Update comments about output offsets and merged input sections.
[deliverable/binutils-gdb.git] / gold / merge.cc
1 // merge.cc -- handle section merging for gold
2
3 // Copyright 2006, 2007 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@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 <cstdlib>
26 #include <algorithm>
27
28 #include "merge.h"
29
30 namespace gold
31 {
32
33 // For each object with merge sections, we store an Object_merge_map.
34 // This is used to map locations in input sections to a merged output
35 // section. The output section itself is not recorded here--it can be
36 // found in the map_to_output_ field of the Object.
37
38 class Object_merge_map
39 {
40 public:
41 Object_merge_map()
42 : first_shnum_(-1U), first_map_(),
43 second_shnum_(-1U), second_map_(),
44 section_merge_maps_()
45 { }
46
47 ~Object_merge_map();
48
49 // Add a mapping for MERGE_MAP, for the bytes from OFFSET to OFFSET
50 // + LENGTH in the input section SHNDX to OUTPUT_OFFSET in the
51 // output section. An OUTPUT_OFFSET of -1 means that the bytes are
52 // discarded. OUTPUT_OFFSET is relative to the start of the merged
53 // data in the output section.
54 void
55 add_mapping(const Merge_map*, unsigned int shndx, section_offset_type offset,
56 section_size_type length, section_offset_type output_offset);
57
58 // Get the output offset for an input address in MERGE_MAP. The
59 // input address is at offset OFFSET in section SHNDX. This sets
60 // *OUTPUT_OFFSET to the offset in the output section; this will be
61 // -1 if the bytes are not being copied to the output. This returns
62 // true if the mapping is known, false otherwise. *OUTPUT_OFFSET is
63 // relative to the start of the merged data in the output section.
64 bool
65 get_output_offset(const Merge_map*, unsigned int shndx,
66 section_offset_type offset,
67 section_offset_type *output_offset);
68
69 private:
70 // Map input section offsets to a length and an output section
71 // offset. An output section offset of -1 means that this part of
72 // the input section is being discarded.
73 struct Input_merge_entry
74 {
75 // The offset in the input section.
76 section_offset_type input_offset;
77 // The length.
78 section_size_type length;
79 // The offset in the output section.
80 section_offset_type output_offset;
81 };
82
83 // A less-than comparison routine for Input_merge_entry.
84 struct Input_merge_compare
85 {
86 bool
87 operator()(const Input_merge_entry& i1, const Input_merge_entry& i2) const
88 { return i1.input_offset < i2.input_offset; }
89 };
90
91 // A list of entries for a particular input section.
92 struct Input_merge_map
93 {
94 // We store these with the Relobj, and we look them up by input
95 // section. It is possible to have two different merge maps
96 // associated with a single output section. For example, this
97 // happens routinely with .rodata, when merged string constants
98 // and merged fixed size constants are both put into .rodata. The
99 // output offset that we store is not the offset from the start of
100 // the output section; it is the offset from the start of the
101 // merged data in the output section. That means that the caller
102 // is going to add the offset of the merged data within the output
103 // section, which means that the caller needs to know which set of
104 // merged data it found the entry in. So it's not enough to find
105 // this data based on the input section and the output section; we
106 // also have to find it based on a set of merged data in the
107 // output section. In order to verify that we are looking at the
108 // right data, we store a pointer to the Merge_map here, and we
109 // pass in a pointer when looking at the data. If we are asked to
110 // look up information for a different Merge_map, we report that
111 // we don't have it, rather than trying a lookup and returning an
112 // answer which will receive the wrong offset.
113 const Merge_map* merge_map;
114 // The list of mappings.
115 std::vector<Input_merge_entry> entries;
116 // Whether the ENTRIES field is sorted by input_offset.
117 bool sorted;
118
119 Input_merge_map()
120 : merge_map(NULL), entries(), sorted(true)
121 { }
122 };
123
124 // Map input section indices to merge maps.
125 typedef std::map<unsigned int, Input_merge_map*> Section_merge_maps;
126
127 // Return a pointer to the Input_merge_map to use for the input
128 // section SHNDX, or NULL.
129 Input_merge_map*
130 get_input_merge_map(unsigned int shndx);
131
132 // Get or make the the Input_merge_map to use for the section SHNDX
133 // with MERGE_MAP.
