Re: 16-bit relocations in 32-bit code

[Anders Norlander <ano at bahnhof dot se>]

What about something like this?
This is all constant expressions requiring no relocations,
regardless of the actual addresses.
Assuming it's all in the same section and file of course.

tab:
	.short PLUS-START
	.short MINUS-START
	.short TIMES-START
...
START:
	# index by edi and add START address to offset
	movzwl tab(%edi,%edi),%edi
	addl $START,%edi
	jmp *%edi

PLUS:
MINUS:
TIMES:

-Anders

On 2010-03-15, at 23:43 , Andrew Zabolotny wrote:

> From Mon, 15 Mar 2010 12:44:26 -0700
> Richard Henderson <rth@redhat.com> wrote:
> 
>> There is no solution that is generally portable.
> I understand, but I can live with platform-specific macros.
> 
>> However, given the existance of R_386_PC16, you should be able to
>> represent ".short SYM-." for any SYM on both i386 and x86_64.
> I don't quite get the idea. ".short SYM-." won't generate any
> relocation at all, how this can help me? But I tried a simple example
> like this:
> 
>     .reloc  .,R_386_PC16,PLUS
>     .short  0
> 
> and this at least does not generate the linker warning about overflow.
> Can I turn my initial statement:
> 
> 	.short	PLUS-0x610000
> 
> into a PC16 relocation somehow?
> 
> Some platforms support the modificator lo16(addr). This is the ideal
> solution for my problem; unfortunately, i386 does not support it.
> 
>> For portability to other machines, you might consider writing
>> your own trivial assembler and linker for the forth.  You
>> can then include the linked forth program into the native 
>> binary via .incbin or the like.
> That's what I'm trying to avoid by using binutils as the swiss army
> knife :)
> 
> -- 
> Andrew