SIGSEGV in netsvc/hn_rxtx.c:1075 (DPDK 23.07)

SIGSEGV in netsvc/hn_rxtx.c:1075 (DPDK 23.07)

[José Manuel Díaz Rodríguez]

[-- Attachment #1: Type: text/plain, Size: 3541 bytes --]

Hi, everyone:

We're trying to make VPP work over Hyper-V Netvsc (DPDK 23.07).
We don't have a DPDK implementing HW NIC, but Hyper-V's paravirtualization of relatively cheap 1G NICs (tested with an HP server and an HP laptop, both fail), so we depend on Hyper-V/uio_hv_generic/netvsc for providing us with a fully virtualized DPDK implementation on top of which to place VPP.
The VPP process aborts after receiving some packets if we configure any worker thread (in contrast, the configuration with all VPP, control and user plane, in a single thread works).
This is the backtrace (deterministic, the same in both machines with different HW):

Thread 3 "vpp_wk_0" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffa7d2a640 (LWP 1296)]
0x00007fffb0190710 in rte_spinlock_trylock (sl=0x20) at ../src-dpdk/lib/eal/x86/include/rte_spinlock.h:63
63              asm volatile (
(gdb) bt
#0  0x00007fffb0190710 in rte_spinlock_trylock (sl=0x20) at ../src-dpdk/lib/eal/x86/include/rte_spinlock.h:63
#1  0x00007fffb01bd531 in hn_process_events (hv=0xac03a7bc0, queue_id=1, tx_limit=0) at ../src-dpdk/drivers/net/netvsc/hn_rxtx.c:1075
#2  0x00007fffb01c55d3 in hn_xmit_pkts (ptxq=0xac08a5040, tx_pkts=0x7fffb7b32e00, nb_pkts=1) at ../src-dpdk/drivers/net/netvsc/hn_rxtx.c:1497
#3  0x00007fffb0b5bbf2 in rte_eth_tx_burst (port_id=0, queue_id=1, tx_pkts=0x7fffb7b32e00, nb_pkts=1)
    at /home/ubuntu/vpp/build-root/install-vpp_debug-native/external/include/rte_ethdev.h:6320
#4  0x00007fffb0b57799 in tx_burst_vector_internal (vm=0x7fffb7af8b80, xd=0x7fffb7b37c40, mb=0x7fffb7b32e00, n_left=1, queue_id=1, is_shared=0 '\000')
    at /home/ubuntu/vpp/src/plugins/dpdk/device/device.c:173
#5  0x00007fffb0b56ac2 in dpdk_device_class_tx_fn_hsw (vm=0x7fffb7af8b80, node=0x7fffb7b54540, f=0x7fffb7bcfdc0)
    at /home/ubuntu/vpp/src/plugins/dpdk/device/device.c:423
#6  0x00007ffff7062a32 in dispatch_node (vm=0x7fffb7af8b80, node=0x7fffb7b54540, type=VLIB_NODE_TYPE_INTERNAL, dispatch_state=VLIB_NODE_STATE_POLLING,
    frame=0x7fffb7bcfdc0, last_time_stamp=29933360537300) at /home/ubuntu/vpp/src/vlib/main.c:960
#7  0x00007ffff7063452 in dispatch_pending_node (vm=0x7fffb7af8b80, pending_frame_index=9, last_time_stamp=29933360537300)
    at /home/ubuntu/vpp/src/vlib/main.c:1119
#8  0x00007ffff705e7d2 in vlib_main_or_worker_loop (vm=0x7fffb7af8b80, is_main=0) at /home/ubuntu/vpp/src/vlib/main.c:1608
#9  0x00007ffff705ded7 in vlib_worker_loop (vm=0x7fffb7af8b80) at /home/ubuntu/vpp/src/vlib/main.c:1741
#10 0x00007ffff709b050 in vlib_worker_thread_fn (arg=0x7fffb865d600) at /home/ubuntu/vpp/src/vlib/threads.c:1604
#11 0x00007ffff7096236 in vlib_worker_thread_bootstrap_fn (arg=0x7fffb865d600) at /home/ubuntu/vpp/src/vlib/threads.c:418
#12 0x00007ffff6c86ac3 in start_thread (arg=<optimized out>) at ./nptl/pthread_create.c:442
#13 0x00007ffff6d18850 in clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:81
(gdb)

Before posting on VPP lists, we want to confirm there aren't any known issues with netvsc that may be causing this, or any advanced configuration or parameter we can try.
Our current configuration is pretty simple.
The problem can be easily reproduced on an Ubuntu 22.04 VM over Hyper-V (Hyper-V Server 2019 or Windows 11), with VPP 23.10/DPDK 23.07, two interfaces, 1 worker thread, DPDK and ping plugins enabled, and sending a bunch of ICMP Echo Requests on a loop from the host machine to one of the VPP interfaces.

Thank you very much.

Regards,

José Manuel

[-- Attachment #2: Type: text/html, Size: 6910 bytes --]