[IR] Introduce llvm.errno.tbaa metadata for errno alias disambiguation

llvm/docs/LangRef.rst

@@ -8864,6 +8864,28 @@ For example, the following metadata section contains two library specifiers::
 Each library specifier will be handled independently by the consuming linker.
 The effect of the library specifiers are defined by the consuming linker.
 
+'``llvm.errno.tbaa``' Named Metadata
+====================================
+
+The module-level ``!llvm.errno.tbaa`` metadata specifies the TBAA nodes used
+for accessing ``errno``. These nodes are guaranteed to represent int-compatible
+accesses according to C/C++ strict aliasing rules. This should let LLVM alias
+analyses to reason about aliasing with ``errno`` when calling library functions
+that may set ``errno``, allowing optimizations such as store-to-load forwarding
+across such routines.
+
+For example, the following is a valid metadata specifying the TBAA information
+for an integer access:
+
+    !llvm.errno.tbaa = !{!0}
+    !0 = !{!1, !1, i64 0}
+    !1 = !{!"int", !2, i64 0}
+    !2 = !{!"omnipotent char", !3, i64 0}
+    !3 = !{!"Simple C/C++ TBAA"}
+
+Multiple TBAA operands are allowed to support merging of modules that may use
+different TBAA hierarchies (e.g., when mixing C and C++).
+
 .. _summary:
 
 ThinLTO Summary

llvm/include/llvm/IR/Verifier.h

@@ -75,9 +75,9 @@ class TBAAVerifier {
 public:
   TBAAVerifier(VerifierSupport *Diagnostic = nullptr)
       : Diagnostic(Diagnostic) {}
-  /// Visit an instruction and return true if it is valid, return false if an
-  /// invalid TBAA is attached.
-  LLVM_ABI bool visitTBAAMetadata(Instruction &I, const MDNode *MD);
+  /// Visit an instruction, or a TBAA node itself as part of a metadata, and
+  /// return true if it is valid, return false if an invalid TBAA is attached.
+  LLVM_ABI bool visitTBAAMetadata(Instruction *I, const MDNode *MD);
 };
 
 /// Check a function for errors, useful for use when debugging a

llvm/lib/Bitcode/Reader/BitcodeReader.cpp

@@ -7024,7 +7024,7 @@ Error BitcodeReader::materialize(GlobalValue *GV) {
   if (!MDLoader->isStrippingTBAA()) {
     for (auto &I : instructions(F)) {
       MDNode *TBAA = I.getMetadata(LLVMContext::MD_tbaa);
-      if (!TBAA || TBAAVerifyHelper.visitTBAAMetadata(I, TBAA))
+      if (!TBAA || TBAAVerifyHelper.visitTBAAMetadata(&I, TBAA))
         continue;
       MDLoader->setStripTBAA(true);
       stripTBAA(F->getParent());

llvm/lib/IR/Verifier.cpp

@@ -480,6 +480,7 @@ class Verifier : public InstVisitor<Verifier>, VerifierSupport {
     visitModuleFlags();
     visitModuleIdents();
     visitModuleCommandLines();
+    visitModuleErrnoTBAA();
 
     verifyCompileUnits();
 
@@ -516,6 +517,7 @@ class Verifier : public InstVisitor<Verifier>, VerifierSupport {
   void visitComdat(const Comdat &C);
   void visitModuleIdents();
   void visitModuleCommandLines();
+  void visitModuleErrnoTBAA();
   void visitModuleFlags();
   void visitModuleFlag(const MDNode *Op,
                        DenseMap<const MDString *, const MDNode *> &SeenIDs,
@@ -1815,6 +1817,18 @@ void Verifier::visitModuleCommandLines() {
   }
 }
 
+void Verifier::visitModuleErrnoTBAA() {
+  const NamedMDNode *ErrnoTBAA = M.getNamedMetadata("llvm.errno.tbaa");
+  if (!ErrnoTBAA)
+    return;
+
+  Check(ErrnoTBAA->getNumOperands() >= 1,
+        "llvm.errno.tbaa must have at least one operand", ErrnoTBAA);
+
+  for (const MDNode *N : ErrnoTBAA->operands())
+    TBAAVerifyHelper.visitTBAAMetadata(nullptr, N);
+}
+
 void Verifier::visitModuleFlags() {
   const NamedMDNode *Flags = M.getModuleFlagsMetadata();
   if (!Flags) return;
@@ -5537,7 +5551,7 @@ void Verifier::visitInstruction(Instruction &I) {
     visitNofreeMetadata(I, MD);
 
   if (MDNode *TBAA = I.getMetadata(LLVMContext::MD_tbaa))
-    TBAAVerifyHelper.visitTBAAMetadata(I, TBAA);
+    TBAAVerifyHelper.visitTBAAMetadata(&I, TBAA);
 
   if (MDNode *MD = I.getMetadata(LLVMContext::MD_noalias))
     visitAliasScopeListMetadata(MD);
@@ -7879,21 +7893,22 @@ static bool isNewFormatTBAATypeNode(llvm::MDNode *Type) {
   return isa_and_nonnull<MDNode>(Type->getOperand(0));
 }
 
-bool TBAAVerifier::visitTBAAMetadata(Instruction &I, const MDNode *MD) {
-  CheckTBAA(MD->getNumOperands() > 0, "TBAA metadata cannot have 0 operands",
-            &I, MD);
+bool TBAAVerifier::visitTBAAMetadata(Instruction *I, const MDNode *MD) {
+  CheckTBAA(MD->getNumOperands() > 0, "TBAA metadata cannot have 0 operands", I,
+            MD);
 
-  CheckTBAA(isa<LoadInst>(I) || isa<StoreInst>(I) || isa<CallInst>(I) ||
-                isa<VAArgInst>(I) || isa<AtomicRMWInst>(I) ||
-                isa<AtomicCmpXchgInst>(I),
-            "This instruction shall not have a TBAA access tag!", &I);
+  if (I)
+    CheckTBAA(isa<LoadInst>(I) || isa<StoreInst>(I) || isa<CallInst>(I) ||
+                  isa<VAArgInst>(I) || isa<AtomicRMWInst>(I) ||
+                  isa<AtomicCmpXchgInst>(I),
+              "This instruction shall not have a TBAA access tag!", I);
 
   bool IsStructPathTBAA =
       isa<MDNode>(MD->getOperand(0)) && MD->getNumOperands() >= 3;
 
   CheckTBAA(IsStructPathTBAA,
             "Old-style TBAA is no longer allowed, use struct-path TBAA instead",
-            &I);
+            I);
 
   MDNode *BaseNode = dyn_cast_or_null<MDNode>(MD->getOperand(0));
   MDNode *AccessType = dyn_cast_or_null<MDNode>(MD->getOperand(1));
@@ -7902,17 +7917,17 @@ bool TBAAVerifier::visitTBAAMetadata(Instruction &I, const MDNode *MD) {
 
   if (IsNewFormat) {
     CheckTBAA(MD->getNumOperands() == 4 || MD->getNumOperands() == 5,
-              "Access tag metadata must have either 4 or 5 operands", &I, MD);
+              "Access tag metadata must have either 4 or 5 operands", I, MD);
   } else {
     CheckTBAA(MD->getNumOperands() < 5,
-              "Struct tag metadata must have either 3 or 4 operands", &I, MD);
+              "Struct tag metadata must have either 3 or 4 operands", I, MD);
   }
 
   // Check the access size field.
   if (IsNewFormat) {
     auto *AccessSizeNode = mdconst::dyn_extract_or_null<ConstantInt>(
         MD->getOperand(3));
-    CheckTBAA(AccessSizeNode, "Access size field must be a constant", &I, MD);
+    CheckTBAA(AccessSizeNode, "Access size field must be a constant", I, MD);
   }
 
   // Check the immutability flag.
@@ -7921,45 +7936,45 @@ bool TBAAVerifier::visitTBAAMetadata(Instruction &I, const MDNode *MD) {
     auto *IsImmutableCI = mdconst::dyn_extract_or_null<ConstantInt>(
         MD->getOperand(ImmutabilityFlagOpNo));
     CheckTBAA(IsImmutableCI,
-              "Immutability tag on struct tag metadata must be a constant", &I,
+              "Immutability tag on struct tag metadata must be a constant", I,
               MD);
     CheckTBAA(
         IsImmutableCI->isZero() || IsImmutableCI->isOne(),
-        "Immutability part of the struct tag metadata must be either 0 or 1",
-        &I, MD);
+        "Immutability part of the struct tag metadata must be either 0 or 1", I,
+        MD);
   }
 
   CheckTBAA(BaseNode && AccessType,
             "Malformed struct tag metadata: base and access-type "
             "should be non-null and point to Metadata nodes",
-            &I, MD, BaseNode, AccessType);
+            I, MD, BaseNode, AccessType);
 
   if (!IsNewFormat) {
     CheckTBAA(isValidScalarTBAANode(AccessType),
-              "Access type node must be a valid scalar type", &I, MD,
+              "Access type node must be a valid scalar type", I, MD,
               AccessType);
   }
 
   auto *OffsetCI = mdconst::dyn_extract_or_null<ConstantInt>(MD->getOperand(2));
-  CheckTBAA(OffsetCI, "Offset must be constant integer", &I, MD);
+  CheckTBAA(OffsetCI, "Offset must be constant integer", I, MD);
 
   APInt Offset = OffsetCI->getValue();
   bool SeenAccessTypeInPath = false;
 
   SmallPtrSet<MDNode *, 4> StructPath;
 
   for (/* empty */; BaseNode && !IsRootTBAANode(BaseNode);
-       BaseNode = getFieldNodeFromTBAABaseNode(I, BaseNode, Offset,
-                                               IsNewFormat)) {
+       BaseNode =
+           getFieldNodeFromTBAABaseNode(*I, BaseNode, Offset, IsNewFormat)) {
     if (!StructPath.insert(BaseNode).second) {
-      CheckFailed("Cycle detected in struct path", &I, MD);
+      CheckFailed("Cycle detected in struct path", I, MD);
       return false;
     }
 
     bool Invalid;
     unsigned BaseNodeBitWidth;
-    std::tie(Invalid, BaseNodeBitWidth) = verifyTBAABaseNode(I, BaseNode,
-                                                             IsNewFormat);
+    std::tie(Invalid, BaseNodeBitWidth) =
+        verifyTBAABaseNode(*I, BaseNode, IsNewFormat);
 
     // If the base node is invalid in itself, then we've already printed all the
     // errors we wanted to print.
@@ -7969,20 +7984,20 @@ bool TBAAVerifier::visitTBAAMetadata(Instruction &I, const MDNode *MD) {
     SeenAccessTypeInPath |= BaseNode == AccessType;
 
     if (isValidScalarTBAANode(BaseNode) || BaseNode == AccessType)
-      CheckTBAA(Offset == 0, "Offset not zero at the point of scalar access",
-                &I, MD, &Offset);
+      CheckTBAA(Offset == 0, "Offset not zero at the point of scalar access", I,
+                MD, &Offset);
 
     CheckTBAA(BaseNodeBitWidth == Offset.getBitWidth() ||
                   (BaseNodeBitWidth == 0 && Offset == 0) ||
                   (IsNewFormat && BaseNodeBitWidth == ~0u),
-              "Access bit-width not the same as description bit-width", &I, MD,
+              "Access bit-width not the same as description bit-width", I, MD,
               BaseNodeBitWidth, Offset.getBitWidth());
 
     if (IsNewFormat && SeenAccessTypeInPath)
       break;
   }
 
-  CheckTBAA(SeenAccessTypeInPath, "Did not see access type in access path!", &I,
+  CheckTBAA(SeenAccessTypeInPath, "Did not see access type in access path!", I,
             MD);
   return true;
 }

llvm/test/Linker/Inputs/errno-tbaa-cxx-metadata.ll

@@ -0,0 +1,5 @@
+!llvm.errno.tbaa = !{!0}
+!0 = !{!1, !1, i64 0}
+!1 = !{!"int", !2, i64 0}
+!2 = !{!"omnipotent char", !3, i64 0}
+!3 = !{!"Simple C++ TBAA"}

llvm/test/Linker/Inputs/errno-tbaa-metadata.ll

@@ -0,0 +1,5 @@
+!llvm.errno.tbaa = !{!0}
+!0 = !{!1, !1, i64 0}
+!1 = !{!"int", !2, i64 0}
+!2 = !{!"omnipotent char", !3, i64 0}
+!3 = !{!"Simple C/C++ TBAA"}

llvm/test/Linker/link-errno-tbaa-metadata.ll

@@ -0,0 +1,8 @@
+; RUN: llvm-link %S/Inputs/errno-tbaa-metadata.ll %S/Inputs/errno-tbaa-cxx-metadata.ll -S -o - | FileCheck %s --check-prefix=CHECK-MERGE
+; RUN: llvm-link %S/Inputs/errno-tbaa-metadata.ll %S/Inputs/errno-tbaa-metadata.ll -S -o - | FileCheck %s --check-prefix=CHECK-DEDUP
+
+; Ensure merging when linking modules w/ different errno TBAA hierarchies.
+; CHECK-MERGE: !llvm.errno.tbaa = !{![[NODE0:[0-9]+]], ![[NODE1:[0-9]+]]}
+
+; Ensure deduplication when linking modules w/ identical errno TBAA nodes.
+; CHECK-DEDUP: !llvm.errno.tbaa = !{![[NODE:[0-9]+]]}

llvm/test/Verifier/errno-tbaa-metadata-1.ll

@@ -0,0 +1,5 @@
+; RUN: not llvm-as < %s -o /dev/null 2>&1 | FileCheck %s
+
+; CHECK: assembly parsed, but does not verify as correct!
+; CHECK-NEXT: llvm.errno.tbaa must have at least one operand
+!llvm.errno.tbaa = !{}

llvm/test/Verifier/errno-tbaa-metadata-2.ll

@@ -0,0 +1,9 @@
+; RUN: not llvm-as < %s -o /dev/null 2>&1 | FileCheck %s
+
+; CHECK: assembly parsed, but does not verify as correct!
+; CHECK-NEXT: Malformed struct tag metadata: base and access-type should be non-null and point to Metadata nodes
+!llvm.errno.tbaa = !{!0}
+!0 = !{!1, i64 0, !1}
+!1 = !{!"int", !2, i64 0}
+!2 = !{!"omnipotent char", !3, i64 0}
+!3 = !{!"Simple C/C++ TBAA"}