2 Copyright 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009
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 = (struct bfd_hash_entry *)
112 bfd_hash_allocate (table, sizeof (struct sec_merge_hash_entry));
116 /* Call the allocation method of the superclass. */
117 entry = bfd_hash_newfunc (entry, table, string);
121 /* Initialize the local fields. */
122 struct sec_merge_hash_entry *ret = (struct sec_merge_hash_entry *) entry;
124 ret->u.suffix = NULL;
133 /* Look up an entry in a section merge hash table. */
135 static struct sec_merge_hash_entry *
136 sec_merge_hash_lookup (struct sec_merge_hash *table, const char *string,
137 unsigned int alignment, bfd_boolean create)
139 const unsigned char *s;
142 struct sec_merge_hash_entry *hashp;
148 s = (const unsigned char *) string;
151 if (table->entsize == 1)
153 while ((c = *s++) != '\0')
155 hash += c + (c << 17);
159 hash += len + (len << 17);
165 for (i = 0; i < table->entsize; ++i)
168 if (i == table->entsize)
170 for (i = 0; i < table->entsize; ++i)
173 hash += c + (c << 17);
178 hash += len + (len << 17);
179 len *= table->entsize;
182 len += table->entsize;
186 for (i = 0; i < table->entsize; ++i)
189 hash += c + (c << 17);
192 len = table->entsize;
195 _index = hash % table->table.size;
196 for (hashp = (struct sec_merge_hash_entry *) table->table.table[_index];
198 hashp = (struct sec_merge_hash_entry *) hashp->root.next)
200 if (hashp->root.hash == hash
202 && memcmp (hashp->root.string, string, len) == 0)
204 /* If the string we found does not have at least the required
205 alignment, we need to insert another copy. */
206 if (hashp->alignment < alignment)
210 /* Mark the less aligned copy as deleted. */
212 hashp->alignment = 0;
223 hashp = ((struct sec_merge_hash_entry *)
224 bfd_hash_insert (&table->table, string, hash));
228 hashp->alignment = alignment;
232 /* Create a new hash table. */
234 static struct sec_merge_hash *
235 sec_merge_init (unsigned int entsize, bfd_boolean strings)
237 struct sec_merge_hash *table;
239 table = (struct sec_merge_hash *) bfd_malloc (sizeof (struct sec_merge_hash));
243 if (! bfd_hash_table_init_n (&table->table, sec_merge_hash_newfunc,
244 sizeof (struct sec_merge_hash_entry), 16699))
253 table->entsize = entsize;
254 table->strings = strings;
259 /* Get the index of an entity in a hash table, adding it if it is not
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 struct sec_merge_hash_entry *entry;
268 entry = sec_merge_hash_lookup (tab, str, alignment, TRUE);
272 if (entry->secinfo == NULL)
275 entry->secinfo = secinfo;
276 if (tab->first == NULL)
279 tab->last->next = entry;
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;
292 bfd_size_type off = 0;
293 int alignment_power = sec->output_section->alignment_power;
297 pad = (char *) bfd_zmalloc ((bfd_size_type) 1 << alignment_power);
302 for (; entry != NULL && entry->secinfo == secinfo; entry = entry->next)
307 len = -off & (entry->alignment - 1);
310 if (bfd_bwrite (pad, len, abfd) != len)
315 str = entry->root.string;
318 if (bfd_bwrite (str, len, abfd) != len)
324 /* Trailing alignment needed? */
325 off = sec->size - off;
327 && bfd_bwrite (pad, off, abfd) != off)
340 /* Register a SEC_MERGE section as a candidate for merging.
