Re: arm-wince-pe-ld

[Nick Clifton <nickc at redhat dot com>]

Hi Rainer,

>> > 	"Ignoring ARM_26D: sec %s secvma 0x%x "
>> 
>> The presence of these relocs is a known problem, which has not been
>> fixed yet.  It would be interesting to know why they are being
>> produced in the first place.
>
> What is the meaning of these relcos? If I knew, it would make it easier
> to find the problem.

They are a special form of the ARM_26 reloc which has already been
processed by the linker when it was performing a partial link.  See
the function aoutarm_fix_pcrel_26() in bfd/coff-arm.c.

The ARM_26 reloc is used when the assembler cannot compute the value
to insert into the offset field of a B, BL or BLX instruction.  It
tells the linker that it has to compute this offset once it knows
where the target symbol will be located.


> Is it possible that these are generated because the compiler generated
> code with -mapcs-26? I have not checked again after compiling with
> -mapcs-32. But my old gas did not create these relocs from the same
> assembler files...

The support for the arm-wince-pe target has bit-rotted over the last
few years, so it probably is a bug in the assembler of linker
somewhere.  A patch has been applied recently to gcc to make the
default for the arm-wince-pe target be -mapcs-32, but it was found
that the Thumb multilibs did not compile...


> We might come up with some changes for gcj for arm-wince-pe (libffi
> etc)  Can I send them to you for review/integration in the code
> base?

Well please send them to the appropriate mailing list (gcc, binutils,
gdb etc).  I may welln be the person who reviews them.  Do you have an
FSF copyright assignment in place ?

Cheers
        Nick