[LLD][COFF] Allow overriding EC alias symbols with lazy archive symbo…

…ls (llvm#113283) On ARM64EC, external function calls emit a pair of weak-dependency aliases: `func` to `#func` and `#func` to the `func` guess exit thunk (instead of a single undefined `func` symbol, which would be emitted on other targets). Allow such aliases to be overridden by lazy archive symbols, just as we would for undefined symbols.
ROCm · Oct 23, 2024 · 9b88792 · 9b88792
1 parent 2c5208a
commit 9b88792
Show file tree

Hide file tree

Showing 6 changed files with 132 additions and 12 deletions.
diff --git a/lld/COFF/InputFiles.cpp b/lld/COFF/InputFiles.cpp
@@ -455,11 +455,35 @@ void ObjFile::initializeSymbols() {
     COFFSymbolRef coffSym = check(coffObj->getSymbol(i));
     bool prevailingComdat;
     if (coffSym.isUndefined()) {
-      symbols[i] = createUndefined(coffSym);
+      symbols[i] = createUndefined(coffSym, false);
     } else if (coffSym.isWeakExternal()) {
-      symbols[i] = createUndefined(coffSym);
-      weakAliases.emplace_back(symbols[i],
-                               coffSym.getAux<coff_aux_weak_external>());
+      auto aux = coffSym.getAux<coff_aux_weak_external>();
+      bool overrideLazy = true;
+
+      // On ARM64EC, external function calls emit a pair of weak-dependency
+      // aliases: func to #func and #func to the func guess exit thunk
+      // (instead of a single undefined func symbol, which would be emitted on
+      // other targets). Allow such aliases to be overridden by lazy archive
+      // symbols, just as we would for undefined symbols.
+      if (isArm64EC(getMachineType()) &&
+          aux->Characteristics == IMAGE_WEAK_EXTERN_ANTI_DEPENDENCY) {
+        COFFSymbolRef targetSym = check(coffObj->getSymbol(aux->TagIndex));
+        if (!targetSym.isAnyUndefined()) {
+          // If the target is defined, it may be either a guess exit thunk or
+          // the actual implementation. If it's the latter, consider the alias
+          // to be part of the implementation and override potential lazy
+          // archive symbols.
+          StringRef targetName = check(coffObj->getSymbolName(targetSym));
+          StringRef name = check(coffObj->getSymbolName(coffSym));
+          std::optional<std::string> mangledName =
+              getArm64ECMangledFunctionName(name);
+          overrideLazy = mangledName == targetName;
+        } else {
+          overrideLazy = false;
+        }
+      }
+      symbols[i] = createUndefined(coffSym, overrideLazy);
+      weakAliases.emplace_back(symbols[i], aux);
     } else if (std::optional<Symbol *> optSym =
                    createDefined(coffSym, comdatDefs, prevailingComdat)) {
       symbols[i] = *optSym;
@@ -508,9 +532,9 @@ void ObjFile::initializeSymbols() {
   decltype(sparseChunks)().swap(sparseChunks);
 }
 
-Symbol *ObjFile::createUndefined(COFFSymbolRef sym) {
+Symbol *ObjFile::createUndefined(COFFSymbolRef sym, bool overrideLazy) {
   StringRef name = check(coffObj->getSymbolName(sym));
-  return ctx.symtab.addUndefined(name, this, sym.isWeakExternal());
+  return ctx.symtab.addUndefined(name, this, overrideLazy);
 }
 
 static const coff_aux_section_definition *findSectionDef(COFFObjectFile *obj,

diff --git a/lld/COFF/InputFiles.h b/lld/COFF/InputFiles.h
@@ -272,7 +272,7 @@ class ObjFile : public InputFile {
                     &comdatDefs,
                 bool &prevailingComdat);
   Symbol *createRegular(COFFSymbolRef sym);
-  Symbol *createUndefined(COFFSymbolRef sym);
+  Symbol *createUndefined(COFFSymbolRef sym, bool overrideLazy);
 
   std::unique_ptr<COFFObjectFile> coffObj;
 

diff --git a/lld/COFF/SymbolTable.cpp b/lld/COFF/SymbolTable.cpp
@@ -620,9 +620,9 @@ void SymbolTable::initializeECThunks() {
 }
 
