n64/n32 - build support patches

[C. Jayachandran]

On Fri, Mar 5, 2010 at 8:49 PM, M. Warner Losh <imp at bsdimp.com> wrote:
> CJ,
>
> Thanks for all the work in this area.  I know that Julie Mallett is
> also working in this area, and I've done some work to clean up the
> build system.  I'm afraid that we now have some conflicts between the
> different efforts, and that it may take a little to resolve the
> conflicts.

This is okay, I can rebase the patch when the other efforts are in,
(or can merge other efforts after this goes in).  I need a base on
which to start work on n64/n32 and I'm making my changes available as
it may be useful for others working in the same area.

Regards,
JC.


> In message: <98a59be81003050215m753799dbu35d2b2eab65e75db at mail.gmail.com>
>            "C. Jayachandran" <c.jayachandran at gmail.com> writes:
> : I've updated the patches for building n64 and n32 kernel and
> : userspace,  please review. The patches are:
> :
> : Tool chain support for n64 and n32::
> : http://sites.google.com/site/cjayachandran/files/n64-n32-toolchain.patch
>
> : This updates the gcc configuration  to provide the default ABI, linker
> : scripts and endianness flags during n64/n32 and o32 builds. With this
> : change, we don't need the -Wl, flags for linker emulation, the mabi
> : flags or the -EB/-EL for compiler.
>
> This part of the patch looks good.
>
> : I had to add a make variable in TARGET_DEFINES in
> : gnu/usr.bin/cc/Makefile.tgt and use that to pass the defaults while
> : creating 'tm.h' - this change is not MIPS-specific.
>
> TARGET_DEFINES is OK.  I'm not sure about the other Makefile variables
> and would like to find some way to not need them.
>
> : User/kernel build with n64 and n32:
> : http://sites.google.com/site/cjayachandran/files/n64-n32-build.patch
> :
> : updated bsd.cpu.mk which does not set LDFLAGS, adds ldscript.mips.n32
> : and removes unnecessary flags from Makefile.mips.
>
> +.if ${MACHINE_ARCH} == "mips"
> +. if defined(TARGET_64BIT)
> +.  if defined(TARGET_BIG_ENDIAN)
> +LD += -melf64btsmip_fbsd
> +.  else
> +LD +=  -melf64ltsmip_fbsd
> +.  endif
> +. elif defined(TARGET_N32)
> +.  if defined(TARGET_BIG_ENDIAN)
> +LD += -melf32btsmipn32_fbsd
> +.  else
> +LD +=  -melf32ltsmipn32_fbsd
> +.  endif
>
> I'd rather find some way to avoid adding TARGET_64BIT, TARGET_N32.
> I've been working to eliminate TARGET_BIG_ENDIAN because it is proving
> to be unworkable in practice, especially as these machines become more
> self-hosting.
>
> If you don't mind, I'll pull in the parts that don't conflict with the
> other work in this area, and we can then look at the harder problem of
> which pieces of that goes in.
>
> Warner
>



-- 
C. Jayachandran    c.jayachandran at gmail.com