15. L2 Forwarding Sample Application (in Real and Virtualized Environments)

15.1.1. Virtual Function Setup Instructions

This application can use the virtual function available in the system and therefore can be used in a virtual machine without passing through the whole Network Device into a guest machine in a virtualized scenario. The virtual functions can be enabled in the host machine or the hypervisor with the respective physical function driver.

For example, in a Linux* host machine, it is possible to enable a virtual function using the following command:

modprobe ixgbe max_vfs=2,2

This command enables two Virtual Functions on each of Physical Function of the NIC, with two physical ports in the PCI configuration space. It is important to note that enabled Virtual Function 0 and 2 would belong to Physical Function 0 and Virtual Function 1 and 3 would belong to Physical Function 1, in this case enabling a total of four Virtual Functions.

15.4.1. Command Line Arguments

The L2 Forwarding sample application takes specific parameters, in addition to Environment Abstraction Layer (EAL) arguments. The preferred way to parse parameters is to use the getopt() function, since it is part of a well-defined and portable library.

The parsing of arguments is done in the l2fwd_parse_args() function. The method of argument parsing is not described here. Refer to the glibc getopt(3) man page for details.

EAL arguments are parsed first, then application-specific arguments. This is done at the beginning of the main() function:

/* init EAL */

ret = rte_eal_init(argc, argv);
if (ret < 0)
    rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");

argc -= ret;
argv += ret;

/* parse application arguments (after the EAL ones) */

ret = l2fwd_parse_args(argc, argv);
if (ret < 0)
    rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n");

15.4.2. Mbuf Pool Initialization

Once the arguments are parsed, the mbuf pool is created. The mbuf pool contains a set of mbuf objects that will be used by the driver and the application to store network packet data:

/* create the mbuf pool */

l2fwd_pktmbuf_pool = rte_mempool_create("mbuf_pool", NB_MBUF, MBUF_SIZE, 32, sizeof(struct rte_pktmbuf_pool_private),
    rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL, SOCKET0, 0);

if (l2fwd_pktmbuf_pool == NULL)
    rte_panic("Cannot init mbuf pool\n");

The rte_mempool is a generic structure used to handle pools of objects. In this case, it is necessary to create a pool that will be used by the driver, which expects to have some reserved space in the mempool structure, sizeof(struct rte_pktmbuf_pool_private) bytes. The number of allocated pkt mbufs is NB_MBUF, with a size of MBUF_SIZE each. A per-lcore cache of 32 mbufs is kept. The memory is allocated in NUMA socket 0, but it is possible to extend this code to allocate one mbuf pool per socket.

Two callback pointers are also given to the rte_mempool_create() function:

• The first callback pointer is to rte_pktmbuf_pool_init() and is used to initialize the private data of the mempool, which is needed by the driver. This function is provided by the mbuf API, but can be copied and extended by the developer.
• The second callback pointer given to rte_mempool_create() is the mbuf initializer. The default is used, that is, rte_pktmbuf_init(), which is provided in the rte_mbuf library. If a more complex application wants to extend the rte_pktmbuf structure for its own needs, a new function derived from rte_pktmbuf_init( ) can be created.

15.4.3. Driver Initialization

The main part of the code in the main() function relates to the initialization of the driver. To fully understand this code, it is recommended to study the chapters that related to the Poll Mode Driver in the DPDK Programmer’s Guide - Rel 1.4 EAR and the DPDK API Reference.

if (rte_eal_pci_probe() < 0)
    rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");

nb_ports = rte_eth_dev_count();

if (nb_ports == 0)
    rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");

/* reset l2fwd_dst_ports */

for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++)
    l2fwd_dst_ports[portid] = 0;

last_port = 0;

/*
 * Each logical core is assigned a dedicated TX queue on each port.
 */

for (portid = 0; portid < nb_ports; portid++) {
    /* skip ports that are not enabled */

    if ((l2fwd_enabled_port_mask & (1 << portid)) == 0)
       continue;

    if (nb_ports_in_mask % 2) {
        l2fwd_dst_ports[portid] = last_port;
        l2fwd_dst_ports[last_port] = portid;
    }
    else
       last_port = portid;

    nb_ports_in_mask++;

    rte_eth_dev_info_get((uint8_t) portid, &dev_info);
}

Observe that:

• rte_igb_pmd_init_all() simultaneously registers the driver as a PCI driver and as an Ethernet* Poll Mode Driver.
• rte_eal_pci_probe() parses the devices on the PCI bus and initializes recognized devices.

