Re: [PATCH] Fix distinction of 32/64bit addresses in MIPS gas

[Thiemo Seufer <ica2_ts at csv dot ica dot uni-stuttgart dot de>]

Ian Lance Taylor wrote:
> Thiemo Seufer <ica2_ts@csv.ica.uni-stuttgart.de> writes:
> 
> > > I can't see how a non SysV MIPS ABI can be 
> > > expected to comply with a SysV standard.
> > 
> > AFAICS this use of R_MIPS_64 isn't covered by any ABI.
> 
> To make my point in a different way, the GNU tools do a lot of things
> which aren't covered by any ABI.  That doesn't make them wrong.

I agree.

> For example, the i386 ELF backend supports 8-bit and 16-bit
> relocations.  They are not covered by any ABI, but they are required
> to build 16-bit code used by PC bootstrap code.

The problem with R_MIPS_64 is that it is part of ABI 64 and ABI N32,
being unused in the latter. We now have a GNU extension which uses
this relocation with different semantics. As long as this extension
isn't used in n32 we are probably safe, otherwise it may break
linking. The clean way would have been to invent a R_MIPS_GNU_64
relocation for this purpose. Well, OTOH there's no urgent need
to change current behaviour.


Thiemo