134 Input_merge_map*
135 get_or_make_input_merge_map(const Merge_map* merge_map, unsigned int shndx);
136
137 // Any given object file will normally only have a couple of input
138 // sections with mergeable contents. So we keep the first two input
139 // section numbers inline, and push any further ones into a map. A
140 // value of -1U in first_shnum_ or second_shnum_ means that we don't
141 // have a corresponding entry.
142 unsigned int first_shnum_;
143 Input_merge_map first_map_;
144 unsigned int second_shnum_;
145 Input_merge_map second_map_;
146 Section_merge_maps section_merge_maps_;
147 };
148
149 // Destructor.
150
151 Object_merge_map::~Object_merge_map()
152 {
153 for (Section_merge_maps::iterator p = this->section_merge_maps_.begin();
154 p != this->section_merge_maps_.end();
155 ++p)
156 delete p->second;
157 }
158
159 // Get the Input_merge_map to use for an input section, or NULL.
160
161 Object_merge_map::Input_merge_map*
162 Object_merge_map::get_input_merge_map(unsigned int shndx)
163 {
164 gold_assert(shndx != -1U);
165 if (shndx == this->first_shnum_)
166 return &this->first_map_;
167 if (shndx == this->second_shnum_)
168 return &this->second_map_;
169 Section_merge_maps::const_iterator p = this->section_merge_maps_.find(shndx);
170 if (p != this->section_merge_maps_.end())
171 return p->second;
172 return NULL;
173 }
174
175 // Get or create the Input_merge_map to use for an input section.
176
177 Object_merge_map::Input_merge_map*
178 Object_merge_map::get_or_make_input_merge_map(const Merge_map* merge_map,
179 unsigned int shndx)
180 {
181 Input_merge_map* map = this->get_input_merge_map(shndx);
182 if (map != NULL)
183 {
184 // For a given input section in a given object, every mapping
185 // must be done with the same Merge_map.
186 gold_assert(map->merge_map == merge_map);
187 return map;
188 }
189
190 // We need to create a new entry.
191 if (this->first_shnum_ == -1U)
192 {
193 this->first_shnum_ = shndx;
194 this->first_map_.merge_map = merge_map;
195 return &this->first_map_;
196 }
197 if (this->second_shnum_ == -1U)
198 {
199 this->second_shnum_ = shndx;
200 this->second_map_.merge_map = merge_map;
201 return &this->second_map_;
202 }
203
204 Input_merge_map* new_map = new Input_merge_map;
205 new_map->merge_map = merge_map;
206 this->section_merge_maps_[shndx] = new_map;
207 return new_map;
208 }
209
210 // Add a mapping.
211
212 void
213 Object_merge_map::add_mapping(const Merge_map* merge_map, unsigned int shndx,
214 section_offset_type input_offset,
215 section_size_type length,
216 section_offset_type output_offset)
217 {
218 Input_merge_map* map = this->get_or_make_input_merge_map(merge_map, shndx);
219
220 // Try to merge the new entry in the last one we saw.
221 if (!map->entries.empty())
222 {
223 Input_merge_entry& entry(map->entries.back());
224
225 // Use section_size_type to avoid signed/unsigned warnings.
226 section_size_type input_offset_u = input_offset;
227 section_size_type output_offset_u = output_offset;
228
229 // If this entry is not in order, we need to sort the vector
230 // before looking anything up.
231 if (input_offset_u < entry.input_offset + entry.length)
232 {
233 gold_assert(input_offset < entry.input_offset);
234 gold_assert(input_offset_u + length
235 <= static_cast<section_size_type>(entry.input_offset));
236 map->sorted = false;
237 }
238 else if (entry.input_offset + entry.length == input_offset_u
239 && (output_offset == -1
240 ? entry.output_offset == -1
241 : entry.output_offset + entry.length == output_offset_u))
242 {
243 entry.length += length;
244 return;
245 }
246 }
247
248 Input_merge_entry entry;
249 entry.input_offset = input_offset;
250 entry.length = length;
251 entry.output_offset = output_offset;
252 map->entries.push_back(entry);
253 }
254
255 // Get the output offset for an input address.
