[Lldb-commits] [lldb] [lldb] Add SubtargetFeatures to ArchSpec (PR #173046)

[via lldb-commits]

================
@@ -1405,6 +1409,117 @@ void ObjectFileELF::ParseARMAttributes(DataExtractor &data, uint64_t length,
   }
 }
 
+static std::optional<lldb::offset_t>
+FindSubSectionOffsetByName(const DataExtractor &data, lldb::offset_t offset,
+                           uint32_t length, llvm::StringRef name) {
+  uint32_t section_length = 0;
+  llvm::StringRef section_name;
+  do {
+    offset += section_length;
+    // Sub-section's size and name are included in the total sub-section length.
+    // Don't shift the offset here, so it will point at the beginning of the
+    // sub-section and could be used as a return value.
+    auto tmp_offset = offset;
+    section_length = data.GetU32(&tmp_offset);
+    section_name = data.GetCStr(&tmp_offset);
+  } while (section_name != name && offset + section_length < length);
+  if (section_name == name)
+    return offset;
+  return std::nullopt;
+}
+
+static std::optional<lldb::offset_t>
+FindSubSubSectionOffsetByTag(const DataExtractor &data, lldb::offset_t offset,
+                             unsigned tag) {
+  // Consume a sub-section size and name to shift the offset at the beginning of
+  // the sub-sub-sections list.
+  auto upper_section_length = data.GetU32(&offset);
+  data.GetCStr(&offset);
+  auto upper_section_end_offset = offset + upper_section_length;
+
+  uint32_t section_length = 0;
+  unsigned section_tag = 0;
+  do {
+    offset += section_length;
+    // Similar to sub-section sub-sub-section's tag and size are included in the
+    // total sub-sub-section length.
+    auto tmp_offset = offset;
+    section_tag = data.GetULEB128(&tmp_offset);
+    section_length = data.GetU32(&tmp_offset);
+  } while (section_tag != tag &&
+           offset + section_length < upper_section_end_offset);
+  if (section_tag == tag)
+    return offset;
+  return std::nullopt;
+}
+
+static std::optional<std::variant<uint64_t, llvm::StringRef>>
+GetSubSubSubSectionValue(const DataExtractor &data, lldb::offset_t offset,
+                         unsigned tag) {
+  // Consume a sub-sub-section tag and size to shift the offset at the beginning
+  // of the sub-sub-sub-sections list.
+  data.GetULEB128(&offset);
+  auto upper_section_length = data.GetU32(&offset);
+  auto upper_section_end_offset = offset + upper_section_length;
+
+  std::variant<uint64_t, llvm::StringRef> result;
+  unsigned section_tag = 0;
+  do {
+    section_tag = data.GetULEB128(&offset);
+    // From the riscv psABI document:
+    // RISC-V attributes have a string value if the tag number is odd and an
+    // integer value if the tag number is even.
+    if (section_tag % 2)
+      result = data.GetCStr(&offset);
+    else
+      result = data.GetULEB128(&offset);
+  } while (section_tag != tag && offset < upper_section_end_offset);
+  if (section_tag == tag)
+    return result;
+  return std::nullopt;
+}
+
+void ObjectFileELF::ParseRISCVAttributes(DataExtractor &data, uint64_t length,
+                                         ArchSpec &arch_spec) {
+  Log *log = GetLog(LLDBLog::Modules);
+
+  lldb::offset_t offset = 0;
+
+  uint8_t format_version = data.GetU8(&offset);
+  if (format_version != llvm::ELFAttrs::Format_Version)
+    return;
+
+  auto subsection_or_opt =
+      FindSubSectionOffsetByName(data, offset, length, "riscv");
+  if (!subsection_or_opt) {
+    LLDB_LOGF(log,
+              "ObjectFileELF::%s Ill-formed .riscv.attributes section: "
+              "mandatory 'riscv' sub-section was not preserved",
+              __FUNCTION__);
+    return;
+  }
+
+  auto subsubsection_or_opt = FindSubSubSectionOffsetByTag(
+      data, *subsection_or_opt, llvm::ELFAttrs::File);
+  if (!subsubsection_or_opt)
+    return;
+
+  auto value_or_opt = GetSubSubSubSectionValue(data, *subsubsection_or_opt,
----------------
daniilavdeev wrote:

Indeed, that sounds a bit weird. The riscv-ps-abi defines a sub-section and a sub-sub-section:

<img width="924" height="325" alt="Screenshot from 2026-02-03 17-45-47" src="https://github.com/user-attachments/assets/36e76974-1a84-4630-ad16-555e10bd1dfc" />


According to the specification, the .riscv.attributes section has the following hierarchy:

**Section:** .riscv.attributes (SHT_RISCV_ATTRIBUTES)
**Sub-section:** "riscv\0"
**Sub-sub-section:** FILE
**Sub-sub-section attribute:** Tag_RISCV_arch, which holds the arch string

For some reason, I have named the Sub-sub-section attribute as `sub-sub-sub-section`, so it might be preferable to rename it to `SubSubSectionValue` and `GetSubSubSectionValueByTag`, or something similar.

https://github.com/llvm/llvm-project/pull/173046