[LLD][COFF] Introduce hybrid symbol table for EC input files on ARM64X (#119294)

lld/COFF/COFFLinkerContext.h

@@ -32,6 +32,27 @@ public:
   SymbolTable symtab;
   COFFOptTable optTable;
 
+  // A hybrid ARM64EC symbol table on ARM64X target.
+  std::optional<SymbolTable> hybridSymtab;
+
+  // Pointer to the ARM64EC symbol table: either symtab for an ARM64EC target or
+  // hybridSymtab for an ARM64X target.
+  SymbolTable *symtabEC = nullptr;
+
+  // Returns the appropriate symbol table for the specified machine type.
+  SymbolTable &getSymtab(llvm::COFF::MachineTypes machine) {
+    if (hybridSymtab && (machine == ARM64EC || machine == AMD64))
+      return *hybridSymtab;
+    return symtab;
+  }
+
+  // Invoke the specified callback for each symbol table.
+  void forEachSymtab(std::function<void(SymbolTable &symtab)> f) {
+    f(symtab);
+    if (hybridSymtab)
+      f(*hybridSymtab);
+  }
+
   std::vector<ObjFile *> objFileInstances;
   std::map<std::string, PDBInputFile *> pdbInputFileInstances;
   std::vector<ImportFile *> importFileInstances;

lld/COFF/Driver.cpp

@@ -595,7 +595,17 @@ void LinkerDriver::setMachine(MachineTypes machine) {
   assert(machine != IMAGE_FILE_MACHINE_UNKNOWN);
 
   ctx.config.machine = machine;
-  ctx.symtab.machine = machine;
+
+  if (machine != ARM64X) {
+    ctx.symtab.machine = machine;
+    if (machine == ARM64EC)
+      ctx.symtabEC = &ctx.symtab;
+  } else {
+    ctx.symtab.machine = ARM64;
+    ctx.hybridSymtab.emplace(ctx, ARM64EC);
+    ctx.symtabEC = &*ctx.hybridSymtab;
+  }
+
   addWinSysRootLibSearchPaths();
 }
 
@@ -2518,54 +2528,56 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
   if (config->imageBase == uint64_t(-1))
     config->imageBase = getDefaultImageBase();
 
-  ctx.symtab.addSynthetic(mangle("__ImageBase"), nullptr);
-  if (config->machine == I386) {
-    ctx.symtab.addAbsolute("___safe_se_handler_table", 0);
-    ctx.symtab.addAbsolute("___safe_se_handler_count", 0);
-  }
+  ctx.forEachSymtab([&](SymbolTable &symtab) {
+    symtab.addSynthetic(mangle("__ImageBase"), nullptr);
+    if (symtab.machine == I386) {
+      symtab.addAbsolute("___safe_se_handler_table", 0);
+      symtab.addAbsolute("___safe_se_handler_count", 0);
+    }
 
-  ctx.symtab.addAbsolute(mangle("__guard_fids_count"), 0);
-  ctx.symtab.addAbsolute(mangle("__guard_fids_table"), 0);
-  ctx.symtab.addAbsolute(mangle("__guard_flags"), 0);
-  ctx.symtab.addAbsolute(mangle("__guard_iat_count"), 0);
-  ctx.symtab.addAbsolute(mangle("__guard_iat_table"), 0);
-  ctx.symtab.addAbsolute(mangle("__guard_longjmp_count"), 0);
-  ctx.symtab.addAbsolute(mangle("__guard_longjmp_table"), 0);
-  // Needed for MSVC 2017 15.5 CRT.
-  ctx.symtab.addAbsolute(mangle("__enclave_config"), 0);
-  // Needed for MSVC 2019 16.8 CRT.
-  ctx.symtab.addAbsolute(mangle("__guard_eh_cont_count"), 0);
-  ctx.symtab.addAbsolute(mangle("__guard_eh_cont_table"), 0);
+    symtab.addAbsolute(mangle("__guard_fids_count"), 0);
+    symtab.addAbsolute(mangle("__guard_fids_table"), 0);
+    symtab.addAbsolute(mangle("__guard_flags"), 0);
+    symtab.addAbsolute(mangle("__guard_iat_count"), 0);
+    symtab.addAbsolute(mangle("__guard_iat_table"), 0);
+    symtab.addAbsolute(mangle("__guard_longjmp_count"), 0);
+    symtab.addAbsolute(mangle("__guard_longjmp_table"), 0);
+    // Needed for MSVC 2017 15.5 CRT.
+    symtab.addAbsolute(mangle("__enclave_config"), 0);
+    // Needed for MSVC 2019 16.8 CRT.
+    symtab.addAbsolute(mangle("__guard_eh_cont_count"), 0);
+    symtab.addAbsolute(mangle("__guard_eh_cont_table"), 0);
 
