[PowerPC, Mach-O] PPCMachObjectWriter.cpp, at long last

David Fang fang at csl.cornell.edu
Thu Aug 1 13:00:51 PDT 2013


Hi,
 	Attached is another revision of the patch.  I think I've 
addressed your points.  Does this look better to you?

Fang

On Wed, 31 Jul 2013, Rafael EspĂ­ndola wrote:

> + * PPC relocation types from <mach-o/ppc/reloc.h>
> + * (renamed, following conventions in this header)
>
> That file is under a different license, so you cannot just copy and
> past from it.
>
>
>    llvm_unreachable("Relocation emission for MachO/PPC64
> unimplemented, until Fang gets around to hacking...");
>
> report_fatal_error is probably better. We normally don't list devs in
> the source, so just "Relocation emission for MachO/PPC64
> unimplemented".
>
>
> +       Log2Size is used for relocation_info::r_length.
> +       See "include/llvm/MC/MCFixup.h"
> +       and "lib/Target/PowerPC/MCTargetDesc/PPCFixupKinds.h"
> +       See "lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp":
> +               adjustFixupValue() for cases of handling fixups.
> +               getFixupKindInfo() for sizes.
>
> Just saying "computes the log2 of the size of the relocation" is
> probably better.
>
> +      llvm_unreachable("Unimplemented (relative)");
>
> If a user can get here, than report_fatal_error is better. If it is
> something that should not happen, keep the llvm_unrechable, but change
> the message :-)
>
> +// inline
>
> remove this comment.
>
>
>
> +// experimenting with order (endian fishiness on PPC?)
>
> What does this mean?
>
> +/**
>
> We normally use /// for comments before a function
>
> +     * odcctools-20090808:as/write_object.c:
>
> please don't copy this.
>
> +#if 0
>
> No commented out code please.
>
> +#if 1
>
> Please remove.
>
> +//      llvm_unreachable("Unhandled PPC scattered relocation type.");
>
> No commented out code.
>
>
> Please clang format this patch. With git you can just run
> clang/tools/clang-format/git-clang-format.
>
> +; This test is paired with test/CodeGen/PowerPC/hello-reloc.s,
> +; which tests llvm-mc.
>
> Please remove.
>
> Please remove the use of tee from the run lines.
>
> +; RUN-WORKS-BUT-SKIPPING: llc -filetype=obj -relocation-model=pic
> -mcpu=g4 -mtriple=powerpc-apple-darwin8 %s -o - | tee %t2 | macho-dump
> | tee %t3 | FileCheck -check-prefix=DARWIN-G4-DUMP %s
>
> Please remove.
>
> +; FIXME: validating .s->.o requires darwin asm syntax support in PPCAsmParser
> +; RUN-XFAIL: llvm-mc -relocation-model=pic -mcpu=g4
> -triple=powerpc-apple-darwin8 %t1 -o - | tee %t4 | macho-dump | tee
> %t5 | FileCheck -check-prefix=DARWIN-G4-DUMP %s
> +; RUN-XFAIL: diff -u %t2 %t4 || diff -u %t3 %t5
>
> Please remove. Again, the idea of llvm-mc tests is *not* to feed it
> the output of llc. That is a valid way to start creating a test, but
> what should be checked in is a reduced testcase.
>
> Please remove all the DARWIN-G4-DUMP lines since they are not being used.
>
> What is this test testing for? It should not be testing for every
> assembly line printed by llc. In fact, this patch doesn't change the
> printed assembly, does it? If the patch just changes the output of
> llvm-mc, it should only have an llvm-mc test.
>
>
> Cheers,
> Rafael
>

-- 
David Fang
http://www.csl.cornell.edu/~fang/
-------------- next part --------------
diff --git a/include/llvm/Object/MachOFormat.h b/include/llvm/Object/MachOFormat.h
index 96ee8a7..e938b1e 100644
--- a/include/llvm/Object/MachOFormat.h
+++ b/include/llvm/Object/MachOFormat.h
@@ -425,6 +425,36 @@ namespace macho {
 
   };
 
+  /// PPC relocation types from <mach-o/ppc/reloc.h>
+  enum RelocationInfoTypePPC {
+    RIT_PPC_VANILLA, /* generic relocation as discribed above */
+    RIT_PPC_PAIR,    /* the second relocation entry of a pair */
+    RIT_PPC_BR14,    /* 14 bit branch displacement (to a word address) */
+    RIT_PPC_BR24,    /* 24 bit branch displacement (to a word address) */
+    RIT_PPC_HI16,    /* a PAIR follows with the low half */
+    RIT_PPC_LO16,    /* a PAIR follows with the high half */
+    RIT_PPC_HA16,    /* Same as the RELOC_HI16 except the low 16 bits and the
+                     * high 16 bits are added together with the low 16 bits
+                     * sign extened first.  This means if bit 15 of the low
+                     * 16 bits is set the high 16 bits stored in the
+                     * instruction will be adjusted.
+                     */
+    RIT_PPC_LO14, /* Same as the LO16 except that the low 2 bits are not
+                     * stored in the instruction and are always zero.  This
+                     * is used in double word load/store instructions.
+                     */
+    RIT_PPC_SECTDIFF,      /* a PAIR follows with subtract symbol value */
+    RIT_PPC_PB_LA_PTR,     /* prebound lazy pointer */
+    RIT_PPC_HI16_SECTDIFF, /* section difference forms of above.  a PAIR */
+    RIT_PPC_LO16_SECTDIFF, /* follows these with subtract symbol value */
+    RIT_PPC_HA16_SECTDIFF,
+    RIT_PPC_JBSR,
+    RIT_PPC_LO14_SECTDIFF,
+    RIT_PPC_LOCAL_SECTDIFF, /* like PPC_SECTDIFF, but the symbol
+                                referenced was local.  */
+    RIT_PPC_TLV
+  };
+
 } // end namespace macho
 
