Re: If not CONFIG_ALL_PPC, you're broken

On Tue, Oct 17, 2000 at 12:57:39AM -0400, Dan Malek wrote:

"Mark A. Greer" wrote:

quoted

I've haven't been paying attention to this but it seems as though any
platform that doesn't fit in CONFIG_ALL_PPC doesn't build anymore.

We don't even need CONFIG_ALL_PPC anymore.  We can just rename
arch/ppc to arch/powermac_workstation, as that is all that works
right now.  It sucks to have to build a 2 Mbyte kernel with USB,
keyboard, mouse, and tons of other stuff just to get a kernel that
will link when all you want is a serial and Ethernet port......and
then hack init functions that you didn't want anyway so they don't
try to touch hardware you don't have.

quoted

........  Yes, I know, I am lazy
and should contribute more.  Anyway, I think we all need to pay closer
attention to the changes that are being put in, including me.

Well, putting in the time is only part of it.  I am trying to support
about a dozen different PowerPC boards, none of which look like a
PowerMac.  By the time I test a kernel across all of these platforms
and check in a few changes I am completely out of date and have to
start over.

quoted

What's the best way to get the other platforms working again?

Fortunately, it looks like some people (Matt, Troy, others) have pushed
a few PReP and CHRP updates, and Paul has a CHRP workstation for testing.
Maybe when the PMac stuff slows down, we can catch up.  Unfortunately,
we still have to build huge kernels with lots of stuff we don't need,
and I don't know how to fix that (well, I do but I don't think it
is likely to happen :-).

My solution for creating a slim embedded PPC kernel in 2.2 was to
stub out all the calls to Mac/CHRP specific routines that were made
in head.S and setup.c.  It doesn't require major surgery to the existing
code and let me generate a kernel with only the arch/ppc/kernel/ objects
that were required for the port.

I was planning on taking this same approach when I get around to
merging the K2 port into the real world.  I think it's a clean
approach if you don't mind stubbed functions.  I'm sure somebody
won't like it though...

--
Matt Porter
MontaVista Software, Inc.
mporter@mvista.com

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Matt Porter wrote:

My solution for creating a slim embedded PPC kernel in 2.2 was to
stub out all the calls to Mac/CHRP specific routines that were made
in head.S and setup.c.

I think that is fine, but we have lost the configuration options that
indicate this should happen.  I first want to get a set of options
that makes sense, rework the code so we can selectively build the
kernels again (with CONFIG_ALL_PPC a sum of the options), and be able
to select processor core/processor type separately (i.e. I need to
select CONFIG_6xx and CONFIG_8260 at the same time...there are more
of these coming from both Mot and IBM).


	-- Dan

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/