[DwarfDebug] Refuse to emit DW_OP_LLVM_arg values wider than 64 bits

DwarfExpression::addUnsignedConstant(const APInt &Value) only supports
wider-than-64-bit values when it is used to emit a top-level DWARF
expression representing the location of a variable. Before this change,
it was possible to call addUnsignedConstant on >64 bit values within a
subexpression when substituting DW_OP_LLVM_arg values.

This can trigger an assertion failure (e.g. PR52584, PR52333) when it
happens in a fragment (DW_OP_LLVM_fragment) expression, as
addUnsignedConstant on >64 bit values splits the constant into separate
DW_OP_pieces, which modifies DwarfExpression::OffsetInBits.

This change papers over the assertion errors by bailing on overly wide
DW_OP_LLVM_arg values. A more comprehensive fix might be to be to split
wide values into pointer-sized fragments.

[0] https://github.com/llvm/llvm-project/blob/e71fa03/llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp#L799-L805

Patch by Ricky Zhou!

Differential Revision: https://reviews.llvm.org/D115343

Author: adrian-prantl

llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp

@@ -789,7 +789,7 @@ DIE *DwarfCompileUnit::constructVariableDIEImpl(const DbgVariable &DV,
     const TargetRegisterInfo &TRI = *Asm->MF->getSubtarget().getRegisterInfo();
 
     auto AddEntry = [&](const DbgValueLocEntry &Entry,
-                                            DIExpressionCursor &Cursor) {
+                        DIExpressionCursor &Cursor) {
       if (Entry.isLocation()) {
         if (!DwarfExpr.addMachineRegExpression(TRI, Cursor,
                                                Entry.getLoc().getReg()))
@@ -798,11 +798,19 @@ DIE *DwarfCompileUnit::constructVariableDIEImpl(const DbgVariable &DV,
         // If there is an expression, emit raw unsigned bytes.
         DwarfExpr.addUnsignedConstant(Entry.getInt());
       } else if (Entry.isConstantFP()) {
+        // DwarfExpression does not support arguments wider than 64 bits
+        // (see PR52584).
+        // TODO: Consider chunking expressions containing overly wide
+        // arguments into separate pointer-sized fragment expressions.
         APInt RawBytes = Entry.getConstantFP()->getValueAPF().bitcastToAPInt();
-        DwarfExpr.addUnsignedConstant(RawBytes);
+        if (RawBytes.getBitWidth() > 64)
+          return false;
+        DwarfExpr.addUnsignedConstant(RawBytes.getZExtValue());
       } else if (Entry.isConstantInt()) {
         APInt RawBytes = Entry.getConstantInt()->getValue();
-        DwarfExpr.addUnsignedConstant(RawBytes);
+        if (RawBytes.getBitWidth() > 64)
+          return false;
+        DwarfExpr.addUnsignedConstant(RawBytes.getZExtValue());
       } else if (Entry.isTargetIndexLocation()) {
         TargetIndexLocation Loc = Entry.getTargetIndexLocation();
         // TODO TargetIndexLocation is a target-independent. Currently only the
@@ -815,11 +823,12 @@ DIE *DwarfCompileUnit::constructVariableDIEImpl(const DbgVariable &DV,
       return true;
     };
 
-    DwarfExpr.addExpression(
-        std::move(Cursor),
-        [&](unsigned Idx, DIExpressionCursor &Cursor) -> bool {
-          return AddEntry(DVal->getLocEntries()[Idx], Cursor);
-        });
+    if (!DwarfExpr.addExpression(
+            std::move(Cursor),
+            [&](unsigned Idx, DIExpressionCursor &Cursor) -> bool {
+              return AddEntry(DVal->getLocEntries()[Idx], Cursor);
+            }))
+      return VariableDie;
 
     // Now attach the location information to the DIE.
     addBlock(*VariableDie, dwarf::DW_AT_location, DwarfExpr.finalize());

llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp

@@ -463,15 +463,14 @@ static bool isMemoryLocation(DIExpressionCursor ExprCursor) {
   return true;
 }
 
-void DwarfExpression::addExpression(DIExpressionCursor &&ExprCursor,
-                                    unsigned FragmentOffsetInBits) {
+void DwarfExpression::addExpression(DIExpressionCursor &&ExprCursor) {
   addExpression(std::move(ExprCursor),
                 [](unsigned Idx, DIExpressionCursor &Cursor) -> bool {
                   llvm_unreachable("unhandled opcode found in expression");
                 });
 }
 
