[AIX][XCOFF] Write Function descriptors and TOC base to data section

[jasonliu]

Summary

This patch implements writing function descriptors and TOC base into data section,
and also add function descriptors(both csect and label) and TOC base symbols to the symbol table.

[DiggerLin]

llvm/lib/MC/XCOFFObjectWriter.cpp

256

I am not sure whether all function description data will be group together and than put in the .data section of xcoffobjectfile? all the variable data and function description will be interleaved into data section.

As I remember, I read some document , it talk about that , we put all variables in the data section first. then function description,s final is TOC entries. Maybe I wrong.

[daltenty]

Do we care about asserting the uniqueness of the TOC anchor somewhere?

[DiggerLin]

llvm/lib/MC/XCOFFObjectWriter.cpp

256

I prefer added CsectGroup DSCsects; for function description csect.

[jasonliu]

In D70271#1747732, @daltenty wrote:

Do we care about asserting the uniqueness of the TOC anchor somewhere?

Sure. I will do it somewhere.

llvm/lib/MC/XCOFFObjectWriter.cpp

256

Interleaving is acceptable for XCOFF format. What it hurts is readability of the symbol table.
But I think the current implementation would naturally put all the global datas first and then function descriptor later, because we always emit global data first and then process each function.

[jasonliu]

llvm/lib/MC/XCOFFObjectWriter.cpp

256

I can add a separate CsectGroup to make the contract more explicit.

[DiggerLin]

llvm/lib/MC/XCOFFObjectWriter.cpp

256

thanks Jason for clarification.

[Xiangling_L]

llvm/lib/MC/XCOFFObjectWriter.cpp

154

I am not quite sure, but my understanding is that we only have one TOC base csect in one translation unit? If I am correct, do we want to name this TOCCsect or add a comment to specify that point? It's a little confused to see plural here.

[sfertile]

llvm/lib/MC/XCOFFObjectWriter.cpp

154

Seconded. Do you intend to map the toc entries into this csect group as well?

256

If we intend to map the toc entries into TOCCsects along side the toc base csect then its probably best to follow suit and have a separate csect group for the descriptors and make this ordering explicit. If not then we need to have lit test that explicitly check that descriptor data is mapped to after read/write data. It might be worthwhile to check said tests with LLVM_REVERSE_ITERATION enabled as well to see if that has any affect.

[jasonliu]

Added separate CsectGroup to contain function descriptor csects.
Added assertion to assert uniqueness TOC base.

[jasonliu]

llvm/lib/MC/XCOFFObjectWriter.cpp

154

The current intend is to map toc entries into this csect group. Hence named it TOCCsects. And I put an assertion in getCsectGroup() to make sure TOC-base is unique and is the first in the group.
If later we decide it's better to separate them out. I could rename the csect.

256

I map the function descriptor to a separate csect group now.

[jasonliu]

I believe I addressed all the comments. Thanks.

[DiggerLin]

LGTM

[jasonliu]

Landed with https://github.com/llvm/llvm-project/commit/c9edaa828eb84a62af2042a1a64168e32ea32924

llvm/lib/MC/XCOFFObjectWriter.cpp

class XCOFFObjectWriter : public MCObjectWriter {
   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 ProgramCodeCsects;
   CsectGroup DataCsects;
+  CsectGroup FuncDSCsects;
+  CsectGroup TOCCsects;
   CsectGroup BSSCsects;
 
   // The Predefined sections.
   Section Text;
   Section Data;
   Section BSS;
 
   // All the XCOFF sections, in the order they will appear in the section header

 
 XCOFFObjectWriter::XCOFFObjectWriter(
     std::unique_ptr<MCXCOFFObjectTargetWriter> MOTW, raw_pwrite_stream &OS)
     : W(OS, support::big), TargetObjectWriter(std::move(MOTW)),
       Strings(StringTableBuilder::XCOFF),
       Text(".text", XCOFF::STYP_TEXT, /* IsVirtual */ false,
            CsectGroups{&ProgramCodeCsects}),
       Data(".data", XCOFF::STYP_DATA, /* IsVirtual */ false,
-           CsectGroups{&DataCsects}),
+           CsectGroups{&DataCsects, &FuncDSCsects, &TOCCsects}),
       BSS(".bss", XCOFF::STYP_BSS, /* IsVirtual */ true,
           CsectGroups{&BSSCsects}) {}
 
 void XCOFFObjectWriter::reset() {
   // Reset any sections we have written to, and empty the section header table.
   for (auto *Sec : Sections)
     Sec->reset();
 
