2 Copyright 2001, 2002, 2003, 2004, 2005, 2006, 2007
3 Free Software Foundation, Inc.
4 Written by Jakub Jelinek <jakub@redhat.com>.
6 This file is part of BFD, the Binary File Descriptor library.
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. */
24 /* This file contains support for merging duplicate entities within sections,
25 as used in ELF SHF_MERGE. */
31 #include "libiberty.h"
33 struct sec_merge_sec_info;
35 /* An entry in the section merge hash table. */
37 struct sec_merge_hash_entry
39 struct bfd_hash_entry root;
40 /* Length of this entry. This includes the zero terminator. */
42 /* Start of this string needs to be aligned to
43 alignment octets (not 1 << align). */
44 unsigned int alignment;
47 /* Index within the merged section. */
49 /* Entry this is a suffix of (if alignment is 0). */
50 struct sec_merge_hash_entry *suffix;
52 /* Which section is it in. */
53 struct sec_merge_sec_info *secinfo;
54 /* Next entity in the hash table. */
55 struct sec_merge_hash_entry *next;
58 /* The section merge hash table. */
62 struct bfd_hash_table table;
63 /* Next available index. */
65 /* First entity in the SEC_MERGE sections of this type. */
66 struct sec_merge_hash_entry *first;
67 /* Last entity in the SEC_MERGE sections of this type. */
68 struct sec_merge_hash_entry *last;
71 /* Are entries fixed size or zero terminated strings? */
77 /* Chain of sec_merge_infos. */
78 struct sec_merge_info *next;
79 /* Chain of sec_merge_sec_infos. */
80 struct sec_merge_sec_info *chain;
81 /* A hash table used to hold section content. */
82 struct sec_merge_hash *htab;
85 struct sec_merge_sec_info
87 /* Chain of sec_merge_sec_infos. */
88 struct sec_merge_sec_info *next;
89 /* The corresponding section. */
91 /* Pointer to merge_info pointing to us. */
93 /* A hash table used to hold section content. */
94 struct sec_merge_hash *htab;
95 /* First string in this section. */
96 struct sec_merge_hash_entry *first_str;
97 /* Original section content. */
98 unsigned char contents[1];
102 /* Routine to create an entry in a section merge hashtab. */
104 static struct bfd_hash_entry *
105 sec_merge_hash_newfunc (struct bfd_hash_entry *entry,
106 struct bfd_hash_table *table, const char *string)
108 /* Allocate the structure if it has not already been allocated by a
111 entry = bfd_hash_allocate (table, sizeof (struct sec_merge_hash_entry));
115 /* Call the allocation method of the superclass. */
116 entry = bfd_hash_newfunc (entry, table, string);
120 /* Initialize the local fields. */
121 struct sec_merge_hash_entry *ret = (struct sec_merge_hash_entry *) entry;
123 ret->u.suffix = NULL;
132 /* Look up an entry in a section merge hash table. */
134 static struct sec_merge_hash_entry *
135 sec_merge_hash_lookup (struct sec_merge_hash *table, const char *string,
136 unsigned int alignment, bfd_boolean create)
138 register const unsigned char *s;
139 register unsigned long hash;
140 register unsigned int c;
141 struct sec_merge_hash_entry *hashp;
147 s = (const unsigned char *) string;
150 if (table->entsize == 1)
152 while ((c = *s++) != '\0')
154 hash += c + (c << 17);
158 hash += len + (len << 17);
164 for (i = 0; i < table->entsize; ++i)
167 if (i == table->entsize)
169 for (i = 0; i < table->entsize; ++i)
172 hash += c + (c << 17);
177 hash += len + (len << 17);
178 len *= table->entsize;
181 len += table->entsize;
185 for (i = 0; i < table->entsize; ++i)
188 hash += c + (c << 17);
191 len = table->entsize;
194 index = hash % table->table.size;
