[WinEH] Fix stack alignment in funclets and ParentFrameOffset calculation

Our previous value of "16 + 8 + MaxCallFrameSize" for ParentFrameOffset
is incorrect when CSRs are involved. We were supposed to have a test
case to catch this, but it wasn't very rigorous.

The main effect here is that calling _CxxThrowException inside a
catchpad doesn't immediately crash on MOVAPS when you have an odd number
of CSRs.

llvm-svn: 250583

5 files changed, 182 insertions, 12 deletions

diff --git a/llvm/include/llvm/Target/TargetFrameLowering.h b/llvm/include/llvm/Target/TargetFrameLowering.h
index 9585f7c3498..398c91ef562 100644
--- a/llvm/include/llvm/Target/TargetFrameLowering.h
+++ b/llvm/include/llvm/Target/TargetFrameLowering.h
@@ -265,6 +265,10 @@ public:
                                              RegScavenger *RS = nullptr) const {
   }
 
+  virtual unsigned getWinEHParentFrameOffset(const MachineFunction &MF) const {
+    report_fatal_error("WinEH not implemented for this target");
+  }
+
   /// eliminateCallFramePseudoInstr - This method is called during prolog/epilog
   /// code insertion to eliminate call frame setup and destroy pseudo
   /// instructions (but only if the Target is using them).  It is responsible
diff --git a/llvm/lib/CodeGen/AsmPrinter/WinException.cpp b/llvm/lib/CodeGen/AsmPrinter/WinException.cpp
index 319320a09e0..fab5b3a59eb 100644
--- a/llvm/lib/CodeGen/AsmPrinter/WinException.cpp
+++ b/llvm/lib/CodeGen/AsmPrinter/WinException.cpp
@@ -710,6 +710,13 @@ void WinException::emitCXXFrameHandler3Table(const MachineFunction *MF) {
       OS.EmitValue(create32bitRef(HandlerMapXData), 4);   // HandlerArray
     }
 
+    // All funclets use the same parent frame offset currently.
+    unsigned ParentFrameOffset = 0;
+    if (shouldEmitPersonality) {
+      const TargetFrameLowering *TFI = MF->getSubtarget().getFrameLowering();
+      ParentFrameOffset = TFI->getWinEHParentFrameOffset(*MF);
+    }
+
     for (size_t I = 0, E = FuncInfo.TryBlockMap.size(); I != E; ++I) {
       WinEHTryBlockMapEntry &TBME = FuncInfo.TryBlockMap[I];
       MCSymbol *HandlerMapXData = HandlerMaps[I];
@@ -749,13 +756,8 @@ void WinException::emitCXXFrameHandler3Table(const MachineFunction *MF) {
         OS.EmitValue(create32bitRef(HT.TypeDescriptor), 4); // Type
         OS.EmitValue(FrameAllocOffsetRef, 4);               // CatchObjOffset
         OS.EmitValue(create32bitRef(HandlerSym), 4);        // Handler
-
-        if (shouldEmitPersonality) {
-          // Keep this in sync with X86FrameLowering::emitPrologue.
-          int ParentFrameOffset =
-              16 + 8 + MF->getFrameInfo()->getMaxCallFrameSize();
+        if (shouldEmitPersonality)
           OS.EmitIntValue(ParentFrameOffset, 4); // ParentFrameOffset
-        }
       }
     }
   }
diff --git a/llvm/lib/Target/X86/X86FrameLowering.cpp b/llvm/lib/Target/X86/X86FrameLowering.cpp
index 83a4acd7a4a..b4dccb60ccc 100644
--- a/llvm/lib/Target/X86/X86FrameLowering.cpp
+++ b/llvm/lib/Target/X86/X86FrameLowering.cpp
@@ -786,7 +786,7 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF,
   // NumBytes value that we would've used for the parent frame.
   unsigned ParentFrameNumBytes = NumBytes;
   if (IsFunclet)
-    NumBytes = MFI->getMaxCallFrameSize();
+    NumBytes = getWinEHFuncletFrameSize(MF);
 
   // Skip the callee-saved push instructions.
   bool PushedRegs = false;
@@ -1039,6 +1039,22 @@ static bool isFuncletReturnInstr(MachineInstr *MI) {
   llvm_unreachable("impossible");
 }
 
