Re: [PATCH v5 4/4] sched/psi: allow unprivileged polling of N*2s period

[Johannes Weiner <hannes@cmpxchg.org>]

Hi Domenico,

thanks, the destruction path looks right to me now.

On Thu, Mar 30, 2023 at 10:11:00AM +0200, Domenico Cerasuolo wrote:
> @@ -1319,11 +1353,16 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group,
>  		rcu_assign_pointer(group->rtpoll_task, task);
>  	}
>  
> -	list_add(&t->node, &group->rtpoll_triggers);
> -	group->rtpoll_min_period = min(group->rtpoll_min_period,
> -		div_u64(t->win.size, UPDATES_PER_WINDOW));
> -	group->rtpoll_nr_triggers[t->state]++;
> -	group->rtpoll_states |= (1 << t->state);
> +	if (privileged) {
> +		list_add(&t->node, &group->rtpoll_triggers);
> +		group->rtpoll_min_period = min(group->rtpoll_min_period,
> +			div_u64(t->win.size, UPDATES_PER_WINDOW));
> +		group->rtpoll_nr_triggers[t->state]++;
> +		group->rtpoll_states |= (1 << t->state);
> +	} else {
> +		list_add(&t->node, &group->avg_triggers);
> +		group->avg_nr_triggers[t->state]++;
> +	}
>  
>  	mutex_unlock(&group->rtpoll_trigger_lock);

But creation needs the same treatment. group->avg_triggers and
avg_nr_triggers are protected by the avgs_lock, not the
rtpoll_trigger_lock.

I suppose the lock taking and the group->rtpoll_task creation
(anything with rtpoll in the name, really) should be within a big

	if (privileged) {
		rtpoll_trigger_lock()
		kthread_create()
		init-and-link()
		rtpoll_trigger_unlock()
	} else {
		avgs_lock()
		...
	}