[lld] Add target support for SystemZ (s390x)

This patch adds full support for linking SystemZ (ELF s390x) object
files.  Support should be generally complete:
- All relocation types are supported.
- Full shared library support (DYNAMIC, GOT, PLT, ifunc).
- Relaxation of TLS and GOT relocations where appropriate.
- Platform-specific test cases.

In addition to new platform code and the obvious changes, there were
a few additional changes to common code:

- Add three new RelExpr members (R_GOTPLT_OFF, R_GOTPLT_PC, and
  R_PLT_GOTREL) needed to support certain s390x relocations.
  I chose not to use a platform-specific name since nothing in
  the definition of these relocs is actually platform-specific;
  it is well possible that other platforms will need the same.

- A couple of tweaks to TLS relocation handling, as the particular
  semantics of the s390x versions differ slightly.  See comments
  in the code.

This was tested by building and testing >1500 Fedora packages,
with only a handful of failures; as these also have issues when
building with LLD on other architectures, they seem unrelated.

Co-authored-by: Tulio Magno Quites Machado Filho <[email protected]>

Author: uweigand

lld/ELF/Arch/SystemZ.cpp

@@ -33,6 +33,7 @@ add_lld_library(lldELF
   Arch/PPC64.cpp
   Arch/RISCV.cpp
   Arch/SPARCV9.cpp
+  Arch/SystemZ.cpp
   Arch/X86.cpp
   Arch/X86_64.cpp
   ARMErrataFix.cpp

lld/ELF/CMakeLists.txt

@@ -199,6 +199,7 @@ static std::tuple<ELFKind, uint16_t, uint8_t> parseEmulation(StringRef emul) {
           .Case("msp430elf", {ELF32LEKind, EM_MSP430})
           .Case("elf64_amdgpu", {ELF64LEKind, EM_AMDGPU})
           .Case("elf64loongarch", {ELF64LEKind, EM_LOONGARCH})
+          .Case("elf64_s390", {ELF64BEKind, EM_S390})
           .Default({ELFNoneKind, EM_NONE});
 
   if (ret.first == ELFNoneKind)
@@ -1171,7 +1172,7 @@ static bool getIsRela(opt::InputArgList &args) {
   uint16_t m = config->emachine;
   return m == EM_AARCH64 || m == EM_AMDGPU || m == EM_HEXAGON ||
          m == EM_LOONGARCH || m == EM_PPC || m == EM_PPC64 || m == EM_RISCV ||
-         m == EM_X86_64;
+         m == EM_S390 || m == EM_X86_64;
 }
 
 static void parseClangOption(StringRef opt, const Twine &msg) {

lld/ELF/Driver.cpp

@@ -1612,6 +1612,8 @@ static uint16_t getBitcodeMachineKind(StringRef path, const Triple &t) {
     return EM_RISCV;
   case Triple::sparcv9:
     return EM_SPARCV9;
+  case Triple::systemz:
+    return EM_S390;
   case Triple::x86:
     return t.isOSIAMCU() ? EM_IAMCU : EM_386;
   case Triple::x86_64:

lld/ELF/InputFiles.cpp

@@ -652,6 +652,7 @@ static int64_t getTlsTpOffset(const Symbol &s) {
 
     // Variant 2.
   case EM_HEXAGON:
+  case EM_S390:
   case EM_SPARCV9:
   case EM_386:
   case EM_X86_64:
@@ -705,6 +706,8 @@ uint64_t InputSectionBase::getRelocTargetVA(const InputFile *file, RelType type,
   case R_GOT_OFF:
   case R_RELAX_TLS_GD_TO_IE_GOT_OFF:
     return sym.getGotOffset() + a;
+  case R_GOTPLT_OFF:
+    return sym.getGotPltOffset() + a;
   case R_AARCH64_GOT_PAGE_PC:
   case R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC:
     return getAArch64Page(sym.getGotVA() + a) - getAArch64Page(p);
@@ -713,6 +716,8 @@ uint64_t InputSectionBase::getRelocTargetVA(const InputFile *file, RelType type,
   case R_GOT_PC:
   case R_RELAX_TLS_GD_TO_IE:
     return sym.getGotVA() + a - p;
+  case R_GOTPLT_PC:
+    return sym.getGotPltVA() + a - p;
   case R_LOONGARCH_GOT_PAGE_PC:
     if (sym.hasFlag(NEEDS_TLSGD))
       return getLoongArchPageDelta(in.got->getGlobalDynAddr(sym) + a, p);
@@ -804,6 +809,8 @@ uint64_t InputSectionBase::getRelocTargetVA(const InputFile *file, RelType type,
     return getLoongArchPageDelta(sym.getPltVA() + a, p);
   case R_PLT_GOTPLT:
     return sym.getPltVA() + a - in.gotPlt->getVA();
+  case R_PLT_GOTREL:
+    return sym.getPltVA() + a - in.got->getVA();
   case R_PPC32_PLTREL:
     // R_PPC_PLTREL24 uses the addend (usually 0 or 0x8000) to indicate r30
     // stores _GLOBAL_OFFSET_TABLE_ or .got2+0x8000. The addend is ignored for

