Reland Implement .eh_frame writer and disassembler.

src/eh-frame.cc

 // Copyright 2016 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/eh-frame.h"
-#include "src/objects-inl.h"
-#include "src/objects.h"
+
+#include <iomanip>
+#include <ostream>
+
+#if V8_TARGET_ARCH_X64
+#include "src/x64/eh-frame-x64.h"
+#elif V8_TARGET_ARCH_ARM
+#include "src/arm/eh-frame-arm.h"
+#elif V8_TARGET_ARCH_ARM64
+#include "src/arm64/eh-frame-arm64.h"
+#else
+
+// Dummy placeholders
 
 namespace v8 {
 namespace internal {
 
-static const int DW_EH_PE_pcrel = 0x10;
-static const int DW_EH_PE_datarel = 0x30;
-static const int DW_EH_PE_udata4 = 0x03;
-static const int DW_EH_PE_sdata4 = 0x0b;
-
-const int EhFrameHdr::kCIESize = 0;
-
-static const int kVersionSize = 1;
-static const int kEncodingSpecifiersSize = 3;
+static const Register kInitialBaseRegister = {Register::kCode_no_reg};
+static const int kInitialBaseOffset = 0;
+static const int kDataAlignmentFactor = 1;
+static const byte kCIE[8] = {0};
+#if ENABLE_DISASSEMBLER
+static const int kInitialStateOffsetInCIE = 0;
+#endif
+
+const char* DwarfRegisterCodeToString(int) {
+  UNIMPLEMENTED();
+  return nullptr;
+}
+
+int RegisterToDwarfCode(Register) {
+  UNIMPLEMENTED();
+  return -1;
+}
+
+}  // namespace internal
+}  // namespace v8
+
+#endif
+
+namespace v8 {
+namespace internal {
+
+STATIC_CONST_MEMBER_DEFINITION const uint32_t EhFrameWriter::kInt32Placeholder;
+
+STATIC_CONST_MEMBER_DEFINITION const int EhFrameWriter::kCIESize = sizeof(kCIE);
+static_assert(EhFrameWriter::kCIESize % kPointerSize == 0,
+              "CIE size must be a multiple of pointer size");
+
+STATIC_CONST_MEMBER_DEFINITION const int EhFrameWriter::kFDEOffset = kCIESize;
+STATIC_CONST_MEMBER_DEFINITION const int
+    EhFrameWriter::kProcedureAddressOffset = kFDEOffset + 2 * kInt32Size;

[rmcilroy, 2016/06/30 15:23:10]

I think it would be better if these were relative to the start of the FDEOffset
(so not taking into account CIE size). That way, you can build the eh_frame, and
keep track of where the FDE starts (in a field), and apply the offsets from
there. That avoids you having to know about sizeof(kCIE) field and make these
STATIC_CONST_MEMBER_DEFINITION (also it will be necessary if you write the CIE
like the FDE as I propose below in eh-frame-x64.h.

[Stefano Sanfilippo, 2016/07/04 18:21:14]

Done.

+STATIC_CONST_MEMBER_DEFINITION const int EhFrameWriter::kProcedureSizeOffset =
+    kFDEOffset + 3 * kInt32Size;
+#ifdef ENABLE_DISASSEMBLER
+STATIC_CONST_MEMBER_DEFINITION const int EhFrameWriter::kFDEDirectivesOffset =

[rmcilroy, 2016/06/30 15:23:10]

This is only used in DisassembleToStream, just define it there

[Stefano Sanfilippo, 2016/07/04 18:21:14]

Done.

+    kFDEOffset + 4 * kInt32Size + 1;
+#endif
+
+const char* DwarfRegisterCodeToString(int code);
+int RegisterToDwarfCode(Register name);

[rmcilroy, 2016/06/30 15:23:10]

these should be methods in EhFrameWriter

[Stefano Sanfilippo, 2016/07/04 18:21:14]

Done.

