Change AArch64 Windows EH UnwindHelp object to be a fixed object

[danielframpton]

Summary

The UnwindHelp object is used during exception handling by runtime code. It must be findable from a fixed offset from FP.

This change allocates the UnwindHelp object as a fixed object (as is done for x86_64) to ensure that both the generated code and runtime agree on the location of the object.

Fixes: https://bugs.llvm.org/show_bug.cgi?id=45346

[efriedma]

If you're not using arcanist, please upload patches with full context (-U10000).

llvm/lib/Target/AArch64/AArch64FrameLowering.cpp

219

Maybe elaborate a bit; this is specifically the first space allocated in the function, next to sp on entry.

1011

Maybe move the "Var args are accounted for in the containing function" comment into getFixedObjectSize?

2683

This loop is strange. There's space explicitly allocated for the object in getFixedObjectSize, so you should be able to directly compute its offset.

2690

Why does UnwindHelp need to be 16-byte aligned?

llvm/test/CodeGen/AArch64/wineh-unwindhelp-via-fp.ll

10

This testcase seems overly complicated; if you're only checking the assembly output for ?func2@@YAXXZ, no need to include the other definitions.

[danielframpton]

Updates from review feedback.

[danielframpton]

Thanks for the review!

[danielframpton]

Added responses to the review and pointed out places where there are important codegen differences.

llvm/lib/Target/AArch64/AArch64FrameLowering.cpp

219

Is this good?

1011

Done.

2683

This loop exists in the corresponding X86 code, but the use of FixedObject there is different.

I have changed to use the calculation from above instead.

2690

The unwind help object itself only needs 8 byte alignment, but the code was expecting the FixedObject area to be 16 byte aligned.

llvm/test/CodeGen/AArch64/seh-finally.ll

94

This is the key difference in codegen (using fp to store the -2).

213

This is the key difference in codegen (using fp to store the -2).

[efriedma]

LGTM

I assume you don't have commit access, so I'll commit it for you. How would you like me to credit you in the commit message (name and email address)?

(The unnecessary alignment is probably something we could fix, but not worth the complexity.)

[danielframpton]

In D77016#1950918, @efriedma wrote:

LGTM

I assume you don't have commit access, so I'll commit it for you. How would you like me to credit you in the commit message (name and email address)?

(The unnecessary alignment is probably something we could fix, but not worth the complexity.)

I do not have commit access, so that would be very helpful, Thanks! You can use Daniel Frampton <[e-mail]>

[danielframpton]

In D77016#1950944, @danielframpton wrote:

In D77016#1950918, @efriedma wrote:

LGTM

I assume you don't have commit access, so I'll commit it for you. How would you like me to credit you in the commit message (name and email address)?

(The unnecessary alignment is probably something we could fix, but not worth the complexity.)

I do not have commit access, so that would be very helpful, Thanks! You can use Daniel Frampton <[e-mail]>

I am doing some final validation of the version you accepted (using it to self-host rustc on windows). I will let you know that is complete.

[danielframpton]

In D77016#1951715, @danielframpton wrote:

In D77016#1950944, @danielframpton wrote:

In D77016#1950918, @efriedma wrote:

LGTM

I assume you don't have commit access, so I'll commit it for you. How would you like me to credit you in the commit message (name and email address)?

(The unnecessary alignment is probably something we could fix, but not worth the complexity.)

I do not have commit access, so that would be very helpful, Thanks! You can use Daniel Frampton <[e-mail]>

I am doing some final validation of the version you accepted (using it to self-host rustc on windows). I will let you know that is complete.