CsectGroup &XCOFFObjectWriter::getCsectGroup(const MCSectionXCOFF *MCSec) {
   case XCOFF::XMC_RW:
     if (XCOFF::XTY_CM == MCSec->getCSectType())
       return BSSCsects;
 
     if (XCOFF::XTY_SD == MCSec->getCSectType())
       return DataCsects;
 
     report_fatal_error("Unhandled mapping of read-write csect to section.");
+  case XCOFF::XMC_DS:
+    return FuncDSCsects;
   case XCOFF::XMC_BS:
     assert(XCOFF::XTY_CM == MCSec->getCSectType() &&
            "Mapping invalid csect. CSECT with bss storage class must be "
            "common type.");
     return BSSCsects;
+  case XCOFF::XMC_TC0:
+    assert(XCOFF::XTY_SD == MCSec->getCSectType() &&
+           "Only an initialized csect can contain TOC-base.");
+    assert(TOCCsects.empty() &&
+           "We should have only one TOC-base, and it should be the first csect "
+           "in this CsectGroup.");
+    return TOCCsects;
   default:
     report_fatal_error("Unhandled mapping of csect to section.");
   }
 }
 
 void XCOFFObjectWriter::executePostLayoutBinding(MCAssembler &Asm,
                                                  const MCAsmLayout &Layout) {
   if (TargetObjectWriter->is64Bit())
for (const auto &S : Asm) {
     assert(WrapperMap.find(MCSec) == WrapperMap.end() &&
            "Cannot add a csect twice.");
 
     // 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());
 
-    // TODO FIXME Handle emiting the TOC base.
-    if (MCSec->getMappingClass() == XCOFF::XMC_TC0)
-      continue;
-
     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())
void XCOFFObjectWriter::executePostLayoutBinding(MCAssembler &Asm,
 
   Strings.finalize();
   assignAddressesAndIndices(Layout);
 }
 
 void XCOFFObjectWriter::recordRelocation(MCAssembler &, const MCAsmLayout &,
                                          const MCFragment *, const MCFixup &,
                                          MCValue, uint64_t &) {
-  report_fatal_error("XCOFF relocations not supported.");
+  // TODO: recordRelocation is not yet implemented.
 }
 
 void XCOFFObjectWriter::writeSections(const MCAssembler &Asm,
                                       const MCAsmLayout &Layout) {
   uint32_t CurrentAddressLocation = 0;
   for (const auto *Section : Sections) {
     // Nothing to write for this Section.
     if (Section->Index == Section::UninitializedIndex || Section->IsVirtual)

