Home | History | Annotate | Download | only in gold
      1 // stringpool.cc -- a string pool for gold
      2 
      3 // Copyright (C) 2006-2014 Free Software Foundation, Inc.
      4 // Written by Ian Lance Taylor <iant (at) google.com>.
      5 
      6 // This file is part of gold.
      7 
      8 // This program is free software; you can redistribute it and/or modify
      9 // it under the terms of the GNU General Public License as published by
     10 // the Free Software Foundation; either version 3 of the License, or
     11 // (at your option) any later version.
     12 
     13 // This program is distributed in the hope that it will be useful,
     14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
     15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
     16 // GNU General Public License for more details.
     17 
     18 // You should have received a copy of the GNU General Public License
     19 // along with this program; if not, write to the Free Software
     20 // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
     21 // MA 02110-1301, USA.
     22 
     23 #include "gold.h"
     24 
     25 #include <cstring>
     26 #include <algorithm>
     27 #include <vector>
     28 
     29 #include "output.h"
     30 #include "parameters.h"
     31 #include "stringpool.h"
     32 
     33 namespace gold
     34 {
     35 
     36 template<typename Stringpool_char>
     37 Stringpool_template<Stringpool_char>::Stringpool_template(uint64_t addralign)
     38   : string_set_(), key_to_offset_(), strings_(), strtab_size_(0),
     39     zero_null_(true), optimize_(false), offset_(sizeof(Stringpool_char)),
     40     addralign_(addralign)
     41 {
     42   if (parameters->options_valid()
     43       && parameters->options().optimize() >= 2
     44       && addralign <= sizeof(Stringpool_char))
     45     this->optimize_ = true;
     46 }
     47 
     48 template<typename Stringpool_char>
     49 void
     50 Stringpool_template<Stringpool_char>::clear()
     51 {
     52   for (typename std::list<Stringdata*>::iterator p = this->strings_.begin();
     53        p != this->strings_.end();
     54        ++p)
     55     delete[] reinterpret_cast<char*>(*p);
     56   this->strings_.clear();
     57   this->key_to_offset_.clear();
     58   this->string_set_.clear();
     59 }
     60 
     61 template<typename Stringpool_char>
     62 Stringpool_template<Stringpool_char>::~Stringpool_template()
     63 {
     64   this->clear();
     65 }
     66 
     67 // Resize the internal hashtable with the expectation we'll get n new
     68 // elements.  Note that the hashtable constructor takes a "number of
     69 // buckets you'd like," rather than "number of elements you'd like,"
     70 // but that's the best we can do.
     71 
     72 template<typename Stringpool_char>
     73 void
     74 Stringpool_template<Stringpool_char>::reserve(unsigned int n)
     75 {
     76   this->key_to_offset_.reserve(n);
     77 
     78 #if defined(HAVE_UNORDERED_MAP)
     79   this->string_set_.rehash(this->string_set_.size() + n);
     80   return;
     81 #elif defined(HAVE_TR1_UNORDERED_MAP)
     82   // rehash() implementation is broken in gcc 4.0.3's stl
     83   //this->string_set_.rehash(this->string_set_.size() + n);
     84   //return;
     85 #elif defined(HAVE_EXT_HASH_MAP)
     86   this->string_set_.resize(this->string_set_.size() + n);
     87   return;
     88 #endif
     89 
     90   // This is the generic "reserve" code, if no #ifdef above triggers.
     91   String_set_type new_string_set(this->string_set_.size() + n);
     92   new_string_set.insert(this->string_set_.begin(), this->string_set_.end());
     93   this->string_set_.swap(new_string_set);
     94 }
     95 
     96 // Compare two strings of arbitrary character type for equality.
     97 
     98 template<typename Stringpool_char>
     99 bool
    100 Stringpool_template<Stringpool_char>::string_equal(const Stringpool_char* s1,
    101 						   const Stringpool_char* s2)
    102 {
    103   while (*s1 != 0)
    104     if (*s1++ != *s2++)
    105       return false;
    106   return *s2 == 0;
    107 }
    108 
    109 // Specialize string_equal for char.