My additional validation showed no issues. Please let me know if there is anything you need me to do.

llvm/lib/Target/AArch64/AArch64FrameLowering.cpp

   return TargetStackID::SVEVector;
 }
 
+/// Returns the size of fixed object (to determine the prologue size).
+static unsigned getFixedObjectSize(const MachineFunction &MF,
+                                   const AArch64FunctionInfo *AFI, bool IsWin64,
+                                   bool IsFunclet) {
+  if (!IsWin64 || IsFunclet) {
+    return 0;
+  } else {
+    const unsigned VarArgsArea = alignTo(AFI->getVarArgsGPRSize(), 16);
+    const unsigned UnwindHelpObject = (MF.hasEHFunclets() ? 16 : 0);
+    return VarArgsArea + UnwindHelpObject;
+  }
+}
+
 /// Returns the size of the entire SVE stackframe (calleesaves + spills).
 static StackOffset getSVEStackSize(const MachineFunction &MF) {
   const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();

       Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv());
   // Var args are accounted for in the containing function, so don't
   // include them for funclets.
-  unsigned FixedObject = (IsWin64 && !IsFunclet) ?
-                         alignTo(AFI->getVarArgsGPRSize(), 16) : 0;
+  unsigned FixedObject = getFixedObjectSize(MF, AFI, IsWin64, IsFunclet);
 
   auto PrologueSaveSize = AFI->getCalleeSavedStackSize() + FixedObject;
   // All of the remaining stack allocations are for locals.

       Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv());
   // Var args are accounted for in the containing function, so don't
   // include them for funclets.
-  unsigned FixedObject =
-      (IsWin64 && !IsFunclet) ? alignTo(AFI->getVarArgsGPRSize(), 16) : 0;
+  unsigned FixedObject = getFixedObjectSize(MF, AFI, IsWin64, IsFunclet);
 
   uint64_t AfterCSRPopSize = ArgumentPopSize;
   auto PrologueSaveSize = AFI->getCalleeSavedStackSize() + FixedObject;

   const auto &Subtarget = MF.getSubtarget<AArch64Subtarget>();
   bool IsWin64 =
       Subtarget.isCallingConvWin64(MF.getFunction().getCallingConv());
-  unsigned FixedObject = IsWin64 ? alignTo(AFI->getVarArgsGPRSize(), 16) : 0;
+
+  unsigned FixedObject =
+      getFixedObjectSize(MF, AFI, IsWin64, /*IsFunclet=*/false);
   unsigned FPAdjust = isTargetDarwin(MF)
                         ? 16 : AFI->getCalleeSavedStackSize(MF.getFrameInfo());
   return {ObjectOffset + FixedObject + FPAdjust, MVT::i8};

   while (MBBI != MBB.end() && MBBI->getFlag(MachineInstr::FrameSetup))
     ++MBBI;
 
+  // Find the minimum fixed object offset already allocated.
+  int64_t MinFixedObjOffset = -getOffsetOfLocalArea();
+  for (int I = MFI.getObjectIndexBegin(); I < 0; ++I)
+    MinFixedObjOffset = std::min(MinFixedObjOffset, MFI.getObjectOffset(I));
+
+  // Allocate space for the UnwindHelp object
+  assert(MinFixedObjOffset <= 0 && "Unexpected MinFixedObjOffset");
+  int64_t UnwindHelpOffset = MinFixedObjOffset -= 8;
+
+  // Ensure alignment.
+  UnwindHelpOffset = -alignTo(-UnwindHelpOffset, 16);
+
   // Create an UnwindHelp object.
-  int UnwindHelpFI =
-      MFI.CreateStackObject(/*size*/8, /*alignment*/16, false);
+  int UnwindHelpFI = MFI.CreateFixedObject(/*size*/ 8, UnwindHelpOffset,
+                                           /*IsImmutable=*/false);
   EHInfo.UnwindHelpFrameIdx = UnwindHelpFI;
+
   // We need to store -2 into the UnwindHelp object at the start of the
   // function.
   DebugLoc DL;

     const MachineFunction &MF, int FI, unsigned &FrameReg,
     bool IgnoreSPUpdates) const {
   const MachineFrameInfo &MFI = MF.getFrameInfo();
-  LLVM_DEBUG(dbgs() << "Offset from the SP for " << FI << " is "
-                    << MFI.getObjectOffset(FI) << "\n");
-  FrameReg = AArch64::SP;
-  return MFI.getObjectOffset(FI);
+  if (IgnoreSPUpdates) {
+    LLVM_DEBUG(dbgs() << "Offset from the SP for " << FI << " is "
+                      << MFI.getObjectOffset(FI) << "\n");
+    FrameReg = AArch64::SP;
+    return MFI.getObjectOffset(FI);
+  }
+
+  return getFrameIndexReference(MF, FI, FrameReg);
 }
 
