COFF/ELF: Place llvm.global_ctors elements in llvm.used if comdat is used

[MaskRay]

Summary

On ELF, an SHT_INIT_ARRAY outside a section group is a GC root. The current
codegen abuses SHT_INIT_ARRAY in a section group to mean a GC root.

On PE/COFF, the dynamic initialization for __declspec(selectany) in a comdat
can be garbage collected by -opt:ref.

Call addUsedGlobal for the two cases to fix the abuse/bug.

[rnk]

lgtm

So the main impact here is that, on ELF, linker GC will no longer be able to GC vague linkage global variables with dynamic initializers. Those are things like

• C++17 inline globals
• selectany globals
• static data members of class template instantiations

Seems reasonable to me.

clang/lib/CodeGen/CGDeclCXX.cpp

559

My first thought is that the selectany attribute should control the GVA_Linkage in the AST, but it looks like that doesn't happen right now. This is just refactoring the condition below.

[MaskRay]

In D106925#2910638, @rnk wrote:

lgtm

So the main impact here is that, on ELF, linker GC will no longer be able to GC vague linkage global variables with dynamic initializers. Those are things like

• C++17 inline globals
• selectany globals
• static data members of class template instantiations

Seems reasonable to me.

That is it! The ELF linker GC is similar to link.exe GC in this regard: ctor sections in a group/COMDAT are not special (GC roots).

clang/lib/CodeGen/CGDeclCXX.cpp

if (D->getTLSKind()) {
     CXXThreadLocalInitVars.push_back(D);
   } else if (PerformInit && ISA) {
     EmitPointerToInitFunc(D, Addr, Fn, ISA);
   } else if (auto *IPA = D->getAttr<InitPriorityAttr>()) {
     OrderGlobalInitsOrStermFinalizers Key(IPA->getPriority(),
                                           PrioritizedCXXGlobalInits.size());
     PrioritizedCXXGlobalInits.push_back(std::make_pair(Key, Fn));
   } else if (isTemplateInstantiation(D->getTemplateSpecializationKind()) ||
-             getContext().GetGVALinkageForVariable(D) == GVA_DiscardableODR) {
+             getContext().GetGVALinkageForVariable(D) == GVA_DiscardableODR ||
+             D->hasAttr<SelectAnyAttr>()) {
     // C++ [basic.start.init]p2:
     //   Definitions of explicitly specialized class template static data
     //   members have ordered initialization. Other class template static data
     //   members (i.e., implicitly or explicitly instantiated specializations)
     //   have unordered initialization.
     //
     // As a consequence, we can put them into their own llvm.global_ctors entry.
     //
     // If the global is externally visible, put the initializer into a COMDAT
     // group with the global being initialized.  On most platforms, this is a
     // minor startup time optimization.  In the MS C++ ABI, there are no guard
     // variables, so this COMDAT key is required for correctness.
-    AddGlobalCtor(Fn, 65535, COMDATKey);
-    if (getTarget().getCXXABI().isMicrosoft() && COMDATKey) {
-      // In The MS C++, MS add template static data member in the linker
-      // drective.
-      addUsedGlobal(COMDATKey);
-    }
-  } else if (D->hasAttr<SelectAnyAttr>()) {
+    //
     // SelectAny globals will be comdat-folded. Put the initializer into a
     // COMDAT group associated with the global, so the initializers get folded
     // too.
+
     AddGlobalCtor(Fn, 65535, COMDATKey);
+    if (COMDATKey && (getTriple().isOSBinFormatELF() ||
+                      getTarget().getCXXABI().isMicrosoft())) {
+      // When COMDAT is used on ELF or in the MS C++ ABI, the key must be in
+      // llvm.used to prevent linker GC.
+      addUsedGlobal(COMDATKey);
+    }
   } else {
     I = DelayedCXXInitPosition.find(D); // Re-do lookup in case of re-hash.
     if (I == DelayedCXXInitPosition.end()) {
       CXXGlobalInits.push_back(Fn);
     } else if (I->second != ~0U) {
       assert(I->second < CXXGlobalInits.size() &&
              CXXGlobalInits[I->second] == nullptr);
       CXXGlobalInits[I->second] = Fn;

clang/test/CodeGenCXX/microsoft-abi-template-static-init.cpp

 {
   inline static A aoo; // C++17 inline variable, thus also a definition
 };
 
 int foo();
 inline int zoo = foo();
 inline static int boo = foo();
 