    110 
    111 template<>
    112 inline bool
    113 Stringpool_template<char>::string_equal(const char* s1, const char* s2)
    114 {
    115   return strcmp(s1, s2) == 0;
    116 }
    117 
    118 // Equality comparison function for the hash table.
    119 
    120 template<typename Stringpool_char>
    121 inline bool
    122 Stringpool_template<Stringpool_char>::Stringpool_eq::operator()(
    123     const Hashkey& h1,
    124     const Hashkey& h2) const
    125 {
    126   return (h1.hash_code == h2.hash_code
    127 	  && h1.length == h2.length
    128 	  && (h1.string == h2.string
    129 	      || memcmp(h1.string, h2.string,
    130 			h1.length * sizeof(Stringpool_char)) == 0));
    131 }
    132 
    133 // Hash function.  The length is in characters, not bytes.
    134 
    135 template<typename Stringpool_char>
    136 size_t
    137 Stringpool_template<Stringpool_char>::string_hash(const Stringpool_char* s,
    138 						  size_t length)
    139 {
    140   return gold::string_hash<Stringpool_char>(s, length);
    141 }
    142 
    143 // Add the string S to the list of canonical strings.  Return a
    144 // pointer to the canonical string.  If PKEY is not NULL, set *PKEY to
    145 // the key.  LENGTH is the length of S in characters.  Note that S may
    146 // not be NUL terminated.
    147 
    148 template<typename Stringpool_char>
    149 const Stringpool_char*
    150 Stringpool_template<Stringpool_char>::add_string(const Stringpool_char* s,
    151 						 size_t len)
    152 {
    153   // We are in trouble if we've already computed the string offsets.
    154   gold_assert(this->strtab_size_ == 0);
    155 
    156   // The size we allocate for a new Stringdata.
    157   const size_t buffer_size = 1000;
    158   // The amount we multiply the Stringdata index when calculating the
    159   // key.
    160   const size_t key_mult = 1024;
    161   gold_assert(key_mult >= buffer_size);
    162 
    163   // Convert len to the number of bytes we need to allocate, including
    164   // the null character.
    165   len = (len + 1) * sizeof(Stringpool_char);
    166 
    167   size_t alc;
    168   bool front = true;
    169   if (len > buffer_size)
    170     {
    171       alc = sizeof(Stringdata) + len;
    172       front = false;
    173     }
    174   else if (this->strings_.empty())
    175     alc = sizeof(Stringdata) + buffer_size;
    176   else
    177     {
    178       Stringdata* psd = this->strings_.front();
    179       if (len > psd->alc - psd->len)
    180 	alc = sizeof(Stringdata) + buffer_size;
    181       else
    182 	{
    183 	  char* ret = psd->data + psd->len;
    184 	  memcpy(ret, s, len - sizeof(Stringpool_char));
    185 	  memset(ret + len - sizeof(Stringpool_char), 0,
    186 		 sizeof(Stringpool_char));
    187 
    188 	  psd->len += len;
    189 
    190 	  return reinterpret_cast<const Stringpool_char*>(ret);
    191 	}
    192     }
    193 
    194   Stringdata* psd = reinterpret_cast<Stringdata*>(new char[alc]);
    195   psd->alc = alc - sizeof(Stringdata);
    196   memcpy(psd->data, s, len - sizeof(Stringpool_char));
    197   memset(psd->data + len - sizeof(Stringpool_char), 0,
    198 	 sizeof(Stringpool_char));
    199   psd->len = len;
    200 
    201   if (front)
    202     this->strings_.push_front(psd);
    203   else
    204     this->strings_.push_back(psd);
    205 
    206   return reinterpret_cast<const Stringpool_char*>(psd->data);
    207 }
    208 
    209 // Add a string to a string pool.
