[sanitizer] Add plumbing for -fsanitize-add-pseudo-functions and partly replace '-mllvm -array-bounds-pseudofn'

Florian1 introduced '-mllvm -array-bounds-pseudofn'
(llvm#128977) to make it easier to
see why crashes occurred, and to estimate with a profiler the cycles spent on these
array-bounds checks. This functionality could be usefully generalized to other
checks in future work.

This patch adds the plumbing for -fsanitize-add-pseudo-functions, and
connects it to the existing array-bounds-pseudo-fn functionality i.e.,
-fsanitize-add-pseudo-functions=array-bounds can be used as a
replacement for '-mllvm -array-bounds-pseudofn', though we do not yet
delete the latter.

Note: we replaced '-mllvm -array-bounds-pseudofn' in clang/test/CodeGen/bounds-checking-debuginfo.c, because adding test cases would modify the line numbers in the test assertions, and therefore obscure that the test output is the same between '-mllvm -array-bounds-pseudofn' and -fsanitize-add-pseudo-functions=array-bounds.

Author: thurstond

clang/include/clang/Basic/CodeGenOptions.h

@@ -399,6 +399,9 @@ class CodeGenOptions : public CodeGenOptionsBase {
   /// (0.0 [default] to skip none, 1.0 to skip all).
   SanitizerMaskCutoffs SanitizeSkipHotCutoffs;
 
+  /// Set of sanitizer checks that will be wrapped inside pseudofunctions.
+  SanitizerSet SanitizeAddPseudoFunctions;
+
   /// List of backend command-line options for -fembed-bitcode.
   std::vector<uint8_t> CmdArgs;
 

clang/include/clang/Driver/Options.td

@@ -2533,6 +2533,21 @@ def fno_sanitize_merge_handlers : Flag<["-"], "fno-sanitize-merge">, Group<f_cla
                         Alias<fno_sanitize_merge_handlers_EQ>, AliasArgs<["all"]>,
                         Visibility<[ClangOption, CLOption]>,
                         HelpText<"Do not allow compiler to merge handlers for any sanitizers">;
+def fsanitize_add_pseudo_functions_EQ
+    : CommaJoined<["-"], "fsanitize-add-pseudo-functions=">,
+      Group<f_clang_Group>,
+      HelpText<"Add pseudo-functions to checks for specified sanitizers">;
+def fno_sanitize_add_pseudo_functions_EQ
+    : CommaJoined<["-"], "fno-sanitize-add-pseudo-functions=">,
+      Group<f_clang_Group>,
+      HelpText<"Do not allow compiler to add pseudo-functions to checks for specified sanitizers">;
+def fsanitize_add_pseudo_functions : Flag<["-"], "fsanitize-add-pseudo-functions">, Group<f_clang_Group>,
+                     Alias<fsanitize_add_pseudo_functions_EQ>, AliasArgs<["all"]>,
+                     HelpText<"Allow compiler to add pseudo-functions to checks for all sanitizers">;
+def fno_sanitize_add_pseudo_functions : Flag<["-"], "fno-sanitize-add-pseudo-functions">, Group<f_clang_Group>,
+                        Alias<fno_sanitize_add_pseudo_functions_EQ>, AliasArgs<["all"]>,
+                        Visibility<[ClangOption, CLOption]>,
+                        HelpText<"Do not allow compiler to add pseudo-functions to checks 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

@@ -27,6 +27,7 @@ class SanitizerArgs {
   SanitizerSet TrapSanitizers;
   SanitizerSet MergeHandlers;
   SanitizerMaskCutoffs SkipHotCutoffs;
+  SanitizerSet AddPseudoFunctions;
 
   std::vector<std::string> UserIgnorelistFiles;
   std::vector<std::string> SystemIgnorelistFiles;

clang/lib/CodeGen/CGExpr.cpp

@@ -1228,7 +1228,7 @@ void CodeGenFunction::EmitBoundsCheckImpl(const Expr *E, llvm::Value *Bound,
   SanitizerScope SanScope(this);
 
   llvm::DILocation *CheckDI = Builder.getCurrentDebugLocation();
-  if (ClArrayBoundsPseudoFn && CheckDI) {
+  if ((ClArrayBoundsPseudoFn || CGM.getCodeGenOpts().SanitizeAddPseudoFunctions.has(SanitizerKind::SO_ArrayBounds)) && CheckDI) {
     CheckDI = getDebugInfo()->CreateSyntheticInlineAt(
         Builder.getCurrentDebugLocation(), "__ubsan_check_array_bounds");
   }

clang/lib/Driver/SanitizerArgs.cpp

@@ -76,6 +76,7 @@ static const SanitizerMask MergeDefault =
     SanitizerKind::Undefined | SanitizerKind::Vptr;
 static const SanitizerMask TrappingDefault =
     SanitizerKind::CFI | SanitizerKind::LocalBounds;
+static const SanitizerMask AddPseudoFunctionsDefault;
 static const SanitizerMask CFIClasses =
     SanitizerKind::CFIVCall | SanitizerKind::CFINVCall |
     SanitizerKind::CFIMFCall | SanitizerKind::CFIDerivedCast |
@@ -738,6 +739,13 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
   // Parse -fno-sanitize-top-hot flags
   SkipHotCutoffs = parseSanitizeSkipHotCutoffArgs(D, Args, DiagnoseErrors);
 
