Re: IA64 GNU Linux : data segment overflowed when linking with cvs head version 040923 20040923 and not with SLES 2.15.90.0.1.1 20040303

[James E Wilson]

On Wed, 2004-10-06 at 07:49, Dupont, MichaelX wrote:
> ld: BLAH: short data segment overflowed (0x4f0b58 >= 0x400000)

There are a great many ways to overflow the short data segment.  I don't
think you will get much help by trying to google for answers.

We would need a testcase that we can use to reproduce the problem in
order to do anything at all here.  You probably want to talk to HJ about
this if you haven't already,  address@hidden  He does IA-64
binutils work for Intel.

> This error message also occurs in the following bug reports :
> http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=33354

I think this is of no use to you. 

> Here is a proposed workaround  to link with –G0 (which did not have an
effect)
> http://www.ccs.ornl.gov/Ram/issues.html#ld_overflow

It isn't clear that this has anything to do with IA-64.  -G0 is really
much more a compiler option than a linker option.  I doubt that it will
have much affect if only used when linking.  If you want it to do
something, you have to use it when compiling.
>  
>Does anyone have any ideas as to why this links with SuSe’s version and
not with CVS head?

Hard to say.  There could be secondary issues.  Maybe different gcc
versions or different glibc versions use different amounts of short data
segment space, thus causing overflows when using some gcc/glibc versions
but not others.  It isn't clear if you did all of your testing on the
same machine will all of the same gcc/glibc versions.

Assuming no secondary issues, then the problem could be that mainline
has a bug fix that the SuSE linker does not have, and this bug fix as a
side-effect causes changes to the short data section, and these changes
cause the short data section to overflow for your example.

It isn't clear if there is a binutils problem here at all.  The Intel
compiler may be trying to use more short data space than is available
after linking in gcc/glibc startup files and libraries.

Without a testcase, I really can't say anything more useful than this.

> Is there a special flag to pass to configure or some patch that SuSe
applied that is not in the CVS version?

I seriously doubt that.  However, I don't have a copy of SuSE IA-64
linux, so I can't say for sure.
-- 
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com