URL
https://opencores.org/ocsvn/open8_urisc/open8_urisc/trunk
Subversion Repositories open8_urisc
[/] [open8_urisc/] [trunk/] [gnu/] [binutils/] [bfd/] [xcofflink.c] - Rev 47
Go to most recent revision | Compare with Previous | Blame | View Log
/* POWER/PowerPC XCOFF linker support. Copyright 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc. Written by Ian Lance Taylor <ian@cygnus.com>, Cygnus Support. This file is part of BFD, the Binary File Descriptor library. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA. */ #include "sysdep.h" #include "bfd.h" #include "bfdlink.h" #include "libbfd.h" #include "coff/internal.h" #include "coff/xcoff.h" #include "libcoff.h" #include "libxcoff.h" #include "libiberty.h" /* This file holds the XCOFF linker code. */ #undef STRING_SIZE_SIZE #define STRING_SIZE_SIZE 4 /* We reuse the SEC_ROM flag as a mark flag for garbage collection. This flag will only be used on input sections. */ #define SEC_MARK (SEC_ROM) /* The list of import files. */ struct xcoff_import_file { /* The next entry in the list. */ struct xcoff_import_file *next; /* The path. */ const char *path; /* The file name. */ const char *file; /* The member name. */ const char *member; }; /* Information we keep for each section in the output file during the final link phase. */ struct xcoff_link_section_info { /* The relocs to be output. */ struct internal_reloc *relocs; /* For each reloc against a global symbol whose index was not known when the reloc was handled, the global hash table entry. */ struct xcoff_link_hash_entry **rel_hashes; /* If there is a TOC relative reloc against a global symbol, and the index of the TOC symbol is not known when the reloc was handled, an entry is added to this linked list. This is not an array, like rel_hashes, because this case is quite uncommon. */ struct xcoff_toc_rel_hash { struct xcoff_toc_rel_hash *next; struct xcoff_link_hash_entry *h; struct internal_reloc *rel; } *toc_rel_hashes; }; /* Information that the XCOFF linker collects about an archive. */ struct xcoff_archive_info { /* The archive described by this entry. */ bfd *archive; /* The import path and import filename to use when referring to this archive in the .loader section. */ const char *imppath; const char *impfile; /* True if the archive contains a dynamic object. */ unsigned int contains_shared_object_p : 1; /* True if the previous field is valid. */ unsigned int know_contains_shared_object_p : 1; }; struct xcoff_link_hash_table { struct bfd_link_hash_table root; /* The .debug string hash table. We need to compute this while reading the input files, so that we know how large the .debug section will be before we assign section positions. */ struct bfd_strtab_hash *debug_strtab; /* The .debug section we will use for the final output. */ asection *debug_section; /* The .loader section we will use for the final output. */ asection *loader_section; /* A count of non TOC relative relocs which will need to be allocated in the .loader section. */ size_t ldrel_count; /* The .loader section header. */ struct internal_ldhdr ldhdr; /* The .gl section we use to hold global linkage code. */ asection *linkage_section; /* The .tc section we use to hold toc entries we build for global linkage code. */ asection *toc_section; /* The .ds section we use to hold function descriptors which we create for exported symbols. */ asection *descriptor_section; /* The list of import files. */ struct xcoff_import_file *imports; /* Required alignment of sections within the output file. */ unsigned long file_align; /* Whether the .text section must be read-only. */ bfd_boolean textro; /* Whether -brtl was specified. */ bfd_boolean rtld; /* Whether garbage collection was done. */ bfd_boolean gc; /* A linked list of symbols for which we have size information. */ struct xcoff_link_size_list { struct xcoff_link_size_list *next; struct xcoff_link_hash_entry *h; bfd_size_type size; } *size_list; /* Information about archives. */ htab_t archive_info; /* Magic sections: _text, _etext, _data, _edata, _end, end. */ asection *special_sections[XCOFF_NUMBER_OF_SPECIAL_SECTIONS]; }; /* Information that we pass around while doing the final link step. */ struct xcoff_final_link_info { /* General link information. */ struct bfd_link_info *info; /* Output BFD. */ bfd *output_bfd; /* Hash table for long symbol names. */ struct bfd_strtab_hash *strtab; /* Array of information kept for each output section, indexed by the target_index field. */ struct xcoff_link_section_info *section_info; /* Symbol index of last C_FILE symbol (-1 if none). */ long last_file_index; /* Contents of last C_FILE symbol. */ struct internal_syment last_file; /* Symbol index of TOC symbol. */ long toc_symindx; /* Start of .loader symbols. */ bfd_byte *ldsym; /* Next .loader reloc to swap out. */ bfd_byte *ldrel; /* File position of start of line numbers. */ file_ptr line_filepos; /* Buffer large enough to hold swapped symbols of any input file. */ struct internal_syment *internal_syms; /* Buffer large enough to hold output indices of symbols of any input file. */ long *sym_indices; /* Buffer large enough to hold output symbols for any input file. */ bfd_byte *outsyms; /* Buffer large enough to hold external line numbers for any input section. */ bfd_byte *linenos; /* Buffer large enough to hold any input section. */ bfd_byte *contents; /* Buffer large enough to hold external relocs of any input section. */ bfd_byte *external_relocs; }; static bfd_boolean xcoff_mark (struct bfd_link_info *, asection *); /* Routines to read XCOFF dynamic information. This don't really belong here, but we already have the ldsym manipulation routines here. */ /* Read the contents of a section. */ static bfd_boolean xcoff_get_section_contents (bfd *abfd, asection *sec) { if (coff_section_data (abfd, sec) == NULL) { bfd_size_type amt = sizeof (struct coff_section_tdata); sec->used_by_bfd = bfd_zalloc (abfd, amt); if (sec->used_by_bfd == NULL) return FALSE; } if (coff_section_data (abfd, sec)->contents == NULL) { bfd_byte *contents; if (! bfd_malloc_and_get_section (abfd, sec, &contents)) { if (contents != NULL) free (contents); return FALSE; } coff_section_data (abfd, sec)->contents = contents; } return TRUE; } /* Get the size required to hold the dynamic symbols. */ long _bfd_xcoff_get_dynamic_symtab_upper_bound (bfd *abfd) { asection *lsec; bfd_byte *contents; struct internal_ldhdr ldhdr; if ((abfd->flags & DYNAMIC) == 0) { bfd_set_error (bfd_error_invalid_operation); return -1; } lsec = bfd_get_section_by_name (abfd, ".loader"); if (lsec == NULL) { bfd_set_error (bfd_error_no_symbols); return -1; } if (! xcoff_get_section_contents (abfd, lsec)) return -1; contents = coff_section_data (abfd, lsec)->contents; bfd_xcoff_swap_ldhdr_in (abfd, (void *) contents, &ldhdr); return (ldhdr.l_nsyms + 1) * sizeof (asymbol *); } /* Get the dynamic symbols. */ long _bfd_xcoff_canonicalize_dynamic_symtab (bfd *abfd, asymbol **psyms) { asection *lsec; bfd_byte *contents; struct internal_ldhdr ldhdr; const char *strings; bfd_byte *elsym, *elsymend; coff_symbol_type *symbuf; if ((abfd->flags & DYNAMIC) == 0) { bfd_set_error (bfd_error_invalid_operation); return -1; } lsec = bfd_get_section_by_name (abfd, ".loader"); if (lsec == NULL) { bfd_set_error (bfd_error_no_symbols); return -1; } if (! xcoff_get_section_contents (abfd, lsec)) return -1; contents = coff_section_data (abfd, lsec)->contents; coff_section_data (abfd, lsec)->keep_contents = TRUE; bfd_xcoff_swap_ldhdr_in (abfd, contents, &ldhdr); strings = (char *) contents + ldhdr.l_stoff; symbuf = bfd_zalloc (abfd, ldhdr.l_nsyms * sizeof (* symbuf)); if (symbuf == NULL) return -1; elsym = contents + bfd_xcoff_loader_symbol_offset(abfd, &ldhdr); elsymend = elsym + ldhdr.l_nsyms * bfd_xcoff_ldsymsz(abfd); for (; elsym < elsymend; elsym += bfd_xcoff_ldsymsz(abfd), symbuf++, psyms++) { struct internal_ldsym ldsym; bfd_xcoff_swap_ldsym_in (abfd, elsym, &ldsym); symbuf->symbol.the_bfd = abfd; if (ldsym._l._l_l._l_zeroes == 0) symbuf->symbol.name = strings + ldsym._l._l_l._l_offset; else { char *c; c = bfd_alloc (abfd, (bfd_size_type) SYMNMLEN + 1); if (c == NULL) return -1; memcpy (c, ldsym._l._l_name, SYMNMLEN); c[SYMNMLEN] = '\0'; symbuf->symbol.name = c; } if (ldsym.l_smclas == XMC_XO) symbuf->symbol.section = bfd_abs_section_ptr; else symbuf->symbol.section = coff_section_from_bfd_index (abfd, ldsym.l_scnum); symbuf->symbol.value = ldsym.l_value - symbuf->symbol.section->vma; symbuf->symbol.flags = BSF_NO_FLAGS; if ((ldsym.l_smtype & L_EXPORT) != 0) { if ((ldsym.l_smtype & L_WEAK) != 0) symbuf->symbol.flags |= BSF_WEAK; else symbuf->symbol.flags |= BSF_GLOBAL; } /* FIXME: We have no way to record the other information stored with the loader symbol. */ *psyms = (asymbol *) symbuf; } *psyms = NULL; return ldhdr.l_nsyms; } /* Get the size required to hold the dynamic relocs. */ long _bfd_xcoff_get_dynamic_reloc_upper_bound (bfd *abfd) { asection *lsec; bfd_byte *contents; struct internal_ldhdr ldhdr; if ((abfd->flags & DYNAMIC) == 0) { bfd_set_error (bfd_error_invalid_operation); return -1; } lsec = bfd_get_section_by_name (abfd, ".loader"); if (lsec == NULL) { bfd_set_error (bfd_error_no_symbols); return -1; } if (! xcoff_get_section_contents (abfd, lsec)) return -1; contents = coff_section_data (abfd, lsec)->contents; bfd_xcoff_swap_ldhdr_in (abfd, (struct external_ldhdr *) contents, &ldhdr); return (ldhdr.l_nreloc + 1) * sizeof (arelent *); } /* Get the dynamic relocs. */ long _bfd_xcoff_canonicalize_dynamic_reloc (bfd *abfd, arelent **prelocs, asymbol **syms) { asection *lsec; bfd_byte *contents; struct internal_ldhdr ldhdr; arelent *relbuf; bfd_byte *elrel, *elrelend; if ((abfd->flags & DYNAMIC) == 0) { bfd_set_error (bfd_error_invalid_operation); return -1; } lsec = bfd_get_section_by_name (abfd, ".loader"); if (lsec == NULL) { bfd_set_error (bfd_error_no_symbols); return -1; } if (! xcoff_get_section_contents (abfd, lsec)) return -1; contents = coff_section_data (abfd, lsec)->contents; bfd_xcoff_swap_ldhdr_in (abfd, contents, &ldhdr); relbuf = bfd_alloc (abfd, ldhdr.l_nreloc * sizeof (arelent)); if (relbuf == NULL) return -1; elrel = contents + bfd_xcoff_loader_reloc_offset(abfd, &ldhdr); elrelend = elrel + ldhdr.l_nreloc * bfd_xcoff_ldrelsz(abfd); for (; elrel < elrelend; elrel += bfd_xcoff_ldrelsz(abfd), relbuf++, prelocs++) { struct internal_ldrel ldrel; bfd_xcoff_swap_ldrel_in (abfd, elrel, &ldrel); if (ldrel.l_symndx >= 3) relbuf->sym_ptr_ptr = syms + (ldrel.l_symndx - 3); else { const char *name; asection *sec; switch (ldrel.l_symndx) { case 0: name = ".text"; break; case 1: name = ".data"; break; case 2: name = ".bss"; break; default: abort (); break; } sec = bfd_get_section_by_name (abfd, name); if (sec == NULL) { bfd_set_error (bfd_error_bad_value); return -1; } relbuf->sym_ptr_ptr = sec->symbol_ptr_ptr; } relbuf->address = ldrel.l_vaddr; relbuf->addend = 0; /* Most dynamic relocs have the same type. FIXME: This is only correct if ldrel.l_rtype == 0. In other cases, we should use a different howto. */ relbuf->howto = bfd_xcoff_dynamic_reloc_howto(abfd); /* FIXME: We have no way to record the l_rsecnm field. */ *prelocs = relbuf; } *prelocs = NULL; return ldhdr.l_nreloc; } /* Hash functions for xcoff_link_hash_table's archive_info. */ static hashval_t xcoff_archive_info_hash (const void *data) { const struct xcoff_archive_info *info; info = (const struct xcoff_archive_info *) data; return htab_hash_pointer (info->archive); } static int xcoff_archive_info_eq (const void *data1, const void *data2) { const struct xcoff_archive_info *info1; const struct xcoff_archive_info *info2; info1 = (const struct xcoff_archive_info *) data1; info2 = (const struct xcoff_archive_info *) data2; return info1->archive == info2->archive; } /* Return information about archive ARCHIVE. Return NULL on error. */ static struct xcoff_archive_info * xcoff_get_archive_info (struct bfd_link_info *info, bfd *archive) { struct xcoff_link_hash_table *htab; struct xcoff_archive_info *entryp, entry; void **slot; htab = xcoff_hash_table (info); entry.archive = archive; slot = htab_find_slot (htab->archive_info, &entry, INSERT); if (!slot) return NULL; entryp = *slot; if (!entryp) { entryp = bfd_zalloc (archive, sizeof (entry)); if (!entryp) return NULL; entryp->archive = archive; *slot = entryp; } return entryp; } /* Routine to create an entry in an XCOFF link hash table. */ static struct bfd_hash_entry * xcoff_link_hash_newfunc (struct bfd_hash_entry *entry, struct bfd_hash_table *table, const char *string) { struct xcoff_link_hash_entry *ret = (struct xcoff_link_hash_entry *) entry; /* Allocate the structure if it has not already been allocated by a subclass. */ if (ret == NULL) ret = bfd_hash_allocate (table, sizeof (* ret)); if (ret == NULL) return NULL; /* Call the allocation method of the superclass. */ ret = ((struct xcoff_link_hash_entry *) _bfd_link_hash_newfunc ((struct bfd_hash_entry *) ret, table, string)); if (ret != NULL) { /* Set local fields. */ ret->indx = -1; ret->toc_section = NULL; ret->u.toc_indx = -1; ret->descriptor = NULL; ret->ldsym = NULL; ret->ldindx = -1; ret->flags = 0; ret->smclas = XMC_UA; } return (struct bfd_hash_entry *) ret; } /* Create a XCOFF link hash table. */ struct bfd_link_hash_table * _bfd_xcoff_bfd_link_hash_table_create (bfd *abfd) { struct xcoff_link_hash_table *ret; bfd_size_type amt = sizeof (* ret); ret = bfd_malloc (amt); if (ret == NULL) return NULL; if (!