256
257 inline bool
258 Object_merge_map::get_output_offset(const Merge_map* merge_map,
259 unsigned int shndx,
260 section_offset_type input_offset,
261 section_offset_type *output_offset)
262 {
263 Input_merge_map* map = this->get_input_merge_map(shndx);
264 if (map == NULL || map->merge_map != merge_map)
265 return false;
266
267 if (!map->sorted)
268 {
269 std::sort(map->entries.begin(), map->entries.end(),
270 Input_merge_compare());
271 map->sorted = true;
272 }
273
274 Input_merge_entry entry;
275 entry.input_offset = input_offset;
276 std::vector<Input_merge_entry>::const_iterator p =
277 std::lower_bound(map->entries.begin(), map->entries.end(),
278 entry, Input_merge_compare());
279 if (p == map->entries.end() || p->input_offset > input_offset)
280 {
281 if (p == map->entries.begin())
282 return false;
283 --p;
284 gold_assert(p->input_offset <= input_offset);
285 }
286
287 if (input_offset - p->input_offset
288 >= static_cast<section_offset_type>(p->length))
289 return false;
290
291 *output_offset = p->output_offset;
292 if (*output_offset != -1)
293 *output_offset += (input_offset - p->input_offset);
294 return true;
295 }
296
297 // Class Merge_map.
298
299 // Add a mapping for the bytes from OFFSET to OFFSET + LENGTH in input
300 // section SHNDX in object OBJECT to an OUTPUT_OFFSET in merged data
301 // in an output section.
302
303 void
304 Merge_map::add_mapping(Relobj* object, unsigned int shndx,
305 section_offset_type offset, section_size_type length,
306 section_offset_type output_offset)
307 {
308 Object_merge_map* object_merge_map = object->merge_map();
309 if (object_merge_map == NULL)
310 {
311 object_merge_map = new Object_merge_map();
312 object->set_merge_map(object_merge_map);
313 }
314
315 object_merge_map->add_mapping(this, shndx, offset, length, output_offset);
316 }
317
318 // Return the output offset for an input address. The input address
319 // is at offset OFFSET in section SHNDX in OBJECT. This sets
320 // *OUTPUT_OFFSET to the offset in the merged data in the output
321 // section. This returns true if the mapping is known, false
322 // otherwise.
323
324 bool
325 Merge_map::get_output_offset(const Relobj* object, unsigned int shndx,
326 section_offset_type offset,
327 section_offset_type* output_offset) const
328 {
329 Object_merge_map* object_merge_map = object->merge_map();
330 if (object_merge_map == NULL)
331 return false;
332 return object_merge_map->get_output_offset(this, shndx, offset,
333 output_offset);
334 }
335
336 // Class Output_merge_base.
337
338 // Return the output offset for an input offset. The input address is
339 // at offset OFFSET in section SHNDX in OBJECT. If we know the
340 // offset, set *POUTPUT and return true. Otherwise return false.
341
342 bool
343 Output_merge_base::do_output_offset(const Relobj* object,
344 unsigned int shndx,
345 section_offset_type offset,
346 section_offset_type* poutput) const
347 {
348 return this->merge_map_.get_output_offset(object, shndx, offset, poutput);
349 }
350
351 // Class Output_merge_data.
352
353 // Compute the hash code for a fixed-size constant.
354
355 size_t
356 Output_merge_data::Merge_data_hash::operator()(Merge_data_key k) const
357 {
358 const unsigned char* p = this->pomd_->constant(k);
359 section_size_type entsize =
360 convert_to_section_size_type(this->pomd_->entsize());
361
362 // Fowler/Noll/Vo (FNV) hash (type FNV-1a).
363 if (sizeof(size_t) == 8)
364 {
365 size_t result = static_cast<size_t>(14695981039346656037ULL);
366 for (section_size_type i = 0; i < entsize; ++i)
367 {
368 result &= (size_t) *p++;
369 result *= 1099511628211ULL;
370 }
371 return result;
372 }
373 else
374 {
375 size_t result = 2166136261UL;
376 for (section_size_type i = 0; i < entsize; ++i)
377 {
378 result ^= (size_t) *p++;
379 result *= 16777619UL;
380 }
381 return result;
382 }
383 }
384
385 // Return whether one hash table key equals another.