+unsigned X86FrameLowering::getWinEHFuncletFrameSize(const MachineFunction &MF) const {
+  // This is the size of the pushed CSRs.
+  unsigned CSSize =
+      MF.getInfo<X86MachineFunctionInfo>()->getCalleeSavedFrameSize();
+  // This is the amount of stack a funclet needs to allocate.
+  unsigned MaxCallSize = MF.getFrameInfo()->getMaxCallFrameSize();
+  // RBP is not included in the callee saved register block. After pushing RBP,
+  // everything is 16 byte aligned. Everything we allocate before an outgoing
+  // call must also be 16 byte aligned.
+  unsigned FrameSizeMinusRBP =
+      RoundUpToAlignment(CSSize + MaxCallSize, getStackAlignment());
+  // Subtract out the size of the callee saved registers. This is how much stack
+  // each funclet will allocate.
+  return FrameSizeMinusRBP - CSSize;
+}
+
 void X86FrameLowering::emitEpilogue(MachineFunction &MF,
                                     MachineBasicBlock &MBB) const {
   const MachineFrameInfo *MFI = MF.getFrameInfo();
@@ -1067,7 +1083,7 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
   uint64_t NumBytes = 0;
 
   if (MBBI->getOpcode() == X86::CATCHRET) {
-    NumBytes = MFI->getMaxCallFrameSize();
+    NumBytes = getWinEHFuncletFrameSize(MF);
     assert(hasFP(MF) && "EH funclets without FP not yet implemented");
     MachineBasicBlock *TargetMBB = MBBI->getOperand(0).getMBB();
 
@@ -1107,7 +1123,7 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
           .addMBB(TargetMBB);
     }
   } else if (MBBI->getOpcode() == X86::CLEANUPRET) {
-    NumBytes = MFI->getMaxCallFrameSize();
+    NumBytes = getWinEHFuncletFrameSize(MF);
     assert(hasFP(MF) && "EH funclets without FP not yet implemented");
     BuildMI(MBB, MBBI, DL, TII.get(Is64Bit ? X86::POP64r : X86::POP32r),
             MachineFramePtr)
@@ -2211,3 +2227,15 @@ MachineBasicBlock::iterator X86FrameLowering::restoreWin32EHStackPointers(
   }
   return MBBI;
 }
+
+unsigned X86FrameLowering::getWinEHParentFrameOffset(const MachineFunction &MF) const {
+  // RDX, the parent frame pointer, is homed into 16(%rsp) in the prologue.
+  unsigned Offset = 16;
+  // RBP is immediately pushed.
+  Offset += SlotSize;
+  // All callee-saved registers are then pushed.
+  Offset += MF.getInfo<X86MachineFunctionInfo>()->getCalleeSavedFrameSize();
+  // Every funclet allocates enough stack space for the largest outgoing call.
+  Offset += getWinEHFuncletFrameSize(MF);
+  return Offset;
+}
diff --git a/llvm/lib/Target/X86/X86FrameLowering.h b/llvm/lib/Target/X86/X86FrameLowering.h
index b9cf3a99bb3..59c6a062810 100644
--- a/llvm/lib/Target/X86/X86FrameLowering.h
+++ b/llvm/lib/Target/X86/X86FrameLowering.h
@@ -101,6 +101,8 @@ public:
                                  MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator MI) const override;
 
+  unsigned getWinEHParentFrameOffset(const MachineFunction &MF) const override;
+
   /// Check the instruction before/after the passed instruction. If
   /// it is an ADD/SUB/LEA instruction it is deleted argument and the
   /// stack adjustment is returned as a positive value for ADD/LEA and
@@ -152,6 +154,8 @@ private:
   restoreWin32EHStackPointers(MachineBasicBlock &MBB,
                               MachineBasicBlock::iterator MBBI, DebugLoc DL,
                               bool RestoreSP = false) const;
+
+  unsigned getWinEHFuncletFrameSize(const MachineFunction &MF) const;
 };
 