+  // Parse -f(no-)?sanitize-add-pseudo-functions flags
+  SanitizerMask AddPseudoFunctionsKinds =
+      parseSanitizeArgs(D, Args, DiagnoseErrors, AddPseudoFunctionsDefault, {}, {},
+                        options::OPT_fsanitize_add_pseudo_functions_EQ,
+                        options::OPT_fno_sanitize_add_pseudo_functions_EQ);
+  AddPseudoFunctionsKinds &= Kinds;
+
   // Setup ignorelist files.
   // Add default ignorelist from resource directory for activated sanitizers,
   // and validate special case lists format.
@@ -1157,6 +1165,8 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
 
   MergeHandlers.Mask |= MergeKinds;
 
+  AddPseudoFunctions.Mask |= AddPseudoFunctionsKinds;
+
   // Zero out SkipHotCutoffs for unused sanitizers
   SkipHotCutoffs.clear(~Sanitizers.Mask);
 }
@@ -1335,6 +1345,10 @@ void SanitizerArgs::addArgs(const ToolChain &TC, const llvm::opt::ArgList &Args,
     CmdArgs.push_back(
         Args.MakeArgString("-fsanitize-skip-hot-cutoff=" + SkipHotCutoffsStr));
 
+  if (!AddPseudoFunctions.empty())
+    CmdArgs.push_back(
+        Args.MakeArgString("-fsanitize-add-pseudo-functions=" + toString(AddPseudoFunctions)));
+
   addSpecialCaseListOpt(Args, CmdArgs,
                         "-fsanitize-ignorelist=", UserIgnorelistFiles);
   addSpecialCaseListOpt(Args, CmdArgs,
@@ -1518,7 +1532,9 @@ SanitizerMask parseArgValues(const Driver &D, const llvm::opt::Arg *A,
        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)) &&
+       A->getOption().matches(options::OPT_fno_sanitize_merge_handlers_EQ) ||
+       A->getOption().matches(options::OPT_fsanitize_add_pseudo_functions_EQ) ||
+       A->getOption().matches(options::OPT_fno_sanitize_add_pseudo_functions_EQ)) &&
       "Invalid argument in parseArgValues!");
   SanitizerMask Kinds;
   for (int i = 0, n = A->getNumValues(); i != n; ++i) {

clang/lib/Frontend/CompilerInvocation.cpp

@@ -1838,6 +1838,10 @@ void CompilerInvocationBase::GenerateCodeGenArgs(const CodeGenOptions &Opts,
   for (std::string Sanitizer : Values)
     GenerateArg(Consumer, OPT_fsanitize_skip_hot_cutoff_EQ, Sanitizer);
 
+  for (StringRef Sanitizer :
+       serializeSanitizerKinds(Opts.SanitizeAddPseudoFunctions))
+    GenerateArg(Consumer, OPT_fsanitize_add_pseudo_functions_EQ, Sanitizer);
+
   if (!Opts.EmitVersionIdentMetadata)
     GenerateArg(Consumer, OPT_Qn);
 
@@ -2332,6 +2336,10 @@ bool CompilerInvocation::ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args,
       "-fsanitize-skip-hot-cutoff=",
       Args.getAllArgValues(OPT_fsanitize_skip_hot_cutoff_EQ), Diags);
 
+  parseSanitizerKinds("-fsanitize-add-pseudo-functions=",
+                      Args.getAllArgValues(OPT_fsanitize_add_pseudo_functions_EQ),
+                      Diags, Opts.SanitizeAddPseudoFunctions);
+
   Opts.EmitVersionIdentMetadata = Args.hasFlag(OPT_Qy, OPT_Qn, true);
 
   if (!LangOpts->CUDAIsDevice)

clang/test/CodeGen/bounds-checking-debuginfo.c

@@ -1,7 +1,7 @@
 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
-// RUN: %clang_cc1 -mllvm -array-bounds-pseudofn -emit-llvm -fdebug-prefix-map=%S/= -fno-ident -fdebug-compilation-dir=%S -fsanitize=array-bounds -fsanitize-trap=array-bounds -triple x86_64 -debug-info-kind=limited %s -o - | FileCheck --check-prefix=CHECK-TRAP %s
-// RUN: %clang_cc1 -mllvm -array-bounds-pseudofn -emit-llvm -fdebug-prefix-map=%S/= -fno-ident -fdebug-compilation-dir=%S -fsanitize=array-bounds                              -triple x86_64 -debug-info-kind=limited %s -o - | FileCheck --check-prefix=CHECK-NOTRAP %s
 
+// RUN: %clang_cc1 -emit-llvm -fdebug-prefix-map=%S/= -fno-ident -fdebug-compilation-dir=%S -fsanitize=array-bounds -fsanitize-trap=array-bounds -fsanitize-add-pseudo-functions=array-bounds -triple x86_64 -debug-info-kind=limited %s -o - | FileCheck --check-prefix=CHECK-TRAP %s
+// RUN: %clang_cc1 -emit-llvm -fdebug-prefix-map=%S/= -fno-ident -fdebug-compilation-dir=%S -fsanitize=array-bounds                              -fsanitize-add-pseudo-functions=array-bounds -triple x86_64 -debug-info-kind=limited %s -o - | FileCheck --check-prefix=CHECK-NOTRAP %s
 
 int f();
 void d(double*);