+
+EhFrameWriter::EhFrameWriter()
+    : last_pc_offset_(0),
+#if DEBUG
+      eh_frame_finalised_(false),
+#endif
+      base_register_(kInitialBaseRegister),
+      base_offset_(kInitialBaseOffset) {
+  WriteBytes(&kCIE[0], sizeof(kCIE));
+  WriteFDEHeader();
+}
+
+void EhFrameWriter::WriteFDEHeader() {
+  DCHECK_EQ(eh_frame_offset(), kFDEOffset);
+  WriteInt32(kInt32Placeholder);      // Placeholder for size of the FDE
+  WriteInt32(kCIESize + kInt32Size);  // Backwards offset to the CIE
+  DCHECK_EQ(eh_frame_offset(), kProcedureAddressOffset);
+  WriteInt32(kInt32Placeholder);  // Placeholder for pointer to procedure
+  DCHECK_EQ(eh_frame_offset(), kProcedureSizeOffset);
+  WriteInt32(kInt32Placeholder);  // Placeholder for size of the procedure
+  WriteByte(0);                   // No augmentation data
+}
+
+void EhFrameWriter::AdvanceLocation(int pc_offset) {
+  DCHECK_GE(pc_offset, last_pc_offset_);
+  uint32_t delta = pc_offset - last_pc_offset_;
+
+  if (delta <= kLocationMask) {
+    WriteByte((kLocationTag << kLocationMaskSize) | (delta & kLocationMask));
+  } else if (delta <= kMaxUInt8) {
+    WriteByte(kAdvanceLoc1);
+    WriteByte(delta);
+  } else if (delta <= kMaxUInt16) {
+    WriteByte(kAdvanceLoc2);
+    WriteInt16(delta);
+  } else {
+    WriteByte(kAdvanceLoc4);
+    WriteInt32(delta);
+  }
+
+  last_pc_offset_ = pc_offset;
+}
+
+void EhFrameWriter::DefineBaseAddressOffset(int base_offset) {
+  DCHECK_GE(base_offset, 0);
+  WriteByte(kDefCFAOffset);
+  WriteULEB128(base_offset);
+  base_offset_ = base_offset;
+}
+
+void EhFrameWriter::DefineBaseAddressRegister(Register name) {
+  int code = RegisterToDwarfCode(name);
+  WriteByte(kDefCFARegister);
+  WriteULEB128(code);
+  base_register_ = name;
+}
+
+void EhFrameWriter::DefineBaseAddressRegisterOffset(Register name,
+                                                    int base_offset) {
+  int code = RegisterToDwarfCode(name);
+  WriteByte(kDefCFA);
+  WriteULEB128(code);
+  WriteULEB128(base_offset);
+  base_offset_ = base_offset;
+  base_register_ = name;
+}
+
+void EhFrameWriter::SaveRegisterToStack(Register name, int offset) {
+  int code = RegisterToDwarfCode(name);
+  DCHECK_GE(code, 0);
+  DCHECK_LE(code, kSavedRegisterMask);
+  DCHECK_EQ(offset % kDataAlignmentFactor, 0);
+  WriteByte((kSavedRegisterTag << kSavedRegisterMaskSize) |
+            (code & kSavedRegisterMask));
+  WriteULEB128(offset / std::abs(kDataAlignmentFactor));
+}
+
+void EhFrameWriter::RegisterIsValid(Register name) {
+  int code = RegisterToDwarfCode(name);
+  DCHECK_GE(code, 0);
+  WriteByte(kSameValue);
+  WriteULEB128(code);
+}
+
+void EhFrameWriter::Finish(int code_size) {
+  DCHECK_GE(eh_frame_buffer_.length(), kCIESize);
+

[rmcilroy, 2016/06/30 15:23:10]

DCHECK(!eh_frame_finalized_)

[Stefano Sanfilippo, 2016/07/04 18:21:14]

Done.

