daily update
[binutils.git] / gold / merge.h
blobfb6721d2925da70c63a5f5e2f1c785af8fec1cb6
1 // merge.h -- handle section merging for gold -*- C++ -*-
3 // Copyright 2006, 2007, 2008 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
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.
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.
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.
23 #ifndef GOLD_MERGE_H
24 #define GOLD_MERGE_H
26 #include <climits>
27 #include <map>
28 #include <vector>
30 #include "stringpool.h"
31 #include "output.h"
33 namespace gold
36 class Merge_map;
38 // For each object with merge sections, we store an Object_merge_map.
39 // This is used to map locations in input sections to a merged output
40 // section. The output section itself is not recorded here--it can be
41 // found in the output_sections_ field of the Object.
43 class Object_merge_map
45 public:
46 Object_merge_map()
47 : first_shnum_(-1U), first_map_(),
48 second_shnum_(-1U), second_map_(),
49 section_merge_maps_()
50 { }
52 ~Object_merge_map();
54 // Add a mapping for MERGE_MAP, for the bytes from OFFSET to OFFSET
55 // + LENGTH in the input section SHNDX to OUTPUT_OFFSET in the
56 // output section. An OUTPUT_OFFSET of -1 means that the bytes are
57 // discarded. OUTPUT_OFFSET is relative to the start of the merged
58 // data in the output section.
59 void
60 add_mapping(const Merge_map*, unsigned int shndx, section_offset_type offset,
61 section_size_type length, section_offset_type output_offset);
63 // Get the output offset for an input address. MERGE_MAP is the map
64 // we are looking for, or NULL if we don't care. The input address
65 // is at offset OFFSET in section SHNDX. This sets *OUTPUT_OFFSET
66 // to the offset in the output section; this will be -1 if the bytes
67 // are not being copied to the output. This returns true if the
68 // mapping is known, false otherwise. *OUTPUT_OFFSET is relative to
69 // the start of the merged data in the output section.
70 bool
71 get_output_offset(const Merge_map*, unsigned int shndx,
72 section_offset_type offset,
73 section_offset_type *output_offset);
75 // Return whether this is the merge map for section SHNDX.
76 bool
77 is_merge_section_for(const Merge_map*, unsigned int shndx);
79 // Initialize an mapping from input offsets to output addresses for
80 // section SHNDX. STARTING_ADDRESS is the output address of the
81 // merged section.
82 template<int size>
83 void
84 initialize_input_to_output_map(
85 unsigned int shndx,
86 typename elfcpp::Elf_types<size>::Elf_Addr starting_address,
87 Unordered_map<section_offset_type,
88 typename elfcpp::Elf_types<size>::Elf_Addr>*);
90 private:
91 // Map input section offsets to a length and an output section
92 // offset. An output section offset of -1 means that this part of
93 // the input section is being discarded.
94 struct Input_merge_entry
96 // The offset in the input section.
97 section_offset_type input_offset;
98 // The length.
99 section_size_type length;
100 // The offset in the output section.
101 section_offset_type output_offset;
104 // A less-than comparison routine for Input_merge_entry.
105 struct Input_merge_compare
107 bool
108 operator()(const Input_merge_entry& i1, const Input_merge_entry& i2) const
109 { return i1.input_offset < i2.input_offset; }
112 // A list of entries for a particular input section.
113 struct Input_merge_map
115 typedef std::vector<Input_merge_entry> Entries;
117 // We store these with the Relobj, and we look them up by input
118 // section. It is possible to have two different merge maps
119 // associated with a single output section. For example, this
120 // happens routinely with .rodata, when merged string constants
121 // and merged fixed size constants are both put into .rodata. The
122 // output offset that we store is not the offset from the start of
123 // the output section; it is the offset from the start of the
124 // merged data in the output section. That means that the caller
125 // is going to add the offset of the merged data within the output
126 // section, which means that the caller needs to know which set of
127 // merged data it found the entry in. So it's not enough to find
128 // this data based on the input section and the output section; we
129 // also have to find it based on a set of merged data in the
130 // output section. In order to verify that we are looking at the
131 // right data, we store a pointer to the Merge_map here, and we
132 // pass in a pointer when looking at the data. If we are asked to
133 // look up information for a different Merge_map, we report that
134 // we don't have it, rather than trying a lookup and returning an
135 // answer which will receive the wrong offset.
136 const Merge_map* merge_map;
137 // The list of mappings.
138 Entries entries;
139 // Whether the ENTRIES field is sorted by input_offset.
140 bool sorted;
142 Input_merge_map()
143 : merge_map(NULL), entries(), sorted(true)
147 // Map input section indices to merge maps.
148 typedef std::map<unsigned int, Input_merge_map*> Section_merge_maps;
150 // Return a pointer to the Input_merge_map to use for the input
151 // section SHNDX, or NULL.
