daily update
[deliverable/binutils-gdb.git] / gold / dwarf_reader.cc
index 1873091b4973ab2ba0d45b89bdb33d083dfc72f5..e8fe04e792fc043802aa80379e1ca8b45b68ce47 100644 (file)
@@ -1,6 +1,6 @@
 // dwarf_reader.cc -- parse dwarf2/3 debug information
 
-// Copyright 2007 Free Software Foundation, Inc.
+// Copyright 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
 // Written by Ian Lance Taylor <iant@google.com>.
 
 // This file is part of gold.
 
 #include "gold.h"
 
+#include <algorithm>
+#include <vector>
+
 #include "elfcpp_swap.h"
 #include "dwarf.h"
+#include "object.h"
+#include "parameters.h"
+#include "reloc.h"
 #include "dwarf_reader.h"
-
-namespace {
-
-// Read an unsigned LEB128 number.  Each byte contains 7 bits of
-// information, plus one bit saying whether the number continues or
-// not.
-
-uint64_t
-read_unsigned_LEB_128(const unsigned char* buffer, size_t* len)
-{
-  uint64_t result = 0;
-  size_t num_read = 0;
-  unsigned int shift = 0;
-  unsigned char byte;
-
-  do
-    {
-      byte = *buffer++;
-      num_read++;
-      result |= (static_cast<uint64_t>(byte & 0x7f)) << shift;
-      shift += 7;
-    }
-  while (byte & 0x80);
-
-  *len = num_read;
-
-  return result;
-}
-
-// Read a signed LEB128 number.  These are like regular LEB128
-// numbers, except the last byte may have a sign bit set.
-
-int64_t
-read_signed_LEB_128(const unsigned char* buffer, size_t* len)
-{
-  int64_t result = 0;
-  int shift = 0;
-  size_t num_read = 0;
-  unsigned char byte;
-
-  do
-    {
-      byte = *buffer++;
-      num_read++;
-      result |= (static_cast<uint64_t>(byte & 0x7f) << shift);
-      shift += 7;
-    }
-  while (byte & 0x80);
-
-  if ((shift < 8 * static_cast<int>(sizeof(result))) && (byte & 0x40))
-    result |= -((static_cast<int64_t>(1)) << shift);
-  *len = num_read;
-  return result;
-}
-
-} // End anonymous namespace.
-
+#include "int_encoding.h"
+#include "compressed_output.h"
 
 namespace gold {
 
-// This is the format of a DWARF2/3 line state machine that we process
-// opcodes using.  There is no need for anything outside the lineinfo
-// processor to know how this works.
-
 struct LineStateMachine
 {
   int file_num;
@@ -114,14 +61,92 @@ ResetLineStateMachine(struct LineStateMachine* lsm, bool default_is_stmt)
   lsm->end_sequence = false;
 }
 
