bfd/ChangeLog
[binutils.git] / bfd / merge.c
blob3efbd43773ee3f61318d2b2064a2b09c13902462
1 /* SEC_MERGE support.
2 Copyright 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
3 Written by Jakub Jelinek <jakub@redhat.com>.
5 This file is part of BFD, the Binary File Descriptor library.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
21 /* This file contains support for merging duplicate entities within sections,
22 as used in ELF SHF_MERGE. */
24 #include "bfd.h"
25 #include "sysdep.h"
26 #include "libbfd.h"
27 #include "hashtab.h"
28 #include "libiberty.h"
30 struct sec_merge_sec_info;
32 /* An entry in the section merge hash table. */
34 struct sec_merge_hash_entry
36 struct bfd_hash_entry root;
37 /* Length of this entry. This includes the zero terminator. */
38 unsigned int len;
39 /* Start of this string needs to be aligned to
40 alignment octets (not 1 << align). */
41 unsigned int alignment;
42 union
44 /* Index within the merged section. */
45 bfd_size_type index;
46 /* Entry this is a suffix of (if alignment is 0). */
47 struct sec_merge_hash_entry *suffix;
48 } u;
49 /* Which section is it in. */
50 struct sec_merge_sec_info *secinfo;
51 /* Next entity in the hash table. */
52 struct sec_merge_hash_entry *next;
55 /* The section merge hash table. */
57 struct sec_merge_hash
59 struct bfd_hash_table table;
60 /* Next available index. */
61 bfd_size_type size;
62 /* First entity in the SEC_MERGE sections of this type. */
63 struct sec_merge_hash_entry *first;
64 /* Last entity in the SEC_MERGE sections of this type. */
65 struct sec_merge_hash_entry *last;
66 /* Entity size. */
67 unsigned int entsize;
68 /* Are entries fixed size or zero terminated strings? */
69 bfd_boolean strings;
72 struct sec_merge_info
74 /* Chain of sec_merge_infos. */
75 struct sec_merge_info *next;
76 /* Chain of sec_merge_sec_infos. */
77 struct sec_merge_sec_info *chain;
78 /* A hash table used to hold section content. */
79 struct sec_merge_hash *htab;
82 struct sec_merge_sec_info
84 /* Chain of sec_merge_sec_infos. */
85 struct sec_merge_sec_info *next;
86 /* The corresponding section. */
87 asection *sec;
88 /* Pointer to merge_info pointing to us. */
89 void **psecinfo;
90 /* A hash table used to hold section content. */
91 struct sec_merge_hash *htab;
92 /* First string in this section. */
93 struct sec_merge_hash_entry *first_str;
94 /* Original section content. */
95 unsigned char contents[1];
99 /* Routine to create an entry in a section merge hashtab. */
101 static struct bfd_hash_entry *
102 sec_merge_hash_newfunc (struct bfd_hash_entry *entry,
103 struct bfd_hash_table *table, const char *string)
105 /* Allocate the structure if it has not already been allocated by a
106 subclass. */
107 if (entry == NULL)
108 entry = bfd_hash_allocate (table, sizeof (struct sec_merge_hash_entry));
109 if (entry == NULL)
110 return NULL;
112 /* Call the allocation method of the superclass. */
113 entry = bfd_hash_newfunc (entry, table, string);
115 if (entry != NULL)
117 /* Initialize the local fields. */
118 struct sec_merge_hash_entry *ret = (struct sec_merge_hash_entry *) entry;
120 ret->u.suffix = NULL;
121 ret->alignment = 0;
122 ret->secinfo = NULL;
123 ret->next = NULL;
126 return entry;
129 /* Look up an entry in a section merge hash table. */
131 static struct sec_merge_hash_entry *
132 sec_merge_hash_lookup (struct sec_merge_hash *table, const char *string,
133 unsigned int alignment, bfd_boolean create)
135 register const unsigned char *s;
136 register unsigned long hash;
137 register unsigned int c;
138 struct sec_merge_hash_entry *hashp;
139 unsigned int len, i;
140 unsigned int index;
142 hash = 0;
143 len = 0;
144 s = (const unsigned char *) string;
145 if (table->strings)
147 if (table->entsize == 1)
