Problem:

When resuming after a suspend, the driver disconnects. Later on, (next
usb interrupt?) or suspend/resume, the kernel oopses.

Cause:

hcd.state is never set to 'USB_STATE_READY' when resuming. The following
patch fixes this. Maybe Johannes might want to do it differently, i.e.
by adding hcd.state = USB_STATE_READY to uhci_resume instead of moving
the state change into start_hc.