max 64 sections??

max 64 sections??

[Doug Evans]

I'm seeing an error in gdb where a symbol is being added to the
symbol table with a bogus address.

In debugging this, I found this in symfile.c:default_symfile_offsets

  objfile->num_sections = SECT_OFF_MAX;
  objfile->section_offsets = (struct section_offsets *)
    obstack_alloc (&objfile->psymbol_obstack, SIZEOF_SECTION_OFFSETS);
  memset (objfile->section_offsets, 0, SIZEOF_SECTION_OFFSETS);

The bogus address is coming from objfile->section_offsets->offsets[228].
SECT_OFF_MAX is only 64.

The elf file in question has _many_ sections.

So what's up?  Should objfile->section_offsets have been allocated
with slots for the same number of sections as in the objfile?

Or is this line in elfread.c:elf_symtab_read  wrong?

          offset = ANOFFSET (objfile->section_offsets, sym->section->index);

Here's where the bogus address gets computed, later in elf_symtab_read:

	      /* Relocate all non-absolute symbols by the section offset.  */
	      if (sym->section != &bfd_abs_section)
		{
		  symaddr += offset;
		}

Or have I overlooked something?

Re: max 64 sections??

[Elena Zannoni]

Doug Evans writes:
 > I'm seeing an error in gdb where a symbol is being added to the
 > symbol table with a bogus address.
 > 
 > In debugging this, I found this in symfile.c:default_symfile_offsets
 > 
 >   objfile->num_sections = SECT_OFF_MAX;
 >   objfile->section_offsets = (struct section_offsets *)
 >     obstack_alloc (&objfile->psymbol_obstack, SIZEOF_SECTION_OFFSETS);
 >   memset (objfile->section_offsets, 0, SIZEOF_SECTION_OFFSETS);
 > 
 > The bogus address is coming from objfile->section_offsets->offsets[228].
 > SECT_OFF_MAX is only 64.
 > 
 > The elf file in question has _many_ sections.
 > 
 > So what's up?  Should objfile->section_offsets have been allocated
 > with slots for the same number of sections as in the objfile?
 > 

Doug, 
Yes, you are not hallucinating. This is a long standing problem.
Feel free to enter a bug report in the gnat database.

Elena


 > Or is this line in elfread.c:elf_symtab_read  wrong?
 > 
 >           offset = ANOFFSET (objfile->section_offsets, sym->section->index);
 > 
 > Here's where the bogus address gets computed, later in elf_symtab_read:
 > 
 > 	      /* Relocate all non-absolute symbols by the section offset.  */
 > 	      if (sym->section != &bfd_abs_section)
 > 		{
 > 		  symaddr += offset;
 > 		}
 > 
 > Or have I overlooked something?