release/21.x: ThinLTOBitcodeWriter: Emit __cfi_check to full LTO part of bitcode file.

[llvmbot]

Backport ff85dbd

Requested by: @efriedma-quic

[llvmbot]

@efriedma-quic What do you think about merging this PR to the release branch?

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: None (llvmbot)

Changes

Backport ff85dbd

Requested by: @efriedma-quic

---

Full diff: https://github.com/llvm/llvm-project/pull/154859.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/IPO/ThinLTOBitcodeWriter.cpp (+11-1)
• (added) llvm/test/Transforms/ThinLTOBitcodeWriter/cfi-check.ll (+19)

diff --git a/llvm/lib/Transforms/IPO/ThinLTOBitcodeWriter.cpp b/llvm/lib/Transforms/IPO/ThinLTOBitcodeWriter.cpp
index e276376f21583..0d631734ca968 100644
--- a/llvm/lib/Transforms/IPO/ThinLTOBitcodeWriter.cpp
+++ b/llvm/lib/Transforms/IPO/ThinLTOBitcodeWriter.cpp
@@ -350,12 +350,20 @@ void splitAndWriteThinLTOBitcode(
       });
     }
 
+  auto MustEmitToMergedModule = [](const GlobalValue *GV) {
+    // The __cfi_check definition is filled in by the CrossDSOCFI pass which
+    // runs only in the merged module.
+    return GV->getName() == "__cfi_check";
+  };
+
   ValueToValueMapTy VMap;
   std::unique_ptr<Module> MergedM(
       CloneModule(M, VMap, [&](const GlobalValue *GV) -> bool {
         if (const auto *C = GV->getComdat())
           if (MergedMComdats.count(C))
             return true;
+        if (MustEmitToMergedModule(GV))
+          return true;
         if (auto *F = dyn_cast<Function>(GV))
           return EligibleVirtualFns.count(F);
         if (auto *GVar =
@@ -372,7 +380,7 @@ void splitAndWriteThinLTOBitcode(
   cloneUsedGlobalVariables(M, *MergedM, /*CompilerUsed*/ true);
 
   for (Function &F : *MergedM)
-    if (!F.isDeclaration()) {
+    if (!F.isDeclaration() && !MustEmitToMergedModule(&F)) {
       // Reset the linkage of all functions eligible for virtual constant
       // propagation. The canonical definitions live in the thin LTO module so
       // that they can be imported.
@@ -394,6 +402,8 @@ void splitAndWriteThinLTOBitcode(
     if (const auto *C = GV->getComdat())
       if (MergedMComdats.count(C))
         return false;
+    if (MustEmitToMergedModule(GV))
+      return false;
     return true;
   });
 
diff --git a/llvm/test/Transforms/ThinLTOBitcodeWriter/cfi-check.ll b/llvm/test/Transforms/ThinLTOBitcodeWriter/cfi-check.ll
new file mode 100644
index 0000000000000..b927af6b92f7c
--- /dev/null
+++ b/llvm/test/Transforms/ThinLTOBitcodeWriter/cfi-check.ll
@@ -0,0 +1,19 @@
+; RUN: opt -thinlto-bc -thinlto-split-lto-unit -o %t %s
+; RUN: llvm-modextract -b -n 0 -o - %t | llvm-dis | FileCheck --check-prefix=M0 %s
+; RUN: llvm-modextract -b -n 1 -o - %t | llvm-dis | FileCheck --check-prefix=M1 %s
+
+; Check that __cfi_check is emitted on the full LTO side with
+; attributes preserved.
+
+; M0: define void @f()
+define void @f() !type !{!"f1", i32 0} {
+  ret void 
+}
+
+; M1: define void @__cfi_check() #0
+define void @__cfi_check() #0 {
+  ret void
+}
+
+; M1: attributes #0 = { "branch-target-enforcement" }
+attributes #0 = { "branch-target-enforcement" }

[efriedma-quic]

LGTM

The change only affects CFI, and it should be a safe fix.

[github-actions]

@efriedma-quic (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR.