Re: Etherboot 5.5

On Sat, 21 May 2005, Georg Baum wrote:
> > Your gcc versions > 3.2.3 are suffering from the same problem with
> > unnamed struct elements as Tim; I will fix these in the morning, and
> > then look at the pre-3.2.3 builds.
> Meanwhile I have a fix for gcc 3.3, 3.4 and 4.0, it is attached. It is 
> ready to be applied IMO, with one exception:
> gcc 4.0 does not allow casts on lvalues. Therefore I got rid of them at 
> several places. The change in src/proto/http.c leads to cleaner code 
> actually, but the other changes are not optimal: I am not sure whether 
> the changes in src/drivers/net/natsemi.c and src/drivers/net/sis900.c are 
> correct, and the change in src/proto/nmb.c should work but is pretty 
> ugly.

Thanks.  I've applied a (very) slightly modified version of this patch.

> Unfortunately I get now an internal linker error:
> 
> $ make bin/e1000.dsk
> ld -N  -T arch/i386/scripts/i386.lds  -u obj_dskprefix --defsym 
> check_obj_dskprefix=obj_dskprefix   -u obj_e1000 --defsym 
> check_obj_e1000=obj_e1000   -u obj_config --defsym 
> check_obj_config=obj_config   bin/blib.a -o bin/e1000.dsk.tmp \
>         -Map bin/e1000.dsk.tmp.map
> ld: internal error ../../ld/ldlang.c 1974
> make: *** [bin/e1000.dsk.tmp] Error 1
> 
> $ ld -v
> GNU ld version 2.15
> 
> Do I need a newer version?

Shouldn't do; I'm using 2.13.90.0.18.

This error occurs in the following portion of code in ldlang.c:

  /* Return the iteration when the definition of NAME was last updated.  A
     value of -1 means that the symbol is not defined in the linker script
     or the command line, but may be defined in the linker symbol table.  */

  int
  lang_symbol_definition_iteration (const char *name)
  {
    struct lang_definedness_hash_entry *defentry
      = (struct lang_definedness_hash_entry *)
      bfd_hash_lookup (&lang_definedness_table, name, FALSE, FALSE);

    /* We've already created this one on the presence of DEFINED in the
       script, so it can't be NULL unless something is borked elsewhere in
       the code.  */
    if (defentry == NULL)
=====>  FAIL (); <======

    return defentry->iteration;
  }


So, it looks as though something is borked in ld.  You could try the 
attached patch to arch/i386/scripts/i386.lds, which temporarily removes 
all uses of DEFINED() in the linker script.  If this fixes the problem, 
then we're closer to identifying what's wrong with ld.

Michael

Index: arch/i386/scripts/i386.lds
===================================================================
RCS file: /cvsroot/etherboot/etherboot/etherboot-5.5/src/arch/i386/scripts/i386.lds,v
retrieving revision 1.10
diff -u -r1.10 i386.lds
--- arch/i386/scripts/i386.lds	27 Apr 2005 12:24:39 -0000	1.10
+++ arch/i386/scripts/i386.lds	23 May 2005 23:59:03 -0000
@@ -63,7 +63,7 @@
      * The prefix
      */
 
-    _prefix_link_addr = DEFINED ( _prefix_link_addr ) ? _prefix_link_addr : 0;
+    _prefix_link_addr = 0;
     . = _prefix_link_addr;
     _prefix = .;
 
@@ -80,8 +80,7 @@
      * The decompressor (may be absent)
      */
 
-    _decompress_link_addr = DEFINED ( _decompress_link_addr ) ?
-			      _decompress_link_addr : 0;
+    _decompress_link_addr = 0;
     . = _decompress_link_addr;
     _decompress = .;
 
@@ -97,7 +96,7 @@
      * The text sections
      */
 
-    _text_link_addr = DEFINED ( _text_link_addr ) ? _text_link_addr : 0;
+    _text_link_addr = 0;
     . = _text_link_addr;
     _text = .;
 
@@ -126,7 +125,7 @@
      * The data sections
      */
 
-    _data_link_addr = DEFINED ( _data_link_addr ) ? _data_link_addr : .;
+    _data_link_addr = .;
     . = _data_link_addr;
     _data = .;
 
@@ -179,8 +178,8 @@
      * location counter.
      */
 
-    _max_align		    = DEFINED ( _max_align ) ? _max_align : 16;
-    _load_addr		    = DEFINED ( _load_addr ) ? _load_addr : 0;
+    _max_align		    = 16;
+    _load_addr		    = 0;
 
     .			    = _load_addr;