The next step is to configure the RX and TX queues. For each port, there is only one RX queue (only one lcore is able to poll a given port). The number of TX queues depends on the number of available lcores. The rte_eth_dev_configure() function is used to configure the number of queues for a port:

ret = rte_eth_dev_configure((uint8_t)portid, 1, 1, &port_conf);
if (ret < 0)
    rte_exit(EXIT_FAILURE, "Cannot configure device: "
        "err=%d, port=%u\n",
        ret, portid);

The global configuration is stored in a static structure:

static const struct rte_eth_conf port_conf = {
    .rxmode = {
        .split_hdr_size = 0,
        .header_split = 0,   /**< Header Split disabled */
        .hw_ip_checksum = 0, /**< IP checksum offload disabled */
        .hw_vlan_filter = 0, /**< VLAN filtering disabled */
        .jumbo_frame = 0,    /**< Jumbo Frame Support disabled */
        .hw_strip_crc= 0,    /**< CRC stripped by hardware */
    },

    .txmode = {
        .mq_mode = ETH_DCB_NONE
    },
};

15.4.4. RX Queue Initialization

The application uses one lcore to poll one or several ports, depending on the -q option, which specifies the number of queues per lcore.

For example, if the user specifies -q 4, the application is able to poll four ports with one lcore. If there are 16 ports on the target (and if the portmask argument is -p ffff ), the application will need four lcores to poll all the ports.

ret = rte_eth_rx_queue_setup((uint8_t) portid, 0, nb_rxd, SOCKET0, &rx_conf, l2fwd_pktmbuf_pool);
if (ret < 0)

    rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
        "err=%d, port=%u\n",
        ret, portid);

The list of queues that must be polled for a given lcore is stored in a private structure called struct lcore_queue_conf.

struct lcore_queue_conf {
    unsigned n_rx_port;
    unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
    struct mbuf_table tx_mbufs[L2FWD_MAX_PORTS];
} rte_cache_aligned;

struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];

The values n_rx_port and rx_port_list[] are used in the main packet processing loop (see Receive, Process and Transmit Packets).

The global configuration for the RX queues is stored in a static structure:

static const struct rte_eth_rxconf rx_conf = {
    .rx_thresh = {
        .pthresh = RX_PTHRESH,
        .hthresh = RX_HTHRESH,
        .wthresh = RX_WTHRESH,
    },
};

15.4.5. TX Queue Initialization

Each lcore should be able to transmit on any port. For every port, a single TX queue is initialized.

/* init one TX queue on each port */

fflush(stdout);

ret = rte_eth_tx_queue_setup((uint8_t) portid, 0, nb_txd, rte_eth_dev_socket_id(portid), &tx_conf);
if (ret < 0)
    rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n", ret, (unsigned) portid);

The global configuration for TX queues is stored in a static structure:

static const struct rte_eth_txconf tx_conf = {
    .tx_thresh = {
        .pthresh = TX_PTHRESH,
        .hthresh = TX_HTHRESH,
        .wthresh = TX_WTHRESH,
    },
    .tx_free_thresh = RTE_TEST_TX_DESC_DEFAULT + 1, /* disable feature */
};

15.4.6. Receive, Process and Transmit Packets

In the l2fwd_main_loop() function, the main task is to read ingress packets from the RX queues. This is done using the following code:

/*
 * Read packet from RX queues
 */

for (i = 0; i < qconf->n_rx_port; i++) {
    portid = qconf->rx_port_list[i];
    nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,  pkts_burst, MAX_PKT_BURST);

    for (j = 0; j < nb_rx; j++) {
        m = pkts_burst[j];
        rte_prefetch0[rte_pktmbuf_mtod(m, void *)); l2fwd_simple_forward(m, portid);
    }
}

Packets are read in a burst of size MAX_PKT_BURST. The rte_eth_rx_burst() function writes the mbuf pointers in a local table and returns the number of available mbufs in the table.