-void DwarfExpression::addExpression(
+bool DwarfExpression::addExpression(
     DIExpressionCursor &&ExprCursor,
     llvm::function_ref<bool(unsigned, DIExpressionCursor &)> InsertArg) {
   // Entry values can currently only cover the initial register location,
@@ -496,7 +495,7 @@ void DwarfExpression::addExpression(
     case dwarf::DW_OP_LLVM_arg:
       if (!InsertArg(Op->getArg(0), ExprCursor)) {
         LocationKind = Unknown;
-        return;
+        return false;
       }
       break;
     case dwarf::DW_OP_LLVM_fragment: {
@@ -527,7 +526,7 @@ void DwarfExpression::addExpression(
       setSubRegisterPiece(0, 0);
       // Reset the location description kind.
       LocationKind = Unknown;
-      return;
+      return true;
     }
     case dwarf::DW_OP_plus_uconst:
       assert(!isRegisterLocation());
@@ -630,6 +629,8 @@ void DwarfExpression::addExpression(
   if (isImplicitLocation() && !isParameterValue())
     // Turn this into an implicit location description.
     addStackValue();
+
+  return true;
 }
 
 /// add masking operations to stencil out a subregister.

llvm/lib/CodeGen/AsmPrinter/DwarfExpression.h

@@ -340,16 +340,17 @@ class DwarfExpression {
   /// create one if necessary.
   unsigned getOrCreateBaseType(unsigned BitSize, dwarf::TypeKind Encoding);
 
+  /// Emit all remaining operations in the DIExpressionCursor. The
+  /// cursor must not contain any DW_OP_LLVM_arg operations.
+  void addExpression(DIExpressionCursor &&Expr);
+
   /// Emit all remaining operations in the DIExpressionCursor.
-  ///
-  /// \param FragmentOffsetInBits     If this is one fragment out of multiple
-  ///                                 locations, this is the offset of the
-  ///                                 fragment inside the entire variable.
-  void addExpression(DIExpressionCursor &&Expr,
-                     unsigned FragmentOffsetInBits = 0);
-  void
-  addExpression(DIExpressionCursor &&Expr,
-                llvm::function_ref<bool(unsigned, DIExpressionCursor &)> InsertArg);
+  /// DW_OP_LLVM_arg operations are resolved by calling (\p InsertArg).
+  //
+  /// \return false if any call to (\p InsertArg) returns false.
+  bool addExpression(
+      DIExpressionCursor &&Expr,
+      llvm::function_ref<bool(unsigned, DIExpressionCursor &)> InsertArg);
 
   /// If applicable, emit an empty DW_OP_piece / DW_OP_bit_piece to advance to
   /// the fragment described by \c Expr.

llvm/test/DebugInfo/X86/pr52584.ll

@@ -0,0 +1,33 @@
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -filetype=obj -o - %s \
+; RUN:     | llvm-dwarfdump - \
+; RUN:     | FileCheck %s
+
+; CHECK:      DW_TAG_variable
+; CHECK-NOT:    DW_AT_location
+; CHECK-NEXT:   DW_AT_name  ("arr")
+; CHECK-NOT:    DW_AT_location
+; CHECK:      DW_TAG
+define dso_local void @test() !dbg !4 {
+entry:
+  call void @llvm.dbg.value(metadata !DIArgList(i128 0), metadata !7, metadata !DIExpression(DW_OP_LLVM_arg, 0, DW_OP_stack_value, DW_OP_LLVM_fragment, 0, 127)), !dbg !12
+  ret void, !dbg !12
+}
+
+declare void @llvm.dbg.value(metadata, metadata, metadata)
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "test.c", directory: "")
+!2 = !{i32 7, !"Dwarf Version", i32 4}
+!3 = !{i32 2, !"Debug Info Version", i32 3}
+!4 = distinct !DISubprogram(name: "test", scope: !1, file: !1, line: 1, type: !5, unit: !0)
+!5 = !DISubroutineType(types: !6)
+!6 = !{null}
+!7 = !DILocalVariable(name: "arr", scope: !4, file: !1, line: 1, type: !8)
+!8 = !DICompositeType(tag: DW_TAG_array_type, baseType: !9, size: 128, elements: !10)
+!9 = !DIBasicType(name: "char", size: 8, encoding: DW_ATE_signed_char)
+!10 = !{!11}
+!11 = !DISubrange(count: 16)
+!12 = !DILocation(line: 1, column: 1, scope: !4)