Re: [PATCH net-next] net/sched: don't use dynamic lockdep keys with clsact/ingress/noqueue

[Davide Caratti <dcaratti@redhat.com>]

On Fri, Jan 30, 2026 at 07:29:21PM -0800, Jakub Kicinski wrote:
> On Thu, 29 Jan 2026 16:24:35 +0100 Davide Caratti wrote:
> > +static inline void qdisc_lock_init(struct Qdisc *sch, const struct Qdisc_ops *ops)
> 
> wrap at 80 please

hi Jakub, thanks for reading. Sure, will fix in v2.
 
> > +{
> > +	bool skip_dynamic_key = (ops->static_flags & TCQ_F_INGRESS) ||
> > +				(ops == &noqueue_qdisc_ops);
> > +
> > +	if (!skip_dynamic_key)
> > +		lockdep_register_key(&sch->root_lock_key);
> > +
> > +	spin_lock_init(&sch->q.lock);
> > +	if (!skip_dynamic_key)
> > +		lockdep_set_class(&sch->q.lock, &sch->root_lock_key);
> 
> is there a reason for the order of things here? 

I was self-convinced that CONFIG_DEBUG_SPINLOCK and similar would
complain with the ordering below, but I'm now seeing empirically
that it doesn't.

> The code flow would be far more natural with:
> 
> {
> 	spin_lock_init(&sch->q.lock);
> 
> 	/* Skip dynamic keys for qdiscs which can't nest */
> 	if (ops->static_flags & TCQ_F_INGRESS ||
>             ops == &noqueue_qdisc_ops)
> 		return;
> 
> 	lockdep_register_key(&sch->root_lock_key);
> 	lockdep_set_class(&sch->q.lock, &sch->root_lock_key);
> }

will do some additional tests and re-send this way if it doesn't
complain.

-- 
davide