 /// The parent frame offset (aka dispFrame) is only used on X86_64 to retrieve

llvm/test/CodeGen/AArch64/seh-finally.ll

 ; CHECK-LABEL: simple_seh
 ; CHECK: add     x29, sp, #16
 ; CHECK: mov     x0, #-2
-; CHECK: stur    x0, [x29, #-16]
+; CHECK: stur    x0, [x29, #16]
 ; CHECK: .set .Lsimple_seh$frame_escape_0, -8
 ; CHECK: ldur    w0, [x29, #-8]
 ; CHECK: bl      foo

 entry:
 ; CHECK-LABEL: stack_realign
 ; CHECK: mov     x29, sp
-; CHECK: sub     x9, sp, #64
+; CHECK: sub     x9, sp, #16
 ; CHECK: and     sp, x9, #0xffffffffffffffe0
 ; CHECK: mov     x19, sp
 ; CHECK: mov     x0, #-2
-; CHECK: stur    x0, [x19, #16]
-; CHECK: .set .Lstack_realign$frame_escape_0, 32
-; CHECK: ldr     w0, [x19, #32]
+; CHECK: stur    x0, [x29, #32]
+; CHECK: .set .Lstack_realign$frame_escape_0, 0
+; CHECK: ldr     w0, [x19]
 ; CHECK: bl      foo
 
   %o = alloca %struct.S, align 32

 ; CHECK-LABEL: vla_present
 ; CHECK: add     x29, sp, #32
 ; CHECK: mov     x1, #-2
-; CHECK: stur    x1, [x29, #-32]
+; CHECK: stur    x1, [x29, #16]
 ; CHECK: .set .Lvla_present$frame_escape_0, -4
 ; CHECK: stur    w0, [x29, #-4]
 ; CHECK: ldur    w8, [x29, #-4]

 entry:
 ; CHECK-LABEL: vla_and_realign
 ; CHECK: mov     x29, sp
-; CHECK: sub     x9, sp, #64
+; CHECK: sub     x9, sp, #48
 ; CHECK: and     sp, x9, #0xffffffffffffffe0
 ; CHECK: mov     x19, sp
 ; CHECK: mov     x1, #-2
-; CHECK: stur    x1, [x19]
+; CHECK: stur    x1, [x29, #32]
 ; CHECK: .set .Lvla_and_realign$frame_escape_0, 32
-; CHECK: str     w0, [x29, #28]
-; CHECK: ldr     w8, [x29, #28]
+; CHECK: str     w0, [x29, #44]
+; CHECK: ldr     w8, [x29, #44]
 ; CHECK: mov     x9, sp
-; CHECK: str     x9, [x19, #24]
-; CHECK: str     x8, [x19, #16]
+; CHECK: str     x9, [x29, #24]
+; CHECK: str     x8, [x19, #24]
 ; CHECK: ldr     w0, [x19, #32]
 ; CHECK: bl      foo
 

llvm/test/CodeGen/AArch64/wineh-try-catch-cbz.ll

 ; but the original issue only reproduced if the cbz was immediately
 ; after the frame setup.)
 
