Re: cr16-elf tools error: no .eh_frame_hdr table will be created.

["M R Swami Reddy" <MR dot Swami dot Reddy at nsc dot com>]

Hi,
It looks like an issue with GAS generating the object file.
--Testcase: test.s:
==
      .file   "t12.c"
       .globl  __Stdin
       .section        .data
       .type   __Stdin, @object
       .size   __Stdin, 72
       .long   0
__Stdin:
       .short  1
       .byte   0
       .byte   0
       .long   __Stdin+0x44
       .zero   40
       .ident  "GCC: (GNU) 4.1.2"
====
Command line:
> cr16-elf-as test.s -o test.o
> cr16-elf-objdump -Dr test.o
==
..
Disassembly of section .data:
00000000 <__Stdin-0x4>:
  0:   00 00           illegal
       ...

00000004 <__Stdin>:
  4:   01 00           illegal
  6:   00 00           illegal
  8:   48 00 00 00     andd    $0x0:l,(r9,r8) // ERROR: it should 44 00 
00 00
  c:   00 00
                        8: R_CR16_NUM32 __Stdin
...
====

Instead of including the 0x44, gas include "0x44 + _Stdin". Is this 
expected?

Any inputs are appreciated.

Thanks
Swami

M R Swami Reddy wrote:
> Hi Alan,
>
> With below patch, GAS does resolve the relocs partially. But final 
> linking
> a.out relocs are not performed correctly, so .data section has 
> incorrect data/info.
>
> Attached test case for describing the problem:
> t12.s -> Test case
> t12.dr -> Dump output with -Dr option with below (email) patch on 
> t12.o file.
> t12.dr.old -> Dump output with -DR option without patch on t12.o file.
> s.new  -> Dump output with -s option for a.out (Fully linked)
> s.old -> Dump output with -s option for a.out (Fully linked) With out 
> below patch.
>
> Any idea who cause this type of problem. I guess, it could an issue with
> R_CR16_NUM32 relocation type flags setting.
>
>
> Thanks
> Swami
>
> Alan Modra wrote:
> > On Fri, Jul 03, 2009 at 03:53:45PM +0530, M R Swami Reddy wrote:
> >  
> > > Btw, I think, no need to do or fix in TC_LINKER_RELAX_FIXUP here.
> > > It looks like an issue with section flags setting or else.
> > >     
> >
> > No, if you make .eh_frame a non-allocated section, it won't be in
> > memory.  How will the unwinder work without its unwinding info?
> >
> > The following patch will cause non-code section fixups to be resolved
> > at assembly time, while code fixups will result in relocations that
> > the linker will be able to use for relaxation.  Since the linker
> > doesn't even relax non-code sections (see elf32_cr16_relax_section)
> > there is no point in emitting extra relocs for sections like
> > .eh_frame.  Applied.
> >
> >     * config/tc-cr16.h (TC_LINKRELAX_FIXUP): Set only for code sections.
> >
> > Index: gas/config/tc-cr16.h
> > ===================================================================
> > RCS file: /cvs/src/src/gas/config/tc-cr16.h,v
> > retrieving revision 1.3
> > diff -u -p -r1.3 tc-cr16.h
> > --- gas/config/tc-cr16.h    27 Nov 2008 11:57:29 -0000    1.3
> > +++ gas/config/tc-cr16.h    4 Jul 2009 01:32:57 -0000
> > @@ -1,5 +1,5 @@
> >  /* tc-cr16.h -- Header file for tc-cr16.c, the CR16 GAS port.
> > -   Copyright 2007 Free Software Foundation, Inc.
> > +   Copyright 2007, 2009 Free Software Foundation, Inc.
> >  
> >     Contributed by M R Swami Reddy <MR.Swami.Reddy@nsc.com>
> >  
> > @@ -49,8 +49,9 @@ extern const struct relax_type md_relax_
> >  #define TC_FORCE_RELOCATION(FIXP) cr16_force_relocation (FIXP)
> >  extern int cr16_force_relocation (struct fix *);
> >  
> > -/* Fixup debug sections since we will never relax them.  */
> > -#define TC_LINKRELAX_FIXUP(seg) (seg->flags & SEC_ALLOC)
> > +/* Fixup non-code sections since we will never relax them.  */
> > +#define TC_LINKRELAX_FIXUP(seg) \
> > +  ((seg->flags & (SEC_ALLOC | SEC_CODE)) == (SEC_ALLOC | SEC_CODE))
> >  
> >  /* CR16 instructions, with operands included, are a multiple
> >     of two bytes long.  */
> >
> >