Re: linkwatch bustage in git-net

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 9 May 2007 15:45:58 +1000 Herbert Xu <herbert@gondor.apana.org.au> wrote:

> On Wed, May 09, 2007 at 03:31:55PM +1000, Herbert Xu wrote:
> > 
> > Hmm, I don't see it here (probably because we use different NICs).
> > But does this help?
> 
> Thinking about it again I don't think it will help you because if your
> carrier started out as off then it would've been considered an urgent
> event anyway.
> 
> So what NIC are you using? And where abouts in the boot process is it
> hanging? For exmaple, is it hanging when obtaining a DHCP address?
> 
> In any case, this patch can't hurt.  So here's one with a changelog:
> 
> [NET] link_watch: Eliminate potential delay on wrap-around

hm, that fixed it.  Do we know why? ;)



btw, looking at the code:

	clear_bit(LW_RUNNING, &linkwatch_flags);

	spin_lock_irq(&lweventlist_lock);
	next = lweventlist;
	lweventlist = NULL;
	spin_unlock_irq(&lweventlist_lock);

	while (next) {
		struct net_device *dev = next;

		next = dev->link_watch_next;

lweventlist_lock protects lweventlist and every netdev's ->link_watch_next.
 But this code is walking that singly-linked list outside the lock and
after clearing LW_RUNNING.  What stops this singly-linked list from getting
altered by another thread of control while this code is traversing it?