-; CHECK:      sub     sp, sp, #32
-; CHECK-NEXT: stp     x29, x30, [sp, #16]
-; CHECK-NEXT: add     x29, sp, #16
+; CHECK: stp     x29, x30, [sp, #-32]!
+; CHECK-NEXT: mov     x29, sp
 ; CHECK-NEXT: mov     x1, #-2
-; CHECK-NEXT: stur    x1, [x29, #-16]
+; CHECK-NEXT: stur    x1, [x29, #16]
 ; CHECK-NEXT: cbz     w0, .LBB0_2
 
 target datalayout = "e-m:w-p:64:64-i32:32-i64:64-i128:128-n32:64-S128"

llvm/test/CodeGen/AArch64/wineh-try-catch-realign.ll

 ; CHECK:      stp     x29, x30, [sp, #-32]!
 ; CHECK-NEXT: str     x28, [sp, #16]
 ; CHECK-NEXT: str     x19, [sp, #24]
-; CHECK-NEXT: add     x0, x19, #64
+; CHECK-NEXT: add     x0, x19, #0
 ; CHECK-NEXT: mov     w1, wzr
 ; CHECK-NEXT: bl      "?bb@@YAXPEAHH@Z"
 ; CHECK-NEXT: adrp    x0, .LBB0_1

llvm/test/CodeGen/AArch64/wineh-try-catch.ll

 ;    and the parent function.
 
 ; The following checks that the unwind help object has -2 stored into it at
-; fp - 400 - 256 = fp - 656, which is on-entry sp - 48 + 32 - 656 =
-; on-entry sp - 672.  We check this offset in the table later on.
+; fp + 16, which is on-entry sp - 16.
+; We check this offset in the table later on.
 
 ; CHECK-LABEL: "?func@@YAHXZ":
-; CHECK:       stp     x29, x30, [sp, #-48]!
+; CHECK:       stp     x29, x30, [sp, #-64]!
 ; CHECK:       str     x28, [sp, #16]
 ; CHECK:       str     x21, [sp, #24]
 ; CHECK:       stp     x19, x20, [sp, #32]

 ; CHECK:       sub     sp, sp, #624
 ; CHECK:       mov     x19, sp
 ; CHECK:       mov     x0, #-2
-; CHECK:       stur    x0, [x19]
+; CHECK:       stur    x0, [x29, #48]
 
 ; Now check that x is stored at fp - 20.  We check that this is the same
 ; location accessed from the funclet to retrieve x.

 
 ; Now check that the offset of the unwind help object from the stack pointer on
 ; entry to func is encoded in cppxdata that is passed to __CxxFrameHandler3.  As
-; computed above, this comes to -672.
+; computed above, this comes to -16.
 ; CHECK-LABEL:        "$cppxdata$?func@@YAHXZ":
 ; CHECK-NEXT:         .word   429065506               ; MagicNumber
 ; CHECK-NEXT:         .word   2                       ; MaxState

 ; CHECK-NEXT:         .word   ("$tryMap$?func@@YAHXZ")@IMGREL ; TryBlockMap
 ; CHECK-NEXT:         .word   4                       ; IPMapEntries
 ; CHECK-NEXT:         .word   ("$ip2state$?func@@YAHXZ")@IMGREL ; IPToStateXData
-; CHECK-NEXT:         .word   -672                    ; UnwindHelp
+; CHECK-NEXT:         .word   -16                     ; UnwindHelp
 
 ; UNWIND: Function: ?func@@YAHXZ (0x0)
 ; UNWIND: Prologue [

 ; UNWIND-NEXT: ; stp x19, x20, [sp, #32]
 ; UNWIND-NEXT: ; str x21, [sp, #24]
 ; UNWIND-NEXT: ; str x28, [sp, #16]
-; UNWIND-NEXT: ; stp x29, x30, [sp, #-48]!
+; UNWIND-NEXT: ; stp x29, x30, [sp, #-64]!
 ; UNWIND-NEXT: ; end
 ; UNWIND: Function: ?catch$2@?0??func@@YAHXZ@4HA
 ; UNWIND: Prologue [

