2013/4/29 Lucas Stach [off-list ref]:

What is the exact situation which caused a
deadlock here? I tried to see what's happening from your changelogs, but
I'm not sure I can follow here. Wasn't the problem more of the used
spinlocks not being IRQ-save and thus potentially creating a deadlock?

You can view below email thread.
http://www.spinics.net/lists/netdev/msg225240.html

The main problem network API use spin_lock_bh only now. So it does not
prefer use irq to handle skb process.