[asan] Fix a false positive ODR violation due to LTO ConstantMerge pass

[kubamracek]

Summary

This fixes a false positive ODR violation that is reported by ASan when using LTO. In cases, where two constant globals have the same value, LTO will merge them, which breaks ASan's ODR detection. See the included testcase for an example.

[steven_wu]

LGTM.

The idea is to prevent LTO (ConstMerge pass) from merging global constants that has the same value, which has the same side-effect as linker merging two weak symbols that has ODR violation.

To be extra safe, there are some linkers that are capable of dedup functions and globals. Maybe a separate investigation to see if there is any linker dedup optimization can break this? Currently, ld64 doesn't dedup none weak symbols so it should be fine for now. Maybe check if lld does that?

[kubamracek]

Thanks, yes, AFAIK, ld64 is fine and I don't really know how to test lld on Darwin (is it even supposed to work?).

[steven_wu]

In D43959#1024157, @kubamracek wrote:

Thanks, yes, AFAIK, ld64 is fine and I don't really know how to test lld on Darwin (is it even supposed to work?).

no, lld is not working yet.

llvm/trunk/lib/Transforms/Instrumentation/AddressSanitizer.cpp

for (size_t i = 0; i < n; i++) {
 
     if (ClInitializers && MD.IsDynInit) HasDynamicallyInitializedGlobals = true;
 
     DEBUG(dbgs() << "NEW GLOBAL: " << *NewGlobal << "\n");
 
     Initializers[i] = Initializer;
   }
 
+  // Add instrumented globals to llvm.compiler.used list to avoid LTO from
+  // ConstantMerge'ing them.
+  appendToCompilerUsed(M,
+                       ArrayRef<GlobalValue *>(
+                           reinterpret_cast<GlobalValue **>(NewGlobals.data()),
+                           NewGlobals.size()));
+
   std::string ELFUniqueModuleId =
       (UseGlobalsGC && TargetTriple.isOSBinFormatELF()) ? getUniqueModuleId(&M)
                                                         : "";
 
   if (!ELFUniqueModuleId.empty()) {
     InstrumentGlobalsELF(IRB, M, NewGlobals, Initializers, ELFUniqueModuleId);
     *CtorComdat = true;
   } else if (UseGlobalsGC && TargetTriple.isOSBinFormatCOFF()) {

llvm/trunk/test/Instrumentation/AddressSanitizer/global_lto_merge.ll

+; RUN: opt < %s -asan -asan-module -S | FileCheck %s
+; RUN: opt < %s -asan -asan-module -constmerge -S | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.11.0"
+
+%struct = type { i64, i64 }
+
+@a = private unnamed_addr constant %struct { i64 16, i64 16 }, align 8
+@b = private unnamed_addr constant %struct { i64 16, i64 16 }, align 8
+
+; CHECK: @a = {{.*}} %struct
+; CHECK: @b = {{.*}} %struct
+
+; CHECK: @llvm.compiler.used =
+; CHECK-SAME: i8* bitcast ({ %struct, [48 x i8] }* @a to i8*)
+; CHECK-SAME: i8* bitcast ({ %struct, [48 x i8] }* @b to i8*)
+
+define i32 @main(i32, i8** nocapture readnone) {
+  %3 = alloca %struct, align 8
+  %4 = alloca %struct, align 8
+  %5 = bitcast %struct* %3 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull %5, i8* bitcast (%struct* @a to i8*), i64 16, i32 8, i1 false)
+  %6 = bitcast %struct* %4 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull %6, i8* bitcast (%struct* @b to i8*), i64 16, i32 8, i1 false)
+  call void asm sideeffect "", "r,r,~{dirflag},~{fpsr},~{flags}"(%struct* nonnull %3, %struct* nonnull %4)
+  ret i32 0
+}
+
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i32, i1)