[llvm][AArch64] Insert "bti j" after call to setjmp

Some implementations of setjmp will end with a br instead of a ret.
This means that the next instruction after a call to setjmp must be
a "bti j" (j for jump) to make this work when branch target identification
is enabled.

The BTI extension was added in armv8.5-a but the bti instruction is in the
hint space. This means we can emit it for any architecture version as long
as branch target enforcement flags are passed.

The starting point for the hint number is 32 then call adds 2, jump adds 4.
Hence "hint #36" for a "bti j" (and "hint #34" for the "bti c" you see
at the start of functions).

The existing Arm command line option -mno-bti-at-return-twice has been
applied to AArch64 as well.

Support is added to SelectionDAG Isel and GlobalIsel. FastIsel will
defer to SelectionDAG.

Based on the change done for M profile Arm in https://reviews.llvm.org/D112427

Fixes #48888

Reviewed By: danielkiss

Differential Revision: https://reviews.llvm.org/D121707

clang/docs/ClangCommandLineReference.rst

@@ -3329,7 +3329,7 @@ Work around VLLDM erratum CVE-2021-35465 (ARM only)
 
 .. option:: -mno-bti-at-return-twice
 
-Do not add a BTI instruction after a setjmp or other return-twice construct (Arm only)
+Do not add a BTI instruction after a setjmp or other return-twice construct (AArch32/AArch64 only)
 
 .. option:: -mno-movt
 

clang/docs/ReleaseNotes.rst

@@ -193,6 +193,11 @@ DWARF Support in Clang
 Arm and AArch64 Support in Clang
 --------------------------------
 
+- When using ``-mbranch-protection=bti`` with AArch64, calls to setjmp will
+  now be followed by a BTI instruction. This is done to be compatible with
+  setjmp implementations that return with a br instead of a ret. You can
+  disable this behaviour using the ``-mno-bti-at-return-twice`` option.
+
 Floating Point Support in Clang
 -------------------------------
 

clang/include/clang/Driver/Options.td

@@ -3414,7 +3414,7 @@ def mmark_bti_property : Flag<["-"], "mmark-bti-property">,
 def mno_bti_at_return_twice : Flag<["-"], "mno-bti-at-return-twice">,
   Group<m_arm_Features_Group>,
   HelpText<"Do not add a BTI instruction after a setjmp or other"
-           " return-twice construct (Arm only)">;
+           " return-twice construct (Arm/AArch64 only)">;
 
 foreach i = {1-31} in
   def ffixed_x#i : Flag<["-"], "ffixed-x"#i>, Group<m_Group>,

clang/lib/Driver/ToolChains/Arch/AArch64.cpp

@@ -588,4 +588,7 @@ void aarch64::getAArch64TargetFeatures(const Driver &D,
     // Enabled A53 errata (835769) workaround by default on android
     Features.push_back("+fix-cortex-a53-835769");
   }
+
+  if (Args.getLastArg(options::OPT_mno_bti_at_return_twice))
+    Features.push_back("+no-bti-at-return-twice");
 }

llvm/lib/Target/AArch64/AArch64.td

@@ -466,6 +466,11 @@ def FeatureEL3 : SubtargetFeature<"el3", "HasEL3", "true",
 def FeatureFixCortexA53_835769 : SubtargetFeature<"fix-cortex-a53-835769",
   "FixCortexA53_835769", "true", "Mitigate Cortex-A53 Erratum 835769">;
 
+def FeatureNoBTIAtReturnTwice : SubtargetFeature<"no-bti-at-return-twice",
+                                                 "NoBTIAtReturnTwice", "true",
+                                                 "Don't place a BTI instruction "
+                                                 "after a return-twice">;
+
 //===----------------------------------------------------------------------===//
 // Architectures.
 //

llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp

@@ -86,6 +86,7 @@ class AArch64ExpandPseudo : public MachineFunctionPass {
                           unsigned N);
   bool expandCALL_RVMARKER(MachineBasicBlock &MBB,
                            MachineBasicBlock::iterator MBBI);
+  bool expandCALL_BTI(MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI);
   bool expandStoreSwiftAsyncContext(MachineBasicBlock &MBB,
                                     MachineBasicBlock::iterator MBBI);
 };
@@ -759,6 +760,37 @@ bool AArch64ExpandPseudo::expandCALL_RVMARKER(
   return true;
 }
 