+template<int size, bool big_endian>
+Sized_dwarf_line_info<size, big_endian>::Sized_dwarf_line_info(Object* object,
+                                                               unsigned int read_shndx)
+  : data_valid_(false), buffer_(NULL), symtab_buffer_(NULL),
+    directories_(), files_(), current_header_index_(-1)
+{
+  unsigned int debug_shndx;
+  for (debug_shndx = 1; debug_shndx < object->shnum(); ++debug_shndx)
+    {
+      // FIXME: do this more efficiently: section_name() isn't super-fast
+      std::string name = object->section_name(debug_shndx);
+      if (name == ".debug_line" || name == ".zdebug_line")
+       {
+         section_size_type buffer_size;
+         this->buffer_ = object->section_contents(debug_shndx, &buffer_size,
+                                                  false);
+         this->buffer_end_ = this->buffer_ + buffer_size;
+         break;
+       }
+    }
+  if (this->buffer_ == NULL)
+    return;
+
+  section_size_type uncompressed_size = 0;
+  unsigned char* uncompressed_data = NULL;
+  if (object->section_is_compressed(debug_shndx, &uncompressed_size))
+    {
+      uncompressed_data = new unsigned char[uncompressed_size];
+      if (!decompress_input_section(this->buffer_,
+                                   this->buffer_end_ - this->buffer_,
+                                   uncompressed_data,
+                                   uncompressed_size))
+       object->error(_("could not decompress section %s"),
+                     object->section_name(debug_shndx).c_str());
+      this->buffer_ = uncompressed_data;
+      this->buffer_end_ = this->buffer_ + uncompressed_size;
+    }
+
+  // Find the relocation section for ".debug_line".
+  // We expect these for relobjs (.o's) but not dynobjs (.so's).
+  bool got_relocs = false;
+  for (unsigned int reloc_shndx = 0;
+       reloc_shndx < object->shnum();
+       ++reloc_shndx)
+    {
+      unsigned int reloc_sh_type = object->section_type(reloc_shndx);
+      if ((reloc_sh_type == elfcpp::SHT_REL
+          || reloc_sh_type == elfcpp::SHT_RELA)
+         && object->section_info(reloc_shndx) == debug_shndx)
+       {
+         got_relocs = this->track_relocs_.initialize(object, reloc_shndx,
+                                                      reloc_sh_type);
+         this->track_relocs_type_ = reloc_sh_type;
+         break;
+       }
+    }
+
+  // Finally, we need the symtab section to interpret the relocs.
+  if (got_relocs)
+    {
+      unsigned int symtab_shndx;
+      for (symtab_shndx = 0; symtab_shndx < object->shnum(); ++symtab_shndx)
+        if (object->section_type(symtab_shndx) == elfcpp::SHT_SYMTAB)
+          {
+            this->symtab_buffer_ = object->section_contents(
+                symtab_shndx, &this->symtab_buffer_size_, false);
+            break;
+          }
+      if (this->symtab_buffer_ == NULL)
+        return;
+    }
+
+  // Now that we have successfully read all the data, parse the debug
+  // info.
+  this->data_valid_ = true;
+  this->read_line_mappings(object, read_shndx);
+}
+
 // Read the DWARF header.
 
 template<int size, bool big_endian>
 const unsigned char*
-Dwarf_line_info<size, big_endian>::read_header_prolog(
+Sized_dwarf_line_info<size, big_endian>::read_header_prolog(
     const unsigned char* lineptr)
 {
-  uint32_t initial_length = elfcpp::Swap<32, big_endian>::readval(lineptr);
+  uint32_t initial_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
   lineptr += 4;
 
   // In DWARF2/3, if the initial length is all 1 bits, then the offset
@@ -129,7 +154,7 @@ Dwarf_line_info<size, big_endian>::read_header_prolog(
   if (initial_length == 0xffffffff)
     {
       header_.offset_size = 8;
-      initial_length = elfcpp::Swap<64, big_endian>::readval(lineptr);
+      initial_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
       lineptr += 8;
     }
   else
@@ -139,13 +164,13 @@ Dwarf_line_info<size, big_endian>::read_header_prolog(
 
   gold_assert(lineptr + header_.total_length <= buffer_end_);
 
-  header_.version = elfcpp::Swap<16, big_endian>::readval(lineptr);
+  header_.version = elfcpp::Swap_unaligned<16, big_endian>::readval(lineptr);
   lineptr += 2;
 
   if (header_.offset_size == 4)
-    header_.prologue_length = elfcpp::Swap<32, big_endian>::readval(lineptr);
+    header_.prologue_length = elfcpp::Swap_unaligned<32, big_endian>::readval(lineptr);
   else
-    header_.prologue_length = elfcpp::Swap<64, big_endian>::readval(lineptr);
+    header_.prologue_length = elfcpp::Swap_unaligned<64, big_endian>::readval(lineptr);
   lineptr += header_.offset_size;
 
   header_.min_insn_length = *lineptr;
@@ -179,19 +204,32 @@ Dwarf_line_info<size, big_endian>::read_header_prolog(
 
 template<int size, bool big_endian>
 const unsigned char*
-Dwarf_line_info<size, big_endian>::read_header_tables(
+Sized_dwarf_line_info<size, big_endian>::read_header_tables(
     const unsigned char* lineptr)
 {
+  ++this->current_header_index_;
+
+  // Create a new directories_ entry and a new files_ entry for our new
+  // header.  We initialize each with a single empty element, because
+  // dwarf indexes directory and filenames starting at 1.
+  gold_assert(static_cast<int>(this->directories_.size())
+             == this->current_header_index_);
+  gold_assert(static_cast<int>(this->files_.size())
+             == this->current_header_index_);
+  this->directories_.push_back(std::vector<std::string>(1));
+  this->files_.push_back(std::vector<std::pair<int, std::string> >(1));
+
   // It is legal for the directory entry table to be empty.
   if (*lineptr)
     {
       int dirindex = 1;
       while (*lineptr)
         {
-          const unsigned char* dirname = lineptr;
-          gold_assert(dirindex == static_cast<int>(directories_.size()));
-          directories_.push_back(reinterpret_cast<const char*>(dirname));
-          lineptr += directories_.back().size() + 1;
+         const char* dirname = reinterpret_cast<const char*>(lineptr);
+          gold_assert(dirindex
+                     == static_cast<int>(this->directories_.back().size()));
+          this->directories_.back().push_back(dirname);
+          lineptr += this->directories_.back().back().size() + 1;
           dirindex++;
         }
     }
@@ -208,18 +246,21 @@ Dwarf_line_info<size, big_endian>::read_header_tables(
           lineptr += strlen(filename) + 1;
 
