[lld] [lld][ELF] Implement merged .debug_names section. (PR #86508)
Fangrui Song via llvm-commits
llvm-commits at lists.llvm.org
Mon Mar 25 23:02:05 PDT 2024
================
@@ -2690,6 +2693,759 @@ static uint32_t computeGdbHash(StringRef s) {
return h;
}
+DebugNamesSection::DebugNamesSection()
+ : SyntheticSection(0, SHT_PROGBITS, 1, ".debug_names") {}
+
+template <class ELFT, class RelTy>
+void DebugNamesSection::getNameRelocsImpl(
+ InputSection *sec, ArrayRef<RelTy> rels,
+ llvm::DenseMap<uint32_t, uint32_t> &relocs) {
+ for (auto &rel : rels) {
+ Symbol &sym = sec->getFile<ELFT>()->getRelocTargetSym(rel);
+ relocs[rel.r_offset] = sym.getVA(getAddend<ELFT>(rel));
+ }
+}
+
+template <class ELFT>
+void DebugNamesSection::getNameRelocs(
+ InputSectionBase *base, llvm::DenseMap<uint32_t, uint32_t> &relocs) {
+ auto *sec = cast<InputSection>(base);
+ const RelsOrRelas<ELFT> rels = sec->template relsOrRelas<ELFT>();
+ if (rels.areRelocsRel())
+ getNameRelocsImpl<ELFT>(sec, rels.rels, relocs);
+ else
+ getNameRelocsImpl<ELFT>(sec, rels.relas, relocs);
+}
+
+void DebugNamesSection::writeTo(uint8_t *buf) { invokeELFT(writeToImpl, buf); }
+
+template <class ELFT> void DebugNamesSection::writeToImpl(uint8_t *buf) {
+ SmallVector<uint32_t, 0> mergedCuOffsets;
+ SmallVector<uint32_t, 0> mergedTuOffsets;
+ llvm::DenseMap<uint32_t, uint32_t> strOffsets;
+ SmallVector<llvm::DenseMap<uint32_t, uint32_t>, 0> chunksRelocs;
+ chunksRelocs.reserve(numChunks);
+
+ for (size_t i = 0; i < numChunks; ++i) {
+ DebugNamesOutputChunk &chunk = outputChunks[i];
+ InputSectionBase *base = inputDebugNamesSections[i];
+ llvm::DenseMap<uint32_t, uint32_t> relocs;
+ getNameRelocs<ELFT>(base, relocs);
+ chunksRelocs.push_back(std::move(relocs));
+
+ // Update CuOffsets list with new data
+ for (uint32_t cuOffset : chunk.compilationUnits)
+ mergedCuOffsets.push_back(chunk.sec->outSecOff + cuOffset);
+
+ // TODO: Update TuOffsets list with new data
+ }
+
+ // Update the entries with the relocated string offsets.
+ for (auto &stringEntry : mergedEntries) {
+ uint32_t oldOffset = stringEntry.stringOffsetOffset;
+ uint32_t idx = stringEntry.chunkIdx;
+ stringEntry.relocatedEntryOffset = chunksRelocs[idx][oldOffset];
+ }
+
+ // Write out bytes for merged section.
+
+ // Write the header.
+ endian::write32<ELFT::TargetEndianness>(buf + 0, mergedHdr.UnitLength);
+ endian::write16<ELFT::TargetEndianness>(buf + 4, mergedHdr.Version);
+ endian::write32<ELFT::TargetEndianness>(buf + 8, mergedHdr.CompUnitCount);
+ endian::write32<ELFT::TargetEndianness>(buf + 12,
+ mergedHdr.LocalTypeUnitCount);
+ endian::write32<ELFT::TargetEndianness>(buf + 16,
+ mergedHdr.ForeignTypeUnitCount);
+ endian::write32<ELFT::TargetEndianness>(buf + 20, mergedHdr.BucketCount);
+ endian::write32<ELFT::TargetEndianness>(buf + 24, mergedHdr.NameCount);
+ endian::write32<ELFT::TargetEndianness>(buf + 28, mergedHdr.AbbrevTableSize);
+ endian::write32<ELFT::TargetEndianness>(buf + 32,
+ mergedHdr.AugmentationStringSize);
+ buf += 36;
+ memcpy(buf, mergedHdr.AugmentationString.c_str(), 8);
+ buf += 8;
+
+ // Write the CU list.
+ for (uint32_t offset : mergedCuOffsets) {
+ endian::write32<ELFT::TargetEndianness>(buf + 0, offset);
+ buf += 4;
+ }
+
+ // Write the local TU list.
