Re: [PATCH net-next v2 0/3] Replace xt_recseq with u64_stats.

[Pablo Neira Ayuso <pablo@netfilter.org>]

Hi Sebastian,

On Thu, Mar 20, 2025 at 01:41:26PM +0100, Pablo Neira Ayuso wrote:
> On Thu, Mar 13, 2025 at 09:34:40AM +0100, Sebastian Andrzej Siewior wrote:
> > On 2025-03-13 00:16:03 [+0100], Pablo Neira Ayuso wrote:
> > > Kconfig !PREEMPT_RT for this is not an option, right?
> > 
> > That bad? I though it would make you happy ;)
> > Making it !PREEMPT_RT would essentially disable the whole nf-legacy
> > interface. Given that it is intended to get rid of it eventually it
> > might be an option. I mean there is nothing you can do with
> > iptables-legacy that you can't do with iptables-nft? 
> > I mean if this is not going to happen because of $reasons then that
> > would be the next best thing.
> 
> We could give a try to this series and see.

I have been discussing this with Florian, our proposal:

1. Make ipatbles legacy depend on !PREEMPT_RT which effectively
   disabled iptables classic for RT.

This should be ok, iptables-nft should work for RT.

2. make iptables-legacy user-selectable.

these two are relatively simple.

If this does not make you happy, it should be possible to take your
patches plus hide synchronize_rcu() latency behind deferred free
(call_rcu+workqueue).

As this looks now, I am afraid chances are high that this series will
require a follow up.

Thanks.