On Sat, Jan 31, 2026 at 12:00 PM Paul Moses [off-list ref] wrote:

Ok, just to confirm the intended behavior changes compared to what is currently in the tree:

  create missing entry list      FAIL (got -22, expected 0)
  create empty entry list        FAIL (got -22, expected 0)
  replace append entries         REPLACE append failed: expected 2 entries, got 1
                                  FAIL (got -22, expected 0)

- CREATE with missing or empty entry list now returns -EINVAL
  Previously, CREATE could appear to succeed if cycle_time was
  provided even with no entries, but it still left an
  empty schedule and later called list_first_entry() at
  net/sched/act_gate.c:552, which is unsafe. Returning -EINVAL here is the
  correct behavior fix.

- REPLACE now replaces the schedule, it does not append
  The old append behavior was an accident caused by reusing the same list and
  never clearing it. With the RCU snapshot change, a fresh schedule is built
  and swapped atomically, so providing a new entry list on REPLACE replaces
  the old one and avoids stale data.

- REPLACE with an empty entry list keeps the old schedule

At first glance it looks ok.
However I can only be sure once you send it and I run some
tests. Just in case I am missing something.

cheers,
Victor