Re: [PATCH net-next] net/sched: introduce TCQ_F_IN_HW to track successful qdisc offload

[David Yang]

On Thu, May 14, 2026 at 8:42 AM Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
>
> I read this 3 times and I'm still not 100% sure what the main problem
> you're trying to fix is. You don't know if the DELETE is for the
> offloaded qdisc? Can't you just store the handle of what's offloaded
> in the driver?

The point is exactly NOT to store the handle in the driver. Only mlxsw
and nfp handle TCQ_F_OFFLOADED correctly today. Fixing the others
would mean repeating the same store-and-lookup logic in every driver,
even ones that don't care about the cookie at all.

But I'm open to other approaches, for example driver priv in struct
Qdisc, or removing the seldom-working TCQ_F_OFFLOADED from
qdisc_offload_dump_helper().

> The qdisc offload is a bit of a mess without a dedicated maintainer.
> Sweeping changes like this are a bit burdensome for the core
> maintainers..
> --
> pw-bot: cr