[dpdk-dev] Segfault when eal thread executing mlx5 nic's lsc event

[Weifeng LI <liweifeng96@126.com>]

hi

     I am using the dpdk bond of mlx5. There is a segment error in the 
process of starting the bond port. This is because EAL interrupt thread 
is processing LSC interrupt when slave_configure is executing the 
rte_eth_dev_rss_reta_update. rte_eth_dev_rss_reta_update will also use 
mlx5 flow list.

     I've also found another discussion about this 
issue.https://mails.dpdk.org/archives/dev/2019-March/125929.html 
<https://mails.dpdk.org/archives/dev/2019-March/125929.html>

     Do it need a lock to protect the mlx5 flow list?


int
slave_configure(struct rte_eth_dev *bonded_eth_dev,
         struct rte_eth_dev *slave_eth_dev)

{

     ...

     /* Start device */
     errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
     if (errval != 0) {
         RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
                 slave_eth_dev->data->port_id, errval);
         return -1;
     }

     /* If RSS is enabled for bonding, synchronize RETA */
     if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
         int i;
         struct bond_dev_private *internals;

         internals = bonded_eth_dev->data->dev_private;

         for (i = 0; i < internals->slave_count; i++) {
             if (internals->slaves[i].port_id == 
slave_eth_dev->data->port_id) {
                 errval = rte_eth_dev_rss_reta_update(
                         slave_eth_dev->data->port_id,
                         &internals->reta_conf[0],
                         internals->slaves[i].reta_size);
                 if (errval != 0) {
                     RTE_BOND_LOG(WARNING,
                              "rte_eth_dev_rss_reta_update on slave port 
%d fails (err %d)."
                              " RSS Configuration for bonding may be 
inconsistent.",
                              slave_eth_dev->data->port_id, errval);
                 }
                 break;
             }
         }
     }