Add support for Dwarf 5 debug_info section in MCCAS

This includes, parsing the Dwarf 5 Compile Unit header correctly
and correctly creating and materializing from the CAS with the larger
DWARFv5 header.

Author: rastogishubham

llvm/include/llvm/MCCAS/MCCASDebugV1.h

@@ -20,6 +20,7 @@ namespace mccasformats {
 namespace v1 {
 
 constexpr unsigned Dwarf4HeaderSize32Bit = 11;
+constexpr unsigned Dwarf5HeaderSize32Bit = 12;
 
 /// Returns true if the values associated with a combination of Form and Attr
 /// are not expected to deduplicate.

llvm/include/llvm/MCCAS/MCCASObjectV1.h

@@ -581,6 +581,10 @@ class MCCASBuilder {
   struct CUSplit {
     SmallVector<MutableArrayRef<char>> SplitCUData;
     SmallVector<size_t> AbbrevOffsets;
+    /// A list of Dwarf versions of the compile units, to help determine what
+    /// the Compile Unit header format and size is like when creating
+    /// CASObjects.
+    SmallVector<uint16_t> DwarfVersions;
   };
   /// Split the data of the __debug_info section it into multiple pieces, one
   /// per Compile Unit(CU) and return them. The abbreviation offset for each CU

llvm/lib/MCCAS/MCCASObjectV1.cpp

@@ -112,12 +112,13 @@ class InMemoryCASDWARFObject : public DWARFObject {
   /// unit.
   Error partitionCUData(ArrayRef<char> DebugInfoData, uint64_t AbbrevOffset,
                         DWARFContext *Ctx, MCCASBuilder &Builder,
-                        AbbrevSetWriter &AbbrevWriter);
+                        AbbrevSetWriter &AbbrevWriter, uint16_t DwarfVersion);
 };
 
 struct CUInfo {
   uint64_t CUSize;
   uint32_t AbbrevOffset;
+  uint16_t DwarfVersion;
 };
 static Expected<CUInfo> getAndSetDebugAbbrevOffsetAndSkip(
     MutableArrayRef<char> CUData, support::endianness Endian,
@@ -1685,10 +1686,25 @@ getAndSetDebugAbbrevOffsetAndSkip(MutableArrayRef<char> CUData,
   if (auto E = Reader.readInteger(DwarfVersion))
     return std::move(E);
 
-  // TODO: Dwarf 5 has a different order for the next fields.
-  if (DwarfVersion != 4)
-    return createStringError(inconvertibleErrorCode(),
-                             "Expected Dwarf 4 input");
+  if (DwarfVersion >= 5) {
+    // From Dwarf 5 Section 7.5.1.1:
+    // Compile Unit Header Format is now changed with unit_type and address_size
+    // after the version. Parse both values from the header.
+    uint8_t UnitType;
+    if (auto E = Reader.readInteger(UnitType))
+      return std::move(E);
+    if (UnitType != dwarf::DW_UT_compile)
+      return createStringError(
+          inconvertibleErrorCode(),
+          "Unit type is not DW_UT_compile, and is incompatible with MCCAS!");
+    uint8_t AddressSize;
+    if (auto E = Reader.readInteger(AddressSize))
+      return std::move(E);
+    if (AddressSize != 8)
+      return createStringError(
+          inconvertibleErrorCode(),
+          "Address size is not 8 bytes, unsupported architecture for MCCAS!");
+  }
 
   // TODO: Handle Dwarf 64 format, which uses 8 bytes.
   size_t AbbrevPosition = Reader.getOffset();
@@ -1710,7 +1726,7 @@ getAndSetDebugAbbrevOffsetAndSkip(MutableArrayRef<char> CUData,
   if (auto E = Reader.skip(*Size))
     return std::move(E);
 
-  return CUInfo{Reader.getOffset(), AbbrevOffset};
+  return CUInfo{Reader.getOffset(), AbbrevOffset, DwarfVersion};
 }
 
 /// Given a list of MCFragments, return a vector with the concatenation of their
@@ -1750,6 +1766,7 @@ MCCASBuilder::splitDebugInfoSectionData(MutableArrayRef<char> DebugInfoData) {
       return Info.takeError();
     Split.SplitCUData.push_back(DebugInfoData.take_front(Info->CUSize));
     Split.AbbrevOffsets.push_back(Info->AbbrevOffset);
+    Split.DwarfVersions.push_back(Info->DwarfVersion);
     DebugInfoData = DebugInfoData.drop_front(Info->CUSize);
   }
 
@@ -1875,7 +1892,8 @@ Error InMemoryCASDWARFObject::partitionCUData(ArrayRef<char> DebugInfoData,
                                               uint64_t AbbrevOffset,
                                               DWARFContext *Ctx,
                                               MCCASBuilder &Builder,
-                                              AbbrevSetWriter &AbbrevWriter) {
+                                              AbbrevSetWriter &AbbrevWriter,
+                                              uint16_t DwarfVersion) {
   StringRef AbbrevSectionContribution =
       getAbbrevSection().drop_front(AbbrevOffset);
   DataExtractor Data(AbbrevSectionContribution, isLittleEndian(), 8);
@@ -1894,13 +1912,24 @@ Error InMemoryCASDWARFObject::partitionCUData(ArrayRef<char> DebugInfoData,
 
