Re: LD 2.20 absolute / relative symbol assignment issue

[Nick Clifton <nickc at redhat dot com>]

Hi Mark,

> I have two issues:
>
> 1) In most cases the values assigned to symbols such as __EXAMPLE_START
> are absolute ones whether I use ABSOLUTE() or not; I cannot get
> section-relative symbol values.

I think that this is a documentation problem.  Essentially I think that 
your problems lie in the difference between a relative link and a final 
link.  For example consider this script:

  SECTIONS
  {
    . = 0x1000 ;

    sym1 = . ;
    sym2 = ABSOLUTE (.);

    .foo :
    {
       sym3 = .;
       sym4 = ABSOLUTE (.);

       LONG (1);

       sym5 = .;
       sym6 = ABSOLUTE (.);

       . += 0x100;

       sym7 = .;
       sym8 = ABSOLUTE (.);
    }

    sym9 = .;
    sym10 = ABSOLUTE (.);

    /DISCARD/ : { *(*) }
  }

If you run the following commands:

  ld -T script.t foo.o

    (You have to specify at least one input file, even
    though it is going to be discarded).

  readelf --syms a.out

Then you will get output like this:

   Num:    Value  Size Type    Bind   Vis      Ndx Name
     0: 00000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 00000000     0 SECTION LOCAL  DEFAULT    1
     2: 00001004     0 NOTYPE  GLOBAL DEFAULT  ABS sym6
     3: 00001000     0 NOTYPE  GLOBAL DEFAULT  ABS sym1
     4: 00001000     0 NOTYPE  GLOBAL DEFAULT  ABS sym4
     5: 00001104     0 NOTYPE  GLOBAL DEFAULT  ABS sym10
     6: 00001104     0 NOTYPE  GLOBAL DEFAULT  ABS sym9
     7: 00000000     0 NOTYPE  GLOBAL DEFAULT    1 sym3
     8: 00000104     0 NOTYPE  GLOBAL DEFAULT    1 sym7
     9: 00001104     0 NOTYPE  GLOBAL DEFAULT  ABS sym8
    10: 00001000     0 NOTYPE  GLOBAL DEFAULT  ABS sym2
    11: 00000004     0 NOTYPE  GLOBAL DEFAULT    1 sym5

Note how symbols 3, 5 and 7 are section based symbols whereas all of the 
rest are absolute symbols.  The absolute symbols have the values that 
you would expect, but the section-relative ones do not.  But try these 
commands next:

  ld -T script.t -r foo.o
  readelf --syms a.out

     0: 00000000     0 NOTYPE  LOCAL  DEFAULT  UND
     1: 00000000     0 SECTION LOCAL  DEFAULT    1
     2: 00001004     0 NOTYPE  GLOBAL DEFAULT  ABS sym6
     3: 00001000     0 NOTYPE  GLOBAL DEFAULT  ABS sym1
     4: 00001000     0 NOTYPE  GLOBAL DEFAULT  ABS sym4
     5: 00001104     0 NOTYPE  GLOBAL DEFAULT  ABS sym10
     6: 00001104     0 NOTYPE  GLOBAL DEFAULT  ABS sym9
     7: 00000000     0 NOTYPE  GLOBAL DEFAULT    1 sym3
     8: 00000104     0 NOTYPE  GLOBAL DEFAULT    1 sym7
     9: 00001104     0 NOTYPE  GLOBAL DEFAULT  ABS sym8
    10: 00001000     0 NOTYPE  GLOBAL DEFAULT  ABS sym2
    11: 00000004     0 NOTYPE  GLOBAL DEFAULT    1 sym5

Now the section-relative symbols have the values that you might expect. 
 The reason is that the "-r" command line switch instructs the linker 
to perform a relocatable link.  This is one where the final location of 
output sections is not fixed, and so section relative symbols cannot be 
fully evaluated.

Once a final link is performed (one without the -r command line switch) 
the location of all output sections is fixed and the values of all 
symbols, section based or absolute, can be computed.

Now I agree that reading the linker documentation gives the impression 
that section-relative symbols always contain values that are relative to 
the start of their containing section.  But this is not true.  It only 
applies before a final link has been invoked.

Does this make sense ?  If so then I will try to come up with a 
documentation patch to explain the situation more clearly.


> 2) In some cases the use of the ABSOLUTE() keyword actually results in
> the section-relative value being assigned!

This worries me - can you give an example of when this happens ?

Cheers
  Nick