 } // end namespace object
diff --git a/lib/Object/MachOObjectFile.cpp b/lib/Object/MachOObjectFile.cpp
index 19c8421..5218b05 100644
--- a/lib/Object/MachOObjectFile.cpp
+++ b/lib/Object/MachOObjectFile.cpp
@@ -1059,7 +1059,8 @@ MachOObjectFile::getRelocationValueString(DataRefImpl Rel,
         break;
     }
   // X86 and ARM share some relocation types in common.
-  } else if (Arch == Triple::x86 || Arch == Triple::arm) {
+  } else if (Arch == Triple::x86 || Arch == Triple::arm ||
+             Arch == Triple::ppc) {
     // Generic relocation types...
     switch (Type) {
       case macho::RIT_Pair: // GENERIC_RELOC_PAIR - prints no info
@@ -1084,7 +1085,7 @@ MachOObjectFile::getRelocationValueString(DataRefImpl Rel,
       }
     }
 
-    if (Arch == Triple::x86) {
+    if (Arch == Triple::x86 || Arch == Triple::ppc) {
       // All X86 relocations that need special printing were already
       // handled in the generic code.
       switch (Type) {
@@ -1177,7 +1178,7 @@ MachOObjectFile::getRelocationHidden(DataRefImpl Rel, bool &Result) const {
 
   // On arches that use the generic relocations, GENERIC_RELOC_PAIR
   // is always hidden.
-  if (Arch == Triple::x86 || Arch == Triple::arm) {
+  if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) {
     if (Type == macho::RIT_Pair) Result = true;
   } else if (Arch == Triple::x86_64) {
     // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows
diff --git a/lib/Target/PowerPC/MCTargetDesc/CMakeLists.txt b/lib/Target/PowerPC/MCTargetDesc/CMakeLists.txt
index 45be471..3efa5ec 100644
--- a/lib/Target/PowerPC/MCTargetDesc/CMakeLists.txt
+++ b/lib/Target/PowerPC/MCTargetDesc/CMakeLists.txt
@@ -5,6 +5,7 @@ add_llvm_library(LLVMPowerPCDesc
   PPCMCCodeEmitter.cpp
   PPCMCExpr.cpp
   PPCPredicates.cpp
+  PPCMachObjectWriter.cpp
   PPCELFObjectWriter.cpp
   )
 
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp b/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp
index b2a8701..9cd6ce4 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCAsmBackend.cpp
@@ -20,6 +20,7 @@
 #include "llvm/Support/ELF.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TargetRegistry.h"
+
 using namespace llvm;
 
 static uint64_t adjustFixupValue(unsigned Kind, uint64_t Value) {
@@ -69,19 +70,6 @@ static unsigned getFixupKindNumBytes(unsigned Kind) {
 }
 
 namespace {
-class PPCMachObjectWriter : public MCMachObjectTargetWriter {
-public:
-  PPCMachObjectWriter(bool Is64Bit, uint32_t CPUType,
-                      uint32_t CPUSubtype)
-    : MCMachObjectTargetWriter(Is64Bit, CPUType, CPUSubtype) {}
-
-  void RecordRelocation(MachObjectWriter *Writer,
-                        const MCAssembler &Asm, const MCAsmLayout &Layout,
-                        const MCFragment *Fragment, const MCFixup &Fixup,
-                        MCValue Target, uint64_t &FixedValue) {
-    llvm_unreachable("Relocation emission for MachO/PPC unimplemented!");
-  }
-};
 
 class PPCAsmBackend : public MCAsmBackend {
 const Target &TheTarget;
@@ -174,12 +162,11 @@ namespace {
 
     MCObjectWriter *createObjectWriter(raw_ostream &OS) const {
       bool is64 = getPointerSize() == 8;
-      return createMachObjectWriter(new PPCMachObjectWriter(
-                                      /*Is64Bit=*/is64,
-                                      (is64 ? object::mach::CTM_PowerPC64 :
-                                       object::mach::CTM_PowerPC),
-                                      object::mach::CSPPC_ALL),
-                                    OS, /*IsLittleEndian=*/false);
+      return createPPCMachObjectWriter(
+          OS,
+          /*Is64Bit=*/is64,
+          (is64 ? object::mach::CTM_PowerPC64 : object::mach::CTM_PowerPC),
+          object::mach::CSPPC_ALL);
     }
 
     virtual bool doesSectionRequireSymbols(const MCSection &Section) const {
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h b/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h
index 9f29132..16c3cb4 100644
--- a/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCMCTargetDesc.h
@@ -46,6 +46,10 @@ MCAsmBackend *createPPCAsmBackend(const Target &T, StringRef TT, StringRef CPU);
 MCObjectWriter *createPPCELFObjectWriter(raw_ostream &OS,
                                          bool Is64Bit,
                                          uint8_t OSABI);
+/// createPPCELFObjectWriter - Construct a PPC Mach-O object writer.
+MCObjectWriter *createPPCMachObjectWriter(raw_ostream &OS, bool Is64Bit,
+                                          uint32_t CPUType,
+                                          uint32_t CPUSubtype);
 } // End llvm namespace
 
