Add flag to write symbols for PLT / GOT entries

[davidlattimore]

Split from #308.

The idea is to write symtab entries with NAME_OF_SYMBOL$got that points to the GOT entry for the symbol. Similarly, NAME_OF_SYMBOL$plt for the PLT entry of the symbol.

e.g. __gmon_start__$got should point to the GOT entry for __gmon_start__.

GNU ld and LLD as far as I know don't do this, but mold does. I'm not sure if we want to do this by default, at least not initially, so I'm inclined to say that we should put it behind a flag. For want of a better name, how about --gen-got-plt-syms, but open to alternative names.

The main places that will need changing are:

• Argument parsing in args.rs to parse the new flag.
• SymbolRequestHandler::finalise_symbol_sizes. This will need to allocate space in the .symstr and .symtab sections for the extra symbols. You can determine if there's GOT / PLT entries for the symbol by checking current_res_flags.
• ObjectLayout::write_symbols (in elf_writer.rs). This is where the symbols will actually get written. The variable res holds a Resolution which will have most of the info you'll need like the GOT and PLT addresses.
• The easiest way to test this is probably to add an extra config to an existing test where you enable the flag. I think the ifunc test (ifunc.c) is probably a reasonable choice, since ifuncs are the main place where Wild uses PLT entries. You can check for the symbol using a test directive like //#ExpectSym:foo$plt.

If you'd like to work on this, please let us know so that we can assign it to you and we don't end up with multiple people working on it at the same time.

[Mayank85Y]

Hey @davidlattimore I looking forward to resolve this issue. Please assign it to me

[davidlattimore]

Great! Don't hesitate to reach out if you have any questions

[lapla-cogito]

Hi @Mayank85Y, are you still working on this?

[lapla-cogito]

I am currently working on implementing this option, but after conducting some experiments using mold locally, I noticed that in addition to NAME_OF_SYMBOL$plt and NAME_OF_SYMBOL$got, symbols like NAME_OF_SYMBOL$pltgot may also exist. For the first pair, it's clear they refer to PLT and GOT entries respectively, but I don't understand how to handle the NAME_OF_SYMBOL$pltgot case.

Plus, when using symbol_write.define_symbol() to actually write the symbol, is this the appropriate approach?

[davidlattimore]

These symbols are just to make debugging a bit nicer, since they give addresses to things that otherwise might not have addresses. So they don't need to conform to any particular standard. In the case of $pltgot, I'm guessing that's the GOT entry for a PLT entry. Some linkers have a separate GOT entry for use by the PLT. So a symbol might have two GOT entries, one for direct use and one for use by the PLT entry for the symbol. Wild doesn't do that, so there's no need for such a debug symbol.