Steve Woodford <steve@mctavish.co.uk> writes:

> On 30 Nov 2001, Aymeric Vincent wrote:
> 
> > On Nov, 11th, I did a full ELF build for the amiga port. I just had to
> > tweak a couple of makefiles in order to avoid descending into gdb and
> > the amiga bootblocks stuff.
> 
> Are you sure this was using the new toolchain? The old toolchain works
> works fine on m68k ELF for those platforms which have made The Switch.

Yes, it was with these versions (may be the fact that I was
cross-compiling from i386 is important?) :

vincent@fairness[12]%   ./m68k--netbsdelf-as -V
GNU assembler version 2.11.2 (m68k--netbsdelf) using BFD version 2.11.2
Assembler messages:
FATAL: Can't create a.out: Permission denied
vincent@fairness[13]%   ./m68k--netbsdelf-gcc -v
Reading specs from /usr/tools/lib/gcc-lib/m68k--netbsdelf/2.95.3/specs
gcc version 2.95.3 20010315 (release)
vincent@fairness[14]%

> What was the problem with gdb?

I'll post the log when I re-do a build. Basically, ISTR some m68k md
files were missing.

> Presumably the only thing holding up Amiga officially switching is
> bootblocks...

Yes, Michael Hitch has done some work on the bootblocks.

> 1) The compiler mis-optimises with -fstrength-reduce enabled (which is the
> default for -O2). I've asked Mycroft to look into this, but anyone with
> cc1 clue is welcome to track down the cause. Hint: It also seems to affect
> Vax and, on both platforms, causes cc1 to SEGV when compiling
> sys/dev/scsipi/ch.c ...

OK, I could build a kernel (which I couldn't boot) but it didn't
include any scsipi code. I'll try that later.

However, I could run most of the ELF userland from an a.out kernel
with EXEC_ELF32 enabled, but ls' output looks funny because of the
struct stat padding. Plus many programs dump core when going
multi-user, certainly because of ioctl()s...

 Aymeric