341 This function is called for all non-dynamic SEC_MERGE input sections. */
344 _bfd_add_merge_section (bfd *abfd, void **psinfo, asection *sec,
347 struct sec_merge_info *sinfo;
348 struct sec_merge_sec_info *secinfo;
353 if ((abfd->flags & DYNAMIC) != 0
354 || (sec->flags & SEC_MERGE) == 0)
358 || (sec->flags & SEC_EXCLUDE) != 0
359 || sec->entsize == 0)
362 if ((sec->flags & SEC_RELOC) != 0)
364 /* We aren't prepared to handle relocations in merged sections. */
368 align = sec->alignment_power;
369 if ((sec->entsize < (unsigned) 1 << align
370 && ((sec->entsize & (sec->entsize - 1))
371 || !(sec->flags & SEC_STRINGS)))
372 || (sec->entsize > (unsigned) 1 << align
373 && (sec->entsize & (((unsigned) 1 << align) - 1))))
375 /* Sanity check. If string character size is smaller than
376 alignment, then we require character size to be a power
377 of 2, otherwise character size must be integer multiple
378 of alignment. For non-string constants, alignment must
379 be smaller than or equal to entity size and entity size
380 must be integer multiple of alignment. */
384 for (sinfo = (struct sec_merge_info *) *psinfo; sinfo; sinfo = sinfo->next)
385 if ((secinfo = sinfo->chain)
386 && ! ((secinfo->sec->flags ^ sec->flags) & (SEC_MERGE | SEC_STRINGS))
387 && secinfo->sec->entsize == sec->entsize
388 && secinfo->sec->alignment_power == sec->alignment_power
389 && secinfo->sec->output_section == sec->output_section)
394 /* Initialize the information we need to keep track of. */
395 sinfo = (struct sec_merge_info *)
396 bfd_alloc (abfd, sizeof (struct sec_merge_info));
399 sinfo->next = (struct sec_merge_info *) *psinfo;
402 sinfo->htab = sec_merge_init (sec->entsize, (sec->flags & SEC_STRINGS));
403 if (sinfo->htab == NULL)
407 /* Read the section from abfd. */
409 amt = sizeof (struct sec_merge_sec_info) - 1 + sec->size;
410 if (sec->flags & SEC_STRINGS)
411 /* Some versions of gcc may emit a string without a zero terminator.
412 See http://gcc.gnu.org/ml/gcc-patches/2006-06/msg01004.html
413 Allocate space for an extra zero. */
415 *psecinfo = bfd_alloc (abfd, amt);
416 if (*psecinfo == NULL)
419 secinfo = (struct sec_merge_sec_info *) *psecinfo;
422 secinfo->next = sinfo->chain->next;
423 sinfo->chain->next = secinfo;
426 secinfo->next = secinfo;
427 sinfo->chain = secinfo;
429 secinfo->psecinfo = psecinfo;
430 secinfo->htab = sinfo->htab;
431 secinfo->first_str = NULL;
433 sec->rawsize = sec->size;
434 if (sec->flags & SEC_STRINGS)
435 memset (secinfo->contents + sec->size, 0, sec->entsize);
436 contents = secinfo->contents;
437 if (! bfd_get_full_section_contents (sec->owner, sec, &contents))
447 /* Record one section into the hash table. */
449 record_section (struct sec_merge_info *sinfo,
450 struct sec_merge_sec_info *secinfo)
452 asection *sec = secinfo->sec;
453 struct sec_merge_hash_entry *entry;
455 unsigned char *p, *end;
456 bfd_vma mask, eltalign;
457 unsigned int align, i;
459 align = sec->alignment_power;
460 end = secinfo->contents + sec->size;
462 mask = ((bfd_vma) 1 << align) - 1;
463 if (sec->flags & SEC_STRINGS)
465 for (p = secinfo->contents; p < end; )
467 eltalign = p - secinfo->contents;
468 eltalign = ((eltalign ^ (eltalign - 1)) + 1) >> 1;
469 if (!eltalign || eltalign > mask)
471 entry = sec_merge_add (sinfo->htab, (char *) p, (unsigned) eltalign,
476 if (sec->entsize == 1)
478 while (p < end && *p == 0)
480 if (!nul && !((p - secinfo->contents) & mask))
483 entry = sec_merge_add (sinfo->htab, "",
484 (unsigned) mask + 1, secinfo);
495 for (i = 0; i < sec->entsize; i++)
498 if (i != sec->entsize)
500 if (!nul && !((p - secinfo->contents) & mask))
503 entry = sec_merge_add (sinfo->htab, (char *) p,
504 (unsigned) mask + 1, secinfo);
515 for (p = secinfo->contents; p < end; p += sec->entsize)
517 entry = sec_merge_add (sinfo->htab, (char *) p, 1, secinfo);
526 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
527 *secinfo->psecinfo = NULL;
532 strrevcmp (const void *a, const void *b)
534 struct sec_merge_hash_entry *A = *(struct sec_merge_hash_entry **) a;
535 struct sec_merge_hash_entry *B = *(struct sec_merge_hash_entry **) b;
536 unsigned int lenA = A->len;
537 unsigned int lenB = B->len;
538 const unsigned char *s = (const unsigned char *) A->root.string + lenA - 1;
539 const unsigned char *t = (const unsigned char *) B->root.string + lenB - 1;
