Re: [PATCH 02/02] Elf: Align elf notes properly

["Magnus Damm" <magnus.damm@gmail.com>]

On 11/10/06, Horms <horms@verge.net.au> wrote:
> On Thu, Nov 09, 2006 at 07:00:22AM -0700, Eric W. Biederman wrote:
> > Magnus Damm <magnus@valinux.co.jp> writes:
> >
> > > elf: Align elf notes properly
> > >
> > > The kernel currently contains several elf note aligment implementations. Most
> > > implementations follow the spec on 32-bit platforms, but none current aligns
> > > the notes correctly on 64-bit platforms. This patch tries to fix this by
> > > interpreting the 64-bit and 32-bit elf specs as the following:
> > >
> > > offset bytes name
> > > 0      4     n_namesz -+                  -+
> > > 4      4     n_descsz  | elf note header   |
> > > 8      4     n_type   -+                   | elf note entry size - N4
> > > 12     N1    name                          |
> > > N2     N3    desc                         -+
> > >
> > > WS = word size in bytes (4 for 32 bit, 8 for 64 bit)
> > > N1 = roundup(n_namesz + sizeof(elf note header), WS) - sizeof(elf note header)
> > > N2 = sizeof(elf note header) + N1
> > > N3 = roundup(n_descsz, WS)
> > > N4 = sizeof(elf note header) + N1 + N2
> > >
> > > The elf note header contains three 32-bit values on 32-bit and 64-bit systems.
> > > The header is followed by name and desc data together with padding. The
> > > alignment and padding varies depending on the word size.
> >
> > I see your point and I disagree.  The notes in a kernel generated
> > core dump do not vary in size.  Find me some implementation evidence that
> > anyone ever added the extra 4 bytes of alignment to the description and the
> > padding fields and I will be ready to consider this.  Currently this
> > just appears to be reading a draft spec that doesn't match reality.
>
> Or perhaps a spec that hasn't been implemented correctly.
> I guess that the real question is, what padding is correct?

I see no point in aligning to 32-bit boundaries on 64-bit platforms.
Their intention was most likely to align to the word size IMO, so this
is most likely a "thinko" left over from whoever ported the code from
32-bit to 64-bit.

How we chose to align on 64-bit platforms is another issue, either we
fix it soon or live with the fact that we are not following the 64-bit
draft spec. Either way is fine with me.

/ magnus