[X86] Add Indirect Thunk Support to X86 to mitigate Load Value Inject…

…ion (LVI)

This pass replaces each indirect call/jump with a direct call to a thunk that looks like:

lfence
jmpq *%r11

This ensures that if the value in register %r11 was loaded from memory, then
the value in %r11 is (architecturally) correct prior to the jump.
Also adds a new target feature to X86: +lvi-cfi
("cfi" meaning control-flow integrity)
The feature can be added via clang CLI using -mlvi-cfi.

This is an alternate implementation to https://reviews.llvm.org/D75934 That merges the thunk insertion functionality with the existing X86 retpoline code.

Differential Revision: https://reviews.llvm.org/D76812

clang/docs/ClangCommandLineReference.rst

@@ -2625,6 +2625,10 @@ Use Intel MCU ABI
 
 Generate branches with extended addressability, usually via indirect jumps.
 
+.. option:: -mlvi-cfi, -mno-lvi-cfi
+
+Enable only control-flow mitigations for Load Value Injection (LVI)
+
 .. option:: -mmacosx-version-min=<arg>, -mmacos-version-min=<arg>
 
 Set Mac OS X deployment target

clang/include/clang/Driver/Options.td

@@ -2309,6 +2309,10 @@ def mspeculative_load_hardening : Flag<["-"], "mspeculative-load-hardening">,
   Group<m_Group>, Flags<[CoreOption,CC1Option]>;
 def mno_speculative_load_hardening : Flag<["-"], "mno-speculative-load-hardening">,
   Group<m_Group>, Flags<[CoreOption]>;
+def mlvi_cfi : Flag<["-"], "mlvi-cfi">, Group<m_Group>, Flags<[CoreOption,DriverOption]>,
+  HelpText<"Enable only control-flow mitigations for Load Value Injection (LVI)">;
+def mno_lvi_cfi : Flag<["-"], "mno-lvi-cfi">, Group<m_Group>, Flags<[CoreOption,DriverOption]>,
+  HelpText<"Disable control-flow mitigations for Load Value Injection (LVI)">;
 
 def mrelax : Flag<["-"], "mrelax">, Group<m_riscv_Features_Group>,
   HelpText<"Enable linker relaxation">;

clang/lib/Driver/ToolChains/Arch/X86.cpp

@@ -146,26 +146,43 @@ void x86::getX86TargetFeatures(const Driver &D, const llvm::Triple &Triple,
   // flags). This is a bit hacky but keeps existing usages working. We should
   // consider deprecating this and instead warn if the user requests external
   // retpoline thunks and *doesn't* request some form of retpolines.
+  auto SpectreOpt = clang::driver::options::ID::OPT_INVALID;
   if (Args.hasArgNoClaim(options::OPT_mretpoline, options::OPT_mno_retpoline,
                          options::OPT_mspeculative_load_hardening,
                          options::OPT_mno_speculative_load_hardening)) {
     if (Args.hasFlag(options::OPT_mretpoline, options::OPT_mno_retpoline,
                      false)) {
       Features.push_back("+retpoline-indirect-calls");
       Features.push_back("+retpoline-indirect-branches");
+      SpectreOpt = options::OPT_mretpoline;
     } else if (Args.hasFlag(options::OPT_mspeculative_load_hardening,
                             options::OPT_mno_speculative_load_hardening,
                             false)) {
       // On x86, speculative load hardening relies on at least using retpolines
       // for indirect calls.
       Features.push_back("+retpoline-indirect-calls");
+      SpectreOpt = options::OPT_mspeculative_load_hardening;
     }
   } else if (Args.hasFlag(options::OPT_mretpoline_external_thunk,
                           options::OPT_mno_retpoline_external_thunk, false)) {
     // FIXME: Add a warning about failing to specify `-mretpoline` and
     // eventually switch to an error here.
     Features.push_back("+retpoline-indirect-calls");
     Features.push_back("+retpoline-indirect-branches");
+    SpectreOpt = options::OPT_mretpoline_external_thunk;
+  }
+
+  auto LVIOpt = clang::driver::options::ID::OPT_INVALID;
+  if (Args.hasFlag(options::OPT_mlvi_cfi, options::OPT_mno_lvi_cfi, false)) {
+    Features.push_back("+lvi-cfi");
+    LVIOpt = options::OPT_mlvi_cfi;
+  }
+
+  if (SpectreOpt != clang::driver::options::ID::OPT_INVALID &&
+      LVIOpt != clang::driver::options::ID::OPT_INVALID) {
+    D.Diag(diag::err_drv_argument_not_allowed_with)
+        << D.getOpts().getOptionName(SpectreOpt)
+        << D.getOpts().getOptionName(LVIOpt);
   }
 
