diff options
Diffstat (limited to '3rdParty/Breakpad/src/common/dwarf/dwarf2reader.cc')
-rw-r--r-- | 3rdParty/Breakpad/src/common/dwarf/dwarf2reader.cc | 617 |
1 files changed, 527 insertions, 90 deletions
diff --git a/3rdParty/Breakpad/src/common/dwarf/dwarf2reader.cc b/3rdParty/Breakpad/src/common/dwarf/dwarf2reader.cc index 7c1a29d..8774122 100644 --- a/3rdParty/Breakpad/src/common/dwarf/dwarf2reader.cc +++ b/3rdParty/Breakpad/src/common/dwarf/dwarf2reader.cc @@ -44,6 +44,8 @@ #include <string> #include <utility> +#include <sys/stat.h> + #include "common/dwarf/bytereader-inl.h" #include "common/dwarf/bytereader.h" #include "common/dwarf/line_state_machine.h" @@ -51,11 +53,38 @@ namespace dwarf2reader { -CompilationUnit::CompilationUnit(const SectionMap& sections, uint64 offset, +CompilationUnit::CompilationUnit(const string& path, + const SectionMap& sections, uint64 offset, ByteReader* reader, Dwarf2Handler* handler) - : offset_from_section_start_(offset), reader_(reader), - sections_(sections), handler_(handler), abbrevs_(NULL), - string_buffer_(NULL), string_buffer_length_(0) {} + : path_(path), offset_from_section_start_(offset), reader_(reader), + sections_(sections), handler_(handler), abbrevs_(), + string_buffer_(NULL), string_buffer_length_(0), + str_offsets_buffer_(NULL), str_offsets_buffer_length_(0), + addr_buffer_(NULL), addr_buffer_length_(0), + is_split_dwarf_(false), dwo_id_(0), dwo_name_(), + skeleton_dwo_id_(0), ranges_base_(0), addr_base_(0), + have_checked_for_dwp_(false), dwp_path_(), + dwp_byte_reader_(), dwp_reader_() {} + +// Initialize a compilation unit from a .dwo or .dwp file. +// In this case, we need the .debug_addr section from the +// executable file that contains the corresponding skeleton +// compilation unit. We also inherit the Dwarf2Handler from +// the executable file, and call it as if we were still +// processing the original compilation unit. + +void CompilationUnit::SetSplitDwarf(const uint8_t* addr_buffer, + uint64 addr_buffer_length, + uint64 addr_base, + uint64 ranges_base, + uint64 dwo_id) { + is_split_dwarf_ = true; + addr_buffer_ = addr_buffer; + addr_buffer_length_ = addr_buffer_length; + addr_base_ = addr_base; + ranges_base_ = ranges_base; + skeleton_dwo_id_ = dwo_id; +} // Read a DWARF2/3 abbreviation section. // Each abbrev consists of a abbreviation number, a tag, a byte @@ -83,9 +112,9 @@ void CompilationUnit::ReadAbbrevs() { // The only way to check whether we are reading over the end of the // buffer would be to first compute the size of the leb128 data by // reading it, then go back and read it again. - const char* abbrev_start = iter->second.first + + const uint8_t *abbrev_start = iter->second.first + header_.abbrev_offset; - const char* abbrevptr = abbrev_start; + const uint8_t *abbrevptr = abbrev_start; #ifndef NDEBUG const uint64 abbrev_length = iter->second.second - header_.abbrev_offset; #endif @@ -132,8 +161,8 @@ void CompilationUnit::ReadAbbrevs() { } // Skips a single DIE's attributes. -const char* CompilationUnit::SkipDIE(const char* start, - const Abbrev& abbrev) { +const uint8_t *CompilationUnit::SkipDIE(const uint8_t* start, + const Abbrev& abbrev) { for (AttributeList::const_iterator i = abbrev.attributes.begin(); i != abbrev.attributes.end(); i++) { @@ -143,8 +172,8 @@ const char* CompilationUnit::SkipDIE(const char* start, } // Skips a single attribute form's data. -const char* CompilationUnit::SkipAttribute(const char* start, - enum DwarfForm form) { +const uint8_t *CompilationUnit::SkipAttribute(const uint8_t *start, + enum DwarfForm form) { size_t len; switch (form) { @@ -171,9 +200,11 @@ const char* CompilationUnit::SkipAttribute(const char* start, case DW_FORM_ref_sig8: return start + 8; case DW_FORM_string: - return start + strlen(start) + 1; + return start + strlen(reinterpret_cast<const char *>(start)) + 1; case DW_FORM_udata: case DW_FORM_ref_udata: + case DW_FORM_GNU_str_index: + case DW_FORM_GNU_addr_index: reader_->ReadUnsignedLEB128(start, &len); return start + len; @@ -183,14 +214,15 @@ const char* CompilationUnit::SkipAttribute(const