[llvm-debuginfo-analyzer] Fix crash with WebAssembly dead code (llvm#141616)

llvm#136772
Incorrect handling of 'tombstone' value for WebAssembly.

llvm-debuginfo-analyzer already uses the tombstone approach
to identify dead code. Currently, the tombstone value is
evaluated as std::numeric_limits<uint64_t>::max(). Which is
wrong as it does not take into account the 'Address Byte Size'
from the Compile Unit header.

Author: CarlosAlbertoEnciso

llvm/include/llvm/DebugInfo/LogicalView/Core/LVObject.h

@@ -84,6 +84,14 @@ using LVTypes = SmallVector<LVType *, 8>;
 
 using LVOffsets = SmallVector<LVOffset, 8>;
 
+// The following DWARF documents detail the 'tombstone' concept:
+//   https://dwarfstd.org/issues/231013.1.html
+//   https://dwarfstd.org/issues/200609.1.html
+//
+// The value of the largest representable address offset (for example,
+// 0xffffffff when the size of an address is 32 bits).
+//
+// -1 (0xffffffff) => Valid tombstone
 const LVAddress MaxAddress = std::numeric_limits<uint64_t>::max();
 
 enum class LVBinaryType { NONE, ELF, COFF };

llvm/include/llvm/DebugInfo/LogicalView/Readers/LVDWARFReader.h

@@ -58,6 +58,9 @@ class LVDWARFReader final : public LVBinaryReader {
   bool FoundLowPC = false;
   bool FoundHighPC = false;
 
+  // The value is updated for each Compile Unit that is processed.
+  std::optional<LVAddress> TombstoneAddress;
+
   // Cross references (Elements).
   using LVElementSet = std::unordered_set<LVElement *>;
   struct LVElementEntry {
@@ -127,6 +130,12 @@ class LVDWARFReader final : public LVBinaryReader {
   LVAddress getCUHighAddress() const { return CUHighAddress; }
   void setCUHighAddress(LVAddress Address) { CUHighAddress = Address; }
 
+  void setTombstoneAddress(LVAddress Address) { TombstoneAddress = Address; }
+  LVAddress getTombstoneAddress() const {
+    assert(TombstoneAddress && "Unset tombstone value");
+    return TombstoneAddress.value();
+  }
+
   const LVSymbols &GetSymbolsWithLocations() const {
     return SymbolsWithLocations;
   }

llvm/lib/DebugInfo/LogicalView/Readers/LVDWARFReader.cpp

@@ -214,10 +214,11 @@ void LVDWARFReader::processOneAttribute(const DWARFDie &Die,
         }
       }
       if (FoundLowPC) {
-        if (CurrentLowPC == MaxAddress)
+        if (CurrentLowPC == getTombstoneAddress())
           CurrentElement->setIsDiscarded();
-        // Consider the case of WebAssembly.
-        CurrentLowPC += WasmCodeSectionOffset;
+        else
+          // Consider the case of WebAssembly.
+          CurrentLowPC += WasmCodeSectionOffset;
         if (CurrentElement->isCompileUnit())
           setCUBaseAddress(CurrentLowPC);
       }
@@ -271,7 +272,8 @@ void LVDWARFReader::processOneAttribute(const DWARFDie &Die,
       DWARFAddressRangesVector Ranges = RangesOrError.get();
       for (DWARFAddressRange &Range : Ranges) {
         // This seems to be a tombstone for empty ranges.
-        if (Range.LowPC == Range.HighPC)
+        if ((Range.LowPC == Range.HighPC) ||
+            (Range.LowPC = getTombstoneAddress()))
           continue;
         // Store the real upper limit for the address range.
         if (UpdateHighAddress && Range.HighPC > 0)
@@ -629,6 +631,11 @@ Error LVDWARFReader::createScopes() {
                                          : DwarfContext->dwo_compile_units();
   for (const std::unique_ptr<DWARFUnit> &CU : CompileUnits) {
 
+    // Take into account the address byte size for a correct 'tombstone'
+    // value identification.
+    setTombstoneAddress(
+        dwarf::computeTombstoneAddress(CU->getAddressByteSize()));
+
     // Deduction of index used for the line records.
     //
     // For the following test case: test.cpp

