On Fri, 18 Aug 2006 21:14:09 -0700
Daniel Phillips [off-list ref] wrote:

So rather than just the word deadlock, let us add "or atomic 0 order
alloc failure during TCP receive" to the challenge.  Fair?

If it's significantly performance-affecting in any way which is at all likely to 
affect anyone, sure.

You can get those warnings now with regular networking using e1000, due to
a combination of excessive default rx ringsize and incorrect VM tuning.