[llvm-commits] [PATCH] Teach llvm-objdump to dump Win64 exception tables

Kai kai at redstar.de
Fri Nov 2 10:56:40 PDT 2012


Hi!

I re-used a test case which generates unwind info with llvm-mc. During 
construction of the test I noticed some small problems which I fixed. I 
also silenced some GCC warnings and added decoding of the flags field. 
The attached patches are updated accordingly.

@Michael
I reused the content of the MC/Object/seh.s test. While studying the 
generated tables I noticed that there is a UOP_PushMachFrame generated 
as last unwind code in the table. Is this correct? I can't related this 
unwind code to the .seh_* operations in the source. The entry order 
seems also wrong. From the first function:

Unwind Codes:
     0x00: UOP_SetFPReg
     0x0f: UOP_PushNonVol RBX
     0x0e: UOP_SaveXMM128 XMM8 [0x0000]
     0x09: UOP_SaveNonVol RSI [0x0010]
     0x04: UOP_AllocSmall 24
     0x00: UOP_PushMachFrame w/o error code

 From the source I would expect the following:

Unwind Codes:
     0x0f: UOP_PushNonVol RBX
     0x0e: UOP_SaveXMM128 XMM8 [0x0000]
     0x09: UOP_SaveNonVol RSI [0x0010]
     0x04: UOP_AllocSmall 24
     0x00: UOP_SetFPReg

Regards
Kai

On 02.11.2012 15:31, João Matos wrote:
> On Fri, Nov 2, 2012 at 1:33 PM, Kai
> <kai at redstar.de
> <mailto:kai at redstar.de>> wrote:
>
>     I am going to place a test into folder test/Other (hopefully it fits
>     there) but I still have trouble to create some good input. Do you
>     have a suggestion how to do it?
>
>
> We discussed this once on IRC and IIRC the usage of the COFF YAML tools
> was suggested so there would be no need to commit binary COFF files.
>
> --
> João Matos
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>

