Re: [PATCH -v3 5/9] workqueue: Tag bound workers with KTHREAD_IS_PER_CPU

[Valentin Schneider]

On 21/01/21 11:17, Peter Zijlstra wrote:
> @@ -4972,9 +4977,11 @@ static void rebind_workers(struct worker
>  	 * of all workers first and then clear UNBOUND.  As we're called
>  	 * from CPU_ONLINE, the following shouldn't fail.
>  	 */
> -	for_each_pool_worker(worker, pool)
> +	for_each_pool_worker(worker, pool) {
> +		kthread_set_per_cpu(worker->task, pool->cpu);
>  		WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task,
>  						  pool->attrs->cpumask) < 0);

At the end of this series, is_cpu_allowed() allows kthreads with
KTHREAD_IS_PER_CPU on any online CPU, even if it isn't the designated
kthread->cpu.

I thought there might be a race here, given this gives us a window where a
pcpu kworker has the flag but is still affined to cpus_possible_mask. Now,
given cpus_write_lock(), we can't have a CPU going up while another goes
down. So I think it's actually fine, and I've been chasing ghosts yet again.

> +	}
>  
>  	raw_spin_lock_irq(&pool->lock);
>