+  // Add padding
+  int unpadded_fde_size = eh_frame_buffer_.length() - kCIESize;
+  int padded_fde_size = RoundUp(unpadded_fde_size, 8);
+  int fde_padding_size = padded_fde_size - unpadded_fde_size;
+
+  static const byte kFDEPadding[] = {kNop, kNop, kNop, kNop,
+                                     kNop, kNop, kNop, kNop};
+  DCHECK_LT(fde_padding_size, static_cast<int>(sizeof(kFDEPadding)));
+  WriteBytes(&kFDEPadding[0], fde_padding_size);
+
+  // Write the size of the FDE now that we know it.
+  PatchInt32(kFDEOffset, padded_fde_size);
+
+  // Write the size and offset to procedure.
+  PatchInt32(kProcedureAddressOffset,
+             -(RoundUp(code_size, 8) + kProcedureAddressOffset));
+  PatchInt32(kProcedureSizeOffset, code_size);
+
+  // Terminate the .eh_frame.
+  static const byte kEhFrameTerminator[kEhFrameTerminatorSize] = {0};
+  WriteBytes(&kEhFrameTerminator[0], kEhFrameTerminatorSize);
+
+  // Write .eh_frame_hdr
+  EhFrameHdr eh_frame_hdr(code_size, eh_frame_offset());
+  WriteBytes(reinterpret_cast<const byte*>(&eh_frame_hdr),
+             EhFrameHdr::kRecordSize);
+
+#if DEBUG
+  eh_frame_finalised_ = true;
+#endif
+}
+
+void EhFrameWriter::GetEhFrame(CodeDesc* desc) {
+  DCHECK(eh_frame_finalised_);
+  desc->unwinding_info_size = eh_frame_buffer_.length();
+  desc->unwinding_info = eh_frame_buffer_.begin();
+}
+
+void EhFrameWriter::WriteULEB128(uint32_t value) {
+  do {
+    byte chunk = value & 0x7f;
+    value >>= 7;
+    if (value != 0) chunk |= 0x80;
+    eh_frame_buffer_.Add(chunk);
+  } while (value != 0);
+}
+
+#ifdef ENABLE_DISASSEMBLER
+
+namespace {
+
+uint32_t DecodeULEB128(const byte* encoded, int* encoded_size) {
+  const byte* cur = encoded;
+  uint32_t decoded_value = 0;
+
+  do {
+    decoded_value <<= 7;
+    decoded_value += static_cast<uint32_t>(static_cast<unsigned>(*cur & 0x7f));
+  } while (*cur++ >= 0x80);
+
+  *encoded_size = static_cast<int>(cur - encoded);
+  return decoded_value;
+}
+
+class StreamModifiersScope final {
+ public:
+  explicit StreamModifiersScope(std::ostream* stream)
+      : stream_(stream), flags_(stream->flags()) {}
+  ~StreamModifiersScope() { stream_->flags(flags_); }
+
+ private:
+  std::ostream* stream_;
+  std::ios::fmtflags flags_;
+};
+
+}  // namespace
+
+// static
+void EhFrameWriter::DumpDWARFDirectives(std::ostream& stream,  // NOLINT
+                                        const byte* begin, const byte* end) {
+  StreamModifiersScope modifiers_scope(&stream);
+
+  const byte* cur = begin;
+  uint32_t offset_in_procedure = 0;
+
+  while (cur != end) {
+    stream << reinterpret_cast<const void*>(cur) << "  ";
+
+    if (((*cur >> kLocationMaskSize) & 0xff) == kLocationTag) {
+      int value = *cur & kLocationMask;
+      cur += sizeof(byte);
+      offset_in_procedure += value;
+      stream << "| pc_offset=" << std::dec << offset_in_procedure
+             << " (delta=0x" << std::hex << value << ")\n";
+      continue;
+    }
+
+    if (((*cur >> kSavedRegisterMaskSize) & 0xff) == kSavedRegisterTag) {
+      stream << "| " << DwarfRegisterCodeToString(*cur & kLocationMask);
+      cur += sizeof(byte);
+      int decoded_size = 0;
+      int decoded_offset = static_cast<int>(DecodeULEB128(cur, &decoded_size));
+      cur += decoded_size;
+      stream << " saved at base" << std::showpos << std::dec
+             << decoded_offset * kDataAlignmentFactor << '\n';
+      continue;
+    }
+
+    uint8_t bytecode = *cur;
+    cur += sizeof(byte);
+
+    switch (bytecode) {
+      case kAdvanceLoc1: {
+        unsigned value = *reinterpret_cast<const uint8_t*>(cur);
+        cur += sizeof(uint8_t);
+        offset_in_procedure += value;
+        stream << "| pc_offset=" << std::dec << offset_in_procedure
+               << " (delta=0x" << std::hex << value << ")\n";
+        break;
+      }
+      case kAdvanceLoc2: {
+        uint16_t value = ReadUnalignedUInt16(cur);
+        cur += sizeof(uint16_t);
+        offset_in_procedure += value;
+        stream << "| pc_offset=" << std::dec << offset_in_procedure
+               << " (delta=0x" << std::hex << value << ")\n";
+        break;
+      }
+      case kAdvanceLoc4: {
+        uint32_t value = ReadUnalignedUInt32(cur);
+        offset_in_procedure += value;
+        cur += sizeof(uint32_t);
+        stream << "| pc_offset=" << std::dec << offset_in_procedure
+               << " (delta=0x" << std::hex << value << ")\n";
+        break;
+      }
+      case kDefCFA: {
+        int decoded_size = 0;
+        int base_register = DecodeULEB128(cur, &decoded_size);
+        cur += decoded_size;
+        int base_offset = DecodeULEB128(cur, &decoded_size);
+        cur += decoded_size;
+        stream << "| base_register=" << DwarfRegisterCodeToString(base_register)
+               << ", base_offset=0x" << std::hex << base_offset << '\n';
+        break;
+      }
+      case kDefCFAOffset: {
+        int decoded_size = 0;
+        stream << "| base_offset=0x" << std::hex
+               << DecodeULEB128(cur, &decoded_size) << '\n';
+        cur += decoded_size;
+        break;
+      }
+      case kDefCFARegister: {
+        int decoded_size = 0;
+        stream << "| base_register="
+               << DwarfRegisterCodeToString(DecodeULEB128(cur, &decoded_size))
+               << '\n';
+        cur += decoded_size;
+        break;
+      }
+      case kSameValue: {
+        int decoded_size = 0;
+        stream << "| "
+               << DwarfRegisterCodeToString(DecodeULEB128(cur, &decoded_size))
+               << " to initial value\n";
+        cur += decoded_size;
+        break;
+      }
+      case kNop:
+        stream << "| nop\n";
+        break;
+      default:
+        UNREACHABLE();
+        return;
+    }
+  }
+}
+
+// static
+void EhFrameWriter::DisassembleToStream(std::ostream& stream,  // NOLINT
+                                        const byte* start, const byte* end) {
+  const byte* cie_directives_start = start + kInitialStateOffsetInCIE;
+  const byte* cie_directives_end = start + kCIESize;
+  DCHECK_LE(cie_directives_start, cie_directives_end);
+
+  stream << reinterpret_cast<const void*>(start) << "  .eh_frame: CIE\n";
+  DumpDWARFDirectives(stream, cie_directives_start, cie_directives_end);
+
+  const byte* procedure_offset_address = start + kProcedureAddressOffset;
+  int32_t procedure_offset =
+      ReadUnalignedValue<int32_t>(procedure_offset_address);
+
+  const byte* procedure_size_address = start + kProcedureSizeOffset;
+  uint32_t procedure_size = ReadUnalignedUInt32(procedure_size_address);
+
+  const byte* fde_start = start + kCIESize;
+  stream << reinterpret_cast<const void*>(fde_start) << "  .eh_frame: FDE\n"
+         << reinterpret_cast<const void*>(procedure_offset_address)
+         << "  | procedure offset=" << procedure_offset << '\n'
+         << reinterpret_cast<const void*>(procedure_size_address)
+         << "  | procedure size=" << procedure_size << '\n';
+
+  const byte* fde_directives_start = start + kFDEDirectivesOffset;
+  const byte* fde_directives_end =
+      end - EhFrameHdr::kRecordSize - kEhFrameTerminatorSize;
+  DCHECK_LE(fde_directives_start, fde_directives_end);
+
+  DumpDWARFDirectives(stream, fde_directives_start, fde_directives_end);
+
+  const byte* fde_terminator_start = fde_directives_end;
+  stream << reinterpret_cast<const void*>(fde_terminator_start)
+         << "  .eh_frame: terminator\n";
+
+  const byte* eh_frame_hdr_start =
+      fde_terminator_start + kEhFrameTerminatorSize;
+  stream << reinterpret_cast<const void*>(eh_frame_hdr_start)
+         << "  .eh_frame_hdr: placeholder\n";
+}
+
+#endif
 