 // Generated files will use "namespace PPC". To avoid symbol clash,
diff --git a/lib/Target/PowerPC/MCTargetDesc/PPCMachObjectWriter.cpp b/lib/Target/PowerPC/MCTargetDesc/PPCMachObjectWriter.cpp
new file mode 100644
index 0000000..f0dcf33
--- /dev/null
+++ b/lib/Target/PowerPC/MCTargetDesc/PPCMachObjectWriter.cpp
@@ -0,0 +1,384 @@
+//===-- PPCMachObjectWriter.cpp - PPC Mach-O Writer -----------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "MCTargetDesc/PPCMCTargetDesc.h"
+#include "MCTargetDesc/PPCFixupKinds.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/MC/MCAsmLayout.h"
+#include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCMachObjectWriter.h"
+#include "llvm/MC/MCSectionMachO.h"
+#include "llvm/MC/MCValue.h"
+#include "llvm/Object/MachOFormat.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/Format.h"
+
+using namespace llvm;
+using namespace llvm::object;
+
+namespace {
+class PPCMachObjectWriter : public MCMachObjectTargetWriter {
+  bool RecordScatteredRelocation(MachObjectWriter *Writer,
+                                 const MCAssembler &Asm,
+                                 const MCAsmLayout &Layout,
+                                 const MCFragment *Fragment,
+                                 const MCFixup &Fixup, MCValue Target,
+                                 unsigned Log2Size, uint64_t &FixedValue);
+
+  void RecordPPCRelocation(MachObjectWriter *Writer, const MCAssembler &Asm,
+                           const MCAsmLayout &Layout,
+                           const MCFragment *Fragment, const MCFixup &Fixup,
+                           MCValue Target, uint64_t &FixedValue);
+
+public:
+  PPCMachObjectWriter(bool Is64Bit, uint32_t CPUType, uint32_t CPUSubtype)
+      : MCMachObjectTargetWriter(Is64Bit, CPUType, CPUSubtype,
+                                 /*UseAggressiveSymbolFolding=*/Is64Bit) {}
+
+  void RecordRelocation(MachObjectWriter *Writer, const MCAssembler &Asm,
+                        const MCAsmLayout &Layout, const MCFragment *Fragment,
+                        const MCFixup &Fixup, MCValue Target,
+                        uint64_t &FixedValue) {
+    if (Writer->is64Bit()) {
+      report_fatal_error("Relocation emission for MachO/PPC64 unimplemented.");
+    } else
+      RecordPPCRelocation(Writer, Asm, Layout, Fragment, Fixup, Target,
+                          FixedValue);
+  }
+};
+}
+
+/// computes the log2 of the size of the relocation,
+/// used for relocation_info::r_length.
+static unsigned getFixupKindLog2Size(unsigned Kind) {
+  switch (Kind) {
+  default:
+    report_fatal_error("log2size(FixupKind): Unhandled fixup kind!");
+  case FK_PCRel_1:
+  case FK_Data_1:
+    return 0;
+  case FK_PCRel_2:
+  case FK_Data_2:
+    return 1;
+  case FK_PCRel_4:
+  case PPC::fixup_ppc_brcond14:
+  case PPC::fixup_ppc_half16:
+  case PPC::fixup_ppc_br24:
+  case FK_Data_4:
+    return 2;
+  case FK_PCRel_8:
+  case FK_Data_8:
+    return 3;
+  }
+  return 0;
+}
+
+/// Translates generic PPC fixup kind to Mach-O/PPC relocation type enum.
+/// Outline based on PPCELFObjectWriter::getRelocTypeInner().
+static unsigned getRelocType(const MCValue &Target,
+                             const MCFixupKind FixupKind, // from
+                                                          // Fixup.getKind()
+                             const bool IsPCRel) {
+  const MCSymbolRefExpr::VariantKind Modifier =
+      Target.isAbsolute() ? MCSymbolRefExpr::VK_None
+                          : Target.getSymA()->getKind();
+  // determine the type of the relocation
+  unsigned Type = macho::RIT_PPC_VANILLA;
+  if (IsPCRel) { // relative to PC
+    switch (FixupKind) {
+    default:
+      report_fatal_error("Unimplemented fixup kind (relative)");
+    case PPC::fixup_ppc_br24:
+      Type = macho::RIT_PPC_BR24; // R_PPC_REL24
+      break;
+    case PPC::fixup_ppc_brcond14:
+      Type = macho::RIT_PPC_BR14;
+      break;
+    case PPC::fixup_ppc_half16:
+      switch (Modifier) {
+      default:
+        llvm_unreachable("Unsupported modifier for half16 fixup");
+      case MCSymbolRefExpr::VK_PPC_HA:
+        Type = macho::RIT_PPC_HA16;
+        break;
+      case MCSymbolRefExpr::VK_PPC_LO:
+        Type = macho::RIT_PPC_LO16;
+        break;
+      case MCSymbolRefExpr::VK_PPC_HI:
+        Type = macho::RIT_PPC_HI16;
+        break;
+      }
+      break;
+    }
+  } else {
+    switch (FixupKind) {
+    default:
+      report_fatal_error("Unimplemented fixup kind (absolute)!");
+    case PPC::fixup_ppc_half16:
+      switch (Modifier) {
+      default:
+        llvm_unreachable("Unsupported modifier for half16 fixup");
+      case MCSymbolRefExpr::VK_PPC_HA:
+        Type = macho::RIT_PPC_HA16_SECTDIFF;
+        break;
+      case MCSymbolRefExpr::VK_PPC_LO:
+        Type = macho::RIT_PPC_LO16_SECTDIFF;
+        break;
+      case MCSymbolRefExpr::VK_PPC_HI:
+        Type = macho::RIT_PPC_HI16_SECTDIFF;
+        break;
+      }
+      break;
+    case FK_Data_4:
+      break;
+    case FK_Data_2:
+      break;
+    }
+  }
+  return Type;
+}
+
+static void makeRelocationInfo(macho::RelocationEntry &MRE,
+                               const uint32_t FixupOffset, const uint32_t Index,
+                               const unsigned IsPCRel, const unsigned Log2Size,
+                               const unsigned IsExtern, const unsigned Type) {
+  MRE.Word0 = FixupOffset;
+  // The bitfield offsets that work (as determined by trial-and-error)
+  // are different than what is documented in the mach-o manuals.
+  // Is this an endianness issue w/ PPC?
+  MRE.Word1 = ((Index << 8) |    // was << 0
+               (IsPCRel << 7) |  // was << 24
+               (Log2Size << 5) | // was << 25
+               (IsExtern << 4) | // was << 27
+               (Type << 0));     // was << 28
+}
+
+static void
+makeScatteredRelocationInfo(macho::RelocationEntry &MRE, const uint32_t Addr,
+                            const unsigned Type, const unsigned Log2Size,
+                            const unsigned IsPCRel, const uint32_t Value2) {
+  // FIXME: see <mach-o/reloc.h> for note on endianness
+  MRE.Word0 = ((Addr << 0) | (Type << 24) | (Log2Size << 28) | (IsPCRel << 30) |
+               macho::RF_Scattered);
+  MRE.Word1 = Value2;
+}
+
+/// Compute fixup offset (address).
+static uint32_t getFixupOffset(const MCAsmLayout &Layout,
+                               const MCFragment *Fragment,
+                               const MCFixup &Fixup) {
+  uint32_t FixupOffset = Layout.getFragmentOffset(Fragment) + Fixup.getOffset();
+  // On Mach-O, ppc_fixup_half16 relocations must refer to the
+  // start of the instruction, not the second halfword, as ELF does
+  if (Fixup.getKind() == PPC::fixup_ppc_half16)
+    FixupOffset &= ~uint32_t(3);
+  return FixupOffset;
+}
+
+/// \return false if falling back to using non-scattered relocation,
+/// otherwise true for normal scattered relocation.
+/// based on X86MachObjectWriter::RecordScatteredRelocation
+/// and ARMMachObjectWriter::RecordScatteredRelocation
+bool PPCMachObjectWriter::RecordScatteredRelocation(
+    MachObjectWriter *Writer, const MCAssembler &Asm, const MCAsmLayout &Layout,
+    const MCFragment *Fragment, const MCFixup &Fixup, MCValue Target,
+    unsigned Log2Size, uint64_t &FixedValue) {
+  // caller already computes these, can we just pass and reuse?
+  const uint32_t FixupOffset = getFixupOffset(Layout, Fragment, Fixup);
+  const MCFixupKind FK = Fixup.getKind();
+  const unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, FK);
+  const unsigned Type = getRelocType(Target, FK, IsPCRel);
+
+  // Is this a local or SECTDIFF relocation entry?
+  // SECTDIFF relocation entries have symbol subtractions,
+  // and require two entries, the first for the add-symbol value,
+  // the second for the subtract-symbol value.
+
+  // See <reloc.h>.
+  const MCSymbol *A = &Target.getSymA()->getSymbol();
+  MCSymbolData *A_SD = &Asm.getSymbolData(*A);
+
+  if (!A_SD->getFragment())
+    report_fatal_error("symbol '" + A->getName() +
+                       "' can not be undefined in a subtraction expression");
+
+  uint32_t Value = Writer->getSymbolAddress(A_SD, Layout);
+  uint64_t SecAddr =
