DPVS segfaults at rte_eth_rx_burst with DPDK 18.11, but works fine with 17.11

[anshulthakur]

Hi,

While running DPVS with DPDK 18.11 stable (Ubuntu 18.04 machine), the following trace is seen when the lcores start their poll loop:

0x0000555555a535a9 in rte_eth_rx_burst (port_id=0, queue_id=0, rx_pkts=0x55555645eb10 <lcore_conf+144>, nb_pkts=32)
    at /home/anshul/loadbalancer/dpdk-stable-18.11.10/build/include/rte_ethdev.h:3879
3879	/home/anshul/loadbalancer/dpdk-stable-18.11.10/build/include/rte_ethdev.h: No such file or directory.
(gdb) bt
#0  0x0000555555a535a9 in rte_eth_rx_burst (port_id=0, queue_id=0, rx_pkts=0x55555645eb10 <lcore_conf+144>, nb_pkts=32)
    at /home/anshul/loadbalancer/dpdk-stable-18.11.10/build/include/rte_ethdev.h:3879
#1  0x0000555555a611df in netif_rx_burst (pid=0, qconf=0x55555645eb00 <lcore_conf+128>)
    at /home/anshul/loadbalancer/dpvs/src/netif.c:1599
#2  0x0000555555a65148 in lcore_job_recv_fwd (arg=0x0) at /home/anshul/loadbalancer/dpvs/src/netif.c:2408
#3  0x0000555555a7c0bb in do_lcore_job (job=0x555556852940 <netif_jobs>) at /home/anshul/loadbalancer/dpvs/src/scheduler.c:150
#4  0x0000555555a7c22c in dpvs_job_loop (arg=0x0) at /home/anshul/loadbalancer/dpvs/src/scheduler.c:198
#5  0x00005555557301b1 in eal_thread_loop ()
#6  0x00007ffff6f466db in start_thread (arg=0x7ffff5940700) at pthread_create.c:463
#7  0x00007ffff6a6471f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95

In the logs before, the following is also seen:

NETIF: dpdk0:dst_port_mask=0
NETIF: Ethdev port_id=0 invalid rss_hf: 0x3afbc, valid value: 0x0
NETIF: Ethdev port_id=0 invalid rx_offload: 0x3, valid value: 0xa1d
NETIF: Ethdev port_id=0 invalid tx_offload: 0x10002, valid value: 0x802d
virtio_dev_configure(): Unsupported Rx multi queue mode 1
Port0 dev_configure = -22
NETIF: netif_port_start: fail to config dpdk0
DPVS: Start dpdk0 failed, skipping ...
DPVS: 

The same codebase works fine with DPDK 17.11.

Please help!

[anshulthakur]

Update: Setting the Receive mq_mode to ETH_MQ_RX_NONE removes the Segfault.

The setup was for a single arm operation with just one dpdk port (and a single rx-tx queue) using virtio drivers. Today, I'll check if further problems come. However, I am not sure if this is the right way to go about this (if RSS flags are disabled, then how would distribution across queues happen?)