195 for (hashp = (struct sec_merge_hash_entry *) table->table.table[index];
197 hashp = (struct sec_merge_hash_entry *) hashp->root.next)
199 if (hashp->root.hash == hash
201 && memcmp (hashp->root.string, string, len) == 0)
203 /* If the string we found does not have at least the required
204 alignment, we need to insert another copy. */
205 if (hashp->alignment < alignment)
209 /* Mark the less aligned copy as deleted. */
211 hashp->alignment = 0;
222 hashp = ((struct sec_merge_hash_entry *)
223 sec_merge_hash_newfunc (NULL, &table->table, string));
226 hashp->root.string = string;
227 hashp->root.hash = hash;
229 hashp->alignment = alignment;
230 hashp->root.next = table->table.table[index];
231 table->table.table[index] = (struct bfd_hash_entry *) hashp;
236 /* Create a new hash table. */
238 static struct sec_merge_hash *
239 sec_merge_init (unsigned int entsize, bfd_boolean strings)
241 struct sec_merge_hash *table;
243 table = bfd_malloc (sizeof (struct sec_merge_hash));
247 if (! bfd_hash_table_init_n (&table->table, sec_merge_hash_newfunc,
248 sizeof (struct sec_merge_hash_entry), 16699))
257 table->entsize = entsize;
258 table->strings = strings;
263 /* Get the index of an entity in a hash table, adding it if it is not
266 static struct sec_merge_hash_entry *
267 sec_merge_add (struct sec_merge_hash *tab, const char *str,
268 unsigned int alignment, struct sec_merge_sec_info *secinfo)
270 register struct sec_merge_hash_entry *entry;
272 entry = sec_merge_hash_lookup (tab, str, alignment, TRUE);
276 if (entry->secinfo == NULL)
279 entry->secinfo = secinfo;
280 if (tab->first == NULL)
283 tab->last->next = entry;
291 sec_merge_emit (bfd *abfd, struct sec_merge_hash_entry *entry)
293 struct sec_merge_sec_info *secinfo = entry->secinfo;
294 asection *sec = secinfo->sec;
296 bfd_size_type off = 0;
297 int alignment_power = sec->output_section->alignment_power;
301 pad = bfd_zmalloc ((bfd_size_type) 1 << alignment_power);
306 for (; entry != NULL && entry->secinfo == secinfo; entry = entry->next)
311 len = -off & (entry->alignment - 1);
314 if (bfd_bwrite (pad, len, abfd) != len)
319 str = entry->root.string;
322 if (bfd_bwrite (str, len, abfd) != len)
328 /* Trailing alignment needed? */
329 off = sec->size - off;
331 && bfd_bwrite (pad, off, abfd) != off)
344 /* Register a SEC_MERGE section as a candidate for merging.
345 This function is called for all non-dynamic SEC_MERGE input sections. */
348 _bfd_add_merge_section (bfd *abfd, void **psinfo, asection *sec,
351 struct sec_merge_info *sinfo;
352 struct sec_merge_sec_info *secinfo;
356 if ((abfd->flags & DYNAMIC) != 0
357 || (sec->flags & SEC_MERGE) == 0)
361 || (sec->flags & SEC_EXCLUDE) != 0
362 || sec->entsize == 0)
365 if ((sec->flags & SEC_RELOC) != 0)
367 /* We aren't prepared to handle relocations in merged sections. */
371 align = sec->alignment_power;
372 if ((sec->entsize < (unsigned) 1 << align
373 && ((sec->entsize & (sec->entsize - 1))
374 || !(sec->flags & SEC_STRINGS)))
375 || (sec->entsize > (unsigned) 1 << align
376 && (sec->entsize & (((unsigned) 1 << align) - 1))))
378 /* Sanity check. If string character size is smaller than
379 alignment, then we require character size to be a power
380 of 2, otherwise character size must be integer multiple
381 of alignment. For non-string constants, alignment must
382 be smaller than or equal to entity size and entity size
383 must be integer multiple of alignment. */
387 for (sinfo = (struct sec_merge_info *) *psinfo; sinfo; sinfo = sinfo->next)
388 if ((secinfo = sinfo->chain)
