Re: ARM mapping symbols and --strip-unneeded

[Nick Clifton <nickc at redhat dot com>]

Hi Michael,

> So, is it reasonable to assume that ld relies on the mapping symbols
> when producing output for big-endian ARM

Yes.  See bfd/elf32-arm.c:elf32_arm_write_section() for more details.

> Based on this I'm wondering about the definition of "unneeded" (being:
> "Remove all symbols that are not needed for relocation processing."):
> Should the mapping symbols perhaps be considered needed and not
> stripped?

For big-endian ARM binaries, yes.

> In the meaintime: Is there another way than --preserve-symbol= to tell
> strip to preserve mapping symbols?

If you are able to add command line options to the build system then you
could try adding --no-strip-unneeded.  Alternatively you could add "@extra_stript_options"
and then in the file extra_strip_options have --preserve-symbol=$a etc.
That should avoid the shell processing problems.

(I would also argue that the build system is broken in that it is stripping
files too early.  But there is probably nothing that can be done about that).

As an alternative, you might like to try out the attached patch, which
should prevent the mapping symbols from being stripped in the first place.
Unfortunately this does have the drawback of making stripped big endian binaries
bigger than their little endian counterparts.

If this patch does work for you, and you are unable to use any of the workarounds
suggested above, then please could you open a bug report for this problem, so
that we have somewhere to record the patch and the problem that it is fixing.

Cheers
  Nick

diff --git a/bfd/elf32-arm.c b/bfd/elf32-arm.c
index a7964c1..1380a04 100644
--- a/bfd/elf32-arm.c
+++ b/bfd/elf32-arm.c
@@ -18266,6 +18266,25 @@ elf32_arm_copy_special_section_fields (const bfd *ibfd ATTRIBUTE_UNUSED,
   return FALSE;
 }
 
+/* If we are going to byteswap code then keep the mapping symbols.  Without
+   this "strip --strip-unneeded" will remove them and that would prevent
+   stripped binaries from being byteswapped correctly.  Unfortunately we do
+   not have access to link_info->globals->byteswap_code so we just test for
+   big endianness here.  */
+
+static void
+elf32_arm_backend_symbol_processing (bfd *abfd, asymbol *sym)
+{
+  if (bfd_big_endian (abfd)
+      && sym->name != NULL
+      && sym->section != bfd_abs_section_ptr
+      && (strcmp (sym->name, "$a") == 0
+	  || strcmp (sym->name, "$t") == 0
+	  || strcmp (sym->name, "$d") == 0))
+    sym->flags |= BSF_KEEP;
+}
+
+/* SPU ELF linker hash table.  */
 #undef  elf_backend_copy_special_section_fields
 #define elf_backend_copy_special_section_fields elf32_arm_copy_special_section_fields
 
@@ -18324,6 +18343,7 @@ elf32_arm_copy_special_section_fields (const bfd *ibfd ATTRIBUTE_UNUSED,
 #define elf_backend_begin_write_processing      elf32_arm_begin_write_processing
 #define elf_backend_add_symbol_hook		elf32_arm_add_symbol_hook
 #define elf_backend_count_additional_relocs	elf32_arm_count_additional_relocs
+#define elf_backend_symbol_processing		elf32_arm_backend_symbol_processing
 
 #define elf_backend_can_refcount       1
 #define elf_backend_can_gc_sections    1