llvm/test/CodeGen/AArch64/wineh-unwindhelp-via-fp.ll

+; RUN: llc -o - %s -mtriple=aarch64-windows | FileCheck %s
+; Check that we allocate the unwind help stack object in a fixed location from fp
+; so that the runtime can find it when handling an exception
+target datalayout = "e-m:w-p:64:64-i32:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-pc-windows-msvc19.25.28611"
+
+; Check that the store to the unwind help object for func2 is via FP
+; CHECK-LABEL: ?func2@@YAXXZ
+; CHECK: mov x[[#SCRATCH_REG:]], #-2
+; CHECK: stur x[[#SCRATCH_REG:]], [x29, #[[#]]]
+;
+; // struct that requires greater than stack alignment
+; struct alignas(32) A
+; {
+;     // data that would be invalid for unwind help (> 0)
+;     int _x[4]{42, 42, 42, 42};
+;     ~A() {}
+; };
+; 
+; // cause us to run the funclet in func2
+; void func3()
+; {
+;     throw 1;
+; }
+; 
+; // the funclet that ensures we have the unwind help correct
+; void func2()
+; {
+;     A a;
+;     func3();
+; }
+; 
+; // function to ensure we are misaligned in func2
+; void func1()
+; {
+;     func2();
+; }
+; 
+; // set things up and ensure alignment for func1
+; void test()
+; {
+;     try {
+;         A a;
+;         func1();
+;     } catch(...) {}
+; }
+
+%rtti.TypeDescriptor2 = type { i8**, i8*, [3 x i8] }
+%eh.CatchableType = type { i32, i32, i32, i32, i32, i32, i32 }
+%eh.CatchableTypeArray.1 = type { i32, [1 x i32] }
+%eh.ThrowInfo = type { i32, i32, i32, i32 }
+%struct.A = type { [4 x i32], [16 x i8] }
+
+$"??0A@@QEAA@XZ" = comdat any
+
+$"??1A@@QEAA@XZ" = comdat any
+
+$"??_R0H@8" = comdat any
+
+$"_CT??_R0H@84" = comdat any
+
+$_CTA1H = comdat any
+
+$_TI1H = comdat any
+
+@"??_7type_info@@6B@" = external constant i8*
+@"??_R0H@8" = linkonce_odr global %rtti.TypeDescriptor2 { i8** @"??_7type_info@@6B@", i8* null, [3 x i8] c".H\00" }, comdat
+@__ImageBase = external dso_local constant i8
+@"_CT??_R0H@84" = linkonce_odr unnamed_addr constant %eh.CatchableType { i32 1, i32 trunc (i64 sub nuw nsw (i64 ptrtoint (%rtti.TypeDescriptor2* @"??_R0H@8" to i64), i64 ptrtoint (i8* @__ImageBase to i64)) to i32), i32 0, i32 -1, i32 0, i32 4, i32 0 }, section ".xdata", comdat
+@_CTA1H = linkonce_odr unnamed_addr constant %eh.CatchableTypeArray.1 { i32 1, [1 x i32] [i32 trunc (i64 sub nuw nsw (i64 ptrtoint (%eh.CatchableType* @"_CT??_R0H@84" to i64), i64 ptrtoint (i8* @__ImageBase to i64)) to i32)] }, section ".xdata", comdat
+@_TI1H = linkonce_odr unnamed_addr constant %eh.ThrowInfo { i32 0, i32 0, i32 0, i32 trunc (i64 sub nuw nsw (i64 ptrtoint (%eh.CatchableTypeArray.1* @_CTA1H to i64), i64 ptrtoint (i8* @__ImageBase to i64)) to i32) }, section ".xdata", comdat
+
+; Function Attrs: noinline optnone uwtable
+define dso_local void @"?func3@@YAXXZ"() #0 {
+  %1 = alloca i32, align 4
+  store i32 1, i32* %1, align 4
+  %2 = bitcast i32* %1 to i8*
+  call void @_CxxThrowException(i8* %2, %eh.ThrowInfo* @_TI1H) #2
+  unreachable
+}
+
+declare dso_local void @_CxxThrowException(i8*, %eh.ThrowInfo*)
+
+; Function Attrs: noinline optnone uwtable
