[ELF] Make __ehdr_start point to ELF file headers

[phosek]

Summary

__ehdr_start should be pointing to ELF file headers, not program headers.

[pcc]

LGTM

This should probably be cherry-picked into the 4.0 branch as well.

[phosek]

This appears to be broken on Windows for some reason (http://lab.llvm.org:8011/builders/lld-x86_64-win7/builds/5154), so reverting the change.

[ruiu]

I have a Windows machine. I can try this patch for you.

[phosek]

When comparing the output from gold, ld and lld, one difference I noticed is that gold emits this symbol as SHN_ABS, ld uses index of .text, while lld uses SHN_UNDEF. It's the same for other reserved symbols like _end, _etext, _edata. Is there a particular reason why lld uses SHN_UNDEF or is that a bug?

[ruiu]

It's reproducible on my machine. This is an output of llvm-readobj.

D:\b>bin\llvm-readobj -symbols \b\tools\lld\test\ELF\Output\ehdr_start.s.tmp

Error reading file: invalid section index
.

File: \b\tools\lld\test\ELF\Output\ehdr_start.s.tmp
Format: ELF64-x86-64
Arch: x86_64
AddressSize: 64bit
LoadName:
Symbols [

Symbol {
  Name:  (0)
  Value: 0x0
  Size: 0
  Binding: Local (0x0)
  Type: None (0x0)
  Other: 0
  Section: Undefined (0x0)
}

[pcc]

These symbols don't really belong to a section, so there's no appropriate section index for them. Nor do they participate in dynamic loading (as they have hidden visibility). A section index of 1 would probably be appropriate (it doesn't matter what value it is, as long as it is not SHN_UNDEF or SHN_ABS).

Anyway, I think the problem is that we do not initialize Out<ELFT>::ElfHeader->SectionIndex, and the fix is to change the code here: http://llvm-cs.pcc.me.uk/tools/lld/ELF/Writer.cpp#1137

lld/trunk/ELF/Writer.cpp

template <class ELFT> void Writer<ELFT>::addReservedSymbols() {
   if (!In<ELFT>::DynSymTab &&
       (Config->EMachine != EM_MIPS && Config->EMachine != EM_ARM))
     Symtab<ELFT>::X->addIgnored("__tls_get_addr");
 
   // If linker script do layout we do not need to create any standart symbols.
   if (ScriptConfig->HasSections)
     return;
 
-  // __ehdr_start is the location of program headers.
+  // __ehdr_start is the location of ELF file headers.
   ElfSym<ELFT>::EhdrStart =
-      addOptionalSynthetic<ELFT>("__ehdr_start", Out<ELFT>::ProgramHeaders, 0);
+      addOptionalSynthetic<ELFT>("__ehdr_start", Out<ELFT>::ElfHeader, 0);
 
   auto Define = [](StringRef S, DefinedSynthetic *&Sym1,
                    DefinedSynthetic *&Sym2) {
     Sym1 = addOptionalSynthetic<ELFT>(S, nullptr, 0, STV_DEFAULT);
     assert(S.startswith("_"));
     S = S.substr(1);
     Sym2 = addOptionalSynthetic<ELFT>(S, nullptr, 0, STV_DEFAULT);
   };

lld/trunk/test/ELF/ehdr_start.s

 # REQUIRES: x86
 
 # RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t.o
 # RUN: ld.lld %t.o -o %t
-# RUN: llvm-objdump -t %t | FileCheck %s
-# CHECK: 0000000000200040 .text 00000000 .hidden __ehdr_start
+# RUN: llvm-readobj -symbols %t | FileCheck %s
+# CHECK:    Name: __ehdr_start (1)
+# CHECK-NEXT:    Value: 0x200000
+# CHECK-NEXT:    Size: 0
+# CHECK-NEXT:    Binding: Local (0x0)
+# CHECK-NEXT:    Type: None (0x0)
+# CHECK-NEXT:    Other [ (0x2)
+# CHECK-NEXT:      STV_HIDDEN (0x2)
+# CHECK-NEXT:    ]
+# CHECK-NEXT:    Section: Undefined (0x0)
 
 .text
 .global _start, __ehdr_start
 _start:
   .quad __ehdr_start
 
 # RUN: ld.lld -r %t.o -o %t.r
-# RUN: llvm-objdump -t %t.r | FileCheck %s --check-prefix=RELOCATABLE
+# RUN: llvm-readobj -symbols %t.r | FileCheck %s --check-prefix=RELOCATABLE
 
-# RELOCATABLE: 0000000000000000 *UND* 00000000 __ehdr_start
+# RELOCATABLE:    Name: __ehdr_start (1)
+# RELOCATABLE-NEXT:    Value: 0x0
+# RELOCATABLE-NEXT:    Size: 0
+# RELOCATABLE-NEXT:    Binding: Global (0x1)
+# RELOCATABLE-NEXT:    Type: None (0x0)
+# RELOCATABLE-NEXT:    Other: 0
+# RELOCATABLE-NEXT:    Section: Undefined (0x0)