Subject: Re: kern/11627
To: gabriel rosenkoetter <gr@eclipsed.net>
From: Nathan J. Williams <nathanw@MIT.EDU>
List: current-users
Date: 01/31/2001 16:36:02
gabriel rosenkoetter <gr@eclipsed.net> writes:

> On Wed, Jan 31, 2001 at 10:27:36AM -0500, Rob Quinn wrote:
> > > Is it necessary to panic under this situation?
> > 
> >  Until there's a real solution, can't we at least change the default i386
> > INSTALL config?
> 
> Rather a bad idea for machines which really *do* have multiple ISA
> buses, don't you think?

On an i386, there's no such thing. ISA isn't a "real" bus anymore;
it's simply "that which responds to I/O cycles and possibly memory
cycles < 16M".

Even with multiple PCI-ISA cards (a bad idea!), only one of them can
be programmed as subtractive decode; the other would have to be
programmed to know about the address ranges of the devices behind
it, and they'd have to not overlap with the other ISA devices. So
there's effectively only one ISA bus, regardless of what PCI-ISA
adapters you plug in.

I suppose it would be possible for someone to make a PCI-ISA bridge
that offset the ISA space to somewhere else, but we certainly have no
support for such a beast.

Until and unless we do, everything that can arrange for multiple ISA
buses on an i386 should be ripped out, because it's wrong.

(I'm in favor of ripping out "isa* at pcib?" and "isa* at pceb?" and
just leaving "isa0 at mainbus0", since the former two give the
not-very-useful impression that the entire ISA bus is behind them.)

        - Nathan