+bool AArch64ExpandPseudo::expandCALL_BTI(MachineBasicBlock &MBB,
+                                         MachineBasicBlock::iterator MBBI) {
+  // Expand CALL_BTI pseudo to:
+  // - a branch to the call target
+  // - a BTI instruction
+  // Mark the sequence as a bundle, to avoid passes moving other code in
+  // between.
+
+  MachineInstr &MI = *MBBI;
+  MachineOperand &CallTarget = MI.getOperand(0);
+  assert((CallTarget.isGlobal() || CallTarget.isReg()) &&
+         "invalid operand for regular call");
+  unsigned Opc = CallTarget.isGlobal() ? AArch64::BL : AArch64::BLR;
+  MachineInstr *Call =
+      BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc)).getInstr();
+  Call->addOperand(CallTarget);
+
+  MachineInstr *BTI =
+      BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(AArch64::HINT))
+          // BTI J so that setjmp can to BR to this.
+          .addImm(36)
+          .getInstr();
+
+  if (MI.shouldUpdateCallSiteInfo())
+    MBB.getParent()->moveCallSiteInfo(&MI, Call);
+
+  MI.eraseFromParent();
+  finalizeBundle(MBB, Call->getIterator(), std::next(BTI->getIterator()));
+  return true;
+}
+
 bool AArch64ExpandPseudo::expandStoreSwiftAsyncContext(
     MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI) {
   Register CtxReg = MBBI->getOperand(0).getReg();
@@ -1238,6 +1270,8 @@ bool AArch64ExpandPseudo::expandMI(MachineBasicBlock &MBB,
      return expandSVESpillFill(MBB, MBBI, AArch64::LDR_ZXI, 2);
    case AArch64::BLR_RVMARKER:
      return expandCALL_RVMARKER(MBB, MBBI);
+   case AArch64::BLR_BTI:
+     return expandCALL_BTI(MBB, MBBI);
    case AArch64::StoreSwiftAsyncContext:
      return expandStoreSwiftAsyncContext(MBB, MBBI);
   }

llvm/lib/Target/AArch64/AArch64FastISel.cpp

@@ -14,6 +14,7 @@
 
 #include "AArch64.h"
 #include "AArch64CallingConvention.h"
+#include "AArch64MachineFunctionInfo.h"
 #include "AArch64RegisterInfo.h"
 #include "AArch64Subtarget.h"
 #include "MCTargetDesc/AArch64AddressingModes.h"
@@ -3127,6 +3128,13 @@ bool AArch64FastISel::fastLowerCall(CallLoweringInfo &CLI) {
   if (!Callee && !Symbol)
     return false;
 
+  // Allow SelectionDAG isel to handle calls to functions like setjmp that need
+  // a bti instruction following the call.
+  if (CLI.CB && CLI.CB->hasFnAttr(Attribute::ReturnsTwice) &&
+      !Subtarget->noBTIAtReturnTwice() &&
+      MF->getInfo<AArch64FunctionInfo>()->branchTargetEnforcement())
+    return false;
+
   // Allow SelectionDAG isel to handle tail calls.
   if (IsTailCall)
     return false;

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -2274,6 +2274,7 @@ const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
     MAKE_CASE(AArch64ISD::MOPS_MEMSET_TAGGING)
     MAKE_CASE(AArch64ISD::MOPS_MEMCOPY)
     MAKE_CASE(AArch64ISD::MOPS_MEMMOVE)
+    MAKE_CASE(AArch64ISD::CALL_BTI)
   }
 #undef MAKE_CASE
   return nullptr;
@@ -6178,6 +6179,12 @@ AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
   AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();
   bool TailCallOpt = MF.getTarget().Options.GuaranteedTailCallOpt;
   bool IsSibCall = false;
+  bool GuardWithBTI = false;
+
+  if (CLI.CB && CLI.CB->getAttributes().hasFnAttr(Attribute::ReturnsTwice) &&
+      !Subtarget->noBTIAtReturnTwice()) {
+    GuardWithBTI = FuncInfo->branchTargetEnforcement();
+  }
 
   // Check callee args/returns for SVE registers and set calling convention
   // accordingly.
@@ -6612,7 +6619,8 @@ AArch64TargetLowering::LowerCall(CallLoweringInfo &CLI,
     Function *ARCFn = *objcarc::getAttachedARCFunction(CLI.CB);
     auto GA = DAG.getTargetGlobalAddress(ARCFn, DL, PtrVT);
     Ops.insert(Ops.begin() + 1, GA);
-  }
+  } else if (GuardWithBTI)
+    CallOpc = AArch64ISD::CALL_BTI;
 
   // Returns a chain and a flag for retval copy to use.
   Chain = DAG.getNode(CallOpc, DL, NodeTys, Ops);

llvm/lib/Target/AArch64/AArch64ISelLowering.h

@@ -55,6 +55,8 @@ enum NodeType : unsigned {
   // x29, x29` marker instruction.
   CALL_RVMARKER,
 
