Re: Creation of #debian-68k on irc.debian.org (OFTC)

[Finn Thain <fthain@telegraphics.com.au>]

On Thu, 22 Jun 2006, Christian T. Steigies wrote:

> On Thu, Jun 22, 2006 at 02:54:28PM +1000, Finn Thain wrote:
> > 
> > Yes, it is lots of "fun". What was the failure? What versions of 
> > binutils, gcc, glibc and kernel headers did you use?
> 
> I tried to build it with cross-tool, binutils-2.16.1, gcc-4.0.3, 
> glibc-2.3.6 Somebody suggested to me to use official debian sources, but 
> toolchain-source is not a big help here. Then I realised that the debian 
> packages have instructions for cross-compling, which actually works,

Yes, I needed Anton Blanchard's patch for glibc-2.3.6 from the Debian 
sources:

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=339415

> so now I have debian packages with:
> 
> GNU assembler version 2.16.91 (m68k-linux-gnu) using BFD version 2.16.91 20060413 Debian GNU/Linux
> gcc version 4.1.2 20060613 (prerelease) (Debian 4.1.1-5)
> 
> But linux-image fails here:
> 
> arch/m68k/math-emu/fp_scan.S: Assembler messages:
> arch/m68k/math-emu/fp_scan.S:67: Error: Unknown operator -- statement `getuser.b (%a0),%d0,fp_err_ua1,%a0' ignored
> arch/m68k/math-emu/fp_scan.S:75: Error: Unknown operator -- statement `getuser.l (%a0)+,%d2,fp_err_ua1,%a0' ignored

Yep. That's the problem with recent binutils, for which you need the 
patches I mentioned from Al Viro. I guess I was wrong about binutils-2.16 
being OK.

> >   CC      drivers/video/fbmem.o
> > {standard input}: Assembler messages:
> > {standard input}:2113: Error: operands mismatch -- statement `moveq.l #8,%a4' ignored
> > make[2]: *** [drivers/video/fbmem.o] Error 1
> > make[1]: *** [drivers/video] Error 2
> > make: *** [drivers] Error 2
> 
> Yup, same error here with my old and trusted gcc-3.3.6.

I've no idea whether to blame gcc or binutils for this. It seems that 
gcc-4.{0,1} are OK, whereas gcc-3.{3,4} are not -- regardless of binutils. 

Which doesn't actually vindicate either binutils or gcc.

> BTW, I am not using the CVS directly, I diffed the linux-CVS against Linus'
> tree and apply that patch to the debian sources. But I can try a clean CVS
> checkout, maybe my diff is bad.
> 
> Lets move to l-m68k?

I'll send a patch to l-m68k to fix the other build failure (undefined 
reference to `set_page_count').

-f

> Christian
> 
> 
>