Re: more efficient current_is_keventd macro? [was Re: [lhcs-devel] Re: Kthread_create() never returns when called from worker_thread]

[Andrew Morton <akpm@osdl.org>]

Srivatsa Vaddagiri <vatsa@in.ibm.com> wrote:
>
> 	current_is_keventd() macro checks "current" with the per-cpu
> thread of _all_ possible cpus attached to keventd_wq. Can't it just check 
> against the per-cpu thread of _current_ cpu alone (since the per-cpu workqueue
> threads are anyway bound only to their cpus)?  

Seems that way, yes.

> int current_is_keventd(void)
> {
>         struct cpu_workqueue_struct *cwq;
> -       int cpu;
> +       int cpu = smp_processor_id();
> 
>         BUG_ON(!keventd_wq);
> 
> -       for_each_cpu(cpu) {
> -               cwq = keventd_wq->cpu_wq + cpu;
> -               if (current == cwq->thread)
> -                       return 1;
> -       }
> 
> -       return 0;
> 
> +	cwq = keventd_wq->cpu_wq + cpu;
> +	if (current == cwq->thread)
> +		return 1;
> +	else
> +		return 0;
> }

Is racy in the presence of preemption.  Please replace smp_processor_id()
with get_cpu(), stick a put_cpu() at the end, avoid having two function
return points, test it and send me the diff?