Update year range in copyright notice of all files.
[external/binutils.git] / gold / icf.cc
index a132875..a49ce2e 100644 (file)
@@ -1,6 +1,6 @@
 // icf.cc -- Identical Code Folding.
 //
-// Copyright 2009, 2010 Free Software Foundation, Inc.
+// Copyright (C) 2009-2017 Free Software Foundation, Inc.
 // Written by Sriraman Tallam <tmsriram@google.com>.
 
 // This file is part of gold.
@@ -213,6 +213,45 @@ preprocess_for_unique_sections(const std::vector<Section_id>& id_section,
     }
 }
 
+// For SHF_MERGE sections that use REL relocations, the addend is stored in
+// the text section at the relocation offset.  Read  the addend value given
+// the pointer to the addend in the text section and the addend size.
+// Update the addend value if a valid addend is found.
+// Parameters:
+// RELOC_ADDEND_PTR   : Pointer to the addend in the text section.
+// ADDEND_SIZE        : The size of the addend.
+// RELOC_ADDEND_VALUE : Pointer to the addend that is updated.
+
+inline void
+get_rel_addend(const unsigned char* reloc_addend_ptr,
+              const unsigned int addend_size,
+              uint64_t* reloc_addend_value)
+{
+  switch (addend_size)
+    {
+    case 0:
+      break;
+    case 1:
+      *reloc_addend_value =
+        read_from_pointer<8>(reloc_addend_ptr);
+      break;
+    case 2:
+      *reloc_addend_value =
+          read_from_pointer<16>(reloc_addend_ptr);
+      break;
+    case 4:
+      *reloc_addend_value =
+        read_from_pointer<32>(reloc_addend_ptr);
+      break;
+    case 8:
+      *reloc_addend_value =
+        read_from_pointer<64>(reloc_addend_ptr);
+      break;
+    default:
+      gold_unreachable();
+    }
+}
+
 // This returns the buffer containing the section's contents, both
 // text and relocs.  Relocs are differentiated as those pointing to
 // sections that could be folded and those that cannot.  Only relocs
@@ -237,20 +276,16 @@ get_section_contents(bool first_iteration,
                      const std::vector<unsigned int>& kept_section_id,
                      std::vector<std::string>* section_contents)
 {
+  // Lock the object so we can read from it.  This is only called
+  // single-threaded from queue_middle_tasks, so it is OK to lock.
+  // Unfortunately we have no way to pass in a Task token.
+  const Task* dummy_task = reinterpret_cast<const Task*>(-1);
+  Task_lock_obj<Object> tl(dummy_task, secn.first);
+
   section_size_type plen;
   const unsigned char* contents = NULL;
-
   if (first_iteration)
-    {
-      // Lock the object so we can read from it.  This is only called
-      // single-threaded from queue_middle_tasks, so it is OK to lock.
-      // Unfortunately we have no way to pass in a Task token.
-      const Task* dummy_task = reinterpret_cast<const Task*>(-1);
-      Task_lock_obj<Object> tl(dummy_task, secn.first);
-      contents = secn.first->section_contents(secn.second,
-                                              &plen,
-                                              false);
-    }
+    contents = secn.first->section_contents(secn.second, &plen, false);
 
   // The buffer to hold all the contents including relocs.  A checksum
   // is then computed on this buffer.
@@ -273,27 +308,46 @@ get_section_contents(bool first_iteration,
 
   if (it_reloc_info_list != reloc_info_list.end())
     {
-      Icf::Sections_reachable_info v =
+      Icf::Sections_reachable_info &v =
         (it_reloc_info_list->second).section_info;
       // Stores the information of the symbol pointed to by the reloc.
-      Icf::Symbol_info s = (it_reloc_info_list->second).symbol_info;
+      const Icf::Symbol_info &s = (it_reloc_info_list->second).symbol_info;
       // Stores the addend and the symbol value.
-      Icf::Addend_info a = (it_reloc_info_list->second).addend_info;
+      Icf::Addend_info &a = (it_reloc_info_list->second).addend_info;
       // Stores the offset of the reloc.
-      Icf::Offset_info o = (it_reloc_info_list->second).offset_info;
-      Icf::Reloc_addend_size_info reloc_addend_size_info =
+      const Icf::Offset_info &o = (it_reloc_info_list->second).offset_info;
+      const Icf::Reloc_addend_size_info &reloc_addend_size_info =
         (it_reloc_info_list->second).reloc_addend_size_info;
       Icf::Sections_reachable_info::iterator it_v = v.begin();
-      Icf::Symbol_info::iterator it_s = s.begin();
+      Icf::Symbol_info::const_iterator it_s = s.begin();
       Icf::Addend_info::iterator it_a = a.begin();
-      Icf::Offset_info::iterator it_o = o.begin();
-      Icf::Reloc_addend_size_info::iterator it_addend_size =
+      Icf::Offset_info::const_iterator it_o = o.begin();
+      Icf::Reloc_addend_size_info::const_iterator it_addend_size =
         reloc_addend_size_info.begin();
 
