From:   Jakub Kicinski <kuba@kernel.org>
Date:   Thu, 1 Sep 2022 14:49:36 -0700

On Thu, 1 Sep 2022 14:25:20 -0700 Kuniyuki Iwashima wrote:

quoted

quoted

I looks to me that the above chunks are functionally a no-op and I
think that omitting the 2 drivers from the v2:

https://lore.kernel.org/netdev/20220829161920.99409-4-kuniyu@amazon.com/ (local)

should break mlx5/nfp inside a netns. I don't understand why including
the above and skipping the latters?!? I guess is a question mostly for
Eric :)  

My best guess is that it's ok unless it does not touch TCP stack deeply
and if it does, the driver developer must catch up with the core changes
not to burden maintainers...?

If so, I understand that take.  OTOH, I also don't want to break anything
when we know the change would do.

So, I'm fine to either stay as is or add the change in v4 again.

FWIW I share Paolo's concern. If we don't want the drivers to be
twiddling with the hash tables we should factor out that code to
a common helper in net/tls/

That makes sense.
For the moment, I'll add the changes back in v4.