PATCH: Don't allow ia64 unwind section to point to section in different files

PATCH: Don't allow ia64 unwind section to point to section in different files

[H. J. Lu]

On Fri, May 13, 2005 at 02:05:56PM -0700, David Mosberger wrote:
> >>>>> On Fri, 13 May 2005 14:01:11 -0700, "H. J. Lu" <hjl@lucon.org> said:
> 
>   >> However, there still seems to be a binutils issues here: if this
>   >> is something binutils cannot properly support, it should issue an
>   >> error, not silently generate wrong code, no?
> 
>   HJ> I will see what I can do.
> 

When weak functions are used on ia64, part of the unwind section may
point to the strong definition in a different file. This will lead to
wrong unwind info. Basically, on ia64, we have to use comdat to get the
right unwind info. This patch will check it.


H.J.
---
2005-05-13  H.J. Lu  <hongjiu.lu@intel.com>

	* elfxx-ia64.c (elfNN_ia64_relocate_section): Don't allow
	unwind section to point to section in different files.

--- bfd/elfxx-ia64.c.weak	2005-05-13 10:47:56.000000000 -0700
+++ bfd/elfxx-ia64.c	2005-05-13 14:44:52.000000000 -0700
@@ -4151,6 +4151,27 @@ elfNN_ia64_relocate_section (output_bfd,
       value += rel->r_addend;
       dynamic_symbol_p = elfNN_ia64_dynamic_symbol_p (h, info, r_type);
 
+      /* The unwind section and the corresponding text section have
+	 to come from the same file.  When the strong function overides
+	 a weak function, part of the unwind section may point to the
+	 wrong place.  */
+      if (h
+	  && (h->root.type = bfd_link_hash_defined
+	      || h->root.type = bfd_link_hash_defweak)
+	  && (elf_section_data (input_section)->this_hdr.sh_type
+	      = SHT_IA_64_UNWIND)
+	  && input_section->owner != h->root.u.def.section->owner)
+	{
+	  (*_bfd_error_handler)
+	    (_("%B: unwind section `%A' points to symbol `%s' defined in section '%A' in %B"),
+	     input_bfd, input_section, h->root.u.def.section,
+	     h->root.u.def.section->owner,
+	     h ? h->root.root.string
+	       : bfd_elf_sym_name (input_bfd, symtab_hdr, sym, sym_sec));
+	  ret_val = FALSE;
+	  continue;
+	}
+
       switch (r_type)
 	{
 	case R_IA64_NONE:

Re: PATCH: Don't allow ia64 unwind section to point to section in different files

[H. J. Lu]

On Fri, May 13, 2005 at 02:46:12PM -0700, H. J. Lu wrote:
> On Fri, May 13, 2005 at 02:05:56PM -0700, David Mosberger wrote:
> > >>>>> On Fri, 13 May 2005 14:01:11 -0700, "H. J. Lu" <hjl@lucon.org> said:
> > 
> >   >> However, there still seems to be a binutils issues here: if this
> >   >> is something binutils cannot properly support, it should issue an
> >   >> error, not silently generate wrong code, no?
> > 
> >   HJ> I will see what I can do.
> > 
> 
> When weak functions are used on ia64, part of the unwind section may
> point to the strong definition in a different file. This will lead to
> wrong unwind info. Basically, on ia64, we have to use comdat to get the
> right unwind info. This patch will check it.
> 
> 

Well, it isn't very useful. The problem is libc.a has many weak
functions and user can override them. My patch will make it impossible
on ia64. I guess we may have to live with the imperfect unwind info
when weak functions are used. The unwinder may have to deal with it.


H.J.

Re: PATCH: Don't allow ia64 unwind section to point to section in different files

[David Mosberger]

>>>>> On Fri, 13 May 2005 14:58:15 -0700, "H. J. Lu" <hjl@lucon.org> said:

  HJ> Well, it isn't very useful. The problem is libc.a has many weak
  HJ> functions and user can override them. My patch will make it
  HJ> impossible on ia64. I guess we may have to live with the
  HJ> imperfect unwind info when weak functions are used. The unwinder
  HJ> may have to deal with it.

Huh?  How do you propose it deal with it?

Wrong unwind info will break exception handling.  I don't think that's
acceptable.

	--david

Re: PATCH: Don't allow ia64 unwind section to point to section in different files

[H. J. Lu]

On Mon, May 16, 2005 at 02:26:36AM -0700, David Mosberger wrote:
> >>>>> On Fri, 13 May 2005 14:58:15 -0700, "H. J. Lu" <hjl@lucon.org> said:
> 
>   HJ> Well, it isn't very useful. The problem is libc.a has many weak
>   HJ> functions and user can override them. My patch will make it
>   HJ> impossible on ia64. I guess we may have to live with the
>   HJ> imperfect unwind info when weak functions are used. The unwinder
>   HJ> may have to deal with it.
> 
> Huh?  How do you propose it deal with it?
> 
> Wrong unwind info will break exception handling.  I don't think that's
> acceptable.

Since the weak function is still in the executable, its unwind info is
also there. That is what you see in kernel.


H.J.