Implement function type checker for the undefined behavior sanitizer.

[pcc]

Summary

This uses function prefix data to store
function type information at the function pointer.

[pcc]

• Also test for the RTTI pointer check code

[pcc]

Rebase.

[pcc]

The LLVM-side dependency for this has landed now.

[pcc]

Rebase.

[rsmith]

Clever! The approach only makes me feel a little uneasy =)

It would be polite to issue a diagnostic if someone explicitly asks for this (not via -fsanitize=undefined) and we don't support it for their target.

lib/CodeGen/CGExpr.cpp

3120

This is a bit hard to parse. (!TargetDecl || !isa<FunctionDecl>(TargetDecl)) maybe?

3121

* on the right.

3125

Some indication of what "PD" stands for here would be useful.

3125–3126

We usually format this as

llvm::Type *PDStructTyElems[] = {
  PDSig->getType(),
  FTRTTIConst->getType()
};

or similar.

3130–3134

What happens if the callee address is at the end of a page and doesn't have the extra data? Could this load trigger a segfault?

3147–3148

Likewise.

lib/CodeGen/CodeGenFunction.cpp

523–524

It'd be nice if we could only do this for address-taken functions. When we take the address of a function, we could emit linkonce_odr thunk with the extra data, and use that instead of the original address... except that would break address-of-function comparisons between instrumented and uninstrumented code. Can we provide an option to enable that behavior for the case where we're OK with an ABI change?

529–530

Does this include the calling convention attributes?

531

Spaces around braces.

lib/CodeGen/TargetInfo.cpp

602–603

What do 'F' and 'T' demangle as? An invalid instruction encoding would give me more confidence here.

[pcc]

• Address code review comments

[pcc]

It would be polite to issue a diagnostic if someone explicitly asks for this (not via -fsanitize=undefined) and we don't support it for their target.

Hmm, we should probably be doing this for other sanitizers too (e.g. MSan and DFSan are only supported on 64-bit Linux).

lib/CodeGen/CGExpr.cpp

3120

Done.

3121

Done.

3125

Globally replaced 'PD' with 'Prefix'.

3125–3126

Done. clang-format prefers something closer to the formatting I originally used; I'll raise this with the clang-format folks.

3130–3134

It could in principle. In practice, Clang always aligns function bodies to 16 bytes, GCC does it at -O>=2 and GCC (4.4 and 4.6) codegens an empty function body at -O0 with >4 bytes. A quick survey of a number of system libraries (libc, libm, libstdc++) on an Ubuntu machine reveals that the functions are suitably aligned. I think the circumstances in which a segfault might occur are sufficiently rare as to justify doing a single load. If it does turn out to be a problem in practice we could consider splitting the load (perhaps conditional on a command line flag).

3147–3148

Done.

lib/CodeGen/CodeGenFunction.cpp

523–524

I like the idea of doing this as an option, but I would prefer to land this version first.

529–530

Not with the Itanium ABI, which does not have a representation for calling conventions, but the Microsoft ABI (for which calling conventions matter more) does, so once RTTI has been implemented for that ABI, this should respect calling conventions there.

531

Done.

lib/CodeGen/TargetInfo.cpp

602–603

"rex.RX push %rsp", according to objdump. But I don't think it really matters what this decodes to. If the instruction pointer finds itself here the situation isn't dissimilar to jumping into the middle of a multibyte instruction. (Besides, the RTTI pointer could decode to anything.)

[rsmith]

LGTM

lib/CodeGen/CGExpr.cpp

3120

It'd be nice if we could erase these checks if the optimizer resolves the call.

lib/CodeGen/TargetInfo.cpp

602–603

I think it matters: it's not unreasonable for a function to start as:

jmp +8
6 byte loop body
check loop condition
conditional jmp back to loop body

IIUC, your check would flag a false positive on indirect calls to such a function.

However, since the 'F' is an instruction prefix that is redundant on a 'T' (pushq %rsp) instruction, I think it's reasonable to assume that the FT signature will not occur in uninstrumented code.

[pcc]

What about the compiler-rt changes?

lib/CodeGen/CGExpr.cpp

3120

Sure. A design point of function prefix data is that it permits this sort of optimization. Maybe if I get a chance I'll see if I can teach the optimizer to do this.

lib/CodeGen/TargetInfo.cpp

602–603

I see. I think I agree with your reasoning.

[pcc]

Closed by commit rL193058 (authored by @pcc).

docs/UsersManual.rst

**-f[no-]sanitize=check1,check2,...**
    -  ``-fsanitize=enum``: Load of a value of an enumerated type which
       is not in the range of representable values for that enumerated
       type.
    -  ``-fsanitize=float-cast-overflow``: Conversion to, from, or
       between floating-point types which would overflow the
       destination.
    -  ``-fsanitize=float-divide-by-zero``: Floating point division by
       zero.
+   -  ``-fsanitize=function``: Indirect call of a function through a
+      function pointer of the wrong type (C++ and x86/x86_64 only).
    -  ``-fsanitize=integer-divide-by-zero``: Integer division by zero.
    -  ``-fsanitize=null``: Use of a null pointer or creation of a null
       reference.
    -  ``-fsanitize=object-size``: An attempt to use bytes which the
       optimizer can determine are not part of the object being
       accessed. The sizes of objects are determined using
       ``__builtin_object_size``, and consequently may be able to detect
       more problems at higher optimization levels.

include/clang/Basic/Sanitizers.def

 
 // UndefinedBehaviorSanitizer
 SANITIZER("alignment", Alignment)
 SANITIZER("bool", Bool)
 SANITIZER("bounds", Bounds)
 SANITIZER("enum", Enum)
 SANITIZER("float-cast-overflow", FloatCastOverflow)
 SANITIZER("float-divide-by-zero", FloatDivideByZero)
+SANITIZER("function", Function)
 SANITIZER("integer-divide-by-zero", IntegerDivideByZero)
 SANITIZER("null", Null)
 SANITIZER("object-size", ObjectSize)
 SANITIZER("return", Return)
 SANITIZER("shift", Shift)
 SANITIZER("signed-integer-overflow", SignedIntegerOverflow)
 SANITIZER("unreachable", Unreachable)
 SANITIZER("vla-bound", VLABound)
 SANITIZER("vptr", Vptr)
 
 // IntegerSanitizer
 SANITIZER("unsigned-integer-overflow", UnsignedIntegerOverflow)
 
 // DataFlowSanitizer
 SANITIZER("dataflow", DataFlow)
 
 // -fsanitize=undefined includes all the sanitizers which have low overhead, no
 // ABI or address space layout implications, and only catch undefined behavior.
 SANITIZER_GROUP("undefined", Undefined,
                 Alignment | Bool | Bounds | Enum | FloatCastOverflow |
-                FloatDivideByZero | IntegerDivideByZero | Null | ObjectSize |
-                Return | Shift | SignedIntegerOverflow | Unreachable |
-                VLABound | Vptr)
+                FloatDivideByZero | Function | IntegerDivideByZero | Null |
+                ObjectSize | Return | Shift | SignedIntegerOverflow |
+                Unreachable | VLABound | Vptr)
 
 // -fsanitize=undefined-trap (and its alias -fcatch-undefined-behavior) includes
 // all sanitizers included by -fsanitize=undefined, except those that require
 // runtime support.  This group is generally used in conjunction with the
 // -fsanitize-undefined-trap-on-error flag.
 SANITIZER_GROUP("undefined-trap", UndefinedTrap,
                 Alignment | Bool | Bounds | Enum | FloatCastOverflow |
                 FloatDivideByZero | IntegerDivideByZero | Null | ObjectSize |

lib/CodeGen/CGBuiltin.cpp

llvm::FunctionType *FT = llvm::FunctionType::get(V->getType(), V->getType(),
                                                    false);
   llvm::Value *Fn = CGF.CGM.CreateRuntimeFunction(FT, FnName);
 
   return CGF.EmitNounwindRuntimeCall(Fn, V, "abs");
 }
 
 static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *Fn,
                               const CallExpr *E, llvm::Value *calleeValue) {
-  return CGF.EmitCall(E->getCallee()->getType(), calleeValue,
+  return CGF.EmitCall(E->getCallee()->getType(), calleeValue, E->getLocStart(),
                       ReturnValueSlot(), E->arg_begin(), E->arg_end(), Fn);
 }
 
 /// \brief Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.*
 /// depending on IntrinsicID.
 ///
 /// \arg CGF The current codegen function.
 /// \arg IntrinsicID The ID for the Intrinsic we wish to generate.

lib/CodeGen/CGCUDARuntime.cpp

RValue CGCUDARuntime::EmitCUDAKernelCallExpr(CodeGenFunction &CGF,
   const Decl *TargetDecl = 0;
   if (const ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E->getCallee())) {
     if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(CE->getSubExpr())) {
       TargetDecl = DRE->getDecl();
     }
   }
 
   llvm::Value *Callee = CGF.EmitScalarExpr(E->getCallee());
-  CGF.EmitCall(E->getCallee()->getType(), Callee, ReturnValue,
-               E->arg_begin(), E->arg_end(), TargetDecl);
+  CGF.EmitCall(E->getCallee()->getType(), Callee, E->getLocStart(),
+               ReturnValue, E->arg_begin(), E->arg_end(), TargetDecl);
   CGF.EmitBranch(ContBlock);
 
   CGF.EmitBlock(ContBlock);
   eval.end(CGF);
 
   return RValue::get(0);
 }

lib/CodeGen/CGExpr.cpp

if (getLangOpts().ObjCAutoRefCount &&
       //   arrow.
       EmitScalarExpr(E->getCallee());
     }
 
     return RValue::get(0);
   }
 
   llvm::Value *Callee = EmitScalarExpr(E->getCallee());
-  return EmitCall(E->getCallee()->getType(), Callee, ReturnValue,
-                  E->arg_begin(), E->arg_end(), TargetDecl);
+  return EmitCall(E->getCallee()->getType(), Callee, E->getLocStart(),
+                  ReturnValue, E->arg_begin(), E->arg_end(), TargetDecl);
 }
 
 LValue CodeGenFunction::EmitBinaryOperatorLValue(const BinaryOperator *E) {
   // Comma expressions just emit their LHS then their RHS as an l-value.
   if (E->getOpcode() == BO_Comma) {
     EmitIgnoredExpr(E->getLHS());
     EnsureInsertPoint();
     return EmitLValue(E->getRHS());

 
 LValue CodeGenFunction::EmitStmtExprLValue(const StmtExpr *E) {
   // Can only get l-value for message expression returning aggregate type
   RValue RV = EmitAnyExprToTemp(E);
   return MakeAddrLValue(RV.getAggregateAddr(), E->getType());
 }
 
 RValue CodeGenFunction::EmitCall(QualType CalleeType, llvm::Value *Callee,
+                                 SourceLocation CallLoc,
                                  ReturnValueSlot ReturnValue,
                                  CallExpr::const_arg_iterator ArgBeg,
                                  CallExpr::const_arg_iterator ArgEnd,
                                  const Decl *TargetDecl) {
   // Get the actual function type. The callee type will always be a pointer to
   // function type or a block pointer type.
   assert(CalleeType->isFunctionPointerType() &&
          "Call must have function pointer type!");
 
   CalleeType = getContext().getCanonicalType(CalleeType);
 
   const FunctionType *FnType
     = cast<FunctionType>(cast<PointerType>(CalleeType)->getPointeeType());
 
   // Force column info to differentiate multiple inlined call sites on
   // the same line, analoguous to EmitCallExpr.
   bool ForceColumnInfo = false;
   if (const FunctionDecl* FD = dyn_cast_or_null<const FunctionDecl>(TargetDecl))
     ForceColumnInfo = FD->isInlineSpecified();
 
+  if (getLangOpts().CPlusPlus && SanOpts->Function &&
+      !dyn_cast_or_null<FunctionDecl>(TargetDecl)) {
+    if (llvm::Constant* PDSig =
+            CGM.getTargetCodeGenInfo().getUBSanFunctionSignature(CGM)) {
+      llvm::Constant *FTRTTIConst =
+          CGM.GetAddrOfRTTIDescriptor(QualType(FnType, 0), /*ForEH=*/true);
+      llvm::Type *PDStructTyElems[] = {PDSig->getType(),
+                                       FTRTTIConst->getType()};
+      llvm::StructType *PDStructTy = llvm::StructType::get(
+          CGM.getLLVMContext(), PDStructTyElems, /*isPacked=*/true);
+
+      llvm::Value *CalleePDStruct = Builder.CreateBitCast(
+          Callee, llvm::PointerType::getUnqual(PDStructTy));
+      llvm::Value *CalleeSigPtr =
+          Builder.CreateConstGEP2_32(CalleePDStruct, 0, 0);
+      llvm::Value *CalleeSig = Builder.CreateLoad(CalleeSigPtr);
+      llvm::Value *CalleeSigMatch = Builder.CreateICmpEQ(CalleeSig, PDSig);
+
+      llvm::BasicBlock *Cont = createBasicBlock("cont");
+      llvm::BasicBlock *TypeCheck = createBasicBlock("typecheck");
+      Builder.CreateCondBr(CalleeSigMatch, TypeCheck, Cont);
+
+      EmitBlock(TypeCheck);
+      llvm::Value *CalleeRTTIPtr =
+          Builder.CreateConstGEP2_32(CalleePDStruct, 0, 1);
+      llvm::Value *CalleeRTTI = Builder.CreateLoad(CalleeRTTIPtr);
+      llvm::Value *CalleeRTTIMatch =
+          Builder.CreateICmpEQ(CalleeRTTI, FTRTTIConst);
+      llvm::Constant *StaticData[] = {EmitCheckSourceLocation(CallLoc),
+                                      EmitCheckTypeDescriptor(CalleeType)};
+      EmitCheck(CalleeRTTIMatch,
+                "function_type_mismatch",
+                StaticData,
+                Callee,
+                CRK_Recoverable);
+
+      Builder.CreateBr(Cont);
+      EmitBlock(Cont);
+    }
+  }
+
   CallArgList Args;
   EmitCallArgs(Args, dyn_cast<FunctionProtoType>(FnType), ArgBeg, ArgEnd,
                ForceColumnInfo);
 
   const CGFunctionInfo &FnInfo =
     CGM.getTypes().arrangeFreeFunctionCall(Args, FnType);
 
   // C99 6.5.2.2p6:

lib/CodeGen/CGExprCXX.cpp

RValue CodeGenFunction::EmitCXXMemberCallExpr(const CXXMemberCallExpr *CE,
 
   const MemberExpr *ME = cast<MemberExpr>(callee);
   const CXXMethodDecl *MD = cast<CXXMethodDecl>(ME->getMemberDecl());
 
   if (MD->isStatic()) {
     // The method is static, emit it as we would a regular call.
     llvm::Value *Callee = CGM.GetAddrOfFunction(MD);
     return EmitCall(getContext().getPointerType(MD->getType()), Callee,
-                    ReturnValue, CE->arg_begin(), CE->arg_end());
+                    CE->getLocStart(), ReturnValue, CE->arg_begin(),
+                    CE->arg_end());
   }
 
   // Compute the object pointer.
   const Expr *Base = ME->getBase();
   bool CanUseVirtualCall = MD->isVirtual() && !ME->hasQualifier();
 
   const CXXMethodDecl *DevirtualizedMethod = NULL;
   if (CanUseVirtualCall && CanDevirtualizeMemberFunctionCall(Base, MD)) {

lib/CodeGen/CodeGenFunction.h

#endif
   RValue EmitCall(const CGFunctionInfo &FnInfo,
                   llvm::Value *Callee,
                   ReturnValueSlot ReturnValue,
                   const CallArgList &Args,
                   const Decl *TargetDecl = 0,
                   llvm::Instruction **callOrInvoke = 0);
 
   RValue EmitCall(QualType FnType, llvm::Value *Callee,
+                  SourceLocation CallLoc,
                   ReturnValueSlot ReturnValue,
                   CallExpr::const_arg_iterator ArgBeg,
                   CallExpr::const_arg_iterator ArgEnd,
                   const Decl *TargetDecl = 0);
   RValue EmitCallExpr(const CallExpr *E,
                       ReturnValueSlot ReturnValue = ReturnValueSlot());
 
   llvm::CallInst *EmitRuntimeCall(llvm::Value *callee,

lib/CodeGen/CodeGenFunction.cpp

 //
 //===----------------------------------------------------------------------===//
 
 #include "CodeGenFunction.h"
 #include "CGCUDARuntime.h"
 #include "CGCXXABI.h"
 #include "CGDebugInfo.h"
 #include "CodeGenModule.h"
+#include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/Decl.h"
 #include "clang/AST/DeclCXX.h"
 #include "clang/AST/StmtCXX.h"
 #include "clang/Basic/OpenCL.h"
 #include "clang/Basic/TargetInfo.h"
 #include "clang/Frontend/CodeGenOptions.h"
 #include "llvm/IR/DataLayout.h"
if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D))
         }
 
   if (getLangOpts().OpenCL) {
     // Add metadata for a kernel function.
     if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D))
       EmitOpenCLKernelMetadata(FD, Fn);
   }
 
+  // If we are checking function types, emit a function type signature as
+  // prefix data.
+  if (getLangOpts().CPlusPlus && SanOpts->Function) {
+    if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D)) {
+      if (llvm::Constant *PDSig =
+              CGM.getTargetCodeGenInfo().getUBSanFunctionSignature(CGM)) {
+        llvm::Constant *FTRTTIConst =
+            CGM.GetAddrOfRTTIDescriptor(FD->getType(), /*ForEH=*/true);
+        llvm::Constant *PDStructElems[] = {PDSig, FTRTTIConst};
+        llvm::Constant *PDStructConst =
+            llvm::ConstantStruct::getAnon(PDStructElems, /*Packed=*/true);
+        Fn->setPrefixData(PDStructConst);
+      }
+    }
+  }
+
   llvm::BasicBlock *EntryBB = createBasicBlock("entry", CurFn);
 
   // Create a marker to make it easy to insert allocas into the entryblock
   // later.  Don't create this with the builder, because we don't want it
   // folded.
   llvm::Value *Undef = llvm::UndefValue::get(Int32Ty);
   AllocaInsertPt = new llvm::BitCastInst(Undef, Int32Ty, "", EntryBB);
   if (Builder.isNamePreserving())

lib/CodeGen/TargetInfo.h

 #define CLANG_CODEGEN_TARGETINFO_H
 
 #include "clang/AST/Type.h"
 #include "clang/Basic/LLVM.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/SmallString.h"
 
 namespace llvm {
+  class Constant;
   class GlobalValue;
   class Type;
   class Value;
 }
 
 namespace clang {
   class ABIInfo;
   class Decl;
public:
     /// relocation model, and so on some targets it instead sniffs for
     /// a particular instruction sequence.  This functions returns
     /// that instruction sequence in inline assembly, which will be
     /// empty if none is required.
     virtual StringRef getARCRetainAutoreleasedReturnValueMarker() const {
       return "";
     }
 
+    /// Return a constant used by UBSan as a signature to identify functions
+    /// possessing type information, or 0 if the platform is unsupported.
+    virtual llvm::Constant *getUBSanFunctionSignature(
+        CodeGen::CodeGenModule &CGM) const {
+      return 0;
+    }
+
     /// Determine whether a call to an unprototyped functions under
     /// the given calling convention should use the variadic
     /// convention or the non-variadic convention.
     ///
     /// There's a good reason to make a platform's variadic calling
     /// convention be different from its non-variadic calling
     /// convention: the non-variadic arguments can be passed in
     /// registers (better for performance), and the variadic arguments

lib/CodeGen/TargetInfo.cpp

bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                llvm::Value *Address) const;
 
   llvm::Type* adjustInlineAsmType(CodeGen::CodeGenFunction &CGF,
                                   StringRef Constraint,
                                   llvm::Type* Ty) const {
     return X86AdjustInlineAsmType(CGF, Constraint, Ty);
   }
 
+  llvm::Constant *getUBSanFunctionSignature(CodeGen::CodeGenModule &CGM) const {
+    unsigned Sig = (0xeb << 0) |  // jmp rel8
+                   (0x06 << 8) |  //           .+0x08
+                   ('F' << 16) |
+                   ('T' << 24);
+    return llvm::ConstantInt::get(CGM.Int32Ty, Sig);
+  }
+
 };
 
 }
 
 /// shouldReturnTypeInRegister - Determine if the given type should be
 /// passed in a register (for the Darwin ABI).
 bool X86_32ABIInfo::shouldReturnTypeInRegister(QualType Ty,
                                                ASTContext &Context,
if (fnType->getCallConv() == CC_C) {
 
       if (!HasAVXType)
         return true;
     }
 
     return TargetCodeGenInfo::isNoProtoCallVariadic(args, fnType);
   }
 
+  llvm::Constant *getUBSanFunctionSignature(CodeGen::CodeGenModule &CGM) const {
+    unsigned Sig = (0xeb << 0) |  // jmp rel8
+                   (0x0a << 8) |  //           .+0x0c
+                   ('F' << 16) |
+                   ('T' << 24);
+    return llvm::ConstantInt::get(CGM.Int32Ty, Sig);
+  }
+
 };
 
 static std::string qualifyWindowsLibrary(llvm::StringRef Lib) {
   // If the argument does not end in .lib, automatically add the suffix. This
   // matches the behavior of MSVC.
   std::string ArgStr = Lib;
   if (Lib.size() <= 4 ||
       Lib.substr(Lib.size() - 4).compare_lower(".lib") != 0) {

test/CodeGenCXX/catch-undef-behavior.cpp

-// RUN: %clang_cc1 -std=c++11 -fsanitize=signed-integer-overflow,integer-divide-by-zero,float-divide-by-zero,shift,unreachable,return,vla-bound,alignment,null,vptr,object-size,float-cast-overflow,bool,enum,bounds -emit-llvm %s -o - -triple x86_64-linux-gnu | FileCheck %s
+// RUN: %clang_cc1 -std=c++11 -fsanitize=signed-integer-overflow,integer-divide-by-zero,float-divide-by-zero,shift,unreachable,return,vla-bound,alignment,null,vptr,object-size,float-cast-overflow,bool,enum,bounds,function -emit-llvm %s -o - -triple x86_64-linux-gnu | FileCheck %s
 
 struct S {
   double d;
   int a, b;
   virtual int f();
 };
 
 struct T : S {};
void downcast_reference(B &b) {
   // CHECK:      [[C_INT:%[0-9]*]] = ptrtoint %class.C* [[C]] to i64
   // CHECK-NEXT: [[MASKED:%[0-9]*]] = and i64 [[C_INT]], 15
   // CHECK-NEXT: [[TEST:%[0-9]*]] = icmp eq i64 [[MASKED]], 0
   // AND the alignment test with the objectsize test.
   // CHECK-NEXT: [[AND:%[0-9]*]] = and i1 {{.*}}, [[TEST]]
   // CHECK-NEXT: br i1 [[AND]]
 }
 
+// CHECK-LABEL: @_Z22indirect_function_callPFviE({{.*}} prefix <{ i32, i8* }> <{ i32 1413876459, i8* bitcast ({ i8*, i8* }* @_ZTIFvPFviEE to i8*) }>
+void indirect_function_call(void (*p)(int)) {
+  // CHECK: [[PTR:%[0-9]*]] = bitcast void (i32)* {{.*}} to <{ i32, i8* }>*
+
+  // Signature check
+  // CHECK-NEXT: [[SIGPTR:%[0-9]*]] = getelementptr <{ i32, i8* }>* [[PTR]], i32 0, i32 0
+  // CHECK-NEXT: [[SIG:%[0-9]*]] = load i32* [[SIGPTR]]
+  // CHECK-NEXT: [[SIGCMP:%[0-9]*]] = icmp eq i32 [[SIG]], 1413876459
+  // CHECK-NEXT: br i1 [[SIGCMP]]
+
+  // RTTI pointer check
+  // CHECK: [[RTTIPTR:%[0-9]*]] = getelementptr <{ i32, i8* }>* [[PTR]], i32 0, i32 1
+  // CHECK-NEXT: [[RTTI:%[0-9]*]] = load i8** [[RTTIPTR]]
+  // CHECK-NEXT: [[RTTICMP:%[0-9]*]] = icmp eq i8* [[RTTI]], bitcast ({ i8*, i8* }* @_ZTIFviE to i8*)
+  // CHECK-NEXT: br i1 [[RTTICMP]]
+  p(42);
+}
+
 namespace CopyValueRepresentation {
   // CHECK-LABEL: define {{.*}} @_ZN23CopyValueRepresentation2S3aSERKS0_
   // CHECK-NOT: call {{.*}} @__ubsan_handle_load_invalid_value
   // CHECK-LABEL: define {{.*}} @_ZN23CopyValueRepresentation2S4aSEOS0_
   // CHECK-NOT: call {{.*}} @__ubsan_handle_load_invalid_value
   // CHECK-LABEL: define {{.*}} @_ZN23CopyValueRepresentation2S5C2ERKS0_
   // CHECK-NOT: call {{.*}} __ubsan_handle_load_invalid_value
   // CHECK-LABEL: define {{.*}} @_ZN23CopyValueRepresentation2S2C2ERKS0_

test/Driver/fsanitize.c

 // RUN: %clang -target x86_64-linux-gnu -fcatch-undefined-behavior %s -### 2>&1 | FileCheck %s --check-prefix=CHECK-UNDEFINED-TRAP
 // RUN: %clang -target x86_64-linux-gnu -fsanitize=undefined-trap -fsanitize-undefined-trap-on-error %s -### 2>&1 | FileCheck %s --check-prefix=CHECK-UNDEFINED-TRAP
 // RUN: %clang -target x86_64-linux-gnu -fsanitize-undefined-trap-on-error -fsanitize=undefined-trap %s -### 2>&1 | FileCheck %s --check-prefix=CHECK-UNDEFINED-TRAP
 // CHECK-UNDEFINED-TRAP: "-fsanitize={{((signed-integer-overflow|integer-divide-by-zero|float-divide-by-zero|shift|unreachable|return|vla-bound|alignment|null|object-size|float-cast-overflow|bounds|enum|bool),?){14}"}}
 // CHECK-UNDEFINED-TRAP: "-fsanitize-undefined-trap-on-error"
 
 // RUN: %clang -target x86_64-linux-gnu -fsanitize=undefined %s -### 2>&1 | FileCheck %s --check-prefix=CHECK-UNDEFINED
-// CHECK-UNDEFINED: "-fsanitize={{((signed-integer-overflow|integer-divide-by-zero|float-divide-by-zero|shift|unreachable|return|vla-bound|alignment|null|vptr|object-size|float-cast-overflow|bounds|enum|bool),?){15}"}}
+// CHECK-UNDEFINED: "-fsanitize={{((signed-integer-overflow|integer-divide-by-zero|float-divide-by-zero|function|shift|unreachable|return|vla-bound|alignment|null|vptr|object-size|float-cast-overflow|bounds|enum|bool),?){16}"}}
 
 // RUN: %clang -target x86_64-linux-gnu -fsanitize=integer %s -### 2>&1 | FileCheck %s --check-prefix=CHECK-INTEGER
 // CHECK-INTEGER: "-fsanitize={{((signed-integer-overflow|unsigned-integer-overflow|integer-divide-by-zero|shift),?){4}"}}
 
 // RUN: %clang -target x86_64-linux-gnu -fsanitize=thread,undefined -fno-thread-sanitizer -fno-sanitize=float-cast-overflow,vptr,bool,enum %s -### 2>&1 | FileCheck %s --check-prefix=CHECK-PARTIAL-UNDEFINED
-// CHECK-PARTIAL-UNDEFINED: "-fsanitize={{((signed-integer-overflow|integer-divide-by-zero|float-divide-by-zero|shift|unreachable|return|vla-bound|alignment|null|object-size|bounds),?){11}"}}
+// CHECK-PARTIAL-UNDEFINED: "-fsanitize={{((signed-integer-overflow|integer-divide-by-zero|float-divide-by-zero|function|shift|unreachable|return|vla-bound|alignment|null|object-size|bounds),?){12}"}}
 
 // RUN: %clang -target x86_64-linux-gnu -fsanitize=address-full %s -### 2>&1 | FileCheck %s --check-prefix=CHECK-ASAN-FULL
 // CHECK-ASAN-FULL: "-fsanitize={{((address|init-order|use-after-return|use-after-scope),?){4}"}}
 
 // RUN: %clang -target x86_64-linux-gnu -fno-sanitize=init-order -fsanitize=address %s -### 2>&1 |  FileCheck %s --check-prefix=CHECK-ASAN-IMPLIED-INIT-ORDER
 // CHECK-ASAN-IMPLIED-INIT-ORDER: "-fsanitize={{((address|init-order),?){2}"}}
 
 // RUN: %clang -target x86_64-linux-gnu -fsanitize=address -fno-sanitize=init-order %s -### 2>&1 |  FileCheck %s --check-prefix=CHECK-ASAN-NO-IMPLIED-INIT-ORDER