[PATCH] D38625: [ELF] - Ignore non-absolute R_386_GOTPC relocation in debug sections.

[George Rimar via Phabricator via llvm-commits]

grimar created this revision.
Herald added a subscriber: emaste.

This is PR34852.

Looks gcc-5.4 has a bug and produce R_386_GOTPC relocation in '.rel.debug_info':
Only i386 target is affected it seems, issue did not reproduce for me when using gcc-5.4.1+x64 target.
Issue also did not reproduce with clang of version 6.0.0 (trunk 310596) or gcc of version 7.1.0 or
even gcc-4.8 (according to bug note).

ld.bfd behavior is to silently resolve the relocation. Given the code we have, I think we can try to ignore it in LLD.
Does not seem much sence to have PC relocation against _GLOBAL_OFFSET_TABLE_ in .debug_info section.


https://reviews.llvm.org/D38625

Files:
  ELF/InputSection.cpp
  test/ELF/i386-debug-noabs.test


Index: test/ELF/i386-debug-noabs.test
===================================================================
--- test/ELF/i386-debug-noabs.test
+++ test/ELF/i386-debug-noabs.test
@@ -0,0 +1,33 @@
+# REQUIRES: x86
+
+# RUN: yaml2obj %s -o %t.o
+# RUN: ld.lld %t.o -o %t.exe
+
+## It is for PR34852 (https://bugs.llvm.org//show_bug.cgi?id=34852)
+## gcc-5.4 has a bug which creates non-absolute R_386_GOTPC relocation
+## against _GLOBAL_OFFSET_TABLE_ when producing objects for i386 target.
+## Here we check that LLD is able to use such objects.
+
+--- !ELF
+FileHeader:      
+  Class:           ELFCLASS32
+  Data:            ELFDATA2LSB
+  Type:            ET_REL
+  Machine:         EM_386
+Sections:        
+  - Name:            .debug_info
+    Type:            SHT_PROGBITS
+    AddressAlign:    0x0000000000000001
+    Content:         0000000000000000
+  - Name:            .rel.debug_info
+    Type:            SHT_REL
+    Link:            .symtab
+    AddressAlign:    0x0000000000000004
+    Info:            .debug_info
+    Relocations:     
+      - Offset:          0x000000000000041F
+        Symbol:          _GLOBAL_OFFSET_TABLE_
+        Type:            R_386_GOTPC
+Symbols:
+  Global:
+    - Name:            _GLOBAL_OFFSET_TABLE_
Index: ELF/InputSection.cpp
===================================================================
--- ELF/InputSection.cpp
+++ ELF/InputSection.cpp
@@ -683,6 +683,12 @@
     if (Expr == R_NONE)
       continue;
     if (Expr != R_ABS) {
+      // gcc-5.4 has a bug which creates R_386_GOTPC relocation against
+      // _GLOBAL_OFFSET_TABLE_ in .debug_info. We ignore such relocation.
+      // TODO: remove this hack once we stop supporting gcc-5.4.
+      if (Config->EMachine == EM_386 && Expr == R_GOTONLY_PC_FROM_END &&
+          Sym.getName() == "_GLOBAL_OFFSET_TABLE_")
+        continue;
       error(this->getLocation<ELFT>(Offset) + ": has non-ABS reloc");
       return;
     }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D38625.117989.patch
Type: text/x-patch
Size: 1941 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20171006/c482129f/attachment.bin>