Re: [PATCH 0/2] printk vs rq->lock and xtime lock

[Linus Torvalds <torvalds@linux-foundation.org>]

On Fri, 8 Aug 2008, Peter Zijlstra wrote:
> +
> +	local_irq_save(flags);
> +	kws = &__get_cpu_var(kws);
> +	if (!kws->pending) {
> +		call_rcu(&kws->head, __wake_up_klogd);
> +		kws->pending = 1;

In case we have a recursive oops in the RCU paths, I'd think we'd be 
*much* better off with

	if (!kws->pending) {  
		kws->pending = 1;
		call_rcu(&kws->head, __wake_up_klogd);
	}

in order to not re-enter the RCU paths again and causing endless oopses 
along with list corruption.

Hmm?

Oh, and I don't see the point of the irqsave/restore in __wake_up_klogd(). 
Or the BUG_ON(), for that matter (even if it ever were to trigger, we'd be 
better off _not_ triggering it - I agree with it in a "verifying the new 
implementation works at all" sense, but not in a "this is actually for 
merging" kind of patch).

			Linus