Re: [PATCH v2] mm/slab: don't use kfree_rcu sheaves on PREEMPT_RT in kvfree_call_rcu()

[ThangNN99 <ngocthang2710.1999@gmail.com>]

Thanks, both flagged items are pre-existing and out of scope for this
patch:

- The headless kvfree_rcu(ptr) might_sleep()/synchronize_rcu() path is
  never reached by the caller this patch fixes: set_cpus_allowed_force()
  uses the headed two-argument kfree_rcu(ptr, rcu) form specifically to
  avoid it (see its own comment), so it's unrelated to the sheaf-bypass
  this patch adds.

- kfree_call_rcu_nolock() still hitting local_trylock(&s->cpu_sheaves->lock)
  unconditionally is exactly what this patch's commit message already
  calls out as unsafe under a raw spinlock. This patch intentionally
  leaves kfree_call_rcu_nolock() untouched; fixing its 'any context'
  guarantee belongs in a separate patch.

This patch stays scoped to the specific syzbot-reported splat in
kvfree_call_rcu(). Happy to look at the two pre-existing issues
separately if a maintainer wants a follow-up patch for either.

Thanks,
ThangNN99