    210 
    211 template<typename Stringpool_char>
    212 const Stringpool_char*
    213 Stringpool_template<Stringpool_char>::add(const Stringpool_char* s, bool copy,
    214                                           Key* pkey)
    215 {
    216   return this->add_with_length(s, string_length(s), copy, pkey);
    217 }
    218 
    219 // Add a new key offset entry.
    220 
    221 template<typename Stringpool_char>
    222 void
    223 Stringpool_template<Stringpool_char>::new_key_offset(size_t length)
    224 {
    225   section_offset_type offset;
    226   if (this->zero_null_ && length == 0)
    227     offset = 0;
    228   else
    229     {
    230       offset = this->offset_;
    231       // Align non-zero length strings.
    232       if (length != 0)
    233 	offset = align_address(offset, this->addralign_);
    234       this->offset_ = offset + (length + 1) * sizeof(Stringpool_char);
    235     }
    236   this->key_to_offset_.push_back(offset);
    237 }
    238 
    239 template<typename Stringpool_char>
    240 const Stringpool_char*
    241 Stringpool_template<Stringpool_char>::add_with_length(const Stringpool_char* s,
    242 						      size_t length,
    243 						      bool copy,
    244 						      Key* pkey)
    245 {
    246   typedef std::pair<typename String_set_type::iterator, bool> Insert_type;
    247 
    248   // We add 1 so that 0 is always invalid.
    249   const Key k = this->key_to_offset_.size() + 1;
    250 
    251   if (!copy)
    252     {
    253       // When we don't need to copy the string, we can call insert
    254       // directly.
    255 
    256       std::pair<Hashkey, Hashval> element(Hashkey(s, length), k);
    257 
    258       Insert_type ins = this->string_set_.insert(element);
    259 
    260       typename String_set_type::const_iterator p = ins.first;
    261 
    262       if (ins.second)
    263 	{
    264 	  // We just added the string.  The key value has now been
    265 	  // used.
    266 	  this->new_key_offset(length);
    267 	}
    268       else
    269 	{
    270 	  gold_assert(k != p->second);
    271 	}
    272 
    273       if (pkey != NULL)
    274 	*pkey = p->second;
    275       return p->first.string;
    276     }
    277 
    278   // When we have to copy the string, we look it up twice in the hash
    279   // table.  The problem is that we can't insert S before we
    280   // canonicalize it by copying it into the canonical list. The hash
    281   // code will only be computed once.
    282 
    283   Hashkey hk(s, length);
    284   typename String_set_type::const_iterator p = this->string_set_.find(hk);
    285   if (p != this->string_set_.end())
    286     {
    287       if (pkey != NULL)
    288 	*pkey = p->second;
    289       return p->first.string;
    290     }
    291 
    292   this->new_key_offset(length);
    293 
    294   hk.string = this->add_string(s, length);
    295   // The contents of the string stay the same, so we don't need to
    296   // adjust hk.hash_code or hk.length.
    297 
    298   std::pair<Hashkey, Hashval> element(hk, k);
    299 
    300   Insert_type ins = this->string_set_.insert(element);
    301   gold_assert(ins.second);
    302 
    303   if (pkey != NULL)
    304     *pkey = k;
    305   return hk.string;
    306 }
    307 
    308 template<typename Stringpool_char>
    309 const Stringpool_char*
    310 Stringpool_template<Stringpool_char>::find(const Stringpool_char* s,
    311 					   Key* pkey) const
    312 {
    313   Hashkey hk(s);
    314   typename String_set_type::const_iterator p = this->string_set_.find(hk);
    315   if (p == this->string_set_.end())
    316     return NULL;
    317 
    318   if (pkey != NULL)
    319     *pkey = p->second;
    320 
    321   return p->first.string;
    322 }
    323 
    324 // Comparison routine used when sorting into an ELF strtab.  We want
    325 // to sort this so that when one string is a suffix of another, we
    326 // always see the shorter string immediately after the longer string.
    327 // For example, we want to see these strings in this order:
    328 //   abcd
    329 //   cd
    330 //   d
    331 // When strings are not suffixes, we don't care what order they are
    332 // in, but we need to ensure that suffixes wind up next to each other.