   // Now add any that the user explicitly requested on the command line,

clang/test/Driver/x86-target-features.c

@@ -154,6 +154,11 @@
 // SLH: "-mspeculative-load-hardening"
 // NO-SLH-NOT: retpoline
 
+// RUN: %clang -target i386-linux-gnu -mlvi-cfi %s -### -o %t.o 2>&1 | FileCheck -check-prefix=LVICFI %s
+// RUN: %clang -target i386-linux-gnu -mno-lvi-cfi %s -### -o %t.o 2>&1 | FileCheck -check-prefix=NO-LVICFI %s
+// LVICFI: "-target-feature" "+lvi-cfi"
+// NO-LVICFI-NOT: lvi-cfi
+
 // RUN: %clang -target i386-linux-gnu -mwaitpkg %s -### -o %t.o 2>&1 | FileCheck -check-prefix=WAITPKG %s
 // RUN: %clang -target i386-linux-gnu -mno-waitpkg %s -### -o %t.o 2>&1 | FileCheck -check-prefix=NO-WAITPKG %s
 // WAITPKG: "-target-feature" "+waitpkg"

llvm/lib/Target/X86/X86.td

@@ -433,6 +433,15 @@ def FeatureRetpolineExternalThunk
           "ourselves. Only has effect when combined with some other retpoline "
           "feature", [FeatureRetpolineIndirectCalls]>;
 
+// Mitigate LVI attacks against indirect calls/branches and call returns
+def FeatureLVIControlFlowIntegrity
+    : SubtargetFeature<
+          "lvi-cfi", "UseLVIControlFlowIntegrity", "true",
+          "Prevent indirect calls/branches from using a memory operand, and "
+          "precede all indirect calls/branches from a register with an "
+          "LFENCE instruction to serialize control flow. Also decompose RET "
+          "instructions into a POP+LFENCE+JMP sequence.">;
+
 // Direct Move instructions.
 def FeatureMOVDIRI  : SubtargetFeature<"movdiri", "HasMOVDIRI", "true",
                                        "Support movdiri instruction">;

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -31972,6 +31972,11 @@ static const char *getIndirectThunkSymbol(const X86Subtarget &Subtarget,
     }
     llvm_unreachable("unexpected reg for retpoline");
   }
+
+  if (Subtarget.useLVIControlFlowIntegrity()) {
+    assert(Subtarget.is64Bit() && "Should not be using a 64-bit thunk!");
+    return "__llvm_lvi_thunk_r11";
+  }
   llvm_unreachable("getIndirectThunkSymbol() invoked without thunk feature");
 }
 

llvm/lib/Target/X86/X86IndirectThunks.cpp

@@ -14,6 +14,8 @@
 ///
 /// Currently supported thunks include:
 /// - Retpoline -- A RET-implemented trampoline that lowers indirect calls
+/// - LVI Thunk -- A CALL/JMP-implemented thunk that forces load serialization
+///   before making an indirect call/jump
 ///
 /// Note that the reason that this is implemented as a MachineFunctionPass and
 /// not a ModulePass is that ModulePasses at this point in the LLVM X86 pipeline
@@ -44,11 +46,14 @@ using namespace llvm;
 #define DEBUG_TYPE "x86-retpoline-thunks"
 
 static const char RetpolineNamePrefix[] = "__llvm_retpoline_";
-static const char R11RetpolineName[]    = "__llvm_retpoline_r11";
-static const char EAXRetpolineName[]    = "__llvm_retpoline_eax";
-static const char ECXRetpolineName[]    = "__llvm_retpoline_ecx";
-static const char EDXRetpolineName[]    = "__llvm_retpoline_edx";
-static const char EDIRetpolineName[]    = "__llvm_retpoline_edi";
+static const char R11RetpolineName[] = "__llvm_retpoline_r11";
+static const char EAXRetpolineName[] = "__llvm_retpoline_eax";
+static const char ECXRetpolineName[] = "__llvm_retpoline_ecx";
+static const char EDXRetpolineName[] = "__llvm_retpoline_edx";
+static const char EDIRetpolineName[] = "__llvm_retpoline_edi";
+
+static const char LVIThunkNamePrefix[] = "__llvm_lvi_thunk_";
+static const char R11LVIThunkName[] = "__llvm_lvi_thunk_r11";
 
 namespace {
 template <typename Derived> class ThunkInserter {
@@ -80,6 +85,38 @@ struct RetpolineThunkInserter : ThunkInserter<RetpolineThunkInserter> {
   void populateThunk(MachineFunction &MF);
 };
 