-------------- next part --------------
diff --git a/test/Other/llvm-objdump-unwind.s b/test/Other/llvm-objdump-unwind.s
new file mode 100644
index 0000000..1e4c742
--- /dev/null
+++ b/test/Other/llvm-objdump-unwind.s
@@ -0,0 +1,106 @@
+// This test checks that the unwind data is dumped by llvm-objdump.
+// RUN: llvm-mc -triple x86_64-pc-win32 -filetype=obj %s | llvm-objdump -u - | FileCheck %s
+
+// CHECK:      Unwind info:
+// CHECK:      Function Table:
+// CHECK-NEXT: Start Address: .text
+// CHECK-NEXT: End Address: .text + 0x001b
+// CHECK-NEXT: Unwind Info Address: .xdata
+// CHECK-NEXT: Version: 1
+// CHECK-NEXT: Flags: 1 UNW_ExceptionHandler
+// CHECK-NEXT: Size of prolog: 18
+// CHECK-NEXT: Number of Codes: 8
+// CHECK-NEXT: Frame register: RBX
+// CHECK-NEXT: Frame offset: 0
+// CHECK-NEXT: Unwind Codes:
+// CHECK-NEXT: 0x00: UOP_SetFPReg
+// CHECK-NEXT: 0x0f: UOP_PushNonVol RBX
+// CHECK-NEXT: 0x0e: UOP_SaveXMM128 XMM8 [0x0000]
+// CHECK-NEXT: 0x09: UOP_SaveNonVol RSI [0x0010]
+// CHECK-NEXT: 0x04: UOP_AllocSmall 24
+// CHECK-NEXT: 0x00: UOP_PushMachFrame w/o error code
+// CHECK:      Function Table:
+// CHECK-NEXT: Start Address: .text + 0x0012
+// CHECK-NEXT: End Address: .text + 0x0012
+// CHECK-NEXT: Unwind Info Address: .xdata + 0x001c
+// CHECK-NEXT: Version: 1
+// CHECK-NEXT: Flags: 4 UNW_ChainInfo
+// CHECK-NEXT: Size of prolog: 0
+// CHECK-NEXT: Number of Codes: 0
+// CHECK-NEXT: No frame pointer used
+// CHECK:      Function Table:
+// CHECK-NEXT: Start Address: .text + 0x001b
+// CHECK-NEXT: End Address: .text + 0x001c
+// CHECK-NEXT: Unwind Info Address: .xdata + 0x002c
+// CHECK-NEXT: Version: 1
+// CHECK-NEXT: Flags: 0
+// CHECK-NEXT: Size of prolog: 0
+// CHECK-NEXT: Number of Codes: 0
+// CHECK-NEXT: No frame pointer used
+// CHECK:      Function Table:
+// CHECK-NEXT: Start Address: .text + 0x001c
+// CHECK-NEXT: End Address: .text + 0x0039
+// CHECK-NEXT: Unwind Info Address: .xdata + 0x0034
+// CHECK-NEXT: Version: 1
+// CHECK-NEXT: Flags: 0
+// CHECK-NEXT: Size of prolog: 14
+// CHECK-NEXT: Number of Codes: 6
+// CHECK-NEXT: No frame pointer used
+// CHECK-NEXT: Unwind Codes:
+// CHECK-NEXT: 0x0e: UOP_AllocLarge 8454128
+// CHECK-NEXT: 0x07: UOP_AllocLarge 8190
+// CHECK-NEXT: 0x00: UOP_PushMachFrame w/o error code
+
+    .text
+    .globl func
+    .def func; .scl 2; .type 32; .endef
+    .seh_proc func
+func:
+    .seh_pushframe @code
+    subq $24, %rsp
+    .seh_stackalloc 24
+    movq %rsi, 16(%rsp)
+    .seh_savereg %rsi, 16
+    movups %xmm8, (%rsp)
+    .seh_savexmm %xmm8, 0
+    pushq %rbx
+    .seh_pushreg 3
+    mov %rsp, %rbx
+    .seh_setframe 3, 0
+    .seh_endprologue
+    .seh_handler __C_specific_handler, @except
+    .seh_handlerdata
+    .long 0
+    .text
+    .seh_startchained
+    .seh_endprologue
+    .seh_endchained
+    lea (%rbx), %rsp
+    pop %rbx
+    addq $24, %rsp
+    ret
+    .seh_endproc
+
+// Test emission of small functions.
+    .globl smallFunc
+    .def smallFunc; .scl 2; .type 32; .endef
+    .seh_proc smallFunc
+smallFunc:
+    ret
+    .seh_endproc
+
+// Function with big stack allocation.
+    .globl smallFunc
+    .def allocFunc; .scl 2; .type 32; .endef
+    .seh_proc smallFunc
+allocFunc:
+    .seh_pushframe @code
+    subq $65520, %rsp
+    .seh_stackalloc 65520
+    sub $8454128, %rsp
+    .seh_stackalloc 8454128
+    .seh_endprologue
+    add $8454128, %rsp
+    addq $65520, %rsp
+    ret
+    .seh_endproc
diff --git a/tools/llvm-objdump/COFFDump.cpp b/tools/llvm-objdump/COFFDump.cpp
new file mode 100644
index 0000000..6237355
--- /dev/null
+++ b/tools/llvm-objdump/COFFDump.cpp
@@ -0,0 +1,310 @@