   DWARFDie CUDie = DCU.getUnitDIE(false);
   assert(CUDie);
-  // Copy 11 bytes which represents the 32-bit DWARF Header for DWARF4.
-  if (DebugInfoData.size() < Dwarf4HeaderSize32Bit)
-    return createStringError(inconvertibleErrorCode(),
-                             "DebugInfoData is too small, it doesn't even "
-                             "contain a 32-bit DWARF Header");
+  ArrayRef<char> HeaderData;
+  if (DwarfVersion >= 5) {
+    // Copy 12 bytes which represents the 32-bit DWARF Header for DWARF5.
+    if (DebugInfoData.size() < Dwarf5HeaderSize32Bit)
+      return createStringError(inconvertibleErrorCode(),
+                               "DebugInfoData is too small, it doesn't even "
+                               "contain a 32-bit DWARF5 Header");
 
-  ArrayRef<char> HeaderData = DebugInfoData.take_front(Dwarf4HeaderSize32Bit);
+    HeaderData = DebugInfoData.take_front(Dwarf5HeaderSize32Bit);
+  } else {
+    // Copy 11 bytes which represents the 32-bit DWARF Header for DWARF4.
+    if (DebugInfoData.size() < Dwarf4HeaderSize32Bit)
+      return createStringError(inconvertibleErrorCode(),
+                               "DebugInfoData is too small, it doesn't even "
+                               "contain a 32-bit DWARF4 Header");
+
+    HeaderData = DebugInfoData.take_front(Dwarf4HeaderSize32Bit);
+  }
   Expected<DIETopLevelRef> Converted =
       DIEToCASConverter(DebugInfoData, Builder)
           .convert(CUDie, HeaderData, AbbrevWriter);
@@ -1930,20 +1959,32 @@ Error MCCASBuilder::splitDebugInfoAndAbbrevSections() {
 
   Expected<SmallVector<char, 0>> FullAbbrevData =
       mergeMCFragmentContents(AbbrevFragmentList);
+
   if (!FullAbbrevData)
     return FullAbbrevData.takeError();
 
-  InMemoryCASDWARFObject CASObj(
-      *FullAbbrevData, Asm.getBackend().Endian == support::endianness::little);
+  const MCSection::FragmentListType &StringOffsetsFragmentList =
+      DwarfSections.StrOffsets->getFragmentList();
+
+  Expected<SmallVector<char, 0>> FullStringOffsetsData =
+      mergeMCFragmentContents(StringOffsetsFragmentList);
+
+  if (!FullStringOffsetsData)
+    return FullStringOffsetsData.takeError();
+
+  InMemoryCASDWARFObject CASObj(*FullAbbrevData, *FullStringOffsetsData,
+                                Asm.getBackend().Endian ==
+                                    support::endianness::little);
   auto DWARFObj = std::make_unique<InMemoryCASDWARFObject>(CASObj);
   auto DWARFContextHolder = std::make_unique<DWARFContext>(std::move(DWARFObj));
   auto *DWARFCtx = DWARFContextHolder.get();
 
   AbbrevSetWriter AbbrevWriter;
-  for (auto [CUData, AbbrevOffset] :
-       llvm::zip(SplitInfo->SplitCUData, SplitInfo->AbbrevOffsets)) {
+  for (auto [CUData, AbbrevOffset, DwarfVersion] :
+       llvm::zip(SplitInfo->SplitCUData, SplitInfo->AbbrevOffsets,
+                 SplitInfo->DwarfVersions)) {
     if (auto E = CASObj.partitionCUData(CUData, AbbrevOffset, DWARFCtx, *this,
-                                        AbbrevWriter))
+                                        AbbrevWriter, DwarfVersion))
       return E;
   }
   return Error::success();
@@ -3074,7 +3115,22 @@ Error mccasformats::v1::visitDebugInfo(
   StringRef DistinctData = LoadedTopRef->DistinctData.getData();
   BinaryStreamReader DistinctReader(DistinctData, support::endianness::little);
   ArrayRef<char> HeaderData;
-  if (auto E = DistinctReader.readArray(HeaderData, Dwarf4HeaderSize32Bit))
+
+  auto BeginOffset = DistinctReader.getOffset();
+  auto Size = getSizeFromDwarfHeader(DistinctReader);
+  if (!Size)
+    return Size.takeError();
+
+  // 2-byte Dwarf version identifier.
+  uint16_t DwarfVersion;
+  if (auto E = DistinctReader.readInteger(DwarfVersion))
+    return E;
+
+  DistinctReader.setOffset(BeginOffset);
+
+  if (auto E = DistinctReader.readArray(
+          HeaderData,
+          DwarfVersion >= 5 ? Dwarf5HeaderSize32Bit : Dwarf4HeaderSize32Bit))
     return E;
   HeaderCallback(toStringRef(HeaderData));