_bfd_link_hash_table_init (&ret->root, abfd, xcoff_link_hash_newfunc, sizeof (struct xcoff_link_hash_entry))) { free (ret); return NULL; } ret->debug_strtab = _bfd_xcoff_stringtab_init (); ret->debug_section = NULL; ret->loader_section = NULL; ret->ldrel_count = 0; memset (&ret->ldhdr, 0, sizeof (struct internal_ldhdr)); ret->linkage_section = NULL; ret->toc_section = NULL; ret->descriptor_section = NULL; ret->imports = NULL; ret->file_align = 0; ret->textro = FALSE; ret->gc = FALSE; ret->archive_info = htab_create (37, xcoff_archive_info_hash, xcoff_archive_info_eq, NULL); memset (ret->special_sections, 0, sizeof ret->special_sections); /* The linker will always generate a full a.out header. We need to record that fact now, before the sizeof_headers routine could be called. */ xcoff_data (abfd)->full_aouthdr = TRUE; return &ret->root; } /* Free a XCOFF link hash table. */ void _bfd_xcoff_bfd_link_hash_table_free (struct bfd_link_hash_table *hash) { struct xcoff_link_hash_table *ret = (struct xcoff_link_hash_table *) hash; _bfd_stringtab_free (ret->debug_strtab); bfd_hash_table_free (&ret->root.table); free (ret); } /* Read internal relocs for an XCOFF csect. This is a wrapper around _bfd_coff_read_internal_relocs which tries to take advantage of any relocs which may have been cached for the enclosing section. */ static struct internal_reloc * xcoff_read_internal_relocs (bfd *abfd, asection *sec, bfd_boolean cache, bfd_byte *external_relocs, bfd_boolean require_internal, struct internal_reloc *internal_relocs) { if (coff_section_data (abfd, sec) != NULL && coff_section_data (abfd, sec)->relocs == NULL && xcoff_section_data (abfd, sec) != NULL) { asection *enclosing; enclosing = xcoff_section_data (abfd, sec)->enclosing; if (enclosing != NULL && (coff_section_data (abfd, enclosing) == NULL || coff_section_data (abfd, enclosing)->relocs == NULL) && cache && enclosing->reloc_count > 0) { if (_bfd_coff_read_internal_relocs (abfd, enclosing, TRUE, external_relocs, FALSE, NULL) == NULL) return NULL; } if (enclosing != NULL && coff_section_data (abfd, enclosing) != NULL && coff_section_data (abfd, enclosing)->relocs != NULL) { size_t off; off = ((sec->rel_filepos - enclosing->rel_filepos) / bfd_coff_relsz (abfd)); if (! require_internal) return coff_section_data (abfd, enclosing)->relocs + off; memcpy (internal_relocs, coff_section_data (abfd, enclosing)->relocs + off, sec->reloc_count * sizeof (struct internal_reloc)); return internal_relocs; } } return _bfd_coff_read_internal_relocs (abfd, sec, cache, external_relocs, require_internal, internal_relocs); } /* Split FILENAME into an import path and an import filename, storing them in *IMPPATH and *IMPFILE respectively. */ bfd_boolean bfd_xcoff_split_import_path (bfd *abfd, const char *filename, const char **imppath, const char **impfile) { const char *base; size_t length; char *path; base = lbasename (filename); length = base - filename; if (length == 0) /* The filename has no directory component, so use an empty path. */ *imppath = ""; else if (length == 1) /* The filename is in the root directory. */ *imppath = "/"; else { /* Extract the (non-empty) directory part. Note that we don't need to strip duplicate directory separators from any part of the string; the native linker doesn't do that either. */ path = bfd_alloc (abfd, length); if (path == NULL) return FALSE; memcpy (path, filename, length - 1); path[length - 1] = 0; *imppath = path; } *impfile = base; return TRUE; } /* Set ARCHIVE's import path as though its filename had been given as FILENAME. */ bfd_boolean bfd_xcoff_set_archive_import_path (struct bfd_link_info *info, bfd *archive, const char *filename) { struct xcoff_archive_info *archive_info; archive_info = xcoff_get_archive_info (info, archive); return (archive_info != NULL && bfd_xcoff_split_import_path (archive, filename, &archive_info->imppath, &archive_info->impfile)); } /* H is an imported symbol. Set the import module's path, file and member to IMPATH, IMPFILE and IMPMEMBER respectively. All three are null if no specific import module is specified. */ static bfd_boolean xcoff_set_import_path (struct bfd_link_info *info, struct xcoff_link_hash_entry *h, const char *imppath, const char *impfile, const char *impmember) { unsigned int c; struct xcoff_import_file **pp; /* We overload the ldindx field to hold the l_ifile value for this symbol. */ BFD_ASSERT (h->ldsym == NULL); BFD_ASSERT ((h->flags & XCOFF_BUILT_LDSYM) == 0); if (imppath == NULL) h->ldindx = -1; else { /* We start c at 1 because the first entry in the import list is reserved for the library search path. */ for (pp = &xcoff_hash_table (info)->imports, c = 1; *pp != NULL; pp = &(*pp)->next, ++c) { if (filename_cmp ((*pp)->path, imppath) == 0 && filename_cmp ((*pp)->file, impfile) == 0 && filename_cmp ((*pp)->member, impmember) == 0) break; } if (*pp == NULL) { struct xcoff_import_file *n; bfd_size_type amt = sizeof (* n); n = bfd_alloc (info->output_bfd, amt); if (n == NULL) return FALSE; n->next = NULL; n->path = imppath; n->file = impfile; n->member = impmember; *pp = n; } h->ldindx = c; } return TRUE; } /* H is the bfd symbol associated with exported .loader symbol LDSYM. Return true if LDSYM defines H. */ static bfd_boolean xcoff_dynamic_definition_p (struct xcoff_link_hash_entry *h, struct internal_ldsym *ldsym) { /* If we didn't know about H before processing LDSYM, LDSYM definitely defines H. */ if (h->root.type == bfd_link_hash_new) return TRUE; /* If H is currently a weak dynamic symbol, and if LDSYM is a strong dynamic symbol, LDSYM trumps the current definition of H. */ if ((ldsym->l_smtype & L_WEAK) == 0 && (h->flags & XCOFF_DEF_DYNAMIC) != 0 && (h->flags & XCOFF_DEF_REGULAR) == 0 && (h->root.type == bfd_link_hash_defweak || h->root.type == bfd_link_hash_undefweak)) return TRUE; /* If H is currently undefined, LDSYM defines it. */ if ((h->flags & XCOFF_DEF_DYNAMIC) == 0 && (h->root.type == bfd_link_hash_undefined || h->root.type == bfd_link_hash_undefweak)) return TRUE; return FALSE; } /* This function is used to add symbols from a dynamic object to the global symbol table. */ static bfd_boolean xcoff_link_add_dynamic_symbols (bfd *abfd, struct bfd_link_info *info) { asection *lsec; bfd_byte *contents; struct internal_ldhdr ldhdr; const char *strings; bfd_byte *elsym, *elsymend; struct xcoff_import_file *n; unsigned int c; struct