       for (; it_v != v.end(); ++it_v, ++it_s, ++it_a, ++it_o, ++it_addend_size)
         {
+         if (first_iteration
+             && it_v->first != NULL)
+           {
+             Symbol_location loc;
+             loc.object = it_v->first;
+             loc.shndx = it_v->second;
+             loc.offset = convert_types<off_t, long long>(it_a->first
+                                                          + it_a->second);
+             // Look through function descriptors
+             parameters->target().function_location(&loc);
+             if (loc.shndx != it_v->second)
+               {
+                 it_v->second = loc.shndx;
+                 // Modify symvalue/addend to the code entry.
+                 it_a->first = loc.offset;
+                 it_a->second = 0;
+               }
+           }
+
           // ADDEND_STR stores the symbol value and addend and offset,
-          // each atmost 16 hex digits long.  it_a points to a pair
+          // each at most 16 hex digits long.  it_a points to a pair
           // where first is the symbol value and second is the
           // addend.
           char addend_str[50];
@@ -373,73 +427,45 @@ get_section_contents(bool first_iteration,
               if (!first_iteration)
                 continue;
 
-              // Lock the object so we can read from it.  This is only called
-              // single-threaded from queue_middle_tasks, so it is OK to lock.
-              // Unfortunately we have no way to pass in a Task token.
-              const Task* dummy_task = reinterpret_cast<const Task*>(-1);
-              Task_lock_obj<Object> tl(dummy_task, it_v->first);
-
               uint64_t secn_flags = (it_v->first)->section_flags(it_v->second);
               // This reloc points to a merge section.  Hash the
               // contents of this section.
               if ((secn_flags & elfcpp::SHF_MERGE) != 0
-                 && parameters->target().can_icf_inline_merge_sections ())
+                 && parameters->target().can_icf_inline_merge_sections())
                 {
                   uint64_t entsize =
                     (it_v->first)->section_entsize(it_v->second);
                  long long offset = it_a->first;
-
-                  unsigned long long addend = it_a->second;
-                  // Ignoring the addend when it is a negative value.  See the 
-                  // comments in Merged_symbol_value::Value in object.h.
-                  if (addend < 0xffffff00)
-                    offset = offset + addend;
-
-                 // For SHT_REL relocation sections, the addend is stored in the
-                 // text section at the relocation offset.
-                 uint64_t reloc_addend_value = 0;
+                 // Handle SHT_RELA and SHT_REL addends, only one of these
+                 // addends exists.
+                 // Get the SHT_RELA addend.  For RELA relocations, we have
+                 // the addend from the relocation.
+                 uint64_t reloc_addend_value = it_a->second;
+
+                 // Handle SHT_REL addends.
+                 // For REL relocations, we need to fetch the addend from the
+                 // section contents.
                   const unsigned char* reloc_addend_ptr =
                    contents + static_cast<unsigned long long>(*it_o);
-                 switch(*it_addend_size)
-                   {
-                     case 0:
-                       {
-                          break;
-                        }
-                      case 1:
-                        {
-                          reloc_addend_value =
-                            read_from_pointer<8>(reloc_addend_ptr);
-                         break;
-                        }
-                      case 2:
-                        {
-                          reloc_addend_value =
-                            read_from_pointer<16>(reloc_addend_ptr);
-                         break;
-                        }
-                      case 4:
-                        {
-                          reloc_addend_value =
-                            read_from_pointer<32>(reloc_addend_ptr);
-                         break;
-                        }
-                      case 8:
-                        {
-                          reloc_addend_value =
-                            read_from_pointer<64>(reloc_addend_ptr);
-                         break;
-                        }
-                     default:
-                       gold_unreachable();
-                   }
-                 offset = offset + reloc_addend_value;
+
+                 // Update the addend value with the SHT_REL addend if
+                 // available.
+                 get_rel_addend(reloc_addend_ptr, *it_addend_size,
+                                &reloc_addend_value);
+
+                 // Ignore the addend when it is a negative value.  See the
+                 // comments in Merged_symbol_value::value in object.h.
+                 if (reloc_addend_value < 0xffffff00)
+                   offset = offset + reloc_addend_value;
 
