Re: [PATCH 01/14] kthread: Fix invalid wakeup in kthreadd

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Thu, Aug 29, 2013 at 09:57:36PM +0800, Libin wrote:
> If kthreadd is preempted at(or before) location a, and the other thread,
> such as calling kthread_create_on_node(), adds a list item to
> the kthread_create_list followed with wake_up_process(kthread). After that
> when kthreadd is re-scheduled, calling set_current_state to set itself as
> state TASK_INTERRUPTIBLE, if it is preempted again after that and before
> __set_current_state(TASK_RUNNING), it triggers the invalid wakeup problem.
> ------------------------
> kthreadd()
> ------------------------
> ...
> for (;;) {
> 	//location a
> 	set_current_state(TASK_INTERRUPTIBLE);
> 	if (list_empty(&kthread_create_list)) {
> 		//location b
> 		schedule();
> 		//location c
> 	}
> 	__set_current_state(TASK_RUNNING);
> 	//location d
> ...
> ------------------------
> kthread_create_on_node()
> ------------------------
> ...
> spin_lock(&kthread_create_lock);
> list_add_tail(&create.list, &kthread_create_list);
> spin_unlock(&kthread_create_lock);
> ...
> wake_up_process(kthreadd_task);
> ...
> 
> To solve this problem, using preempt_disable() to bound the operaion that
> setting the task state and the conditions(set by the wake thread) validation.
> ------------------------
> kthreadd()
> ------------------------
> ...
> for (;;) {
> 	preempt_disable();
> 	set_current_state(TASK_INTERRUPTIBLE);
> 	if (list_empty(&kthread_create_list)) {
> 		preempt_enable();
> 		schedule();
> 		preempt_disable();
> 	}
> ...
> 
> Signed-off-by: Libin <huawei.libin@huawei.com>
> ---
>  kernel/kthread.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/kthread.c b/kernel/kthread.c
> index 760e86d..25c3fed 100644
> --- a/kernel/kthread.c
> +++ b/kernel/kthread.c
> @@ -456,10 +456,15 @@ int kthreadd(void *unused)
>  	current->flags |= PF_NOFREEZE;
> 
>  	for (;;) {
> +		preempt_disable();
>  		set_current_state(TASK_INTERRUPTIBLE);
> -		if (list_empty(&kthread_create_list))

And this one already has the list_empty() check after the call to
set_current_state(), so no change should be needed here.

On the other hand, if your testing shows that you are losing wakeups
with this exact code, please let us know!

							Thanx, Paul

> +		if (list_empty(&kthread_create_list)) {
> +			preempt_enable();
>  			schedule();
> +			preempt_disable();
> +		}
>  		__set_current_state(TASK_RUNNING);
> +		preempt_enable();
> 
>  		spin_lock(&kthread_create_lock);
>  		while (!list_empty(&kthread_create_list)) {
> -- 
> 1.8.2.1
> 
>