PR42182: Allow thread-local to use __cxa_thread_atexit when -fno-use-cxx-atexit is used

[erichkeane]

Summary

This matches the GCC behavior, __cxa_thread_atexit should be permissible
even though cxa_atexit is disabled.

[rjmccall]

clang/lib/CodeGen/ItaniumCXXABI.cpp

2373

It's fine to match the GCC behavior, and it's even somewhat explicable. Could you update the comment to explain, first of all, that this path handles both __cxa_atexit and __cxa_thread_atexit, and secondly that the option is only meant to control the former?

[erichkeane]

Update the comment as requested by @rjmccall

[rjmccall]

Thanks!

clang/lib/CodeGen/ItaniumCXXABI.cpp

void ItaniumCXXABI::EmitGuardedInit(CodeGenFunction &CGF,
 
   CGF.EmitBlock(EndBlock);
 }
 
 /// Register a global destructor using __cxa_atexit.
 static void emitGlobalDtorWithCXAAtExit(CodeGenFunction &CGF,
                                         llvm::FunctionCallee dtor,
                                         llvm::Constant *addr, bool TLS) {
+  assert((TLS || CGF.getTypes().getCodeGenOpts().CXAAtExit) &&
+         "__cxa_atexit is disabled");
   const char *Name = "__cxa_atexit";
   if (TLS) {
     const llvm::Triple &T = CGF.getTarget().getTriple();
     Name = T.isOSDarwin() ?  "_tlv_atexit" : "__cxa_thread_atexit";
   }
 
   // We're assuming that the destructor function is something we can
   // reasonably call with the default CC.  Go ahead and cast it to the

 
 /// Register a global destructor as best as we know how.
 void ItaniumCXXABI::registerGlobalDtor(CodeGenFunction &CGF, const VarDecl &D,
                                        llvm::FunctionCallee dtor,
                                        llvm::Constant *addr) {
   if (D.isNoDestroy(CGM.getContext()))
     return;
 
-  // Use __cxa_atexit if available.
-  if (CGM.getCodeGenOpts().CXAAtExit)
+  // emitGlobalDtorWithCXAAtExit will emit a call to either __cxa_thread_atexit
+  // or __cxa_atexit depending on whether this VarDecl is a thread-local storage
+  // or not. CXAAtExit controls only __cxa_atexit, so use it if it is enabled.
+  // We can always use __cxa_thread_atexit.
+  if (CGM.getCodeGenOpts().CXAAtExit || D.getTLSKind())
     return emitGlobalDtorWithCXAAtExit(CGF, dtor, addr, D.getTLSKind());
 
-  if (D.getTLSKind())
-    CGM.ErrorUnsupported(&D, "non-trivial TLS destruction");
-
   // In Apple kexts, we want to add a global destructor entry.
   // FIXME: shouldn't this be guarded by some variable?
   if (CGM.getLangOpts().AppleKext) {
     // Generate a global destructor entry.
     return CGM.AddCXXDtorEntry(dtor, addr);
   }
 
   CGF.registerGlobalDtorWithAtExit(D, dtor, addr);

clang/test/CodeGenCXX/cxx11-thread-local.cpp

 // RUN: %clang_cc1 -std=c++11 -emit-llvm %s -o - -triple x86_64-linux-gnu | FileCheck --check-prefix=CHECK --check-prefix=LINUX %s
 // RUN: %clang_cc1 -std=c++11 -emit-llvm %s -O2 -disable-llvm-passes -o - -triple x86_64-linux-gnu | FileCheck --check-prefix=CHECK --check-prefix=LINUX --check-prefix=CHECK-OPT %s
 // RUN: %clang_cc1 -std=c++11 -femulated-tls -emit-llvm %s -o - \
 // RUN:     -triple x86_64-linux-gnu 2>&1 | FileCheck --check-prefix=CHECK --check-prefix=LINUX %s
 // RUN: %clang_cc1 -std=c++11 -emit-llvm %s -o - -triple x86_64-apple-darwin12 | FileCheck --check-prefix=CHECK --check-prefix=DARWIN %s
 
+// RUN: %clang_cc1 -std=c++11 -fno-use-cxa-atexit -emit-llvm %s -o - -triple x86_64-linux-gnu | FileCheck --check-prefix=CHECK --check-prefix=LINUX %s
+// RUN: %clang_cc1 -std=c++11 -fno-use-cxa-atexit -emit-llvm %s -O2 -disable-llvm-passes -o - -triple x86_64-linux-gnu | FileCheck --check-prefix=CHECK --check-prefix=LINUX --check-prefix=CHECK-OPT %s
+// RUN: %clang_cc1 -std=c++11 -fno-use-cxa-atexit -femulated-tls -emit-llvm %s -o - \
+// RUN:     -triple x86_64-linux-gnu 2>&1 | FileCheck --check-prefix=CHECK --check-prefix=LINUX %s
+// RUN: %clang_cc1 -std=c++11 -fno-use-cxa-atexit -emit-llvm %s -o - -triple x86_64-apple-darwin12 | FileCheck --check-prefix=CHECK --check-prefix=DARWIN %s
+
 int f();
 int g();
 
 // LINUX-DAG: @a = thread_local global i32 0
 // DARWIN-DAG: @a = internal thread_local global i32 0
 thread_local int a = f();
 extern thread_local int b;
 // CHECK-DAG: @c = global i32 0