This patch set switches from SRCU to using normal locks.  This required
pulling most of the IPMI message handler up to run in kernel threads,
which was a good thing, anyway.  It also handles some fallout from
that change.

I didn't have a way to thoroughly test the IPMI driver, most of the
work for this was creating a test framework, which required work in
QEMU, the OpenIPMI simulator, and a few other things.  But that's
working now, and it shook out surprisingly few bugs, one in the new
changes and one driver unload issue.

Anyway, here's the first candidate for this.