[lld/mac] Always reference dyld_stub_binder when linked with libSystem

[thakis]

Summary

lld currently only references dyld_stub_binder when it's needed.
ld64 always references it when libSystem is linked.
Match ld64.

The (somewhat lame) motivation is that nm on a binary without any
export writes a "no symbols" warning to stderr, and this change makes
it so that every binary in practice has at least a reference to
dyld_stub_binder, which suppresses that.

Every "real" output file will reference dyld_stub_binder, so most
of the time this shouldn't make much of a difference. And if you
really don't want to have this reference for whatever reason, you
can stop passing -lSystem, like you have to for ld64 anyways.

(After linking any dylib, we dump the exported list of symbols to
a txt file with nm and only relink downstream deps if that txt
file changes. A nicer fix is to make lld optionally write .tbd files
with the public interface of a linked dylib and use that instead,
but for now the txt files are what we do.)

[int3]

lld/MachO/Driver.cpp

1049

can we make it clear that this isn't needed for correctness (as far as we know)?

[thakis]

Thanks!

lld/MachO/Driver.cpp

   // We always create a cStringSection, regardless of whether dedupLiterals is
   // true. If it isn't, we simply create a non-deduplicating CStringSection.
   // Either way, we must unconditionally finalize it here.
   in.cStringSection->finalizeContents();
   if (in.wordLiteralSection)
     in.wordLiteralSection->finalizeContents();
 }
 
+static void referenceStubBinder() {
+  bool needsStubHelper = config->outputType == MH_DYLIB ||
+                         config->outputType == MH_EXECUTE ||
+                         config->outputType == MH_BUNDLE;
+  if (!needsStubHelper || !symtab->find("dyld_stub_binder"))
+    return;
+
+  // dyld_stub_binder is in libSystem.dylib and used by dyld to resolve lazy
+  // bindings. This code here adds a opportunistic reference to it if the
+  // symbol happens to exist, which is always when libSystem.dylib is linked
+  // in.
+  // StubHelperSection::setup() adds a reference and errors out if the symbol
+  // isn't exist in case the symbol is actually needed.
+  symtab->addUndefined("dyld_stub_binder", /*file=*/nullptr, /*isWeak=*/false);
+}
+
 bool macho::link(ArrayRef<const char *> argsArr, bool canExitEarly,
                  raw_ostream &stdoutOS, raw_ostream &stderrOS) {
   lld::stdoutOS = &stdoutOS;
   lld::stderrOS = &stderrOS;
 
   errorHandler().cleanupCallback = []() { freeArena(); };
 
   errorHandler().logName = args::getFilenameWithoutExe(argsArr[0]);

     // patterns need not match.
     for (const CachedHashStringRef &cachedName :
          config->exportedSymbols.literals) {
       if (const Symbol *sym = symtab->find(cachedName))
         if (const auto *undefined = dyn_cast<Undefined>(sym))
           treatUndefinedSymbol(*undefined, "-exported_symbol(s_list)");
     }
 
+    referenceStubBinder();
+
     // FIXME: should terminate the link early based on errors encountered so
     // far?
 
     createSyntheticSections();
     createSyntheticSymbols();
 
     if (!config->exportedSymbols.empty()) {
       for (Symbol *sym : symtab->getSymbols()) {

lld/test/MachO/dyld-stub-binder.s

 # REQUIRES: aarch64
 # RUN: rm -rf %t; split-file %s %t
 # RUN: llvm-mc -filetype=obj -triple=arm64-apple-darwin %t/foo.s -o %t/foo.o
 # RUN: llvm-mc -filetype=obj -triple=arm64-apple-darwin %t/bar.s -o %t/bar.o
 # RUN: llvm-mc -filetype=obj -triple=arm64-apple-darwin %t/test.s -o %t/test.o
 
 ## Dylibs that don't do lazy dynamic calls don't need dyld_stub_binder.
 # RUN: %lld -arch arm64 -dylib %t/foo.o -o %t/libfoo.dylib
 # RUN: llvm-nm -m %t/libfoo.dylib | FileCheck --check-prefix=NOSTUB %s
 
+## Binaries that don't do lazy dynamic calls but are linked against
+## libSystem.dylib get a reference to dyld_stub_binder even if it's
+## not needed.
+# RUN: %lld -arch arm64 -lSystem -dylib %t/foo.o -o %t/libfoo.dylib
+# RUN: llvm-nm -m %t/libfoo.dylib | FileCheck --check-prefix=STUB %s
+
+
 ## Dylibs that do lazy dynamic calls do need dyld_stub_binder.
 # RUN: not %lld -arch arm64 -dylib %t/bar.o %t/libfoo.dylib \
 # RUN:     -o %t/libbar.dylib 2>&1 | FileCheck --check-prefix=MISSINGSTUB %s
 # RUN: %lld -arch arm64 -lSystem -dylib %t/bar.o  %t/libfoo.dylib \
 # RUN:     -o %t/libbar.dylib
 # RUN: llvm-nm -m %t/libbar.dylib | FileCheck --check-prefix=STUB %s
 
 ## As do executables.

lld/test/MachO/lto-internalize.ll

 ;; internalized.
 ; CHECK: define internal void @baz()
 
 ; Check foo and bar are not emitted to the .symtab
 ; SYMTAB-LABEL: SYMBOL TABLE:
 ; SYMTAB-NEXT:  g     F __TEXT,__text _main
 ; SYMTAB-NEXT:  g     F __TEXT,__text _used_in_regular_obj
 ; SYMTAB-NEXT:  g     F __TEXT,__text __mh_execute_header
+; SYMTAB-NEXT:          *UND* dyld_stub_binder
 ; SYMTAB-EMPTY:
 
 ;--- test.s
 target triple = "x86_64-apple-macosx10.15.0"
 target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 
 declare void @baz()
 

lld/test/MachO/stabs.s

 # CHECK-DAG:  (     SECT EXT) [[#%.2d,TEXT_ID + 1]]      0000   [[#BAZ2]]          '_baz2'
 # CHECK-DAG:  (     SECT EXT) [[#%.2d,TEXT_ID + 1]]      0000   [[#QUX]]           '_qux'
 # CHECK-DAG:  (     SECT EXT) [[#%.2d,TEXT_ID + 1]]      0000   [[#QUUX]]          '_quux'
 # CHECK-DAG:  (     SECT EXT) [[#%.2d,DATA_ID + 1]]      0000   [[#GLOB]]          '_global_var'
 # CHECK-DAG:  (     SECT EXT) [[#%.2d,COMM_ID + 1]]      0000   [[#ZERO]]          '_zero'
 # CHECK-DAG:  (     SECT EXT) [[#%.2d,MORE_TEXT_ID + 1]] 0000   [[#FUN]]           '_fun'
 # CHECK-DAG:  (     SECT EXT) [[#%.2d,TEXT_ID + 1]]      0000   {{[0-9a-f]+}}      '_no_debug'
 # CHECK-DAG:  (       {{.*}}) {{[0-9]+}}                 0010   {{[0-9a-f]+}}      '__mh_execute_header'
+# CHECK-DAG:  (       {{.*}}) {{[0-9]+}}                 0100   0000000000000000   'dyld_stub_binder'
 # CHECK-EMPTY:
 
 ## Check that we don't attempt to emit rebase opcodes for the debug sections
 ## when building a PIE (since we have filtered the sections out).
 # RUN: %lld -lSystem -pie %t/test.o %t/foo.a %t/no-debug.o -o %t/test
 # RUN: llvm-objdump --macho --rebase %t/test | FileCheck %s --check-prefix=PIE
 # PIE:       Rebase table:
 # PIE-NEXT:  segment  section            address     type