                   section_size_type secn_len;
+
                   const unsigned char* str_contents =
                   (it_v->first)->section_contents(it_v->second,
                                                   &secn_len,
                                                   false) + offset;
+                 gold_assert (offset < (long long) secn_len);
+
                   if ((secn_flags & elfcpp::SHF_STRINGS) != 0)
                     {
                       // String merge section.
@@ -480,10 +506,14 @@ get_section_contents(bool first_iteration,
                     }
                   else
                     {
-                      // Use the entsize to determine the length.
-                      buffer.append(reinterpret_cast<const 
+                      // Use the entsize to determine the length to copy.
+                     uint64_t bufsize = entsize;
+                     // If entsize is too big, copy all the remaining bytes.
+                     if ((offset + entsize) > secn_len)
+                       bufsize = secn_len - offset;
+                      buffer.append(reinterpret_cast<const
                                                      char*>(str_contents),
-                                    entsize);
+                                    bufsize);
                     }
                  buffer.append("@");
                 }
@@ -515,7 +545,7 @@ get_section_contents(bool first_iteration,
     {
       buffer.append("Contents = ");
       buffer.append(reinterpret_cast<const char*>(contents), plen);
-      // Store the section contents that dont change to avoid recomputing
+      // Store the section contents that don't change to avoid recomputing
       // during the next call to this function.
       (*section_contents)[section_num] = buffer;
     }
@@ -550,7 +580,7 @@ get_section_contents(bool first_iteration,
 // KEPT_SECTION_ID    : Vector which maps folded sections to kept sections.
 // ID_SECTION         : Vector mapping a section to an unique integer.
 // IS_SECN_OR_GROUP_UNIQUE : To check if a section or a group of identical
-//                            sectionsis already known to be unique.
+//                            sections is already known to be unique.
 // SECTION_CONTENTS   : Store the section's text and relocs to non-ICF
 //                      sections.
 
@@ -560,6 +590,7 @@ match_sections(unsigned int iteration_num,
                std::vector<unsigned int>* num_tracked_relocs,
                std::vector<unsigned int>* kept_section_id,
                const std::vector<Section_id>& id_section,
+              const std::vector<uint64_t>& section_addraligns,
                std::vector<bool>* is_secn_or_group_unique,
                std::vector<std::string>* section_contents)
 {
@@ -600,13 +631,7 @@ match_sections(unsigned int iteration_num,
         {
           if ((*kept_section_id)[i] != i)
             {
-              // This section is already folded into something.  See
-              // if it should point to a different kept section.
-              unsigned int kept_section = (*kept_section_id)[i];
-              if (kept_section != (*kept_section_id)[kept_section])
-                {
-                  (*kept_section_id)[i] = (*kept_section_id)[kept_section];
-                }
+              // This section is already folded into something.
               continue;
             }
           this_secn_contents = get_section_contents(false, secn, i, NULL,
@@ -641,7 +666,25 @@ match_sections(unsigned int iteration_num,
                          this_secn_contents.c_str(),
                          this_secn_contents.length()) != 0)
                   continue;
-              (*kept_section_id)[i] = kept_section;
+
+             // Check section alignment here.
+             // The section with the larger alignment requirement
+             // should be kept.  We assume alignment can only be 
+             // zero or positive integral powers of two.
+             uint64_t align_i = section_addraligns[i];
+             uint64_t align_kept = section_addraligns[kept_section];
+             if (align_i <= align_kept)
+               {
+                 (*kept_section_id)[i] = kept_section;
+               }
+             else
+               {
+                 (*kept_section_id)[kept_section] = i;
+                 it->second = i;
+                 full_section_contents[kept_section].swap(
+                     full_section_contents[i]);
+               }
+
               converged = false;
               break;
             }
@@ -658,20 +701,41 @@ match_sections(unsigned int iteration_num,
         (*is_secn_or_group_unique)[i] = true;
     }
 
+  // If a section was folded into another section that was later folded
+  // again then the former has to be updated.
+  for (unsigned int i = 0; i < id_section.size(); i++)
+    {
+      // Find the end of the folding chain
+      unsigned int kept = i;
+      while ((*kept_section_id)[kept] != kept)
+        {
+          kept = (*kept_section_id)[kept];
+        }
+      // Update every element of the chain
+      unsigned int current = i;
+      while ((*kept_section_id)[current] != kept)
+        {
+          unsigned int next = (*kept_section_id)[current];
+          (*kept_section_id)[current] = kept;
+          current = next;
+        }
+    }
+
   return converged;
 }
 
 // During safe icf (--icf=safe), only fold functions that are ctors or dtors.
-// This function returns true if the mangled function name is a ctor or a
-// dtor.
+// This function returns true if the section name is that of a ctor or a dtor.
 