+      Writer->getSectionAddress(A_SD->getFragment()->getParent());
+  FixedValue += SecAddr;
+  uint32_t Value2 = 0;
+
+  if (const MCSymbolRefExpr *B = Target.getSymB()) {
+    MCSymbolData *B_SD = &Asm.getSymbolData(B->getSymbol());
+
+    if (!B_SD->getFragment())
+      report_fatal_error("symbol '" + B->getSymbol().getName() +
+                         "' can not be undefined in a subtraction expression");
+
+    // FIXME: is Type correct? see include/llvm/Object/MachOFormat.h
+    Value2 = Writer->getSymbolAddress(B_SD, Layout);
+    FixedValue -= Writer->getSectionAddress(B_SD->getFragment()->getParent());
+  }
+  // FIXME: does FixedValue get used??
+
+  // Relocations are written out in reverse order, so the PAIR comes first.
+  if (Type == macho::RIT_PPC_SECTDIFF || Type == macho::RIT_PPC_HI16_SECTDIFF ||
+      Type == macho::RIT_PPC_LO16_SECTDIFF ||
+      Type == macho::RIT_PPC_HA16_SECTDIFF ||
+      Type == macho::RIT_PPC_LO14_SECTDIFF ||
+      Type == macho::RIT_PPC_LOCAL_SECTDIFF) {
+    // X86 had this piece, but ARM does not
+    // If the offset is too large to fit in a scattered relocation,
+    // we're hosed. It's an unfortunate limitation of the MachO format.
+    if (FixupOffset > 0xffffff) {
+      char Buffer[32];
+      format("0x%x", FixupOffset).print(Buffer, sizeof(Buffer));
+      Asm.getContext().FatalError(Fixup.getLoc(),
+                                  Twine("Section too large, can't encode "
+                                        "r_address (") +
+                                      Buffer + ") into 24 bits of scattered "
+                                               "relocation entry.");
+      llvm_unreachable("fatal error returned?!");
+    }
+
+    // Is this supposed to follow MCTarget/PPCAsmBackend.cpp:adjustFixupValue()?
+    // see PPCMCExpr::EvaluateAsRelocatableImpl()
+    uint32_t other_half = 0;
+    switch (Type) {
+    case macho::RIT_PPC_LO16_SECTDIFF:
+      other_half = (FixedValue >> 16) & 0xffff;
+      // applyFixupOffset longer extracts the high part because it now assumes
+      // this was already done.
+      // It looks like this is not true for the FixedValue needed with Mach-O
+      // relocs.
+      // So we need to adjust FixedValue again here.
+      FixedValue &= 0xffff;
+      break;
+    case macho::RIT_PPC_HA16_SECTDIFF:
+      other_half = FixedValue & 0xffff;
+      FixedValue =
+          ((FixedValue >> 16) + ((FixedValue & 0x8000) ? 1 : 0)) & 0xffff;
+      break;
+    case macho::RIT_PPC_HI16_SECTDIFF:
+      other_half = FixedValue & 0xffff;
+      FixedValue = (FixedValue >> 16) & 0xffff;
+      break;
+    default:
+      llvm_unreachable("Invalid PPC scattered relocation type.");
+      break;
+    }
+
+    macho::RelocationEntry MRE;
+    makeScatteredRelocationInfo(MRE, other_half, macho::RIT_PPC_PAIR, Log2Size,
+                                IsPCRel, Value2);
+    Writer->addRelocation(Fragment->getParent(), MRE);
+  } else {
+    // If the offset is more than 24-bits, it won't fit in a scattered
+    // relocation offset field, so we fall back to using a non-scattered
+    // relocation. This is a bit risky, as if the offset reaches out of
+    // the block and the linker is doing scattered loading on this
+    // symbol, things can go badly.
+    //
+    // Required for 'as' compatibility.
+    if (FixupOffset > 0xffffff)
+      return false;
+  }
+  macho::RelocationEntry MRE;
+  makeScatteredRelocationInfo(MRE, FixupOffset, Type, Log2Size, IsPCRel, Value);
+  Writer->addRelocation(Fragment->getParent(), MRE);
+  return true;
+}
+
+// see PPCELFObjectWriter for a general outline of cases
+void PPCMachObjectWriter::RecordPPCRelocation(
+    MachObjectWriter *Writer, const MCAssembler &Asm, const MCAsmLayout &Layout,
+    const MCFragment *Fragment, const MCFixup &Fixup, MCValue Target,
+    uint64_t &FixedValue) {
+  const MCFixupKind FK = Fixup.getKind(); // unsigned
+  const unsigned Log2Size = getFixupKindLog2Size(FK);
+  const bool IsPCRel = Writer->isFixupKindPCRel(Asm, FK);
+  const unsigned RelocType = getRelocType(Target, FK, IsPCRel);
+
