I think there is some weird racy stuff going on with rf-kill. In the i3945 init code you can see that it disables the PCI device right before it registers the rf-kill handler. This doesn't prevent the handler from being called which looks like it might access the device, even though its disabled. I imagine the side effects of this are hardware dependent.
Incidentally, the init code in Hardy is completely different.
I think there is some weird racy stuff going on with rf-kill. In the i3945 init code you can see that it disables the PCI device right before it registers the rf-kill handler. This doesn't prevent the handler from being called which looks like it might access the device, even though its disabled. I imagine the side effects of this are hardware dependent.
Incidentally, the init code in Hardy is completely different.