locore32@gaea.ocn.ne.jp ("Toru Nishimura") writes:
> You missed my point, entirely.  There is no need to maintain cpu_attach()
> routine as long as new processor support is required for a specific target.
> For DECstation, it's the dead weight to have, say, Vr stuff TX stuff.
> Or, are you willing to maintain cpu_idenitfy() bloat for ever-increasing
> MIPS diversity in terms of MI commonness? Com'on, remember the
> engineering sense of balance.

as opposed to bloat for cache ops, or irrelevant pmap bits, or...

if you want to go making it smaller, conditionalize support for
specific CPUs in what should be an almost completely table-driven
operation using #ifdefs, and have ports only include the CPU defines
that they need.

you would, i think, want a couple of convenience #ifdefs which turn
approximately everything on.

That way not only can you get the completely generic kernel for things
that need it -- and a bunch of things need it -- but you can also trim
down in specific kernels all the way to your hardware.


but that's only if it's needed.  I'm not convinced that it is...



cgd