Re: [RFC PATCH net-next] netpoll: hold RCU while walking napi_list

[Runyu Xiao]

Hi Jakub, Breno,

On Mon, 29 Jun 2026 15:58:33 -0700 Jakub Kicinski wrote:
> Yes, like Breno says, we need an in-kernel path that can trigger the
> issue. Out-of-tree reproducers can be useful to validate the fix, but
> they are not useful to prove that a problem actually exists.
>
> We try to avoid defensive programming in the kernel.

Understood, thanks for the clarification.

The out-of-tree reproducer was only meant for our internal
CONFIG_PROVE_RCU_LIST triage, not as proof that the upstream netpoll
path is buggy.

After your feedback, I retested this on latest net-next with a real
in-kernel netconsole path. In that setup I could hit
netpoll_send_udp(), netpoll_send_skb(), and netpoll_poll_dev(), but I
could not reproduce the PROVE_RCU_LIST warning on the real path.

Given that result, I am not going to continue this RFC in its current
form.

Thanks,
Runyu