389 && ! ((secinfo->sec->flags ^ sec->flags) & (SEC_MERGE | SEC_STRINGS))
390 && secinfo->sec->entsize == sec->entsize
391 && secinfo->sec->alignment_power == sec->alignment_power
392 && secinfo->sec->output_section == sec->output_section)
397 /* Initialize the information we need to keep track of. */
398 sinfo = bfd_alloc (abfd, sizeof (struct sec_merge_info));
401 sinfo->next = (struct sec_merge_info *) *psinfo;
404 sinfo->htab = sec_merge_init (sec->entsize, (sec->flags & SEC_STRINGS));
405 if (sinfo->htab == NULL)
409 /* Read the section from abfd. */
411 amt = sizeof (struct sec_merge_sec_info) + sec->size - 1;
412 *psecinfo = bfd_alloc (abfd, amt);
413 if (*psecinfo == NULL)
416 secinfo = (struct sec_merge_sec_info *) *psecinfo;
419 secinfo->next = sinfo->chain->next;
420 sinfo->chain->next = secinfo;
423 secinfo->next = secinfo;
424 sinfo->chain = secinfo;
426 secinfo->psecinfo = psecinfo;
427 secinfo->htab = sinfo->htab;
428 secinfo->first_str = NULL;
430 sec->rawsize = sec->size;
431 if (! bfd_get_section_contents (sec->owner, sec, secinfo->contents,
442 /* Record one section into the hash table. */
444 record_section (struct sec_merge_info *sinfo,
445 struct sec_merge_sec_info *secinfo)
447 asection *sec = secinfo->sec;
448 struct sec_merge_hash_entry *entry;
450 unsigned char *p, *end;
451 bfd_vma mask, eltalign;
452 unsigned int align, i;
454 align = sec->alignment_power;
455 end = secinfo->contents + sec->size;
457 mask = ((bfd_vma) 1 << align) - 1;
458 if (sec->flags & SEC_STRINGS)
460 for (p = secinfo->contents; p < end; )
462 eltalign = p - secinfo->contents;
463 eltalign = ((eltalign ^ (eltalign - 1)) + 1) >> 1;
464 if (!eltalign || eltalign > mask)
466 entry = sec_merge_add (sinfo->htab, (char *) p, (unsigned) eltalign,
471 if (sec->entsize == 1)
473 while (p < end && *p == 0)
475 if (!nul && !((p - secinfo->contents) & mask))
478 entry = sec_merge_add (sinfo->htab, "",
479 (unsigned) mask + 1, secinfo);
490 for (i = 0; i < sec->entsize; i++)
493 if (i != sec->entsize)
495 if (!nul && !((p - secinfo->contents) & mask))
498 entry = sec_merge_add (sinfo->htab, (char *) p,
499 (unsigned) mask + 1, secinfo);
510 for (p = secinfo->contents; p < end; p += sec->entsize)
512 entry = sec_merge_add (sinfo->htab, (char *) p, 1, secinfo);
521 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
522 *secinfo->psecinfo = NULL;
527 strrevcmp (const void *a, const void *b)
529 struct sec_merge_hash_entry *A = *(struct sec_merge_hash_entry **) a;
530 struct sec_merge_hash_entry *B = *(struct sec_merge_hash_entry **) b;
531 unsigned int lenA = A->len;
532 unsigned int lenB = B->len;
533 const unsigned char *s = (const unsigned char *) A->root.string + lenA - 1;
534 const unsigned char *t = (const unsigned char *) B->root.string + lenB - 1;
535 int l = lenA < lenB ? lenA : lenB;
540 return (int) *s - (int) *t;
548 /* Like strrevcmp, but for the case where all strings have the same
549 alignment > entsize. */
552 strrevcmp_align (const void *a, const void *b)
554 struct sec_merge_hash_entry *A = *(struct sec_merge_hash_entry **) a;
555 struct sec_merge_hash_entry *B = *(struct sec_merge_hash_entry **) b;
556 unsigned int lenA = A->len;
557 unsigned int lenB = B->len;
558 const unsigned char *s = (const unsigned char *) A->root.string + lenA - 1;
559 const unsigned char *t = (const unsigned char *) B->root.string + lenB - 1;
560 int l = lenA < lenB ? lenA : lenB;
561 int tail_align = (lenA & (A->alignment - 1)) - (lenB & (A->alignment - 1));
569 return (int) *s - (int) *t;
