striking out at the eabi

[Mike Taht <mike at ackley dot net>]

I have been working on coaxing my arm ep9302 based board (a ts7250) into a 
modern state. The other week I coaxed (well, with tons of help from 
#ep93xx on freenode) the current linux 2.6.17 git kernel + the patches 
from http://www.wantstofly.org/~buytenh/ep93xx/ to not only boot, but run 
stably with busybox and a few apps, and recognise absolutely tons of usb 
devices. This is the first low end processor I've ever had that could do 
host usb. woohoo, I'm on the bleeding edge!

I used a hard float gcc-4.0 2.3.5 non-eabi arm toolchain for that. Have 
also got a 4.1 based hardfloat toolchain to work. (also, the first time 
ever I've had a FC5 x86_64 based cross toolchain work and it's nice to 
think I can finally retire that old rh9 box)

Flushed with success I set out, following instructions on this list, to 
try getting softfloat/ntpl/eabi to work. (my end game is to finally get to 
where I can use the maverick crunch instructions on this processor, and 
yea, I know that's a longer way out)

One of my noneabi softfloat attempts got to where it could build the 
kernel but the floating point emulator was the wrong size (need to specify 
hard float on the kernel build?), and userspace would just hang in an 
init=/hello_world case.

More recently I've managed to get myself into a hole using some patches 
posted on this list for eabi (latest.dat latest-earm.sh etc and 
crosstool-0.42-mg2.patch), with great fear moving aside my 
/usr/include/{asm,linux} for the 2.6.17 headers, and still the compile 
blows up saying my headers are too old.

so,
1) has anyone got a known good arm eabi cross-compiler tarball for x86 or 
x86_64 I could just try?
2) anyone else insane enough to try and do builds from x86_64?
3) got a script?

I am also curious in the general case if the relaxed alignment issues in 
the eabi are going to cause performance problems for armv5 and below.

I'm redoing last night's build with sh -x now, maybe I just missed 
something stupid

TIA 
_______________ 
Michael Taht 



--
For unsubscribe information see http://sourceware.org/lists.html#faq