+//===-- llvm-objdump.cpp - Object file dumping utility for llvm -----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the COFF-specific dumper for llvm-objdump.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm-objdump.h"
+#include "llvm/Object/COFF.h"
+#include "llvm/Object/ObjectFile.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/SourceMgr.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/system_error.h"
+#include "llvm/Support/Win64EH.h"
+#include <algorithm>
+#include <cstring>
+using namespace llvm;
+using namespace object;
+
+namespace {
+  using namespace llvm::Win64EH;
+
+  llvm::raw_ostream &writeHexNumber(llvm::raw_ostream &Out,
+                                    unsigned long long N) {
+    if (N >= 10)
+      Out << "0x";
+    Out.write_hex(N);
+    return Out;
+  }
+
+  StringRef GetCOFFUnwindCodeTypeName(uint8_t Code) {
+    switch(Code) {
+    default: llvm_unreachable("Invalid unwind code");
+    case UOP_PushNonVol: return "UOP_PushNonVol";
+    case UOP_AllocLarge: return "UOP_AllocLarge";
+    case UOP_AllocSmall: return "UOP_AllocSmall";
+    case UOP_SetFPReg: return "UOP_SetFPReg";
+    case UOP_SaveNonVol: return "UOP_SaveNonVol";
+    case UOP_SaveNonVolBig: return "UOP_SaveNonVolBig";
+    case UOP_SaveXMM128: return "UOP_SaveXMM128";
+    case UOP_SaveXMM128Big: return "UOP_SaveXMM128Big";
+    case UOP_PushMachFrame: return "UOP_PushMachFrame";
+    }
+  }
+
+  StringRef GetCOFFUnwindRegisterName(uint8_t Reg) {
+    switch(Reg) {
+    default: llvm_unreachable("Invalid register");
+    case 0: return "RAX";
+    case 1: return "RCX";
+    case 2: return "RDX";
+    case 3: return "RBX";
+    case 4: return "RSP";
+    case 5: return "RBP";
+    case 6: return "RSI";
+    case 7: return "RDI";
+    case 8: return "R8";
+    case 9: return "R9";
+    case 10: return "R10";
+    case 11: return "R11";
+    case 12: return "R12";
+    case 13: return "R13";
+    case 14: return "R14";
+    case 15: return "R15";
+    }
+  }
+
+  void PrintCOFFUnwindCode(const UnwindCode* UnwindCodes, unsigned numCodes) {
+    for (unsigned i = 0; i < numCodes; i++) {
+      const UnwindCode& UC = UnwindCodes[i];
+      outs() <<  format("    0x%02x: ", unsigned(UC.u.codeOffset))
+             << GetCOFFUnwindCodeTypeName(UC.u.unwindOp);
+      switch (UC.u.unwindOp) {
+      case UOP_PushNonVol:
+        outs() << " " << GetCOFFUnwindRegisterName(UC.u.opInfo);
+        break;
+      case UOP_AllocLarge: {
+        uint32_t size;
+        if (UC.u.opInfo == 0) {
+          size = UnwindCodes[++i].frameOffset;
+        }
+        else {
+          size = UnwindCodes[i+1].frameOffset
+                 + (((uint32_t) UnwindCodes[i+2].frameOffset) << 16);
+          i += 2;
+        }
+        outs() << " " << size;
+        break;
+      }
+      case UOP_AllocSmall:
+        outs() << " " << ((UC.u.opInfo+1) * 8);
+        break;
+      case UOP_SetFPReg:
+        outs() << " ";
+        break;
+      case UOP_SaveNonVol:
+        outs() << " " << GetCOFFUnwindRegisterName(UC.u.opInfo)
+               << format(" [0x%04x]", 8 * UnwindCodes[++i].frameOffset);
+        break;
+      case UOP_SaveNonVolBig: {
+        uint32_t ofs = UnwindCodes[i+1].frameOffset
+                        + (((uint32_t) UnwindCodes[i+2].frameOffset) << 16);
+        i += 2;
+        outs() << " " << GetCOFFUnwindRegisterName(UC.u.opInfo)
+               << format(" [0x%08x]", ofs);
+        break;
+      }
+      case UOP_SaveXMM128:
+        outs() << " XMM" << static_cast<uint32_t>(UC.u.opInfo)
+               << format(" [0x%04x]", 16 * UnwindCodes[++i].frameOffset);
+        break;
+      case UOP_SaveXMM128Big: {
