Re: [PATCH] kthread: Enhance kthread_stop to abort interruptible sleeps

[Oleg Nesterov <oleg@tv-sign.ru>]

On 04/13, Eric W. Biederman wrote:
>
> +static inline int __kthread_should_stop(struct task_struct *tsk)
> +{
> +	return test_tsk_thread_flag(tsk, TIF_KTHREAD_STOP);
> +}

Am I blind? Where does copy_process/dup_task_struct clears unwanted
flags in thread_info->flags ?

> +int kthread_stop(struct task_struct *tsk)
>  {
>  	int ret;
>  
> -	mutex_lock(&kthread_stop_lock);
> -
> -	/* It could exit after stop_info.k set, but before wake_up_process. */
> -	get_task_struct(k);
> +	/* Ensure the task struct persists until I read the exit code. */
> +	get_task_struct(tsk);
>  
> -	/* Must init completion *before* thread sees kthread_stop_info.k */
> -	init_completion(&kthread_stop_info.done);
> -	smp_wmb();
> +	set_tsk_thread_flag(tsk, TIF_KTHREAD_STOP);
> +	spin_lock_irq(&tsk->sighand->siglock);
> +	signal_wake_up(tsk, 1);
> +	spin_unlock_irq(&tsk->sighand->siglock);

Off-topic again, the comment above signal_wake_up() is very confusing...
I think the only reason for ->siglock is to prevent the case when
TIF_SIGPENDING may be cleared by recalc_sigpending(). Or something else?

This changes the current API, kthread_stop() doesn't wake up UNINTERRUPTIBLE
tasks any longer. I'd say this is good, but may break things ... For example,
kthred_stop(kthread_create(...)) can't work now.

> -	/* Now set kthread_should_stop() to true, and wake it up. */
> -	kthread_stop_info.k = k;
> -	wake_up_process(k);
> -	put_task_struct(k);
> +	wait_for_completion(tsk->vfork_done);
> +	ret = tsk->exit_code;

This is really good. Now the kernel thread can exit() at any point without
fear to break kthread_stop().

> @@ -218,7 +219,7 @@ static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
>  fastcall void recalc_sigpending_tsk(struct task_struct *t)
>  {
>  	if (t->signal->group_stop_count > 0 ||
> -	    (freezing(t)) ||
> +	    (freezing(t)) || __kthread_should_stop(t) ||
>  	    PENDING(&t->pending, &t->blocked) ||
>  	    PENDING(&t->signal->shared_pending, &t->blocked))
>  		set_tsk_thread_flag(t, TIF_SIGPENDING);

Aha, now I understand your point about interruptible sleeps (the previous
message). What is the reason for this change?

Oleg.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/