Illegal instruction on m68k nommu (solved. C and C++ now work.)

[David McCullough]

Jivin Lennart Sorensen lays it down ...
> On Tue, Apr 21, 2009 at 09:47:04AM +1000, David McCullough wrote:
> > If someone wants to work up a patch for the uclinux.org version that
> > fixes this (but still has a configure option or something to allow it to
> > be used for older toolchains with busted CTOR/DTOR support) I will gladly
> > add it.
> 
> Would be neat if it could detect it somehow at runtime.

Originally we had thought about having ld-elf2flt remove the mostly
likely offending .o's from the link line so that elf2flt's CTOR/DTOR
support would always work whether the toolchain included working
CTOR/DTOR support or not.

Somehow this never eventuated.

We could add a '-old-ctor-dtor-support' option to ld-elf2flt to enable
adding the support in and mark it in elf2flt.ld like SINGLE_LINK or the
various *DAT options.

> > Not everyone using it is running gcc-4.3, linux-2.6, m68k or blackfin,  and
> > this is why it's not always easy getting patches changes back in.
> 
> Certainly true, but at the same time until it is fixed no one is going
> to use gcc 4.3 either.  No one is forcing anyone to update to the newest
> version of elf2flt, but it sure would be nice that if you upgrade you
> get one that works with current tools.

Yep,  and I would be happy with a configure option to enable the old
behaviour and have the new behaviour be the default.

Really,  most solutions I can think of would be ok.  How different is
the bfin version ?

> > If a patch is relatively easy to review and fairly obviously safe or in the
> > right direction,  it goes in :-)
> > 
> > As for the elf2flt version numbers,  they were previously done as
> > date-stamped releases shipped with toolchains off uclinux.org.  But I
> > haven't done anything with toolchains for some time,  and the CVS
> > version would be the "ideal" starting point unless you are looking
> > at old toolchains.
> 
> Of course.  Hard to test things though given I only have a couple
> of platforms I could test on, which is nowhere near all the possible
> combinations.

Thats fair enough,  I don't expect it to be tested,  as long as it
passes the "looks ok test" it's usually close enough,

Cheers,
Davidm

-- 
David McCullough,  david_mccullough at securecomputing.com,  Ph:+61 734352815
McAfee - SnapGear  http://www.snapgear.com                http://www.uCdot.org