149 while ((c = *s++) != '\0')
151 hash += c + (c << 17);
152 hash ^= hash >> 2;
153 ++len;
155 hash += len + (len << 17);
157 else
159 for (;;)
161 for (i = 0; i < table->entsize; ++i)
162 if (s[i] != '\0')
163 break;
164 if (i == table->entsize)
165 break;
166 for (i = 0; i < table->entsize; ++i)
168 c = *s++;
169 hash += c + (c << 17);
170 hash ^= hash >> 2;
172 ++len;
174 hash += len + (len << 17);
175 len *= table->entsize;
177 hash ^= hash >> 2;
178 len += table->entsize;
180 else
182 for (i = 0; i < table->entsize; ++i)
184 c = *s++;
185 hash += c + (c << 17);
186 hash ^= hash >> 2;
188 len = table->entsize;
191 index = hash % table->table.size;
192 for (hashp = (struct sec_merge_hash_entry *) table->table.table[index];
193 hashp != NULL;
194 hashp = (struct sec_merge_hash_entry *) hashp->root.next)
196 if (hashp->root.hash == hash
197 && len == hashp->len
198 && memcmp (hashp->root.string, string, len) == 0)
200 /* If the string we found does not have at least the required
201 alignment, we need to insert another copy. */
202 if (hashp->alignment < alignment)
204 if (create)
206 /* Mark the less aligned copy as deleted. */
207 hashp->len = 0;
208 hashp->alignment = 0;
210 break;
212 return hashp;
216 if (! create)
217 return NULL;
219 hashp = ((struct sec_merge_hash_entry *)
220 sec_merge_hash_newfunc (NULL, &table->table, string));
221 if (hashp == NULL)
222 return NULL;
223 hashp->root.string = string;
224 hashp->root.hash = hash;
225 hashp->len = len;
226 hashp->alignment = alignment;
227 hashp->root.next = table->table.table[index];
228 table->table.table[index] = (struct bfd_hash_entry *) hashp;
230 return hashp;
233 /* Create a new hash table. */
235 static struct sec_merge_hash *
236 sec_merge_init (unsigned int entsize, bfd_boolean strings)
238 struct sec_merge_hash *table;
240 table = bfd_malloc (sizeof (struct sec_merge_hash));
241 if (table == NULL)
242 return NULL;
244 if (! bfd_hash_table_init (&table->table, sec_merge_hash_newfunc))
246 free (table);
247 return NULL;
250 table->size = 0;
251 table->first = NULL;
252 table->last = NULL;
253 table->entsize = entsize;
254 table->strings = strings;
256 return table;
259 /* Get the index of an entity in a hash table, adding it if it is not
260 already present. */
262 static struct sec_merge_hash_entry *
263 sec_merge_add (struct sec_merge_hash *tab, const char *str,
264 unsigned int alignment, struct sec_merge_sec_info *secinfo)
266 register struct sec_merge_hash_entry *entry;
268 entry = sec_merge_hash_lookup (tab, str, alignment, TRUE);
269 if (entry == NULL)
270 return NULL;
272 if (entry->secinfo == NULL)
274 tab->size++;
275 entry->secinfo = secinfo;
276 if (tab->first == NULL)
277 tab->first = entry;
278 else
279 tab->last->next = entry;
280 tab->last = entry;
283 return entry;
286 static bfd_boolean
287 sec_merge_emit (bfd *abfd, struct sec_merge_hash_entry *entry)
289 struct sec_merge_sec_info *secinfo = entry->secinfo;
290 asection *sec = secinfo->sec;
291 char *pad = "";
292 bfd_size_type off = 0;
293 int alignment_power = sec->output_section->alignment_power;
295 if (alignment_power)
296 pad = bfd_zmalloc ((bfd_size_type) 1 << alignment_power);
298 for (; entry != NULL && entry->secinfo == secinfo; entry = entry->next)
300 register const char *str;
301 register size_t len;
303 len = off & (entry->alignment - 1);
304 if (len)
306 len = entry->alignment - len;
307 if (bfd_bwrite (pad, len, abfd) != len)
308 break;
309 off += len;
312 str = entry->root.string;
313 len = entry->len;
315 if (bfd_bwrite (str, len, abfd) != len)
316 break;
318 off += len;
321 if (alignment_power)
322 free (pad);
324 return entry == NULL || entry->secinfo != secinfo;
327 /* Register a SEC_MERGE section as a candidate for merging.
