Re: Patches to fix building m68k with CPUFLAGS

[Rin Okuyama <rokuyama.rk%gmail.com@localhost>]

Hi,

On 2020/07/20 12:14, Jake Hamby wrote:
> After further testing, I found that GCC 8.4.0 generates bigger, and
> very buggy code, for m68k. I've seen no freezes, uvm errors, or panics
> on illegal instructions with the NetBSD-current build I made with
> "-m68040" and my previous patch, minus the part to change HAVE_GCC
> from 7 to 8.

Yes, amiga kernel does not work correctly if compiled by GCC8. See more
details in this thread:

http://mail-index.netbsd.org/port-m68k/2020/06/04/msg000760.html

On 2020/07/20 3:20, Jake Hamby wrote:
> One more note on the "libgcc/config/m68k/lb1sf68.S" patches (to gcc
> and gcc.old):
...
> The reason you only see a failure
> here with CPUFLAGS set is because, I suspect, "defined (__mcoldfire__)
> && !defined (__mcfisab__)" must be defined in the default
> configuration, and not if you use, e.g. "-m68040" or "-m68020-60"
> (which should be the ideal default build flag for targets like Atari
> ST and Amiga).

This is not correct. __mcoldfire__ is not defined for our m68k ports.
You can confirm this by:

% m68k--netbsdelf-gcc -dM -E - < /dev/null | grep coldfire

Also you can see braw is used for the library compiled without
optimization flags as:

% m68k--netbsdelf-objdump -d libgcc_s.so | grep _exception_handler

I guess that libgcc_s can be compiled without -m68040 or -m68060, just
because the generated code is small enough. In order to work this around,
you can simply replace bra with bral in that file. Alternatively, you
can build the library by skipping tsort and lorder:

https://mail-index.netbsd.org/tech-toolchain/2020/06/15/msg003823.html

On 2020/07/20 2:35, Jake Hamby wrote:
> --- a/external/gpl3/binutils.old/usr.bin/gas/arch/m68k/config.h
> +++ b/external/gpl3/binutils.old/usr.bin/gas/arch/m68k/config.h
> @@ -298,7 +298,7 @@
>   #define TARGET_ALIAS "m68k--netbsdelf"
>
>   /* Define as 1 if big endian. */
> -/* #undef TARGET_BYTES_BIG_ENDIAN */
> +#define TARGET_BYTES_BIG_ENDIAN 1
>
>   /* Canonical target. */
>   #define TARGET_CANONICAL "m68k--netbsdelf"

This patch is not necessary because m68k is purely big-endian platform,
and TARGET_BYTES_BIG_ENDIAN is unconditionally defined in
src/external/gpl3/binutils{,.old}/dist/gas/dist/config/tc-m68k.h.

Your patch for external/gpl3/gcc/dist/gcc/config/m68k/m68k.h seems
rational. I'd appreciate a lot if you could report this to upstream.

As you pointed out, CPUFLAGS should be neglected in order kernel and
some standalone programs to get compiled. However, I'm not sure whether
share/mk/bsd.sys.mk is the best place to do this.

Thanks,
rin