xcoff_import_file **pp; /* We can only handle a dynamic object if we are generating an XCOFF output file. */ if (info->output_bfd->xvec != abfd->xvec) { (*_bfd_error_handler) (_("%s: XCOFF shared object when not producing XCOFF output"), bfd_get_filename (abfd)); bfd_set_error (bfd_error_invalid_operation); return FALSE; } /* The symbols we use from a dynamic object are not the symbols in the normal symbol table, but, rather, the symbols in the export table. If there is a global symbol in a dynamic object which is not in the export table, the loader will not be able to find it, so we don't want to find it either. Also, on AIX 4.1.3, shr.o in libc.a has symbols in the export table which are not in the symbol table. */ /* Read in the .loader section. FIXME: We should really use the o_snloader field in the a.out header, rather than grabbing the section by name. */ lsec = bfd_get_section_by_name (abfd, ".loader"); if (lsec == NULL) { (*_bfd_error_handler) (_("%s: dynamic object with no .loader section"), bfd_get_filename (abfd)); bfd_set_error (bfd_error_no_symbols); return FALSE; } if (! xcoff_get_section_contents (abfd, lsec)) return FALSE; contents = coff_section_data (abfd, lsec)->contents; /* Remove the sections from this object, so that they do not get included in the link. */ bfd_section_list_clear (abfd); bfd_xcoff_swap_ldhdr_in (abfd, contents, &ldhdr); strings = (char *) contents + ldhdr.l_stoff; elsym = contents + bfd_xcoff_loader_symbol_offset(abfd, &ldhdr); elsymend = elsym + ldhdr.l_nsyms * bfd_xcoff_ldsymsz(abfd); for (; elsym < elsymend; elsym += bfd_xcoff_ldsymsz(abfd)) { struct internal_ldsym ldsym; char nambuf[SYMNMLEN + 1]; const char *name; struct xcoff_link_hash_entry *h; bfd_xcoff_swap_ldsym_in (abfd, elsym, &ldsym); /* We are only interested in exported symbols. */ if ((ldsym.l_smtype & L_EXPORT) == 0) continue; if (ldsym._l._l_l._l_zeroes == 0) name = strings + ldsym._l._l_l._l_offset; else { memcpy (nambuf, ldsym._l._l_name, SYMNMLEN); nambuf[SYMNMLEN] = '\0'; name = nambuf; } /* Normally we could not call xcoff_link_hash_lookup in an add symbols routine, since we might not be using an XCOFF hash table. However, we verified above that we are using an XCOFF hash table. */ h = xcoff_link_hash_lookup (xcoff_hash_table (info), name, TRUE, TRUE, TRUE); if (h == NULL) return FALSE; if (!xcoff_dynamic_definition_p (h, &ldsym)) continue; h->flags |= XCOFF_DEF_DYNAMIC; h->smclas = ldsym.l_smclas; if (h->smclas == XMC_XO) { /* This symbol has an absolute value. */ if ((ldsym.l_smtype & L_WEAK) != 0) h->root.type = bfd_link_hash_defweak; else h->root.type = bfd_link_hash_defined; h->root.u.def.section = bfd_abs_section_ptr; h->root.u.def.value = ldsym.l_value; } else { /* Otherwise, we don't bother to actually define the symbol, since we don't have a section to put it in anyhow. We assume instead that an undefined XCOFF_DEF_DYNAMIC symbol should be imported from the symbol's undef.abfd. */ if ((ldsym.l_smtype & L_WEAK) != 0) h->root.type = bfd_link_hash_undefweak; else h->root.type = bfd_link_hash_undefined; h->root.u.undef.abfd = abfd; } /* If this symbol defines a function descriptor, then it implicitly defines the function code as well. */ if (h->smclas == XMC_DS || (h->smclas == XMC_XO && name[0] != '.')) h->flags |= XCOFF_DESCRIPTOR; if ((h->flags & XCOFF_DESCRIPTOR) != 0) { struct xcoff_link_hash_entry *hds; hds = h->descriptor; if (hds == NULL) { char *dsnm; dsnm = bfd_malloc ((bfd_size_type) strlen (name) + 2); if (dsnm == NULL) return FALSE; dsnm[0] = '.'; strcpy (dsnm + 1, name); hds = xcoff_link_hash_lookup (xcoff_hash_table (info), dsnm, TRUE, TRUE, TRUE); free (dsnm); if (hds == NULL) return FALSE; hds->descriptor = h; h->descriptor = hds; } if (xcoff_dynamic_definition_p (hds, &ldsym)) { hds->root.type = h->root.type; hds->flags |= XCOFF_DEF_DYNAMIC; if (h->smclas == XMC_XO) { /* An absolute symbol appears to actually define code, not a function descriptor. This is how some math functions are implemented on AIX 4.1. */ hds->smclas = XMC_XO; hds->root.u.def.section = bfd_abs_section_ptr; hds->root.u.def.value = ldsym.l_value; } else { hds->smclas = XMC_PR; hds->root.u.undef.abfd = abfd; /* We do not want to add this to the undefined symbol list. */ } } } } if (contents != NULL && ! coff_section_data (abfd, lsec)->keep_contents) { free (coff_section_data (abfd, lsec)->contents); coff_section_data (abfd, lsec)->contents = NULL; } /* Record this file in the import files. */ n = bfd_alloc (abfd, (bfd_size_type) sizeof (struct xcoff_import_file)); if (n == NULL) return FALSE; n->next = NULL; if (abfd->my_archive == NULL) { if (!bfd_xcoff_split_import_path (abfd, abfd->filename, &n->path, &n->file)) return FALSE; n->member = ""; } else { struct xcoff_archive_info *archive_info; archive_info = xcoff_get_archive_info (info, abfd->my_archive); if (!archive_info->impfile) { if (!bfd_xcoff_split_import_path (archive_info->archive, archive_info->archive->filename, &archive_info->imppath, &archive_info->impfile)) return FALSE; } n->path = archive_info->imppath; n->file = archive_info->impfile; n->member = bfd_get_filename (abfd); } /* We start c at 1 because the first import file number is reserved for LIBPATH. */ for (pp = &xcoff_hash_table (info)->imports, c = 1; *pp != NULL; pp = &(*pp)->next, ++c) ; *pp = n; xcoff_data (abfd)->import_file_id = c; return TRUE; } /* xcoff_link_create_extra_sections Takes care of creating the .loader, .gl, .ds, .debug and sections. */ static bfd_boolean xcoff_link_create_extra_sections (bfd * abfd, struct bfd_link_info *info) { bfd_boolean return_value = FALSE; if (info->output_bfd->xvec == abfd->xvec) { /* We need to build a .loader section, so we do it here. This won't work if we're producing an XCOFF output file with no XCOFF input files. FIXME. */ if (!info->relocatable && xcoff_hash_table (info)->loader_section == NULL) { asection *lsec; flagword flags = SEC_HAS_CONTENTS | SEC_IN_MEMORY; lsec = bfd_make_section_anyway_with_flags (abfd, ".loader", flags); if (lsec == NULL) goto end_return; xcoff_hash_table (info)->loader_section = lsec; } /* Likewise for the linkage section. */ if (xcoff_hash_table (info)->linkage_section == NULL) { asection *lsec; flagword flags = (SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS | SEC_IN_MEMORY); lsec = bfd_make_section_anyway_with_flags (abfd, ".gl", flags); if (lsec == NULL) goto end_return; xcoff_hash_table (info)->linkage_section = lsec; lsec->alignment_power = 2; } /* Likewise for the TOC section. */ if (xcoff_hash_table (info)->toc_section == NULL) { asection *tsec; flagword flags = (SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS | SEC_IN_MEMORY); tsec = bfd_make_section_anyway_with_flags (abfd, ".tc", flags); if (tsec == NULL) goto end_return; xcoff_hash_table (info)->toc_section = tsec; tsec->alignment_power = 2; } /* Likewise for the descriptor section. */ if (xcoff_hash_table (info)->descriptor_section == NULL) { asection *dsec; flagword flags = (SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS | SEC_IN_MEMORY); dsec = bfd_make_section_anyway_with_flags (abfd, ".ds", flags); if (dsec == NULL) goto end_return; xcoff_hash_table (info)->descriptor_section = dsec; dsec->alignment_power = 2; } /* Likewise for the .debug section. */ if (xcoff_hash_table (info)->debug_section == NULL && info->strip != strip_all) { asection *dsec; flagword flags = SEC_HAS_CONTENTS | SEC_IN_MEMORY; dsec = bfd_make_section_anyway_with_flags (abfd, ".debug", flags); if (dsec == NULL) goto end_return; xcoff_hash_table (info)->debug_section = dsec; } } return_value = TRUE; end_return: return return_value; } /* Returns the index of reloc in RELOCS with the least address greater than or equal to ADDRESS. The relocs are sorted by address. */ static bfd_size_type xcoff_find_reloc (struct internal_reloc *relocs, bfd_size_type count, bfd_vma address) { bfd_size_type min, max, this; if (count < 2) { if (count == 1 && relocs[0].r_vaddr < address) return 1; else return 0; } min = 0; max = count; /* Do a binary search over (min,max]. */ while (min + 1 < max) { bfd_vma raddr; this = (max + min) / 2; raddr = relocs[this].r_vaddr; if (raddr > address) max = this; else if (raddr < address) min = this; else { min = this; break; } } if (relocs[min].r_vaddr < address) return min + 1; while (min > 0 && relocs[min - 1].r_vaddr == address) --min; return min; } /* Add all the symbols from an object file to the hash table. XCOFF is a weird format. A normal XCOFF .o files will have three COFF sections--.text, .data, and .bss--but each COFF section will contain many csects. These csects are described in the symbol table. From the linker's point of view, each csect must be considered a section in its own right. For example, a TOC entry is handled as a small XMC_TC csect. The linker must be able to merge different TOC entries together, which means that it must be able to extract the XMC_TC csects from the .data section of the input .o file. From the point of view of our linker, this is, of course, a hideous nightmare. We cope by actually creating sections for each csect, and discarding the original sections. We then have to handle the relocation entries carefully, since the only way to tell which csect they belong to is to examine the address. */ static bfd_boolean xcoff_link_add_symbols (bfd *abfd, struct bfd_link_info *info) { unsigned int n_tmask; unsigned int n_btshft; bfd_boolean default_copy; bfd_size_type symcount; struct xcoff_link_hash_entry **sym_hash; asection **csect_cache; unsigned int *lineno_counts; bfd_size_type linesz; asection *o; asection *last_real; bfd_boolean keep_syms; asection *csect; unsigned int csect_index; asection *first_csect; bfd_size_type symesz; bfd_byte *esym; bfd_byte *esym_end; struct reloc_info_struct { struct internal_reloc *relocs; asection **csects; bfd_byte *linenos; } *reloc_info = NULL; bfd_size_type amt; keep_syms = obj_coff_keep_syms (abfd); if ((abfd->flags & DYNAMIC) != 0 && ! info->static_link) { if (! xcoff_link_add_dynamic_symbols (abfd, info)) return FALSE; } /* Create the loader, toc, gl, ds and debug sections, if needed. */ if (! xcoff_link_create_extra_sections (abfd, info)) goto error_return; if ((abfd->flags & DYNAMIC) != 0 && ! info->static_link) return TRUE; n_tmask = coff_data (abfd)->local_n_tmask; n_btshft = coff_data (abfd)->local_n_btshft; /* Define macros so that ISFCN, et. al., macros work correctly. */ #define N_TMASK n_tmask #define N_BTSHFT n_btshft if (info->keep_memory) default_copy = FALSE; else default_copy = TRUE; symcount = obj_raw_syment_count (abfd); /* We keep a list of the linker hash table entries that correspond to each external symbol. */ amt = symcount * sizeof (struct xcoff_link_hash_entry *); sym_hash = bfd_zalloc (abfd, amt); if (sym_hash == NULL && symcount != 0) goto error_return; coff_data (abfd)->sym_hashes = (struct coff_link_hash_entry **) sym_hash; /* Because of the weird stuff we are doing with XCOFF csects, we can not easily determine which section a symbol is in, so we store the information in the tdata for the input file. */ amt = symcount * sizeof (asection *); csect_cache = bfd_zalloc (abfd, amt); if (csect_cache == NULL && symcount != 0) goto error_return; xcoff_data (abfd)->csects = csect_cache; /* We garbage-collect line-number information on a symbol-by-symbol basis, so we need to have quick access to the number of entries per symbol. */ amt = symcount * sizeof (unsigned int); lineno_counts = bfd_zalloc (abfd, amt); if (lineno_counts == NULL && symcount != 0) goto error_return; xcoff_data (abfd)->lineno_counts = lineno_counts; /* While splitting sections into csects, we need to assign the relocs correctly. The relocs and the csects must both be in order by VMA within a given section, so we handle this by scanning along the relocs as we process the csects. We index into reloc_info using the section target_index. */ amt = abfd->section_count + 1; amt *= sizeof (struct reloc_info_struct); reloc_info = bfd_zmalloc (amt); if (reloc_info == NULL) goto error_return; /* Read in the relocs and line numbers for each section. */ linesz = bfd_coff_linesz (abfd); last_real = NULL; for (o = abfd->sections; o != NULL; o = o->next) { last_real = o; if ((o->flags & SEC_RELOC) != 0) { reloc_info[o->target_index].relocs = xcoff_read_internal_relocs (abfd, o, TRUE, NULL, FALSE, NULL); amt = o->reloc_count; amt *= sizeof (asection *); reloc_info[o->target_index].csects = bfd_zmalloc (amt); if (reloc_info[o->target_index].csects == NULL) goto error_return; } if ((info->strip == strip_none || info->strip == strip_some) && o->lineno_count > 0) { bfd_byte *linenos; amt = linesz * o->lineno_count; linenos = bfd_malloc (amt); if (linenos == NULL) goto error_return; reloc_info[o->target_index].linenos = linenos; if (bfd_seek (abfd, o->line_filepos, SEEK_SET) != 0 || bfd_bread (linenos, amt, abfd) != amt) goto error_return; } } /* Don't let the linker relocation routines discard the symbols. */ obj_coff_keep_syms (abfd) = TRUE; csect = NULL; csect_index = 0; first_csect = NULL; symesz = bfd_coff_symesz (abfd); BFD_ASSERT (symesz == bfd_coff_auxesz (abfd)); esym = (bfd_byte *) obj_coff_external_syms (abfd); esym_end = esym + symcount * symesz; while (esym < esym_end) { struct internal_syment sym; union internal_auxent aux; const char *name; char buf[SYMNMLEN + 1]; int smtyp; asection *section; bfd_vma value; struct xcoff_link_hash_entry *set_toc; bfd_coff_swap_sym_in (abfd, (void *) esym, (void *) &sym); /* In this pass we are only interested in symbols with csect information. */ if (!CSECT_SYM_P (sym.n_sclass)) { /* Set csect_cache, Normally csect is a .pr, .rw etc. created in the loop If C_FILE or first time, handle special Advance esym, sym_hash, csect_hash ptrs. */ if (sym.n_sclass == C_FILE || sym.n_sclass == C_DWARF) csect = NULL; if (csect != NULL) *csect_cache = csect; else if (first_csect == NULL || sym.n_sclass == C_FILE || sym.n_sclass == C_DWARF) *csect_cache = coff_section_from_bfd_index (abfd, sym.n_scnum); else *csect_cache = NULL; esym += (sym.n_numaux + 1) * symesz; sym_hash += sym.n_numaux + 1; csect_cache += sym.n_numaux + 1; lineno_counts += sym.n_numaux + 1; continue; } name = _bfd_coff_internal_syment_name (abfd, &sym, buf); if (name == NULL) goto error_return; /* If this symbol has line number information attached to it, and we're not stripping it, count the number of entries and add them to the count for this csect. In the final link pass we are going to attach line number information by symbol, rather than by section, in order to more easily handle garbage collection. */ if ((info->strip == strip_none || info->strip == strip_some) && sym.n_numaux > 1 && csect != NULL && ISFCN (sym.n_type)) { union internal_auxent auxlin; bfd_coff_swap_aux_in (abfd, (void *) (esym + symesz), sym.n_type, sym.n_sclass, 0, sym.n_numaux, (void *) &auxlin); if (auxlin.x_sym.x_fcnary.x_fcn.x_lnnoptr != 0) { asection *enclosing; bfd_signed_vma linoff; enclosing = xcoff_section_data (abfd, csect)->enclosing; if (enclosing == NULL) { (*_bfd_error_handler) (_("%B: `%s' has line numbers but no enclosing section"), abfd, name); bfd_set_error (bfd_error_bad_value); goto error_return; } linoff = (auxlin.x_sym.x_fcnary.x_fcn.x_lnnoptr - enclosing->line_filepos); /* Explicit cast to bfd_signed_vma for compiler. */ if (linoff < (bfd_signed_vma) (enclosing->lineno_count * linesz)) { struct internal_lineno lin; bfd_byte *linpstart; linpstart = (reloc_info[enclosing->target_index].linenos + linoff); bfd_coff_swap_lineno_in (abfd, (void *) linpstart, (void *) &lin); if (lin.l_lnno == 0 && ((bfd_size_type) lin.l_addr.l_symndx == ((esym - (bfd_byte *) obj_coff_external_syms (abfd)) / symesz))) { bfd_byte *linpend, *linp; linpend = (reloc_info[enclosing->target_index].linenos + enclosing->lineno_count * linesz); for (linp = linpstart + linesz; linp < linpend; linp += linesz) { bfd_coff_swap_lineno_in (abfd, (void *) linp, (void *) &lin); if (lin.l_lnno == 0) break; } *lineno_counts = (linp - linpstart) / linesz; /* The setting of line_filepos will only be useful if all the line number entries for a csect are contiguous; this only matters for error reporting. */ if (csect->line_filepos == 0) csect->line_filepos = auxlin.x_sym.x_fcnary.x_fcn.x_lnnoptr; } } } } /* Pick up the csect auxiliary information. */ if (sym.n_numaux == 0) { (*_bfd_error_handler) (_("%B: class %d symbol `%s' has no aux entries"), abfd, sym.n_sclass, name); bfd_set_error (bfd_error_bad_value); goto error_return; } bfd_coff_swap_aux_in (abfd, (void *) (esym + symesz * sym.n_numaux), sym.n_type, sym.n_sclass, sym.n_numaux - 1, sym.n_numaux, (void *) &aux); smtyp = SMTYP_SMTYP (aux.x_csect.x_smtyp); section = NULL; value = 0; set_toc = NULL; switch (smtyp) { default: (*_bfd_error_handler) (_("%B: symbol `%s' has unrecognized csect type %d"), abfd, name, smtyp); bfd_set_error (bfd_error_bad_value); goto error_return; case XTY_ER: /* This is an external reference. */ if (sym.n_sclass == C_HIDEXT || sym.n_scnum != N_UNDEF || aux.x_csect.x_scnlen.l != 0) { (*_bfd_error_handler) (_("%B: bad XTY_ER symbol `%s': class %d scnum %d scnlen %d"), abfd, name, sym.n_sclass, sym.n_scnum, aux.x_csect.x_scnlen.l); bfd_set_error (bfd_error_bad_value); goto error_return; } /* An XMC_XO external reference is actually a reference to an absolute location. */ if (aux.x_csect.x_smclas != XMC_XO) section = bfd_und_section_ptr; else { section = bfd_abs_section_ptr; value = sym.n_value; } break; case XTY_SD: csect = NULL; csect_index = -(unsigned) 1; /* When we see a TOC anchor, we record the TOC value. */ if (aux.x_csect.x_smclas == XMC_TC0) { if (sym.n_sclass != C_HIDEXT || aux.x_csect.x_scnlen.l != 0) { (*_bfd_error_handler) (_("%B: XMC_TC0 symbol `%s' is class %d scnlen %d"), abfd, name, sym.n_sclass, aux.x_csect.x_scnlen.l); bfd_set_error (bfd_error_bad_value); goto error_return; } xcoff_data (abfd)->toc = sym.n_value; } /* We must merge TOC entries for the same symbol. We can merge two TOC entries if they are both C_HIDEXT, they both have the same name, they are both 4 or 8 bytes long, and they both have a relocation table entry for an external symbol with the same name. Unfortunately, this means that we must look through the relocations. Ick. Logic for 32 bit vs 64 bit. 32 bit has a csect length of 4 for TOC 64 bit has a csect length of 8 for TOC The conditions to get past the if-check are not that bad. They are what is used to create the TOC csects in the first place. */ if (aux.x_csect.x_smclas == XMC_TC && sym.n_sclass == C_HIDEXT && info->output_bfd->xvec == abfd->xvec && ((bfd_xcoff_is_xcoff32 (abfd) && aux.x_csect.x_scnlen.l == 4) || (bfd_xcoff_is_xcoff64 (abfd) && aux.x_csect.x_scnlen.l == 8))) { asection *enclosing; struct internal_reloc *relocs; bfd_size_type relindx; struct internal_reloc *rel; enclosing = coff_section_from_bfd_index (abfd, sym.n_scnum); if (enclosing == NULL) goto error_return; relocs = reloc_info[enclosing->target_index].relocs; amt = enclosing->reloc_count; relindx = xcoff_find_reloc (relocs, amt, sym.n_value); rel = relocs + relindx; /* 32 bit R_POS r_size is 31 64 bit R_POS r_size is 63 */ if (relindx < enclosing->reloc_count && rel->r_vaddr == (bfd_vma) sym.n_value && rel->r_type == R_POS && ((bfd_xcoff_is_xcoff32 (abfd) && rel->r_size == 31) || (bfd_xcoff_is_xcoff64 (abfd) && rel->r_size == 63))) { bfd_byte *erelsym; struct internal_syment relsym; erelsym = ((bfd_byte *) obj_coff_external_syms (abfd) + rel->r_symndx * symesz); bfd_coff_swap_sym_in (abfd, (void *) erelsym, (void *) &relsym); if (EXTERN_SYM_P (relsym.n_sclass)) { const char *relname; char relbuf[SYMNMLEN + 1]; bfd_boolean copy; struct xcoff_link_hash_entry *h; /* At this point we know that the TOC entry is for an externally visible symbol. */ relname = _bfd_coff_internal_syment_name (abfd, &relsym, relbuf); if (relname == NULL) goto error_return; /* We only merge TOC entries if the TC name is the same as the symbol name. This handles the normal case, but not common cases like SYM.P4 which gcc generates to store SYM + 4 in the TOC. FIXME. */ if (strcmp (name, relname) == 0) { copy = (! info->keep_memory || relsym._n._n_n._n_zeroes != 0 || relsym._n._n_n._n_offset == 0); h = xcoff_link_hash_lookup (xcoff_hash_table (info), relname, TRUE, copy, FALSE); if (h == NULL) goto error_return; /* At this point h->root.type could be bfd_link_hash_new. That should be OK, since we know for sure that we will come across this symbol as we step through the file. */ /* We store h in *sym_hash for the convenience of the relocate_section function. */ *sym_hash = h; if (h->toc_section != NULL) { asection **rel_csects; /* We already have a TOC entry for this symbol, so we can just ignore this one. */ rel_csects = reloc_info[enclosing->target_index].csects; rel_csects[relindx] = bfd_und_section_ptr; break; } /* We are about to create a TOC entry for this symbol. */ set_toc = h; } } } } { asection *enclosing; /* We need to create a new section. We get the name from the csect storage mapping class, so that the linker can accumulate similar csects together. */ csect = bfd_xcoff_create_csect_from_smclas(abfd, &aux, name); if (NULL == csect) goto error_return; /* The enclosing section is the main section : .data, .text or .bss that the csect is coming from. */ enclosing = coff_section_from_bfd_index (abfd, sym.n_scnum); if (enclosing == NULL) goto error_return; if (! bfd_is_abs_section (enclosing) && ((bfd_vma) sym.n_value < enclosing->vma || ((bfd_vma) sym.n_value + aux.x_csect.x_scnlen.l > enclosing->vma + enclosing->size))) { (*_bfd_error_handler) (_("%B: csect `%s' not in enclosing section"), abfd, name); bfd_set_error (bfd_error_bad_value); goto error_return; } csect->vma = sym.n_value; csect->filepos = (enclosing->filepos + sym.n_value - enclosing->vma); csect->size = aux.x_csect.x_scnlen.l; csect->flags |= SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS; csect->alignment_power = SMTYP_ALIGN (aux.x_csect.x_smtyp); /* Record the enclosing section in the tdata for this new section. */ amt = sizeof (struct coff_section_tdata); csect->used_by_bfd = bfd_zalloc (abfd, amt); if (csect->used_by_bfd == NULL) goto error_return; amt = sizeof (struct xcoff_section_tdata); coff_section_data (abfd, csect)->tdata = bfd_zalloc (abfd, amt); if (coff_section_data (abfd, csect)->tdata == NULL) goto error_return; xcoff_section_data (abfd, csect)->enclosing = enclosing; xcoff_section_data (abfd, csect)->lineno_count = enclosing->lineno_count; if (enclosing->owner == abfd) { struct internal_reloc *relocs; bfd_size_type relindx; struct internal_reloc *rel; asection **rel_csect; relocs = reloc_info[enclosing->target_index].relocs; amt = enclosing->reloc_count; relindx = xcoff_find_reloc (relocs, amt, csect->vma); rel = relocs + relindx; rel_csect = (reloc_info[enclosing->target_index].csects + relindx); csect->rel_filepos = (enclosing->rel_filepos + relindx * bfd_coff_relsz (abfd)); while (relindx < enclosing->reloc_count && *rel_csect == NULL && rel->r_vaddr < csect->vma + csect->size) { *rel_csect = csect; csect->flags |= SEC_RELOC; ++csect->reloc_count; ++relindx; ++rel; ++rel_csect; } } /* There are a number of other fields and section flags which we do not bother to set. */ csect_index = ((esym - (bfd_byte *) obj_coff_external_syms (abfd)) / symesz); xcoff_section_data (abfd, csect)->first_symndx = csect_index; if (first_csect == NULL) first_csect = csect; /* If this symbol is external, we treat it as starting at the beginning of the newly created section. */ if (EXTERN_SYM_P (sym.n_sclass)) { section = csect; value = 0; } /* If this is a TOC section for a symbol, record it. */ if (set_toc != NULL) set_toc->toc_section = csect; } break; case XTY_LD: /* This is a label definition. The x_scnlen field is the symbol index of the csect. Usually the XTY_LD symbol will follow its appropriate XTY_SD symbol. The .set pseudo op can cause the XTY_LD to not follow the XTY_SD symbol. */ { bfd_boolean bad; bad = FALSE; if (aux.x_csect.x_scnlen.l < 0 || (aux.x_csect.x_scnlen.l >= esym - (bfd_byte *) obj_coff_external_syms (abfd))) bad = TRUE; if (! bad) { section = xcoff_data (abfd)->csects[aux.x_csect.x_scnlen.l]; if (section == NULL || (section->flags & SEC_HAS_CONTENTS) == 0) bad = TRUE; } if (bad) { (*_bfd_error_handler) (_("%B: misplaced XTY_LD `%s'"), abfd, name); bfd_set_error (bfd_error_bad_value); goto error_return; } csect = section; value = sym.n_value - csect->vma; } break; case XTY_CM: /* This is an unitialized csect. We could base the name on the storage mapping class, but we don't bother except for an XMC_TD symbol. If this csect is externally visible, it is a common symbol. We put XMC_TD symbols in sections named .tocbss, and rely on the linker script to put that in the TOC area. */ if (aux.x_csect.x_smclas == XMC_TD) { /* The linker script puts the .td section in the data section after the .tc section. */ csect = bfd_make_section_anyway_with_flags (abfd, ".td", SEC_ALLOC); } else csect = bfd_make_section_anyway_with_flags (abfd, ".bss", SEC_ALLOC); if (csect == NULL) goto error_return; csect->vma = sym.n_value; csect->size = aux.x_csect.x_scnlen.l; csect->alignment_power = SMTYP_ALIGN (aux.x_csect.x_smtyp); /* There are a number of other fields and section flags which we do not bother to set. */ csect_index = ((esym - (bfd_byte *) obj_coff_external_syms (abfd)) / symesz); amt = sizeof (struct coff_section_tdata); csect->used_by_bfd = bfd_zalloc (abfd, amt); if (csect->used_by_bfd == NULL) goto error_return; amt = sizeof (struct xcoff_section_tdata); coff_section_data (abfd, csect)->tdata = bfd_zalloc (abfd, amt); if (coff_section_data (abfd, csect)->tdata == NULL) goto error_return; xcoff_section_data (abfd, csect)->first_symndx = csect_index; if (first_csect == NULL) first_csect = csect; if (EXTERN_SYM_P (sym.n_sclass)) { csect->flags |= SEC_IS_COMMON; csect->size = 0; section = csect; value = aux.x_csect.x_scnlen.l; } break; } /* Check for magic symbol names. */ if ((smtyp == XTY_SD || smtyp == XTY_CM) && aux.x_csect.x_smclas != XMC_TC && aux.x_csect.x_smclas != XMC_TD) { int i = -1; if (name[0] == '_') { if (strcmp (name, "_text") == 0) i = XCOFF_SPECIAL_SECTION_TEXT; else if (strcmp (name, "_etext") == 0) i = XCOFF_SPECIAL_SECTION_ETEXT; else if (strcmp (name, "_data") == 0) i = XCOFF_SPECIAL_SECTION_DATA; else if (strcmp (name, "_edata") == 0) i = XCOFF_SPECIAL_SECTION_EDATA; else if (strcmp (name, "_end") == 0) i = XCOFF_SPECIAL_SECTION_END; } else