328 This function is called for all non-dynamic SEC_MERGE input sections. */
330 bfd_boolean
331 _bfd_add_merge_section (bfd *abfd, void **psinfo, asection *sec,
332 void **psecinfo)
334 struct sec_merge_info *sinfo;
335 struct sec_merge_sec_info *secinfo;
336 unsigned int align;
337 bfd_size_type amt;
339 if ((abfd->flags & DYNAMIC) != 0
340 || (sec->flags & SEC_MERGE) == 0)
341 abort ();
343 if (sec->size == 0
344 || (sec->flags & SEC_EXCLUDE) != 0
345 || sec->entsize == 0)
346 return TRUE;
348 if ((sec->flags & SEC_RELOC) != 0)
350 /* We aren't prepared to handle relocations in merged sections. */
351 return TRUE;
354 align = sec->alignment_power;
355 if ((sec->entsize < (unsigned) 1 << align
356 && ((sec->entsize & (sec->entsize - 1))
357 || !(sec->flags & SEC_STRINGS)))
358 || (sec->entsize > (unsigned) 1 << align
359 && (sec->entsize & (((unsigned) 1 << align) - 1))))
361 /* Sanity check. If string character size is smaller than
362 alignment, then we require character size to be a power
363 of 2, otherwise character size must be integer multiple
364 of alignment. For non-string constants, alignment must
365 be smaller than or equal to entity size and entity size
366 must be integer multiple of alignment. */
367 return TRUE;
370 for (sinfo = (struct sec_merge_info *) *psinfo; sinfo; sinfo = sinfo->next)
371 if ((secinfo = sinfo->chain)
372 && ! ((secinfo->sec->flags ^ sec->flags) & (SEC_MERGE | SEC_STRINGS))
373 && secinfo->sec->entsize == sec->entsize
374 && secinfo->sec->alignment_power == sec->alignment_power
375 && secinfo->sec->output_section == sec->output_section)
376 break;
378 if (sinfo == NULL)
380 /* Initialize the information we need to keep track of. */
381 sinfo = bfd_alloc (abfd, sizeof (struct sec_merge_info));
382 if (sinfo == NULL)
383 goto error_return;
384 sinfo->next = (struct sec_merge_info *) *psinfo;
385 sinfo->chain = NULL;
386 *psinfo = sinfo;
387 sinfo->htab = sec_merge_init (sec->entsize, (sec->flags & SEC_STRINGS));
388 if (sinfo->htab == NULL)
389 goto error_return;
392 /* Read the section from abfd. */
394 amt = sizeof (struct sec_merge_sec_info) + sec->size - 1;
395 *psecinfo = bfd_alloc (abfd, amt);
396 if (*psecinfo == NULL)
397 goto error_return;
399 secinfo = (struct sec_merge_sec_info *) *psecinfo;
400 if (sinfo->chain)
402 secinfo->next = sinfo->chain->next;
403 sinfo->chain->next = secinfo;
405 else
406 secinfo->next = secinfo;
407 sinfo->chain = secinfo;
408 secinfo->sec = sec;
409 secinfo->psecinfo = psecinfo;
410 secinfo->htab = sinfo->htab;
411 secinfo->first_str = NULL;
413 sec->rawsize = sec->size;
414 if (! bfd_get_section_contents (sec->owner, sec, secinfo->contents,
415 0, sec->size))
416 goto error_return;
418 return TRUE;
420 error_return:
421 *psecinfo = NULL;
422 return FALSE;
425 /* Record one section into the hash table. */
426 static bfd_boolean
427 record_section (struct sec_merge_info *sinfo,
428 struct sec_merge_sec_info *secinfo)
430 asection *sec = secinfo->sec;
431 struct sec_merge_hash_entry *entry;
432 bfd_boolean nul;
433 unsigned char *p, *end;
434 bfd_vma mask, eltalign;
435 unsigned int align, i;
437 align = sec->alignment_power;
438 end = secinfo->contents + sec->size;
439 nul = FALSE;
440 mask = ((bfd_vma) 1 << align) - 1;
441 if (sec->flags & SEC_STRINGS)
443 for (p = secinfo->contents; p < end; )
445 eltalign = p - secinfo->contents;
446 eltalign = ((eltalign ^ (eltalign - 1)) + 1) >> 1;
447 if (!eltalign || eltalign > mask)
448 eltalign = mask + 1;
449 entry = sec_merge_add (sinfo->htab, (char *) p, (unsigned) eltalign,
450 secinfo);
451 if (! entry)
452 goto error_return;
453 p += entry->len;
454 if (sec->entsize == 1)
456 while (p < end && *p == 0)
458 if (!nul && !((p - secinfo->contents) & mask))
460 nul = TRUE;
461 entry = sec_merge_add (sinfo->htab, "",
462 (unsigned) mask + 1, secinfo);
463 if (! entry)
464 goto error_return;
466 p++;
469 else
471 while (p < end)
473 for (i = 0; i < sec->entsize; i++)
474 if (p[i] != '\0')
475 break;
476 if (i != sec->entsize)
477 break;
478 if (!nul && !((p - secinfo->contents) & mask))
480 nul = TRUE;
481 entry = sec_merge_add (sinfo->htab, (char *) p,
482 (unsigned) mask + 1, secinfo);
483 if (! entry)
484 goto error_return;
486 p += sec->entsize;
491 else
493 for (p = secinfo->contents; p < end; p += sec->entsize)
495 entry = sec_merge_add (sinfo->htab, (char *) p, 1, secinfo);
496 if (! entry)
497 goto error_return;
501 return TRUE;
503 error_return:
504 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
505 *secinfo->psecinfo = NULL;
506 return FALSE;
509 static int
510 strrevcmp (const void *a, const void *b)
512 struct sec_merge_hash_entry *A = *(struct sec_merge_hash_entry **) a;
513 struct sec_merge_hash_entry *B = *(struct sec_merge_hash_entry **) b;
514 unsigned int lenA = A->len;
515 unsigned int lenB = B->len;
516 const unsigned char *s = (const unsigned char *) A->root.string + lenA - 1;
517 const unsigned char *t = (const unsigned char *) B->root.string + lenB - 1;
518 int l = lenA < lenB ? lenA : lenB;
520 while (l)
522 if (*s != *t)
523 return (int) *s - (int) *t;
524 s--;
525 t--;
526 l--;
528 return lenA - lenB;
531 /* Like strrevcmp, but for the case where all strings have the same
532 alignment > entsize. */
534 static int
535 strrevcmp_align (const void *a, const void *b)
537 struct sec_merge_hash_entry *A = *(struct sec_merge_hash_entry **) a;
538 struct sec_merge_hash_entry *B = *(struct sec_merge_hash_entry **) b;
539 unsigned int lenA = A->len;
540 unsigned int lenB = B->len;
541 const unsigned char *s = (const unsigned char *) A->root.string + lenA - 1;
542 const unsigned char *t = (const unsigned char *) B->root.string + lenB - 1;
543 int l = lenA < lenB ? lenA : lenB;
544 int tail_align = (lenA & (A->alignment - 1)) - (lenB & (A->alignment - 1));
546 if (tail_align != 0)
547 return tail_align;
549 while (l)
551 if (*s != *t)
552 return (int) *s - (int) *t;
553 s--;
554 t--;
555 l--;
557 return lenA - lenB;
560 static inline int
561 is_suffix (const struct sec_merge_hash_entry *A,
562 const struct sec_merge_hash_entry *B)
564 if (A->len <= B->len)
565 /* B cannot be a suffix of A unless A is equal to B, which is guaranteed
566 not to be equal by the hash table. */
567 return 0;
569 return memcmp (A->root.string + (A->len - B->len),
570 B->root.string, B->len) == 0;
573 /* This is a helper function for _bfd_merge_sections. It attempts to
574 merge strings matching suffixes of longer strings. */
575 static void
576 merge_strings (struct sec_merge_info *sinfo)
578 struct sec_merge_hash_entry **array, **a, *e;
579 struct sec_merge_sec_info *secinfo;
580 bfd_size_type size, amt;
