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

DavidSpickett · tstellar · commit 571c7d8f6dae · 2022-04-18T19:37:41.000-07:00
Cherry-picked from c3b9819 which was originally reviewed as https://reviews.llvm.org/D121707. This reverts commit edb7ba7. This changes BLR_BTI to take variable_ops meaning that we can accept a register or a label. The pattern still expects one argument so we'll never get more than one. Then later we can check the type of the operand to choose BL or BLR to emit. (this is what BLR_RVMARKER does but I missed this detail of it first time around) Also require NoSLSBLRMitigation which I missed in the first version.
diff --git a/clang/docs/ClangCommandLineReference.rst b/clang/docs/ClangCommandLineReference.rst
@@ -3293,7 +3293,7 @@ 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
 
diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
@@ -381,6 +381,11 @@ Arm and AArch64 Support in Clang
 - The ``attribute((target("branch-protection=...)))`` attributes will now also
   work for the ARM backend.
 
+- 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.
+
 SPIR-V Support in Clang
 -----------------------
 
@@ -391,7 +396,6 @@ SPIR-V Support in Clang
 - Added linking of separate object files in SPIR-V format using external
   ``spirv-link`` tool.
 
-
 Floating Point Support in Clang
 -------------------------------
 - The default setting of FP contraction (FMA) is now -ffp-contract=on (for
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
@@ -3372,7 +3372,7 @@ 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>,
diff --git a/clang/lib/Driver/ToolChains/Arch/AArch64.cpp b/clang/lib/Driver/ToolChains/Arch/AArch64.cpp
@@ -592,4 +592,7 @@ 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");
 }
diff --git a/llvm/lib/Target/AArch64/AArch64.td b/llvm/lib/Target/AArch64/AArch64.td
@@ -464,6 +464,11 @@ 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.
 //
diff --git a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp b/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);
   }
diff --git a/llvm/lib/Target/AArch64/AArch64FastISel.cpp b/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;
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -2278,6 +2278,7 @@ 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;
@@ -6106,6 +6107,12 @@ 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.
@@ -6540,7 +6547,8 @@ 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);
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/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,
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/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,
@@ -2320,6 +2325,8 @@ 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 variable_ops), []>,
+                Sched<[WriteBrReg]>;
 } // isCall
 
 def : Pat<(AArch64call GPR64:$Rn),
@@ -2333,6 +2340,10 @@ def : Pat<(AArch64call_rvmarker (i64 tglobaladdr:$rvfunc), GPR64:$Rn),
           (BLR_RVMARKER tglobaladdr:$rvfunc, GPR64:$Rn)>,
       Requires<[NoSLSBLRMitigation]>;
 
+def : Pat<(AArch64call_bti GPR64:$Rn),
+          (BLR_BTI GPR64:$Rn)>,
+      Requires<[NoSLSBLRMitigation]>;
+
 let isBranch = 1, isTerminator = 1, isBarrier = 1, isIndirectBranch = 1 in {
 def BR  : BranchReg<0b0000, "br", [(brind GPR64:$Rn)]>;
 } // isBranch, isTerminator, isBarrier, isIndirectBranch
diff --git a/llvm/lib/Target/AArch64/AArch64Subtarget.h b/llvm/lib/Target/AArch64/AArch64Subtarget.h
@@ -243,6 +243,10 @@ class AArch64Subtarget final : public AArch64GenSubtargetInfo {
   // Enable 64-bit vectorization in SLP.
   unsigned MinVectorRegisterBitWidth = 64;
 
+  // Do not place a BTI instruction after a call to a return twice function like
+  // setjmp.
+  bool NoBTIAtReturnTwice = false;
+
   bool OutlineAtomics = false;
   bool PredictableSelectIsExpensive = false;
   bool BalanceFPOps = false;
@@ -588,6 +592,8 @@ class AArch64Subtarget final : public AArch64GenSubtargetInfo {
 
   bool fixCortexA53_835769() const { return FixCortexA53_835769; }
 
+  bool noBTIAtReturnTwice() const { return NoBTIAtReturnTwice; }
+
   bool addrSinkUsingGEPs() const override {
     // Keeping GEPs inbounds is important for exploiting AArch64
     // addressing-modes in ILP32 mode.
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64CallLowering.cpp b/llvm/lib/Target/AArch64/GISel/AArch64CallLowering.cpp
@@ -1127,14 +1127,22 @@ bool AArch64CallLowering::lowerCall(MachineIRBuilder &MIRBuilder,
 
   // Create a temporarily-floating call instruction so we can add the implicit
   // uses of arg registers.
-  unsigned Opc = getCallOpcode(MF, Info.Callee.isReg(), false);
+  const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>();
+  unsigned Opc = 0;
+  // A call to a returns twice function like setjmp must be followed by a bti
+  // instruction.
+  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);
 
   auto MIB = MIRBuilder.buildInstrNoInsert(Opc);
   MIB.add(Info.Callee);
 
   // 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,
diff --git a/llvm/test/CodeGen/AArch64/setjmp-bti-no-enforcement.ll b/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}
diff --git a/llvm/test/CodeGen/AArch64/setjmp-bti-outliner.ll b/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}
diff --git a/llvm/test/CodeGen/AArch64/setjmp-bti.ll b/llvm/test/CodeGen/AArch64/setjmp-bti.ll

Original file line number	Diff line number	Diff line change
`@@ -592,4 +592,7 @@ void aarch64::getAArch64TargetFeatures(const Driver &D,`
`592`	`592`	`// Enabled A53 errata (835769) workaround by default on android`
`593`	`593`	`Features.push_back("+fix-cortex-a53-835769");`
`594`	`594`	`}`
	`595`	`+`
	`596`	`+ if (Args.getLastArg(options::OPT_mno_bti_at_return_twice))`
	`597`	`+ Features.push_back("+no-bti-at-return-twice");`
`595`	`598`	`}`