386
387 bool
388 Output_merge_data::Merge_data_eq::operator()(Merge_data_key k1,
389 Merge_data_key k2) const
390 {
391 const unsigned char* p1 = this->pomd_->constant(k1);
392 const unsigned char* p2 = this->pomd_->constant(k2);
393 return memcmp(p1, p2, this->pomd_->entsize()) == 0;
394 }
395
396 // Add a constant to the end of the section contents.
397
398 void
399 Output_merge_data::add_constant(const unsigned char* p)
400 {
401 section_size_type entsize = convert_to_section_size_type(this->entsize());
402 section_size_type addralign =
403 convert_to_section_size_type(this->addralign());
404 section_size_type addsize = std::max(entsize, addralign);
405 if (this->len_ + addsize > this->alc_)
406 {
407 if (this->alc_ == 0)
408 this->alc_ = 128 * addsize;
409 else
410 this->alc_ *= 2;
411 this->p_ = static_cast<unsigned char*>(realloc(this->p_, this->alc_));
412 if (this->p_ == NULL)
413 gold_nomem();
414 }
415
416 memcpy(this->p_ + this->len_, p, entsize);
417 if (addsize > entsize)
418 memset(this->p_ + this->len_ + entsize, 0, addsize - entsize);
419 this->len_ += addsize;
420 }
421
422 // Add the input section SHNDX in OBJECT to a merged output section
423 // which holds fixed length constants. Return whether we were able to
424 // handle the section; if not, it will be linked as usual without
425 // constant merging.
426
427 bool
428 Output_merge_data::do_add_input_section(Relobj* object, unsigned int shndx)
429 {
430 section_size_type len;
431 const unsigned char* p = object->section_contents(shndx, &len, false);
432
433 section_size_type entsize = convert_to_section_size_type(this->entsize());
434
435 if (len % entsize != 0)
436 return false;
437
438 this->input_count_ += len / entsize;
439
440 for (section_size_type i = 0; i < len; i += entsize, p += entsize)
441 {
442 // Add the constant to the section contents. If we find that it
443 // is already in the hash table, we will remove it again.
444 Merge_data_key k = this->len_;
445 this->add_constant(p);
446
447 std::pair<Merge_data_hashtable::iterator, bool> ins =
448 this->hashtable_.insert(k);
449
450 if (!ins.second)
451 {
452 // Key was already present. Remove the copy we just added.
453 this->len_ -= entsize;
454 k = *ins.first;
455 }
456
457 // Record the offset of this constant in the output section.
458 this->add_mapping(object, shndx, i, entsize, k);
459 }
460
461 return true;
462 }
463
464 // Set the final data size in a merged output section with fixed size
465 // constants.
466
467 void
468 Output_merge_data::set_final_data_size()
469 {
470 // Release the memory we don't need.
471 this->p_ = static_cast<unsigned char*>(realloc(this->p_, this->len_));
472 gold_assert(this->p_ != NULL);
473 this->set_data_size(this->len_);
474 }
475
476 // Write the data of a merged output section with fixed size constants
477 // to the file.
478
479 void
480 Output_merge_data::do_write(Output_file* of)
481 {
482 of->write(this->offset(), this->p_, this->len_);
483 }
484
485 // Write the data to a buffer.
486
487 void
488 Output_merge_data::do_write_to_buffer(unsigned char* buffer)
489 {
490 memcpy(buffer, this->p_, this->len_);
491 }
492
493 // Print merge stats to stderr.
494
495 void
496 Output_merge_data::do_print_merge_stats(const char* section_name)
497 {
498 fprintf(stderr,
499 _("%s: %s merged constants size: %lu; input: %zu; output: %zu\n"),
500 program_name, section_name,
501 static_cast<unsigned long>(this->entsize()),
502 this->input_count_, this->hashtable_.size());
503 }
504
505 // Class Output_merge_string.
506
507 // Add an input section to a merged string section.