581 unsigned int alignment = 0;
583 /* Now sort the strings */
584 amt = sinfo->htab->size * sizeof (struct sec_merge_hash_entry *);
585 array = bfd_malloc (amt);
586 if (array == NULL)
587 goto alloc_failure;
589 for (e = sinfo->htab->first, a = array; e; e = e->next)
590 if (e->alignment)
592 *a++ = e;
593 /* Adjust the length to not include the zero terminator. */
594 e->len -= sinfo->htab->entsize;
595 if (alignment != e->alignment)
597 if (alignment == 0)
598 alignment = e->alignment;
599 else
600 alignment = (unsigned) -1;
604 sinfo->htab->size = a - array;
605 if (sinfo->htab->size != 0)
607 qsort (array, (size_t) sinfo->htab->size,
608 sizeof (struct sec_merge_hash_entry *),
609 (alignment != (unsigned) -1 && alignment > sinfo->htab->entsize
610 ? strrevcmp_align : strrevcmp));
612 /* Loop over the sorted array and merge suffixes */
613 e = *--a;
614 e->len += sinfo->htab->entsize;
615 while (--a >= array)
617 struct sec_merge_hash_entry *cmp = *a;
619 cmp->len += sinfo->htab->entsize;
620 if (e->alignment >= cmp->alignment
621 && !((e->len - cmp->len) & (cmp->alignment - 1))
622 && is_suffix (e, cmp))
624 cmp->u.suffix = e;
625 cmp->alignment = 0;
627 else
628 e = cmp;
632 alloc_failure:
633 if (array)
634 free (array);
636 /* Now assign positions to the strings we want to keep. */
637 size = 0;
638 secinfo = sinfo->htab->first->secinfo;
639 for (e = sinfo->htab->first; e; e = e->next)
641 if (e->secinfo != secinfo)
643 secinfo->sec->size = size;
644 secinfo = e->secinfo;
646 if (e->alignment)
648 if (e->secinfo->first_str == NULL)
650 e->secinfo->first_str = e;
651 size = 0;
653 size = (size + e->alignment - 1) & ~((bfd_vma) e->alignment - 1);
654 e->u.index = size;
655 size += e->len;
658 secinfo->sec->size = size;
660 /* And now adjust the rest, removing them from the chain (but not hashtable)
661 at the same time. */
662 for (a = &sinfo->htab->first, e = *a; e; e = e->next)
663 if (e->alignment)
664 a = &e->next;
665 else
667 *a = e->next;
668 if (e->len)
670 e->secinfo = e->u.suffix->secinfo;
671 e->alignment = e->u.suffix->alignment;
672 e->u.index = e->u.suffix->u.index + (e->u.suffix->len - e->len);
677 /* This function is called once after all SEC_MERGE sections are registered
678 with _bfd_merge_section. */
680 bfd_boolean
681 _bfd_merge_sections (bfd *abfd ATTRIBUTE_UNUSED, struct bfd_link_info *info,
682 void *xsinfo, void (*remove_hook) (bfd *, asection *))
684 struct sec_merge_info *sinfo;
686 for (sinfo = (struct sec_merge_info *) xsinfo; sinfo; sinfo = sinfo->next)
688 struct sec_merge_sec_info * secinfo;
690 if (! sinfo->chain)
691 continue;
693 /* Move sinfo->chain to head of the chain, terminate it. */
694 secinfo = sinfo->chain;
695 sinfo->chain = secinfo->next;
696 secinfo->next = NULL;
698 /* Record the sections into the hash table. */
699 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
700 if (secinfo->sec->flags & SEC_EXCLUDE)
702 *secinfo->psecinfo = NULL;
703 if (remove_hook)
704 (*remove_hook) (abfd, secinfo->sec);
706 else if (! record_section (sinfo, secinfo))
707 break;
709 if (secinfo)
710 continue;
712 if (sinfo->htab->first == NULL)
713 continue;
715 if (sinfo->htab->strings)
716 merge_strings (sinfo);
717 else
719 struct sec_merge_hash_entry *e;
720 bfd_size_type size = 0;
722 /* Things are much simpler for non-strings.