+define dso_local void @"?func2@@YAXXZ"() #0 personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
+  %1 = alloca %struct.A, align 32
+  %2 = call %struct.A* @"??0A@@QEAA@XZ"(%struct.A* %1) #3
+  invoke void @"?func3@@YAXXZ"()
+          to label %3 unwind label %4
+
+3:                                                ; preds = %0
+  call void @"??1A@@QEAA@XZ"(%struct.A* %1) #3
+  ret void
+
+4:                                                ; preds = %0
+  %5 = cleanuppad within none []
+  call void @"??1A@@QEAA@XZ"(%struct.A* %1) #3 [ "funclet"(token %5) ]
+  cleanupret from %5 unwind to caller
+}
+
+; Function Attrs: noinline nounwind optnone uwtable
+define linkonce_odr dso_local %struct.A* @"??0A@@QEAA@XZ"(%struct.A* returned %0) unnamed_addr #1 comdat align 2 {
+  %2 = alloca %struct.A*, align 8
+  store %struct.A* %0, %struct.A** %2, align 8
+  %3 = load %struct.A*, %struct.A** %2, align 8
+  %4 = getelementptr inbounds %struct.A, %struct.A* %3, i32 0, i32 0
+  %5 = getelementptr inbounds [4 x i32], [4 x i32]* %4, i64 0, i64 0
+  store i32 42, i32* %5, align 4
+  %6 = getelementptr inbounds i32, i32* %5, i64 1
+  store i32 42, i32* %6, align 4
+  %7 = getelementptr inbounds i32, i32* %6, i64 1
+  store i32 42, i32* %7, align 4
+  %8 = getelementptr inbounds i32, i32* %7, i64 1
+  store i32 42, i32* %8, align 4
+  ret %struct.A* %3
+}
+
+declare dso_local i32 @__CxxFrameHandler3(...)
+
+; Function Attrs: noinline nounwind optnone uwtable
+define linkonce_odr dso_local void @"??1A@@QEAA@XZ"(%struct.A* %0) unnamed_addr #1 comdat align 2 {
+  %2 = alloca %struct.A*, align 8
+  store %struct.A* %0, %struct.A** %2, align 8
+  %3 = load %struct.A*, %struct.A** %2, align 8
+  ret void
+}
+
+; Function Attrs: noinline optnone uwtable
+define dso_local void @"?func1@@YAXXZ"() #0 {
+  call void @"?func2@@YAXXZ"()
+  ret void
+}
+
+; Function Attrs: noinline optnone uwtable
+define dso_local void @"?test@@YAXXZ"() #0 personality i8* bitcast (i32 (...)* @__CxxFrameHandler3 to i8*) {
+  %1 = alloca %struct.A, align 32
+  %2 = call %struct.A* @"??0A@@QEAA@XZ"(%struct.A* %1) #3
+  invoke void @"?func1@@YAXXZ"()
+          to label %3 unwind label %4
+
+3:                                                ; preds = %0
+  call void @"??1A@@QEAA@XZ"(%struct.A* %1) #3
+  br label %11
+
+4:                                                ; preds = %0
+  %5 = cleanuppad within none []
+  call void @"??1A@@QEAA@XZ"(%struct.A* %1) #3 [ "funclet"(token %5) ]
+  cleanupret from %5 unwind label %6
+
+6:                                                ; preds = %4
+  %7 = catchswitch within none [label %8] unwind to caller
+
+8:                                                ; preds = %6
+  %9 = catchpad within %7 [i8* null, i32 64, i8* null]
+  catchret from %9 to label %10
+
+10:                                               ; preds = %8
+  br label %11
+
+11:                                               ; preds = %10, %3
+  ret void
+}
+
+!llvm.module.flags = !{!0}
+
+!0 = !{i32 1, !"wchar_size", i32 2}