152 Input_merge_map*
153 get_input_merge_map(unsigned int shndx);
155 // Get or make the the Input_merge_map to use for the section SHNDX
156 // with MERGE_MAP.
157 Input_merge_map*
158 get_or_make_input_merge_map(const Merge_map* merge_map, unsigned int shndx);
160 // Any given object file will normally only have a couple of input
161 // sections with mergeable contents. So we keep the first two input
162 // section numbers inline, and push any further ones into a map. A
163 // value of -1U in first_shnum_ or second_shnum_ means that we don't
164 // have a corresponding entry.
165 unsigned int first_shnum_;
166 Input_merge_map first_map_;
167 unsigned int second_shnum_;
168 Input_merge_map second_map_;
169 Section_merge_maps section_merge_maps_;
172 // This class manages mappings from input sections to offsets in an
173 // output section. This is used where input sections are merged. The
174 // actual data is stored in fields in Object.
176 class Merge_map
178 public:
179 Merge_map()
182 // Add a mapping for the bytes from OFFSET to OFFSET + LENGTH in the
183 // input section SHNDX in object OBJECT to OUTPUT_OFFSET in the
184 // output section. An OUTPUT_OFFSET of -1 means that the bytes are
185 // discarded. OUTPUT_OFFSET is not the offset from the start of the
186 // output section, it is the offset from the start of the merged
187 // data within the output section.
188 void
189 add_mapping(Relobj* object, unsigned int shndx,
190 section_offset_type offset, section_size_type length,
191 section_offset_type output_offset);
193 // Return the output offset for an input address. The input address
194 // is at offset OFFSET in section SHNDX in OBJECT. This sets
195 // *OUTPUT_OFFSET to the offset in the output section; this will be
196 // -1 if the bytes are not being copied to the output. This returns
197 // true if the mapping is known, false otherwise. This returns the
198 // value stored by add_mapping, namely the offset from the start of
199 // the merged data within the output section.
200 bool
201 get_output_offset(const Relobj* object, unsigned int shndx,
202 section_offset_type offset,
203 section_offset_type *output_offset) const;
205 // Return whether this is the merge mapping for section SHNDX in
206 // OBJECT. This should return true when get_output_offset would
207 // return true for some input offset.
208 bool
209 is_merge_section_for(const Relobj* object, unsigned int shndx) const;
212 // A general class for SHF_MERGE data, to hold functions shared by
213 // fixed-size constant data and string data.
215 class Output_merge_base : public Output_section_data
217 public:
218 Output_merge_base(uint64_t entsize, uint64_t addralign)
219 : Output_section_data(addralign), merge_map_(), entsize_(entsize)
222 protected:
223 // Return the output offset for an input offset.
224 bool
225 do_output_offset(const Relobj* object, unsigned int shndx,
226 section_offset_type offset,
227 section_offset_type* poutput) const;
229 // Return whether this is the merge section for an input section.
230 bool
231 do_is_merge_section_for(const Relobj*, unsigned int shndx) const;
233 // Return the entry size.
234 uint64_t
235 entsize() const
236 { return this->entsize_; }
238 // Add a mapping from an OFFSET in input section SHNDX in object
239 // OBJECT to an OUTPUT_OFFSET in the output section. OUTPUT_OFFSET
240 // is the offset from the start of the merged data in the output
241 // section.
242 void
243 add_mapping(Relobj* object, unsigned int shndx, section_offset_type offset,
244 section_size_type length, section_offset_type output_offset)
246 this->merge_map_.add_mapping(object, shndx, offset, length, output_offset);
249 private:
250 // A mapping from input object/section/offset to offset in output
251 // section.
252 Merge_map merge_map_;
253 // The entry size. For fixed-size constants, this is the size of
254 // the constants. For strings, this is the size of a character.
255 uint64_t entsize_;
258 // Handle SHF_MERGE sections with fixed-size constant data.
260 class Output_merge_data : public Output_merge_base
262 public:
263 Output_merge_data(uint64_t entsize, uint64_t addralign)
264 : Output_merge_base(entsize, addralign), p_(NULL), len_(0), alc_(0),
265 input_count_(0),
266 hashtable_(128, Merge_data_hash(this), Merge_data_eq(this))
269 protected:
270 // Add an input section.
271 bool
272 do_add_input_section(Relobj* object, unsigned int shndx);
274 // Set the final data size.
275 void
276 set_final_data_size();
278 // Write the data to the file.
279 void
280 do_write(Output_file*);
282 // Write the data to a buffer.
283 void
284 do_write_to_buffer(unsigned char*);
286 // Write to a map file.
287 void
288 do_print_to_mapfile(Mapfile* mapfile) const
289 { mapfile->print_output_data(this, _("** merge constants")); }
291 // Print merge stats to stderr.
292 void
293 do_print_merge_stats(const char* section_name);
295 private:
296 // We build a hash table of the fixed-size constants. Each constant
297 // is stored as a pointer into the section data we are accumulating.
