Fix failure to invoke exception handler on Win64

When the last instruction prior to a function epilogue is a call, we
need to emit a nop so that the return address is not in the epilogue IP
range.  This is consistent with MSVC's behavior, and may be a workaround
for a bug in the Win64 unwinder.

Differential Revision: http://reviews.llvm.org/D4751

Patch by Vadim Chugunov!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@214775 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/X86/X86FrameLowering.cpp

@@ -849,6 +849,11 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
   unsigned FramePtr = RegInfo->getFrameRegister(MF);
   unsigned StackPtr = RegInfo->getStackRegister();
 
+  bool IsWinEH =
+      MF.getTarget().getMCAsmInfo()->getExceptionHandlingType() ==
+      ExceptionHandling::WinEH;
+  bool NeedsWinEH = IsWinEH && MF.getFunction()->needsUnwindTableEntry();
+
   switch (RetOpcode) {
   default:
     llvm_unreachable("Can only insert epilog into returning blocks");
@@ -933,17 +938,29 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
       unsigned Opc = getLEArOpcode(IsLP64);
       addRegOffset(BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr),
                    FramePtr, false, -CSSize);
+      --MBBI;
     } else {
       unsigned Opc = (Is64Bit ? X86::MOV64rr : X86::MOV32rr);
       BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr)
         .addReg(FramePtr);
+      --MBBI;
     }
   } else if (NumBytes) {
     // Adjust stack pointer back: ESP += numbytes.
     emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, IsLP64, UseLEA,
                  TII, *RegInfo);
+    --MBBI;
   }
 
+  // Windows unwinder will not invoke function's exception handler if IP is
+  // either in prologue or in epilogue.  This behavior causes a problem when a
+  // call immediately precedes an epilogue, because the return address points
+  // into the epilogue.  To cope with that, we insert an epilogue marker here,
+  // then replace it with a 'nop' if it ends up immediately after a CALL in the
+  // final emitted code.
+  if (NeedsWinEH)
+    BuildMI(MBB, MBBI, DL, TII.get(X86::SEH_Epilogue));
+
   // We're returning from function via eh_return.
   if (RetOpcode == X86::EH_RETURN || RetOpcode == X86::EH_RETURN64) {
     MBBI = MBB.getLastNonDebugInstr();

lib/Target/X86/X86InstrCompiler.td

@@ -214,6 +214,8 @@ let isPseudo = 1 in {
                             "#SEH_PushFrame $mode", []>;
   def SEH_EndPrologue : I<0, Pseudo, (outs), (ins),
                             "#SEH_EndPrologue", []>;
+  def SEH_Epilogue : I<0, Pseudo, (outs), (ins),
+                            "#SEH_Epilogue", []>;
 }
 
 //===----------------------------------------------------------------------===//

lib/Target/X86/X86MCInstLower.cpp

@@ -826,6 +826,20 @@ void X86AsmPrinter::LowerPATCHPOINT(const MachineInstr &MI) {
            getSubtargetInfo());
 }
 
+// Returns instruction preceding MBBI in MachineFunction.
+// If MBBI is the first instruction of the first basic block, returns null.
+static MachineBasicBlock::const_iterator
+PrevCrossBBInst(MachineBasicBlock::const_iterator MBBI) {
+  const MachineBasicBlock *MBB = MBBI->getParent();
+  while (MBBI == MBB->begin()) {
+    if (MBB == MBB->getParent()->begin())
+      return nullptr;
+    MBB = MBB->getPrevNode();
+    MBBI = MBB->end();
+  }
+  return --MBBI;
+}
+
 void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
   X86MCInstLower MCInstLowering(*MF, *this);
   const X86RegisterInfo *RI =
@@ -967,6 +981,21 @@ void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
     OutStreamer.EmitWinCFIEndProlog();
     return;
 
+  case X86::SEH_Epilogue: {
+    MachineBasicBlock::const_iterator MBBI(MI);
+    // Check if preceded by a call and emit nop if so.
+    for (MBBI = PrevCrossBBInst(MBBI); MBBI; MBBI = PrevCrossBBInst(MBBI)) {
+      // Conservatively assume that pseudo instructions don't emit code and keep
+      // looking for a call. We may emit an unnecessary nop in some cases.
+      if (!MBBI->isPseudo()) {
+        if (MBBI->isCall())
+          EmitAndCountInstruction(MCInstBuilder(X86::NOOP));
+        break;
+      }
+    }
+    return;
+  }
+
   case X86::PSHUFBrm:
   case X86::VPSHUFBrm:
     // Lower PSHUFB normally but add a comment if we can find a constant

test/CodeGen/X86/win64_call_epi.ll

@@ -0,0 +1,65 @@
+; RUN: llc < %s -mtriple=x86_64-pc-mingw32 | FileCheck %s -check-prefix=WIN64
+
+declare void @bar()
+declare void @baz()
+declare i32 @personality(...)
+
+; Check for 'nop' between the last call and the epilogue.
+define void @foo1() {
+
+    invoke void @bar()
+        to label %normal
+        unwind label %catch
+
+normal:
+    ret void
+
+catch:
+    %1 = landingpad { i8*, i32 } personality i32 (...)* @personality cleanup
+    resume { i8*, i32 } %1
+}
+; WIN64-LABEL: foo1:
+; WIN64: .seh_proc foo1
+; WIN64: callq bar
+; WIN64: nop
+; WIN64: addq ${{[0-9]+}}, %rsp
+; WIN64: retq
+; Check for 'ud2' after noreturn call
+; WIN64: callq _Unwind_Resume
+; WIN64-NEXT: ud2
+; WIN64: .seh_endproc
+
+
+; Check it still works when blocks are reordered.
+@something = global i32 0
+define void @foo2(i1 zeroext %cond ) {
+    br i1 %cond, label %a, label %b, !prof !0
+a:
+    call void @bar()
+    br label %done
+b:
+    call void @baz()
+    store i32 0, i32* @something
+    br label %done
+done:
+    ret void
+}
+!0 = metadata !{metadata !"branch_weights", i32 100, i32 0}
+; WIN64-LABEL: foo2:
+; WIN64: callq bar
+; WIN64: nop
+; WIN64: addq ${{[0-9]+}}, %rsp
+; WIN64: retq
+
+
+; Check nop is not emitted when call is not adjacent to epilogue.
+define i32 @foo3() {
+    call void @bar()
+    ret i32 0
+}
+; WIN64-LABEL: foo3:
+; WIN64: callq bar
+; WIN64: xorl
+; WIN64-NOT: nop
+; WIN64: addq ${{[0-9]+}}, %rsp
+; WIN64: retq