char* start, case DW_FORM_addr: return start + reader_->AddressSize(); case DW_FORM_ref_addr: - // DWARF2 and 3 differ on whether ref_addr is address size or + // DWARF2 and 3/4 differ on whether ref_addr is address size or // offset size. - assert(header_.version == 2 || header_.version == 3); + assert(header_.version >= 2); if (header_.version == 2) { return start + reader_->AddressSize(); - } else if (header_.version == 3) { + } else if (header_.version >= 3) { return start + reader_->OffsetSize(); } + break; case DW_FORM_block1: return start + 1 + reader_->ReadOneByte(start); @@ -217,7 +249,7 @@ const char* CompilationUnit::SkipAttribute(const char* start, // the offset in the .debug_abbrev section for our abbrevs, and an // address size. void CompilationUnit::ReadHeader() { - const char* headerptr = buffer_; + const uint8_t *headerptr = buffer_; size_t initial_length_size; assert(headerptr + 4 < buffer_ + buffer_length_); @@ -234,7 +266,9 @@ void CompilationUnit::ReadHeader() { header_.abbrev_offset = reader_->ReadOffset(headerptr); headerptr += reader_->OffsetSize(); - assert(headerptr + 1 < buffer_ + buffer_length_); + // Compare against less than or equal because this may be the last + // section in the file. + assert(headerptr + 1 <= buffer_ + buffer_length_); header_.address_size = reader_->ReadOneByte(headerptr); reader_->SetAddressSize(header_.address_size); headerptr += 1; @@ -295,17 +329,39 @@ uint64 CompilationUnit::Start() { string_buffer_length_ = iter->second.second; } + // Set the string offsets section if we have one. + iter = sections_.find(".debug_str_offsets"); + if (iter != sections_.end()) { + str_offsets_buffer_ = iter->second.first; + str_offsets_buffer_length_ = iter->second.second; + } + + // Set the address section if we have one. + iter = sections_.find(".debug_addr"); + if (iter != sections_.end()) { + addr_buffer_ = iter->second.first; + addr_buffer_length_ = iter->second.second; + } + // Now that we have our abbreviations, start processing DIE's. ProcessDIEs(); + // If this is a skeleton compilation unit generated with split DWARF, + // and the client needs the full debug info, we need to find the full + // compilation unit in a .dwo or .dwp file. + if (!is_split_dwarf_ + && dwo_name_ != NULL + && handler_->NeedSplitDebugInfo()) + ProcessSplitDwarf(); + return ourlength; } // If one really wanted, you could merge SkipAttribute and // ProcessAttribute // This is all boring data manipulation and calling of the handler. -const char* CompilationUnit::ProcessAttribute( - uint64 dieoffset, const char* start, enum DwarfAttribute attr, +const uint8_t *CompilationUnit::ProcessAttribute( + uint64 dieoffset, const uint8_t *start, enum DwarfAttribute attr, enum DwarfForm form) { size_t len; @@ -319,48 +375,46 @@ const char* CompilationUnit::ProcessAttribute( return ProcessAttribute(dieoffset, start, attr, form); case DW_FORM_flag_present: - handler_->ProcessAttributeUnsigned(dieoffset, attr, form, 1); + ProcessAttributeUnsigned(dieoffset, attr, form, 1); return start; case DW_FORM_data1: case DW_FORM_flag: - handler_->ProcessAttributeUnsigned(dieoffset, attr, form, - reader_->ReadOneByte(start)); + ProcessAttributeUnsigned(dieoffset, attr, form, + reader_->ReadOneByte(start)); return start + 1; case DW_FORM_data2: - handler_->ProcessAttributeUnsigned(dieoffset, attr, form, - reader_->ReadTwoBytes(start)); + ProcessAttributeUnsigned(dieoffset, attr, form, + reader_->ReadTwoBytes(start)); return start + 2; case DW_FORM_data4: - handler_->ProcessAttributeUnsigned(dieoffset, attr, form, - reader_->ReadFourBytes(start)); + ProcessAttributeUnsigned(dieoffset, attr, form, + reader_->ReadFourBytes(start)); return start + 4; case DW_FORM_data8: - handler_->ProcessAttributeUnsigned(dieoffset, attr, form, - reader_->ReadEightBytes(start)); + ProcessAttributeUnsigned(dieoffset, attr, form, + reader_->ReadEightBytes(start)); return start + 8; case DW_FORM_string: { - const char* str = start; - handler_->ProcessAttributeString(dieoffset, attr, form, - str); + const char *str = reinterpret_cast<const char *>(start); + ProcessAttributeString(dieoffset, attr, form, str); return start + strlen(str) + 1; } case DW_FORM_udata: - handler_->ProcessAttributeUnsigned(dieoffset, attr, form, - reader_->ReadUnsignedLEB128(start, - &len)); + ProcessAttributeUnsigned(dieoffset, attr, form, + reader_->ReadUnsignedLEB128(start, &len)); return start + len; case DW_FORM_sdata: - handler_->ProcessAttributeSigned(dieoffset, attr, form, - reader_->ReadSignedLEB128(start, &len)); + ProcessAttributeSigned(dieoffset, attr, form, + reader_->ReadSignedLEB128(start, &len)); return start + len; case DW_FORM_addr: - handler_->ProcessAttributeUnsigned(dieoffset, attr, form, - reader_->ReadAddress(start)); + ProcessAttributeUnsigned(dieoffset, attr, form, + reader_->ReadAddress(start)); return start + reader_->AddressSize(); case DW_FORM_sec_offset: - handler_->ProcessAttributeUnsigned(dieoffset, attr, form, - reader_->ReadOffset(start)); + ProcessAttributeUnsigned(dieoffset, attr, form, + reader_->ReadOffset(start)); return start + reader_->OffsetSize(); case DW_FORM_ref1: @@ -390,14 +444,14 @@ const char* CompilationUnit::ProcessAttribute( + offset_from_section_start_); return start + len; case DW_FORM_ref_addr: - // DWARF2 and 3 differ on whether ref_addr is address size or + // DWARF2 and 3/4 differ on whether ref_addr is address size or // offset size. - assert(header_.version == 2 || header_.version == 3); + assert(header_.version >= 2); if (header_.version == 2) { handler_->ProcessAttributeReference(dieoffset, attr, form, reader_->ReadAddress(start)); return start + reader_->AddressSize(); - } else if (header_.version == 3) { + } else if (header_.version >= 3) { handler_->ProcessAttributeReference(dieoffset, attr, form, reader_->ReadOffset(start)); return start + reader_->OffsetSize(); @@ -439,34 +493,66 @@ const char* CompilationUnit::ProcessAttribute( const uint64 offset = reader_->ReadOffset(start); assert(string_buffer_ + offset < string_buffer_ + string_buffer_length_); - const char* str = string_buffer_ + offset; - handler_->ProcessAttributeString(dieoffset, attr, form, - str); + const char *str = reinterpret_cast<const char *>(string_buffer_ + offset); + ProcessAttributeString(dieoffset, attr, form, str); return start + reader_->OffsetSize(); } + + case DW_FORM_GNU_str_index: { + uint64 str_index = reader_->ReadUnsignedLEB128(start, &len); + const uint8_t* offset_ptr = + str_offsets_buffer_ + str_index * reader_->OffsetSize(); + const uint64 offset = reader_->ReadOffset(offset_ptr); + if (offset >= string_buffer_length_) { + return NULL; + } + + const char* str = reinterpret_cast<const char *>(string_buffer_) + offset; + ProcessAttributeString(dieoffset, attr, form, str); + return start + len; + break; + } + case DW_FORM_GNU_addr_index: { + uint64 addr_index = reader_->ReadUnsignedLEB128(start, &len); + const uint8_t* addr_ptr = + addr_buffer_ + addr_base_ + addr_index * reader_->AddressSize(); + ProcessAttributeUnsigned(dieoffset, attr, form, + reader_->ReadAddress(addr_ptr)); + return start + len; + } } fprintf(stderr, "Unhandled form type\n"); return NULL; } -const char* CompilationUnit::ProcessDIE(uint64 dieoffset, - const char* start, - const Abbrev& abbrev) { +const uint8_t *CompilationUnit::ProcessDIE(uint64 dieoffset, + const uint8_t *start, + const Abbrev& abbrev) { for (AttributeList::const_iterator i = abbrev.attributes.begin(); i != abbrev.attributes.end(); i++) { start = ProcessAttribute(dieoffset, start, i->first, i->second); } + + // If this is a compilation unit in a split DWARF object, verify that + // the dwo_id matches. If it does not match, we will ignore this + // compilation unit. + if (abbrev.tag == DW_TAG_compile_unit + && is_split_dwarf_ + && dwo_id_ != skeleton_dwo_id_) { + return NULL; + } + return start; } void CompilationUnit::ProcessDIEs() { - const char* dieptr = after_header_; + const uint8_t *dieptr = after_header_; size_t len; // lengthstart is the place the length field is based on. // It is the point in the header after the initial length field - const char* lengthstart = buffer_; + const uint8_t *lengthstart = buffer_; // In 64 bit