I didn't read the patch yet (will do), just a minor nit right now,

On 03/28, Roland McGrath wrote:
>
> @@ -1528,19 +1534,11 @@ static int do_wait_thread(struct task_struct *tsk, int *retval,
>  			return 1;
>  	}
> 
> -	/*
> -	 * If we never saw an eligile child, check for children stolen by
> -	 * ptrace.  We don't leave -ECHILD in *@retval if there are any,
> -	 * because we will eventually be allowed to wait for them again.
> -	 */
> -	if (*retval)
> -		list_for_each_entry(p, &tsk->ptrace_children, ptrace_list) {
> -			int ret = eligible_child(type, pid, options, p);
> -			if (ret) {
> -				*retval = unlikely(ret < 0) ? ret : 0;
> -				break;
> -			}
> -		}
> +	list_for_each_entry(p, &tsk->ptrace_attach, ptrace_list) {
> +		if (wait_consider_task(tsk, p, retval, type, pid, options,
> +				       infop, stat_addr, ru))
> +			return 1;
> +	}

Afaics, this adds a minor pessimization. We shouldn't scan ->ptrace_attach
list if it was already found that another thread has a "hidden" task.

Oleg.