It's not a regression. It turns out I was wrong in my other comment - this
never worked. Userspace was working around the problem by unloading the
driver before hibernation and loading it back in on resume (I switched to
module free kernel a few days ago). Building a kernel that prevents module
unloading shows similar results all the way back to 2.6.25 (although the
error switches to "SIOCSIFFLAGS: No buffer space available".
--
Sitsofe | http://sucs.org/~sits/
--