-  if (isArm64EC(config->machine)) {
-    ctx.symtab.addAbsolute("__arm64x_extra_rfe_table", 0);
-    ctx.symtab.addAbsolute("__arm64x_extra_rfe_table_size", 0);
-    ctx.symtab.addAbsolute("__arm64x_redirection_metadata", 0);
-    ctx.symtab.addAbsolute("__arm64x_redirection_metadata_count", 0);
-    ctx.symtab.addAbsolute("__hybrid_auxiliary_delayload_iat_copy", 0);
-    ctx.symtab.addAbsolute("__hybrid_auxiliary_delayload_iat", 0);
-    ctx.symtab.addAbsolute("__hybrid_auxiliary_iat", 0);
-    ctx.symtab.addAbsolute("__hybrid_auxiliary_iat_copy", 0);
-    ctx.symtab.addAbsolute("__hybrid_code_map", 0);
-    ctx.symtab.addAbsolute("__hybrid_code_map_count", 0);
-    ctx.symtab.addAbsolute("__hybrid_image_info_bitfield", 0);
-    ctx.symtab.addAbsolute("__x64_code_ranges_to_entry_points", 0);
-    ctx.symtab.addAbsolute("__x64_code_ranges_to_entry_points_count", 0);
-    ctx.symtab.addSynthetic("__guard_check_icall_a64n_fptr", nullptr);
-    ctx.symtab.addSynthetic("__arm64x_native_entrypoint", nullptr);
-  }
+    if (isArm64EC(ctx.config.machine)) {
+      symtab.addAbsolute("__arm64x_extra_rfe_table", 0);
+      symtab.addAbsolute("__arm64x_extra_rfe_table_size", 0);
+      symtab.addAbsolute("__arm64x_redirection_metadata", 0);
+      symtab.addAbsolute("__arm64x_redirection_metadata_count", 0);
+      symtab.addAbsolute("__hybrid_auxiliary_delayload_iat_copy", 0);
+      symtab.addAbsolute("__hybrid_auxiliary_delayload_iat", 0);
+      symtab.addAbsolute("__hybrid_auxiliary_iat", 0);
+      symtab.addAbsolute("__hybrid_auxiliary_iat_copy", 0);
+      symtab.addAbsolute("__hybrid_code_map", 0);
+      symtab.addAbsolute("__hybrid_code_map_count", 0);
+      symtab.addAbsolute("__hybrid_image_info_bitfield", 0);
+      symtab.addAbsolute("__x64_code_ranges_to_entry_points", 0);
+      symtab.addAbsolute("__x64_code_ranges_to_entry_points_count", 0);
+      symtab.addSynthetic("__guard_check_icall_a64n_fptr", nullptr);
+      symtab.addSynthetic("__arm64x_native_entrypoint", nullptr);
+    }
 
-  if (config->pseudoRelocs) {
-    ctx.symtab.addAbsolute(mangle("__RUNTIME_PSEUDO_RELOC_LIST__"), 0);
-    ctx.symtab.addAbsolute(mangle("__RUNTIME_PSEUDO_RELOC_LIST_END__"), 0);
-  }
-  if (config->mingw) {
-    ctx.symtab.addAbsolute(mangle("__CTOR_LIST__"), 0);
-    ctx.symtab.addAbsolute(mangle("__DTOR_LIST__"), 0);
-  }
-  if (config->debug || config->buildIDHash != BuildIDHash::None)
-    if (ctx.symtab.findUnderscore("__buildid"))
-      ctx.symtab.addUndefined(mangle("__buildid"));
+    if (config->pseudoRelocs) {
+      symtab.addAbsolute(mangle("__RUNTIME_PSEUDO_RELOC_LIST__"), 0);
+      symtab.addAbsolute(mangle("__RUNTIME_PSEUDO_RELOC_LIST_END__"), 0);
+    }
+    if (config->mingw) {
+      symtab.addAbsolute(mangle("__CTOR_LIST__"), 0);
+      symtab.addAbsolute(mangle("__DTOR_LIST__"), 0);
+    }
+    if (config->debug || config->buildIDHash != BuildIDHash::None)
+      if (symtab.findUnderscore("__buildid"))
+        symtab.addUndefined(mangle("__buildid"));
+  });
 
   // This code may add new undefined symbols to the link, which may enqueue more
   // symbol resolution tasks, so we need to continue executing tasks until we
@@ -2808,7 +2820,8 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
   if (auto *arg = args.getLastArg(OPT_print_symbol_order))
     config->printSymbolOrder = arg->getValue();
 
-  ctx.symtab.initializeECThunks();
+  if (ctx.symtabEC)
+    ctx.symtabEC->initializeECThunks();
 
   // Identify unreferenced COMDAT sections.
   if (config->doGC) {

lld/COFF/InputFiles.cpp

@@ -114,15 +114,15 @@ void ArchiveFile::parse() {
   file = CHECK(Archive::create(mb), this);
 
   // Try to read symbols from ECSYMBOLS section on ARM64EC.
-  if (isArm64EC(ctx.config.machine)) {
+  if (ctx.symtabEC) {
     iterator_range<Archive::symbol_iterator> symbols =
         CHECK(file->ec_symbols(), this);
     if (!symbols.empty()) {
       for (const Archive::Symbol &sym : symbols)
-        ctx.symtab.addLazyArchive(this, sym);
+        ctx.symtabEC->addLazyArchive(this, sym);
 
       // Read both EC and native symbols on ARM64X.
-      if (ctx.config.machine != ARM64X)
+      if (!ctx.hybridSymtab)
         return;
     }
   }
@@ -177,7 +177,8 @@ ObjFile *ObjFile::create(COFFLinkerContext &ctx, MemoryBufferRef m, bool lazy) {
     Fatal(ctx) << m.getBufferIdentifier() << " is not a COFF file";
 
   bin->release();
-  return make<ObjFile>(ctx.symtab, obj, lazy);
+  return make<ObjFile>(ctx.getSymtab(MachineTypes(obj->getMachine())), obj,
+                       lazy);
 }
 
 void ObjFile::parseLazy() {

lld/COFF/SymbolTable.h

@@ -47,7 +47,9 @@ class Symbol;
 // There is one add* function per symbol type.
 class SymbolTable {
 public:
-  SymbolTable(COFFLinkerContext &c) : ctx(c) {}
+  SymbolTable(COFFLinkerContext &c,
+              llvm::COFF::MachineTypes machine = IMAGE_FILE_MACHINE_UNKNOWN)
+      : ctx(c), machine(machine) {}
 
   void addFile(InputFile *file);
 
@@ -120,7 +122,7 @@ public:
                        uint32_t newSectionOffset = 0);
 
   COFFLinkerContext &ctx;
-  llvm::COFF::MachineTypes machine = IMAGE_FILE_MACHINE_UNKNOWN;
+  llvm::COFF::MachineTypes machine;
 
   bool isEC() const { return machine == ARM64EC; }
 

lld/test/COFF/arm64ec-codemap.test

@@ -9,6 +9,7 @@ RUN: llvm-mc -filetype=obj -triple=arm64ec-windows data-sec2.s -o data-sec2.obj
 RUN: llvm-mc -filetype=obj -triple=arm64ec-windows empty-sec.s -o arm64ec-empty-sec.obj
 RUN: llvm-mc -filetype=obj -triple=x86_64-windows x86_64-func-sym.s -o x86_64-func-sym.obj
 RUN: llvm-mc -filetype=obj -triple=x86_64-windows empty-sec.s -o x86_64-empty-sec.obj
+RUN: llvm-mc -filetype=obj -triple=aarch64-windows %S/Inputs/loadconfig-arm64ec.s -o loadconfig-arm64.obj
 RUN: llvm-mc -filetype=obj -triple=arm64ec-windows %S/Inputs/loadconfig-arm64ec.s -o loadconfig-arm64ec.obj
 
 Link ARM64EC DLL and verify that the code is arranged as expected.
@@ -51,7 +52,7 @@ RUN: llvm-readobj --coff-load-config test2.dll | FileCheck -check-prefix=CODEMAP
 RUN: llvm-objdump -d test2.dll | FileCheck -check-prefix=DISASM %s
 
 RUN: lld-link -out:testx.dll -machine:arm64x arm64-func-sym.obj arm64ec-func-sym.obj \
-RUN:          x86_64-func-sym.obj loadconfig-arm64ec.obj -dll -noentry
+RUN:          x86_64-func-sym.obj loadconfig-arm64.obj loadconfig-arm64ec.obj -dll -noentry
 
 Adding empty chunks does not affect code map ranges.
 

lld/test/COFF/arm64ec-entry-thunk.s

@@ -27,6 +27,7 @@ thunk:
     .rva func
 
 // RUN: llvm-mc -filetype=obj -triple=arm64ec-windows %S/Inputs/loadconfig-arm64ec.s -o loadcfg.obj
+// RUN: llvm-mc -filetype=obj -triple=aarch64-windows %S/Inputs/loadconfig-arm64ec.s -o native-loadcfg.obj
 // RUN: llvm-mc -filetype=obj -triple=arm64ec-windows test-simple.s -o test-simple.obj
 // RUN: lld-link -machine:arm64ec -dll -noentry -out:out-simple.dll loadcfg.obj test-simple.obj
 // RUN: llvm-objdump -d out-simple.dll | FileCheck --check-prefix=DISASM %s
@@ -43,7 +44,7 @@ thunk:
 // RUN: llvm-readobj --sections out-simple.dll | FileCheck --check-prefix=HYBMP %s
 // HYBMP-NOT: .hybmp
 
-// RUN: lld-link -machine:arm64x -dll -noentry -out:out-simplex.dll loadcfg.obj test-simple.obj
+// RUN: lld-link -machine:arm64x -dll -noentry -out:out-simplex.dll native-loadcfg.obj loadcfg.obj test-simple.obj
 // RUN: llvm-objdump -d out-simplex.dll | FileCheck --check-prefix=DISASM %s
 
 #--- test-split-func.s

lld/test/COFF/arm64ec-lib.test

@@ -11,6 +11,7 @@ RUN: llvm-mc -filetype=obj -triple=arm64ec-windows ref-alias.s -o ref-alias.obj
 RUN: llvm-mc -filetype=obj -triple=arm64ec-windows ref-thunk.s -o ref-thunk.obj
 RUN: llvm-mc -filetype=obj -triple=arm64ec-windows func.s -o func.obj
 RUN: llvm-mc -filetype=obj -triple=x86_64-windows func-x86_64.s -o func-x86_64.obj
+RUN: llvm-mc -filetype=obj -triple=aarch64-windows %S/Inputs/loadconfig-arm64ec.s -o loadconfig-arm64.obj
 RUN: llvm-mc -filetype=obj -triple=arm64ec-windows %S/Inputs/loadconfig-arm64ec.s -o loadconfig-arm64ec.obj
 
 RUN: llvm-lib -machine:arm64ec -out:sym-arm64ec.lib sym-arm64ec.obj nsym-aarch64.obj
@@ -26,7 +27,8 @@ Verify that a symbol can be referenced from a regular archive map when ECSYMBOLS
 RUN: lld-link -machine:arm64ec -dll -noentry -out:test2.dll symref-arm64ec.obj sym-x86_64.lib loadconfig-arm64ec.obj
 
 Verify that both native and EC symbols can be referenced in a hybrid target.
-RUN: lld-link -machine:arm64x -dll -noentry -out:test3.dll symref-arm64ec.obj nsymref-aarch64.obj sym-arm64ec.lib loadconfig-arm64ec.obj
+RUN: lld-link -machine:arm64x -dll -noentry -out:test3.dll symref-arm64ec.obj nsymref-aarch64.obj sym-arm64ec.lib \
+RUN:          loadconfig-arm64.obj loadconfig-arm64ec.obj
 
 Ensure that an EC symbol is not resolved using a regular symbol map.
 RUN: not lld-link -machine:arm64ec -dll -noentry -out:test-err.dll nsymref-arm64ec.obj sym-arm64ec.lib loadconfig-arm64ec.obj 2>&1 |\

lld/test/COFF/arm64ec-range-thunks.s

@@ -5,6 +5,7 @@
 # RUN: llvm-mc -filetype=obj -triple=aarch64-windows native-funcs.s -o funcs-aarch64.obj
 # RUN: llvm-mc -filetype=obj -triple=x86_64-windows space.s -o space-x86_64.obj
 # RUN: llvm-mc -filetype=obj -triple=aarch64-windows space.s -o space-aarch64.obj
+# RUN: llvm-mc -filetype=obj -triple=aarch64-windows %S/Inputs/loadconfig-arm64ec.s -o loadconfig-arm64.obj
 # RUN: llvm-mc -filetype=obj -triple=arm64ec-windows %S/Inputs/loadconfig-arm64ec.s -o loadconfig-arm64ec.obj
 
 
@@ -59,8 +60,8 @@
 
 # A similar test using a hybrid binary and native placeholder chunks.
 
-# RUN: lld-link -machine:arm64x -noentry -dll funcs-arm64ec.obj space-aarch64.obj loadconfig-arm64ec.obj -out:testx.dll \
-# RUN:          -verbose 2>&1 | FileCheck -check-prefix=VERBOSE %s
+# RUN: lld-link -machine:arm64x -noentry -dll funcs-arm64ec.obj space-aarch64.obj loadconfig-arm64.obj loadconfig-arm64ec.obj \
+# RUN:          -out:testx.dll -verbose 2>&1 | FileCheck -check-prefix=VERBOSE %s
 # RUN: llvm-objdump -d testx.dll | FileCheck --check-prefix=DISASM %s
 
 # RUN: llvm-readobj --coff-load-config testx.dll | FileCheck --check-prefix=LOADCFGX %s
@@ -74,8 +75,8 @@
 
 # Test a hybrid ARM64X binary which requires range extension thunks for both native and EC relocations.
 
-# RUN: lld-link -machine:arm64x -noentry -dll funcs-arm64ec.obj funcs-aarch64.obj loadconfig-arm64ec.obj -out:testx2.dll \
-# RUN:          -verbose 2>&1 | FileCheck -check-prefix=VERBOSEX %s
+# RUN: lld-link -machine:arm64x -noentry -dll funcs-arm64ec.obj funcs-aarch64.obj loadconfig-arm64.obj loadconfig-arm64ec.obj \
+# RUN:           -out:testx2.dll -verbose 2>&1 | FileCheck -check-prefix=VERBOSEX %s
 # VERBOSEX: Added 5 thunks with margin {{.*}} in 1 passes
 
 # RUN: llvm-objdump -d testx2.dll | FileCheck --check-prefix=DISASMX %s

lld/test/COFF/arm64x-symtab.s

@@ -0,0 +1,38 @@
+// REQUIRES: aarch64, x86
+// RUN: split-file %s %t.dir && cd %t.dir
+
+// RUN: llvm-mc -filetype=obj -triple=aarch64-windows sym.s -o sym-aarch64.obj
+// RUN: llvm-mc -filetype=obj -triple=arm64ec-windows sym.s -o sym-arm64ec.obj
+// RUN: llvm-mc -filetype=obj -triple=x86_64-windows sym.s -o sym-x86_64.obj
+// RUN: llvm-mc -filetype=obj -triple=aarch64-windows symref.s -o symref-aarch64.obj
+// RUN: llvm-mc -filetype=obj -triple=arm64ec-windows symref.s -o symref-arm64ec.obj
+// RUN: llvm-lib -machine:arm64x -out:sym.lib sym-aarch64.obj sym-arm64ec.obj
+
+// Check that native object files can't reference EC symbols.
+
+// RUN: not lld-link -machine:arm64x -dll -noentry -out:err1.dll symref-aarch64.obj sym-arm64ec.obj \
+// RUN:              2>&1 | FileCheck --check-prefix=UNDEF %s
+// UNDEF:      lld-link: error: undefined symbol: sym
+// UNDEF-NEXT: >>> referenced by symref-aarch64.obj:(.data)
+
+// RUN: not lld-link -machine:arm64x -dll -noentry -out:err2.dll symref-aarch64.obj sym-x86_64.obj \
+// RUN:              2>&1 | FileCheck --check-prefix=UNDEF %s
+
+// Check that ARM64X target can have the same symbol names in both native and EC namespaces.
+
+// RUN: lld-link -machine:arm64x -dll -noentry -out:out.dll symref-aarch64.obj sym-aarch64.obj \
+// RUN:           symref-arm64ec.obj sym-x86_64.obj
+
+// Check that ARM64X target can reference both native and EC symbols from an archive.
+
+// RUN: lld-link -machine:arm64x -dll -noentry -out:out2.dll symref-aarch64.obj symref-arm64ec.obj sym.lib
+
+#--- symref.s
+    .data
+    .rva sym
+
+#--- sym.s
+     .data
+     .globl sym
+sym:
+     .word 0