Re: PATCH: Support VER_FLG_INFO in readelf

[David Miller <davem at davemloft dot net>]

From: Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
Date: Mon, 08 Mar 2010 19:01:03 +0100

> David Miller <davem@davemloft.net> writes:
> 
>> From: Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
>> Date: Mon, 08 Mar 2010 12:49:19 +0100
>>
>>> gcctestdir/ld: error: /vol/gcc-4.4/lib/gcc/sparc-sun-solaris2.10/4.4.2/../../../libstdc++.a(class_type_info.o): unsupported reloc 18 against local symbol
>>
>> This relocation is R_SPARC_WPLT30, I wonder why it appears against a
>> local symbol since by definition a local symbol should never go
>> through the PLT.
>>
>> Was libstdc++.a built using Sun's as and ld?
> 
> Indeed, which is what I usually do on SPARC.  I've recently started
> trying either gas and Sun ld or gas and gld to compare results, but I'm
> much further along on x86 here.

Ok, was libstdc++.a built using -fPIC?

The assembler should never emit R_SPARC_WPLT30 for a call instruction
unless PIC was specified on the command line.

R_SPARC_WPLT30 specifically instructs the linker to create a PLT entry
for the symbol reference.

If Sun's assembler is emitting this for a non-PIC build, that's
extremely unfortunate.

Ugh! From bfd/elfxx-sparc.c:

	case R_SPARC_WPLT30:
 ...
	  if (h == NULL)
	    {
	      if (! ABI_64_P (abfd))
		{
		  /* The Solaris native assembler will generate a WPLT30
		     reloc for a local symbol if you assemble a call from
		     one section to another when using -K pic.  We treat
		     it as WDISP30.  */
		  if (ELF32_R_TYPE (rel->r_info) == R_SPARC_PLT32)
		    goto r_sparc_plt32;
		  break;
		}
	      /* PR 7027: We need similar behaviour for 64-bit binaries.  */
	      else if (r_type == R_SPARC_WPLT30)
		break;

	      /* It does not make sense to have a procedure linkage
                 table entry for a local symbol.  */
	      bfd_set_error (bfd_error_bad_value);
	      return FALSE;
	    }

So I guess I'll have to add something to GOLD to handle this too.