-
-// CHECK: @llvm.used = appending global [7 x i8*] [i8* bitcast (i32* @"?x1@selectany_init@@3HA" to i8*), i8* bitcast (i32* @"?x@?$A@H@explicit_template_instantiation@@2HA" to i8*), i8* bitcast (i32* @"?ioo@?$X_@H@@2HA" to i8*), i8* getelementptr inbounds (%struct.A, %struct.A* @"?aoo@S1@@2UA@@A", i32 0, i32 0), i8* bitcast (i32* @"?zoo@@3HA" to i8*), i8* getelementptr inbounds (%struct.S, %struct.S* @"?s@?$ExportedTemplate@H@@2US@@A", i32 0, i32 0), i8* bitcast (i32* @"?x@?$A@H@implicit_template_instantiation@@2HA" to i8*)], section "llvm.metadata"
+// CHECK: @llvm.used = appending global [8 x i8*] [i8* bitcast (i32* @"?x@selectany_init@@3HA" to i8*), i8* bitcast (i32* @"?x1@selectany_init@@3HA" to i8*), i8* bitcast (i32* @"?x@?$A@H@explicit_template_instantiation@@2HA" to i8*), i8* bitcast (i32* @"?ioo@?$X_@H@@2HA" to i8*), i8* getelementptr inbounds (%struct.A, %struct.A* @"?aoo@S1@@2UA@@A", i32 0, i32 0), i8* bitcast (i32* @"?zoo@@3HA" to i8*), i8* getelementptr inbounds (%struct.S, %struct.S* @"?s@?$ExportedTemplate@H@@2US@@A", i32 0, i32 0), i8* bitcast (i32* @"?x@?$A@H@implicit_template_instantiation@@2HA" to i8*)], section "llvm.metadata"

clang/test/CodeGenCXX/static-member-variable-explicit-specialization.cpp

 // RUN: %clang_cc1 %s -std=c++1y -triple=x86_64-pc-linux -emit-llvm -o - | FileCheck --check-prefix=ELF --check-prefix=ALL %s
 // RUN: %clang_cc1 %s -std=c++1y -triple=x86_64-apple-darwin -emit-llvm -o - | FileCheck --check-prefix=MACHO --check-prefix=ALL %s
+// RUN: %clang_cc1 %s -std=c++1y -triple=x86_64-pc-linux -emit-llvm -fdeclspec -DSELECTANY -o - | FileCheck --check-prefix=ELF-SELECTANY %s
+
+#ifdef SELECTANY
+struct S {
+  S();
+  ~S();
+};
+
+int f();
+
+// ELF-SELECTANY: @llvm.global_ctors = appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 65535, void ()* @__cxx_global_var_init, i8* bitcast (i32* @selectany to i8*) }]
+// ELF-SELECTANY: @llvm.used = appending global [1 x i8*] [i8* bitcast (i32* @selectany to i8*)]
+int __declspec(selectany) selectany = f();
+
+#else
 
 // ALL: ; ModuleID
 
 extern "C" int foo();
 
 template<typename T> struct A { static int a; };
 template<typename T> int A<T>::a = foo();
 

 
 // ELF:  { i32, void ()*, i8* } { i32 65535, void ()* @[[unordered6:[^,]*]], i8* @_Z1xIcE },
 // MACHO:  { i32, void ()*, i8* } { i32 65535, void ()* @[[unordered6:[^,]*]], i8* null },
 
 // ALL:  { i32, void ()*, i8* } { i32 65535, void ()* @[[unordered7:[^,]*]], i8* null },
 
 // ALL:  { i32, void ()*, i8* } { i32 65535, void ()* @_GLOBAL__sub_I_static_member_variable_explicit_specialization.cpp, i8* null }]
 
+/// llvm.used ensures SHT_INIT_ARRAY in a section group cannot be GCed.
+// ELF: @llvm.used = appending global [6 x i8*] [i8* bitcast (i32* @_ZN1AIsE1aE to i8*), i8* bitcast (i16* @_Z1xIsE to i8*), i8* bitcast (i32* @_ZN2ns1aIiE1iE to i8*), i8* bitcast (i32* @_ZN2ns1b1iIiEE to i8*), i8* bitcast (i32* @_ZN1AIvE1aE to i8*), i8* @_Z1xIcE]
+
 template int A<short>::a;  // Unordered
 int b = foo();
 int c = foo();
 int d = A<void>::a; // Unordered
 
 // An explicit specialization is ordered, and goes in __GLOBAL_sub_I_static_member_variable_explicit_specialization.cpp.
 template<> struct A<int> { static int a; };
 int A<int>::a = foo();

 }
 
 namespace {
 template<typename T> struct Internal { static int a; };
 template<typename T> int Internal<T>::a = foo();
 }
 int *use_internal_a = &Internal<int>::a;
 
+#endif
+
 // ALL: define internal void @[[unordered1]](
 // ALL: call i32 @foo()
 // ALL: store {{.*}} @_ZN1AIsE1aE
 // ALL: ret
 
 // ALL: define internal void @[[unordered2]](
 // ALL: call i32 @foo()
 // ALL: store {{.*}} @_Z1xIsE