Hi Pierre,
On Wed, Dec 20, 2023 at 09:32:52AM +0100, Pierre Bourdon wrote:
[...]
> Nov 26 00:40:57 aether kernel: Call trace:
> Nov 26 00:40:57 aether kernel: rhashtable_walk_next+0x7c/0xa8
> Nov 26 00:40:57 aether kernel: process_one_work+0x1fc/0x460
> Nov 26 00:40:57 aether kernel: worker_thread+0x170/0x4a8
> Nov 26 00:40:57 aether kernel: kthread+0xec/0xf8
> Nov 26 00:40:57 aether kernel: ret_from_fork+0x10/0x20
>
> By playing a bit with the nftables configuration I've isolated it
> further: it's only happening with IPv6 flow offloading. "ip protocol {
> tcp, udp } flow offload @f;" doesn't cause hangs, but "ip6 nexthdr {
> tcp, udp } flow offload @f;" consistently does.
>
> The device this is happening on is NXP LX2160A based (ARMv8, 16
> cores). This has been happening since at least kernel 6.1 and I've
> tested all the way to 6.6.3.
>
> Has anyone ever reported something similar?
I got a similar report from another user, which also mentioned about
IPv6, I did not manage to reproduce this issue yet.
> What would be good next steps to help track this down further? Any
> useful .config options? I've never debugged hangs like this in the
> Linux kernel, unfortunately. My router doesn't have a JTAG for me to
> plug in a hardware debugger and get stack traces while everything is
> frozen. The fact that it takes 1-4 days for the issue to reproduce
> also doesn't help...
If this is a generic issue in the flowtable with IPv6, probably try to
reproduce this issue in an easier to debug environment, such as qemu
VM with a kernel running CONFIG_KASAN might help.
