There were certain situations where ipmi_register_smi() would
return a failure, but the interface would still be registered
and would need to be unregistered.  This is obviously a bad
design and resulted in an oops in certain failure cases.

If the interface is started up in ipmi_register_smi(), then
an error occurs, shut down the interface there so the
cleanup can be done properly.

Fix the various smi users, too.
Signed-off-by: Corey Minyard <cminyard@mvista.com>
Reported-by: Justin Ernst <justin.ernst@hpe.com>
Tested-by: Justin Ernst <justin.ernst@hpe.com>
Cc: Andrew Banman <abanman@hpe.com>
Cc: Russ Anderson <russ.anderson@hpe.com>
Cc: <stable@vger.kernel.org> # 4.18.x