+  CALL_BTI, // Function call followed by a BTI instruction.
+
   // Produces the full sequence of instructions for getting the thread pointer
   // offset of a variable into X0, using the TLSDesc model.
   TLSDESC_CALLSEQ,

llvm/lib/Target/AArch64/AArch64InstrInfo.td

@@ -473,6 +473,11 @@ def AArch64call          : SDNode<"AArch64ISD::CALL",
                                 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
                                  SDNPVariadic]>;
 
+def AArch64call_bti      : SDNode<"AArch64ISD::CALL_BTI",
+                                SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
+                                [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
+                                 SDNPVariadic]>;
+
 def AArch64call_rvmarker: SDNode<"AArch64ISD::CALL_RVMARKER",
                              SDTypeProfile<0, -1, [SDTCisPtrTy<0>]>,
                              [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
@@ -2328,6 +2333,8 @@ let isCall = 1, Defs = [LR], Uses = [SP] in {
                 PseudoInstExpansion<(BLR GPR64:$Rn)>;
   def BLR_RVMARKER : Pseudo<(outs), (ins variable_ops), []>,
                      Sched<[WriteBrReg]>;
+  def BLR_BTI : Pseudo<(outs), (ins GPR64:$Rn), []>,
+                Sched<[WriteBrReg]>;
 } // isCall
 
 def : Pat<(AArch64call GPR64:$Rn),
@@ -2341,6 +2348,9 @@ def : Pat<(AArch64call_rvmarker (i64 tglobaladdr:$rvfunc), GPR64:$Rn),
           (BLR_RVMARKER tglobaladdr:$rvfunc, GPR64:$Rn)>,
       Requires<[NoSLSBLRMitigation]>;
 
+def : Pat<(AArch64call_bti GPR64:$Rn),
+          (BLR_BTI GPR64:$Rn)>;
+
 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
 def BR  : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
 } // isBranch, isTerminator, isBarrier, isIndirectBranch

llvm/lib/Target/AArch64/GISel/AArch64CallLowering.cpp

@@ -1129,12 +1129,20 @@ bool AArch64CallLowering::lowerCall(MachineIRBuilder &MIRBuilder,
   // Create a temporarily-floating call instruction so we can add the implicit
   // uses of arg registers.
 
+  const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>();
+  unsigned Opc = 0;
   // Calls with operand bundle "clang.arc.attachedcall" are special. They should
   // be expanded to the call, directly followed by a special marker sequence and
   // a call to an ObjC library function.
-  unsigned Opc = 0;
   if (Info.CB && objcarc::hasAttachedCallOpBundle(Info.CB))
     Opc = AArch64::BLR_RVMARKER;
+  // A call to a returns twice function like setjmp must be followed by a bti
+  // instruction.
+  else if (Info.CB &&
+           Info.CB->getAttributes().hasFnAttr(Attribute::ReturnsTwice) &&
+           !Subtarget.noBTIAtReturnTwice() &&
+           MF.getInfo<AArch64FunctionInfo>()->branchTargetEnforcement())
+    Opc = AArch64::BLR_BTI;
   else
     Opc = getCallOpcode(MF, Info.Callee.isReg(), false);
 
@@ -1153,7 +1161,6 @@ bool AArch64CallLowering::lowerCall(MachineIRBuilder &MIRBuilder,
 
   // Tell the call which registers are clobbered.
   const uint32_t *Mask;
-  const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>();
   const auto *TRI = Subtarget.getRegisterInfo();
 
   AArch64OutgoingValueAssigner Assigner(AssignFnFixed, AssignFnVarArg,

llvm/test/CodeGen/AArch64/setjmp-bti-no-enforcement.ll

@@ -0,0 +1,51 @@
+; RUN: llc -mtriple=aarch64-none-linux-gnu < %s | FileCheck %s --check-prefix=NOBTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -global-isel < %s | FileCheck %s --check-prefix=NOBTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -fast-isel < %s | FileCheck %s --check-prefix=NOBTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mattr=+no-bti-at-return-twice < %s | \
+; RUN: FileCheck %s --check-prefix=NOBTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -global-isel -mattr=+no-bti-at-return-twice < %s | \
+; RUN: FileCheck %s --check-prefix=NOBTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -fast-isel -mattr=+no-bti-at-return-twice < %s | \
+; RUN: FileCheck %s --check-prefix=NOBTI
+
+; Same as setjmp-bti.ll except that we do not enable branch target enforcement for this
+; module. There should be no combination of options that leads to a bti being emitted.
+
+; C source
+; --------
+; extern int setjmp(void*);
+; extern void notsetjmp(void);
+;
+; void bbb(void) {
+;   setjmp(0);
+;   int (*fnptr)(void*) = setjmp;
+;   fnptr(0);
+;   notsetjmp();
+; }
+
+define void @bbb() {
+; NOBTI-LABEL: bbb:
+; NOBTI:     bl setjmp
+; NOBTI-NOT: hint #36
+; NOBTI:     blr x{{[0-9]+}}
+; NOBTI-NOT: hint #36
+; NOBTI:     bl notsetjmp
+; NOBTI-NOT: hint #36
+
+entry:
+  %fnptr = alloca i32 (i8*)*, align 8
+  %call = call i32 @setjmp(i8* noundef null) #0
+  store i32 (i8*)* @setjmp, i32 (i8*)** %fnptr, align 8
+  %0 = load i32 (i8*)*, i32 (i8*)** %fnptr, align 8
+  %call1 = call i32 %0(i8* noundef null) #0
+  call void @notsetjmp()
+  ret void
+}
+
+declare i32 @setjmp(i8* noundef) #0
+declare void @notsetjmp()
+
+attributes #0 = { returns_twice }
+
+; !llvm.module.flags = !{!0}
+; !0 = !{i32 1, !"branch-target-enforcement", i32 1}