 static bool
-is_function_ctor_or_dtor(const char* mangled_func_name)
+is_function_ctor_or_dtor(const std::string& section_name)
 {
-  if ((is_prefix_of("_ZN", mangled_func_name)
-       || is_prefix_of("_ZZ", mangled_func_name))
-      && (is_gnu_v3_mangled_ctor(mangled_func_name)
-          || is_gnu_v3_mangled_dtor(mangled_func_name)))
+  const char* mangled_func_name = strrchr(section_name.c_str(), '.');
+  gold_assert(mangled_func_name != NULL);
+  if ((is_prefix_of("._ZN", mangled_func_name)
+       || is_prefix_of("._ZZ", mangled_func_name))
+      && (is_gnu_v3_mangled_ctor(mangled_func_name + 1)
+          || is_gnu_v3_mangled_dtor(mangled_func_name + 1)))
     {
       return true;
     }
@@ -688,6 +752,7 @@ Icf::find_identical_sections(const Input_objects* input_objects,
 {
   unsigned int section_num = 0;
   std::vector<unsigned int> num_tracked_relocs;
+  std::vector<uint64_t> section_addraligns;
   std::vector<bool> is_secn_or_group_unique;
   std::vector<std::string> section_contents;
   const Target& target = parameters->target();
@@ -706,7 +771,7 @@ Icf::find_identical_sections(const Input_objects* input_objects,
 
       for (unsigned int i = 0;i < (*p)->shnum(); ++i)
         {
-         const char* section_name = (*p)->section_name(i).c_str();
+         const std::string section_name = (*p)->section_name(i);
           if (!is_section_foldable_candidate(section_name))
             continue;
           if (!(*p)->is_section_included(i))
@@ -714,13 +779,11 @@ Icf::find_identical_sections(const Input_objects* input_objects,
           if (parameters->options().gc_sections()
               && symtab->gc()->is_section_garbage(*p, i))
               continue;
-         const char* mangled_func_name = strrchr(section_name, '.');
-         gold_assert(mangled_func_name != NULL);
          // With --icf=safe, check if the mangled function name is a ctor
          // or a dtor.  The mangled function name can be obtained from the
          // section name by stripping the section prefix.
          if (parameters->options().icf_safe_folding()
-              && !is_function_ctor_or_dtor(mangled_func_name + 1)
+              && !is_function_ctor_or_dtor(section_name)
              && (!target.can_check_for_function_pointers()
                   || section_has_function_pointers(*p, i)))
             {
@@ -730,6 +793,7 @@ Icf::find_identical_sections(const Input_objects* input_objects,
           this->section_id_[Section_id(*p, i)] = section_num;
           this->kept_section_id_.push_back(section_num);
           num_tracked_relocs.push_back(0);
+         section_addraligns.push_back((*p)->section_addralign(i));
           is_secn_or_group_unique.push_back(false);
           section_contents.push_back("");
           section_num++;
@@ -750,8 +814,8 @@ Icf::find_identical_sections(const Input_objects* input_objects,
       num_iterations++;
       converged = match_sections(num_iterations, symtab,
                                  &num_tracked_relocs, &this->kept_section_id_,
-                                 this->id_section_, &is_secn_or_group_unique,
-                                 &section_contents);
+                                 this->id_section_, section_addraligns,
+                                 &is_secn_or_group_unique, &section_contents);
     }
 
   if (parameters->options().print_icf_sections())
@@ -779,7 +843,7 @@ Icf::find_identical_sections(const Input_objects* input_objects,
       else if (sym->source() == Symbol::FROM_OBJECT 
                && !sym->object()->is_dynamic())
         {
-          Object* obj = sym->object();
+          Relobj* obj = static_cast<Relobj*>(sym->object());
           bool is_ordinary;
           unsigned int shndx = sym->shndx(&is_ordinary);
           if (is_ordinary)
@@ -796,7 +860,7 @@ Icf::find_identical_sections(const Input_objects* input_objects,
 // Unfolds the section denoted by OBJ and SHNDX if folded.
 
 void
-Icf::unfold_section(Object* obj, unsigned int shndx)
+Icf::unfold_section(Relobj* obj, unsigned int shndx)
 {
   Section_id secn(obj, shndx);
   Uniq_secn_id_map::iterator it = this->section_id_.find(secn);
@@ -813,7 +877,7 @@ Icf::unfold_section(Object* obj, unsigned int shndx)
 // is different from this section.
 
 bool
-Icf::is_section_folded(Object* obj, unsigned int shndx)
+Icf::is_section_folded(Relobj* obj, unsigned int shndx)
 {
   Section_id secn(obj, shndx);
   Uniq_secn_id_map::iterator it = this->section_id_.find(secn);
@@ -827,7 +891,7 @@ Icf::is_section_folded(Object* obj, unsigned int shndx)
 // This function returns the folded section for the given section.
 
 Section_id
-Icf::get_folded_section(Object* dup_obj, unsigned int dup_shndx)
+Icf::get_folded_section(Relobj* dup_obj, unsigned int dup_shndx)
 {
   Section_id dup_secn(dup_obj, dup_shndx);
   Uniq_secn_id_map::iterator it = this->section_id_.find(dup_secn);