           uint64_t dirindex = read_unsigned_LEB_128(lineptr, &len);
-          if (dirindex >= directories_.size())
-            dirindex = 0;
           lineptr += len;
 
+          if (dirindex >= this->directories_.back().size())
+            dirindex = 0;
+         int dirindexi = static_cast<int>(dirindex);
+
           read_unsigned_LEB_128(lineptr, &len);   // mod_time
           lineptr += len;
 
           read_unsigned_LEB_128(lineptr, &len);   // filelength
           lineptr += len;
 
-          gold_assert(fileindex == static_cast<int>(files_.size()));
-          files_.push_back(std::pair<int, std::string>(dirindex, filename));
+          gold_assert(fileindex
+                     == static_cast<int>(this->files_.back().size()));
+          this->files_.back().push_back(std::make_pair(dirindexi, filename));
           fileindex++;
         }
     }
@@ -230,13 +271,9 @@ Dwarf_line_info<size, big_endian>::read_header_tables(
 
 // Process a single opcode in the .debug.line structure.
 
-// Templating on size and big_endian would yield more efficient (and
-// simpler) code, but would bloat the binary.  Speed isn't important
-// here.
-
 template<int size, bool big_endian>
 bool
-Dwarf_line_info<size, big_endian>::process_one_opcode(
+Sized_dwarf_line_info<size, big_endian>::process_one_opcode(
     const unsigned char* start, struct LineStateMachine* lsm, size_t* len)
 {
   size_t oplen = 0;
@@ -314,7 +351,7 @@ Dwarf_line_info<size, big_endian>::process_one_opcode(
     case elfcpp::DW_LNS_fixed_advance_pc:
       {
         int advance_address;
-        advance_address = elfcpp::Swap<16, big_endian>::readval(start);
+        advance_address = elfcpp::Swap_unaligned<16, big_endian>::readval(start);
         oplen += 2;
         lsm->address += advance_address;
       }
@@ -342,17 +379,42 @@ Dwarf_line_info<size, big_endian>::process_one_opcode(
         switch (extended_op)
           {
           case elfcpp::DW_LNE_end_sequence:
+            // This means that the current byte is the one immediately
+            // after a set of instructions.  Record the current line
+            // for up to one less than the current address.
+            lsm->line_num = -1;
             lsm->end_sequence = true;
             *len = oplen;
             return true;
 
           case elfcpp::DW_LNE_set_address:
-            // FIXME: modify the address based on the reloc
-            lsm->address = elfcpp::Swap<size, big_endian>::readval(start);
-            // FIXME: set lsm->shndx from the reloc
-            lsm->shndx = 1;
-            break;
-
+            {
+              lsm->address =
+               elfcpp::Swap_unaligned<size, big_endian>::readval(start);
+              typename Reloc_map::const_iterator it
+                  = this->reloc_map_.find(start - this->buffer_);
+              if (it != reloc_map_.end())
+                {
+                 // If this is a SHT_RELA section, then ignore the
+                 // section contents.  This assumes that this is a
+                 // straight reloc which just uses the reloc addend.
+                 // The reloc addend has already been included in the
+                 // symbol value.
+                 if (this->track_relocs_type_ == elfcpp::SHT_RELA)
+                   lsm->address = 0;
+                 // Add in the symbol value.
+                 lsm->address += it->second.second;
+                  lsm->shndx = it->second.first;
+                }
+              else
+                {
+                  // If we're a normal .o file, with relocs, every
+                  // set_address should have an associated relocation.
+                 if (this->input_is_relobj())
+                    this->data_valid_ = false;
+                }
+              break;
+            }
           case elfcpp::DW_LNE_define_file:
             {
               const char* filename  = reinterpret_cast<const char*>(start);
@@ -360,17 +422,19 @@ Dwarf_line_info<size, big_endian>::process_one_opcode(
               start += templen;
 
               uint64_t dirindex = read_unsigned_LEB_128(start, &templen);
-              if (dirindex >= directories_.size())
-                dirindex = 0;
               oplen += templen;
 
+              if (dirindex >= this->directories_.back().size())
+                dirindex = 0;
+             int dirindexi = static_cast<int>(dirindex);
+
               read_unsigned_LEB_128(start, &templen);   // mod_time
               oplen += templen;
 
               read_unsigned_LEB_128(start, &templen);   // filelength
               oplen += templen;
 
-              files_.push_back(std::pair<int, std::string>(dirindex,
+              this->files_.back().push_back(std::make_pair(dirindexi,
                                                           filename));
             }
             break;
@@ -400,7 +464,8 @@ Dwarf_line_info<size, big_endian>::process_one_opcode(
 
 template<int size, bool big_endian>
 unsigned const char*
-Dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr)
+Sized_dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr,
+                                                    unsigned int shndx)
 {
   struct LineStateMachine lsm;
 
@@ -422,11 +487,23 @@ Dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr)
         {
           size_t oplength;
           bool add_line = this->process_one_opcode(lineptr, &lsm, &oplength);
-          if (add_line)
+          if (add_line
+              && (shndx == -1U || lsm.shndx == -1U || shndx == lsm.shndx))
             {
               Offset_to_lineno_entry entry
-                  = { lsm.address, lsm.file_num, lsm.line_num };
-              line_number_map_[lsm.shndx].push_back(entry);
+                  = { lsm.address, this->current_header_index_,
+                      lsm.file_num, true, lsm.line_num };
+             std::vector<Offset_to_lineno_entry>&
+               map(this->line_number_map_[lsm.shndx]);
+             // If we see two consecutive entries with the same
+             // offset and a real line number, then mark the first
+             // one as non-canonical.
+             if (!map.empty()
+                 && (map.back().offset == static_cast<off_t>(lsm.address))
+                 && lsm.line_num != -1
+                 && map.back().line_num != -1)
+               map.back().last_line_for_offset = false;
+             map.push_back(entry);
             }
           lineptr += oplength;
         }
@@ -435,17 +512,71 @@ Dwarf_line_info<size, big_endian>::read_lines(unsigned const char* lineptr)
   return lengthstart + header_.total_length;
 }
 
+// Looks in the symtab to see what section a symbol is in.
+
+template<int size, bool big_endian>
+unsigned int
+Sized_dwarf_line_info<size, big_endian>::symbol_section(
+    Object* object,
+    unsigned int sym,
+    typename elfcpp::Elf_types<size>::Elf_Addr* value,
+    bool* is_ordinary)
+{
+  const int symsize = elfcpp::Elf_sizes<size>::sym_size;
+  gold_assert(sym * symsize < this->symtab_buffer_size_);
+  elfcpp::Sym<size, big_endian> elfsym(this->symtab_buffer_ + sym * symsize);
+  *value = elfsym.get_st_value();
+  return object->adjust_sym_shndx(sym, elfsym.get_st_shndx(), is_ordinary);
+}
+
+// Read the relocations into a Reloc_map.
+
 template<int size, bool big_endian>
 void
-Dwarf_line_info<size, big_endian>::read_line_mappings()
+Sized_dwarf_line_info<size, big_endian>::read_relocs(Object* object)
 {
-  while (buffer_ < buffer_end_)
+  if (this->symtab_buffer_ == NULL)
+    return;
+
+  typename elfcpp::Elf_types<size>::Elf_Addr value;
+  off_t reloc_offset;
+  while ((reloc_offset = this->track_relocs_.next_offset()) != -1)
     {
-      const unsigned char* lineptr = buffer_;
+      const unsigned int sym = this->track_relocs_.next_symndx();
+
+      bool is_ordinary;
+      const unsigned int shndx = this->symbol_section(object, sym, &value,
+                                                     &is_ordinary);
+
+      // There is no reason to record non-ordinary section indexes, or
+      // SHN_UNDEF, because they will never match the real section.
+      if (is_ordinary && shndx != elfcpp::SHN_UNDEF)
+       {
+         value += this->track_relocs_.next_addend();
+         this->reloc_map_[reloc_offset] = std::make_pair(shndx, value);
+       }
+
+      this->track_relocs_.advance(reloc_offset + 1);
+    }
+}
+
+// Read the line number info.
+
+template<int size, bool big_endian>
+void
+Sized_dwarf_line_info<size, big_endian>::read_line_mappings(Object* object,
+                                                           unsigned int shndx)
+{
+  gold_assert(this->data_valid_ == true);
+
+  this->read_relocs(object);
+  while (this->buffer_ < this->buffer_end_)
+    {
+      const unsigned char* lineptr = this->buffer_;
       lineptr = this->read_header_prolog(lineptr);
       lineptr = this->read_header_tables(lineptr);
-      lineptr = this->read_lines(lineptr);
-      buffer_ = lineptr;
+      lineptr = this->read_lines(lineptr, shndx);
+      this->buffer_ = lineptr;
     }
 