+        uint32_t ofs = UnwindCodes[i+1].frameOffset
+                       + (((uint32_t) UnwindCodes[i+2].frameOffset) << 16);
+        i += 2;
+        outs() << " XMM" << UC.u.opInfo << format(" [0x%08x]", ofs);
+        break;
+      }
+      case UOP_PushMachFrame:
+        outs() << " " << (UC.u.opInfo ? "w/o" : "w") << " error code";
+        break;
+      }
+      outs() << "\n";
+    }
+  }
+
+  error_code ResolveCOFFRelocation(const COFFObjectFile* o,
+                                   const SymbolRef& sym,
+                                   const coff_section*& ResolvedSection,
+                                   uint64_t& ResolvedAddr) {
+    error_code ec;
+    if ((ec = sym.getAddress(ResolvedAddr))) return ec;
+    section_iterator iter(o->begin_sections());
+    if ((ec = sym.getSection(iter))) return ec;
+    ResolvedSection = o->getCOFFSection(iter);
+    return object_error::success;
+  }
+
+  error_code GetCOFFUnwindInfo(const COFFObjectFile* o,
+                               const std::vector<RelocationRef> Rels,
+                               uint64_t offset, ArrayRef<uint8_t>& Contents,
+                               uint64_t& addr) {
+    for (std::vector<RelocationRef>::const_iterator R = Rels.begin();
+                                                    R != Rels.end(); R++) {
+      error_code ec;
+      uint64_t ofs;
+      if ((ec = R->getOffset(ofs))) return ec;
+      if (ofs == offset) {
+        SymbolRef sym;
+        if ((ec = R->getSymbol(sym))) return ec;
+        const coff_section* section;
+        ResolveCOFFRelocation(o, sym, section, addr);
+        if ((ec = o->getSectionContents(section, Contents))) return ec;
+        break;
+      }
+    }
+    return object_error::success;
+  }
+
+  error_code GetCOFFSymbol(const std::vector<RelocationRef> Rels,
+                           uint64_t offset, StringRef& name) {
+    for (std::vector<RelocationRef>::const_iterator R = Rels.begin();
+                                                    R != Rels.end(); R++) {
+      error_code ec;
+      uint64_t ofs;
+      if ((ec = R->getOffset(ofs))) return ec;
+      if (ofs == offset) {
+        SymbolRef sym;
+        if ((ec = R->getSymbol(sym))) return ec;
+        if ((ec = sym.getName(name))) return ec;
+        break;
+      }
+    }
+    return object_error::success;
+  }
+
+  void PrintCOFFSymbolAddress(llvm::raw_ostream &Out,
+                              const std::vector<RelocationRef> Rels,
+                              uint64_t offset, uint32_t disp) {
+      StringRef sym;
+      GetCOFFSymbol(Rels, offset, sym);
+      Out << sym;
+      if (disp > 0)
+        Out << format(" + 0x%04x", disp);
+  }
+}
+
+void llvm::PrintCOFFUnwindInfo(const COFFObjectFile* o) {
+  const coff_file_header *header;
+  if (error(o->getHeader(header))) return;
+
+  if (header->Machine != COFF::IMAGE_FILE_MACHINE_AMD64) {
+    errs() << "Unsupported image machine type "
+              "(currently only AMD64 is supported).\n";
+    return;
+  }
+
+  const coff_section* pdata = 0;
+
+  error_code ec;
+  for (section_iterator si = o->begin_sections(),
+                            se = o->end_sections();
+                            si != se; si.increment(ec)) {
+    if (error(ec)) return;
+
+    StringRef Name;
+    if (error(si->getName(Name))) continue;
+
+    if (Name.compare(".pdata") == 0) {
+      pdata = o->getCOFFSection(si);
+      std::vector<RelocationRef> Rels;
+      for (relocation_iterator ri = si->begin_relocations(),
+                               re = si->end_relocations();
+                               ri != re; ri.increment(ec)) {
+        if (error(ec)) break;
