Re: PATCH: PR ld/14272: Mark the plugin symbol undefined

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Thu, Feb 5, 2015 at 8:06 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Thu, Feb 5, 2015 at 5:52 PM, Alan Modra <amodra@gmail.com> wrote:
>> On Thu, Feb 05, 2015 at 05:06:38PM -0800, H.J. Lu wrote:
>>> On Thu, Feb 5, 2015 at 4:59 PM, Alan Modra <amodra@gmail.com> wrote:
>>> > On Wed, Feb 04, 2015 at 04:36:16AM -0800, H.J. Lu wrote:
>>> >> On Wed, Feb 4, 2015 at 3:08 AM, Alan Modra <amodra@gmail.com> wrote:
>>> >> > On Tue, Feb 03, 2015 at 09:06:47AM -0800, H.J. Lu wrote:
>>> >> >>       PR ld/12365
>>> >> >>       * ld-plugin/pr12365a.c: New file.
>>> >> >>       * ld-plugin/pr12365b.c: Likewise.
>>> >> >>       * ld-plugin/pr12365c.c: Likewise.
>>> >> >
>>> >> > The testcase fails when using gcc-5.  Also, gcc-5 objects don't seem
>>> >> > exercise your _bfd_elf_fix_symbol_flags code.  I presume you need an
>>> >> > older version of gcc, with lto bugs, to need those hacks.  Is that
>>> >> > true?  Can you write a testcase that does exercise your
>>> >> > _bfd_elf_fix_symbol_flags code?
>>> >>
>>> >> The -flto-partition default was changed.  I checked in this patch
>>> >> to pass -flto-partition=none to the PR ld/12365 test.
>>> >
>>> > The test fails when using gcc-4.7.2 on powerpc64-linux.  Please fix it.
>>>
>>> Does it fail with gcc 4.8?  What is the error message?
>>
>> No error message.  The link succeeds.  Similarly with gcc-4.9 and
>
> The test is very sensitive to LTO partition.  Also

The test is also sensitive to LTO IR order.  I am testing this patch.

> #define ASMNAME(cname)  ASMNAME2 (__USER_LABEL_PREFIX__, cname)
> #define ASMNAME2(prefix, cname) STRING (prefix) cname
> #define STRING(x)    #x
>
> typedef __SIZE_TYPE__ size_t;
> extern void abort (void);
> extern void *memcpy (void *, const void *, size_t)
>   __asm (ASMNAME ("my_memcpy"));
>
> doesn't work for all platforms.  I will keep digging.
>

I will also limit this test to x86 for now until we find a way to make it
work for other platforms.

-- 
H.J.

diff --git a/bfd/elflink.c b/bfd/elflink.c
index 604cfb6..15b82a9 100644
--- a/bfd/elflink.c
+++ b/bfd/elflink.c
@@ -2423,20 +2423,6 @@ _bfd_elf_fix_symbol_flags (struct elf_link_hash_entry *h,
     }
   else
     {
-      /* If a plugin symbol is referenced from a non-IR file, mark
-         the symbol as undefined, except for symbol for linker
-	 created section.  */
-      if (h->root.non_ir_ref
-	  && (h->root.type == bfd_link_hash_defined
-	      || h->root.type == bfd_link_hash_defweak)
-	  && (h->root.u.def.section->flags & SEC_LINKER_CREATED) == 0
-	  && h->root.u.def.section->owner != NULL
-	  && (h->root.u.def.section->owner->flags & BFD_PLUGIN) != 0)
-	{
-	  h->root.type = bfd_link_hash_undefined;
-	  h->root.u.undef.abfd = h->root.u.def.section->owner;
-	}
-
       /* Unfortunately, NON_ELF is only correct if the symbol
 	 was first seen in a non-ELF file.  Fortunately, if the symbol
 	 was first seen in an ELF file, we're probably OK unless the
@@ -9807,6 +9793,22 @@ elf_link_input_bfd (struct elf_final_link_info *flinfo, bfd *input_bfd)
 
 		  s_type = h->type;
 
+		  /* If a plugin symbol is referenced from a non-IR file, mark
+		     the symbol as undefined, except for symbol for linker
+		     created section.  */
+		  if (h->root.non_ir_ref
+		      && (h->root.type == bfd_link_hash_defined
+			  || h->root.type == bfd_link_hash_defweak)
+		      && (h->root.u.def.section->flags
+			  & SEC_LINKER_CREATED) == 0
+		      && h->root.u.def.section->owner != NULL
+		      && (h->root.u.def.section->owner->flags
+			  & BFD_PLUGIN) != 0)
+		    {
+		      h->root.type = bfd_link_hash_undefined;
+		      h->root.u.undef.abfd = h->root.u.def.section->owner;
+		    }
+
 		  ps = NULL;
 		  if (h->root.type == bfd_link_hash_defined
 		      || h->root.type == bfd_link_hash_defweak)