[Bug ld/19368] [2.26/2.27 regression] IFUNC support not working on arm-linux-gnueabi*

[jiwang at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=19368

--- Comment #3 from Jiong Wang <jiwang at gcc dot gnu.org> ---
This is caused by the following commit which turn on
elf_backend_extern_protected_data.

commit b68a20d6675f1360ea4db50a9835c073675b9889
Author: Szabolcs Nagy <address@hidden>
Date: Mon Jul 27 11:45:27 2015 +0100

[ARM] Fix extern protected data handling

after this change we are turning on PC relocative relocation to absolute
relocation, and thus the following relocation entries change:

--- dump.a.right        2015-12-16 16:13:01.797125100 +0000
+++ dump.a.wrong        2015-12-16 16:10:06.045693098 +0000
 Relocation section '.rel.dyn' at offset 0x4e4 contains 16 entries:
  Offset     Info    Type                Sym. Value  Symbol's Name
 ...     
-00011030  00000017 R_ARM_RELATIVE        
 0001104c  00000017 R_ARM_RELATIVE        
 00011018  000000a0 R_ARM_IRELATIVE       
 0001101c  000000a0 R_ARM_IRELATIVE       
@@ -115,6 +114,7 @@
 00011024  00000415 R_ARM_GLOB_DAT         00000000   __cxa_finalize
 00011028  00000515 R_ARM_GLOB_DAT         00000000   _ITM_registerTMCloneTable
 0001102c  00000615 R_ARM_GLOB_DAT         00000000   ret_foo
+00011030  00001315 R_ARM_GLOB_DAT         00011050   global
 ...

These runtime relocation types change looks ok to me, but their runtime resolve
order will cause problem for ifunc resolver.

You can see "global" is turned from RELATIVE which is resolved before
IRELATIVE, into GLOBA_DAT which will be resolved after the IRELATIVE, while
"global" is referenced in the ifunc function, so that ifunc function will get
uninitialized address zero, then trigger segment fault.

I think ARM backend need the similar fix from:

  https://sourceware.org/bugzilla/show_bug.cgi?id=13302

we need to make sure IRELATIVE relocation happen last.

-- 
You are receiving this mail because:
You are on the CC list for the bug.