   // Sort the lines numbers, so addr2line can use binary search.
@@ -453,37 +584,206 @@ Dwarf_line_info<size, big_endian>::read_line_mappings()
        it != line_number_map_.end();
        ++it)
     // Each vector needs to be sorted by offset.
-    sort(it->second.begin(), it->second.end());
+    std::sort(it->second.begin(), it->second.end());
 }
 
-// Return a string for a file name and line number.
+// Some processing depends on whether the input is a .o file or not.
+// For instance, .o files have relocs, and have .debug_lines
+// information on a per section basis.  .so files, on the other hand,
+// lack relocs, and offsets are unique, so we can ignore the section
+// information.
 
 template<int size, bool big_endian>
-std::string
-Dwarf_line_info<size, big_endian>::addr2line(unsigned int shndx, off_t offset)
+bool
+Sized_dwarf_line_info<size, big_endian>::input_is_relobj()
 {
-  const Offset_to_lineno_entry lookup_key = { offset, 0, 0 };
-  std::vector<Offset_to_lineno_entry>& offsets = line_number_map_[shndx];
-  typename std::vector<Offset_to_lineno_entry>::const_iterator it
-      = std::lower_bound(offsets.begin(), offsets.end(), lookup_key);
+  // Only .o files have relocs and the symtab buffer that goes with them.
+  return this->symtab_buffer_ != NULL;
+}
 
