Re: [trunk + 3.4-branch RFA] don't use empic relocs formips-linux eh

[cgd at broadcom dot com]

At Mon, 14 Jun 2004 13:11:16 +0100, Richard Sandiford wrote:
> cgd@broadcom.com writes:
> > I'd like this approved for the branch as well, so 3.4.1 will work
> > nicely w/ the next major binutils release.
> 
> Agreed that using embedded pic relocations for linux is a bug.
> It's almost certainly a regression from some earlier version of gcc,
> so if you can give a specific version number, I think the patch should
> be OK for 3.4.x.

This was introduced in 3.0.  I'm not sure when the use of DWARF EH 
info for MIPS was introduced into mips-linux gcc, it may have been the
same time.


> > +/* If possible, we should attempt to use GP-relative relocs for this
> > +   (see http://sources.redhat.com/ml/binutils/2004-05/msg00227.html).
> > +   However, until that is implement, this just uses standard, absolute
>                              ^^^^^^^^^
> typo

Thanks.  (Moot, considering the comments below.)


> > +   references.  */
> > +#define ASM_PREFERRED_EH_DATA_FORMAT(CODE, GLOBAL)	DW_EH_PE_absptr
> 
> This is the same as the definition in defaults.h.  Do we really  need it?

It doesn't have to be defined -- yes, it's the same as the default.

I thought that having the define (duplicating the default) here would
be useful, to reference the discussion.


> If not, and given that the comment references a URL anyway, I think it
> would be better to just remove both the linux.h and linux64.h definitions.
> Anyone who wonders why can go to Google.
> 
> A patch to remove both definitions is preapproved if it works.

Heh.  OK, I'll retest w/ that change.

(it'll work -- no-define has been working on mips64-linux -- but
testing good.  8-)



chris