Re: [PATCH] NMI request/release, version 4

[Dipankar Sarma <dipankar@gamebox.net>]

Ok, some more comments -

On Wed, Oct 23, 2002 at 03:14:52PM -0500, Corey Minyard wrote:
> +void release_nmi(struct nmi_handler *handler)
> +{
> +	wait_queue_t  q_ent;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&nmi_handler_lock, flags);
> +	list_del_rcu(&(handler->link));
> +
> +	/* Wait for handler to finish being freed.  This can't be
> +           interrupted, we must wait until it finished. */
> +	init_waitqueue_head(&(handler->wait));
> +	init_waitqueue_entry(&q_ent, current);
> +	add_wait_queue(&(handler->wait), &q_ent);
> +	call_rcu(&(handler->rcu), free_nmi_handler, handler);
> +	for (;;) {
> +		set_current_state(TASK_UNINTERRUPTIBLE);
> +		if (list_empty(&(handler->link)))
> +			break;
> +		spin_unlock_irqrestore(&nmi_handler_lock, flags);
> +		schedule();
> +		spin_lock_irqsave(&nmi_handler_lock, flags);
> +	}
> +	remove_wait_queue(&(handler->wait), &q_ent);
> +	spin_unlock_irqrestore(&nmi_handler_lock, flags);
> +}

Can release_nmi() be done from irq context ? If not, I don't see
why spin_lock_irqsave() is required here. If it can be called
from irq context, then I can't see how you can schedule()
(or wait_for_completion() for that matter :)).

Thanks
Dipankar