-  // If we found an exact match, great, otherwise find the last entry
-  // before the passed-in offset.
-  if (it->offset > offset)
+// Given an Offset_to_lineno_entry vector, and an offset, figure out
+// if the offset points into a function according to the vector (see
+// comments below for the algorithm).  If it does, return an iterator
+// into the vector that points to the line-number that contains that
+// offset.  If not, it returns vector::end().
+
+static std::vector<Offset_to_lineno_entry>::const_iterator
+offset_to_iterator(const std::vector<Offset_to_lineno_entry>* offsets,
+                   off_t offset)
+{
+  const Offset_to_lineno_entry lookup_key = { offset, 0, 0, true, 0 };
+
+  // lower_bound() returns the smallest offset which is >= lookup_key.
+  // If no offset in offsets is >= lookup_key, returns end().
+  std::vector<Offset_to_lineno_entry>::const_iterator it
+      = std::lower_bound(offsets->begin(), offsets->end(), lookup_key);
+
+  // This code is easiest to understand with a concrete example.
+  // Here's a possible offsets array:
+  // {{offset = 3211, header_num = 0, file_num = 1, last, line_num = 16},  // 0
+  //  {offset = 3224, header_num = 0, file_num = 1, last, line_num = 20},  // 1
+  //  {offset = 3226, header_num = 0, file_num = 1, last, line_num = 22},  // 2
+  //  {offset = 3231, header_num = 0, file_num = 1, last, line_num = 25},  // 3
+  //  {offset = 3232, header_num = 0, file_num = 1, last, line_num = -1},  // 4
+  //  {offset = 3232, header_num = 0, file_num = 1, last, line_num = 65},  // 5
+  //  {offset = 3235, header_num = 0, file_num = 1, last, line_num = 66},  // 6
+  //  {offset = 3236, header_num = 0, file_num = 1, last, line_num = -1},  // 7
+  //  {offset = 5764, header_num = 0, file_num = 1, last, line_num = 48},  // 8
+  //  {offset = 5764, header_num = 0, file_num = 1,!last, line_num = 47},  // 9
+  //  {offset = 5765, header_num = 0, file_num = 1, last, line_num = 49},  // 10
+  //  {offset = 5767, header_num = 0, file_num = 1, last, line_num = 50},  // 11
+  //  {offset = 5768, header_num = 0, file_num = 1, last, line_num = 51},  // 12
+  //  {offset = 5773, header_num = 0, file_num = 1, last, line_num = -1},  // 13
+  //  {offset = 5787, header_num = 1, file_num = 1, last, line_num = 19},  // 14
+  //  {offset = 5790, header_num = 1, file_num = 1, last, line_num = 20},  // 15
+  //  {offset = 5793, header_num = 1, file_num = 1, last, line_num = 67},  // 16
+  //  {offset = 5793, header_num = 1, file_num = 1, last, line_num = -1},  // 17
+  //  {offset = 5793, header_num = 1, file_num = 1,!last, line_num = 66},  // 18
+  //  {offset = 5795, header_num = 1, file_num = 1, last, line_num = 68},  // 19
+  //  {offset = 5798, header_num = 1, file_num = 1, last, line_num = -1},  // 20
+  // The entries with line_num == -1 mark the end of a function: the
+  // associated offset is one past the last instruction in the
+  // function.  This can correspond to the beginning of the next
+  // function (as is true for offset 3232); alternately, there can be
+  // a gap between the end of one function and the start of the next
+  // (as is true for some others, most obviously from 3236->5764).
+  //
+  // Case 1: lookup_key has offset == 10.  lower_bound returns
+  //         offsets[0].  Since it's not an exact match and we're
+  //         at the beginning of offsets, we return end() (invalid).
+  // Case 2: lookup_key has offset 10000.  lower_bound returns
+  //         offset[21] (end()).  We return end() (invalid).
+  // Case 3: lookup_key has offset == 3211.  lower_bound matches
+  //         offsets[0] exactly, and that's the entry we return.
+  // Case 4: lookup_key has offset == 3232.  lower_bound returns
+  //         offsets[4].  That's an exact match, but indicates
+  //         end-of-function.  We check if offsets[5] is also an
+  //         exact match but not end-of-function.  It is, so we
+  //         return offsets[5].
+  // Case 5: lookup_key has offset == 3214.  lower_bound returns
+  //         offsets[1].  Since it's not an exact match, we back
+  //         up to the offset that's < lookup_key, offsets[0].
+  //         We note offsets[0] is a valid entry (not end-of-function),
+  //         so that's the entry we return.
+  // Case 6: lookup_key has offset == 4000.  lower_bound returns
+  //         offsets[8].  Since it's not an exact match, we back
+  //         up to offsets[7].  Since offsets[7] indicates
+  //         end-of-function, we know lookup_key is between
+  //         functions, so we return end() (not a valid offset).
+  // Case 7: lookup_key has offset == 5794.  lower_bound returns
+  //         offsets[19].  Since it's not an exact match, we back
+  //         up to offsets[16].  Note we back up to the *first*
+  //         entry with offset 5793, not just offsets[19-1].
+  //         We note offsets[16] is a valid entry, so we return it.
+  //         If offsets[16] had had line_num == -1, we would have
+  //         checked offsets[17].  The reason for this is that
+  //         16 and 17 can be in an arbitrary order, since we sort
+  //         only by offset and last_line_for_offset.  (Note it
+  //         doesn't help to use line_number as a tertiary sort key,
+  //         since sometimes we want the -1 to be first and sometimes
+  //         we want it to be last.)
+
+  // This deals with cases (1) and (2).
+  if ((it == offsets->begin() && offset < it->offset)
+      || it == offsets->end())
+    return offsets->end();
+
+  // This deals with cases (3) and (4).
+  if (offset == it->offset)
     {
-      if (it == offsets.begin())
-        return "";
-      --it;
-      gold_assert(it->offset < offset);
+      while (it != offsets->end()
+             && it->offset == offset
+             && it->line_num == -1)
+        ++it;
+      if (it == offsets->end() || it->offset != offset)
+        return offsets->end();
+      else
+        return it;
     }
 