540 int l = lenA < lenB ? lenA : lenB;
545 return (int) *s - (int) *t;
553 /* Like strrevcmp, but for the case where all strings have the same
554 alignment > entsize. */
557 strrevcmp_align (const void *a, const void *b)
559 struct sec_merge_hash_entry *A = *(struct sec_merge_hash_entry **) a;
560 struct sec_merge_hash_entry *B = *(struct sec_merge_hash_entry **) b;
561 unsigned int lenA = A->len;
562 unsigned int lenB = B->len;
563 const unsigned char *s = (const unsigned char *) A->root.string + lenA - 1;
564 const unsigned char *t = (const unsigned char *) B->root.string + lenB - 1;
565 int l = lenA < lenB ? lenA : lenB;
566 int tail_align = (lenA & (A->alignment - 1)) - (lenB & (A->alignment - 1));
574 return (int) *s - (int) *t;
583 is_suffix (const struct sec_merge_hash_entry *A,
584 const struct sec_merge_hash_entry *B)
586 if (A->len <= B->len)
587 /* B cannot be a suffix of A unless A is equal to B, which is guaranteed
588 not to be equal by the hash table. */
591 return memcmp (A->root.string + (A->len - B->len),
592 B->root.string, B->len) == 0;
595 /* This is a helper function for _bfd_merge_sections. It attempts to
596 merge strings matching suffixes of longer strings. */
598 merge_strings (struct sec_merge_info *sinfo)
600 struct sec_merge_hash_entry **array, **a, *e;
601 struct sec_merge_sec_info *secinfo;
602 bfd_size_type size, amt;
603 unsigned int alignment = 0;
605 /* Now sort the strings */
606 amt = sinfo->htab->size * sizeof (struct sec_merge_hash_entry *);
607 array = (struct sec_merge_hash_entry **) bfd_malloc (amt);
611 for (e = sinfo->htab->first, a = array; e; e = e->next)
615 /* Adjust the length to not include the zero terminator. */
616 e->len -= sinfo->htab->entsize;
617 if (alignment != e->alignment)
620 alignment = e->alignment;
622 alignment = (unsigned) -1;
626 sinfo->htab->size = a - array;
627 if (sinfo->htab->size != 0)
629 qsort (array, (size_t) sinfo->htab->size,
630 sizeof (struct sec_merge_hash_entry *),
631 (alignment != (unsigned) -1 && alignment > sinfo->htab->entsize
632 ? strrevcmp_align : strrevcmp));
634 /* Loop over the sorted array and merge suffixes */
636 e->len += sinfo->htab->entsize;
639 struct sec_merge_hash_entry *cmp = *a;
641 cmp->len += sinfo->htab->entsize;
642 if (e->alignment >= cmp->alignment
643 && !((e->len - cmp->len) & (cmp->alignment - 1))
644 && is_suffix (e, cmp))
658 /* Now assign positions to the strings we want to keep. */
660 secinfo = sinfo->htab->first->secinfo;
661 for (e = sinfo->htab->first; e; e = e->next)
663 if (e->secinfo != secinfo)
665 secinfo->sec->size = size;
666 secinfo = e->secinfo;
670 if (e->secinfo->first_str == NULL)
672 e->secinfo->first_str = e;
675 size = (size + e->alignment - 1) & ~((bfd_vma) e->alignment - 1);
680 secinfo->sec->size = size;
681 if (secinfo->sec->alignment_power != 0)
683 bfd_size_type align = (bfd_size_type) 1 << secinfo->sec->alignment_power;
684 secinfo->sec->size = (secinfo->sec->size + align - 1) & -align;
687 /* And now adjust the rest, removing them from the chain (but not hashtable)
689 for (a = &sinfo->htab->first, e = *a; e; e = e->next)
697 e->secinfo = e->u.suffix->secinfo;
698 e->alignment = e->u.suffix->alignment;
699 e->u.index = e->u.suffix->u.index + (e->u.suffix->len - e->len);
704 /* This function is called once after all SEC_MERGE sections are registered
705 with _bfd_merge_section. */
708 _bfd_merge_sections (bfd *abfd,
709 struct bfd_link_info *info ATTRIBUTE_UNUSED,
711 void (*remove_hook) (bfd *, asection *))
713 struct sec_merge_info *sinfo;
715 for (sinfo = (struct sec_merge_info *) xsinfo; sinfo; sinfo = sinfo->next)
717 struct sec_merge_sec_info * secinfo;
722 /* Move sinfo->chain to head of the chain, terminate it. */
723 secinfo = sinfo->chain;
724 sinfo->chain = secinfo->next;
725 secinfo->next = NULL;
727 /* Record the sections into the hash table. */
728 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
729 if (secinfo->sec->flags & SEC_EXCLUDE)
731 *secinfo->psecinfo = NULL;
733 (*remove_hook) (abfd, secinfo->sec);
735 else if (! record_section (sinfo, secinfo))
741 if (sinfo->htab->first == NULL)
744 if (sinfo->htab->strings)
745 merge_strings (sinfo);
748 struct sec_merge_hash_entry *e;
749 bfd_size_type size = 0;
751 /* Things are much simpler for non-strings.