 //
 // In order to calculate offsets in the .eh_frame_hdr, we must know the layout
 // of the DSO generated by perf inject, which is assumed to be the following:
 //
 //  |      ...      |                        |
 //  +---------------+ <-- (F) ---            |  Larger offsets in file
 //  |               |           ^            |
 //  |  Instructions |           | .text      v
 //  |               |           v
@@ skipping 13 matching lines @@
 //  |    version    |           ^
 //  +---------------+           |
 //  |   encoding    |           |
 //  |  specifiers   |           |
 //  +---------------+ <---(A)   | .eh_frame_hdr
 //  |   offset to   |           |
 //  |   .eh_frame   |           |
 //  +---------------+           |
 //  |      ...      |          ...
 //
-// (F) is aligned at a 16-byte boundary.
-// (D) is aligned at a  8-byte boundary.
-// (B) is aligned at a  4-byte boundary.
-// (E), (C) and (A) have no alignment requirements.
+// (F) is aligned to a 16-byte boundary.
+// (D) is aligned to a  8-byte boundary.
+// (B) is aligned to a  4-byte boundary.
+// (C) is aligned to an addressing unit size boundary.
+// (E) and (A) have no alignment requirements.
 //
 // The distance between (A) and (B) is 4 bytes.
 //
-// The size of the .eh_frame is required to be a multiple of the pointer size,
-// which means that (B) will be naturally aligned to a 4-byte boundary on all
-// the architectures we support.
+// The size of the FDE is required to be a multiple of the pointer size, which
+// means that (B) will be naturally aligned to a 4-byte boundary on all the
+// architectures we support.
 //
 // Because (E) has no alignment requirements, there is padding between (E) and
 // (D). (F) is aligned at a 16-byte boundary, thus to a 8-byte one as well.
 //
-EhFrameHdr::EhFrameHdr(Code* code) {
-  int code_size = code->is_crankshafted() ? code->safepoint_table_offset()
-                                          : code->instruction_size();
-  version_ = 1;
-  eh_frame_ptr_encoding_ = DW_EH_PE_sdata4 | DW_EH_PE_pcrel;
-  lut_size_encoding_ = DW_EH_PE_udata4;
-  lut_entries_encoding_ = DW_EH_PE_sdata4 | DW_EH_PE_datarel;
+EhFrameHdr::EhFrameHdr(int code_size, int eh_frame_size) {
+  version_ = kEhFrameHdrVersion;
 