723 Just assign them slots in the section. */
724 secinfo = NULL;
725 for (e = sinfo->htab->first; e; e = e->next)
727 if (e->secinfo->first_str == NULL)
729 if (secinfo)
730 secinfo->sec->size = size;
731 e->secinfo->first_str = e;
732 size = 0;
734 size = (size + e->alignment - 1)
735 & ~((bfd_vma) e->alignment - 1);
736 e->u.index = size;
737 size += e->len;
738 secinfo = e->secinfo;
740 secinfo->sec->size = size;
743 /* Finally remove all input sections which have not made it into
744 the hash table at all. */
745 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
746 if (secinfo->first_str == NULL)
747 _bfd_strip_section_from_output (info, secinfo->sec);
750 return TRUE;
753 /* Write out the merged section. */
755 bfd_boolean
756 _bfd_write_merged_section (bfd *output_bfd, asection *sec, void *psecinfo)
758 struct sec_merge_sec_info *secinfo;
759 file_ptr pos;
761 secinfo = (struct sec_merge_sec_info *) psecinfo;
763 if (secinfo->first_str == NULL)
764 return TRUE;
766 pos = sec->output_section->filepos + sec->output_offset;
767 if (bfd_seek (output_bfd, pos, SEEK_SET) != 0)
768 return FALSE;
770 if (! sec_merge_emit (output_bfd, secinfo->first_str))
771 return FALSE;
773 return TRUE;
776 /* Adjust an address in the SEC_MERGE section. Given OFFSET within
777 *PSEC, this returns the new offset in the adjusted SEC_MERGE
778 section and writes the new section back into *PSEC. */
780 bfd_vma
781 _bfd_merged_section_offset (bfd *output_bfd ATTRIBUTE_UNUSED, asection **psec,
782 void *psecinfo, bfd_vma offset)
784 struct sec_merge_sec_info *secinfo;
785 struct sec_merge_hash_entry *entry;
786 unsigned char *p;
787 asection *sec = *psec;
789 secinfo = (struct sec_merge_sec_info *) psecinfo;
791 if (offset >= sec->rawsize)
793 if (offset > sec->rawsize)
795 (*_bfd_error_handler)
796 (_("%s: access beyond end of merged section (%ld)"),
797 bfd_get_filename (sec->owner), (long) offset);
799 return secinfo->first_str ? sec->size : 0;
802 if (secinfo->htab->strings)
804 if (sec->entsize == 1)
806 p = secinfo->contents + offset - 1;
807 while (p >= secinfo->contents && *p)
808 --p;
809 ++p;
811 else
813 p = secinfo->contents + (offset / sec->entsize) * sec->entsize;
814 p -= sec->entsize;
815 while (p >= secinfo->contents)
817 unsigned int i;
819 for (i = 0; i < sec->entsize; ++i)
820 if (p[i] != '\0')
821 break;
822 if (i == sec->entsize)
823 break;
824 p -= sec->entsize;
826 p += sec->entsize;
829 else
831 p = secinfo->contents + (offset / sec->entsize) * sec->entsize;
833 entry = sec_merge_hash_lookup (secinfo->htab, (char *) p, 0, FALSE);
834 if (!entry)
836 if (! secinfo->htab->strings)
837 abort ();
838 /* This should only happen if somebody points into the padding
839 after a NUL character but before next entity. */
840 if (*p)
841 abort ();
842 if (! secinfo->htab->first)
843 abort ();
844 entry = secinfo->htab->first;
845 p = (secinfo->contents + (offset / sec->entsize + 1) * sec->entsize
846 - entry->len);
849 *psec = entry->secinfo->sec;
850 return entry->u.index + (secinfo->contents + offset - p);