    333 // So we do a reversed lexicographic sort on the reversed string.
    334 
    335 template<typename Stringpool_char>
    336 bool
    337 Stringpool_template<Stringpool_char>::Stringpool_sort_comparison::operator()(
    338   const Stringpool_sort_info& sort_info1,
    339   const Stringpool_sort_info& sort_info2) const
    340 {
    341   const Hashkey& h1(sort_info1->first);
    342   const Hashkey& h2(sort_info2->first);
    343   const Stringpool_char* s1 = h1.string;
    344   const Stringpool_char* s2 = h2.string;
    345   const size_t len1 = h1.length;
    346   const size_t len2 = h2.length;
    347   const size_t minlen = len1 < len2 ? len1 : len2;
    348   const Stringpool_char* p1 = s1 + len1 - 1;
    349   const Stringpool_char* p2 = s2 + len2 - 1;
    350   for (size_t i = minlen; i > 0; --i, --p1, --p2)
    351     {
    352       if (*p1 != *p2)
    353 	return *p1 > *p2;
    354     }
    355   return len1 > len2;
    356 }
    357 
    358 // Return whether s1 is a suffix of s2.
    359 
    360 template<typename Stringpool_char>
    361 bool
    362 Stringpool_template<Stringpool_char>::is_suffix(const Stringpool_char* s1,
    363                                                 size_t len1,
    364 						const Stringpool_char* s2,
    365                                                 size_t len2)
    366 {
    367   if (len1 > len2)
    368     return false;
    369   return memcmp(s1, s2 + len2 - len1, len1 * sizeof(Stringpool_char)) == 0;
    370 }
    371 
    372 // Turn the stringpool into an ELF strtab: determine the offsets of
    373 // each string in the table.
    374 
    375 template<typename Stringpool_char>
    376 void
    377 Stringpool_template<Stringpool_char>::set_string_offsets()
    378 {
    379   if (this->strtab_size_ != 0)
    380     {
    381       // We've already computed the offsets.
    382       return;
    383     }
    384 
    385   const size_t charsize = sizeof(Stringpool_char);
    386 
    387   // Offset 0 may be reserved for the empty string.
    388   section_offset_type offset = this->zero_null_ ? charsize : 0;
    389 
    390   // Sorting to find suffixes can take over 25% of the total CPU time
    391   // used by the linker.  Since it's merely an optimization to reduce
    392   // the strtab size, and gives a relatively small benefit (it's
    393   // typically rare for a symbol to be a suffix of another), we only
    394   // take the time to sort when the user asks for heavy optimization.
    395   if (!this->optimize_)
    396     {
    397       // If we are not optimizing, the offsets are already assigned.
    398       offset = this->offset_;
    399     }
    400   else
    401     {
    402       size_t count = this->string_set_.size();
    403 
    404       std::vector<Stringpool_sort_info> v;
    405       v.reserve(count);
    406 
    407       for (typename String_set_type::iterator p = this->string_set_.begin();
    408            p != this->string_set_.end();
    409            ++p)
    410         v.push_back(Stringpool_sort_info(p));
    411 
    412       std::sort(v.begin(), v.end(), Stringpool_sort_comparison());
    413 
    414       section_offset_type last_offset = -1;
    415       for (typename std::vector<Stringpool_sort_info>::iterator last = v.end(),
    416              curr = v.begin();
    417            curr != v.end();
    418            last = curr++)
    419         {
    420 	  section_offset_type this_offset;
    421           if (this->zero_null_ && (*curr)->first.string[0] == 0)
    422             this_offset = 0;
    423           else if (last != v.end()
    424                    && is_suffix((*curr)->first.string,
    425 				(*curr)->first.length,
    426                                 (*last)->first.string,
    427 				(*last)->first.length))
    428             this_offset = (last_offset
    429 			   + (((*last)->first.length - (*curr)->first.length)
    430 			      * charsize));
    431           else
    432             {
    433               this_offset = align_address(offset, this->addralign_);
    434               offset = this_offset + ((*curr)->first.length + 1) * charsize;
    435             }
    436 	  this->key_to_offset_[(*curr)->second - 1] = this_offset;
    437 	  last_offset = this_offset;
    438         }
    439     }
    440 
    441   this->strtab_size_ = offset;
    442 }
    443 
    444 // Get the offset of a string in the ELF strtab.  The string must
    445 // exist.