-  // .eh_frame pointer and LUT
-  if (code->has_unwinding_info()) {
-    DCHECK_GE(code->unwinding_info_size(), EhFrameHdr::kRecordSize);
-    int eh_frame_size = code->unwinding_info_size() - EhFrameHdr::kRecordSize;
+  eh_frame_ptr_encoding_ = kSData4 | kPcRel;
+  lut_size_encoding_ = kUData4;
+  lut_entries_encoding_ = kSData4 | kDataRel;
 
-    offset_to_eh_frame_ =
-        -(eh_frame_size + kVersionSize + kEncodingSpecifiersSize);  // A -> D
-    lut_entries_number_ = 1;
-    offset_to_procedure_ = -(RoundUp(code_size, 8) + eh_frame_size);  // B -> F
-    offset_to_fde_ = -(eh_frame_size - kCIESize);                     // B -> C
-  } else {
-    // Create a dummy table
-    offset_to_eh_frame_ = 0;
-    lut_entries_number_ = 0;
-    offset_to_procedure_ = 0;
-    offset_to_fde_ = 0;
-  }
+  offset_to_eh_frame_ = -(eh_frame_size + kFDEVersionSize +
+                          kFDEEncodingSpecifiersSize);  // A -> D
+  lut_entries_number_ = 1;
+  offset_to_procedure_ = -(RoundUp(code_size, 8) + eh_frame_size);  // B -> F
+  offset_to_fde_ = -(eh_frame_size - EhFrameWriter::kCIESize);      // B -> C
+}
+
+// static
+EhFrameHdr EhFrameHdr::MakeDummy() {
+  EhFrameHdr dummy_frame;
+  dummy_frame.version_ = kEhFrameHdrVersion;
+  dummy_frame.eh_frame_ptr_encoding_ = kSData4 | kPcRel;
+  dummy_frame.lut_size_encoding_ = kUData4;
+  dummy_frame.lut_entries_encoding_ = kSData4 | kDataRel;
+  dummy_frame.offset_to_eh_frame_ = 0;
+  dummy_frame.lut_entries_number_ = 0;
+  dummy_frame.offset_to_procedure_ = 0;
+  dummy_frame.offset_to_fde_ = 0;
+  return dummy_frame;
 }
 
 }  // namespace internal
 }  // namespace v8