Re: Fix relro strip test for MIPS

[Daniel Jacobowitz <drow at false dot org>]

On Wed, Sep 19, 2007 at 12:05:54PM -0400, Daniel Jacobowitz wrote:
> There are two unique things about MIPS which conspire to fail the
> ld -z relro -shared tests (which test that objcopy and strip do not
> change the binary).  One is that .dynamic is writable and there is no
> .got.plt section, so a typical shared library has no PT_GNU_RELRO;
> it gets added and then converted to PT_NULL.  The other is that it
> adds a second PT_NULL program header for prelinker support.
> These combine to produce an input file with two PT_NULL segments.
> There's a FIXME in copy_elf_program_header above some code which
> seems clearly wrong for this case.
> 
> Deleting that turns up a second problem.  The input file to strip
> has a PT_NULL segment with alignment of zero (I believe this is the
> aborted PT_GNU_RELRO segment, but it might have been the MIPS-specific
> one; I'm not completely sure).  We were overriding its input alignment
> and I don't see any reason to do so.
> 
> With this patch two more tests pass on mips-linux.  Unless someone
> sees a problem with them, I will apply them in a few days.

I agree with H.J. that we should not generate the extra PT_NULL
segment, and I'm keeping an eye on his patch to prevent that
(thanks!).  Meanwhile, there were no concerns about this patch
for copying odd binaries; I've checked it in.

-- 
Daniel Jacobowitz
CodeSourcery