llvm/test/CodeGen/PowerPC/aix-func-dsc-gen.ll

+; RUN: llc -verify-machineinstrs -mcpu=pwr7 -mtriple powerpc-ibm-aix-xcoff -filetype=obj -o %t.o < %s
+; RUN: llvm-readobj  --symbols %t.o | FileCheck %s
+
+define void @foo() {
+entry:
+  ret void
+}
+
+; CHECK:      File: {{.*}}aix-func-dsc-gen.ll.tmp.o
+; CHECK-NEXT: Format: aixcoff-rs6000
+; CHECK-NEXT: Arch: powerpc
+; CHECK-NEXT: AddressSize: 32bit
+; CHECK:        Symbol {
+; CHECK-NEXT:     Index: [[#Index:]]
+; CHECK-NEXT:     Name: .text
+; CHECK-NEXT:     Value (RelocatableAddress): 0x0
+; CHECK-NEXT:     Section: .text
+; CHECK-NEXT:     Type: 0x0
+; CHECK-NEXT:     StorageClass: C_HIDEXT (0x6B)
+; CHECK-NEXT:     NumberOfAuxEntries: 1
+; CHECK-NEXT:     CSECT Auxiliary Entry {
+; CHECK-NEXT:       Index: [[#Index+1]]
+; CHECK-NEXT:       SectionLen: 4
+; CHECK-NEXT:       ParameterHashIndex: 0x0
+; CHECK-NEXT:       TypeChkSectNum: 0x0
+; CHECK-NEXT:       SymbolAlignmentLog2: 4 
+; CHECK-NEXT:       SymbolType: XTY_SD (0x1)
+; CHECK-NEXT:       StorageMappingClass: XMC_PR (0x0)
+; CHECK-NEXT:       StabInfoIndex: 0x0
+; CHECK-NEXT:       StabSectNum: 0x0
+; CHECK-NEXT:     }
+; CHECK-NEXT:   }
+; CHECK-NEXT:   Symbol {
+; CHECK-NEXT:     Index: [[#Index+2]]
+; CHECK-NEXT:     Name: .foo
+; CHECK-NEXT:     Value (RelocatableAddress): 0x0
+; CHECK-NEXT:     Section: .text
+; CHECK-NEXT:     Type: 0x0
+; CHECK-NEXT:     StorageClass: C_EXT (0x2)
+; CHECK-NEXT:     NumberOfAuxEntries: 1
+; CHECK-NEXT:     CSECT Auxiliary Entry {
+; CHECK-NEXT:       Index: [[#Index+3]]
+; CHECK-NEXT:       ContainingCsectSymbolIndex: [[#Index]]
+; CHECK-NEXT:       ParameterHashIndex: 0x0
+; CHECK-NEXT:       TypeChkSectNum: 0x0
+; CHECK-NEXT:       SymbolAlignmentLog2: 0
+; CHECK-NEXT:       SymbolType: XTY_LD (0x2)
+; CHECK-NEXT:       StorageMappingClass: XMC_PR (0x0)
+; CHECK-NEXT:       StabInfoIndex: 0x0
+; CHECK-NEXT:       StabSectNum: 0x0
+; CHECK-NEXT:     }
+; CHECK-NEXT:   }
+; CHECK-NEXT:   Symbol {
+; CHECK-NEXT:     Index: [[#Index+4]]
+; CHECK-NEXT:     Name: foo
+; CHECK-NEXT:     Value (RelocatableAddress): 0x4
+; CHECK-NEXT:     Section: .data
+; CHECK-NEXT:     Type: 0x0
+; CHECK-NEXT:     StorageClass: C_HIDEXT (0x6B)
+; CHECK-NEXT:     NumberOfAuxEntries: 1
+; CHECK-NEXT:     CSECT Auxiliary Entry {
+; CHECK-NEXT:       Index: [[#Index+5]]
+; CHECK-NEXT:       SectionLen: 12
+; CHECK-NEXT:       ParameterHashIndex: 0x0
+; CHECK-NEXT:       TypeChkSectNum: 0x0
+; CHECK-NEXT:       SymbolAlignmentLog2: 0
+; CHECK-NEXT:       SymbolType: XTY_SD (0x1)
+; CHECK-NEXT:       StorageMappingClass: XMC_DS (0xA)
+; CHECK-NEXT:       StabInfoIndex: 0x0
+; CHECK-NEXT:       StabSectNum: 0x0
+; CHECK-NEXT:     }
+; CHECK-NEXT:   }
+; CHECK-NEXT:   Symbol {
+; CHECK-NEXT:     Index: [[#Index+6]]
+; CHECK-NEXT:     Name: foo
+; CHECK-NEXT:     Value (RelocatableAddress): 0x4
+; CHECK-NEXT:     Section: .data
+; CHECK-NEXT:     Type: 0x0
+; CHECK-NEXT:     StorageClass: C_EXT (0x2)
+; CHECK-NEXT:     NumberOfAuxEntries: 1
+; CHECK-NEXT:     CSECT Auxiliary Entry {
+; CHECK-NEXT:       Index: [[#Index+7]]
+; CHECK-NEXT:       ContainingCsectSymbolIndex: [[#Index+4]]
+; CHECK-NEXT:       ParameterHashIndex: 0x0
+; CHECK-NEXT:       TypeChkSectNum: 0x0
+; CHECK-NEXT:       SymbolAlignmentLog2: 0
+; CHECK-NEXT:       SymbolType: XTY_LD (0x2)
+; CHECK-NEXT:       StorageMappingClass: XMC_DS (0xA)
+; CHECK-NEXT:       StabInfoIndex: 0x0
+; CHECK-NEXT:       StabSectNum: 0x0
+; CHECK-NEXT:     }
+; CHECK-NEXT:   }
+; CHECK-NEXT:   Symbol {
+; CHECK-NEXT:     Index: [[#Index+8]]
+; CHECK-NEXT:     Name: TOC
+; CHECK-NEXT:     Value (RelocatableAddress): 0x10
+; CHECK-NEXT:     Section: .data
+; CHECK-NEXT:     Type: 0x0
+; CHECK-NEXT:     StorageClass: C_HIDEXT (0x6B)
+; CHECK-NEXT:     NumberOfAuxEntries: 1
+; CHECK-NEXT:     CSECT Auxiliary Entry {
+; CHECK-NEXT:       Index: [[#Index+9]]
+; CHECK-NEXT:       SectionLen: 0
+; CHECK-NEXT:       ParameterHashIndex: 0x0
+; CHECK-NEXT:       TypeChkSectNum: 0x0
+; CHECK-NEXT:       SymbolAlignmentLog2: 0
+; CHECK-NEXT:       SymbolType: XTY_SD (0x1)
+; CHECK-NEXT:       StorageMappingClass: XMC_TC0 (0xF)
+; CHECK-NEXT:       StabInfoIndex: 0x0
+; CHECK-NEXT:       StabSectNum: 0x0
+; CHECK-NEXT:     }
+; CHECK-NEXT:   }
+