Re: cross-compiling LM32 internal error while linking

[Federico Vaga <federico dot vaga at gmail dot com>]

I think I got the source of the problem. I was writing the
simple testcase that you asked me:


#include <stdlib.h>

int main(int argc, char *argv[])
{
	exit(0);
}


/bin/lm32-elf-gcc -c -o wrn-crt0.o wrn-crt0.S
/bin/lm32-elf-gcc -c -o rt-mturtle.o rt-mturtle.c
/bin/lm32-elf-gcc -c -o gcclm32-bug.o gcclm32-bug.c
/bin/lm32-elf-gcc -o gcclm32-bug.elf -nostartfiles wrn-crt0.o rt-mturtle.o gcclm32-bug.o -T wrnode.ld -lgcc -lc
/lib/gcc/lm32-elf/5.3.0/../../../../lm32-elf/bin/ld: internal error: addend should be zero for R_LM32_16_GOT
/lib/gcc/lm32-elf/5.3.0/../../../../lm32-elf/bin/ld: internal error: addend should be zero for R_LM32_16_GOT



In the file rt-mturtle.c I have my syscall stubs in order to avoid errors like:

newlib-2.2.0-1/newlib/libc/stdlib/exit.c:70: undefined reference to `_exit'



The problem is with the errno 'variable'. According to:

https://sourceware.org/newlib/libc.html#Stubs

for function like fork(), I followed the example:

#include <errno.h>
#undef errno
extern int errno;
int fork(void) {
 errno = EAGAIN;
 return -1;
}



The errno trick of undefining the macro and set the an external variable causes
the error:

/lib/gcc/lm32-elf/5.3.0/../../../../lm32-elf/bin/ld: internal error: addend should be zero for R_LM32_16_GOT


Of course. Now I'm able to compile without errors if I avoid to use errno. But
I do not thing is the right(tm) solution.

On Fri, Feb 05, 2016 at 11:23:53AM +0000, Nick Clifton wrote:
> Hi Federico,
>
> > When I compile my code I get this message:
> >
> > ld: internal error: addend should be zero for R_LM32_16_GOT
> >
> > Then the compilation complete without errors (O_O)
> > (like if that error is not an error but a warning)
>
> This is a bug.  Either the code being linked is broken somehow, possibly
> due to a compiler or assembler error, and so the link should not have
> succeeded.  Or else the error message is actually a warning about behaviour
> that the programmer (of the linker) was not expecting, and so the linker
> might not have behaved correctly.
>
> A simple testcase to reproduce the problem would certainly be helpful for
> whomever does investigate this problem.
>
>
> > I do not understand what does this mean, and I cannot find information on
> > the web. Any hint?
>
> Well the error message comes from the lm32_elf_relocate_section function in the
> bfd/elf32-lm32.c source file, where it is processing the R_LM32_16_GOT reloc.
>
> Exactly what this relocation is supposed to do and why its addend should be zero
> I do not know.  It is presumably specified by the ABI for the LM32 architecture,
> but I could not find a copy of this document on the web. :-(
>
>
> > If this is not the right place, can you redirect me to the
> > right place?
>
> Probably the best person to ask would be the maintainer for the LM32 code -
> Jon Beniston.  I have added him to the CC list of this email.
>
> Cheers
>  Nick