[JITLink] Add anonymous symbols in LinkGraph for unnamed temporary symbols

[StephenFan]

Summary

In RISCV, temporary symbols will be used to generate dwarf, eh_frame sections..., and will be placed in object code's symbol table. However, LLVM does not use names on these temporary symbols. This patch add anonymous symbols in LinkGraph for these temporary symbols.

[StephenFan]

Amend comments

[StephenFan]

Fix typos

[StephenFan]

Add a testcase

[lhames]

LGTM. Thanks Stefan!

[StephenFan]

rebase

[StephenFan]

Reverted due to miss # REQUIRES: asserts in test file

[lhames]

This was reapplied with fixes in 0283b07746e87.

llvm/include/llvm/ExecutionEngine/JITLink/riscv.h

enum EdgeKind_riscv : Edge::Kind {
   R_RISCV_PCREL_LO12_S,
 
   /// PC relative call
   ///
   /// Fixup expression:
   ///   Fixup <- (Target - Fixup + Addend)
   R_RISCV_CALL,
 
+  /// 32 bits PC relative relocation
+  ///
+  /// Fixup expression:
+  ///   Fixup <- (Target - Fixup + Addend)
+  R_RISCV_32_PCREL,
+
   /// PC relative GOT offset
   ///
   /// Fixup expression:
   ///   Fixup <- (GOT - Fixup + Addend) >> 12
   R_RISCV_GOT_HI20,
 
   /// PC relative call by PLT
   ///
enum EdgeKind_riscv : Edge::Kind {
   /// Fixup expression
   ///   Fixup <- (Target - *{2}Fixup - Addend)
   R_RISCV_SUB16,
 
   /// 8 bits label subtraction
   ///
   /// Fixup expression
   ///   Fixup <- (Target - *{1}Fixup - Addend)
-  R_RISCV_SUB8
+  R_RISCV_SUB8,
 };
 
 /// Returns a string name for the given riscv edge. For debugging purposes
 /// only
 const char *getEdgeKindName(Edge::Kind K);
 } // namespace riscv
 } // namespace jitlink
 } // namespace llvm
 
 #endif

llvm/lib/ExecutionEngine/JITLink/ELFLinkGraphBuilder.h

if (Sym.isDefined() &&
       }
       if (auto *B = getGraphBlock(Shndx)) {
         LLVM_DEBUG({
           dbgs() << "      " << SymIndex
                  << ": Creating defined graph symbol for ELF symbol \"" << *Name
                  << "\"\n";
         });
 
-        // Model the section symbols as anonymous symbol.
+        // In RISCV, temporary symbols (Used to generate dwarf, eh_frame
+        // sections...) will appear in object code's symbol table, and LLVM does
+        // not use names on these temporary symbols (RISCV gnu toolchain uses
+        // names on these temporary symbols). If the symbol is unnamed, add an
+        // anonymous symbol.
         auto &GSym =
-            Sym.getType() == ELF::STT_SECTION
-                ? G->addAnonymousSymbol(*B, Sym.getValue(), Sym.st_size, false,
-                                        false)
+            Name->empty()
+                ? G->addAnonymousSymbol(*B, Sym.getValue(), Sym.st_size,
+                                        false, false)
                 : G->addDefinedSymbol(*B, Sym.getValue(), *Name, Sym.st_size, L,
                                       S, Sym.getType() == ELF::STT_FUNC, false);
         setGraphSymbol(SymIndex, GSym);
       }
     } else if (Sym.isUndefined() && Sym.isExternal()) {
       LLVM_DEBUG({
         dbgs() << "      " << SymIndex
                << ": Creating external graph symbol for ELF symbol \"" << *Name

llvm/lib/ExecutionEngine/JITLink/ELF_riscv.cpp

Error applyFixup(LinkGraph &G, Block &B, const Edge &E) const {
     }
     case R_RISCV_SUB8: {
       int64_t Value =
           *(reinterpret_cast<const uint8_t *>(FixupAddress.getValue())) -
           E.getTarget().getAddress().getValue() - E.getAddend();
       *FixupPtr = static_cast<uint8_t>(Value);
       break;
     }
+    case R_RISCV_32_PCREL: {
+      // FIXME: It seems that R_RISCV_32_PCREL relocation will only appear in debug sections
+      // like eh_frame section. Currently, because of eh_frame will not be processed in JITLink's RISCV
+      // backend, test this relocation is difficult, so here report error if needs to fixup this relocation
+      return make_error<JITLinkError>("Fixup of relocation type R_RISCV_32_PCREL is not supportted");
+    }
     }
     return Error::success();
   }
 };
 
 template <typename ELFT>
 class ELFLinkGraphBuilder_riscv : public ELFLinkGraphBuilder<ELFT> {
 private:
getRelocationKind(const uint32_t Type) {
     case ELF::R_RISCV_SUB64:
       return EdgeKind_riscv::R_RISCV_SUB64;
     case ELF::R_RISCV_SUB32:
       return EdgeKind_riscv::R_RISCV_SUB32;
     case ELF::R_RISCV_SUB16:
       return EdgeKind_riscv::R_RISCV_SUB16;
     case ELF::R_RISCV_SUB8:
       return EdgeKind_riscv::R_RISCV_SUB8;
+    case ELF::R_RISCV_32_PCREL:
+      return EdgeKind_riscv::R_RISCV_32_PCREL;
     }
 
     return make_error<JITLinkError>("Unsupported riscv relocation:" +
                                     formatv("{0:d}", Type));
   }
 
   Error addRelocations() override {
     LLVM_DEBUG(dbgs() << "Processing relocations:\n");

llvm/lib/ExecutionEngine/JITLink/riscv.cpp

const char *getEdgeKindName(Edge::Kind K) {
   case R_RISCV_PCREL_HI20:
     return "R_RISCV_PCREL_HI20";
   case R_RISCV_PCREL_LO12_I:
     return "R_RISCV_PCREL_LO12_I";
   case R_RISCV_PCREL_LO12_S:
     return "R_RISCV_PCREL_LO12_S";
   case R_RISCV_CALL:
     return "R_RISCV_CALL";
+  case R_RISCV_32_PCREL:
+    return "R_RISCV_32_PCREL";
   case R_RISCV_ADD64:
     return "R_RISCV_ADD64";
   case R_RISCV_ADD32:
     return "R_RISCV_ADD32";
   case R_RISCV_ADD16:
     return "R_RISCV_ADD16";
   case R_RISCV_ADD8:
     return "R_RISCV_ADD8";

llvm/test/ExecutionEngine/JITLink/RISCV/anonymous_symbol.s

+# RUN: llvm-mc -triple=riscv64 -filetype=obj -o %t %s
+# RUN: llvm-jitlink -debug-only=jitlink -noexec %t 2>&1 | FileCheck %s
+#
+# Because of the exist of cfi directive, sections like eh_frame section will be emitted
+# in llvm's object code emission phase. Anonymous symbols will also be emitted to indicate
+# the section start and section end. So that by relocating these symbol, the section length
+# can be calculated.
+#
+# CHECK: Creating defined graph symbol for ELF symbol ""
+# CHECK: Creating defined graph symbol for ELF symbol ""
+        .text
+        .globl main
+        .p2align 2
+        .type main,@function
+main:
+        .cfi_startproc
+        ret
+        .Lfunc_end0:
+        .size main, .Lfunc_end0-main
+        .cfi_endproc