[lld-macho] Fold __objc_imageinfo sections

Previously, we treated it as a regular ConcatInputSection. However, ld64
actually parses its contents and uses that to synthesize a single image
info struct, generating one 8-byte section instead of `8 * number of
object files with ObjC code`.

I'm not entirely sure what impact this section has on the runtime, so I
just tried to follow ld64's semantics as closely as possible in this
diff. My main motivation though was to reduce binary size.

No significant perf change on chromium_framework on my 16-core Mac Pro:

             base           diff           difference (95% CI)
  sys_time   1.764 ± 0.062  1.748 ± 0.032  [  -2.4% ..   +0.5%]
  user_time  5.112 ± 0.104  5.106 ± 0.046  [  -0.9% ..   +0.7%]
  wall_time  6.111 ± 0.184  6.085 ± 0.076  [  -1.6% ..   +0.8%]
  samples    30             32

Reviewed By: #lld-macho, thakis

Differential Revision: https://reviews.llvm.org/D130125

lld/MachO/Driver.cpp

@@ -586,7 +586,7 @@ static void initializeSectionRenameMap() {
                              section_names::objcCatList,
                              section_names::objcNonLazyCatList,
                              section_names::objcProtoList,
-                             section_names::objcImageInfo};
+                             section_names::objCImageInfo};
     for (StringRef s : v)
       config->sectionRenameMap[{segment_names::data, s}] = {
           segment_names::dataConst, s};
@@ -1102,6 +1102,8 @@ static void gatherInputSections() {
         }
       }
     }
+    if (!file->objCImageInfo.empty())
+      in.objCImageInfo->addFile(file);
   }
   assert(inputOrder <= UnspecifiedInputOrder);
 }

lld/MachO/InputFiles.cpp

@@ -363,6 +363,9 @@ void ObjFile::parseSections(ArrayRef<SectionHeader> sectionHeaders) {
       // have the same name without causing duplicate symbol errors. To avoid
       // spurious duplicate symbol errors, we do not parse these sections.
       // TODO: Evaluate whether the bitcode metadata is needed.
+    } else if (name == section_names::objCImageInfo &&
+               segname == segment_names::data) {
+      objCImageInfo = data;
     } else {
       if (name == section_names::addrSig)
         addrSigSection = sections.back();

lld/MachO/InputFiles.h

@@ -120,6 +120,7 @@ class InputFile {
 
   std::vector<Symbol *> symbols;
   std::vector<Section *> sections;
+  ArrayRef<uint8_t> objCImageInfo;
 
   // If not empty, this stores the name of the archive containing this file.
   // We use this string for creating error messages.

lld/MachO/InputSection.h

@@ -321,7 +321,7 @@ constexpr const char objcCatList[] = "__objc_catlist";
 constexpr const char objcClassList[] = "__objc_classlist";
 constexpr const char objcClassRefs[] = "__objc_classrefs";
 constexpr const char objcConst[] = "__objc_const";
-constexpr const char objcImageInfo[] = "__objc_imageinfo";
+constexpr const char objCImageInfo[] = "__objc_imageinfo";
 constexpr const char objcNonLazyCatList[] = "__objc_nlcatlist";
 constexpr const char objcNonLazyClassList[] = "__objc_nlclslist";
 constexpr const char objcProtoList[] = "__objc_protolist";

lld/MachO/SyntheticSections.cpp

@@ -1621,6 +1621,86 @@ void WordLiteralSection::writeTo(uint8_t *buf) const {
     memcpy(buf + p.second * 4, &p.first, 4);
 }
 
+ObjCImageInfoSection::ObjCImageInfoSection()
+    : SyntheticSection(segment_names::data, section_names::objCImageInfo) {}
+
+ObjCImageInfoSection::ImageInfo
+ObjCImageInfoSection::parseImageInfo(const InputFile *file) {
+  ImageInfo info;
+  ArrayRef<uint8_t> data = file->objCImageInfo;
+  // The image info struct has the following layout:
+  // struct {
+  //   uint32_t version;
+  //   uint32_t flags;
+  // };
+  if (data.size() < 8) {
+    warn(toString(file) + ": invalid __objc_imageinfo size");
+    return info;
+  }
+
+  auto *buf = reinterpret_cast<const uint32_t *>(data.data());
+  if (read32le(buf) != 0) {
+    warn(toString(file) + ": invalid __objc_imageinfo version");
+    return info;
+  }
+
+  uint32_t flags = read32le(buf + 1);
+  info.swiftVersion = (flags >> 8) & 0xff;
+  info.hasCategoryClassProperties = flags & 0x40;
+  return info;
+}
+
+static std::string swiftVersionString(uint8_t version) {
+  switch (version) {
+    case 1:
+      return "1.0";
+    case 2:
+      return "1.1";
+    case 3:
+      return "2.0";
+    case 4:
+      return "3.0";
+    case 5:
+      return "4.0";
+    default:
+      return ("0x" + Twine::utohexstr(version)).str();
+  }
+}
+
+// Validate each object file's __objc_imageinfo and use them to generate the
+// image info for the output binary. Only two pieces of info are relevant:
+// 1. The Swift version (should be identical across inputs)
+// 2. `bool hasCategoryClassProperties` (true only if true for all inputs)
+void ObjCImageInfoSection::finalizeContents() {
+  assert(files.size() != 0); // should have already been checked via isNeeded()
+
+  info.hasCategoryClassProperties = true;
+  const InputFile *firstFile;
+  for (auto file : files) {
+    ImageInfo inputInfo = parseImageInfo(file);
+    info.hasCategoryClassProperties &= inputInfo.hasCategoryClassProperties;
+
+    if (inputInfo.swiftVersion != 0) {
+      if (info.swiftVersion != 0 &&
+          info.swiftVersion != inputInfo.swiftVersion) {
+        error("Swift version mismatch: " + toString(firstFile) +
+              " has version " + swiftVersionString(info.swiftVersion) +
+              " but " + toString(file) + " has version " +
+              swiftVersionString(inputInfo.swiftVersion));
+      } else {
+        info.swiftVersion = inputInfo.swiftVersion;
+        firstFile = file;
+      }
+    }
+  }
+}
+
+void ObjCImageInfoSection::writeTo(uint8_t *buf) const {
+  uint32_t flags = info.hasCategoryClassProperties ? 0x40 : 0x0;
+  flags |= info.swiftVersion << 8;
+  write32le(buf + 4, flags);
+}
+
 void macho::createSyntheticSymbols() {
   auto addHeaderSymbol = [](const char *name) {
     symtab->addSynthetic(name, in.header->isec, /*value=*/0,

lld/MachO/SyntheticSections.h

@@ -19,6 +19,7 @@
 
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/Hashing.h"
+#include "llvm/ADT/Optional.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/MC/StringTableBuilder.h"
 #include "llvm/Support/MathExtras.h"
@@ -600,6 +601,27 @@ class WordLiteralSection final : public SyntheticSection {
   std::unordered_map<uint32_t, uint64_t> literal4Map;
 };
 
+class ObjCImageInfoSection final : public SyntheticSection {
+public:
+  ObjCImageInfoSection();
+  bool isNeeded() const override { return !files.empty(); }
+  uint64_t getSize() const override { return 8; }
+  void addFile(const InputFile *file) {
+    assert(!file->objCImageInfo.empty());
+    files.push_back(file);
+  }
+  void finalizeContents();
+  void writeTo(uint8_t *buf) const override;
+
+private:
+  struct ImageInfo {
+    uint8_t swiftVersion = 0;
+    bool hasCategoryClassProperties = false;
+  } info;
+  static ImageInfo parseImageInfo(const InputFile *);
+  std::vector<const InputFile *> files; // files with image info
+};
+
 struct InStruct {
   const uint8_t *bufferStart = nullptr;
   MachHeaderSection *header = nullptr;
@@ -616,6 +638,7 @@ struct InStruct {
   StubsSection *stubs = nullptr;
   StubHelperSection *stubHelper = nullptr;
   UnwindInfoSection *unwindInfo = nullptr;
+  ObjCImageInfoSection *objCImageInfo = nullptr;
   ConcatInputSection *imageLoaderCache = nullptr;
 };
 

lld/MachO/Writer.cpp

@@ -1164,6 +1164,10 @@ template <class LP> void Writer::run() {
 
   if (in.stubHelper->isNeeded())
     in.stubHelper->setup();
+
+  if (in.objCImageInfo->isNeeded())
+    in.objCImageInfo->finalizeContents();
+
   // At this point, we should know exactly which output sections are needed,
   // courtesy of scanSymbols() and scanRelocations().
   createOutputSections<LP>();
@@ -1210,6 +1214,7 @@ void macho::createSyntheticSections() {
   in.stubs = make<StubsSection>();
   in.stubHelper = make<StubHelperSection>();
   in.unwindInfo = makeUnwindInfoSection();
+  in.objCImageInfo = make<ObjCImageInfoSection>();
 
   // This section contains space for just a single word, and will be used by
   // dyld to cache an address to the image loader it uses.

lld/test/MachO/builtin-rename.s

@@ -37,7 +37,6 @@
 # NDATA-DAG: __DATA,__objc_catlist __DATA__objc_catlist
 # NDATA-DAG: __DATA,__objc_nlcatlist __DATA__objc_nlcatlist
 # NDATA-DAG: __DATA,__objc_protolist __DATA__objc_protolist
-# NDATA-DAG: __DATA,__objc_imageinfo __DATA__objc_imageinfo
 # NDATA-DAG: __DATA,__nl_symbol_ptr __IMPORT__pointers
 
 # YDATA-DAG: __DATA_CONST,__auth_got __DATA__auth_got
@@ -52,7 +51,6 @@
 # YDATA-DAG: __DATA_CONST,__objc_catlist __DATA__objc_catlist
 # YDATA-DAG: __DATA_CONST,__objc_nlcatlist __DATA__objc_nlcatlist
 # YDATA-DAG: __DATA_CONST,__objc_protolist __DATA__objc_protolist
-# YDATA-DAG: __DATA_CONST,__objc_imageinfo __DATA__objc_imageinfo
 # YDATA-DAG: __DATA_CONST,__nl_symbol_ptr __IMPORT__pointers
 
 ## LLD doesn't support defining symbols in synthetic sections, so we test them
@@ -133,10 +131,14 @@ __DATA__objc_nlcatlist:
 __DATA__objc_protolist:
   .space 8
 
-.section __DATA,__objc_imageinfo
-.global __DATA__objc_imageinfo
-__DATA__objc_imageinfo:
-  .space 8
+## __objc_imageinfo should get moved under __DATA_CONST as well, but symbols
+## within __objc_imageinfo get dropped during link, so we are cannot test this
+## case using the output of `llvm-objdump --syms`. TODO: rewrite test to use
+## `llvm-readobj --section-headers`, which will avoid this issue.
+# .section __DATA,__objc_imageinfo
+# .global __DATA__objc_imageinfo
+# __DATA__objc_imageinfo:
+#   .space 8
 
 .section __IMPORT,__pointers,non_lazy_symbol_pointers
 .global __IMPORT__pointers