[PATCH] Race conditions in QUEUE handling on SMP builds

[Alexander Demenshin]

On Sat, Jul 22, 2000 at 10:11:14AM +1000, Rob Thomas wrote:

> I'm just a lurker here, so I won't get too carried away, but, to me, if
> there -is- a race somewhere, that just masks it. Anyway. I'll go back to
> lurking, I apologise for the un-neccesary frothing.

  Sorry, but you are wrong here. It does not mask it, it _avoids_ race,
  or, in other words, _eliminates_ race. If you know another way to
  eliminate race - let me know... May be it is possible to fix it in
  different way, I don't know, but solution I found is simple and without
  any [visible] performance loss (on the other hand, we can fix the problem
  by introducing additional locking - this would be a problem).
  
  If you will look at code, immediately before my skb_get(), you will
  find out similar call but for netdevice - it is also used to avoid
  race conditions. So why not to do so with skb as well (especially
  when it solves the problem)? Or - do you prefer 1000 lines patch for
  bug fixes? :)
  
/Al