    446 
    447 template<typename Stringpool_char>
    448 section_offset_type
    449 Stringpool_template<Stringpool_char>::get_offset(const Stringpool_char* s)
    450   const
    451 {
    452   return this->get_offset_with_length(s, string_length(s));
    453 }
    454 
    455 template<typename Stringpool_char>
    456 section_offset_type
    457 Stringpool_template<Stringpool_char>::get_offset_with_length(
    458     const Stringpool_char* s,
    459     size_t length) const
    460 {
    461   gold_assert(this->strtab_size_ != 0);
    462   Hashkey hk(s, length);
    463   typename String_set_type::const_iterator p = this->string_set_.find(hk);
    464   if (p != this->string_set_.end())
    465     return this->key_to_offset_[p->second - 1];
    466   gold_unreachable();
    467 }
    468 
    469 // Write the ELF strtab into the buffer.
    470 
    471 template<typename Stringpool_char>
    472 void
    473 Stringpool_template<Stringpool_char>::write_to_buffer(
    474     unsigned char* buffer,
    475     section_size_type bufsize)
    476 {
    477   gold_assert(this->strtab_size_ != 0);
    478   gold_assert(bufsize >= this->strtab_size_);
    479   if (this->zero_null_)
    480     buffer[0] = '\0';
    481   for (typename String_set_type::const_iterator p = this->string_set_.begin();
    482        p != this->string_set_.end();
    483        ++p)
    484     {
    485       const int len = (p->first.length + 1) * sizeof(Stringpool_char);
    486       const section_offset_type offset = this->key_to_offset_[p->second - 1];
    487       gold_assert(static_cast<section_size_type>(offset) + len
    488 		  <= this->strtab_size_);
    489       memcpy(buffer + offset, p->first.string, len);
    490     }
    491 }
    492 
    493 // Write the ELF strtab into the output file at the specified offset.
    494 
    495 template<typename Stringpool_char>
    496 void
    497 Stringpool_template<Stringpool_char>::write(Output_file* of, off_t offset)
    498 {
    499   gold_assert(this->strtab_size_ != 0);
    500   unsigned char* view = of->get_output_view(offset, this->strtab_size_);
    501   this->write_to_buffer(view, this->strtab_size_);
    502   of->write_output_view(offset, this->strtab_size_, view);
    503 }
    504 
    505 // Print statistical information to stderr.  This is used for --stats.
    506 
    507 template<typename Stringpool_char>
    508 void
    509 Stringpool_template<Stringpool_char>::print_stats(const char* name) const
    510 {
    511 #if defined(HAVE_UNORDERED_MAP) || defined(HAVE_TR1_UNORDERED_MAP) || defined(HAVE_EXT_HASH_MAP)
    512   fprintf(stderr, _("%s: %s entries: %zu; buckets: %zu\n"),
    513 	  program_name, name, this->string_set_.size(),
    514 	  this->string_set_.bucket_count());
    515 #else
    516   fprintf(stderr, _("%s: %s entries: %zu\n"),
    517 	  program_name, name, this->table_.size());
    518 #endif
    519   fprintf(stderr, _("%s: %s Stringdata structures: %zu\n"),
    520 	  program_name, name, this->strings_.size());
    521 }
    522 
    523 // Instantiate the templates we need.
    524 
    525 template
    526 class Stringpool_template<char>;
    527 
    528 template
    529 class Stringpool_template<uint16_t>;
    530 
    531 template
    532 class Stringpool_template<uint32_t>;
    533 
    534 } // End namespace gold.
    535