Re: IPv4 NAT doesn't compile in 2.4.4

[Russell King]

On Sat, Apr 28, 2001 at 02:07:53PM -0700, David S. Miller wrote:
> Why would ip_nat_cleanup() be removed by the linker?

Because we explicitly tell the linker to drop all code marked as
__exit:

#define __exit          __attribute__ ((unused, __section__(".text.exit")))

  /* Sections to be discarded */
  /DISCARD/ : {
        *(.text.exit)
        *(.data.exit)
        *(.exitcall.exit)
        }

>  All the "unused"
> attribute should do is shut up gcc if the thing is marked static yet
> not called.  The GCC manual even states "... means that the function
> is meant to be possibly unused.  GNU CC will not produce a warning
> for this function."  It makes no mention of any effect on the actual
> code output, or that the linker will delete it.

I quote from the ld info pages:

   The special output section name `/DISCARD/' may be used to discard
input sections.  Any input sections which are assigned to an output
section named `/DISCARD/' are not included in the output file.

> It doesn't remove the function on any platform I could test this on.

Try x86.

> If the linker removed it, why did it give a relocation truncation
> error instead of a missing symbol error?  And more importantly, what
> specifically was the reason that the linker removed the function on
> ARM, what made this happen?

Architecture independent linker behaviour.

> Please explain this in detail so we don't have to guess as I have
> seen no other report of this.

The reason it shows up on ARM is because the relocation is not 32-bit
long, and therefore the relocation it is trying to encode generates
an error.

(I'm presuming that it was allocating the symbol an address of zero,
but I haven't checked this since trying to call a non-present section
seems a bit stupid to start with).

--
Russell King ([e-mail])                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html