[PATCH] D63220: [ELF][RISCV] Support GD/LD/IE/LE TLS models

[Fangrui Song via Phabricator via llvm-commits]

MaskRay marked 2 inline comments as done.
MaskRay added inline comments.


================
Comment at: ELF/Relocations.cpp:224
+  // If the TLS model can be relaxed to Local-Exec, DTPMOD is constant one.
+  bool IsLocalExec = !Config->Shared && !Sym.IsPreemptible;
+
----------------
jrtc27 wrote:
> The RISC-V behaviour is identical to ARM if I remember correctly, so we should be reusing that code rather than reimplementing it again here.
R_ARM_TLS_GD32 <-> R_RISCV_TLS_GD_HI20
R_ARM_TLS_LDM32 <-> R_RISCV_TLS_GD_HI20 with a local symbol .LANCHOR0
R_ARM_TLS_IE32 <-> R_RISCV_TLS_GOT_HI20
R_ARM_TLS_LE32 <-> R_RISCV_TPREL_{HI20,LO12_I,LO12_S}

If this is what you mean :)

However, there is a big difference: RISC-V doesn't support relaxation. So I have to make sure no R_RELAX_TLS_* code is called.

Having the two variables is the least intrusive approach I can think of. (I complained the lack of linker relaxation turns out to make lld's generic TLS handling uglier. Here it is :))




================
Comment at: test/ELF/riscv-tls-ld.s:1
+# REQUIRES: riscv
+# RUN: echo '.tbss; .globl b, c; b: .zero 4; c:' > %t.s
----------------
jrtc27 wrote:
> This file is the same as GD just with different check prefixes, no? RISC-V’s local dynamic is the same as global dynamic as far as relocations are concerned.
No exactly the same.

`riscv-tls-gd.s` checks preemptable symbols. This file checks `.LANCHOR0`. I do struggle to make the tests simpler while covering useful/possible scenarios.


Repository:
  rLLD LLVM Linker

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63220/new/

https://reviews.llvm.org/D63220