lld/ELF/InputSection.cpp

@@ -1357,8 +1357,8 @@ static unsigned handleTlsRelocation(RelType type, Symbol &sym,
             R_LOONGARCH_GOT_PAGE_PC, R_GOT_OFF, R_TLSIE_HINT>(expr)) {
     ctx.hasTlsIe.store(true, std::memory_order_relaxed);
     // Initial-Exec relocs can be relaxed to Local-Exec if the symbol is locally
-    // defined.
-    if (toExecRelax && isLocalInExecutable) {
+    // defined.  This is not supported on SystemZ.
+    if (toExecRelax && isLocalInExecutable && config->emachine != EM_S390) {
       c.addReloc({R_RELAX_TLS_IE_TO_LE, type, offset, addend, &sym});
     } else if (expr != R_TLSIE_HINT) {
       sym.setFlags(NEEDS_TLSIE);
@@ -1404,6 +1404,26 @@ template <class ELFT, class RelTy> void RelocationScanner::scanOne(RelTy *&i) {
   if (expr == R_NONE)
     return;
 
+  // Like other platforms, calls to the TLS helper routine on SystemZ carry
+  // two relocations, one for the helper routine itself, and a TLS marker
+  // relocation.  When relaxing the TLS model, the helper routine is no longer
+  // needed, and its relocation should be removed.  Unlike other platforms,
+  // on SystemZ the TLS marker routine typically comes *after* the helper
+  // routine relocation, so the getTlsGdRelaxSkip mechanism used by
+  // handleTlsRelocation does not work on this platform.
+  //
+  // Instead, check for this case here: if we are building a main executable
+  // (i.e. TLS relaxation applies), and the relocation *after* the current one
+  // is a TLS call marker instruction matching the current instruction, then
+  // skip this relocation.
+  if (config->emachine == EM_S390 && !config->shared) {
+    if (i < end && getter.get(i->r_offset) == offset - 2) {
+      RelType nextType = i->getType(/*isMips64EL=*/false);
+      if (nextType == R_390_TLS_GDCALL || nextType == R_390_TLS_LDCALL)
+        return;
+    }
+  }
+
   // Error if the target symbol is undefined. Symbol index 0 may be used by
   // marker relocations, e.g. R_*_NONE and R_ARM_V4BX. Don't error on them.
   if (sym.isUndefined() && symIndex != 0 &&

lld/ELF/Relocations.cpp

@@ -40,11 +40,14 @@ enum RelExpr {
   R_GOTPLT,
   R_GOTPLTREL,
   R_GOTREL,
+  R_GOTPLT_OFF,
+  R_GOTPLT_PC,
   R_NONE,
   R_PC,
   R_PLT,
   R_PLT_PC,
   R_PLT_GOTPLT,
+  R_PLT_GOTREL,
   R_RELAX_HINT,
   R_RELAX_GOT_PC,
   R_RELAX_GOT_PC_NOPIC,

lld/ELF/Relocations.h

@@ -444,6 +444,7 @@ static std::pair<ELFKind, uint16_t> parseBfdName(StringRef s) {
       .Case("elf32-msp430", {ELF32LEKind, EM_MSP430})
       .Case("elf32-loongarch", {ELF32LEKind, EM_LOONGARCH})
       .Case("elf64-loongarch", {ELF64LEKind, EM_LOONGARCH})
+      .Case("elf64-s390", {ELF64BEKind, EM_S390})
       .Default({ELFNoneKind, EM_NONE});
 }
 

lld/ELF/ScriptParser.cpp

@@ -1419,6 +1419,9 @@ DynamicSection<ELFT>::computeContents() {
     case EM_MIPS:
       addInSec(DT_MIPS_PLTGOT, *in.gotPlt);
       break;
+    case EM_S390:
+      addInSec(DT_PLTGOT, *in.got);
+      break;
     case EM_SPARCV9:
       addInSec(DT_PLTGOT, *in.plt);
       break;

lld/ELF/SyntheticSections.cpp

@@ -87,6 +87,8 @@ TargetInfo *elf::getTarget() {
     return getRISCVTargetInfo();
   case EM_SPARCV9:
     return getSPARCV9TargetInfo();
+  case EM_S390:
+    return getSystemZTargetInfo();
   case EM_X86_64:
     return getX86_64TargetInfo();
   }