-  // Convert the file_num + line_num into a string.
+  // This handles the first part of case (7) -- we back up to the
+  // *first* entry that has the offset that's behind us.
+  gold_assert(it != offsets->begin());
+  std::vector<Offset_to_lineno_entry>::const_iterator range_end = it;
+  --it;
+  const off_t range_value = it->offset;
+  while (it != offsets->begin() && (it-1)->offset == range_value)
+    --it;
+
+  // This handles cases (5), (6), and (7): if any entry in the
+  // equal_range [it, range_end) has a line_num != -1, it's a valid
+  // match.  If not, we're not in a function.  The line number we saw
+  // last for an offset will be sorted first, so it'll get returned if
+  // it's present.
+  for (; it != range_end; ++it)
+    if (it->line_num != -1)
+      return it;
+  return offsets->end();
+}
+
+// Returns the canonical filename:lineno for the address passed in.
+// If other_lines is not NULL, appends the non-canonical lines
+// assigned to the same address.
+
+template<int size, bool big_endian>
+std::string
+Sized_dwarf_line_info<size, big_endian>::do_addr2line(
+    unsigned int shndx,
+    off_t offset,
+    std::vector<std::string>* other_lines)
+{
+  if (this->data_valid_ == false)
+    return "";
+
+  const std::vector<Offset_to_lineno_entry>* offsets;
+  // If we do not have reloc information, then our input is a .so or
+  // some similar data structure where all the information is held in
+  // the offset.  In that case, we ignore the input shndx.
+  if (this->input_is_relobj())
+    offsets = &this->line_number_map_[shndx];
+  else
+    offsets = &this->line_number_map_[-1U];
+  if (offsets->empty())
+    return "";
+
+  typename std::vector<Offset_to_lineno_entry>::const_iterator it
+      = offset_to_iterator(offsets, offset);
+  if (it == offsets->end())
+    return "";
+
+  std::string result = this->format_file_lineno(*it);
+  if (other_lines != NULL)
+    for (++it; it != offsets->end() && it->offset == offset; ++it)
+      {
+        if (it->line_num == -1)
+          continue;  // The end of a previous function.
+        other_lines->push_back(this->format_file_lineno(*it));
+      }
+  return result;
+}
+
+// Convert the file_num + line_num into a string.
+
+template<int size, bool big_endian>
+std::string
+Sized_dwarf_line_info<size, big_endian>::format_file_lineno(
+    const Offset_to_lineno_entry& loc) const
+{
   std::string ret;
-  gold_assert(it->file_num < static_cast<int>(files_.size()));
-  const std::pair<int, std::string>& filename_pair = files_[it->file_num];
-  gold_assert(filename_pair.first < static_cast<int>(directories_.size()));
-  const std::string& dirname = directories_[filename_pair.first];
+
+  gold_assert(loc.header_num < static_cast<int>(this->files_.size()));
+  gold_assert(loc.file_num
+             < static_cast<int>(this->files_[loc.header_num].size()));
+  const std::pair<int, std::string>& filename_pair
+      = this->files_[loc.header_num][loc.file_num];
   const std::string& filename = filename_pair.second;
+
+  gold_assert(loc.header_num < static_cast<int>(this->directories_.size()));
+  gold_assert(filename_pair.first
+              < static_cast<int>(this->directories_[loc.header_num].size()));
+  const std::string& dirname
+      = this->directories_[loc.header_num][filename_pair.first];
+
   if (!dirname.empty())
     {
       ret += dirname;
@@ -494,31 +794,150 @@ Dwarf_line_info<size, big_endian>::addr2line(unsigned int shndx, off_t offset)
     ret = "(unknown)";
 
   char buffer[64];   // enough to hold a line number
-  snprintf(buffer, sizeof(buffer), "%d", it->line_num);
+  snprintf(buffer, sizeof(buffer), "%d", loc.line_num);
   ret += ":";
   ret += buffer;
 