+struct LVIThunkInserter : ThunkInserter<LVIThunkInserter> {
+  const char *getThunkPrefix() { return LVIThunkNamePrefix; }
+  bool mayUseThunk(const MachineFunction &MF) {
+    return MF.getSubtarget<X86Subtarget>().useLVIControlFlowIntegrity();
+  }
+  void insertThunks(MachineModuleInfo &MMI) {
+    createThunkFunction(MMI, R11LVIThunkName);
+  }
+  void populateThunk(MachineFunction &MF) {
+    // Grab the entry MBB and erase any other blocks. O0 codegen appears to
+    // generate two bbs for the entry block.
+    MachineBasicBlock *Entry = &MF.front();
+    Entry->clear();
+    while (MF.size() > 1)
+      MF.erase(std::next(MF.begin()));
+
+    // This code mitigates LVI by replacing each indirect call/jump with a
+    // direct call/jump to a thunk that looks like:
+    // ```
+    // lfence
+    // jmpq *%r11
+    // ```
+    // This ensures that if the value in register %r11 was loaded from memory,
+    // then the value in %r11 is (architecturally) correct prior to the jump.
+    const TargetInstrInfo *TII = MF.getSubtarget<X86Subtarget>().getInstrInfo();
+    BuildMI(&MF.front(), DebugLoc(), TII->get(X86::LFENCE));
+    BuildMI(&MF.front(), DebugLoc(), TII->get(X86::JMP64r)).addReg(X86::R11);
+    MF.front().addLiveIn(X86::R11);
+    return;
+  }
+};
+
 class X86IndirectThunks : public MachineFunctionPass {
 public:
   static char ID;
@@ -98,7 +135,7 @@ class X86IndirectThunks : public MachineFunctionPass {
   }
 
 private:
-  std::tuple<RetpolineThunkInserter> TIs;
+  std::tuple<RetpolineThunkInserter, LVIThunkInserter> TIs;
 
   // FIXME: When LLVM moves to C++17, these can become folds
   template <typename... ThunkInserterT>

llvm/lib/Target/X86/X86Subtarget.h

@@ -428,6 +428,12 @@ class X86Subtarget final : public X86GenSubtargetInfo {
   /// than emitting one inside the compiler.
   bool UseRetpolineExternalThunk = false;
 
+  /// Prevent generation of indirect call/branch instructions from memory,
+  /// and force all indirect call/branch instructions from a register to be
+  /// preceded by an LFENCE. Also decompose RET instructions into a
+  /// POP+LFENCE+JMP sequence.
+  bool UseLVIControlFlowIntegrity = false;
+
   /// Use software floating point for code generation.
   bool UseSoftFloat = false;
 
@@ -719,13 +725,16 @@ class X86Subtarget final : public X86GenSubtargetInfo {
   // These are generic getters that OR together all of the thunk types
   // supported by the subtarget. Therefore useIndirectThunk*() will return true
   // if any respective thunk feature is enabled.
-  bool useIndirectThunkCalls() const { return useRetpolineIndirectCalls(); }
+  bool useIndirectThunkCalls() const {
+    return useRetpolineIndirectCalls() || useLVIControlFlowIntegrity();
+  }
   bool useIndirectThunkBranches() const {
-    return useRetpolineIndirectBranches();
+    return useRetpolineIndirectBranches() || useLVIControlFlowIntegrity();
   }
 
   bool preferMaskRegisters() const { return PreferMaskRegisters; }
   bool useGLMDivSqrtCosts() const { return UseGLMDivSqrtCosts; }
+  bool useLVIControlFlowIntegrity() const { return UseLVIControlFlowIntegrity; }
 
   unsigned getPreferVectorWidth() const { return PreferVectorWidth; }
   unsigned getRequiredVectorWidth() const { return RequiredVectorWidth; }