> But this is one of the tradeoffs of using NetBSD in the first place.
> NetBSD makes proper multi-platform support a priority over fixes
> for individual machines (as the lack of i386/ISA bounce buffer for
> so long demonstrates). FreeBSD is probably a better choice if you
> are using the 386 architecture and want to see that optimised, even
> at the cost of introducing incompatabilities with other architectures.

I run NetBSD on i386s, Amigas, and Sparcs.  I don't want to run different
OSes on different machines, as long as we have NetBSD.

I don't think there have to be tradeoffs like this.  An I/O driver is a
system specific thing and as such, can be optimized for every platform
(if we want).  The higher level things are not like this, but I believe
the com problems are at a very low level.

> However, that aside, it seems to me that since NetBSD has working
> code for the serial driver at one point, it should be able to be
> made to work again without adverse effects.

Yep.  I hope some guru will try to solve the problem (or provide some
information for us others to do it).

BTW, do the Sparc UART chips have any kind of FIFOs?

  -jm