+ // TODO: Fix this, once we get everything working without TUs.
+ if (mergedHdr.LocalTypeUnitCount != 0)
+ warn(".debug_names: type units are not handled in merged index");
+
+ // Write the foreign TU list.
+ // Currently LLVM does not emit foreign type units, so there should
+ // be nothing here to emit.
+ // TODO: Fix this, once we get everything working wtihout TUs.
+ if (mergedHdr.ForeignTypeUnitCount != 0)
+ warn(".debug_names: type units are not handled in merged index");
+
+ // Write the hash table.
+ // ... Write the buckets
+ uint32_t idx = 1;
+ for (const auto &bucket : bucketList) {
+ if (!bucket.empty())
+ endian::write32<ELFT::TargetEndianness>(buf + 0, idx);
+ idx += bucket.size();
+ buf += 4;
+ }
+
+ // ...Write the hashes
+ for (const auto &bucket : bucketList) {
+ for (const auto &entry : bucket) {
+ uint32_t hashValue = entry->hashValue;
+ endian::write32<ELFT::TargetEndianness>(buf + 0, hashValue);
+ buf += 4;
+ }
+ }
+
+ // Write the string offsets.
+ for (const auto &entry : mergedEntries) {
+ endian::write32<ELFT::TargetEndianness>(buf + 0,
+ entry.relocatedEntryOffset);
+ buf += 4;
+ }
+
+ // Write the entry offsets.
+ for (const auto &entry : mergedEntries) {
+ endian::write32<ELFT::TargetEndianness>(buf + 0, entry.entryOffset);
+ buf += 4;
+ }
+
+ // Write the abbrev table.
+ for (const auto *abbrev : mergedAbbrevTable) {
+ size_t uleb_size = encodeULEB128(abbrev->code, buf);
+ buf += uleb_size;
+ uleb_size = encodeULEB128(abbrev->tag, buf);
+ buf += uleb_size;
+ for (auto attr : abbrev->attributes) {
+ uleb_size = encodeULEB128(attr.Index, buf);
+ buf += uleb_size;
+ uleb_size = encodeULEB128(attr.Form, buf);
+ buf += uleb_size;
+ }
+ endian::write16<ELFT::TargetEndianness>(buf + 0, 0); // attribute sentinels.
+ buf += 2;
+ }
+ *buf++ = 0; // abbrev table sentinel
+
+ // Write the entry pool.
+ for (const auto &stringEntry : mergedEntries) {
+ // Write all the entries for the string.
+ size_t uleb_size;
+ for (const auto &entry : stringEntry.indexEntries) {
+ uleb_size = encodeULEB128(entry->abbrevCode, buf);
+ buf += uleb_size;
+ for (const auto &value : entry->attrValues) {
+ if (value.attrSize > 0) {
+ endian::write32<ELFT::TargetEndianness>(buf + 0, value.attrValue);
+ buf += value.attrSize;
+ }
+ }
+ }
+ *buf++ = 0; // Entry sentinel
+ }
+}
+
+bool DebugNamesSection::isNeeded() const { return numChunks > 0; }
+
+template <class ELFT>
+static void
+readCompileUnitOffsets(struct DebugNamesSection::DebugNamesSectionData &secData,
+ DebugNamesSection::DebugNamesInputChunk &inputChunk,
+ DebugNamesSection::DebugNamesOutputChunk &outputChunk,
+ llvm::DWARFDataExtractor &namesExtractor) {
+ uint64_t offset = secData.locs.CUsBase;
+ uint64_t *offsetPtr = &offset;
+ for (size_t i = 0; i < secData.hdr.CompUnitCount; ++i) {
+ llvm::Error err = llvm::Error::success();
+ uint32_t value = namesExtractor.getU32(offsetPtr, &err);
+ if (err)
+ errorOrWarn(toString(inputChunk.namesSection->sec) +
+ ": error reading CU offsets: " + toString(std::move(err)));
+ else
+ outputChunk.compilationUnits.push_back(value);
+ }
+}
+
+template <class ELFT>
+static void
+readEntryOffsets(struct DebugNamesSection::DebugNamesSectionData &secData,
+ DebugNamesSection::DebugNamesInputChunk &chunk,
+ llvm::DWARFDataExtractor &namesExtractor) {
+ secData.entryOffsets = std::make_unique<uint32_t[]>(secData.hdr.NameCount);
+ if (secData.locs.EntryOffsetsBase >= namesExtractor.getData().size())
+ errorOrWarn(toString(chunk.namesSection->sec) +
+ ": invalid entry offsets input");
+
+ uint64_t offset = secData.locs.EntryOffsetsBase;
+ uint64_t *offsetPtr = &offset;
+ for (size_t i = 0; i < secData.hdr.NameCount; ++i) {
+ llvm::Error err = llvm::Error::success();
+ uint32_t value = namesExtractor.getU32(offsetPtr, &err);
+ if (err)
+ errorOrWarn(toString(chunk.namesSection->sec) +
+ ": error reading entry offsets: " + toString(std::move(err)));
+ else
+ secData.entryOffsets.get()[i] = value;
+ }
+}
+
+template <class ELFT>
+static void readAttributeValues(
+ SmallVector<DebugNamesSection::AttrValueData, 3> &values,
+ DebugNamesSection::DebugNamesInputChunk &chunk, uint64_t &offset,
+ struct DebugNamesSection::DebugNamesSectionData &secData,
+ int32_t &parentOffset, llvm::DWARFDataExtractor &namesExtractor,
+ const llvm::DWARFDebugNames::Abbrev &abbrev) {
+ const LLDDWARFSection &namesSection = *chunk.namesSection;
+ uint64_t *offsetPtr = &offset;
+ DebugNamesSection::AttrValueData cuOrTuAttr = {0, 0};
+ for (auto attr : abbrev.Attributes) {
+ llvm::Error err = llvm::Error::success();
+ DebugNamesSection::AttrValueData newAttr;
+ uint32_t value;
+ switch (attr.Form) {
+ case DW_FORM_flag_present: {
+ // Currently only DW_IDX_parent attributes (in .debug_names) can
+ // have this form. This form does not have a real value (nothing is
+ // emitted for it).
+ break;
+ }
+ case DW_FORM_data1:
+ case DW_FORM_ref1: {
+ newAttr.attrValue = namesExtractor.getU8(offsetPtr, &err);
+ newAttr.attrSize = 1;
+ break;
+ }
+ case DW_FORM_data2:
+ case DW_FORM_ref2: {
+ value = namesExtractor.getU16(offsetPtr, &err);
+ newAttr.attrValue = value;
+ newAttr.attrSize = 2;
+ break;
+ }
+ case DW_FORM_data4:
+ case DW_FORM_ref4: {
+ value = namesExtractor.getU32(offsetPtr, &err);
+ newAttr.attrValue = value;
+ newAttr.attrSize = 4;
+ if (attr.Index == dwarf::DW_IDX_parent)
+ parentOffset = value + secData.locs.EntriesBase;
+ break;
+ }
+ case DW_FORM_data8:
+ case DW_FORM_ref8:
+ case DW_FORM_ref_sig8: {
+ value = namesExtractor.getU64(offsetPtr, &err);
+ newAttr.attrValue = value;
+ newAttr.attrSize = 8;
+ break;
+ }
+ default: {
+ errorOrWarn(toString(namesSection.sec) +
+ Twine(": unrecognized form encoding ") + Twine(attr.Form) +
+ " in .debug_names abbrev table");
+ break;
+ }
+ }
+ if (err)
+ errorOrWarn(
+ toString(namesSection.sec) +
+ ": error while reading attributes: " + toString(std::move(err)));
+ if (attr.Index == DW_IDX_compile_unit || attr.Index == DW_IDX_type_unit)
+ // Save it to put it at the end of the attributes list.
+ cuOrTuAttr = newAttr;
+ else if (attr.Form != DW_FORM_flag_present)
+ values.push_back(newAttr);
+ }
+
+ // Make sure the CU or TU index attribute is the last one in the list.
+ values.push_back(cuOrTuAttr);
+}
+
+template <class ELFT>
+static void readEntry(DebugNamesSection::NamedEntry &stringEntry,
+ DebugNamesSection::DebugNamesInputChunk &chunk,
+ DebugNamesSection::DebugNamesSectionData &secData,
+ llvm::DWARFDataExtractor &namesExtractor,
+ const llvm::DWARFDebugNames::NameIndex &ni) {
+ std::unique_ptr<LLDDWARFSection> &namesSection = chunk.namesSection;
+ uint64_t offset = stringEntry.entryOffset;
+ // Each entry ends with a zero 'sentinel' value
+ while (offset < namesSection->Data.size() &&
+ namesSection->Data[offset] != 0) {
+ // Read & store all entries (for the same string)
+ auto entry = std::make_unique<DebugNamesSection::IndexEntry>();
+ entry->poolOffset = offset;
+ llvm::Error err = llvm::Error::success();
+ uint64_t ulebValue = namesExtractor.getULEB128(&offset, &err);
+ if (err)
+ errorOrWarn(toString(chunk.namesSection->sec) +
+ ": error reading entry: " + toString(std::move(err)));
+ entry->abbrevCode = static_cast<uint32_t>(ulebValue);
+ const auto &abbrevs = ni.getAbbrevs();
+ auto abbrevIt = abbrevs.find_as(entry->abbrevCode);
+ if (abbrevIt != abbrevs.end()) {
+ const llvm::DWARFDebugNames::Abbrev &abbrev = *abbrevIt;
+ readAttributeValues<ELFT>(entry->attrValues, chunk, offset, secData,
+ entry->parentOffset, namesExtractor, abbrev);
+ stringEntry.indexEntries.push_back(std::move(entry));
+ }
+ }
+ if (offset >= namesSection->Data.size())
+ errorOrWarn(toString(chunk.namesSection->sec) +
+ ": encountered unexpected end of section while reading entry");
+}
+
+template <class ELFT>
+static void
+readEntries(struct DebugNamesSection::DebugNamesSectionData &secData,
+ DebugNamesSection::DebugNamesInputChunk &chunk,
+ llvm::DWARFDataExtractor &namesExtractor,
+ llvm::DataExtractor &strExtractor,
+ const llvm::DWARFDebugNames::NameIndex &ni) {
+ // Temporary map from entry offsets to address (pointer) of entry with that
+ // offset; used to find parent entries quickly.
+ DenseMap<uint32_t, DebugNamesSection::IndexEntry *> offsetMap;
+ // Reserve sizes for this chunk's hashes & namedEntries.
+ chunk.hashValues.reserve(secData.hdr.NameCount);
+ secData.namedEntries.reserve(secData.hdr.NameCount);
+ // Calculate the Entry Offsets, create initial records.
+ for (uint32_t i = 0; i < secData.hdr.NameCount; ++i) {
+ // Get string value
+ DebugNamesSection::NamedEntry stringEntry;
+ stringEntry.entryOffset =
+ secData.locs.EntriesBase + secData.entryOffsets[i];
+ uint64_t strOffsetOffset =
+ secData.locs.StringOffsetsBase + i * secData.dwarfSize;
+ stringEntry.stringOffsetOffset = strOffsetOffset;
+ uint64_t strOffset =
+ namesExtractor.getRelocatedValue(secData.dwarfSize, &strOffsetOffset);
+ StringRef name = strExtractor.getCStrRef(&strOffset);
+ stringEntry.name = name.data();
+ // Calculate hash
+ stringEntry.hashValue = caseFoldingDjbHash(name);
+ chunk.hashValues.push_back(stringEntry.hashValue);
+ // Read String Entry
+ readEntry<ELFT>(stringEntry, chunk, secData, namesExtractor, ni);
+ // Add index entries & offsets to our temporary map
+ for (const auto &entry : stringEntry.indexEntries)
+ offsetMap[entry->poolOffset] = entry.get();
+ secData.namedEntries.push_back(std::move(stringEntry));
+ }
+ // Find/assign pointers to any 'real' parent entries (needed to find correct
+ // parent entry offsets in merged data).
+ for (auto &stringEntry : secData.namedEntries)
+ for (auto &entry : stringEntry.indexEntries)
+ if (entry->parentOffset != -1)
+ entry->parentEntry = offsetMap[entry->parentOffset];
+}
+
+static uint16_t computeDebugNamesHeaderSize() {
+ // Size of the .debug_names section header, in bytes, for DWARF32:
+ uint16_t size = /* unit length */ 4 +
+ /* version */ 2 +
+ /* padding */ 2 +
+ /* CU count */ 4 +
+ /* TU count */ 4 +
+ /* Foreign TU count */ 4 +
+ /* Bucket Count */ 4 +
+ /* Name Count */ 4 +
+ /* Abbrev table size */ 4 +
+ /* Augmentation string size */ 4 +
+ /* augmentation string */ 8;
+ return size;
+}
+
+template <class ELFT>
+static void collectDebugNamesSectionData(
+ DebugNamesSection::DebugNamesInputChunk &chunk,
+ DebugNamesSection::DebugNamesOutputChunk &outputChunk,
+ llvm::DWARFDataExtractor &namesExtractor,
+ llvm::DataExtractor &strExtractor) {
+ for (const auto &ni : *chunk.debugNamesData) {
----------------
MaskRay wrote:
`s/auto/DebugNames::NameIndex/`
https://github.com/llvm/llvm-project/pull/86508
More information about the llvm-commits
mailing list