+  // If this is a difference or a defined symbol plus an offset, then we need a
+  // scattered relocation entry. Differences always require scattered
+  // relocations.
+  if (Target.getSymB() &&
+      // Q: are branch targets ever scattered?
+      RelocType != macho::RIT_PPC_BR24 && RelocType != macho::RIT_PPC_BR14) {
+    RecordScatteredRelocation(Writer, Asm, Layout, Fragment, Fixup, Target,
+                              Log2Size, FixedValue);
+    return;
+  }
+
+  // this doesn't seem right for RIT_PPC_BR24
+  // Get the symbol data, if any.
+  MCSymbolData *SD = 0;
+  if (Target.getSymA())
+    SD = &Asm.getSymbolData(Target.getSymA()->getSymbol());
+
+  // See <reloc.h>.
+  const uint32_t FixupOffset = getFixupOffset(Layout, Fragment, Fixup);
+  unsigned Index = 0;
+  unsigned IsExtern = 0;
+  unsigned Type = RelocType;
+
+  if (Target.isAbsolute()) { // constant
+                             // SymbolNum of 0 indicates the absolute section.
+                             //
+    // FIXME: Currently, these are never generated (see code below). I cannot
+    // find a case where they are actually emitted.
+    report_fatal_error("FIXME: relocations to absolute targets "
+                       "not yet implemented");
+    // the above line stolen from ARM, not sure
+  } else {
+    // Resolve constant variables.
+    if (SD->getSymbol().isVariable()) {
+      int64_t Res;
+      if (SD->getSymbol().getVariableValue()->EvaluateAsAbsolute(
+              Res, Layout, Writer->getSectionAddressMap())) {
+        FixedValue = Res;
+        return;
+      }
+    }
+
+    // Check whether we need an external or internal relocation.
+    if (Writer->doesSymbolRequireExternRelocation(SD)) {
+      IsExtern = 1;
+      Index = SD->getIndex();
+      // For external relocations, make sure to offset the fixup value to
+      // compensate for the addend of the symbol address, if it was
+      // undefined. This occurs with weak definitions, for example.
+      if (!SD->Symbol->isUndefined())
+        FixedValue -= Layout.getSymbolOffset(SD);
+    } else {
+      // The index is the section ordinal (1-based).
+      const MCSectionData &SymSD =
+          Asm.getSectionData(SD->getSymbol().getSection());
+      Index = SymSD.getOrdinal() + 1;
+      FixedValue += Writer->getSectionAddress(&SymSD);
+    }
+    if (IsPCRel)
+      FixedValue -= Writer->getSectionAddress(Fragment->getParent());
+  }
+
+  // struct relocation_info (8 bytes)
+  macho::RelocationEntry MRE;
+  makeRelocationInfo(MRE, FixupOffset, Index, IsPCRel, Log2Size, IsExtern,
+                     Type);
+  Writer->addRelocation(Fragment->getParent(), MRE);
+}
+
+MCObjectWriter *llvm::createPPCMachObjectWriter(raw_ostream &OS, bool Is64Bit,
+                                                uint32_t CPUType,
+                                                uint32_t CPUSubtype) {
+  return createMachObjectWriter(
+      new PPCMachObjectWriter(Is64Bit, CPUType, CPUSubtype), OS,
+      /*IsLittleEndian=*/false);
+}
diff --git a/test/CodeGen/PowerPC/hello-reloc.s b/test/CodeGen/PowerPC/hello-reloc.s
new file mode 100644
index 0000000..042534e
--- /dev/null
+++ b/test/CodeGen/PowerPC/hello-reloc.s
@@ -0,0 +1,253 @@
+; I took the asm produced by llc -filetype=asm hello-reloc.ll and 
+; syntactically translated it to the supported PPCAsmParser syntax for 
+; this case.
+; The asm transformation will no longer be needed once darwin-asm syntax 
+; is supported in PPCAsmParser, at which point these tests can be combined.
+
+; RUN: llvm-mc -filetype=obj -relocation-model=pic -mcpu=g4 -triple=powerpc-apple-darwin8 %s -o - | macho-dump | FileCheck -check-prefix=DARWIN-G4-DUMP %s
+
+;	.machine ppc7400
+	.section	__TEXT,__textcoal_nt,coalesced,pure_instructions
+	.section	__TEXT,__picsymbolstub1,symbol_stubs,pure_instructions,32
+	.section	__TEXT,__text,regular,pure_instructions
+	.globl	_main
+	.align	4
+_main:                                  ; @main
+; BB#0:                                 ; %entry
+	mflr 0
+	stw 31, -4(1)
+	stw 0, 8(1)
+	stwu 1, -80(1)
+	bl L0$pb
+L0$pb:
+	mr 31, 1
+	li 5, 0
+	mflr 2
+	stw 3, 68(31)
+	stw 5, 72(31)
