NetBSD on OpenFirmware-less Macintoshes

All,

a few days ago, a Power Macintosh 7100/80 has joined my
collection. I have
looked on the web a bit since, to get an idea of what blocks
the way to
NetBSD support,a nd I'd like to spawn a (yet another ;)
discussion.

Of the two major issues, support for the 601 and the lack of
OpenFirmware,
the former seems to come along in arch/powerpc, right? So I
concentrated on
looking into hardware support for now.

What came up time and again, is the question whether support
for 1st
generation Power Macs should be part of an existing port
(macppc, mac68k),
or a new port. At the time, Bill Stouder-Studenmund strongly
advocated
integrating Nubus Macs with the macppc port similarly to
what happened with
sun3 and sun3x:

<http://groups.google.com/group/so
l.lists.netbsd.port.macppc/browse_thread/threa
d/9a1cae1740a3b1c1/896509ba2496e012>

Earlier, Bill had outlined a possible autoconf structure
that would serve
both non-OF and OF power Macs:

<http://mail-index.netbsd.org/port-macppc/1998/1
0/07/0001.html>

While the sun3/sun3x example sounds appealing, the two
machine classes
share a lot more than just the cpu (68020 vs. 68030). The
macppc port,
OTOH, is closely tied to OpenFirmware support (mainbus
already implies
presence of OF), and PCI (obio attaches to pci, not
mainbus). 1st
generation Power Macs have neither of the two, and because
of their
different configuration mechanism, hardware, interrupt
structure are a lot
closer to mac68k than to macppc. While there have been
discussions to add
Nubus support to LinuxPPC by constructing a fake OF device
tree for this
machine class, the idea was rejected:

<http://groups.google.com/group/co
mp.os.linux.powerpc/browse_thread/thread/12bc8
6ecd3d55134/03a29a4b79f8c488>

In the absence of such a fake OF-like device tree, support
for 1st
generation Power Macs results in either littering code with
#ifdefs where
it refers to hardware information, or separating source
trees (alike
sun3/sun3{,x}/).

When I browsed the macppc/conf/* files, I stumbled on the
fact that
macppc/conf/files.macppc, which is unique per port,
unconditionally pulls
in OF support files both from arch/powerpc, and dev/ofppc.
Unfortunately,
the 'include' statement does not take an attribute (see
sun3/conf/files.sun3). Littering the powerpc files.* with
'of' attributes
doesn't sound too appealing, either.

Comments?

	hauke


--
"It's never straight up and down"     (DEVO)

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

On Jul 7, 2007, at 08:55, Hauke Fath wrote:

> a few days ago, a Power Macintosh 7100/80 has joined my
collection. I  
> have
> looked on the web a bit since, to get an idea of what
blocks the way to
> NetBSD support,a nd I'd like to spawn a (yet another ;)
discussion.
>
> Of the two major issues, support for the 601 and the
lack of  
> OpenFirmware,
> the former seems to come along in arch/powerpc, right?
So I  
> concentrated on
> looking into hardware support for now.
>
> What came up time and again, is the question whether
support for 1st
> generation Power Macs should be part of an existing
port (macppc,  
> mac68k),
> or a new port. At the time, Bill Stouder-Studenmund
strongly advocated
> integrating Nubus Macs with the macppc port similarly
to what happened  
> with
> sun3 and sun3x:
>
> <http://groups.google.com/group/sol.lists.netbsd.port
.macppc/ 
> browse_thread/threa
> d/9a1cae1740a3b1c1/896509ba2496e012>
>
> Earlier, Bill had outlined a possible autoconf
structure that would  
> serve
> both non-OF and OF power Macs:
>
> <http://mail-index.netbsd.org/port-macppc/1998/1
0/07/0001.html>

I don't like that idea.
NuBus and PCI PowerMacs don't have much more in common than
the CPU  
class and ADB ( ok, that's a little bit exaggerated but I'm
sure you  
know what I mean )
As it is today macppc is pretty much CHRP with apple
flavour.

> While the sun3/sun3x example sounds appealing, the two
machine classes
> share a lot more than just the cpu (68020 vs. 68030).
The macppc port,
> OTOH, is closely tied to OpenFirmware support (mainbus
already implies
> presence of OF), and PCI (obio attaches to pci, not
mainbus).

Exactly.

> 1st generation Power Macs have neither of the two, and
because of their
> different configuration mechanism, hardware, interrupt
structure are a  
> lot
> closer to mac68k than to macppc. While there have been
discussions to  
> add
> Nubus support to LinuxPPC by constructing a fake OF
device tree for  
> this
> machine class, the idea was rejected:
>
> <http://groups.google.com/group/comp.os.linux.
powerpc/browse_thread/ 
> thread/12bc8
> 6ecd3d55134/03a29a4b79f8c488>
>
> In the absence of such a fake OF-like device tree,
support for 1st
> generation Power Macs results in either littering code
with #ifdefs  
> where
> it refers to hardware information, or separating source
trees (alike
> sun3/sun3{,x}/).

I'd rather see NuBus code factored out of mac68k, cleaned of
68k-isms (  
if any ) and moved to sys/dev/nubus, same with ADB ( code
exists but so  
far works only on macppc, mainly because I don't have any
other  
ADB-equipped hardware )
Not sure what else could be recycled - interrupt handling
from mac68k?  
mesh and friends from macppc? Any other obio devices that
are similar  
enough?

IMHO faking OF etc. for NuBus PowerMacs is too much work and
additional  
messiness for too little gain. Better start a nice, clean
new port and  
have it share as much code as possible with the other mac
ports.

have fun
Michael
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (Darwin)

iQEVAwUBRpHxAMpnzkX8Yg2nAQLHMggAg0GmXeXroEM0Rk3KOB4NJmyeok+J
vzJX
pmD5HxQhtjKaqXeM6vwckt6gQdygCxZC7PmeFjsayZOHG2u5Fop3u6Kwp4Ip
gu+o
EqiLQZ+tTKnP7BsXT2OCMLn08ql6sZOHnFUcK3OhHByuMluCgr+fvIW8Xfmh
jBS9
3pfARsrK4l1kCyF+bE3/37jon+nxguIZklq+y9PUs6hGDZ7kTy6NRx1y6t6Y
yB/D
BaF4Zru4/3kVnXStkTdzw65X7V2uKilTPT8i/MtmuOvF1vyjk5gAxb92YS8C
piU9
wNAPyjeXV8lcVkPCjYEPdUthl+pPAM8My2TRlckvwLY2erCC/zliSw==
=5Ntg
-----END PGP SIGNATURE-----