Re: PATCH: PR ld/4504: Copy relocation doesn't preserve section alignment

["H. J. Lu" <hjl at lucon dot org>]

On Wed, May 16, 2007 at 11:50:42AM +0930, Alan Modra wrote:
> On Tue, May 15, 2007 at 06:46:20AM -0700, H. J. Lu wrote:
> > +  orig_power_of_two = bfd_get_section_alignment (sec->owner, sec);
> > +
> > +  if (orig_power_of_two > power_of_two)
> > +    {
> > +      /* Adjust the section alignment if needed.  */
> > +      if (! bfd_set_section_alignment (dynbss->owner, dynbss,
> > +				       orig_power_of_two))
> > +	return FALSE;
> > +    }
> > +
> > +  /* We make sure that the symbol will be aligned properly.  Since we
> > +     don't know its alignment requirement, we start with the maximum
> > +     alignment and check low bits of the symbol address for the
> > +     minimum alignment.  */
> > +  mask = ((bfd_vma) 1 << orig_power_of_two) - 1;
> > +  while ((h->root.u.def.value & mask) != 0)
> > +    {
> > +       mask >>= 1;
> > +       --orig_power_of_two;
> > +    }
> 
> You didn't put this loop where I suggested, before setting .dynbss
> alignment.  Why?  There is no need to align .dynbss more than the
> required symbol alignment.
> 

Like this?


H.J.
---
2007-05-15  H.J. Lu  <hongjiu.lu@intel.com>

	* elflink.c (_bfd_elf_adjust_dynamic_copy): Align dynamic bss
	section to the minimum alignment.

--- bfd/elflink.c.foo	2007-05-15 20:53:44.000000000 -0700
+++ bfd/elflink.c	2007-05-15 21:00:23.000000000 -0700
@@ -2656,33 +2656,33 @@ bfd_boolean
 _bfd_elf_adjust_dynamic_copy (struct elf_link_hash_entry *h,
 			      asection *dynbss)
 {
-  unsigned int power_of_two, orig_power_of_two;
+  unsigned int power_of_two;
   bfd_vma mask;
   asection *sec = h->root.u.def.section;
 
   /* The section aligment of definition is the maximum alignment
-     requirement of symbols defined in the section.  */
-  power_of_two = bfd_get_section_alignment (dynbss->owner, dynbss);
-  orig_power_of_two = bfd_get_section_alignment (sec->owner, sec);
+     requirement of symbols defined in the section.  Since we don't
+     know the symbol alignment requirement, we start with the
+     maximum alignment and check low bits of the symbol address
+     for the minimum alignment.  */
+  power_of_two = bfd_get_section_alignment (sec->owner, sec);
+  mask = ((bfd_vma) 1 << power_of_two) - 1;
+  while ((h->root.u.def.value & mask) != 0)
+    {
+       mask >>= 1;
+       --power_of_two;
+    }
 
-  if (orig_power_of_two > power_of_two)
+  if (power_of_two > bfd_get_section_alignment (dynbss->owner,
+						dynbss))
     {
       /* Adjust the section alignment if needed.  */
       if (! bfd_set_section_alignment (dynbss->owner, dynbss,
-				       orig_power_of_two))
+				       power_of_two))
 	return FALSE;
     }
 
-  /* We make sure that the symbol will be aligned properly.  Since we
-     don't know its alignment requirement, we start with the maximum
-     alignment and check low bits of the symbol address for the
-     minimum alignment.  */
-  mask = ((bfd_vma) 1 << orig_power_of_two) - 1;
-  while ((h->root.u.def.value & mask) != 0)
-    {
-       mask >>= 1;
-       --orig_power_of_two;
-    }
+  /* We make sure that the symbol will be aligned properly.  */
   dynbss->size = BFD_ALIGN (dynbss->size, mask + 1);
 
   /* Define the symbol as being at this point in DYNBSS.  */