>    Andrew> that's not entirely accurate...you actually need to worry
>    Andrew> about anything with a major number of 0.  this would
>    Andrew> include the block interfaces for the wd? devices.
>  I'd like to suggest that 32 bit major number 0, indicates a 16
>device.
>  BSDi has solved this, btw.

Details please?  (Or was the first sentence the only needed detail?)

[Actually, as long as whatever ends up on major 0 doesn't add any extra
wierd semantics to the difference between 16 bit numbers and 32 bit numbers,
like I was suggesting for the partition expansion, it probably doesn't matter
that one can't tell the difference between a 16 bit (0,x) and a 32 bit (00,xx)
-- until that major device starts depending on 16+ bit minor numbers that is.]

>  I'd also like to suggest that 16 bits for the major number is over
> doing it. I can understand 8 getting a bit tight in exceptional
> conditions, but 12 ought to it. I suggest major numbers are 8 or 12
> bits in size, and the rest is minor numbers.

UNOS (my favorite dinosaur era near-UNIX) split device numbers 8/24 (or was
it 12/20?), and this allowed all manner of useful device control bits to be
encoded.  12 bits of major number would make for a very large cdevsw[], 
though I guess the table is sized according to what numbers are in use.