578 is_suffix (const struct sec_merge_hash_entry *A,
579 const struct sec_merge_hash_entry *B)
581 if (A->len <= B->len)
582 /* B cannot be a suffix of A unless A is equal to B, which is guaranteed
583 not to be equal by the hash table. */
586 return memcmp (A->root.string + (A->len - B->len),
587 B->root.string, B->len) == 0;
590 /* This is a helper function for _bfd_merge_sections. It attempts to
591 merge strings matching suffixes of longer strings. */
593 merge_strings (struct sec_merge_info *sinfo)
595 struct sec_merge_hash_entry **array, **a, *e;
596 struct sec_merge_sec_info *secinfo;
597 bfd_size_type size, amt;
598 unsigned int alignment = 0;
600 /* Now sort the strings */
601 amt = sinfo->htab->size * sizeof (struct sec_merge_hash_entry *);
602 array = bfd_malloc (amt);
606 for (e = sinfo->htab->first, a = array; e; e = e->next)
610 /* Adjust the length to not include the zero terminator. */
611 e->len -= sinfo->htab->entsize;
612 if (alignment != e->alignment)
615 alignment = e->alignment;
617 alignment = (unsigned) -1;
621 sinfo->htab->size = a - array;
622 if (sinfo->htab->size != 0)
624 qsort (array, (size_t) sinfo->htab->size,
625 sizeof (struct sec_merge_hash_entry *),
626 (alignment != (unsigned) -1 && alignment > sinfo->htab->entsize
627 ? strrevcmp_align : strrevcmp));
629 /* Loop over the sorted array and merge suffixes */
631 e->len += sinfo->htab->entsize;
634 struct sec_merge_hash_entry *cmp = *a;
636 cmp->len += sinfo->htab->entsize;
637 if (e->alignment >= cmp->alignment
638 && !((e->len - cmp->len) & (cmp->alignment - 1))
639 && is_suffix (e, cmp))
653 /* Now assign positions to the strings we want to keep. */
655 secinfo = sinfo->htab->first->secinfo;
656 for (e = sinfo->htab->first; e; e = e->next)
658 if (e->secinfo != secinfo)
660 secinfo->sec->size = size;
661 secinfo = e->secinfo;
665 if (e->secinfo->first_str == NULL)
667 e->secinfo->first_str = e;
670 size = (size + e->alignment - 1) & ~((bfd_vma) e->alignment - 1);
675 secinfo->sec->size = size;
676 if (secinfo->sec->alignment_power != 0)
678 bfd_size_type align = (bfd_size_type) 1 << secinfo->sec->alignment_power;
679 secinfo->sec->size = (secinfo->sec->size + align - 1) & -align;
682 /* And now adjust the rest, removing them from the chain (but not hashtable)
684 for (a = &sinfo->htab->first, e = *a; e; e = e->next)
692 e->secinfo = e->u.suffix->secinfo;
693 e->alignment = e->u.suffix->alignment;
694 e->u.index = e->u.suffix->u.index + (e->u.suffix->len - e->len);
699 /* This function is called once after all SEC_MERGE sections are registered
700 with _bfd_merge_section. */
703 _bfd_merge_sections (bfd *abfd ATTRIBUTE_UNUSED,
704 struct bfd_link_info *info ATTRIBUTE_UNUSED,
706 void (*remove_hook) (bfd *, asection *))
708 struct sec_merge_info *sinfo;
710 for (sinfo = (struct sec_merge_info *) xsinfo; sinfo; sinfo = sinfo->next)
712 struct sec_merge_sec_info * secinfo;
717 /* Move sinfo->chain to head of the chain, terminate it. */
718 secinfo = sinfo->chain;
719 sinfo->chain = secinfo->next;
720 secinfo->next = NULL;
722 /* Record the sections into the hash table. */
723 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
724 if (secinfo->sec->flags & SEC_EXCLUDE)
726 *secinfo->psecinfo = NULL;
728 (*remove_hook) (abfd, secinfo->sec);
730 else if (! record_section (sinfo, secinfo))
736 if (sinfo->htab->first == NULL)
739 if (sinfo->htab->strings)
740 merge_strings (sinfo);
743 struct sec_merge_hash_entry *e;
744 bfd_size_type size = 0;
746 /* Things are much simpler for non-strings.
