kldunload and device detach panics

[Duncan Barclay]

Hi

Whilst starting work on porting the Linux Broadcom 440x 10/100 driver I've
come across a bit of
a problem with kldunload and its interaction with device xxx_detach.

I'm cribbing code from if_bge.c and if_rl.c (picked at random). if_rl.c uses
a mutex for locking the softc. If the rl_detach routine is called twice (for
example failing attach and then kldunload) the mutex gets destroyed twice -
leading to a panic with witness.

Is there a safe sequence of mutex checking etc. to avoid killing the machine
in this way? At present I'm assuming that if the mutex is not initialised
then _detach doesn't need to clean as nothing could have been set up, or it
has already been cleaned.

Duncan