+        Rels.push_back(*ri);
+      }
+
+      // Sort relocations by address.
+      std::sort(Rels.begin(), Rels.end(), RelocAddressLess);
+
+      ArrayRef<uint8_t> Contents;
+      if (error(o->getSectionContents(pdata, Contents))) continue;
+      if (Contents.empty()) continue;
+
+      unsigned i = 0;
+      while ((Contents.size() - i) >= sizeof(RuntimeFunction)) {
+        const RuntimeFunction* RF = 
+            reinterpret_cast<const RuntimeFunction*>(Contents.data() + i);
+
+        outs() << "Function Table:\n";
+
+        outs() << "  Start Address: ";
+        PrintCOFFSymbolAddress(outs(), Rels,
+                               i + offsetof(RuntimeFunction, startAddress),
+                               RF->startAddress);
+        outs() << "\n";
+
+        outs() << "  End Address: ";
+        PrintCOFFSymbolAddress(outs(), Rels,
+                               i + offsetof(RuntimeFunction, endAddress),
+                               RF->endAddress);
+        outs() << "\n";
+
+        outs() << "  Unwind Info Address: ";
+        PrintCOFFSymbolAddress(outs(), Rels,
+                               i + offsetof(RuntimeFunction, unwindInfoOffset),
+                               RF->unwindInfoOffset);
+        outs() << "\n";
+
+        ArrayRef<uint8_t> XContents;
+        uint64_t UnwindInfoOffset = 0;
+        if (error(GetCOFFUnwindInfo(o, Rels,
+                                    i + offsetof(RuntimeFunction, unwindInfoOffset),
+                                    XContents, UnwindInfoOffset))) continue;
+        if (XContents.empty()) continue;
+
+        UnwindInfoOffset += RF->unwindInfoOffset;
+        if (UnwindInfoOffset > XContents.size()) continue;
+
+        const Win64EH::UnwindInfo* UI = reinterpret_cast<const Win64EH::UnwindInfo*>
+                             (XContents.data() + UnwindInfoOffset);
+
+        outs() << "  Version: " << (int) UI->version << "\n";
+        outs() << "  Flags: " << (int) UI->flags;
+        if (UI->flags) {
+            if (UI->flags & UNW_ExceptionHandler) outs() << " UNW_ExceptionHandler";
+            if (UI->flags & UNW_TerminateHandler) outs() << " UNW_TerminateHandler";
+            if (UI->flags & UNW_ChainInfo) outs() << " UNW_ChainInfo";
+        }
+        outs() << "\n";
+        outs() << "  Size of prolog: " << (int) UI->prologSize << "\n";
+        outs() << "  Number of Codes: " << (int) UI->numCodes << "\n";
+        // Maybe this should move to output of UOP_SetFPReg?
+        if (UI->frameRegister) {
+          outs() << "  Frame register: "
+                 << GetCOFFUnwindRegisterName(UI->frameRegister) << "\n";
+          outs() << "  Frame offset: " << 16 * (int) UI->frameOffset << "\n";
+        }
+        else {
+          outs() << "  No frame pointer used\n";
+        }
+        if (UI->flags & (UNW_ExceptionHandler | UNW_TerminateHandler)) {
+          // FIXME: Output exception handler data
+        }
+        else if (UI->flags & UNW_ChainInfo) {
+          // FIXME: Output chained unwind info
+        }
+
+        if (UI->numCodes)
+          outs() << "  Unwind Codes:\n";
+
+        PrintCOFFUnwindCode(&UI->unwindCodes[0], UI->numCodes);
+
+        outs() << "\n\n";
+        outs().flush();
+
+        i += sizeof(RuntimeFunction);
+      }
+      continue;
+    }
+  }
+}
diff --git a/tools/llvm-objdump/llvm-objdump.cpp b/tools/llvm-objdump/llvm-objdump.cpp
index 13ea4e3..1028b48 100644
--- a/tools/llvm-objdump/llvm-objdump.cpp
+++ b/tools/llvm-objdump/llvm-objdump.cpp
@@ -100,9 +100,16 @@ MAttrs("mattr",
   cl::desc("Target specific attributes"),
   cl::value_desc("a1,+a2,-a3,..."));
 
