Relocation data value adjust

Relocation data value adjust

[daniel tian]

hi, guys:
    I have a problem in my binutils porting. Here is the example:
                .file	"mtable.c"
	.globl	funpointer
	.data
	.align	4
	.type	funpointer, @object
	.size	funpointer, 8
funpointer:
	.long	{a, c}
	.text
	.align	4
	.globl	main
	.type	main, @function
main:
	NOP
a:
                NOP
c:	
                NOP
	.size	main, .-main
	.ident	"GCC: (GNU) 4.3.0"
the label a and c will be 4, 8.
Because of the instruction being aligned with 4bytes, I wanna those
two label value right shift 2bit (the result will be 1, 2).

I digged the write.c, and still being puzzled. The difficulty I met is
how I could distinguish those two labels  being in funpointer (data
section) from like "JMP @a" (in .text section, because in jmp section,
there will be a fixup, and it will be resolved in md_apply_fix, the
"a" address value will be right shifted two bit in this function.).
and I noticed "a, c"  reloc_type is BFD_RELOC_32 (TC_CONS_FIX_NEW is
not defined in my target.), So I wonder whether I chould identitify
"a,c" with BFD_RELOC_32. Because when gcc will generate lots of debug
info with "-g", and BFD_RELOC_32/16/8 also exist in those debug info.

Can anyone give me some advice?
Thank you very much.

             daniel

Re: Relocation data value adjust

[Andreas Schwab]

daniel tian <daniel.xntian@gmail.com> writes:

> I digged the write.c, and still being puzzled. The difficulty I met is
> how I could distinguish those two labels  being in funpointer (data
> section) from like "JMP @a" (in .text section, because in jmp section,
> there will be a fixup, and it will be resolved in md_apply_fix, the
> "a" address value will be right shifted two bit in this function.).
> and I noticed "a, c"  reloc_type is BFD_RELOC_32 (TC_CONS_FIX_NEW is
> not defined in my target.), So I wonder whether I chould identitify
> "a,c" with BFD_RELOC_32.

Qualify the expression by using something like a@fptr and attach a
different relocation type to it.  See the powerpc port for an example.

Andreas.

-- 
Andreas Schwab, schwab@redhat.com
GPG Key fingerprint = D4E8 DBE3 3813 BB5D FA84  5EC7 45C6 250E 6F00 984E
"And now for something completely different."

Re: Relocation data value adjust

[daniel tian]

2010/5/21 Andreas Schwab <schwab@redhat.com>:
> daniel tian <daniel.xntian@gmail.com> writes:
>
>> I digged the write.c, and still being puzzled. The difficulty I met is
>> how I could distinguish those two labels  being in funpointer (data
>> section) from like "JMP @a" (in .text section, because in jmp section,
>> there will be a fixup, and it will be resolved in md_apply_fix, the
>> "a" address value will be right shifted two bit in this function.).
>> and I noticed "a, c"  reloc_type is BFD_RELOC_32 (TC_CONS_FIX_NEW is
>> not defined in my target.), So I wonder whether I chould identitify
>> "a,c" with BFD_RELOC_32.
>
> Qualify the expression by using something like a@fptr and attach a
> different relocation type to it.  See the powerpc port for an example.
>
> Andreas.

hi, Andreas:
  I checked the as-internal document, there is no something like fptr
stuff, or this is a self defined thing to handle some special
relocation type, right?
Because I don't know the powerpc assemble language, do you mind to
give me a little more detail about @fptr thing?

Thanks very much.

Re: Relocation data value adjust

[Alan Modra]

On Tue, May 25, 2010 at 11:22:55AM +0800, daniel tian wrote:
> 2010/5/21 Andreas Schwab <schwab@redhat.com>:
> > Qualify the expression by using something like a@fptr and attach a
> > different relocation type to it.  See the powerpc port for an example.
> 
> Because I don't know the powerpc assemble language, do you mind to
> give me a little more detail about @fptr thing?

Have a look at gas/config/tc-ppc.c:ppc_elf_suffix and the places that
this function is called.

-- 
Alan Modra
Australia Development Lab, IBM