[X86] -fno-plt: use GOT __tls_get_addr only if GOTPCRELX is enabled

[MaskRay]

Summary

As of binutils 2.32, ld has a bogus TLS relaxation error when the GD/LD
code sequence using R_X86_64_GOTPCREL (instead of R_X86_64_GOTPCRELX) is
attempted to be relaxed to IE/LE (binutils PR24784). gold and lld are good.

In gcc/config/i386/i386.md, there is a configure-time check of as/ld
support and the GOT relaxation will not be used if as/ld doesn't support
it:

if (flag_plt || !HAVE_AS_IX86_TLS_GET_ADDR_GOT)
  return "call\t%P2";
return "call\t{*%p2@GOT(%1)|[DWORD PTR %p2@GOT[%1]]}";

In clang, -DENABLE_X86_RELAX_RELOCATIONS=OFF is the default. The ld.bfd
bogus error can be reproduced with:

thread_local int a;
int main() { return a; }

clang -fno-plt -fpic a.cc -fuse-ld=bfd

GOTPCRELX gained relative good support in 2016, which is considered
relatively new. It is even difficult to conditionally default to
-DENABLE_X86_RELAX_RELOCATIONS=ON due to cross compilation reasons. So
work around the ld.bfd bug by only using GOT when GOTPCRELX is enabled.

[nikic]

LGTM, this looks like a reasonable workaround to me.

[MaskRay]

In D64304#1578773, @nikic wrote:

LGTM, this looks like a reasonable workaround to me.

Can you answer the question?

@nikic Can you share a reproduce of the Rust code?

[nikic]

In D64304#1579785, @MaskRay wrote:

Can you share a reproduce of the Rust code?

Sorry, I missed that! The rust code is essentially the same as your C code, plus a lot of boilerplate to avoid linking libstd, otherwise the internal TLS use in libstd will already cause the linker error.

// test2.rs
#![crate_type="rlib"]
#![no_std]
#![feature(thread_local)]

#[thread_local]
static TEST: isize = 0;

pub fn get_test() -> isize {
    TEST
}

// test.rs
#![no_std]
#![no_main]
extern crate test2;

use core::panic::PanicInfo;
#[panic_handler]
fn panic(_info: &PanicInfo) -> ! {
    loop {}
}

#[no_mangle]
pub extern "C" fn _start() -> isize {
    test2::get_test()
}

Additionally -C link-arg=-nostartfiles needs to be specified to avoid linking libc. (Equivalents of -fno-plt and -fpic are enabled by default already.)

This requires a rustc build linked against LLVM 9 though, and the only public build with LLVM 9 (rustup-toolchain-install-master 82310f68ee567093e06ad767cc0d3cb6127457ba) already includes your patch here, otherwise rustc would be practically unusable.

[MaskRay]

Add a TODO that we should delete the workaround when GOTPCRELX becomes commonplace

llvm/trunk/lib/Target/X86/X86MCInstLower.cpp

case X86::TLS_base_addr64:
     SRVK = MCSymbolRefExpr::VK_TLSLD;
     break;
   default:
     llvm_unreachable("unexpected opcode");
   }
 
   const MCSymbolRefExpr *Sym = MCSymbolRefExpr::create(
       MCInstLowering.GetSymbolFromOperand(MI.getOperand(3)), SRVK, Ctx);
-  bool UseGot = MMI->getModule()->getRtLibUseGOT();
+
+  // As of binutils 2.32, ld has a bogus TLS relaxation error when the GD/LD
+  // code sequence using R_X86_64_GOTPCREL (instead of R_X86_64_GOTPCRELX) is
+  // attempted to be relaxed to IE/LE (binutils PR24784). Work around the bug by
+  // only using GOT when GOTPCRELX is enabled.
+  // TODO Delete the workaround when GOTPCRELX becomes commonplace.
+  bool UseGot = MMI->getModule()->getRtLibUseGOT() &&
+                Ctx.getAsmInfo()->canRelaxRelocations();
 
   if (Is64Bits) {
     bool NeedsPadding = SRVK == MCSymbolRefExpr::VK_TLSGD;
     if (NeedsPadding)
       EmitAndCountInstruction(MCInstBuilder(X86::DATA16_PREFIX));
     EmitAndCountInstruction(MCInstBuilder(X86::LEA64r)
                                 .addReg(X86::RDI)
                                 .addReg(X86::RIP)

llvm/trunk/test/CodeGen/X86/tls-no-plt.ll

-; RUN: llc < %s -mtriple=i386-linux-musl -relocation-model=pic | FileCheck --check-prefixes=CHECK,X86 %s
-; RUN: llc < %s -mtriple=x86_64-linux-musl -relocation-model=pic | FileCheck --check-prefixes=CHECK,X64 %s
+; RUN: llc < %s -mtriple=i386-linux-musl -relocation-model=pic -relax-elf-relocations=true | FileCheck --check-prefixes=CHECK,X86 %s
+; RUN: llc < %s -mtriple=x86_64-linux-musl -relocation-model=pic -relax-elf-relocations=true | FileCheck --check-prefixes=CHECK,X64 %s
+
+;; If GOTPCRELX is disabled, don't use GOT for __tls_get_addr to work around
+;; a ld.bfd bug (binutils PR24784).
+; RUN: llc < %s -mtriple=i386-linux-musl -relocation-model=pic | FileCheck --check-prefixes=CHECK,X86-PLT %s
+; RUN: llc < %s -mtriple=x86_64-linux-musl -relocation-model=pic | FileCheck --check-prefixes=CHECK,X64-PLT %s
 
 @gd = thread_local global i32 0
 @ld = internal thread_local global i32 0
 
 define i32* @get_gd() {
 entry:
 ; CHECK-LABEL: get_gd:
 ; X86: leal gd@TLSGD(%ebx), %eax
 ; X86: calll *___tls_get_addr@GOT(%ebx)
+; X86-PLT: calll ___tls_get_addr@PLT
 
 ; X64: leaq gd@TLSGD(%rip), %rdi
 ; X64: callq *__tls_get_addr@GOTPCREL(%rip)
+; X64-PLT: callq __tls_get_addr@PLT
   ret i32* @gd
 }
 
 define i32* @get_ld() {
 ; FIXME: This function uses a single thread-local variable, we might want to fall back to general-dynamic.
 ; CHECK-LABEL: get_ld:
 ; X86: leal ld@TLSLDM(%ebx), %eax
 ; X86: calll *___tls_get_addr@GOT(%ebx)
+; X86-PLT: calll ___tls_get_addr@PLT
 
 ; X64: leaq ld@TLSLD(%rip), %rdi
 ; X64: callq *__tls_get_addr@GOTPCREL(%rip)
+; X64-PLT: callq __tls_get_addr@PLT
   ret i32* @ld
 }
 
 !llvm.module.flags = !{!1}
 !1 = !{i32 7, !"RtLibUseGOT", i32 1}