Re: [RFC] kthread_create_on_node is failing to honor the node choice

[Lai Jiangshan <laijs@cn.fujitsu.com>]

On 06/26/2015 05:44 AM, Nathan Zimmer wrote:
> In kthread_create_on_node we set_cpus_allowed to cpu_all_mask
> regardless of what the node is requested.
> This seems incorrect.
> 
> Signed-off-by: Nathan Zimmer <nzimmer@sgi.com>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Nishanth Aravamudan <nacc@linux.vnet.ibm.com>
> Cc: Tejun Heo <tj@kernel.org>
> Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
> Cc: Mel Gorman <mgorman@suse.de>
> Cc: linux-kernel@vger.kernel.org
> 
> ---
>  kernel/kthread.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/kthread.c b/kernel/kthread.c
> index 10e489c..d885d66 100644
> --- a/kernel/kthread.c
> +++ b/kernel/kthread.c
> @@ -318,7 +318,10 @@ struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
>  		 * The kernel thread should not inherit these properties.
>  		 */
>  		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
> -		set_cpus_allowed_ptr(task, cpu_all_mask);
> +		if (node == -1)
> +			set_cpus_allowed_ptr(task, cpu_all_mask);
> +		else
> +			set_cpus_allowed_ptr(task, cpumask_of_node(node));


cpumask_of_node(node) is bad here. It containers only online cpus.

>  	}
>  	kfree(create);
>  	return task;
>