>Regardless, it would be a Good Idea (tm) to perform some of the FIPS
>tests to ensure the RNG hardware at least looks functional rather than

All FIPS 140 requires is a check that the value returned from the RNG 
isn't the same as the last value.  If it is, you can fetch another value 
and check again since there is a non-zero possibility that the first 
"double" was simply statistics.  

Of course the downside is what you have to do if the test fails ;-)
Which is why if you want to use /dev/random to seed your approved RNG
you should probably do so via an api that will impose the RNG test - but 
not otherwise burden /dev/random.

--sjg