[PATCH] D118183: Fix UB in DwarfExpression::emitLegacyZExt()

[Adrian Prantl via Phabricator via llvm-commits]

aprantl updated this revision to Diff 403022.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D118183/new/

https://reviews.llvm.org/D118183

Files:
  llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp
  llvm/test/DebugInfo/X86/convert-debugloc.ll


Index: llvm/test/DebugInfo/X86/convert-debugloc.ll
===================================================================
--- llvm/test/DebugInfo/X86/convert-debugloc.ll
+++ llvm/test/DebugInfo/X86/convert-debugloc.ll
@@ -64,11 +64,17 @@
 ; NOCONV:       DW_AT_location (
 ; NOCONV:         {{.*}}, DW_OP_dup, DW_OP_constu 0x7, DW_OP_shr, DW_OP_lit0, DW_OP_not, DW_OP_mul, DW_OP_constu 0x8, DW_OP_shl, DW_OP_or, DW_OP_stack_value)
 ; NOCONV:       DW_AT_name ("y")
+; NOCONV:     DW_TAG_variable
+; NOCONV:       DW_AT_location (
+; NOCONV:         DW_OP_constu 0x40, DW_OP_lit0, DW_OP_plus, DW_OP_lit1, DW_OP_constu 0x40, DW_OP_shl, DW_OP_lit1, DW_OP_minus, DW_OP_and, DW_OP_stack_value)
+; NOCONV:       DW_AT_name ("z")
 ; NOCONV:     NULL
 ; NOCONV:   DW_TAG_base_type
 ; NOCONV:     DW_AT_name ("signed char")
 ; NOCONV:   DW_TAG_base_type
 ; NOCONV:     DW_AT_name ("int")
+; NOCONV:   DW_TAG_base_type
+; NOCONV:     DW_AT_name ("unsigned long long")
 ; NOCONV:   NULL
 
 
@@ -81,6 +87,7 @@
 ;; will not attempt to eliminate.  At the moment, only "convert" ops are folded.
 ;; If you have to change the expression, the expected DWO_id also changes.
   call void @llvm.dbg.value(metadata i8 32, metadata !13, metadata !DIExpression(DW_OP_lit0, DW_OP_plus, DW_OP_LLVM_convert, 8, DW_ATE_signed, DW_OP_LLVM_convert, 32, DW_ATE_signed, DW_OP_stack_value)), !dbg !15
+  call void @llvm.dbg.value(metadata i8 64, metadata !17, metadata !DIExpression(DW_OP_lit0, DW_OP_plus, DW_OP_LLVM_convert, 64, DW_ATE_unsigned, DW_OP_LLVM_convert, 128, DW_ATE_unsigned, DW_OP_LLVM_convert, 64, DW_ATE_unsigned, DW_OP_stack_value)), !dbg !15
   ret i8 %x, !dbg !16
 }
 
@@ -111,3 +118,5 @@
 !14 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
 !15 = !DILocation(line: 3, column: 14, scope: !7)
 !16 = !DILocation(line: 4, column: 3, scope: !7)
+!17 = !DILocalVariable(name: "z", scope: !7, file: !1, line: 3, type: !18)
+!18 = !DIBasicType(name: "unsigned long long", size: 64, encoding: DW_ATE_unsigned)
Index: llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp
===================================================================
--- llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp
+++ llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp
@@ -681,9 +681,25 @@
 }
 
 void DwarfExpression::emitLegacyZExt(unsigned FromBits) {
-  // (X & (1 << FromBits - 1))
-  emitOp(dwarf::DW_OP_constu);
-  emitUnsigned((1ULL << FromBits) - 1);
+  // Heuristic to decide the most efficient encoding.
+  // A ULEB can encode 7 1-bits per byte.
+  if (FromBits / 7 < 1+1+1+1+1) {
+    // (X & (1 << FromBits - 1))
+    emitOp(dwarf::DW_OP_constu);
+    emitUnsigned((1ULL << FromBits) - 1);
+  } else {
+    // Note that the DWARF 4 stack consists of pointer-sized elements,
+    // so technically it doesn't make sense to shift left more than 64
+    // bits. We leave that for the consumer to decide though. LLDB for
+    // example uses APInt for the stack elements and can still deal
+    // with this.
+    emitOp(dwarf::DW_OP_lit1);
+    emitOp(dwarf::DW_OP_constu);
+    emitUnsigned(FromBits);
+    emitOp(dwarf::DW_OP_shl);
+    emitOp(dwarf::DW_OP_lit1);
+    emitOp(dwarf::DW_OP_minus);
+  }
   emitOp(dwarf::DW_OP_and);
 }
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D118183.403022.patch
Type: text/x-patch
Size: 3211 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220125/af814473/attachment-0001.bin>