   return ret;
 }
 
+// Dwarf_line_info routines.
+
+static unsigned int next_generation_count = 0;
+
+struct Addr2line_cache_entry
+{
+  Object* object;
+  unsigned int shndx;
+  Dwarf_line_info* dwarf_line_info;
+  unsigned int generation_count;
+  unsigned int access_count;
+
+  Addr2line_cache_entry(Object* o, unsigned int s, Dwarf_line_info* d)
+      : object(o), shndx(s), dwarf_line_info(d),
+        generation_count(next_generation_count), access_count(0)
+  {
+    if (next_generation_count < (1U << 31))
+      ++next_generation_count;
+  }
+};
+// We expect this cache to be small, so don't bother with a hashtable
+// or priority queue or anything: just use a simple vector.
+static std::vector<Addr2line_cache_entry> addr2line_cache;
+
+std::string
+Dwarf_line_info::one_addr2line(Object* object,
+                               unsigned int shndx, off_t offset,
+                               size_t cache_size,
+                               std::vector<std::string>* other_lines)
+{
+  Dwarf_line_info* lineinfo = NULL;
+  std::vector<Addr2line_cache_entry>::iterator it;
+
+  // First, check the cache.  If we hit, update the counts.
+  for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
+    {
+      if (it->object == object && it->shndx == shndx)
+        {
+          lineinfo = it->dwarf_line_info;
+          it->generation_count = next_generation_count;
+          // We cap generation_count at 2^31 -1 to avoid overflow.
+          if (next_generation_count < (1U << 31))
+            ++next_generation_count;
+          // We cap access_count at 31 so 2^access_count doesn't overflow
+          if (it->access_count < 31)
+            ++it->access_count;
+          break;
+        }
+    }
+
+  // If we don't hit the cache, create a new object and insert into the
+  // cache.
+  if (lineinfo == NULL)
+  {
+    switch (parameters->size_and_endianness())
+      {
+#ifdef HAVE_TARGET_32_LITTLE
+        case Parameters::TARGET_32_LITTLE:
+          lineinfo = new Sized_dwarf_line_info<32, false>(object, shndx); break;
+#endif
+#ifdef HAVE_TARGET_32_BIG
+        case Parameters::TARGET_32_BIG:
+          lineinfo = new Sized_dwarf_line_info<32, true>(object, shndx); break;
+#endif
+#ifdef HAVE_TARGET_64_LITTLE
+        case Parameters::TARGET_64_LITTLE:
+          lineinfo = new Sized_dwarf_line_info<64, false>(object, shndx); break;
+#endif
+#ifdef HAVE_TARGET_64_BIG
+        case Parameters::TARGET_64_BIG:
+          lineinfo = new Sized_dwarf_line_info<64, true>(object, shndx); break;
+#endif
+        default:
+          gold_unreachable();
+      }
+    addr2line_cache.push_back(Addr2line_cache_entry(object, shndx, lineinfo));
+  }
+
+  // Now that we have our object, figure out the answer
+  std::string retval = lineinfo->addr2line(shndx, offset, other_lines);
+
+  // Finally, if our cache has grown too big, delete old objects.  We
+  // assume the common (probably only) case is deleting only one object.
+  // We use a pretty simple scheme to evict: function of LRU and MFU.
+  while (addr2line_cache.size() > cache_size)
+    {
+      unsigned int lowest_score = ~0U;
+      std::vector<Addr2line_cache_entry>::iterator lowest
+          = addr2line_cache.end();
+      for (it = addr2line_cache.begin(); it != addr2line_cache.end(); ++it)
+        {
+          const unsigned int score = (it->generation_count
+                                      + (1U << it->access_count));
+          if (score < lowest_score)
+            {
+              lowest_score = score;
+              lowest = it;
+            }
+        }
+      if (lowest != addr2line_cache.end())
+        {
+          delete lowest->dwarf_line_info;
+          addr2line_cache.erase(lowest);
+        }
+    }
+
+  return retval;
+}
+
+void
+Dwarf_line_info::clear_addr2line_cache()
+{
+  for (std::vector<Addr2line_cache_entry>::iterator it = addr2line_cache.begin();
+       it != addr2line_cache.end();
+       ++it)
+    delete it->dwarf_line_info;
+  addr2line_cache.clear();
+}
+
 #ifdef HAVE_TARGET_32_LITTLE
 template
-class Dwarf_line_info<32, false>;
+class Sized_dwarf_line_info<32, false>;
 #endif
 
 #ifdef HAVE_TARGET_32_BIG
 template
-class Dwarf_line_info<32, true>;
+class Sized_dwarf_line_info<32, true>;
 #endif
 
 #ifdef HAVE_TARGET_64_LITTLE
 template
-class Dwarf_line_info<64, false>;
+class Sized_dwarf_line_info<64, false>;
 #endif
 
 #ifdef HAVE_TARGET_64_BIG
 template
-class Dwarf_line_info<64, true>;
+class Sized_dwarf_line_info<64, true>;
 #endif
 
 } // End namespace gold.
This page took 0.032341 seconds and 4 git commands to generate.