[PATCH] D59553: [LLD][ELF][DebugInfo] llvm-symbolizer shows incorrect source line info if --gc-sections used

Tue Apr 9 01:57:33 PDT 2019

MaskRay requested changes to this revision.
MaskRay added a comment.
This revision now requires changes to proceed.

lld sets x86_64 image base to 0x200000.
A .debug_info associated with a non-discarded section has DW_AT_low_pc > 0x200000
A .debug_info associated with a discarded section has DW_AT_low_pc = 0. If its size is larger than 0x200000 => DW_AT_high_pc > image_base => the two debug info entries overlap => llvm-symbolizer locates a DIE that's not what the user expects => unexpected filename (`not_used.cpp`)

If you make `".rept 2000000; nop; .endr"` larger, you'll notice bfd linked executable has the same problem (its image base is 0x400000).

> InputSection.cpp:833
>  Target->relocateOne(BufLoc, Type, SignExtend64<Bits>(Sym.getVA(Addend)));

I think the current handling in lld is quite good, no need to special case on `if (!Sym.getOutputSection())`.

After patching in this commit, `DW_AT_low_pc` becomes -1 but `llvm-symbolizer` is still puzzled.

  % llvm-symbolizer -obj res.out 0x2010f0
  __libc_csu_init
  ??:0:0      ######### 

The relevant code is in DWARFDebugAranges.cpp. I think the right fix is to improves the heuristics, rather than change lld. I can take a stab.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D59553/new/

https://reviews.llvm.org/D59553