[PATCH] D47098: [ELF] Make R_GOTONLY_PC_FROM_END R_GOTREL_FROM_END relative to _GLOBAL_OFFSET_TABLE_ instead of end of .got

[Fangrui Song via Phabricator via llvm-commits]

MaskRay marked an inline comment as done.
MaskRay added inline comments.


================
Comment at: ELF/InputSection.cpp:501
+    // If we scanned relocations earlier, GlobalOffsetTable could't be null.
+    return (GOTBase ? GOTBase->getVA()
+                    : InX::Got->getVA() + InX::Got->getSize()) +
----------------
grimar wrote:
> What is your plan to fix this `FIXME`? It does not seem nice approach to add weak _GLOBAL_OFFSET_TABLE_ symbols to test cases
> which affects the calculation.
Changed the description. I think the usage of `InX::Got->getVA() + InX::Got->getSize()` is incorrect. We should use either:

* `ElfSym::GlobalOffsetTable`. but it can be null when `_GLOBAL_OFFSET_TABLE_` is not explicitly referenced.
* If `GlobalOffsetTable` is null, use `InX::Got` or `InX::GotPlt` depending on `GotBaseSymInGotPlt`. We can avoid this if we scan relocations earlier and decide to emit `GlobalOffsetTable`.

To properly address this I also need to study the TLS stuff. This revision is already sort of complicated, I want to keep the changes in a future revision


Repository:
  rLLD LLVM Linker

https://reviews.llvm.org/D47098