dwarf, the initial length is 12 bytes, because of the // 0xffffffff at the start. @@ -500,7 +586,7 @@ void CompilationUnit::ProcessDIEs() { const Abbrev& abbrev = abbrevs_->at(static_cast<size_t>(abbrev_num)); const enum DwarfTag tag = abbrev.tag; - if (!handler_->StartDIE(absolute_offset, tag, abbrev.attributes)) { + if (!handler_->StartDIE(absolute_offset, tag)) { dieptr = SkipDIE(dieptr, abbrev); } else { dieptr = ProcessDIE(absolute_offset, dieptr, abbrev); @@ -514,10 +600,313 @@ void CompilationUnit::ProcessDIEs() { } } -LineInfo::LineInfo(const char* buffer, uint64 buffer_length, +// Check for a valid ELF file and return the Address size. +// Returns 0 if not a valid ELF file. +inline int GetElfWidth(const ElfReader& elf) { + if (elf.IsElf32File()) + return 4; + if (elf.IsElf64File()) + return 8; + return 0; +} + +void CompilationUnit::ProcessSplitDwarf() { + struct stat statbuf; + if (!have_checked_for_dwp_) { + // Look for a .dwp file in the same directory as the executable. + have_checked_for_dwp_ = true; + string dwp_suffix(".dwp"); + dwp_path_ = path_ + dwp_suffix; + if (stat(dwp_path_.c_str(), &statbuf) != 0) { + // Fall back to a split .debug file in the same directory. + string debug_suffix(".debug"); + dwp_path_ = path_; + size_t found = path_.rfind(debug_suffix); + if (found + debug_suffix.length() == path_.length()) + dwp_path_ = dwp_path_.replace(found, debug_suffix.length(), dwp_suffix); + } + if (stat(dwp_path_.c_str(), &statbuf) == 0) { + ElfReader* elf = new ElfReader(dwp_path_); + int width = GetElfWidth(*elf); + if (width != 0) { + dwp_byte_reader_.reset(new ByteReader(reader_->GetEndianness())); + dwp_byte_reader_->SetAddressSize(width); + dwp_reader_.reset(new DwpReader(*dwp_byte_reader_, elf)); + dwp_reader_->Initialize(); + } else { + delete elf; + } + } + } + bool found_in_dwp = false; + if (dwp_reader_) { + // If we have a .dwp file, read the debug sections for the requested CU. + SectionMap sections; + dwp_reader_->ReadDebugSectionsForCU(dwo_id_, §ions); + if (!sections.empty()) { + found_in_dwp = true; + CompilationUnit dwp_comp_unit(dwp_path_, sections, 0, + dwp_byte_reader_.get(), handler_); + dwp_comp_unit.SetSplitDwarf(addr_buffer_, addr_buffer_length_, addr_base_, + ranges_base_, dwo_id_); + dwp_comp_unit.Start(); + } + } + if (!found_in_dwp) { + // If no .dwp file, try to open the .dwo file. + if (stat(dwo_name_, &statbuf) == 0) { + ElfReader elf(dwo_name_); + int width = GetElfWidth(elf); + if (width != 0) { + ByteReader reader(ENDIANNESS_LITTLE); + reader.SetAddressSize(width); + SectionMap sections; + ReadDebugSectionsFromDwo(&elf, §ions); + CompilationUnit dwo_comp_unit(dwo_name_, sections, 0, &reader, + handler_); + dwo_comp_unit.SetSplitDwarf(addr_buffer_, addr_buffer_length_, + addr_base_, ranges_base_, dwo_id_); + dwo_comp_unit.Start(); + } + } + } +} + +void CompilationUnit::ReadDebugSectionsFromDwo(ElfReader* elf_reader, + SectionMap* sections) { + static const char* const section_names[] = { + ".debug_abbrev", + ".debug_info", + ".debug_str_offsets", + ".debug_str" + }; + for (unsigned int i = 0u; + i < sizeof(section_names)/sizeof(*(section_names)); ++i) { + string base_name = section_names[i]; + string dwo_name = base_name + ".dwo"; + size_t section_size; + const char* section_data = elf_reader->GetSectionByName(dwo_name, + §ion_size); + if (section_data != NULL) + sections->insert(std::make_pair( + base_name, std::make_pair( + reinterpret_cast<const uint8_t *>(section_data), + section_size))); + } +} + +DwpReader::DwpReader(const ByteReader& byte_reader, ElfReader* elf_reader) + : elf_reader_(elf_reader), byte_reader_(byte_reader), + cu_index_(NULL), cu_index_size_(0), string_buffer_(NULL), + string_buffer_size_(0), version_(0), ncolumns_(0), nunits_(0), + nslots_(0), phash_(NULL), pindex_(NULL), shndx_pool_(NULL), + offset_table_(NULL), size_table_(NULL), abbrev_data_(NULL), + abbrev_size_(0), info_data_(NULL), info_size_(0), + str_offsets_data_(NULL), str_offsets_size_(0) {} + +DwpReader::~DwpReader() { + if (elf_reader_) delete elf_reader_; +} + +void DwpReader::Initialize() { + cu_index_ = elf_reader_->GetSectionByName(".debug_cu_index", + &cu_index_size_); + if (cu_index_ == NULL) { + return; + } + // The .debug_str.dwo section is shared by all CUs in the file. + string_buffer_ = elf_reader_->GetSectionByName(".debug_str.dwo", + &string_buffer_size_); + + version_ = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(cu_index_)); + + if (version_ == 1) { + nslots_ = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(cu_index_) + + 3 * sizeof(uint32)); + phash_ = cu_index_ + 4 * sizeof(uint32); + pindex_ = phash_ + nslots_ * sizeof(uint64); + shndx_pool_ = pindex_ + nslots_ * sizeof(uint32); + if (shndx_pool_ >= cu_index_ + cu_index_size_) { + version_ = 0; + } + } else if (version_ == 2) { + ncolumns_ = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(cu_index_) + sizeof(uint32)); + nunits_ = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(cu_index_) + 2 * sizeof(uint32)); + nslots_ = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(cu_index_) + 3 * sizeof(uint32)); + phash_ = cu_index_ + 4 * sizeof(uint32); + pindex_ = phash_ + nslots_ * sizeof(uint64); + offset_table_ = pindex_ + nslots_ * sizeof(uint32); + size_table_ = offset_table_ + ncolumns_ * (nunits_ + 1) * sizeof(uint32); + abbrev_data_ = elf_reader_->GetSectionByName(".debug_abbrev.dwo", + &abbrev_size_); + info_data_ = elf_reader_->GetSectionByName(".debug_info.dwo", &info_size_); + str_offsets_data_ = elf_reader_->GetSectionByName(".debug_str_offsets.dwo", + &str_offsets_size_); + if (size_table_ >= cu_index_ + cu_index_size_) { + version_ = 0; + } + } +} + +void DwpReader::ReadDebugSectionsForCU(uint64 dwo_id, + SectionMap* sections) { + if (version_ == 1) { + int slot = LookupCU(dwo_id); + if (slot == -1) { + return; + } + + // The index table points to the section index pool, where we + // can read a list of section indexes for the debug sections + // for the CU whose dwo_id we are looking for. + int index = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(pindex_) + + slot * sizeof(uint32)); + const char* shndx_list = shndx_pool_ + index * sizeof(uint32); + for (;;) { + if (shndx_list >= cu_index_ + cu_index_size_) { + version_ = 0; + return; + } + unsigned int shndx = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(shndx_list)); + shndx_list += sizeof(uint32); + if (shndx == 0) + break; + const char* section_name = elf_reader_->GetSectionName(shndx); + size_t section_size; + const char* section_data; + // We're only interested in these four debug sections. + // The section names in the .dwo file end with ".dwo", but we + // add them to the sections table with their normal names. + if (!strncmp(section_name, ".debug_abbrev", strlen(".debug_abbrev"))) { + section_data = elf_reader_->GetSectionByIndex(shndx, §ion_size); + sections->insert(std::make_pair( + ".debug_abbrev", + std::make_pair(reinterpret_cast<const uint8_t *> (section_data), + section_size))); + } else if (!strncmp(section_name, ".debug_info", strlen(".debug_info"))) { + section_data = elf_reader_->GetSectionByIndex(shndx, §ion_size); + sections->insert(std::make_pair( + ".debug_info", + std::make_pair(reinterpret_cast<const uint8_t *> (section_data), + section_size))); + } else if (!strncmp(section_name, ".debug_str_offsets", + strlen(".debug_str_offsets"))) { + section_data = elf_reader_->GetSectionByIndex(shndx, §ion_size); + sections->insert(std::make_pair( + ".debug_str_offsets", + std::make_pair(reinterpret_cast<const uint8_t *> (section_data), + section_size))); + } + } + sections->insert(std::make_pair( + ".debug_str", + std::make_pair(reinterpret_cast<const uint8_t *> (string_buffer_), + string_buffer_size_))); + } else if (version_ == 2) { + uint32 index = LookupCUv2(dwo_id); + if (index == 0) { + return; + } + + // The index points to a row in each of the section offsets table + // and the section size table, where we can read the offsets and sizes + // of the contributions to each debug section from the CU whose dwo_id + // we are looking for. Row 0 of the section offsets table has the + // section ids for each column of the table. The size