> i have no feeling one way or the other about the encoding other than that
> it's easier in a macro to do a bit test than a <= and a >=.  but since i
> can think of several "tun"-like virtual devices i'd like to add using LKM's,
> i *do* see a reason why we should allow thousands of devices per kernel.

There's no reason that you can't have thousands of LKM devices per
kernel, _without_ devoting a bit to it.

doing _anything_ other than a table is going to be slower than a
table-based approach (though, compared to the work required to do a
device access, that's not really significant).  Using a solely
table-based approach _does not_ prevent you from switching to using a
bit mask later, does not prevent you from having thousands of LKM
devices, etc.

There's no need to do the bit-fiddling now, so why make one large
change which introduces a whole lot of complexity, when there really
is no reason that part of it couldn't be done later?



cgd