+	stw 4, 64(31)
+	addis 2, 2, (L_.str-L0$pb)@ha
+	la 3, (L_.str-L0$pb)@l(2)
+	bl L_puts$stub
+	li 3, 0
+	addi 1, 1, 80
+	lwz 0, 8(1)
+	lwz 31, -4(1)
+	mtlr 0
+	blr
+
+	.section	__TEXT,__picsymbolstub1,symbol_stubs,pure_instructions,32
+	.align	4
+L_puts$stub:
+	.indirect_symbol	_puts
+	mflr 0
+	bcl 20, 31, L_puts$stub$tmp
+L_puts$stub$tmp:
+	mflr 11
+	addis 11, 11, (L_puts$lazy_ptr-L_puts$stub$tmp)@ha
+	mtlr 0
+	lwzu 12, (L_puts$lazy_ptr-L_puts$stub$tmp)@l(11)
+	mtctr 12
+	bctr
+	.section	__DATA,__la_symbol_ptr,lazy_symbol_pointers
+L_puts$lazy_ptr:
+	.indirect_symbol	_puts
+	.long	dyld_stub_binding_helper
+
+.subsections_via_symbols
+	.section	__TEXT,__cstring,cstring_literals
+L_.str:                                 ; @.str
+	.asciz	 "Hello, world!"
+
+
+; DARWIN-G4-DUMP: ('cputype', 18)
+; DARWIN-G4-DUMP: ('cpusubtype', 0)
+; DARWIN-G4-DUMP: ('filetype', 1)
+; DARWIN-G4-DUMP: ('num_load_commands', 3)
+; DARWIN-G4-DUMP: ('load_commands_size', 500)
+; DARWIN-G4-DUMP: ('flag', 8192)
+; DARWIN-G4-DUMP: ('load_commands', [
+; DARWIN-G4-DUMP: # Load Command 0
+; DARWIN-G4-DUMP: (('command', 1)
+; DARWIN-G4-DUMP: ('size', 396)
+; DARWIN-G4-DUMP: ('segment_name', '\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00')
+; DARWIN-G4-DUMP: ('vm_addr', 0)
+; DARWIN-G4-DUMP: ('vm_size', 130)
+; DARWIN-G4-DUMP: ('file_offset', 528)
+; DARWIN-G4-DUMP: ('file_size', 130)
+; DARWIN-G4-DUMP: ('maxprot', 7)
+; DARWIN-G4-DUMP: ('initprot', 7)
+; DARWIN-G4-DUMP: ('num_sections', 5)
+; DARWIN-G4-DUMP: ('flags', 0)
+; DARWIN-G4-DUMP: ('sections', [
+; DARWIN-G4-DUMP: # Section 0
+; DARWIN-G4-DUMP: (('section_name', '__text\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00')
+; DARWIN-G4-DUMP: ('segment_name', '__TEXT\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00')
+; DARWIN-G4-DUMP: ('address', 0)
+; DARWIN-G4-DUMP: ('size', 80)
+; DARWIN-G4-DUMP: ('offset', 528)
+; DARWIN-G4-DUMP: ('alignment', 4)
+; DARWIN-G4-DUMP: ('reloc_offset', 660)
+; DARWIN-G4-DUMP: ('num_reloc', 5)
+; DARWIN-G4-DUMP: ('flags', 0x80000400)
+; DARWIN-G4-DUMP: ('reserved1', 0)
+; DARWIN-G4-DUMP: ('reserved2', 0)
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: ('_relocations', [
+; DARWIN-G4-DUMP: # Relocation 0
+; DARWIN-G4-DUMP: (('word-0', 0x34),
+; DARWIN-G4-DUMP: ('word-1', 0x3c3)),
+; DARWIN-G4-DUMP: # Relocation 1
+; DARWIN-G4-DUMP: (('word-0', 0xab000030),
+; DARWIN-G4-DUMP: ('word-1', 0x74)),
+; DARWIN-G4-DUMP: # Relocation 2
+; DARWIN-G4-DUMP: (('word-0', 0xa1000000),
+; DARWIN-G4-DUMP: ('word-1', 0x14)),
+; DARWIN-G4-DUMP: # Relocation 3
+; DARWIN-G4-DUMP: (('word-0', 0xac00002c),
+; DARWIN-G4-DUMP: ('word-1', 0x74)),
+; DARWIN-G4-DUMP: # Relocation 4
+; DARWIN-G4-DUMP: (('word-0', 0xa1000060),
+; DARWIN-G4-DUMP: ('word-1', 0x14)),
+; DARWIN-G4-DUMP: ])
+; DARWIN-G4-DUMP: # Section 1
+; DARWIN-G4-DUMP: (('section_name', '__textcoal_nt\x00\x00\x00')
+; DARWIN-G4-DUMP: ('segment_name', '__TEXT\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00')
+; DARWIN-G4-DUMP: ('address', 80)
+; DARWIN-G4-DUMP: ('size', 0)
+; DARWIN-G4-DUMP: ('offset', 608)
+; DARWIN-G4-DUMP: ('alignment', 0)
+; DARWIN-G4-DUMP: ('reloc_offset', 0)
+; DARWIN-G4-DUMP: ('num_reloc', 0)
+; DARWIN-G4-DUMP: ('flags', 0x8000000b)
+; DARWIN-G4-DUMP: ('reserved1', 0)
+; DARWIN-G4-DUMP: ('reserved2', 0)
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: ('_relocations', [
+; DARWIN-G4-DUMP: ])
+; DARWIN-G4-DUMP: # Section 2
+; DARWIN-G4-DUMP: (('section_name', '__picsymbolstub1')
+; DARWIN-G4-DUMP: ('segment_name', '__TEXT\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00')
+; DARWIN-G4-DUMP: ('address', 80)
+; DARWIN-G4-DUMP: ('size', 32)
+; DARWIN-G4-DUMP: ('offset', 608)
+; DARWIN-G4-DUMP: ('alignment', 4)
+; DARWIN-G4-DUMP: ('reloc_offset', 700)
+; DARWIN-G4-DUMP: ('num_reloc', 4)
+; DARWIN-G4-DUMP: ('flags', 0x80000408)
+; DARWIN-G4-DUMP: ('reserved1', 0)
+; DARWIN-G4-DUMP: ('reserved2', 32)
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: ('_relocations', [
+; DARWIN-G4-DUMP: # Relocation 0
+; DARWIN-G4-DUMP: (('word-0', 0xab000014),
+; DARWIN-G4-DUMP: ('word-1', 0x70)),
