> If it's made
> a kernel option which is *not* enabled in standard configuration, then
> it's "just" extra cruft (aka "optional facility").  I'd rather have cruft
> like i386-specific bounce buffer support, but for some reason that isn't
> "necessary" enough [see footnote].

Uh, comparing this to i386 bounce buffer support is ... stupid.

(1) this was a part-of-a-weekend hack, done because i was bored and
wanted to see how hard it would be to do.
(2) when i mentioned to a few people that i'd done it and/or was doing
it, about 20-30% indicated that they could _use_ it, and wanted a
copy.

hence, i submitted a PR.


> You can probably tell that I'm among those people who are peeved by the
> fact that this problem, which has been well known (with known solutions)
> since *before* NetBSD started, hasn't been fixed because the solutions
> wasn't "clean" enough.  Well, it's a dirty problem, and that's because
> the hardware sucks, so what do you expect?  "Buy a better motherbord"
> just isn't a solution, to most people. "Switch to linux" is.  End peeve.

It's very, very easy to implement a bounce-buffer solution that works
great on the i386.  It's also quite easy to have that bounce-buffer
solution make isa devices even harder to use on other systems, like
the Alpha.

Given that there are other systems which _are_ using ISA device
drivers, and that they need to have the same issue addressed, why not
allow it to be addressed it a way that all can use?

As far as i know, i'm the only person who does NetBSD development
who's actually sat down and thought about a solution; don't bitch me
out about it.


cgd