Richard,

On Wed, 2010-11-24 at 09:27 +0000, Richard Sandiford wrote:
> Matthias Klose <doko@ubuntu.com> writes:
> > On 23.11.2010 17:32, Tristan Gingold wrote:
> >> Hi,
> >>
> >> here is the result of a first pre-test.  I still plan to update the target list.
> >>
> >> Two comments:
> >>
> >> 1) are the arm-eabi failures expected ?
> >
> > I see these too, these seem to be introduced after 20101028.
> 
> They were (rightly, IMO) introduced by the fix for ld/12001: we now
> complain if a symbol is defined by both -defsym and an input file.
> The patch below seems to work on both arm-linux-gnueabi and arm-eabi.
> Matthew, could you sanity-check it?  I realise these -defsyms must
> be there for a reason, so I'm probably missing something, sorry.

I agree that the tests are wrong and the fix for ld/12001 is correct.

The -defsyms were in the tests originally to ensure that the code was
put at particular locations to make sure we generated the correct
relocation veneers when necessary.

> Richard
> 
> ld/testsuite/
> 	* ld-arm/arm-elf.exp (armeabitests): Remove --defsym argument
> 	from jump-reloc-veneers* tests.
> 
> Index: ld/testsuite/ld-arm/arm-elf.exp
> ===================================================================
> --- ld/testsuite/ld-arm/arm-elf.exp	2010-11-24 08:58:57.000000000 +0000
> +++ ld/testsuite/ld-arm/arm-elf.exp	2010-11-24 09:11:10.000000000 +0000
> @@ -464,19 +464,19 @@ set armeabitests {
>       "farcall-data"}
>  
>      {"R_ARM_THM_JUMP24 Relocation veneers: Short 1" 
> -     "-defsym _start=0x8000 --section-start destsect=0x00009000" 
> +     "--section-start destsect=0x00009000" 
>       "-march=armv7-a -mthumb" 
>       {jump-reloc-veneers.s}
>       {{objdump -d jump-reloc-veneers-short1.d}}
>       "jump-reloc-veneers-short1"}
>      {"R_ARM_THM_JUMP24 Relocation veneers: Short 2" 
> -     "-defsym _start=0x8000 --section-start destsect=0x00900000" 
> +     "--section-start destsect=0x00900000" 
>       "-march=armv7-a -mthumb" 
>       {jump-reloc-veneers.s}
>       {{objdump -d jump-reloc-veneers-short2.d}}
>       "jump-reloc-veneers-short2"}
>      {"R_ARM_THM_JUMP24 Relocation veneers: Long" 
> -     "-defsym _start=0x8000 --section-start destsect=0x09000000" 
> +     "--section-start destsect=0x09000000" 
>       "-march=armv7-a -mthumb" 
>       {jump-reloc-veneers.s}
>       {{objdump -d jump-reloc-veneers-long.d}}

The current default ld script lays the code out as desired for the
tests, and so your proposed change is correct.  But in an effort to keep
these tests future proof I suggest that we force the '.text' section to
start at 0x8000 on the command line as well as the 'destsect' section.
I have attached a patch which does this.  

Any comments welcome, and if there are no issues can someone please
approve.

Thanks,

Matt

ld/testsuite/
	* ld-arm/arm-elf.exp (armeabitests): Replace --defsym argument
	in jump-reloc-veneers* tests with --section-start .text=0x8000.

-- 
Matthew Gretton-Dann
Principal Engineer - PDSW Tools
ARM Ltd