Re: On resolving the MIPS gas branch reloc issue

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

cgd at broadcom dot com wrote:
> At Wed, 26 Feb 2003 08:30:50 +0000 (UTC), "Thiemo Seufer" wrote:
> > The old version, written according to the ABI spec, was useless, therefore
> > it was unused with the exception of some half-working/half-broken behaviour
> > in binutils, and the ABI people at MIPS say it was a typo in the spec and
> > should have never been defined this way.
> 
> in other words, you've _created_ a new element of an ABI.

I implemented it in the way it should have been long before.

> ultimately, the ability to branch to externs or syms in other sections
> is probably the right thing, but, well, if it's gonna be implemented
> by inventing something new in the ABI (because, well, what's there now
> obviously can't be used).  i'd really like to see it in an ABI
> document _first_.

This would be nice, but for MIPS there were many ABIs announced and
few were ever formally specified.

> Otherwise, it'll be implemented, and somebody'll
> change their mind about how it should work, and then some significant
> have the ABI change out from under them.

That's IMHO pretty unlikely, given that MIPS Corp. "owns" the ABI
and they agree on this interpretation of the Spec.

> > This patch fixes the problem of branches to globals in the same file,
> 
> I don't believe you need to actually create a new element of the ABI
> to do this, though...

The alternative would be to simply allow the GNU extension reloc for the
other ABIs, but then GNU binutils won't be compatible to other toolchains.
I forgot to mention that the SDE MIPS Tools (formerly Algorithmics, now
AFAIK the official MIPS toolchain) have used R_MIPS_PC16 for years in the
way my patch does now for binutils.

So, if nobody comes up with an implementation adhering to the ABI wording,
there shouldn't be any problem.


Thiemo