Then, each mbuf in the table is processed by the l2fwd_simple_forward() function. The processing is very simple: process the TX port from the RX port, then replace the source and destination MAC addresses if MAC addresses updating is enabled.

During the initialization process, a static array of destination ports (l2fwd_dst_ports[]) is filled such that for each source port, a destination port is assigned that is either the next or previous enabled port from the portmask. Naturally, the number of ports in the portmask must be even, otherwise, the application exits.

static void
l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid)
{
    struct ether_hdr *eth;
    void *tmp;
    unsigned dst_port;

    dst_port = l2fwd_dst_ports[portid];

    eth = rte_pktmbuf_mtod(m, struct ether_hdr *);

    /* 02:00:00:00:00:xx */

    tmp = &eth->d_addr.addr_bytes[0];

    *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t) dst_port << 40);

    /* src addr */

    ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], &eth->s_addr);

    l2fwd_send_packet(m, (uint8_t) dst_port);
}

Then, the packet is sent using the l2fwd_send_packet (m, dst_port) function. For this test application, the processing is exactly the same for all packets arriving on the same RX port. Therefore, it would have been possible to call the l2fwd_send_burst() function directly from the main loop to send all the received packets on the same TX port, using the burst-oriented send function, which is more efficient.

However, in real-life applications (such as, L3 routing), packet N is not necessarily forwarded on the same port as packet N-1. The application is implemented to illustrate that, so the same approach can be reused in a more complex application.

The l2fwd_send_packet() function stores the packet in a per-lcore and per-txport table. If the table is full, the whole packets table is transmitted using the l2fwd_send_burst() function:

/* Send the packet on an output interface */

static int
l2fwd_send_packet(struct rte_mbuf *m, uint8_t port)
{
    unsigned lcore_id, len;
    struct lcore_queue_conf *qconf;

    lcore_id = rte_lcore_id();
    qconf = &lcore_queue_conf[lcore_id];
    len = qconf->tx_mbufs[port].len;
    qconf->tx_mbufs[port].m_table[len] = m;
    len++;

    /* enough pkts to be sent */

    if (unlikely(len == MAX_PKT_BURST)) {
        l2fwd_send_burst(qconf, MAX_PKT_BURST, port);
        len = 0;
    }

    qconf->tx_mbufs[port].len = len; return 0;
}

To ensure that no packets remain in the tables, each lcore does a draining of TX queue in its main loop. This technique introduces some latency when there are not many packets to send, however it improves performance:

cur_tsc = rte_rdtsc();

/*
 *   TX burst queue drain
 */

diff_tsc = cur_tsc - prev_tsc;

if (unlikely(diff_tsc > drain_tsc)) {
    for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
        if (qconf->tx_mbufs[portid].len == 0)
            continue;

        l2fwd_send_burst(&lcore_queue_conf[lcore_id], qconf->tx_mbufs[portid].len, (uint8_t) portid);

        qconf->tx_mbufs[portid].len = 0;
    }

    /* if timer is enabled */

    if (timer_period > 0) {
        /* advance the timer */

        timer_tsc += diff_tsc;

        /* if timer has reached its timeout */

        if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
            /* do this only on master core */

            if (lcore_id == rte_get_master_lcore()) {
                print_stats();

                /* reset the timer */
                timer_tsc = 0;
            }
        }
    }

    prev_tsc = cur_tsc;
}