 } // End llvm namespace
diff --git a/llvm/test/CodeGen/X86/win-catchpad-csrs.ll b/llvm/test/CodeGen/X86/win-catchpad-csrs.ll
index 5e8c3350bca..32031e2787e 100644
--- a/llvm/test/CodeGen/X86/win-catchpad-csrs.ll
+++ b/llvm/test/CodeGen/X86/win-catchpad-csrs.ll
@@ -125,14 +125,20 @@ catchendblock:                                    ; preds = %catch,
 ; X64: .seh_pushreg 5
 ; X64: pushq %rsi
 ; X64: .seh_pushreg 6
-; X64: subq $32, %rsp
-; X64: .seh_stackalloc 32
+; X64: pushq %rdi
+; X64: .seh_pushreg 7
+; X64: pushq %rbx
+; X64: .seh_pushreg 3
+; X64: subq $40, %rsp
+; X64: .seh_stackalloc 40
 ; X64: leaq 32(%rdx), %rbp
 ; X64: .seh_endprologue
 ; X64: movl $2, %ecx
 ; X64: callq f
 ; X64: leaq [[contbb]](%rip), %rax
-; X64: addq $32, %rsp
+; X64: addq $40, %rsp
+; X64: popq %rbx
+; X64: popq %rdi
 ; X64: popq %rsi
 ; X64: popq %rbp
 ; X64: retq
@@ -142,4 +148,130 @@ catchendblock:                                    ; preds = %catch,
 ; X64:   .long   "??_R0H@8"@IMGREL
 ; X64:   .long   0
 ; X64:   .long   "?catch$[[catch1bb]]@?0?try_catch_catch@4HA"@IMGREL
+; X64:   .long   88
+
+define i32 @try_one_csr() personality i32 (...)* @__CxxFrameHandler3 {
+entry:
+  %a = call i32 @getint()
+  %b = call i32 @getint()
+  call void (...) @useints(i32 %a)
+  invoke void @f(i32 1)
+          to label %try.cont unwind label %catch.dispatch
+
+catch.dispatch:                                   ; preds = %entry
+  %0 = catchpad [%rtti.TypeDescriptor2* @"\01??_R0H@8", i32 0, i8* null]
+          to label %catch unwind label %catchendblock
+
+catch:
+  catchret %0 to label %try.cont
+
+try.cont:                                         ; preds = %entry, %invoke.cont.2, %invoke.cont.3
+  ret i32 0
+
+catchendblock:                                    ; preds = %catch,
+  catchendpad unwind to caller
+}
+
+; X64-LABEL: try_one_csr:
+; X64: pushq %rbp
+; X64: .seh_pushreg 5
+; X64: pushq %rsi
+; X64: .seh_pushreg 6
+; X64-NOT: pushq
+; X64: subq $40, %rsp
+; X64: .seh_stackalloc 40
+; X64: leaq 32(%rsp), %rbp
+; X64: .seh_setframe 5, 32
+; X64: .seh_endprologue
+; X64: callq getint
+; X64: callq getint
+; X64: callq useints
+; X64: movl $1, %ecx
+; X64: callq f
+; X64: [[contbb:\.LBB1_[0-9]+]]: # %try.cont
+; X64: addq $40, %rsp
+; X64-NOT: popq
+; X64: popq %rsi
+; X64: popq %rbp
+; X64: retq
+
+; X64: "?catch$[[catch1bb:[0-9]+]]@?0?try_one_csr@4HA":
+; X64: LBB1_[[catch1bb]]: # %catch.dispatch{{$}}
+; X64: movq %rdx, 16(%rsp)
+; X64: pushq %rbp
+; X64: .seh_pushreg 5
+; X64: pushq %rsi
+; X64: .seh_pushreg 6
+; X64: subq $40, %rsp
+; X64: .seh_stackalloc 40
+; X64: leaq 32(%rdx), %rbp
+; X64: .seh_endprologue
+; X64: leaq [[contbb]](%rip), %rax
+; X64: addq $40, %rsp
+; X64: popq %rsi
+; X64: popq %rbp
+; X64: retq
+
+; X64: $handlerMap$0$try_one_csr:
+; X64:   .long   0
+; X64:   .long   "??_R0H@8"@IMGREL
+; X64:   .long   0
+; X64:   .long   "?catch$[[catch1bb]]@?0?try_one_csr@4HA"@IMGREL
+; X64:   .long   72
+
+define i32 @try_no_csr() personality i32 (...)* @__CxxFrameHandler3 {
+entry:
+  invoke void @f(i32 1)
+          to label %try.cont unwind label %catch.dispatch
+
+catch.dispatch:                                   ; preds = %entry
+  %0 = catchpad [%rtti.TypeDescriptor2* @"\01??_R0H@8", i32 0, i8* null]
+          to label %catch unwind label %catchendblock
+
+catch:
+  catchret %0 to label %try.cont
+
+try.cont:                                         ; preds = %entry, %invoke.cont.2, %invoke.cont.3
+  ret i32 0
+
+catchendblock:                                    ; preds = %catch,
+  catchendpad unwind to caller
+}
+
+; X64-LABEL: try_no_csr:
+; X64: pushq %rbp
+; X64: .seh_pushreg 5
+; X64-NOT: pushq
+; X64: subq $48, %rsp
+; X64: .seh_stackalloc 48
+; X64: leaq 48(%rsp), %rbp
+; X64: .seh_setframe 5, 48
+; X64: .seh_endprologue
+; X64: movl $1, %ecx
+; X64: callq f
+; X64: [[contbb:\.LBB2_[0-9]+]]: # %try.cont
+; X64: addq $48, %rsp
+; X64-NOT: popq
+; X64: popq %rbp
+; X64: retq
+
+; X64: "?catch$[[catch1bb:[0-9]+]]@?0?try_no_csr@4HA":
+; X64: LBB2_[[catch1bb]]: # %catch.dispatch{{$}}
+; X64: movq %rdx, 16(%rsp)
+; X64: pushq %rbp
+; X64: .seh_pushreg 5
+; X64: subq $32, %rsp
+; X64: .seh_stackalloc 32
+; X64: leaq 48(%rdx), %rbp
+; X64: .seh_endprologue
+; X64: leaq [[contbb]](%rip), %rax
+; X64: addq $32, %rsp
+; X64: popq %rbp
+; X64: retq
+
+; X64: $handlerMap$0$try_no_csr:
+; X64:   .long   0
+; X64:   .long   "??_R0H@8"@IMGREL
+; X64:   .long   0
+; X64:   .long   "?catch$[[catch1bb]]@?0?try_no_csr@4HA"@IMGREL
 ; X64:   .long   56