752 Just assign them slots in the section. */
754 for (e = sinfo->htab->first; e; e = e->next)
756 if (e->secinfo->first_str == NULL)
759 secinfo->sec->size = size;
760 e->secinfo->first_str = e;
763 size = (size + e->alignment - 1)
764 & ~((bfd_vma) e->alignment - 1);
767 secinfo = e->secinfo;
769 secinfo->sec->size = size;
772 /* Finally remove all input sections which have not made it into
773 the hash table at all. */
774 for (secinfo = sinfo->chain; secinfo; secinfo = secinfo->next)
775 if (secinfo->first_str == NULL)
776 secinfo->sec->flags |= SEC_EXCLUDE | SEC_KEEP;
782 /* Write out the merged section. */
785 _bfd_write_merged_section (bfd *output_bfd, asection *sec, void *psecinfo)
787 struct sec_merge_sec_info *secinfo;
790 secinfo = (struct sec_merge_sec_info *) psecinfo;
795 if (secinfo->first_str == NULL)
798 /* FIXME: octets_per_byte. */
799 pos = sec->output_section->filepos + sec->output_offset;
800 if (bfd_seek (output_bfd, pos, SEEK_SET) != 0)
803 if (! sec_merge_emit (output_bfd, secinfo->first_str))
809 /* Adjust an address in the SEC_MERGE section. Given OFFSET within
810 *PSEC, this returns the new offset in the adjusted SEC_MERGE
811 section and writes the new section back into *PSEC. */
814 _bfd_merged_section_offset (bfd *output_bfd ATTRIBUTE_UNUSED, asection **psec,
815 void *psecinfo, bfd_vma offset)
817 struct sec_merge_sec_info *secinfo;
818 struct sec_merge_hash_entry *entry;
820 asection *sec = *psec;
822 secinfo = (struct sec_merge_sec_info *) psecinfo;
827 if (offset >= sec->rawsize)
829 if (offset > sec->rawsize)
831 (*_bfd_error_handler)
832 (_("%s: access beyond end of merged section (%ld)"),
833 bfd_get_filename (sec->owner), (long) offset);
835 return secinfo->first_str ? sec->size : 0;
838 if (secinfo->htab->strings)
840 if (sec->entsize == 1)
842 p = secinfo->contents + offset - 1;
843 while (p >= secinfo->contents && *p)
849 p = secinfo->contents + (offset / sec->entsize) * sec->entsize;
851 while (p >= secinfo->contents)
855 for (i = 0; i < sec->entsize; ++i)
858 if (i == sec->entsize)
867 p = secinfo->contents + (offset / sec->entsize) * sec->entsize;
869 entry = sec_merge_hash_lookup (secinfo->htab, (char *) p, 0, FALSE);
872 if (! secinfo->htab->strings)
874 /* This should only happen if somebody points into the padding
875 after a NUL character but before next entity. */
878 if (! secinfo->htab->first)
880 entry = secinfo->htab->first;
881 p = (secinfo->contents + (offset / sec->entsize + 1) * sec->entsize
885 *psec = entry->secinfo->sec;
886 return entry->u.index + (secinfo->contents + offset - p);