747 Just assign them slots in the section. */
749 for (e = sinfo->htab->first; e; e = e->next)
751 if (e->secinfo->first_str == NULL)
754 secinfo->sec->size = size;
755 e->secinfo->first_str = e;
758 size = (size + e->alignment - 1)
759 & ~((bfd_vma) e->alignment - 1);
762 secinfo = e->secinfo;
764 secinfo->sec->size = size;
767 /* Finally remove all input sections which have not made it into
768 the hash table at all. */
769 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
770 if (secinfo->first_str == NULL)
771 secinfo->sec->flags |= SEC_EXCLUDE | SEC_KEEP;
777 /* Write out the merged section. */
780 _bfd_write_merged_section (bfd *output_bfd, asection *sec, void *psecinfo)
782 struct sec_merge_sec_info *secinfo;
785 secinfo = (struct sec_merge_sec_info *) psecinfo;
787 if (secinfo->first_str == NULL)
790 pos = sec->output_section->filepos + sec->output_offset;
791 if (bfd_seek (output_bfd, pos, SEEK_SET) != 0)
794 if (! sec_merge_emit (output_bfd, secinfo->first_str))
800 /* Adjust an address in the SEC_MERGE section. Given OFFSET within
801 *PSEC, this returns the new offset in the adjusted SEC_MERGE
802 section and writes the new section back into *PSEC. */
805 _bfd_merged_section_offset (bfd *output_bfd ATTRIBUTE_UNUSED, asection **psec,
806 void *psecinfo, bfd_vma offset)
808 struct sec_merge_sec_info *secinfo;
809 struct sec_merge_hash_entry *entry;
811 asection *sec = *psec;
813 secinfo = (struct sec_merge_sec_info *) psecinfo;
815 if (offset >= sec->rawsize)
817 if (offset > sec->rawsize)
819 (*_bfd_error_handler)
820 (_("%s: access beyond end of merged section (%ld)"),
821 bfd_get_filename (sec->owner), (long) offset);
823 return secinfo->first_str ? sec->size : 0;
826 if (secinfo->htab->strings)
828 if (sec->entsize == 1)
830 p = secinfo->contents + offset - 1;
831 while (p >= secinfo->contents && *p)
837 p = secinfo->contents + (offset / sec->entsize) * sec->entsize;
839 while (p >= secinfo->contents)
843 for (i = 0; i < sec->entsize; ++i)
846 if (i == sec->entsize)
855 p = secinfo->contents + (offset / sec->entsize) * sec->entsize;
857 entry = sec_merge_hash_lookup (secinfo->htab, (char *) p, 0, FALSE);
860 if (! secinfo->htab->strings)
862 /* This should only happen if somebody points into the padding
863 after a NUL character but before next entity. */
866 if (! secinfo->htab->first)
868 entry = secinfo->htab->first;
869 p = (secinfo->contents + (offset / sec->entsize + 1) * sec->entsize
873 *psec = entry->secinfo->sec;
874 return entry->u.index + (secinfo->contents + offset - p);