table begins + // with row 1. + const char* id_row = offset_table_; + const char* offset_row = offset_table_ + + index * ncolumns_ * sizeof(uint32); + const char* size_row = + size_table_ + (index - 1) * ncolumns_ * sizeof(uint32); + if (size_row + ncolumns_ * sizeof(uint32) > cu_index_ + cu_index_size_) { + version_ = 0; + return; + } + for (unsigned int col = 0u; col < ncolumns_; ++col) { + uint32 section_id = + byte_reader_.ReadFourBytes(reinterpret_cast<const uint8_t *>(id_row) + + col * sizeof(uint32)); + uint32 offset = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(offset_row) + + col * sizeof(uint32)); + uint32 size = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(size_row) + col * sizeof(uint32)); + if (section_id == DW_SECT_ABBREV) { + sections->insert(std::make_pair( + ".debug_abbrev", + std::make_pair(reinterpret_cast<const uint8_t *> (abbrev_data_) + + offset, size))); + } else if (section_id == DW_SECT_INFO) { + sections->insert(std::make_pair( + ".debug_info", + std::make_pair(reinterpret_cast<const uint8_t *> (info_data_) + + offset, size))); + } else if (section_id == DW_SECT_STR_OFFSETS) { + sections->insert(std::make_pair( + ".debug_str_offsets", + std::make_pair(reinterpret_cast<const uint8_t *> (str_offsets_data_) + + offset, size))); + } + } + sections->insert(std::make_pair( + ".debug_str", + std::make_pair(reinterpret_cast<const uint8_t *> (string_buffer_), + string_buffer_size_))); + } +} + +int DwpReader::LookupCU(uint64 dwo_id) { + uint32 slot = static_cast<uint32>(dwo_id) & (nslots_ - 1); + uint64 probe = byte_reader_.ReadEightBytes( + reinterpret_cast<const uint8_t *>(phash_) + slot * sizeof(uint64)); + if (probe != 0 && probe != dwo_id) { + uint32 secondary_hash = + (static_cast<uint32>(dwo_id >> 32) & (nslots_ - 1)) | 1; + do { + slot = (slot + secondary_hash) & (nslots_ - 1); + probe = byte_reader_.ReadEightBytes( + reinterpret_cast<const uint8_t *>(phash_) + slot * sizeof(uint64)); + } while (probe != 0 && probe != dwo_id); + } + if (probe == 0) + return -1; + return slot; +} + +uint32 DwpReader::LookupCUv2(uint64 dwo_id) { + uint32 slot = static_cast<uint32>(dwo_id) & (nslots_ - 1); + uint64 probe = byte_reader_.ReadEightBytes( + reinterpret_cast<const uint8_t *>(phash_) + slot * sizeof(uint64)); + uint32 index = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(pindex_) + slot * sizeof(uint32)); + if (index != 0 && probe != dwo_id) { + uint32 secondary_hash = + (static_cast<uint32>(dwo_id >> 32) & (nslots_ - 1)) | 1; + do { + slot = (slot + secondary_hash) & (nslots_ - 1); + probe = byte_reader_.ReadEightBytes( + reinterpret_cast<const uint8_t *>(phash_) + slot * sizeof(uint64)); + index = byte_reader_.ReadFourBytes( + reinterpret_cast<const uint8_t *>(pindex_) + slot * sizeof(uint32)); + } while (index != 0 && probe != dwo_id); + } + return index; +} + +LineInfo::LineInfo(const uint8_t *buffer, uint64 buffer_length, ByteReader* reader, LineInfoHandler* handler): - handler_(handler), reader_(reader), buffer_(buffer), - buffer_length_(buffer_length) { + handler_(handler), reader_(reader), buffer_(buffer) { +#ifndef NDEBUG + buffer_length_ = buffer_length; +#endif header_.std_opcode_lengths = NULL; } @@ -530,7 +919,7 @@ uint64 LineInfo::Start() { // The header for a debug_line section is mildly complicated, because // the line info is very tightly encoded. void LineInfo::ReadHeader() { - const char* lineptr = buffer_; + const uint8_t *lineptr = buffer_; size_t initial_length_size; const uint64 initial_length @@ -553,6 +942,13 @@ void LineInfo::ReadHeader() { header_.min_insn_length = reader_->ReadOneByte(lineptr); lineptr += 1; + if (header_.version >= 4) { + __attribute__((unused)) uint8 max_ops_per_insn = + reader_->ReadOneByte(lineptr); + ++lineptr; + assert(max_ops_per_insn == 1); + } + header_.default_is_stmt = reader_->ReadOneByte(lineptr); lineptr += 1; @@ -577,7 +973,7 @@ void LineInfo::ReadHeader() { if (*lineptr) { uint32 dirindex = 1; while (*lineptr) { - const char* dirname = lineptr; + const char *dirname = reinterpret_cast<const char *>(lineptr); handler_->DefineDir(dirname, dirindex); lineptr += strlen(dirname) + 1; dirindex++; @@ -590,7 +986,7 @@ void LineInfo::ReadHeader() { uint32 fileindex = 1; size_t len; while (*lineptr) { - const char* filename = lineptr; + const char *filename = reinterpret_cast<const char *>(lineptr); lineptr += strlen(filename) + 1; uint64 dirindex = reader_->ReadUnsignedLEB128(lineptr, &len); @@ -615,7 +1011,7 @@ void LineInfo::ReadHeader() { bool LineInfo::ProcessOneOpcode(ByteReader* reader, LineInfoHandler* handler, const struct LineInfoHeader &header, - const char* start, + const uint8_t *start, struct LineStateMachine* lsm, size_t* len, uintptr pc, @@ -756,7 +1152,7 @@ bool LineInfo::ProcessOneOpcode(ByteReader* reader, } break; case DW_LNE_define_file: { - const char* filename = start; + const char *filename = reinterpret_cast<const char *>(start); templen = strlen(filename) + 1; start += templen; @@ -803,7 +1199,7 @@ void LineInfo::ReadLines() { // lengthstart is the place the length field is based on. // It is the point in the header after the initial length field - const char* lengthstart = buffer_; + const uint8_t *lengthstart = buffer_; // In 64 bit dwarf, the initial length is 12 bytes, because of the // 0xffffffff at the start. @@ -812,7 +1208,7 @@ void LineInfo::ReadLines() { else lengthstart += 4; - const char* lineptr = after_header_; + const uint8_t *lineptr = after_header_; lsm.Reset(header_.default_is_stmt); // The LineInfoHandler interface expects each line's length along @@ -1311,7 +1707,7 @@ class CallFrameInfo::State { const Entry *entry_; // The next instruction to process. - const char *cursor_; + const uint8_t *cursor_; // The current set of rules. RuleMap rules_; @@ -1409,7 +1805,8 @@ bool CallFrameInfo::State::ParseOperands(const char *format, if (len > bytes_left || expression_length > bytes_left - len) return ReportIncomplete(); cursor_ += len; - operands->expression = string(cursor_, expression_length); + operands->expression = string(reinterpret_cast<const char *>(cursor_), + expression_length); cursor_ += expression_length; break; } @@ -1512,16 +1909,19 @@ bool CallFrameInfo::State::DoInstruction() { // Change the base register used to compute the CFA. case DW_CFA_def_cfa_register: { + if (!ParseOperands("r", &ops)) return false; Rule *cfa_rule = rules_.CFARule(); if (!cfa_rule) { - reporter_->NoCFARule(entry_->offset, entry_->kind, CursorOffset()); + if (!DoDefCFA(ops.register_number, ops.offset)) { + reporter_->NoCFARule(entry_->offset, entry_->kind, CursorOffset()); + return false; + } + } else { + cfa_rule->SetBaseRegister(ops.register_number); + if (!cfa_rule->Handle(handler_, address_, + Handler::kCFARegister)) return false; } - if (!ParseOperands("r", &ops)) return false; - cfa_rule->SetBaseRegister(ops.register_number); - if (!cfa_rule->Handle(handler_, address_, - Handler::kCFARegister)) - return false; break; } @@ -1759,8 +2159,8 @@ bool CallFrameInfo::State::DoRestore(unsigned reg) { return DoRule(reg, rule); } -bool CallFrameInfo::ReadEntryPrologue(const char *cursor, Entry *entry) { - const char *buffer_end = buffer_ + buffer_length_; +bool CallFrameInfo::ReadEntryPrologue(const uint8_t *cursor, Entry *entry) { + const uint8_t *buffer_end = buffer_ + buffer_length_; // Initialize enough of ENTRY for use in error reporting. entry->offset = cursor - buffer_; @@ -1838,7 +2238,7 @@ bool CallFrameInfo::ReadEntryPrologue(const char *cursor, Entry *entry) { } bool CallFrameInfo::ReadCIEFields(CIE *cie) { - const char *cursor = cie->fields; + const uint8_t *cursor = cie->fields; size_t len; assert(cie->kind == kCIE); @@ -1860,22 +2260,23 @@ bool CallFrameInfo::ReadCIEFields(CIE *cie) { cursor++; // If we don't recognize the version, we can't parse any more fields of the - // CIE. For DWARF CFI, we handle versions 1 through 3 (there was never a - // version 2 of CFI data). For .eh_frame, we handle versions 1 and 3 as well; + // CIE. For DWARF CFI, we handle versions 1 through 4 (there was never a + // version 2 of CFI data). For .eh_frame, we handle versions 1 and 4 as well; // the difference between those versions seems to be the