[Bug ld/11845] relative constant confusion

[amodra at gmail dot com]

------- Additional Comments From amodra at gmail dot com  2010-07-28 02:34 
-------
There is no problem with the MAX operator as far as I can tell.  Your source of
confusion is twofold.

1)  A constant like 0x1000 inside an ouput section statement is not an absolute
constant, but rather a section relative offset.  So 0x1000 really means <output
section vma>+0x1000.  To evaluate MAX(0x1000,ABSOLUTE(.)) ld first converts the
0x1000 to an absolute value, adding the output section vma.  The result of MAX()
is then an absolute address, so the net effect of your expression is to make the
size of your output section at least 0x1000.  I imagine this isn't what you
require.  The comparison in your alternative conditional expression is evaluated
the same way, so also doesn't do exactly what you want.

2) ld muddies the waters even more in the way the result section of expressions
is calculated.  As I said above, the result of your MAX() is absolute, so
unsurprisingly, the rhs of your expression involving MAX() is absolute. 
However, in the conditional expression the result of 0x1000 - ABSOLUTE(.) is a
section relative offset.  This offset is then added to '.' without first being
converted to absolute.  That oddity is why your second expression almost works.

I don't know what to do about these problems.  Any change in ld script behaviour
will likely annoy more people than it pleases..

You can work around this using something like the following.

SECTIONS
{
  . = 0x500;
  ABS1K = 0x1000;
  .text :
  {
    *(.text)
    . += MAX(ABS1K, ABSOLUTE(.)) - ABSOLUTE(.);
  }
}



-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|MAX operator looks buggy,   |relative constant confusion
                   |returning always the lhs    |
                   |value                       |


http://sourceware.org/bugzilla/show_bug.cgi?id=11845

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.