:Here's my latest patch.  The difference from the last is that I interlocked
:the tsleeps and asserted that the serializer is held in the interrupt
:handler.

    It looks pretty good but I have no way to test it.  Insofar as the
    serializer goes, if you make a mistake you'll probably deadlock a
    thread or three, but it should be readily debuggable.

						-Matt