Fix issues with GlobalMerge on Mach-O.

[jyknight]

As a side-effect of PR #101222, GlobalMerge started making transforms which are unsafe on Mach-O platforms.

Two issues, in particular, are fixed here:

1. We must never merge symbols in the __cfstring section, as the linker assumes each object in this section is only ever referenced directly, and that it can split the section as it likes.

   Previously, we avoided this problem because CFString literals are identified by private-linkage symbols. This patch adds a list of section-names with special behavior, to avoid merging under Mach-O.

2. When GlobalMerge code was originally written, it had to be careful about emitting symbol aliases, due to issues with Mach-O's subsection splitting in the linker with -dead_strip enabled. The underlying cause of this problem was fixed in 2016, via creation of the .alt_entry assembler directive, which allows a symbol to not also imply the start of a new subsection. GlobalMerge's workaround for that issue was never removed.

   In the meantime, Apple's new ld-prime linker was written, and has a bug in .alt_entry handling. Therefore, even though the original issue was fixed, we must continue to be careful not to emit any such symbol aliases. The existing workaround avoided it for InternalLinkage symbols, but after the above-mentioned PR, we also must avoid emitting aliases for PrivateLinkage symbols.

   I will file an Apple bug-report about this issue, so that it can be fixed in a future version of ld-prime. But, in the meantime, the workaround is sufficient for GlobalMerge, unless -global-merge-on-externals is enabled (which it is already not by default, on MachO platforms, due to the original issue).

Fixes #104625

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: James Y Knight (jyknight)

Changes

As a side-effect of PR #101222, GlobalMerge started making transforms which are unsafe on Mach-O platforms.

Two issues, in particular, are fixed here:

1. We must never merge symbols in the __cfstring section, as the linker assumes each object in this section is only ever referenced directly, and that it can split the section as it likes.

   Previously, we avoided this problem because CFString literals are identified by private-linkage symbols. This patch adds a list of section-names with special behavior, to avoid merging under Mach-O.

2. When GlobalMerge code was originally written, it had to be careful about emitting symbol aliases, due to issues with Mach-O's subsection splitting in the linker with -dead_strip enabled. The underlying cause of this problem was fixed in 2016, via creation of the .alt_entry assembler directive, which allows a symbol to not also imply the start of a new subsection. GlobalMerge's workaround for that issue was never removed.

   In the meantime, Apple's new ld-prime linker was written, and has a bug in .alt_entry handling. Therefore, even though the original issue was fixed, we must continue to be careful not to emit any such symbol aliases. The existing workaround avoided it for InternalLinkage symbols, but after the above-mentioned PR, we also must avoid emitting aliases for PrivateLinkage symbols.

   I will file an Apple bug-report about this issue, so that it can be fixed in a future version of ld-prime. But, in the meantime, the workaround is sufficient for GlobalMerge, unless -global-merge-on-externals is enabled (which it is already not by default, on MachO platforms, due to the original issue).

Fixes #104625

---

Full diff: https://github.com/llvm/llvm-project/pull/110046.diff

3 Files Affected:

• (modified) llvm/lib/CodeGen/GlobalMerge.cpp (+27-6)
• (added) llvm/test/Transforms/GlobalMerge/macho-sections.ll (+25)
• (added) llvm/test/Transforms/GlobalMerge/macho-symbols.ll (+28)

diff --git a/llvm/lib/CodeGen/GlobalMerge.cpp b/llvm/lib/CodeGen/GlobalMerge.cpp
index c31ba6b31ad9ac..2003b74a314bac 100644
--- a/llvm/lib/CodeGen/GlobalMerge.cpp
+++ b/llvm/lib/CodeGen/GlobalMerge.cpp
@@ -578,12 +578,18 @@ bool GlobalMergeImpl::doMerge(const SmallVectorImpl<GlobalVariable *> &Globals,
       Globals[k]->replaceAllUsesWith(GEP);
       Globals[k]->eraseFromParent();
 
-      // When the linkage is not internal we must emit an alias for the original
-      // variable name as it may be accessed from another object. On non-Mach-O
-      // we can also emit an alias for internal linkage as it's safe to do so.
-      // It's not safe on Mach-O as the alias (and thus the portion of the
-      // MergedGlobals variable) may be dead stripped at link time.
-      if (Linkage != GlobalValue::InternalLinkage || !IsMachO) {
+      // Emit an alias for the original variable name. This is necessary for an
+      // external symbol, as it may be accessed from another object. For
+      // internal symbols, it's not strictly required, but it's useful.
+      //
+      // This _should_ also work on Mach-O ever since '.alt_entry' support was
+      // added in 2016. Unfortunately, there's a bug in ld-prime (present at
+      // least from Xcode 15.0 through Xcode 16.0), in which -dead_strip doesn't
+      // always honor alt_entry. To workaround this issue, we don't emit aliases
+      // on Mach-O. Except, we _must_ do so for external symbols. That means
+      // MergeExternal is broken with that linker. (That option is currently off
+      // by default on MachO).
+      if (!IsMachO || Linkage == GlobalValue::ExternalLinkage) {
         GlobalAlias *GA = GlobalAlias::create(Tys[StructIdxs[idx]], AddrSpace,
                                               Linkage, Name, GEP, &M);
         GA->setVisibility(Visibility);
@@ -640,6 +646,17 @@ void GlobalMergeImpl::setMustKeepGlobalVariables(Module &M) {
   }
 }
 
+// This function returns true if the given data Section name has custom
+// subsection-splitting semantics in Mach-O (such as splitting by a fixed size)
+//
+// See also ObjFile::parseSections and getRecordSize in lld/MachO/InputFiles.cpp
+static bool isSpecialMachOSection(StringRef Section) {
+  // Uses starts_with, since section attributes can appear at the end of the name.
+  return Section.starts_with("__DATA,__cfstring") ||
+      Section.starts_with("__DATA,__objc_classrefs") ||
+      Section.starts_with("__DATA,__objc_selrefs");
+}
+
 bool GlobalMergeImpl::run(Module &M) {
   if (!EnableGlobalMerge)
     return false;
@@ -678,6 +695,10 @@ bool GlobalMergeImpl::run(Module &M) {
     unsigned AddressSpace = PT->getAddressSpace();
     StringRef Section = GV.getSection();
 
+    // On Mach-O, some section names have special semantics. Don't merge these.
+    if (IsMachO && isSpecialMachOSection(Section))
+      continue;
+
     // Ignore all 'special' globals.
     if (GV.getName().starts_with("llvm.") || GV.getName().starts_with(".llvm."))
       continue;
diff --git a/llvm/test/Transforms/GlobalMerge/macho-sections.ll b/llvm/test/Transforms/GlobalMerge/macho-sections.ll
new file mode 100644
index 00000000000000..389480f41856e3
--- /dev/null
+++ b/llvm/test/Transforms/GlobalMerge/macho-sections.ll
@@ -0,0 +1,25 @@
+; RUN: opt -global-merge -global-merge-max-offset=100 -S -o - %s | FileCheck %s
+; RUN: opt -passes='global-merge<max-offset=100>' -S -o - %s | FileCheck %s
+
+;; Check that we do _not_ merge data with certain special section-names under Mach-O
+
+target datalayout = "e-p:64:64"
+target triple = "x86_64-apple-macos11"
+
+; CHECK-NOT: @_MergedGlobals
+
+@cfstring1 = private global i32 1, section "__DATA,__cfstring"
+@cfstring2 = private global i32 2, section "__DATA,__cfstring"
+@objcclassrefs1 = private global i32 3, section "__DATA,__objc_classrefs,regular,no_dead_strip"
+@objcclassrefs2 = private global i32 4, section "__DATA,__objc_classrefs,regular,no_dead_strip"
+@objcselrefs1 = private global i32 5, section "__DATA,__objc_selrefs,literal_pointers,no_dead_strip"
+@objcselrefs2 = private global i32 6, section "__DATA,__objc_selrefs,literal_pointers,no_dead_strip"
+define void @use() {
+  load ptr, ptr @cfstring1
+  load ptr, ptr @cfstring2
+  load ptr, ptr @objcclassrefs1
+  load ptr, ptr @objcclassrefs2
+  load ptr, ptr @objcselrefs1
+  load ptr, ptr @objcselrefs2
+  ret void
+}
diff --git a/llvm/test/Transforms/GlobalMerge/macho-symbols.ll b/llvm/test/Transforms/GlobalMerge/macho-symbols.ll
new file mode 100644
index 00000000000000..a3e99a55668bae
--- /dev/null
+++ b/llvm/test/Transforms/GlobalMerge/macho-symbols.ll
@@ -0,0 +1,28 @@
+; RUN: opt -global-merge -global-merge-max-offset=100 -S -o - %s | FileCheck %s
+; RUN: opt -passes='global-merge<max-offset=100>' -S -o - %s | FileCheck %s
+
+;; For Mach-O, we do not expect any alias symbols to be created for
+;; internal/private symbols by GlobalMerge.
+
+target datalayout = "e-p:64:64"
+target triple = "x86_64-apple-macos11"
+
+@a = private global i32 1
+@b = private global i32 2
+@c = internal global i32 3
+@d = internal global i32 4
+
+; CHECK: @_MergedGlobals = internal global <{ i32, i32, i32, i32 }> <{ i32 1, i32 2, i32 3, i32 4 }>, align 4
+; CHECK-NOT: alias
+
+define void @use() {
+  ; CHECK: load i32, ptr @_MergedGlobals,
+  %x = load i32, ptr @a
+  ; CHECK: load i32, ptr getelementptr inbounds (<{ i32, i32, i32, i32 }>, ptr @_MergedGlobals, i32 0, i32 1)
+  %y = load i32, ptr @b
+  ; CHECK: load i32, ptr getelementptr inbounds (<{ i32, i32, i32, i32 }>, ptr @_MergedGlobals, i32 0, i32 2)
+  %z1 = load i32, ptr @c
+  ; CHECK: load i32, ptr getelementptr inbounds (<{ i32, i32, i32, i32 }>, ptr @_MergedGlobals, i32 0, i32 3)
+  %z2 = load i32, ptr @d
+  ret void
+}

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[dtellenbach]

@jroelofs

[smeenai]

LGTM, thanks for fixing this.

[amy-kwan]

Post commit review - LGTM.