 Symbol *SymbolTable::addUndefined(StringRef name, InputFile *f,
-                                  bool isWeakAlias) {
+                                  bool overrideLazy) {
   auto [s, wasInserted] = insert(name, f);
-  if (wasInserted || (s->isLazy() && isWeakAlias)) {
+  if (wasInserted || (s->isLazy() && overrideLazy)) {
     replaceSymbol<Undefined>(s, name);
     return s;
   }
@@ -639,7 +639,8 @@ void SymbolTable::addLazyArchive(ArchiveFile *f, const Archive::Symbol &sym) {
     return;
   }
   auto *u = dyn_cast<Undefined>(s);
-  if (!u || u->weakAlias || s->pendingArchiveLoad)
+  if (!u || (u->weakAlias && !u->isECAlias(ctx.config.machine)) ||
+      s->pendingArchiveLoad)
     return;
   s->pendingArchiveLoad = true;
   f->addMember(sym);
@@ -653,7 +654,8 @@ void SymbolTable::addLazyObject(InputFile *f, StringRef n) {
     return;
   }
   auto *u = dyn_cast<Undefined>(s);
-  if (!u || u->weakAlias || s->pendingArchiveLoad)
+  if (!u || (u->weakAlias && !u->isECAlias(ctx.config.machine)) ||
+      s->pendingArchiveLoad)
     return;
   s->pendingArchiveLoad = true;
   f->lazy = false;

diff --git a/lld/COFF/SymbolTable.h b/lld/COFF/SymbolTable.h
@@ -91,7 +91,7 @@ class SymbolTable {
   Symbol *addSynthetic(StringRef n, Chunk *c);
   Symbol *addAbsolute(StringRef n, uint64_t va);
 
-  Symbol *addUndefined(StringRef name, InputFile *f, bool isWeakAlias);
+  Symbol *addUndefined(StringRef name, InputFile *f, bool overrideLazy);
   void addLazyArchive(ArchiveFile *f, const Archive::Symbol &sym);
   void addLazyObject(InputFile *f, StringRef n);
   void addLazyDLLSymbol(DLLFile *f, DLLFile::Symbol *sym, StringRef n);

diff --git a/lld/COFF/Symbols.h b/lld/COFF/Symbols.h
@@ -353,6 +353,10 @@ class Undefined : public Symbol {
     isAntiDep = antiDep;
   }
 
+  bool isECAlias(MachineTypes machine) const {
+    return weakAlias && isAntiDep && isArm64EC(machine);
+  }
+
   // If this symbol is external weak, replace this object with aliased symbol.
   bool resolveWeakAlias();
 };

diff --git a/lld/test/COFF/arm64ec-lib.test b/lld/test/COFF/arm64ec-lib.test
@@ -7,10 +7,16 @@ RUN: llvm-mc -filetype=obj -triple=aarch64-windows nsymref.s -o nsymref-aarch64.
 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 nsym.s -o nsym-aarch64.obj
+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=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
 RUN: llvm-lib -machine:amd64 -out:sym-x86_64.lib sym-x86_64.obj
+RUN: llvm-lib -machine:arm64ec -out:func.lib func.obj
+RUN: llvm-lib -machine:arm64ec -out:func-x86_64.lib func-x86_64.obj
 
 Verify that a symbol can be referenced from ECSYMBOLS.
 RUN: lld-link -machine:arm64ec -dll -noentry -out:test.dll symref-arm64ec.obj sym-arm64ec.lib loadconfig-arm64ec.obj
@@ -26,6 +32,54 @@ RUN: not lld-link -machine:arm64ec -dll -noentry -out:test-err.dll nsymref-arm64
 RUN:              FileCheck --check-prefix=ERR %s
 ERR: error: undefined symbol: nsym
 
+Verify that a library symbol can be referenced, even if its name conflicts with an anti-dependency alias.
+RUN: lld-link -machine:arm64ec -dll -noentry -out:ref-alias-1.dll ref-alias.obj func.lib loadconfig-arm64ec.obj
+RUN: llvm-objdump -d ref-alias-1.dll | FileCheck -check-prefix=DISASM %s
+DISASM:      0000000180001000 <.text>:
+DISASM-NEXT: 180001000: d65f03c0     ret
+DISASM-EMPTY:
+
+RUN: llvm-readobj --hex-dump=.test ref-alias-1.dll | FileCheck -check-prefix=TESTSEC %s
+TESTSEC: 0x180004000 00100000
+
+The same test, but with a different input order.
+RUN: lld-link -machine:arm64ec -dll -noentry -out:ref-alias-2.dll func.lib ref-alias.obj loadconfig-arm64ec.obj
+RUN: llvm-objdump -d ref-alias-2.dll | FileCheck -check-prefix=DISASM %s
+RUN: llvm-readobj --hex-dump=.test ref-alias-2.dll | FileCheck -check-prefix=TESTSEC %s
+
+Verify that when an anti-dependency to a guess exit thunk is present, it is overridden by an archive symbol.
+RUN: lld-link -machine:arm64ec -dll -noentry -out:ref-thunk-1.dll ref-thunk.obj func.lib loadconfig-arm64ec.obj
+RUN: llvm-objdump -d ref-thunk-1.dll | FileCheck -check-prefix=DISASM %s
+RUN: llvm-readobj --hex-dump=.test ref-thunk-1.dll | FileCheck -check-prefix=TESTSEC %s
+
+The same test, but with a different input order.
+RUN: lld-link -machine:arm64ec -dll -noentry -out:ref-thunk-2.dll func.lib ref-thunk.obj loadconfig-arm64ec.obj
+RUN: llvm-objdump -d ref-thunk-2.dll | FileCheck -check-prefix=DISASM %s
+RUN: llvm-readobj --hex-dump=.test ref-thunk-2.dll | FileCheck -check-prefix=TESTSEC %s
+
+Test linking against an x86_64 library (which uses a demangled function name).
+RUN: lld-link -machine:arm64ec -dll -noentry -out:ref-x86-1.dll ref-thunk.obj func-x86_64.lib loadconfig-arm64ec.obj
+RUN: llvm-objdump -d ref-x86-1.dll | FileCheck -check-prefix=DISASM-X86 %s
+RUN: llvm-readobj --hex-dump=.test ref-x86-1.dll | FileCheck -check-prefix=TESTSEC %s
+
+DISASM-X86:      0000000180001000 <.text>:
+DISASM-X86-NEXT: 180001000: c3                           retq
+
+The same test, but with a different input order.
+RUN: lld-link -machine:arm64ec -dll -noentry -out:ref-x86-2.dll func-x86_64.lib ref-thunk.obj loadconfig-arm64ec.obj
+RUN: llvm-objdump -d ref-x86-2.dll | FileCheck -check-prefix=DISASM-X86 %s
+RUN: llvm-readobj --hex-dump=.test ref-x86-2.dll | FileCheck -check-prefix=TESTSEC %s
+
+Check that an alias to the implementation takes precedence over an archive symbol.
+RUN: lld-link -machine:arm64ec -dll -noentry -out:ref-x86-2.dll func-x86_64.lib func.obj ref-thunk.obj loadconfig-arm64ec.obj
+RUN: llvm-objdump -d ref-x86-2.dll | FileCheck -check-prefix=DISASM %s
+RUN: llvm-readobj --hex-dump=.test ref-x86-2.dll | FileCheck -check-prefix=TESTSEC %s
+
+A similar test using -start-lib for linking.
+RUN: lld-link -machine:arm64ec -dll -noentry -out:start-lib-1.dll ref-thunk.obj -start-lib func.obj -end-lib loadconfig-arm64ec.obj
+RUN: llvm-objdump -d start-lib-1.dll | FileCheck -check-prefix=DISASM %s
+RUN: llvm-readobj --hex-dump=.test start-lib-1.dll | FileCheck -check-prefix=TESTSEC %s
+
 #--- symref.s
     .data
     .rva sym
@@ -45,3 +99,39 @@ sym:
      .globl nsym
 nsym:
      .word 0
+
+#--- ref-alias.s
+    .weak_anti_dep func
+.set func,"#func"
+
+    .section .test, "r"
+    .rva func
+
+#--- ref-thunk.s
+    .weak_anti_dep func
+.set func, "#func"
+    .weak_anti_dep "#func"
+.set "#func", thunksym
+
+    .section .test, "r"
+    .rva func
+
+    .section .thnk,"xr",discard,thunksym
+thunksym:
+    mov w0, #2
+    ret
+
+#--- func.s
+    .text
+    .globl "#func"
+"#func":
+    ret
+
+    .weak_anti_dep func
+.set func,"#func"
+
+#--- func-x86_64.s
+    .text
+    .globl func
+func:
+    ret