508
509 template<typename Char_type>
510 bool
511 Output_merge_string<Char_type>::do_add_input_section(Relobj* object,
512 unsigned int shndx)
513 {
514 section_size_type len;
515 const unsigned char* pdata = object->section_contents(shndx, &len, false);
516
517 const Char_type* p = reinterpret_cast<const Char_type*>(pdata);
518 const Char_type* pend = p + len;
519
520 if (len % sizeof(Char_type) != 0)
521 {
522 object->error(_("mergeable string section length not multiple of "
523 "character size"));
524 return false;
525 }
526
527 size_t count = 0;
528
529 // The index I is in bytes, not characters.
530 section_size_type i = 0;
531 while (i < len)
532 {
533 const Char_type* pl;
534 for (pl = p; *pl != 0; ++pl)
535 {
536 if (pl >= pend)
537 {
538 object->error(_("entry in mergeable string section "
539 "not null terminated"));
540 break;
541 }
542 }
543
544 Stringpool::Key key;
545 const Char_type* str = this->stringpool_.add_with_length(p, pl - p, true,
546 &key);
547
548 section_size_type bytelen_with_null = ((pl - p) + 1) * sizeof(Char_type);
549 this->merged_strings_.push_back(Merged_string(object, shndx, i, str,
550 bytelen_with_null, key));
551
552 p = pl + 1;
553 i += bytelen_with_null;
554 ++count;
555 }
556
557 this->input_count_ += count;
558
559 return true;
560 }
561
562 // Finalize the mappings from the input sections to the output
563 // section, and return the final data size.
564
565 template<typename Char_type>
566 section_size_type
567 Output_merge_string<Char_type>::finalize_merged_data()
568 {
569 this->stringpool_.set_string_offsets();
570
571 for (typename Merged_strings::const_iterator p =
572 this->merged_strings_.begin();
573 p != this->merged_strings_.end();
574 ++p)
575 {
576 section_offset_type offset =
577 this->stringpool_.get_offset_from_key(p->stringpool_key);
578 this->add_mapping(p->object, p->shndx, p->offset, p->length, offset);
579 }
580
581 // Save some memory.
582 this->merged_strings_.clear();
583
584 return this->stringpool_.get_strtab_size();
585 }
586
587 template<typename Char_type>
588 void
589 Output_merge_string<Char_type>::set_final_data_size()
590 {
591 const off_t final_data_size = this->finalize_merged_data();
592 this->set_data_size(final_data_size);
593 }
594
595 // Write out a merged string section.
596
597 template<typename Char_type>
598 void
599 Output_merge_string<Char_type>::do_write(Output_file* of)
600 {
601 this->stringpool_.write(of, this->offset());
602 }
603
604 // Write a merged string section to a buffer.
605
606 template<typename Char_type>
607 void
608 Output_merge_string<Char_type>::do_write_to_buffer(unsigned char* buffer)
609 {
610 this->stringpool_.write_to_buffer(buffer, this->data_size());
611 }
612
613 // Return the name of the types of string to use with
614 // do_print_merge_stats.
615
616 template<typename Char_type>
617 const char*
618 Output_merge_string<Char_type>::string_name()
619 {
620 gold_unreachable();
621 return NULL;
622 }
623
624 template<>
625 const char*
626 Output_merge_string<char>::string_name()
627 {
628 return "strings";
629 }
630
631 template<>
632 const char*
633 Output_merge_string<uint16_t>::string_name()
634 {
635 return "16-bit strings";
636 }
637
638 template<>
639 const char*
640 Output_merge_string<uint32_t>::string_name()
641 {
642 return "32-bit strings";
643 }
644
645 // Print merge stats to stderr.
646
647 template<typename Char_type>
648 void
649 Output_merge_string<Char_type>::do_print_merge_stats(const char* section_name)
650 {
651 char buf[200];
652 snprintf(buf, sizeof buf, "%s merged %s", section_name, this->string_name());
653 fprintf(stderr, _("%s: %s input: %zu\n"),
654 program_name, buf, this->input_count_);
655 this->stringpool_.print_stats(buf);
656 }
657
658 // Instantiate the templates we need.
659
660 template
661 class Output_merge_string<char>;
662
663 template
664 class Output_merge_string<uint16_t>;
665
666 template
667 class Output_merge_string<uint32_t>;
668
669 } // End namespace gold.
This page took 0.067138 seconds and 5 git commands to generate.