RFC on removing R_PARISC_DIR21L in DSO for hppa?

RFC on removing R_PARISC_DIR21L in DSO for hppa?

[Carlos O'Donell]

I'm seeing that on occassion a number of DSO's have ended up with
R_PARISC_DIR21L relocations, these relocations are not handled by the
dyanmic loader in glibc (I'm not sure how to handle them to tell you the
truth). There is code here that is '#ifdef 0' with a comment indicating
that it is for debugging purposes.

I assumed that DIR21L and such relocations would endup as DIR32
relocations, those being the only relocs I've ever seen in after a final
link. Would it be valid to warn again when DIR21L relocations are about
to make it into the final link? Or am I missing the point here, do they
finally get turned into DIR32 and halting the link in
elf32_hppa_check_relocs() is not the correct thing to do?

Cheers,
Carlos

Re: RFC on removing R_PARISC_DIR21L in DSO for hppa?

[Alan Modra]

On Sun, Jun 27, 2004 at 11:39:36AM -0400, Carlos O'Donell wrote:
> Would it be valid to warn again when DIR21L relocations are about
> to make it into the final link?

Yes.  Since glibc's ld.so doesn't handle them, you could merge the
DIR17F thru DIR21L cases with the DPREL ones immediately above.  ie.
turn the warning into a hard error.

I think when I wrote that code, I was distinguishing relocs that could
be handled by ld.so from those that can't (DPREL are only valid in the
main app).  It's a rather pedantic distinction when ld.so doesn't
provide support though!

I'm not sure why the warning was turned off for DIR17F .. DIR21L.  It
was probably an over-enthusiastic change when disabling other warnings.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Re: RFC on removing R_PARISC_DIR21L in DSO for hppa?

[Carlos O'Donell]

On Mon, Jun 28, 2004 at 09:15:02AM +0930, Alan Modra wrote:
> On Sun, Jun 27, 2004 at 11:39:36AM -0400, Carlos O'Donell wrote:
> > Would it be valid to warn again when DIR21L relocations are about
> > to make it into the final link?
> 
> Yes.  Since glibc's ld.so doesn't handle them, you could merge the
> DIR17F thru DIR21L cases with the DPREL ones immediately above.  ie.
> turn the warning into a hard error.
> 
> I think when I wrote that code, I was distinguishing relocs that could
> be handled by ld.so from those that can't (DPREL are only valid in the
> main app).  It's a rather pedantic distinction when ld.so doesn't
> provide support though!
> 
> I'm not sure why the warning was turned off for DIR17F .. DIR21L.  It
> was probably an over-enthusiastic change when disabling other warnings.

That's what I thought, I'll stew my patch a bit more to make sure I've
got the hard error in the right place, then I'll submit a patch.

Thanks Alan,

Cheers,
Carlos.