llvm/test/tools/llvm-debuginfo-analyzer/WebAssembly/wasm-32bit-tombstone.s

@@ -0,0 +1,107 @@
+# REQUIRES: x86-registered-target
+
+# Test that DWARF tombstones are correctly detected/respected in wasm
+# 32 bit object files.
+
+# The test case was produced by the following steps:
+#
+# // test-clang.cpp
+# void foo() {
+# }
+#
+# 1) clang --target=wasm32 -S -g test-clang.cpp
+#                             -o Inputs/wasm-32bit-tombstone.s
+#
+# 2) Creating a single function, tombstoning it in the assembly, by
+#    manually changing the DW_AT_low_pc for the DW_TAG_subprogram:
+#    .Lfunc_begin0 to 0xffffffff to mark the function as dead code:
+#
+#	   .int8	2                          # Abbrev [2] 0x26:0x1b DW_TAG_subprogram
+#	   .int32	.Lfunc_begin0              # DW_AT_low_pc  <---------
+#	   .int32	.Lfunc_end0-.Lfunc_begin0  # DW_AT_high_pc
+
+#	   .int8	2                          # Abbrev [2] 0x26:0x1b DW_TAG_subprogram
+#	   .int32	0xffffffff                 # DW_AT_low_pc  <---------
+#	   .int32	.Lfunc_end0-.Lfunc_begin0  # DW_AT_high_pc
+
+# RUN: llvm-mc -arch=wasm32 -filetype=obj       \
+# RUN:         %p/wasm-32bit-tombstone.s        \
+# RUN:         -o %t.wasm-32bit-tombstone.wasm
+
+# RUN: llvm-debuginfo-analyzer --select-elements=Discarded         \
+# RUN:                         --print=elements                    \
+# RUN:                         %t.wasm-32bit-tombstone.wasm 2>&1 | \
+# RUN: FileCheck --strict-whitespace -check-prefix=ONE %s
+
+# ONE: Logical View:
+# ONE-NEXT:           {File} '{{.*}}wasm-32bit-tombstone.wasm'
+# ONE-EMPTY:
+# ONE-NEXT:           {CompileUnit} 'test-clang.cpp'
+# ONE-NEXT:           {Function} not_inlined 'foo' -> 'void'
+
+# RUN: llvm-dwarfdump --debug-info %t.wasm-32bit-tombstone.wasm | \
+# RUN: FileCheck %s --check-prefix=TWO
+
+# TWO:      DW_TAG_subprogram
+# TWO-NEXT:   DW_AT_low_pc	(dead code)
+# TWO-NEXT:   DW_AT_high_pc
+# TWO-NEXT:   DW_AT_name	("foo")
+
+	.text
+	.file	"test-clang.cpp"
+	.functype	_Z3foov () -> ()
+	.section	.text._Z3foov,"",@
+_Z3foov:                                # @_Z3foov
+.Lfunc_begin0:
+	.functype	_Z3foov () -> ()
+	return
+	end_function
+.Lfunc_end0:
+                                        # -- End function
+	.section	.debug_abbrev,"",@
+	.int8	1                               # Abbreviation Code
+	.int8	17                              # DW_TAG_compile_unit
+	.int8	1                               # DW_CHILDREN_yes
+	.int8	3                               # DW_AT_name
+	.int8	14                              # DW_FORM_strp
+	.int8	17                              # DW_AT_low_pc
+	.int8	1                               # DW_FORM_addr
+	.int8	18                              # DW_AT_high_pc
+	.int8	6                               # DW_FORM_data4
+	.int8	0                               # EOM(1)
+	.int8	0                               # EOM(2)
+	.int8	2                               # Abbreviation Code
+	.int8	46                              # DW_TAG_subprogram
+	.int8	0                               # DW_CHILDREN_no
+	.int8	17                              # DW_AT_low_pc
+	.int8	1                               # DW_FORM_addr
+	.int8	18                              # DW_AT_high_pc
+	.int8	6                               # DW_FORM_data4
+	.int8	3                               # DW_AT_name
+	.int8	14                              # DW_FORM_strp
+	.int8	0                               # EOM(1)
+	.int8	0                               # EOM(2)
+	.int8	0                               # EOM(3)
+	.section	.debug_info,"",@
+.Lcu_begin0:
+	.int32	.Ldebug_info_end0-.Ldebug_info_start0 # Length of Unit
+.Ldebug_info_start0:
+	.int16	4                               # DWARF version number
+	.int32	.debug_abbrev0                  # Offset Into Abbrev. Section
+	.int8	4                               # Address Size (in bytes)
+	.int8	1                               # Abbrev [1] 0xb:0x37 DW_TAG_compile_unit
+	.int32	.Linfo_string1                  # DW_AT_name
+	.int32	.Lfunc_begin0                   # DW_AT_low_pc
+	.int32	.Lfunc_end0-.Lfunc_begin0       # DW_AT_high_pc
+	.int8	2                               # Abbrev [2] 0x26:0x1b DW_TAG_subprogram
+	.int32	0xffffffff                   # DW_AT_low_pc
+	.int32	.Lfunc_end0-.Lfunc_begin0       # DW_AT_high_pc
+	.int32	.Linfo_string4                  # DW_AT_name
+	.int8	0                               # End Of Children Mark
+.Ldebug_info_end0:
+	.section	.debug_str,"S",@
+.Linfo_string1:
+	.asciz	"test-clang.cpp"                # string offset=176
+.Linfo_string4:
+	.asciz	"foo"                           # string offset=241
+	.ident	"clang version 19.0.0"