llvm/test/CodeGen/AArch64/setjmp-bti-outliner.ll

@@ -0,0 +1,83 @@
+; RUN: llc -mtriple=aarch64-none-linux-gnu -enable-machine-outliner < %s | FileCheck %s --check-prefix=BTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -global-isel -enable-machine-outliner < %s | \
+; RUN: FileCheck %s --check-prefix=BTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -fast-isel -enable-machine-outliner < %s | \
+; RUN: FileCheck %s --check-prefix=BTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -enable-machine-outliner -mattr=+no-bti-at-return-twice < %s | \
+; RUN: FileCheck %s --check-prefix=NOBTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -global-isel -enable-machine-outliner -mattr=+no-bti-at-return-twice < %s | \
+; RUN: FileCheck %s --check-prefix=NOBTI
+; RUN: llc -mtriple=aarch64-none-linux-gnu -fast-isel -enable-machine-outliner -mattr=+no-bti-at-return-twice < %s | \
+; RUN: FileCheck %s --check-prefix=NOBTI
+
+; Check that the outliner does not split up the call to setjmp and the bti after it.
+; When we do not insert a bti, it is allowed to move the setjmp call into an outlined function.
+
+; C source
+; --------
+; extern int setjmp(void*);
+;
+; int f(int a, int b, int c, int d) {
+;   setjmp(0);
+;   return 1 + a * (a + b) / (c + d);
+; }
+;
+; int g(int a, int b, int c, int d) {
+;   setjmp(0);
+;   return 2 + a * (a + b) / (c + d);
+; }
+
+define i32 @f(i32 noundef %a, i32 noundef %b, i32 noundef %c, i32 noundef %d) {
+; BTI-LABEL: f:
+; BTI:         bl      OUTLINED_FUNCTION_1
+; BTI-NEXT:    bl      setjmp
+; BTI-NEXT:    hint    #36
+; BTI-NEXT:    bl      OUTLINED_FUNCTION_0
+
+; NOBTI:      f:
+; NOBTI:        bl      OUTLINED_FUNCTION_0
+; NOBTI-NEXT:   bl      OUTLINED_FUNCTION_1
+
+entry:
+  %call = call i32 @setjmp(i8* noundef null) #0
+  %add = add nsw i32 %b, %a
+  %mul = mul nsw i32 %add, %a
+  %add1 = add nsw i32 %d, %c
+  %div = sdiv i32 %mul, %add1
+  %add2 = add nsw i32 %div, 1
+  ret i32 %add2
+}
+
+declare i32 @setjmp(i8* noundef) #0
+
+define i32 @g(i32 noundef %a, i32 noundef %b, i32 noundef %c, i32 noundef %d) {
+; BTI-LABEL: g:
+; BTI:         bl      OUTLINED_FUNCTION_1
+; BTI-NEXT:    bl      setjmp
+; BTI-NEXT:    hint    #36
+; BTI-NEXT:    bl      OUTLINED_FUNCTION_0
+
+; NOBTI:      g:
+; NOBTI:        bl      OUTLINED_FUNCTION_0
+; NOBTI-NEXT:   bl      OUTLINED_FUNCTION_1
+
+entry:
+  %call = call i32 @setjmp(i8* noundef null) #0
+  %add = add nsw i32 %b, %a
+  %mul = mul nsw i32 %add, %a
+  %add1 = add nsw i32 %d, %c
+  %div = sdiv i32 %mul, %add1
+  %add2 = add nsw i32 %div, 2
+  ret i32 %add2
+}
+
+; NOBTI-LABEL: OUTLINED_FUNCTION_0:
+; NOBTI:         b       setjmp
+; NOBTI:       OUTLINED_FUNCTION_1:
+; NOBTI-LABEL:   ret
+
+attributes #0 = { returns_twice }
+
+!llvm.module.flags = !{!0}
+
+!0 = !{i32 1, !"branch-target-enforcement", i32 1}