Re: traditional mips vs. little endian?

["H . J . Lu" <hjl at lucon dot org>]

On Sun, Jul 22, 2001 at 01:12:30PM -0700, cgd@sibyte.com wrote:
> "H . J . Lu" <hjl@lucon.org> writes:
> > I don't know which MIPS ABI the other little-endian targets are using.
> > But on the other hand, unless the target is used by SGI, I don't see
> > there is any reason to use the SGI MIPS ABI. We changed Linux to the
> > SVR4 MIPS ABI from the SGI MIPS ABI, which was a mistake to begin with
> > since we have no support fior the SGI MIPS ABI in glibc. My questions
> > are which little-endian MIPS targets use the SGI MIPS ABI and how it is
> > supported.
> 
> At the very least "mips-elf" and friends -- which have been used as-is
> for embedded development by people for who knows how long.  Changing
> their ABI would be a mistake, in my opinion, because it would break
> compatibility with who knows how many people's long-existing binaries.
> 
> But worse, even if it were The Right Thing to change the ABI of the
> little-endian mips targets, that wasn't done (as far as I, and the
> current configury, can tell).  The existing tests assume that it was,
> but the binaries that are built do not match that.

The main differences between the SVR4 MIPS ABI and the SGI MIPS ABI are
in shared libraries and section/symbol orders. When Linux is switched
to the SVR4 ABI, the only things we have to do is to remove some codes
in the dynamic linker which attempted to support the SGI ABI and never
worked. Nothing else needs to be changed.

I believe only IRIX should use the SGI ABI and the rest should use
the SVR4 ABI. But it is not my call. I have no problems to limit
the SVR4 ABI to Linux in the binutils testcases.

H.J.