Re: DWARF 2 sections and padding ?

[Brian Ford <ford at vss dot fsi dot com>]

On Fri, 21 Feb 2003, Daniel Jacobowitz wrote:

> On Fri, Feb 21, 2003 at 10:16:49AM -0600, Brian Ford wrote:
> > I am trying to add DWARF 2 support for PE/COFF on Cygwin.  PE/COFF
> > sections must be aligned to 512 byte boundaries.  This extra padding
> > confuses the DWARF 2 parsing routines because they use the sections
> > raw_size, thus parsing an invalid compilation unit header (all
> > zeros).
> >
> > The actual, smaller size is available in the section header.  I am
> > confused about what the cooked_size is supposed to mean.  It is currently
> > zero, but I think it could contain the smaller virtual size.
> >
> > I have read the DWARF spec several times, and although it discusses this
> > issue briefly for some debugging sections, it is unclear to me how this
> > should really work.  Unless I try to arrange for the producer (gcc) to
> > pad the compilation units with zeros instead of having bfd pad the end of
> > the sections, I don't know how to proceed.
> >
> > I would like to just use an alternate size instead of the raw size (cooked
> > maybe?) but I don't know how this would affect other targets and I don't
> > want a PE specific hack in DWARF parsing code.
> >
> > Any ideas?  Thanks.
>
> I don't see why the raw size should include the padding; that should
> just be used when _placing_ sections.  But this is a binutils/bfd
> issue; redirected to that list.
>
The raw size appears to translate directly to the size in the section
header.  The PE/COFF format states this must be a multiple of the page
size.

Other COFF targets that define ALIGN_SECTIONS_IN_FILE do the same thing,
in separate code, though.  There is one case where these other targets do
what you suggest, if COFF_PAGE_SIZE is defined:

/* In demand paged files the low order bits of the file offset
   must match the low order bits of the virtual address.  */
#ifdef COFF_PAGE_SIZE
      if ((abfd->flags & D_PAGED) != 0
          && (current->flags & SEC_ALLOC) != 0)
        sofar += (current->vma - sofar) % page_size;
#endif

I modified the PE/COFF code to do this instead and, although it violates
the format, a simple test appeared valid and runable on Windows XP.  What
do the binutils maintainers think about making this change?

-- 
Brian Ford
Senior Realtime Software Engineer
VITAL - Visual Simulation Systems
FlightSafety International
Phone: 314-551-8460
Fax:   314-551-8444