299 // A key in the hash table. This is an offset in the section
300 // contents we are building.
301 typedef section_offset_type Merge_data_key;
303 // Compute the hash code. To do this we need a pointer back to the
304 // object holding the data.
305 class Merge_data_hash
307 public:
308 Merge_data_hash(const Output_merge_data* pomd)
309 : pomd_(pomd)
312 size_t
313 operator()(Merge_data_key) const;
315 private:
316 const Output_merge_data* pomd_;
319 friend class Merge_data_hash;
321 // Compare two entries in the hash table for equality. To do this
322 // we need a pointer back to the object holding the data. Note that
323 // we now have a pointer to the object stored in two places in the
324 // hash table. Fixing this would require specializing the hash
325 // table, which would be hard to do portably.
326 class Merge_data_eq
328 public:
329 Merge_data_eq(const Output_merge_data* pomd)
330 : pomd_(pomd)
333 bool
334 operator()(Merge_data_key k1, Merge_data_key k2) const;
336 private:
337 const Output_merge_data* pomd_;
340 friend class Merge_data_eq;
342 // The type of the hash table.
343 typedef Unordered_set<Merge_data_key, Merge_data_hash, Merge_data_eq>
344 Merge_data_hashtable;
346 // Given a hash table key, which is just an offset into the section
347 // data, return a pointer to the corresponding constant.
348 const unsigned char*
349 constant(Merge_data_key k) const
351 gold_assert(k >= 0 && k < static_cast<section_offset_type>(this->len_));
352 return this->p_ + k;
355 // Add a constant to the output.
356 void
357 add_constant(const unsigned char*);
359 // The accumulated data.
360 unsigned char* p_;
361 // The length of the accumulated data.
362 section_size_type len_;
363 // The size of the allocated buffer.
364 section_size_type alc_;
365 // The number of entries seen in input files.
366 size_t input_count_;
367 // The hash table.
368 Merge_data_hashtable hashtable_;
371 // Handle SHF_MERGE sections with string data. This is a template
372 // based on the type of the characters in the string.
374 template<typename Char_type>
375 class Output_merge_string : public Output_merge_base
377 public:
378 Output_merge_string(uint64_t addralign)
379 : Output_merge_base(sizeof(Char_type), addralign), stringpool_(),
380 merged_strings_(), input_count_(0)
382 gold_assert(addralign <= sizeof(Char_type));
383 this->stringpool_.set_no_zero_null();
386 protected:
387 // Add an input section.
388 bool
389 do_add_input_section(Relobj* object, unsigned int shndx);
391 // Do all the final processing after the input sections are read in.
392 // Returns the final data size.
393 section_size_type
394 finalize_merged_data();
396 // Set the final data size.
397 void
398 set_final_data_size();
400 // Write the data to the file.
401 void
402 do_write(Output_file*);
404 // Write the data to a buffer.
405 void
406 do_write_to_buffer(unsigned char*);
408 // Write to a map file.
409 void
410 do_print_to_mapfile(Mapfile* mapfile) const
411 { mapfile->print_output_data(this, _("** merge strings")); }
413 // Print merge stats to stderr.
414 void
415 do_print_merge_stats(const char* section_name);
417 // Writes the stringpool to a buffer.
418 void
419 stringpool_to_buffer(unsigned char* buffer, section_size_type buffer_size)
420 { this->stringpool_.write_to_buffer(buffer, buffer_size); }
422 // Clears all the data in the stringpool, to save on memory.
423 void
424 clear_stringpool()
425 { this->stringpool_.clear(); }
427 private:
428 // The name of the string type, for stats.
429 const char*
430 string_name();
432 // As we see input sections, we build a mapping from object, section
433 // index and offset to strings.
434 struct Merged_string
436 // The input object where the string was found.
437 Relobj* object;
438 // The input section in the input object.
439 unsigned int shndx;
440 // The offset in the input section.
441 section_offset_type offset;
442 // The string itself, a pointer into a Stringpool.
443 const Char_type* string;
444 // The length of the string in bytes, including the null terminator.
445 size_t length;
446 // The key in the Stringpool.
447 Stringpool::Key stringpool_key;
449 Merged_string(Relobj *objecta, unsigned int shndxa,
450 section_offset_type offseta, const Char_type* stringa,
451 size_t lengtha, Stringpool::Key stringpool_keya)
452 : object(objecta), shndx(shndxa), offset(offseta), string(stringa),
453 length(lengtha), stringpool_key(stringpool_keya)
457 typedef std::vector<Merged_string> Merged_strings;
459 // As we see the strings, we add them to a Stringpool.
460 Stringpool_template<Char_type> stringpool_;
461 // Map from a location in an input object to an entry in the
462 // Stringpool.
463 Merged_strings merged_strings_;
464 // The number of entries seen in input files.
465 size_t input_count_;
468 } // End namespace gold.
470 #endif // !defined(GOLD_MERGE_H)