RFA: Small fix to elf-eh-frame.c for MIPS ELF64

RFA: Small fix to elf-eh-frame.c for MIPS ELF64

[Richard Sandiford]

_bfd_elf_discard_section_eh_frame has the following code to handle
the personality data:

		      ENSURE_NO_RELOCS (buf);
		      /* Ensure we have a reloc here, against
			 a global symbol.  */
		      if (GET_RELOC (buf) != NULL)
			{
                          ...
			  cookie->rel++;
			}

This causes problems on MIPS because ELF64 composite relocations are
represented as a sequence of normal relocations against the same address.
There might be more than one relocation to skip here.

One fix would be to replace:

   cookie->rel++;

with:

   SKIP_RELOCS (buf + 1);

This is safe because of the ENSURE_NO_RELOCS() line quoted above,
and I'd be happy to write it like that if preferred.  I thought the
attached patch was slightly more readable and self-contained though.

The testcase is in the main patch that I'm about to post (which also
mentions the testing procedure).  I've split this bit out because it
wasn't really related to the rest.  OK to install?

Richard


bfd/
	* elf-eh-frame.c (_bfd_elf_discard_section_eh_frame): Deal with
	composite relocations against the personality data.

--- bfd/elf-eh-frame.c.orig	Fri Oct  8 13:40:53 2004
+++ bfd/elf-eh-frame.c	Sat Nov 13 09:41:56 2004
@@ -499,7 +499,10 @@ _bfd_elf_discard_section_eh_frame
 
 			      cie.personality = h;
 			    }
-			  cookie->rel++;
+			  /* Cope with MIPS-style composite relocations.  */
+			  do
+			    cookie->rel++;
+			  while (GET_RELOC (buf) != NULL);
 			}
 		      buf += per_width;
 		    }

Re: RFA: Small fix to elf-eh-frame.c for MIPS ELF64

[Alan Modra]

On Sat, Nov 13, 2004 at 12:01:58PM +0000, Richard Sandiford wrote:
> 	* elf-eh-frame.c (_bfd_elf_discard_section_eh_frame): Deal with
> 	composite relocations against the personality data.

OK.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre