[AIX][XCOFF] Generate undefined symbol in symbol table for external function call

[jasonliu]

Summary

This patch sets up the infrastructure for

1. Associate MCSymbolXCOFF with an MCSectionXCOFF when it could not get implicitly associated.
2. Generate undefined symbols. The patch itself generates undefined symbol for external function call only. Generate undefined symbol for external global variable and external function descriptors will be handled in separate patch(s) after this is land.

[daltenty]

llvm/lib/MC/XCOFFObjectWriter.cpp

313

getContainingCsect() already has this assertion internally

[jasonliu]

Removed unnecessary assertion.

[jasonliu]

llvm/lib/MC/XCOFFObjectWriter.cpp

313

You are right. Removed.

[daltenty]

LGTM

[hubert.reinterpretcast]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

5330

Does it make sense to check the linkage on the function? What if it is weak?

[hubert.reinterpretcast]

Minor nits if the weak function case needs to be addressed anyway (probably at least a TODO).

llvm/lib/MC/XCOFFObjectWriter.cpp

573

s/symbol's/symbols'/;
or
Calculate indices for undefined symbols.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

5326

Suggestion:
<ins>an </ins>undefined symbol need<ins>s</ins> to <ins>be </ins>associate<ins>d</ins> with

[jasonliu]

Addressed post review comments in https://github.com/llvm/llvm-project/commit/7707d8aa9db8aa3814593f9c40cc707f306e3ae2

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

5330

You are right. We would have a weak external referenced symbol in that case.
We will also need to do more work for weak function in the future though, for example, emit ".weak .foo" and so on.

llvm/include/llvm/MC/MCSectionXCOFF.h

class MCSectionXCOFF final : public MCSection {
   XCOFF::StorageClass StorageClass;
   MCSymbolXCOFF *const QualName;
 
   MCSectionXCOFF(StringRef Section, XCOFF::StorageMappingClass SMC,
                  XCOFF::SymbolType ST, XCOFF::StorageClass SC, SectionKind K,
                  MCSymbolXCOFF *QualName, MCSymbol *Begin)
       : MCSection(SV_XCOFF, K, Begin), Name(Section), MappingClass(SMC),
         Type(ST), StorageClass(SC), QualName(QualName) {
-    assert((ST == XCOFF::XTY_SD || ST == XCOFF::XTY_CM) &&
+    assert((ST == XCOFF::XTY_SD || ST == XCOFF::XTY_CM || ST == XCOFF::XTY_ER) &&
            "Invalid or unhandled type for csect.");
     assert(QualName != nullptr && "QualName is needed.");
     QualName->setStorageClass(SC);
     QualName->setContainingCsect(this);
   }
 
 public:
   ~MCSectionXCOFF();

llvm/include/llvm/MC/MCSymbolXCOFF.h

public:
   }
 
   MCSectionXCOFF *getContainingCsect() const {
     assert(ContainingCsect &&
            "Trying to get containing csect but none was set.");
     return ContainingCsect;
   }
 
+  bool hasContainingCsect() const { return ContainingCsect != nullptr; }
+
 private:
   Optional<XCOFF::StorageClass> StorageClass;
   MCSectionXCOFF *ContainingCsect = nullptr;
 };
 
 } // end namespace llvm
 
 #endif // LLVM_MC_MCSYMBOLXCOFF_H

llvm/lib/MC/XCOFFObjectWriter.cpp

class XCOFFObjectWriter : public MCObjectWriter {
 
   support::endian::Writer W;
   std::unique_ptr<MCXCOFFObjectTargetWriter> TargetObjectWriter;
   StringTableBuilder Strings;
 
   // CsectGroups. These store the csects which make up different parts of
   // the sections. Should have one for each set of csects that get mapped into
   // the same section and get handled in a 'similar' way.
+  CsectGroup UndefinedCsects;
   CsectGroup ProgramCodeCsects;
   CsectGroup DataCsects;
   CsectGroup FuncDSCsects;
   CsectGroup TOCCsects;
   CsectGroup BSSCsects;
 
   // The Predefined sections.
   Section Text;
: W(OS, support::big), TargetObjectWriter(std::move(MOTW)),
       Text(".text", XCOFF::STYP_TEXT, /* IsVirtual */ false,
            CsectGroups{&ProgramCodeCsects}),
       Data(".data", XCOFF::STYP_DATA, /* IsVirtual */ false,
            CsectGroups{&DataCsects, &FuncDSCsects, &TOCCsects}),
       BSS(".bss", XCOFF::STYP_BSS, /* IsVirtual */ true,
           CsectGroups{&BSSCsects}) {}
 
 void XCOFFObjectWriter::reset() {
+  UndefinedCsects.clear(); 
+
   // Reset any sections we have written to, and empty the section header table.
   for (auto *Sec : Sections)
     Sec->reset();
 
   // Reset the symbol table and string table.
   SymbolTableEntryCount = 0;
   SymbolTableOffset = 0;
   SectionCount = 0;
void XCOFFObjectWriter::executePostLayoutBinding(MCAssembler &Asm,
   // wrapper. Needed for finding the ControlSection to insert an MCSymbol into
   // from its containing MCSectionXCOFF.
   DenseMap<const MCSectionXCOFF *, ControlSection *> WrapperMap;
 
   for (const auto &S : Asm) {
     const auto *MCSec = cast<const MCSectionXCOFF>(&S);
     assert(WrapperMap.find(MCSec) == WrapperMap.end() &&
            "Cannot add a csect twice.");
+    assert(XCOFF::XTY_ER != MCSec->getCSectType() &&
+           "An undefined Csect should not get registered.");
 
     // If the name does not fit in the storage provided in the symbol table
     // entry, add it to the string table.
     if (nameShouldBeInStringTable(MCSec->getSectionName()))
       Strings.add(MCSec->getSectionName());
 
     CsectGroup &Group = getCsectGroup(MCSec);
     Group.emplace_back(MCSec);
     WrapperMap[MCSec] = &Group.back();
   }
 
   for (const MCSymbol &S : Asm.symbols()) {
     // Nothing to do for temporary symbols.
     if (S.isTemporary())
       continue;
     const MCSymbolXCOFF *XSym = cast<MCSymbolXCOFF>(&S);
 
     // Map the symbol into its containing csect.
     const MCSectionXCOFF *ContainingCsect = XSym->getContainingCsect();
-    assert(WrapperMap.find(ContainingCsect) != WrapperMap.end() &&
-           "Expected containing csect to exist in map");
+    assert(ContainingCsect && "Every MCSymbolXCOFF should have a csect.");
    
     // If the symbol is the Csect itself, we don't need to put the symbol
     // into Csect's Syms.
     if (XSym == ContainingCsect->getQualNameSymbol())
       continue;
     
+    if (XSym->isUndefined(false)) {
+      UndefinedCsects.emplace_back(ContainingCsect);
+      continue;
+    }
+
+    assert(WrapperMap.find(ContainingCsect) != WrapperMap.end() &&
+           "Expected containing csect to exist in map");
+
     // Lookup the containing csect and add the symbol to it.
     WrapperMap[ContainingCsect]->Syms.emplace_back(XSym);
 
     // If the name does not fit in the storage provided in the symbol table
     // entry, add it to the string table.
     if (nameShouldBeInStringTable(XSym->getName()))
       Strings.add(XSym->getName());
     }
for (const auto *Sec : Sections) {
     W.write<uint16_t>(0);
     W.write<uint16_t>(0);
 
     W.write<int32_t>(Sec->Flags);
   }
 }
 
 void XCOFFObjectWriter::writeSymbolTable(const MCAsmLayout &Layout) {
+  for (const auto &Csect : UndefinedCsects) {
+    writeSymbolTableEntryForControlSection(
+        Csect, XCOFF::ReservedSectionNum::N_UNDEF, Csect.MCCsect->getStorageClass());
+  }
+
   for (const auto *Section : Sections) {
     // Nothing to write for this Section.
     if (Section->Index == Section::UninitializedIndex)
       continue;
 
     for (const auto *Group : Section->Groups) {
       if (Group->empty())
         continue;
 
       const int16_t SectionIndex = Section->Index;
       for (const auto &Csect : *Group) {
         // Write out the control section first and then each symbol in it.
         writeSymbolTableEntryForControlSection(
             Csect, SectionIndex, Csect.MCCsect->getStorageClass());
 
         for (const auto Sym : Csect.Syms)
           writeSymbolTableEntryForCsectMemberLabel(
               Sym, Csect, SectionIndex, Layout.getSymbolOffset(*(Sym.MCSym)));
       }
     }
   }
 }
 
 void XCOFFObjectWriter::assignAddressesAndIndices(const MCAsmLayout &Layout) {
+  // The first symbol table entry is for the file name. We are not emitting it
+  // yet, so start at index 0.
+  uint32_t SymbolTableIndex = 0;
+
+  // Calculate undefined symbol's indices.
+  for (auto &Csect : UndefinedCsects) {
+    Csect.Size = 0;
+    Csect.Address = 0;
+    Csect.SymbolTableIndex = SymbolTableIndex;
+    // 1 main and 1 auxiliary symbol table entry for each contained symbol.
+    SymbolTableIndex += 2;
+  }
+
   // The address corrresponds to the address of sections and symbols in the
   // object file. We place the shared address 0 immediately after the
   // section header table.
   uint32_t Address = 0;
   // Section indices are 1-based in XCOFF.
   int32_t SectionIndex = 1;
-  // The first symbol table entry is for the file name. We are not emitting it
-  // yet, so start at index 0.
-  uint32_t SymbolTableIndex = 0;
 
   for (auto *Section : Sections) {
     const bool IsEmpty =
         llvm::all_of(Section->Groups,
                      [](const CsectGroup *Group) { return Group->empty(); });
     if (IsEmpty)
       continue;
 

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

SDValue PPCTargetLowering::FinishCall(
   }
 
   if (Subtarget.isAIXABI() && isFunctionGlobalAddress(Callee)) {
     // On AIX, direct function calls reference the symbol for the function's
     // entry point, which is named by inserting a "." before the function's
     // C-linkage name.
     GlobalAddressSDNode *G = cast<GlobalAddressSDNode>(Callee);
     auto &Context = DAG.getMachineFunction().getMMI().getContext();
-    MCSymbol *S = Context.getOrCreateSymbol(Twine(".") +
-                                            Twine(G->getGlobal()->getName()));
+    
+    MCSymbolXCOFF *S = cast<MCSymbolXCOFF>(Context.getOrCreateSymbol(
+        Twine(".") + Twine(G->getGlobal()->getName())));
+    
+    const GlobalValue *GV = G->getGlobal();
+    if (GV && GV->isDeclaration() && !S->hasContainingCsect()) {
+      // On AIX, undefined symbol need to associate with a MCSectionXCOFF to
+      // get the correct storage mapping class. In this case, XCOFF::XMC_PR.
+      MCSectionXCOFF *Sec =
+          Context.getXCOFFSection(S->getName(), XCOFF::XMC_PR, XCOFF::XTY_ER,
+                                  XCOFF::C_EXT, SectionKind::getMetadata());
+      S->setContainingCsect(Sec);
+    }
+
     Callee = DAG.getMCSymbol(S, PtrVT);
     // Replace the GlobalAddressSDNode Callee with the MCSymbolSDNode.
     Ops[1] = Callee;
   }
 
   Chain = DAG.getNode(CallOpc, dl, NodeTys, Ops);
   InFlag = Chain.getValue(1);
 

llvm/test/CodeGen/PowerPC/aix-undef-func-call.ll

+; RUN: llc -mtriple powerpc-ibm-aix-xcoff -filetype=obj -o %t.o < %s
+; RUN: llvm-readobj  --symbols %t.o | FileCheck %s
+
+define void @bar() {
+entry:
+  call void bitcast (void (...)* @foo to void ()*)()
+  ret void
+}
+
+declare void @foo(...)
+
+;CHECK: Symbol {
+;CHECK:   Name: .foo
+;CHECK-NEXT:   Value (RelocatableAddress): 0x0
+;CHECK-NEXT:   Section: N_UNDEF
+;CHECK-NEXT:   Type: 0x0
+;CHECK-NEXT:   StorageClass: C_EXT (0x2)
+;CHECK-NEXT:   NumberOfAuxEntries: 1
+;CHECK-NEXT:   CSECT Auxiliary Entry {
+;CHECK:          SectionLen: 0
+;CHECK-NEXT:     ParameterHashIndex: 0x0
+;CHECK-NEXT:     TypeChkSectNum: 0x0
+;CHECK-NEXT:     SymbolAlignmentLog2: 0
+;CHECK-NEXT:     SymbolType: XTY_ER (0x0)
+;CHECK-NEXT:     StorageMappingClass: XMC_PR (0x0)
+;CHECK-NEXT:     StabInfoIndex: 0x0
+;CHECK-NEXT:     StabSectNum: 0x0
+;CHECK-NEXT:   }
+;CHECK-NEXT: }
+