aboutsummaryrefslogtreecommitdiff
path: root/src/common/dwarf/dwarf2reader.cc
diff options
context:
space:
mode:
authorYunlian Jiang <yunlian@google.com>2016-05-04 11:09:44 -0700
committerYunlian Jiang <yunlian@google.com>2016-05-04 11:09:44 -0700
commit764c21f7529df70a19f1e1cb33bb9ece28e0bf8f (patch)
treed71aaba9a7d47419c05ab7af184e117b5a115592 /src/common/dwarf/dwarf2reader.cc
parentmacho: fix printf type mismatches (diff)
downloadbreakpad-764c21f7529df70a19f1e1cb33bb9ece28e0bf8f.tar.xz
Add debug fission support.
This added debug fission support. It tries to find the dwp file from the debug dir /usr/lib/debug/*/debug and read symbols from them. Most of this patch comes from https://critique.corp.google.com/#review/52048295 and some fixes after that. The elf_reader.cc comes from TOT google code. I just removed some google dependency. Current problems from this patch 1: Some type mismatch: from uint8_t * to char *. 2: Some hack to find the .dwp file. (replace .debug with .dwp) BUG=chromium:604440 R=dehao@google.com, ivanpe@chromium.org Review URL: https://codereview.chromium.org/1884283002 .
Diffstat (limited to 'src/common/dwarf/dwarf2reader.cc')
-rw-r--r--src/common/dwarf/dwarf2reader.cc436
1 files changed, 410 insertions, 26 deletions
diff --git a/src/common/dwarf/dwarf2reader.cc b/src/common/dwarf/dwarf2reader.cc
index fcd29b8c..003ed7b1 100644
--- a/src/common/dwarf/dwarf2reader.cc
+++ b/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
@@ -174,6 +203,8 @@ const uint8_t *CompilationUnit::SkipAttribute(const uint8_t *start,
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;
@@ -296,9 +327,31 @@ 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;
}
@@ -320,48 +373,46 @@ const uint8_t *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 = reinterpret_cast<const char *>(start);
- handler_->ProcessAttributeString(dieoffset, attr, form,
- str);
+ 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:
@@ -441,10 +492,32 @@ const uint8_t *CompilationUnit::ProcessAttribute(
assert(string_buffer_ + offset < string_buffer_ + string_buffer_length_);
const char *str = reinterpret_cast<const char *>(string_buffer_ + offset);
- handler_->ProcessAttributeString(dieoffset, attr, form,
- str);
+ 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;
@@ -458,6 +531,16 @@ const uint8_t *CompilationUnit::ProcessDIE(uint64 dieoffset,
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;
}
@@ -515,6 +598,307 @@ void CompilationUnit::ProcessDIEs() {
}
}
+// 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_ != NULL) {
+ // If we have a .dwp file, read the debug sections for the requested CU.
+ SectionMap sections;
+ dwp_reader_->ReadDebugSectionsForCU(dwo_id_, &sections);
+ 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, &sections);
+ 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,
+ &section_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, &section_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, &section_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, &section_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) {