'final link failed: Bad value' when building Linux/MIPS kernels.

'final link failed: Bad value' when building Linux/MIPS kernels.

[Steven J. Hill]

Greetings.

I have been chasing down a binutils regression that is preventing me from
building a Linux/MIPS kernel using a gcc-4.2 based toolchain. Here is the
snippet of output when building a 2.6.12 kernel for my platform:

     LD      drivers/mtd/maps/built-in.o
     LD      drivers/mtd/nand/built-in.o
     LD      drivers/mtd/built-in.o
   mipsel-linux-uclibc-ld: final link failed: Bad value
   make[1]: *** [drivers/mtd/built-in.o] Error 1
   make: *** [_module_drivers/mtd] Error 2

The versions were:

   binutils-2.17.50.0.12
   gcc-4.2-20070131
   linux-headers-2.6.12

Using the release version of binutils-2.17 works fine and does not
produce the error above. However, binutils-HEAD out of today's CVS
and all of H.J. Lu's development releases > 2.17.50.0.3 will produce
the error. The regression appeared in binutils-2.17.50.0.4 and still
exists. The simple patch is:

   --- binutils-2.17.50.0.12/bfd/elfxx-mips.c
   +++ binutils-2.17.50.0.12-patched/bfd/elfxx-mips.c
   @@ -3603,12 +3603,9 @@
                             const Elf_Internal_Rela *relocation,
                             const Elf_Internal_Rela *relend)
    {
   -  unsigned long r_symndx = ELF_R_SYM (abfd, relocation->r_info);
   -
      while (relocation < relend)
        {
   -      if (ELF_R_TYPE (abfd, relocation->r_info) == r_type
   -         && ELF_R_SYM (abfd, relocation->r_info) == r_symndx)
   +      if (ELF_R_TYPE (abfd, relocation->r_info) == r_type)
           return relocation;

          ++relocation;

I applied to binutils-HEAD and was then able to build a valid Linux/MIPS
kernel and applications. I need to move on to other items, but if someone
could comment or have a look at this, it would be much appreciated. If a
better formatted bug report should be submitted somewhere, just let me
know. Thanks.

-Steve

Re: 'final link failed: Bad value' when building Linux/MIPS kernels.

[H. J. Lu]

On Thu, Mar 15, 2007 at 09:35:42PM -0500, Steven J. Hill wrote:
> Greetings.
> 
> I have been chasing down a binutils regression that is preventing me from
> building a Linux/MIPS kernel using a gcc-4.2 based toolchain. Here is the
> snippet of output when building a 2.6.12 kernel for my platform:
> 
>      LD      drivers/mtd/maps/built-in.o
>      LD      drivers/mtd/nand/built-in.o
>      LD      drivers/mtd/built-in.o
>    mipsel-linux-uclibc-ld: final link failed: Bad value
>    make[1]: *** [drivers/mtd/built-in.o] Error 1
>    make: *** [_module_drivers/mtd] Error 2
> 
> The versions were:
> 
>    binutils-2.17.50.0.12
>    gcc-4.2-20070131
>    linux-headers-2.6.12
> 
> Using the release version of binutils-2.17 works fine and does not
> produce the error above. However, binutils-HEAD out of today's CVS
> and all of H.J. Lu's development releases > 2.17.50.0.3 will produce
> the error. The regression appeared in binutils-2.17.50.0.4 and still
> exists. The simple patch is:
> 
>    --- binutils-2.17.50.0.12/bfd/elfxx-mips.c
>    +++ binutils-2.17.50.0.12-patched/bfd/elfxx-mips.c
>    @@ -3603,12 +3603,9 @@
>                              const Elf_Internal_Rela *relocation,
>                              const Elf_Internal_Rela *relend)
>     {
>    -  unsigned long r_symndx = ELF_R_SYM (abfd, relocation->r_info);
>    -
>       while (relocation < relend)
>         {
>    -      if (ELF_R_TYPE (abfd, relocation->r_info) == r_type
>    -         && ELF_R_SYM (abfd, relocation->r_info) == r_symndx)
>    +      if (ELF_R_TYPE (abfd, relocation->r_info) == r_type)
>            return relocation;
> 
>           ++relocation;
> 
> I applied to binutils-HEAD and was then able to build a valid Linux/MIPS
> kernel and applications. I need to move on to other items, but if someone
> could comment or have a look at this, it would be much appreciated. If a
> better formatted bug report should be submitted somewhere, just let me
> know. Thanks.

I suggest you open a bug report with all input files for
drivers/mtd/built-in.o and drivers/mtd/.built-in.o.cmd. You can find
input filenames in drivers/mtd/.built-in.o.cmd. 


H.J.

Re: 'final link failed: Bad value' when building Linux/MIPS kernels.

[Paul Koning]

>>>>> "Steven" == Steven J Hill <sjhill@realitydiluted.com> writes:

 Steven> Greetings.  I have been chasing down a binutils regression
 Steven> that is preventing me from building a Linux/MIPS kernel using
 Steven> a gcc-4.2 based toolchain. Here is the snippet of output when
 Steven> building a 2.6.12 kernel for my platform:

 Steven>      LD      drivers/mtd/maps/built-in.o
 Steven>      LD      drivers/mtd/nand/built-in.o
 Steven>      LD      drivers/mtd/built-in.o
 Steven>    mipsel-linux-uclibc-ld: final link failed: Bad value
 Steven>    make[1]: *** [drivers/mtd/built-in.o] Error 1
 Steven>    make: *** [_module_drivers/mtd] Error 2

I've run into the same message when building things for a NetBSD/MIPS
target.   I had assumed it was some error on my end.  

The problem with this message is that it is completely content-free.
It doesn't contain a meaningful problem statement, and it doesn't say
anything at all about where the problem is.  It doesn't even indicate
what input file triggered the problem!   How is a user supposed to fix
a failed build when this error message appears?  

    paul

Re: 'final link failed: Bad value' when building Linux/MIPS kernels.

[H. J. Lu]

On Fri, Mar 16, 2007 at 09:09:39AM -0400, Paul Koning wrote:
> >>>>> "Steven" == Steven J Hill <sjhill@realitydiluted.com> writes:
> 
>  Steven> Greetings.  I have been chasing down a binutils regression
>  Steven> that is preventing me from building a Linux/MIPS kernel using
>  Steven> a gcc-4.2 based toolchain. Here is the snippet of output when
>  Steven> building a 2.6.12 kernel for my platform:
> 
>  Steven>      LD      drivers/mtd/maps/built-in.o
>  Steven>      LD      drivers/mtd/nand/built-in.o
>  Steven>      LD      drivers/mtd/built-in.o
>  Steven>    mipsel-linux-uclibc-ld: final link failed: Bad value
>  Steven>    make[1]: *** [drivers/mtd/built-in.o] Error 1
>  Steven>    make: *** [_module_drivers/mtd] Error 2
> 
> I've run into the same message when building things for a NetBSD/MIPS
> target.   I had assumed it was some error on my end.  
> 
> The problem with this message is that it is completely content-free.
> It doesn't contain a meaningful problem statement, and it doesn't say
> anything at all about where the problem is.  It doesn't even indicate
> what input file triggered the problem!   How is a user supposed to fix
> a failed build when this error message appears?  

Please file a bug report with all input files for the linker. I
will at least provide clear diagnostic messages.


H.J.