Re: NMI handling rework

[Zwane Mwaikambo]

On Thu, 14 Nov 2002, Corey Minyard wrote:

> RCU does.  Basically, the code pulls it from the list atomically wrt to 
> the NMI handler, and uses RCU to schedule the actual free of the data to 
> be done after all CPUs have gone to idle or returned from interrupts. 
>  It's subtle, you have to think about it a little.  But it does work.

Still not convinced, i still want to know which interrupt rate and how 
many processors. Are the following functions really protected from say 
NMIs at 300,000/s? request_nmi actually looks dodgiest.

static void free_nmi_handler(void *arg)
{
	struct nmi_handler *handler = arg;

	INIT_LIST_HEAD(&(handler->link));
	complete(&(handler->complete));
}

void release_nmi(struct nmi_handler *handler)
{
	spin_lock(&nmi_handler_lock);
	list_del_rcu(&(handler->link));
	init_completion(&(handler->complete));
	call_rcu(&(handler->rcu), free_nmi_handler, handler);
	spin_unlock(&nmi_handler_lock);
}

-- 
function.linuxpower.ca