Re: [PATCH net v6 1/1] net/sched: act_gate: snapshot parameters with RCU on replace

[Victor Nogueira]

On Sun, Feb 15, 2026 at 8:59 PM Paul Moses <p@xxxxxxx> wrote:
> 1. hrtimer_get_expires() just returns the stored node.expires and
>    hrtimer_cancel() doesn’t clear it, so expires==0 is not a reliable
>    inactivity test. Logic was that although I detected no observable
>    behavior difference, relying on stale expires could theoretically
>    cause infrequent subtle intermittent misses of intended behavior.
>    It's maybe more appropriate to leave it alone for stable or at
>    least not in this patch/series?

Yes, even if it is an issue, we should leave it for another patchset.

> [...]
> 3. It's the same pattern used in sch_taprio and it's documented in
>    Documentation/memory-barriers.txt: the compiler may merge/discard/
>    invent/reorder plain accesses and READ_ONCE()/WRITE_ONCE() exist to
>    make intentional lockless shared variable accesses well defined.
>    Since tk_offset is read with READ_ONCE() outside tcf_lock, the writer
>    uses WRITE_ONCE() to pair with that lockless read.

Ok.

cheers,
Victor