" PRIu8
                " MTU: %s (%d)\n", port_id, rte_strerror(-ret), -ret);
        }
        return ret;
    }

    /* On success, inform vrouter about new MTU */
    for (i = 0; i < router->vr_max_interfaces; i++) {
        vif = __vrouter_get_interface(router, i);
        if (vif && (vif->vif_type == VIF_TYPE_PHYSICAL)) {
            ethdev_port_id = (((struct vr_dpdk_ethdev *)(vif->vif_os))->
                        ethdev_port_id);
            if (ethdev_port_id == port_id) {
                /* Ethernet header size */
                new_mtu += sizeof(struct vr_eth);
                if (vr_dpdk.vlan_tag != VLAN_ID_INVALID) {
                    /* 802.1q header size */
                    new_mtu += sizeof(uint32_t);
                }
                vif->vif_mtu = new_mtu;
                if (vif->vif_bridge)
                    vif->vif_bridge->vif_mtu = new_mtu;
            }
        }
    }

    return 0;
}


/* Configure KNI state callback */
static int
dpdk_knidev_config_network_if(uint8_t port_id, uint8_t if_up)
{
    int ret = 0;

    RTE_LOG(INFO, VROUTER, "Configuring eth device %" PRIu8 " %s\n",
                    port_id, if_up ? "UP" : "DOWN");
    if (port_id >= rte_eth_dev_count() || port_id >= RTE_MAX_ETHPORTS) {
        RTE_LOG(ERR, VROUTER, "Invalid eth device %" PRIu8 "\n", port_id);
        return -EINVAL;
    }

    if (if_up)
        ret = rte_eth_dev_start(port_id);
    else
        rte_eth_dev_stop(port_id);

    if (ret < 0) {
        RTE_LOG(ERR, VROUTER, "Configuring eth device %" PRIu8 " UP"
                    "failed (%d)\n", port_id, ret);
    }

    return ret;
}

/* Init KNI */
int
vr_dpdk_knidev_init(uint8_t port_id, struct vr_interface *vif)
{
    int i;
    struct rte_eth_dev_info dev_info;
    struct rte_kni_conf kni_conf;
    struct rte_kni_ops kni_ops;
    struct rte_kni *kni;
    struct rte_config *rte_conf = rte_eal_get_configuration();

    if (!vr_dpdk.kni_inited) {
        /*
         * If the host does not support KNIs (i.e. RedHat), we'll get
         * a panic here.
         */
        rte_kni_init(VR_DPDK_MAX_KNI_INTERFACES);
        vr_dpdk.kni_inited = true;
    }

    /* get eth device info */
    memset(&dev_info, 0, sizeof(dev_info));
    rte_eth_dev_info_get(port_id, &dev_info);

    /* create KNI configuration */
    memset(&kni_conf, 0, sizeof(kni_conf));
    strncpy(kni_conf.name, (char *)vif->vif_name, sizeof(kni_conf.name) - 1);

    kni_conf.addr = dev_info.pci_dev->addr;
    kni_conf.id = dev_info.pci_dev->id;
    kni_conf.group_id = port_id;
    kni_conf.mbuf_size = VR_DPDK_MAX_PACKET_SZ;
    /*
     * Due to DPDK commit 41a6ebd, now to prevent packet reordering in KNI
     * we have to bind KNI kernel thread to a first online unused CPU.
     */
    for (i = 0; i < RTE_MAX_LCORE; i++) {
        if (lcore_config[i].detected
                && rte_conf->lcore_role[VR_DPDK_FWD_LCORE_ID + i] == ROLE_OFF) {
            kni_conf.force_bind = 1;
            kni_conf.core_id = i;
            RTE_LOG(INFO, VROUTER, "    bind KNI kernel thread to CPU %d\n", i);
            break;
        }
    }

    /* KNI options
     *
     * Changing state of the KNI interface can change state of the physical
     * interface. This is useful for the vhost, but not for the VLAN
     * forwarding interface.
     */
    if (vif->vif_type == VIF_TYPE_VLAN) {
        memset(&kni_ops, 0, sizeof(kni_ops));
    } else {
        kni_ops.port_id = port_id;
        kni_ops.change_mtu = dpdk_knidev_change_mtu;
        kni_ops.config_network_if = dpdk_knidev_config_network_if;
    }

    /* allocate KNI device */
    kni = rte_kni_alloc(vr_dpdk.rss_mempool, &kni_conf, &kni_ops);
    if (kni == NULL) {
        RTE_LOG(ERR, VROUTER, "    error allocation KNI device %s"
            " at eth device %" PRIu8 "\n", vif->vif_name, port_id);
        return -ENOMEM;
    }

    /* store pointer to KNI for further use */
    vif->vif_os = kni;

    /* add interface to the table of KNIs */
    for (i = 0; i < VR_DPDK_MAX_KNI_INTERFACES; i++) {
        if (vr_dpdk.knis[i] == NULL) {
            vr_dpdk.knis[i] = vif->vif_os;
            break;
        }
    }

    return 0;
}