rte_eth_tx_burst无法发送数据包

Question

一个dpdk应用程序，它生成少量的arp请求包并调用rte_eth_tx_burst将它们发送出去，一些数据包没有被对等NIC端口接收(这可以通过使用wireshark从对等NIC捕获数据包来确认)，dpdk-proc-info显示没有错误计数。但是在调用rte_eth_tx_burst之前让应用程序休眠10，它可以发送所有的数据包.

示例代码：

main(){
  port_init();
  sleep(10);
  gen_pkt(mbuf);
  rte_eth_tx_burst(mbuf);
}

系统设置:Ubuntu20.04.2LTS，dpdk-稳定-20.11.3，I350千兆网络连接1521，igb_uio驱动程序

root@k8s-node:/home/dpdk-stable-20.11.3/build/app# ./dpdk-proc-info -- --xstats
EAL: No legacy callbacks, legacy socket not created
###### NIC extended statistics for port 0  #########
####################################################
rx_good_packets: 10
tx_good_packets: 32
rx_good_bytes: 1203
tx_good_bytes: 1920
rx_missed_errors: 0
rx_errors: 0
tx_errors: 0
rx_mbuf_allocation_errors: 0
rx_q0_packets: 0
rx_q0_bytes: 0
rx_q0_errors: 0
tx_q0_packets: 0
tx_q0_bytes: 0
rx_crc_errors: 0
rx_align_errors: 0
rx_symbol_errors: 0
rx_missed_packets: 0
tx_single_collision_packets: 0
tx_multiple_collision_packets: 0
tx_excessive_collision_packets: 0
tx_late_collisions: 0
tx_total_collisions: 0
tx_deferred_packets: 0
tx_no_carrier_sense_packets: 0
rx_carrier_ext_errors: 0
rx_length_errors: 0
rx_xon_packets: 0
tx_xon_packets: 0
rx_xoff_packets: 0
tx_xoff_packets: 0
rx_flow_control_unsupported_packets: 0
rx_size_64_packets: 4
rx_size_65_to_127_packets: 3
rx_size_128_to_255_packets: 3
rx_size_256_to_511_packets: 0
rx_size_512_to_1023_packets: 0
rx_size_1024_to_max_packets: 0
rx_broadcast_packets: 0
rx_multicast_packets: 10
rx_undersize_errors: 0
rx_fragment_errors: 0
rx_oversize_errors: 0
rx_jabber_errors: 0
rx_management_packets: 0
rx_management_dropped: 0
tx_management_packets: 0
rx_total_packets: 10
tx_total_packets: 32
rx_total_bytes: 1203
tx_total_bytes: 1920
tx_size_64_packets: 32
tx_size_65_to_127_packets: 0
tx_size_128_to_255_packets: 0
tx_size_256_to_511_packets: 0
tx_size_512_to_1023_packets: 0
tx_size_1023_to_max_packets: 0
tx_multicast_packets: 0
tx_broadcast_packets: 32
tx_tso_packets: 0
tx_tso_errors: 0
rx_sent_to_host_packets: 0
tx_sent_by_host_packets: 0
rx_code_violation_packets: 0
interrupt_assert_count: 0
####################################################

root@k8s-node:/home/dpdk-stable-20.11.3/build/app# ./dpdk-proc-info -- --stats
EAL: No legacy callbacks, legacy socket not created

  ######################## NIC statistics for port 0  ########################
  RX-packets: 5           RX-errors:  0           RX-bytes:  785       
  RX-nombuf:  0         
  TX-packets: 32          TX-errors:  0           TX-bytes:  1920      

  Stats reg  0 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  1 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  2 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  3 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  4 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  5 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  6 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  7 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  8 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg  9 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg 10 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg 11 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg 12 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg 13 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg 14 RX-packets: 0           RX-errors: 0           RX-bytes: 0         
  Stats reg 15 RX-packets: 0           RX-errors: 0           RX-bytes: 0         

  Stats reg  0 TX-packets: 0           TX-bytes: 0         
  Stats reg  1 TX-packets: 0           TX-bytes: 0         
  Stats reg  2 TX-packets: 0           TX-bytes: 0         
  Stats reg  3 TX-packets: 0           TX-bytes: 0         
  Stats reg  4 TX-packets: 0           TX-bytes: 0         
  Stats reg  5 TX-packets: 0           TX-bytes: 0         
  Stats reg  6 TX-packets: 0           TX-bytes: 0         
  Stats reg  7 TX-packets: 0           TX-bytes: 0         
  Stats reg  8 TX-packets: 0           TX-bytes: 0         
  Stats reg  9 TX-packets: 0           TX-bytes: 0         
  Stats reg 10 TX-packets: 0           TX-bytes: 0         
  Stats reg 11 TX-packets: 0           TX-bytes: 0         
  Stats reg 12 TX-packets: 0           TX-bytes: 0         
  Stats reg 13 TX-packets: 0           TX-bytes: 0         
  Stats reg 14 TX-packets: 0           TX-bytes: 0         
  Stats reg 15 TX-packets: 0           TX-bytes: 0         
  ############################################################################

更新：

谢谢你的回复，我修改了密码：

main(){
  uint32_t port_mask = 0x1;
  port_init();
  check_all_ports_link_status(port_mask);
  gen_pkt(mbuf);
  rte_eth_tx_burst(mbuf);
}

拿到了打印日志：

Checking link status...............................
done
Port0 Link Up. Speed 1000 Mbps - full-duplex

我认为NIC应该已经完全实现了，但是对等NIC端口仍然漏掉了很多数据包。

Answer 1

在大多数工作情况下，物理NIC被枚举为双工(完整/半)、速度(1、10、25、40、50、100、200)，以及在1秒内协商的(自动/禁用)。任何超过2或3秒的标志是连接的机器或开关不能与双工，速度或自动协商。因此，建议是

1. 如果接口是NIC的话，更新驱动程序和固件。
2. 测试不同的连接电缆，因为链路可能达不到正确。
3. 在轮毂或开关的情况下，尝试固定速度和自动协商.
4. 我不建议改变从全双工到半双工(因为它可能是电缆或SFI问题)。

作为暂时的工作，您可以使用到达，这也说明它可能需要It might need to wait up to 9 seconds。

注意:测试是否存在电缆问题的简单方法是在两端运行DPDK，以检查链接启动所需的时间。

修改后的代码片段：

main(){
  port_init();
  
  RTE_ETH_FOREACH_DEV(portid) {
   struct rte_eth_link link;
   memset(&link, 0, sizeof(link));

   do {
        retval = rte_eth_link_get_nowait(port, &link);
        if (retval < 0) {
            printf("Failed link get (port %u): %s\n",
                port, rte_strerror(-retval));
            return retval;
        } else if (link.link_status)
            break;
        printf("Waiting for Link up on port %"PRIu16"\n", port);
        sleep(1);
    } while (!link.link_status);
  }

  gen_pkt(mbuf);
  rte_eth_tx_burst(mbuf);
}

或

main(){
  port_init();
  
  RTE_ETH_FOREACH_DEV(portid) {
   struct rte_eth_link link;
   memset(&link, 0, sizeof(link));
   ret = rte_eth_link_get(portid, &link);
   if (ret < 0) {
            printf("Port %u link get failed: err=%d\n", portid, ret);
            continue;
    }

  gen_pkt(mbuf);
  rte_eth_tx_burst(mbuf);
}

Answer 2

在物理链路上升之前不能发送数据包也就不足为奇了。这需要一些时间，可以使用rte_eth_link_get() API接口自动等待。