Re: elision-conf.c "__pthread_init_array causes a section type conflict"

[Dominik Vogt <vogt at linux dot vnet dot ibm dot com>]

On Mon, Jul 22, 2013 at 02:59:05PM -0700, Roland McGrath wrote:
> Use -save-temps and check the .i file for any other declarations using the
> same section attribute.  I suspect there are none and it's just that your
> compiler configuration uses .init_array directly, so it's internally
> generated a use of the section.  

I get some error message when compiling with -save-temps.  Some
gawk script generates input for gcc on stdin (input file name '-'),
and then at a later place in the makefiles a filename -.i is
generated from that and put on a gcc command line.  Gcc then fails
to interpret that as an option.  If you're interested in the error
message, I can try to reproduce it.

> Trunk GCC makes .init_array writable, which it really should be only in PIC
> code, where really it should be RELRO.  Probably the problem goes away if
> you drop the const in the SHARED case.  But really the compiler is just
> wrong and it should be happy with that declaration, possibly requiring you
> to add attribute_relro.
> 
> So, first just add attribute_relro and see if that works.  If so, then all
> is well.  If not, then you can try it with and without const and/or
> attribute_relro and see if you get a winner.  But if you can't declare it
> const attribute_relro and win then we should file a GCC bug.

Okay, removing const seems to work (I'm not sure the path with
SHARED undefined is ever taken, though):

  #ifdef SHARED
  # define INIT_SECTION ".init_array"
  # define MAYBE_CONST
  #else
  # define INIT_SECTION ".preinit_array"
  # define MAYBE_CONST const
  #endif

  void (*MAYBE_CONST __pthread_init_array []) (int, char **, char **)
    __attribute__ ((section (INIT_SECTION), aligned (sizeof (void *)))) =
  {
    &elision_init
  };

However I'm not sure where you wanted me to put attribute_relro;
should it _replace_ the section attribute?

Ciao

Dominik ^_^  ^_^

P.S.:  There's no need to explicitly cc me; the Reply-To in my
messages is set appropriately.

-- 

Dominik Vogt
IBM Germany