linux@baker-net.org.uk said:

The first thing to try would be to see if rmmod rt61pci; modprobe rt61pci is
enough to recover it. If it is actually a cardbus card then you could also
try unplugging and replugging it. 

It's a rt2500, so I use that module. Should I use rt61pci?
If memory servers, rmmod/modprobe does not help. It is a PCI card with no loose 
parts on it so unplugging/plugging is not really an option.


linux@baker-net.org.uk said:

Neither of those is an ideal solution though. Do you fancy the task of trying
 to diagnose the underlying fault which is likely to require at the very
least  making sure the driver is built with debugfs support enabled and may
require  adding a few printk statements into the kernel modules to reveal
more details  of the behaviour? 

No problem. I've never used debugfs, but I kind of know my way around the 
kernel. The problem has existed at least since the opensource driver got 
merged into mainline. I'm not sure if the closed source one had the same issue.
So bisecting is not really an option.

So, where do I sprinkle the prink's, and what info is useful?

Thanks!
/Anders