linux-mips-fnet
[Top] [All Lists]

Re: m700-10 success

To: linux-mips@fnet.fr
Subject: Re: m700-10 success
From: Systemkennung Linux <linux@mailhost.uni-koblenz.de>
Date: Wed, 8 May 1996 06:17:06 +0200 (MET DST)
In-reply-to: <199605072314.BAA29311@pallas.spacetec.no> from "Tor Arntsen" at May 8, 96 01:14:22 am
Hi,

> Then it's important to fix up the include/linux and include/asm symbolic
> links everywhere, no use having them to point to Ralf's home directory :-)

This is just the urgent request to give me an account on every Linux/MIPS
machine ...

> For binaries I use root-0.01.tar.gz from ftp.fnet.fr, I have shuffled a bit
> around with some stuff (particularly the rc scripts and /etc things) to get
> the network and nfsboot/mount scenario that I want, but this is not really
> of much use for anyone else.

Well, this archive was clearly just a snapshot of my development source
tree as it was at that point.  Looks like a mess but still usefull.

I'll try to replace the userland in root-0.01.tar.gz with lots of smaller
archives as  a first step before we have a real Linux distribution.

Well, distributions are a special thing with Linux/MIPS anyway.  We're about
to endup with two complete binary trees - one little and one big endian
tree.  I think this might be a real problem with some FTP servers mirroring
Linux/MIPS.  Two complete packages for one not that popular architecture
could affect the spreading negativly.  Being little endian many of use
don't have the advantage of conforming to the ABI, exchanging binaries gets
difficult.

For this reasons I'm really thinking hard about making a bi-endian kernel.
This is however Pandora's box and I'd like to keep it closed ...  Some time
ago I've already done some successfull experiments with big endian usercode
on my Acer.  This however was a tiny program; real world apps would
require us to change large parts of the kernel.  In particular ioctl's
will be fun.  The only thing that accounts positivly for us is that
enforced by Intel's "maladia segmenti"  almost every useraccess goes through
get_user() and put_user() which'd be very usefull hooks for making Linux
bi-endian.

So there is a decission we have to make.

  Ralf

PS: Little endian sucks.

<Prev in Thread] Current Thread [Next in Thread>