[llvm-dev] What is an address-significant symbol?

[Peter Smith via llvm-dev]

On Fri, 22 Nov 2019 at 11:20, Gaier, Bjoern via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
>
> Hello LLVM- and Clang-List,
>
>
>
> I’m not sure if this a subject for LLVM or Clang – but there is something I don’t understand. I wrote the following code in C++:
>
>
>
> searchPlanschi(&__stdio_common_vswprintf);
>
>
>
> ‚searchPlanschi‘ is a function I provide. Clang generates the following assembly code for this:
>
>
>
>      lea   rcx, [rip + __stdio_common_vswprintf]
>
>       call  "?searchPlanschi@@YAXPEAX at Z"
>
>
>
> I was surprised to see the register rip there, as far as I know this is the Instruction register, right? Why do I need the rip register to get the address of the function? I searched the assembly file for ‘__stdio_common_vswprintf’ to get some hints about this. The only thing I found was:
>
>
>
>       .addrsig
>
>       .addrsig_sym __stdio_common_vswprintf
>
>
>
> So I googled “.addrsig” and found the following text:
>
> “This section is used to mark symbols as address-significant, i.e. the address of the symbol is used in a comparison or leaks outside the translation unit. It has the same meaning as the absence of the LLVM attributes unnamed_addr and local_unnamed_addr.
> Any sections referred to by symbols that are not marked as address-significant in any object file may be safely merged by a linker without breaking the address uniqueness guarantee provided by the C and C++ language standards.
> The contents of the section are a sequence of ULEB128-encoded integers referring to the symbol table indexes of the address-significant symbols.”
>
> But sadly… this is way over my head. What does that actually mean? Does that explain the code construct with the rip register? Is that a form of optimization?
>

There is a linker optimization called identical code folding (ICF).
The details of the initial implementation in gold is described in
https://ai.google/research/pubs/pub36912 . ICF can cause problems when
the program depends on functions having a unique address, both gold
and LLD have an --icf=safe mode that limits the scope of the
optimization to avoid folding sections that are "address-significant".
The implementation in gold uses linker heuristics such as relocation
type to determine address significance. The implementation in LLD uses
information generated by the compiler, which is placed in .addrsig.

I can't answer the question about code-generation off the top of my
head, my understanding is that .addrsig is primarily used for
implementing --icf=safe in linkers, I don't think it has an affect on
code-generation.

Hope this is of some help

Peter

>
>
> Thank you in advance for any help!
>
>
>
> Kind greetings
>
> Björn
>
> Als GmbH eingetragen im Handelsregister Bad Homburg v.d.H. HRB 9816, USt.ID-Nr. DE 114 165 789 Geschäftsführer: Dr. Hiroshi Nakamura, Dr. Robert Plank, Markus Bode, Heiko Lampert, Takashi Nagano, Takeshi Fukushima. Junichi Tajika
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev