[ubsan] Add -fsanitize-merge (and -fno-sanitize-merge) (llvm#120464)

'-mllvm -ubsan-unique-traps'
(llvm#65972) applies to all UBSan
checks. This patch introduces -fsanitize-merge (defaults to on,
maintaining the status quo behavior) and -fno-sanitize-merge (equivalent
to '-mllvm -ubsan-unique-traps'), with the option to selectively
applying non-merged handlers to a subset of UBSan checks (e.g.,
-fno-sanitize-merge=bool,enum).

N.B. we do not use "trap" in the argument name since
llvm#119302 has generalized
-ubsan-unique-traps to work for non-trap modes (min-rt and regular rt).

This patch does not remove the -ubsan-unique-traps flag; that will
override -f(no-)sanitize-merge.

clang/include/clang/Basic/CodeGenOptions.h

@@ -380,6 +380,10 @@ class CodeGenOptions : public CodeGenOptionsBase {
   /// Set of sanitizer checks that trap rather than diagnose.
   SanitizerSet SanitizeTrap;
 
+  /// Set of sanitizer checks that can merge handlers (smaller code size at
+  /// the expense of debuggability).
+  SanitizerSet SanitizeMergeHandlers;
+
   /// List of backend command-line options for -fembed-bitcode.
   std::vector<uint8_t> CmdArgs;
 

clang/include/clang/Driver/Options.td

@@ -2555,6 +2555,21 @@ def fno_sanitize_trap : Flag<["-"], "fno-sanitize-trap">, Group<f_clang_Group>,
                         Alias<fno_sanitize_trap_EQ>, AliasArgs<["all"]>,
                         Visibility<[ClangOption, CLOption]>,
                         HelpText<"Disable trapping for all sanitizers">;
+def fsanitize_merge_handlers_EQ
+    : CommaJoined<["-"], "fsanitize-merge=">,
+      Group<f_clang_Group>,
+      HelpText<"Allow compiler to merge handlers for specified sanitizers">;
+def fno_sanitize_merge_handlers_EQ
+    : CommaJoined<["-"], "fno-sanitize-merge=">,
+      Group<f_clang_Group>,
+      HelpText<"Do not allow compiler to merge handlers for specified sanitizers">;
+def fsanitize_merge_handlers : Flag<["-"], "fsanitize-merge">, Group<f_clang_Group>,
+                     Alias<fsanitize_merge_handlers_EQ>, AliasArgs<["all"]>,
+                     HelpText<"Allow compiler to merge handlers for all sanitizers">;
+def fno_sanitize_merge_handlers : Flag<["-"], "fno-sanitize-merge">, Group<f_clang_Group>,
+                        Alias<fno_sanitize_merge_handlers_EQ>, AliasArgs<["all"]>,
+                        Visibility<[ClangOption, CLOption]>,
+                        HelpText<"Do not allow compiler to merge handlers for any sanitizers">;
 def fsanitize_undefined_trap_on_error
     : Flag<["-"], "fsanitize-undefined-trap-on-error">, Group<f_clang_Group>,
       Alias<fsanitize_trap_EQ>, AliasArgs<["undefined"]>;

clang/include/clang/Driver/SanitizerArgs.h

@@ -25,6 +25,7 @@ class SanitizerArgs {
   SanitizerSet Sanitizers;
   SanitizerSet RecoverableSanitizers;
   SanitizerSet TrapSanitizers;
+  SanitizerSet MergeHandlers;
 
   std::vector<std::string> UserIgnorelistFiles;
   std::vector<std::string> SystemIgnorelistFiles;

clang/lib/CodeGen/CGExpr.cpp

@@ -3546,7 +3546,7 @@ static void emitCheckHandlerCall(CodeGenFunction &CGF,
                                  ArrayRef<llvm::Value *> FnArgs,
                                  SanitizerHandler CheckHandler,
                                  CheckRecoverableKind RecoverKind, bool IsFatal,
-                                 llvm::BasicBlock *ContBB) {
+                                 llvm::BasicBlock *ContBB, bool NoMerge) {
   assert(IsFatal || RecoverKind != CheckRecoverableKind::Unrecoverable);
   std::optional<ApplyDebugLocation> DL;
   if (!CGF.Builder.getCurrentDebugLocation()) {
@@ -3581,10 +3581,9 @@ static void emitCheckHandlerCall(CodeGenFunction &CGF,
                                llvm::AttributeList::FunctionIndex, B),
       /*Local=*/true);
   llvm::CallInst *HandlerCall = CGF.EmitNounwindRuntimeCall(Fn, FnArgs);
-  bool NoMerge =
-      ClSanitizeDebugDeoptimization ||
-      !CGF.CGM.getCodeGenOpts().OptimizationLevel ||
-      (CGF.CurCodeDecl && CGF.CurCodeDecl->hasAttr<OptimizeNoneAttr>());
+  NoMerge = NoMerge || ClSanitizeDebugDeoptimization ||
+            !CGF.CGM.getCodeGenOpts().OptimizationLevel ||
+            (CGF.CurCodeDecl && CGF.CurCodeDecl->hasAttr<OptimizeNoneAttr>());
   if (NoMerge)
     HandlerCall->addFnAttr(llvm::Attribute::NoMerge);
   if (!MayReturn) {
@@ -3608,6 +3607,7 @@ void CodeGenFunction::EmitCheck(
   llvm::Value *FatalCond = nullptr;
   llvm::Value *RecoverableCond = nullptr;
   llvm::Value *TrapCond = nullptr;
+  bool NoMerge = false;
   for (int i = 0, n = Checked.size(); i < n; ++i) {
     llvm::Value *Check = Checked[i].first;
     // -fsanitize-trap= overrides -fsanitize-recover=.
@@ -3618,6 +3618,9 @@ void CodeGenFunction::EmitCheck(
                   ? RecoverableCond
                   : FatalCond;
     Cond = Cond ? Builder.CreateAnd(Cond, Check) : Check;
+
+    if (!CGM.getCodeGenOpts().SanitizeMergeHandlers.has(Checked[i].second))
+      NoMerge = true;
   }
 
   if (ClSanitizeGuardChecks) {
@@ -3632,7 +3635,7 @@ void CodeGenFunction::EmitCheck(
   }
 
   if (TrapCond)
-    EmitTrapCheck(TrapCond, CheckHandler);
+    EmitTrapCheck(TrapCond, CheckHandler, NoMerge);
   if (!FatalCond && !RecoverableCond)
     return;
 
@@ -3698,7 +3701,7 @@ void CodeGenFunction::EmitCheck(
     // Simple case: we need to generate a single handler call, either
     // fatal, or non-fatal.
     emitCheckHandlerCall(*this, FnType, Args, CheckHandler, RecoverKind,
-                         (FatalCond != nullptr), Cont);
+                         (FatalCond != nullptr), Cont, NoMerge);
   } else {
     // Emit two handler calls: first one for set of unrecoverable checks,
     // another one for recoverable.
@@ -3708,10 +3711,10 @@ void CodeGenFunction::EmitCheck(
     Builder.CreateCondBr(FatalCond, NonFatalHandlerBB, FatalHandlerBB);
     EmitBlock(FatalHandlerBB);
     emitCheckHandlerCall(*this, FnType, Args, CheckHandler, RecoverKind, true,
-                         NonFatalHandlerBB);
+                         NonFatalHandlerBB, NoMerge);
     EmitBlock(NonFatalHandlerBB);
     emitCheckHandlerCall(*this, FnType, Args, CheckHandler, RecoverKind, false,
-                         Cont);
+                         Cont, NoMerge);
   }
 
   EmitBlock(Cont);
@@ -3901,7 +3904,8 @@ void CodeGenFunction::EmitUnreachable(SourceLocation Loc) {
 }
 
 void CodeGenFunction::EmitTrapCheck(llvm::Value *Checked,
-                                    SanitizerHandler CheckHandlerID) {
+                                    SanitizerHandler CheckHandlerID,
+                                    bool NoMerge) {
   llvm::BasicBlock *Cont = createBasicBlock("cont");
 
   // If we're optimizing, collapse all calls to trap down to just one per
@@ -3911,9 +3915,9 @@ void CodeGenFunction::EmitTrapCheck(llvm::Value *Checked,
 
   llvm::BasicBlock *&TrapBB = TrapBBs[CheckHandlerID];
 
-  bool NoMerge = ClSanitizeDebugDeoptimization ||
-                 !CGM.getCodeGenOpts().OptimizationLevel ||
-                 (CurCodeDecl && CurCodeDecl->hasAttr<OptimizeNoneAttr>());
+  NoMerge = NoMerge || ClSanitizeDebugDeoptimization ||
+            !CGM.getCodeGenOpts().OptimizationLevel ||
+            (CurCodeDecl && CurCodeDecl->hasAttr<OptimizeNoneAttr>());
 
   if (TrapBB && !NoMerge) {
     auto Call = TrapBB->begin();

clang/lib/CodeGen/CodeGenFunction.h

@@ -5171,7 +5171,8 @@ class CodeGenFunction : public CodeGenTypeCache {
 
   /// Create a basic block that will call the trap intrinsic, and emit a
   /// conditional branch to it, for the -ftrapv checks.
-  void EmitTrapCheck(llvm::Value *Checked, SanitizerHandler CheckHandlerID);
+  void EmitTrapCheck(llvm::Value *Checked, SanitizerHandler CheckHandlerID,
+                     bool NoMerge = false);
 
   /// Emit a call to trap or debugtrap and attach function attribute
   /// "trap-func-name" if specified.

clang/lib/Driver/SanitizerArgs.cpp

@@ -68,6 +68,7 @@ static const SanitizerMask TrappingSupported =
     SanitizerKind::ImplicitConversion | SanitizerKind::Nullability |
     SanitizerKind::LocalBounds | SanitizerKind::CFI |
     SanitizerKind::FloatDivideByZero | SanitizerKind::ObjCCast;
+static const SanitizerMask MergeDefault = SanitizerKind::Undefined;
 static const SanitizerMask TrappingDefault = SanitizerKind::CFI;
 static const SanitizerMask CFIClasses =
     SanitizerKind::CFIVCall | SanitizerKind::CFINVCall |
@@ -696,6 +697,13 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
   TrappingKinds &= Kinds;
   RecoverableKinds &= ~TrappingKinds;
 
+  // Parse -f(no-)?sanitize-nonmerged-handlers flags
+  SanitizerMask MergeKinds =
+      parseSanitizeArgs(D, Args, DiagnoseErrors, MergeDefault, {}, {},
+                        options::OPT_fsanitize_merge_handlers_EQ,
+                        options::OPT_fno_sanitize_merge_handlers_EQ);
+  MergeKinds &= Kinds;
+
   // Setup ignorelist files.
   // Add default ignorelist from resource directory for activated sanitizers,
   // and validate special case lists format.
@@ -1114,6 +1122,8 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
   TrapSanitizers.Mask |= TrappingKinds;
   assert(!(RecoverableKinds & TrappingKinds) &&
          "Overlap between recoverable and trapping sanitizers");
+
+  MergeHandlers.Mask |= MergeKinds;
 }
 
 static std::string toString(const clang::SanitizerSet &Sanitizers) {
@@ -1275,6 +1285,10 @@ void SanitizerArgs::addArgs(const ToolChain &TC, const llvm::opt::ArgList &Args,
     CmdArgs.push_back(
         Args.MakeArgString("-fsanitize-trap=" + toString(TrapSanitizers)));
 
+  if (!MergeHandlers.empty())
+    CmdArgs.push_back(
+        Args.MakeArgString("-fsanitize-merge=" + toString(MergeHandlers)));
+
   addSpecialCaseListOpt(Args, CmdArgs,
                         "-fsanitize-ignorelist=", UserIgnorelistFiles);
   addSpecialCaseListOpt(Args, CmdArgs,
@@ -1442,13 +1456,16 @@ void SanitizerArgs::addArgs(const ToolChain &TC, const llvm::opt::ArgList &Args,
 
 SanitizerMask parseArgValues(const Driver &D, const llvm::opt::Arg *A,
                              bool DiagnoseErrors) {
-  assert((A->getOption().matches(options::OPT_fsanitize_EQ) ||
-          A->getOption().matches(options::OPT_fno_sanitize_EQ) ||
-          A->getOption().matches(options::OPT_fsanitize_recover_EQ) ||
-          A->getOption().matches(options::OPT_fno_sanitize_recover_EQ) ||
-          A->getOption().matches(options::OPT_fsanitize_trap_EQ) ||
-          A->getOption().matches(options::OPT_fno_sanitize_trap_EQ)) &&
-         "Invalid argument in parseArgValues!");
+  assert(
+      (A->getOption().matches(options::OPT_fsanitize_EQ) ||
+       A->getOption().matches(options::OPT_fno_sanitize_EQ) ||
+       A->getOption().matches(options::OPT_fsanitize_recover_EQ) ||
+       A->getOption().matches(options::OPT_fno_sanitize_recover_EQ) ||
+       A->getOption().matches(options::OPT_fsanitize_trap_EQ) ||
+       A->getOption().matches(options::OPT_fno_sanitize_trap_EQ) ||
+       A->getOption().matches(options::OPT_fsanitize_merge_handlers_EQ) ||
+       A->getOption().matches(options::OPT_fno_sanitize_merge_handlers_EQ)) &&
+      "Invalid argument in parseArgValues!");
   SanitizerMask Kinds;
   for (int i = 0, n = A->getNumValues(); i != n; ++i) {
     const char *Value = A->getValue(i);

clang/lib/Frontend/CompilerInvocation.cpp

@@ -1792,6 +1792,10 @@ void CompilerInvocationBase::GenerateCodeGenArgs(const CodeGenOptions &Opts,
   for (StringRef Sanitizer : serializeSanitizerKinds(Opts.SanitizeTrap))
     GenerateArg(Consumer, OPT_fsanitize_trap_EQ, Sanitizer);
 
+  for (StringRef Sanitizer :
+       serializeSanitizerKinds(Opts.SanitizeMergeHandlers))
+    GenerateArg(Consumer, OPT_fsanitize_merge_handlers_EQ, Sanitizer);
+
   if (!Opts.EmitVersionIdentMetadata)
     GenerateArg(Consumer, OPT_Qn);
 
@@ -2269,6 +2273,9 @@ bool CompilerInvocation::ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args,
   parseSanitizerKinds("-fsanitize-trap=",
                       Args.getAllArgValues(OPT_fsanitize_trap_EQ), Diags,
                       Opts.SanitizeTrap);
+  parseSanitizerKinds("-fsanitize-merge=",
+                      Args.getAllArgValues(OPT_fsanitize_merge_handlers_EQ),
+                      Diags, Opts.SanitizeMergeHandlers);
 
   Opts.EmitVersionIdentMetadata = Args.hasFlag(OPT_Qy, OPT_Qn, true);
 

clang/test/CodeGen/ubsan-trap-merge.c

@@ -1,11 +1,16 @@
 // NOTE: Assertions have mostly been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
 // The most important assertion is the attributes at the end of the file, which
-// shows whether -ubsan-unique-traps attaches 'nomerge' to each ubsan call.
+// shows whether -ubsan-unique-traps and -fno-sanitize-merge attach 'nomerge'
+// to each ubsan call.
 //
 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm -fsanitize=signed-integer-overflow -O3 -mllvm -ubsan-unique-traps %s -o - -fsanitize-trap=signed-integer-overflow | FileCheck %s --check-prefix=TRAP
 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm -fsanitize=signed-integer-overflow -O3 -mllvm -ubsan-unique-traps %s -o -                                         | FileCheck %s --check-prefix=HANDLER
 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm -fsanitize=signed-integer-overflow -O3 -mllvm -ubsan-unique-traps %s -o - -fsanitize-minimal-runtime              | FileCheck %s --check-prefix=MINRT
 //
+// RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm -fsanitize=signed-integer-overflow -O3 -fno-sanitize-merge=signed-integer-overflow  %s -o - -fsanitize-trap=signed-integer-overflow | FileCheck %s --check-prefix=TRAP
+// RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm -fsanitize=signed-integer-overflow -O3 -fno-sanitize-merge=signed-integer-overflow  %s -o -                                         | FileCheck %s --check-prefix=HANDLER
+// RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm -fsanitize=signed-integer-overflow -O3 -fno-sanitize-merge=signed-integer-overflow  %s -o - -fsanitize-minimal-runtime              | FileCheck %s --check-prefix=MINRT
+//
 // REQUIRES: x86-registered-target
 
 // TRAP-LABEL: define dso_local range(i32 -2147483523, -2147483648) i32 @f(