+static cl::opt<bool>
+UnwindInfo("unwind-info", cl::desc("Display unwind information"));
+
+static cl::alias
+UnwindInfoShort("u", cl::desc("Alias for --unwind-info"),
+                cl::aliasopt(::UnwindInfo));
+
 static StringRef ToolName;
 
-static bool error(error_code ec) {
+bool llvm::error(error_code ec) {
   if (!ec) return false;
 
   outs() << ToolName << ": error reading file: " << ec.message() << ".\n";
@@ -161,7 +168,7 @@ void llvm::DumpBytes(StringRef bytes) {
   outs() << output;
 }
 
-static bool RelocAddressLess(RelocationRef a, RelocationRef b) {
+bool llvm::RelocAddressLess(RelocationRef a, RelocationRef b) {
   uint64_t a_addr, b_addr;
   if (error(a.getAddress(a_addr))) return false;
   if (error(b.getAddress(b_addr))) return false;
@@ -566,6 +573,19 @@ static void PrintSymbolTable(const ObjectFile *o) {
   }
 }
 
+static void PrintUnwindInfo(const ObjectFile *o) {
+  outs() << "Unwind info:\n\n";
+
+  if (const COFFObjectFile *coff = dyn_cast<const COFFObjectFile>(o)) {
+    PrintCOFFUnwindInfo(coff);
+  } else {
+    // TODO: Extract DWARF dump tool to objdump.
+    errs() << "This operation is only currently supported "
+              "for COFF object files.\n";
+    return;
+  }
+}
+
 static void DumpObject(const ObjectFile *o) {
   outs() << '\n';
   outs() << o->getFileName()
@@ -581,6 +601,8 @@ static void DumpObject(const ObjectFile *o) {
     PrintSectionContents(o);
   if (SymbolTable)
     PrintSymbolTable(o);
+  if (::UnwindInfo)
+    PrintUnwindInfo(o);
 }
 
 /// @brief Dump each object file in \a a;
@@ -659,7 +681,8 @@ int main(int argc, char **argv) {
       && !Relocations
       && !SectionHeaders
       && !SectionContents
-      && !SymbolTable) {
+      && !SymbolTable
+      && !::UnwindInfo) {
     cl::PrintHelpMessage();
     return 2;
   }
diff --git a/tools/llvm-objdump/llvm-objdump.h b/tools/llvm-objdump/llvm-objdump.h
index aa71b77..bf3d422 100644
--- a/tools/llvm-objdump/llvm-objdump.h
+++ b/tools/llvm-objdump/llvm-objdump.h
@@ -17,12 +17,21 @@
 
 namespace llvm {
 
+namespace object {
+    class COFFObjectFile;
+    class RelocationRef;
+}
+class error_code;
+
 extern cl::opt<std::string> TripleName;
 extern cl::opt<std::string> ArchName;
 
 // Various helper functions.
+bool error(error_code ec);
+bool RelocAddressLess(object::RelocationRef a, object::RelocationRef b);
 void DumpBytes(StringRef bytes);
 void DisassembleInputMachO(StringRef Filename);
+void PrintCOFFUnwindInfo(const object::COFFObjectFile* o);
 
 class StringRefMemoryObject : public MemoryObject {
   virtual void anchor();
-------------- next part --------------
diff --git a/include/llvm/Support/Win64EH.h b/include/llvm/Support/Win64EH.h
index 8d74e10..fb8aeae 100644
--- a/include/llvm/Support/Win64EH.h
+++ b/include/llvm/Support/Win64EH.h
@@ -60,9 +60,9 @@ enum {
 
 /// RuntimeFunction - An entry in the table of functions with unwind info.
 struct RuntimeFunction {
-  uint64_t startAddress;
-  uint64_t endAddress;
-  uint64_t unwindInfoOffset;
+  uint32_t startAddress;
+  uint32_t endAddress;
+  uint32_t unwindInfoOffset;
 };
 
 /// UnwindInfo - An entry in the exception table.
@@ -78,17 +78,17 @@ struct UnwindInfo {
   void *getLanguageSpecificData() {
     return reinterpret_cast<void *>(&unwindCodes[(numCodes+1) & ~1]);
   }
-  uint64_t getLanguageSpecificHandlerOffset() {
-    return *reinterpret_cast<uint64_t *>(getLanguageSpecificData());
+  uint32_t getLanguageSpecificHandlerOffset() {
+    return *reinterpret_cast<uint32_t *>(getLanguageSpecificData());
   }
-  void setLanguageSpecificHandlerOffset(uint64_t offset) {
+  void setLanguageSpecificHandlerOffset(uint32_t offset) {
     *reinterpret_cast<uint64_t *>(getLanguageSpecificData()) = offset;
   }
   RuntimeFunction *getChainedFunctionEntry() {
     return reinterpret_cast<RuntimeFunction *>(getLanguageSpecificData());
   }
   void *getExceptionData() {
-    return reinterpret_cast<void *>(reinterpret_cast<uint64_t *>(
+    return reinterpret_cast<void *>(reinterpret_cast<uint32_t *>(
                                                   getLanguageSpecificData())+1);
   }
 };


More information about the llvm-commits mailing list