Crash at termination with offloading

[PlagueCZ]

Describe the bug
Doing a minimal startup and then terminating dpservice-bin via Ctrl+C leads to a SIGSEGV on termination (in DPDK cleanup).

To Reproduce
Running on a PC with Mellanox Connectx-6 with two VMs running (using a vfio NIC).
dpservice-bin -l0,1 -- --no-stats
dpservice-cli init
dpservice-cli add interface --id test10 --device 0000:03:00.0_representor_vf0 --vni 123 --ipv4 192.168.123.10 --ipv6 fe80::10
Ctrl+C on dpservice

Stacktrace

Thread 1 "dpservice-bin" received signal SIGSEGV, Segmentation fault.
0x00007ffff5e8bab1 in flow_dv_sample_clone_free_cb ()
   from /usr/local/lib/x86_64-linux-gnu/dpdk/pmds-23.0/librte_net_mlx5.so.23.0

(this is part of rte_eal_cleanup())

Additional information
This does not happen once --no-offload is added (given the stacktrace that's expected).

[PlagueCZ]

@byteocean not sure you can test this. I will hopefully get a separated lab setup soon to test on another machine.

[byteocean]

not encounter this so far. has something to do with migrating to DPDK 22? which was not tested on my side yet.

[PlagueCZ]

@byteocean thanks for pointing me the right direction!
The change that causes this is 61cf7a0 (Which removed the DPDK warnings at the end about stopping ports).
So I guess there needs to be something more done before stopping ports, something about the flows I guess?

[byteocean]

maybe the handler thing that wraps the age action. you could try to remove this indirect action part to see if the error still exits. btw, I rebased to main and upgrade dpdk to 22.11.3, and also got this error when I terminated it, after running some ping tests.

[byteocean]

@PlagueCZ fyi, under 22.03, such error (seg fault) does not appear.

[PlagueCZ]

@byteocean removing call to dp_install_default_rule_in_monitoring_group() fixes the error.
I have tested 21.11 right before updating to 22.11 and the error does occur there.