Re: -m68060 breaks stuff in NetBSD-6

["Leonard Schmidt" <Leonard24%gmx.net@localhost>]

> On Jan 28, 2013, at 5:12 AM, Ignatios Souvatzis wrote:
> > On Sun, Jan 27, 2013 at 09:24:21PM +0000, John Klos wrote:
> >
> >> I noticed that -m68040 had issues, but they've been addressed here:
> >>
> >> http://gnats.netbsd.org/45857
> >>
> >> I was having problems with compiling stuff from pkgsrc on my m68060
> >> Amigas which I mentioned here:
> >>
> >> http://mail-index.netbsd.org/port-amiga/2012/05/16/msg007665.html
> > ...
> >> Just wanted to throw that out there in case anyone else is still
> >> using -m68060. One day when someone has time we'll have to see where
> >> this is breaking.
> >
> > not really - at least I can't find a message in the web interface
> > that explains a connection to 68060 compilation.
> >
> > -m68060 shouldn't damage userland, normally. It has *less*, not
> > *more* instructions, than 68020/30. It breaks the kernel
> > because it influences the inline assembler settings; optimization
> > for the kernel is set in higher makefile magic. (Long: We need
> > -m68020-60 for GENERIC, -m68060 for 68060 kernels, but need an
> > additional setting to force the assembler to use 68030 syntax, as
> > we have only one set of inline assembler stuff - and use .word
> > directives for other CPU's (rather: MMU's and FPU's) instructions.
> >
> >> It turns out that this was due to -m68060 optimizations in NetBSD-6.
> >
> > Ah, I wonder: ISTR that there are some FPU coprocessor changes
> > between 020/30 and 040/60, too, so you have to be careful there at
> > least if you use inline assembler.
> >
> > It would help if you'd show the -S output of a single failing file.
> 
> Could this explain some of the compiling problems that I have in  
> NetBSD 6?
> 
Hmm, I usually compile with -O2 -m68060 options and I didn't experience any 
failures due to compiler errors so far. My /etc/mk.conf looks like this:

COPTS+= -O2 -m68060 -Wa,-m68060,-m68851
M68060=YES

Kernel and userland compiles just fine with this setting, last tested with HEAD 
trunk some days ago. I don`t use pkgsrc, I usually download the source archives 
directly from their host sites and do the classical ./configure; make install 
thing.  I also set CFLAGS="-O2 -m68060" in my .profile so that ./configure 
automatically generates Makefiles with this two compiler options. All 
applications compiled this way worked for me so far.  So I cannot confirm that 
there is some gcc problem with -m68060 option. -O3 is another story though.

-Leo24