+; DARWIN-G4-DUMP: # Relocation 1
+; DARWIN-G4-DUMP: (('word-0', 0xa1000000),
+; DARWIN-G4-DUMP: ('word-1', 0x58)),
+; DARWIN-G4-DUMP: # Relocation 2
+; DARWIN-G4-DUMP: (('word-0', 0xac00000c),
+; DARWIN-G4-DUMP: ('word-1', 0x70)),
+; DARWIN-G4-DUMP: # Relocation 3
+; DARWIN-G4-DUMP: (('word-0', 0xa1000018),
+; DARWIN-G4-DUMP: ('word-1', 0x58)),
+; DARWIN-G4-DUMP: ])
+; DARWIN-G4-DUMP: # Section 3
+; DARWIN-G4-DUMP: (('section_name', '__la_symbol_ptr\x00')
+; DARWIN-G4-DUMP: ('segment_name', '__DATA\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00')
+; DARWIN-G4-DUMP: ('address', 112)
+; DARWIN-G4-DUMP: ('size', 4)
+; DARWIN-G4-DUMP: ('offset', 640)
+; DARWIN-G4-DUMP: ('alignment', 0)
+; DARWIN-G4-DUMP: ('reloc_offset', 732)
+; DARWIN-G4-DUMP: ('num_reloc', 1)
+; DARWIN-G4-DUMP: ('flags', 0x7)
+; DARWIN-G4-DUMP: ('reserved1', 1)
+; DARWIN-G4-DUMP: ('reserved2', 0)
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: ('_relocations', [
+; DARWIN-G4-DUMP: # Relocation 0
+; DARWIN-G4-DUMP: (('word-0', 0x0),
+; DARWIN-G4-DUMP: ('word-1', 0x250)),
+; DARWIN-G4-DUMP: ])
+; DARWIN-G4-DUMP: # Section 4
+; DARWIN-G4-DUMP: (('section_name', '__cstring\x00\x00\x00\x00\x00\x00\x00')
+; DARWIN-G4-DUMP: ('segment_name', '__TEXT\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00')
+; DARWIN-G4-DUMP: ('address', 116)
+; DARWIN-G4-DUMP: ('size', 14)
+; DARWIN-G4-DUMP: ('offset', 644)
+; DARWIN-G4-DUMP: ('alignment', 0)
+; DARWIN-G4-DUMP: ('reloc_offset', 0)
+; DARWIN-G4-DUMP: ('num_reloc', 0)
+; DARWIN-G4-DUMP: ('flags', 0x2)
+; DARWIN-G4-DUMP: ('reserved1', 0)
+; DARWIN-G4-DUMP: ('reserved2', 0)
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: ('_relocations', [
+; DARWIN-G4-DUMP: ])
+; DARWIN-G4-DUMP: ])
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: # Load Command 1
+; DARWIN-G4-DUMP: (('command', 2)
+; DARWIN-G4-DUMP: ('size', 24)
+; DARWIN-G4-DUMP: ('symoff', 748)
+; DARWIN-G4-DUMP: ('nsyms', 3)
+; DARWIN-G4-DUMP: ('stroff', 784)
+; DARWIN-G4-DUMP: ('strsize', 40)
+; DARWIN-G4-DUMP: ('_string_data', '\x00_main\x00dyld_stub_binding_helper\x00_puts\x00\x00\x00')
+; DARWIN-G4-DUMP: ('_symbols', [
+; DARWIN-G4-DUMP: # Symbol 0
+; DARWIN-G4-DUMP: (('n_strx', 1)
+; DARWIN-G4-DUMP: ('n_type', 0xf)
+; DARWIN-G4-DUMP: ('n_sect', 1)
+; DARWIN-G4-DUMP: ('n_desc', 0)
+; DARWIN-G4-DUMP: ('n_value', 0)
+; DARWIN-G4-DUMP: ('_string', '_main')
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: # Symbol 1
+; DARWIN-G4-DUMP: (('n_strx', 32)
+; DARWIN-G4-DUMP: ('n_type', 0x1)
+; DARWIN-G4-DUMP: ('n_sect', 0)
+; DARWIN-G4-DUMP: ('n_desc', 1)
+; DARWIN-G4-DUMP: ('n_value', 0)
+; DARWIN-G4-DUMP: ('_string', '_puts')
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: # Symbol 2
+; DARWIN-G4-DUMP: (('n_strx', 7)
+; DARWIN-G4-DUMP: ('n_type', 0x1)
+; DARWIN-G4-DUMP: ('n_sect', 0)
+; DARWIN-G4-DUMP: ('n_desc', 0)
+; DARWIN-G4-DUMP: ('n_value', 0)
+; DARWIN-G4-DUMP: ('_string', 'dyld_stub_binding_helper')
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: ])
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: # Load Command 2
+; DARWIN-G4-DUMP: (('command', 11)
+; DARWIN-G4-DUMP: ('size', 80)
+; DARWIN-G4-DUMP: ('ilocalsym', 0)
+; DARWIN-G4-DUMP: ('nlocalsym', 0)
+; DARWIN-G4-DUMP: ('iextdefsym', 0)
+; DARWIN-G4-DUMP: ('nextdefsym', 1)
+; DARWIN-G4-DUMP: ('iundefsym', 1)
+; DARWIN-G4-DUMP: ('nundefsym', 2)
+; DARWIN-G4-DUMP: ('tocoff', 0)
+; DARWIN-G4-DUMP: ('ntoc', 0)
+; DARWIN-G4-DUMP: ('modtaboff', 0)
+; DARWIN-G4-DUMP: ('nmodtab', 0)
+; DARWIN-G4-DUMP: ('extrefsymoff', 0)
+; DARWIN-G4-DUMP: ('nextrefsyms', 0)
+; DARWIN-G4-DUMP: ('indirectsymoff', 740)
+; DARWIN-G4-DUMP: ('nindirectsyms', 2)
+; DARWIN-G4-DUMP: ('extreloff', 0)
+; DARWIN-G4-DUMP: ('nextrel', 0)
+; DARWIN-G4-DUMP: ('locreloff', 0)
+; DARWIN-G4-DUMP: ('nlocrel', 0)
+; DARWIN-G4-DUMP: ('_indirect_symbols', [
+; DARWIN-G4-DUMP: # Indirect Symbol 0
+; DARWIN-G4-DUMP: (('symbol_index', 0x1),),
+; DARWIN-G4-DUMP: # Indirect Symbol 1
+; DARWIN-G4-DUMP: (('symbol_index', 0x1),),
+; DARWIN-G4-DUMP: ])
+; DARWIN-G4-DUMP: ),
+; DARWIN-G4-DUMP: ])


More information about the llvm-commits mailing list