same as for // .debug_frame. - if (cie->version < 1 || cie->version > 3) { + if (cie->version < 1 || cie->version > 4) { reporter_->UnrecognizedVersion(cie->offset, cie->version); return false; } - const char *augmentation_start = cursor; - const void *augmentation_end = - memchr(augmentation_start, '\0', cie->end - augmentation_start); + const uint8_t *augmentation_start = cursor; + const uint8_t *augmentation_end = + reinterpret_cast<const uint8_t *>(memchr(augmentation_start, '\0', + cie->end - augmentation_start)); if (! augmentation_end) return ReportIncomplete(cie); - cursor = static_cast<const char *>(augmentation_end); - cie->augmentation = string(augmentation_start, - cursor - augmentation_start); + cursor = augmentation_end; + cie->augmentation = string(reinterpret_cast<const char *>(augmentation_start), + cursor - augmentation_start); // Skip the terminating '\0'. cursor++; @@ -1893,16 +2294,36 @@ bool CallFrameInfo::ReadCIEFields(CIE *cie) { } } + if (cie->version >= 4) { + uint8_t address_size = *cursor++; + if (address_size != 8) { + // TODO(scottmg): Only supporting x64 for now. + reporter_->UnexpectedAddressSize(cie->offset, address_size); + return false; + } + + uint8_t segment_size = *cursor++; + if (segment_size != 0) { + // TODO(scottmg): Only supporting x64 for now. + // I would have perhaps expected 4 here, but LLVM emits a 0, near + // http://llvm.org/docs/doxygen/html/MCDwarf_8cpp_source.html#l00606. As + // we are not using the value, only succeed for now if it's the expected + // 0. + reporter_->UnexpectedSegmentSize(cie->offset, segment_size); + return false; + } + } + // Parse the code alignment factor. cie->code_alignment_factor = reader_->ReadUnsignedLEB128(cursor, &len); if (size_t(cie->end - cursor) < len) return ReportIncomplete(cie); cursor += len; - + // Parse the data alignment factor. cie->data_alignment_factor = reader_->ReadSignedLEB128(cursor, &len); if (size_t(cie->end - cursor) < len) return ReportIncomplete(cie); cursor += len; - + // Parse the return address register. This is a ubyte in version 1, and // a ULEB128 in version 3. if (cie->version == 1) { @@ -1921,9 +2342,9 @@ bool CallFrameInfo::ReadCIEFields(CIE *cie) { if (size_t(cie->end - cursor) < len + data_size) return ReportIncomplete(cie); cursor += len; - const char *data = cursor; + const uint8_t *data = cursor; cursor += data_size; - const char *data_end = cursor; + const uint8_t *data_end = cursor; cie->has_z_lsda = false; cie->has_z_personality = false; @@ -2013,9 +2434,9 @@ bool CallFrameInfo::ReadCIEFields(CIE *cie) { return true; } - + bool CallFrameInfo::ReadFDEFields(FDE *fde) { - const char *cursor = fde->fields; + const uint8_t *cursor = fde->fields; size_t size; fde->address = reader_->ReadEncodedPointer(cursor, fde->cie->pointer_encoding, @@ -2081,10 +2502,10 @@ bool CallFrameInfo::ReadFDEFields(FDE *fde) { } bool CallFrameInfo::Start() { - const char *buffer_end = buffer_ + buffer_length_; - const char *cursor; + const uint8_t *buffer_end = buffer_ + buffer_length_; + const uint8_t *cursor; bool all_ok = true; - const char *entry_end; + const uint8_t *entry_end; bool ok; // Traverse all the entries in buffer_, skipping CIEs and offering @@ -2254,6 +2675,22 @@ void CallFrameInfo::Reporter::BadCIEId(uint64 offset, uint64 cie_offset) { filename_.c_str(), offset, section_.c_str(), cie_offset); } +void CallFrameInfo::Reporter::UnexpectedAddressSize(uint64 offset, + uint8_t address_size) { + fprintf(stderr, + "%s: CFI frame description entry at offset 0x%llx in '%s':" + " CIE specifies unexpected address size: %d\n", + filename_.c_str(), offset, section_.c_str(), address_size); +} + +void CallFrameInfo::Reporter::UnexpectedSegmentSize(uint64 offset, + uint8_t segment_size) { + fprintf(stderr, + "%s: CFI frame description entry at offset 0x%llx in '%s':" + " CIE specifies unexpected segment size: %d\n", + filename_.c_str(), offset, section_.c_str(), segment_size); +} + void CallFrameInfo::Reporter::UnrecognizedVersion(uint64 offset, int version) { fprintf(stderr, "%s: CFI frame description entry at offset 0x%llx in '%s':" |