Re: [RFC PATCH] rhashtable: Bounce deferred worker kick through irq_work when insecure_elasticity is set

[Tejun Heo]

On Mon, Apr 20, 2026 at 04:44:26PM +0800, Herbert Xu wrote:
> On Sun, Apr 19, 2026 at 08:19:33AM -1000, Tejun Heo wrote:
> >
> > +/*
> > + * Kick the deferred rehash worker. With insecure_elasticity the caller may
> > + * hold a raw spinlock. schedule_work() under a raw spinlock records
> > + * caller_lock -> pool->lock -> pi_lock -> rq->__lock. If any of these
> > + * locks is acquired in the reverse direction elsewhere, the cycle closes.
> > + * Bounce through irq_work so schedule_work() runs from hard IRQ context
> > + * with the caller's lock no longer held.
> > + */
> > +static void rhashtable_kick_deferred_worker(struct rhashtable *ht)
> > +{
> > +	if (ht->p.insecure_elasticity)
> > +		irq_work_queue(&ht->run_irq_work);
> > +	else
> > +		schedule_work(&ht->run_work);
> > +}
> 
> Can we just do irq_work_queue unconditionally?

Oh yeah, that works too. I'll respin the patch.

Thanks.

-- 
tejun