if (name[0] == 'e' && strcmp (name, "end") == 0) i = XCOFF_SPECIAL_SECTION_END2; if (i != -1) xcoff_hash_table (info)->special_sections[i] = csect; } /* Now we have enough information to add the symbol to the linker hash table. */ if (EXTERN_SYM_P (sym.n_sclass)) { bfd_boolean copy; flagword flags; BFD_ASSERT (section != NULL); /* We must copy the name into memory if we got it from the syment itself, rather than the string table. */ copy = default_copy; if (sym._n._n_n._n_zeroes != 0 || sym._n._n_n._n_offset == 0) copy = TRUE; /* Ignore global linkage code when linking statically. */ if (info->static_link && (smtyp == XTY_SD || smtyp == XTY_LD) && aux.x_csect.x_smclas == XMC_GL) { section = bfd_und_section_ptr; value = 0; } /* The AIX linker appears to only detect multiple symbol definitions when there is a reference to the symbol. If a symbol is defined multiple times, and the only references are from the same object file, the AIX linker appears to permit it. It does not merge the different definitions, but handles them independently. On the other hand, if there is a reference, the linker reports an error. This matters because the AIX <net/net_globals.h> header file actually defines an initialized array, so we have to actually permit that to work. Just to make matters even more confusing, the AIX linker appears to permit multiple symbol definitions whenever the second definition is in an archive rather than an object file. This may be a consequence of the manner in which it handles archives: I think it may load the entire archive in as separate csects, and then let garbage collection discard symbols. We also have to handle the case of statically linking a shared object, which will cause symbol redefinitions, although this is an easier case to detect. */ else if (info->output_bfd->xvec == abfd->xvec) { if (! bfd_is_und_section (section)) *sym_hash = xcoff_link_hash_lookup (xcoff_hash_table (info), name, TRUE, copy, FALSE); else /* Make a copy of the symbol name to prevent problems with merging symbols. */ *sym_hash = ((struct xcoff_link_hash_entry *) bfd_wrapped_link_hash_lookup (abfd, info, name, TRUE, TRUE, FALSE)); if (*sym_hash == NULL) goto error_return; if (((*sym_hash)->root.type == bfd_link_hash_defined || (*sym_hash)->root.type == bfd_link_hash_defweak) && ! bfd_is_und_section (section) && ! bfd_is_com_section (section)) { /* This is a second definition of a defined symbol. */ if (((*sym_hash)->flags & XCOFF_DEF_REGULAR) == 0 && ((*sym_hash)->flags & XCOFF_DEF_DYNAMIC) != 0) { /* The existing symbol is from a shared library. Replace it. */ (*sym_hash)->root.type = bfd_link_hash_undefined; (*sym_hash)->root.u.undef.abfd = (*sym_hash)->root.u.def.section->owner; } else if (abfd->my_archive != NULL) { /* This is a redefinition in an object contained in an archive. Just ignore it. See the comment above. */ section = bfd_und_section_ptr; value = 0; } else if (sym.n_sclass == C_AIX_WEAKEXT || (*sym_hash)->root.type == bfd_link_hash_defweak) { /* At least one of the definitions is weak. Allow the normal rules to take effect. */ } else if ((*sym_hash)->root.u.undef.next != NULL || info->hash->undefs_tail == &(*sym_hash)->root) { /* This symbol has been referenced. In this case, we just continue and permit the multiple definition error. See the comment above about the behaviour of the AIX linker. */ } else if ((*sym_hash)->smclas == aux.x_csect.x_smclas) { /* The symbols are both csects of the same class. There is at least a chance that this is a semi-legitimate redefinition. */ section = bfd_und_section_ptr; value = 0; (*sym_hash)->flags |= XCOFF_MULTIPLY_DEFINED; } } else if (((*sym_hash)->flags & XCOFF_MULTIPLY_DEFINED) != 0 && (*sym_hash)->root.type == bfd_link_hash_defined && (bfd_is_und_section (section) || bfd_is_com_section (section))) { /* This is a reference to a multiply defined symbol. Report the error now. See the comment above about the behaviour of the AIX linker. We could also do this with warning symbols, but I'm not sure the XCOFF linker is wholly prepared to handle them, and that would only be a warning, not an error. */ if (! ((*info->callbacks->multiple_definition) (info, &(*sym_hash)->root, NULL, NULL, (bfd_vma) 0))) goto error_return; /* Try not to give this error too many times. */ (*sym_hash)->flags &= ~XCOFF_MULTIPLY_DEFINED; } } /* _bfd_generic_link_add_one_symbol may call the linker to generate an error message, and the linker may try to read the symbol table to give a good error. Right now, the line numbers are in an inconsistent state, since they are counted both in the real sections and in the new csects. We need to leave the count in the real sections so that the linker can report the line number of the error correctly, so temporarily clobber the link to the csects so that the linker will not try to read the line numbers a second time from the csects. */ BFD_ASSERT (last_real->next == first_csect); last_real->next = NULL; flags = (sym.n_sclass == C_EXT ? BSF_GLOBAL : BSF_WEAK); if (! (_bfd_generic_link_add_one_symbol (info, abfd, name, flags, section, value, NULL, copy, TRUE, (struct bfd_link_hash_entry **) sym_hash))) goto error_return; last_real->next = first_csect; if (smtyp == XTY_CM) { if ((*sym_hash)->root.type != bfd_link_hash_common || (*sym_hash)->root.u.c.p->section != csect) /* We don't need the common csect we just created. */ csect->size = 0; else (*sym_hash)->root.u.c.p->alignment_power = csect->alignment_power; } if (info->output_bfd->xvec == abfd->xvec) { int flag; if (smtyp == XTY_ER || smtyp == XTY_CM || section == bfd_und_section_ptr) flag = XCOFF_REF_REGULAR; else flag = XCOFF_DEF_REGULAR; (*sym_hash)->flags |= flag; if ((*sym_hash)->smclas == XMC_UA || flag == XCOFF_DEF_REGULAR) (*sym_hash)->smclas = aux.x_csect.x_smclas; } } if (smtyp == XTY_ER) *csect_cache = section; else { *csect_cache = csect; if (csect != NULL) xcoff_section_data (abfd, csect)->last_symndx = (esym - (bfd_byte *) obj_coff_external_syms (abfd)) / symesz; } esym += (sym.n_numaux + 1) * symesz; sym_hash += sym.n_numaux + 1; csect_cache += sym.n_numaux + 1; lineno_counts += sym.n_numaux + 1; } BFD_ASSERT (last_real == NULL || last_real->next == first_csect); /* Make sure that we have seen all the relocs. */ for (o = abfd->sections; o != first_csect; o = o->next) { /* Debugging sections have no csects. */ if (bfd_get_section_flags (abfd, o) & SEC_DEBUGGING) continue; /* Reset the section size and the line number count, since the data is now attached to the csects. Don't reset the size of the .debug section, since we need to read it below in bfd_xcoff_size_dynamic_sections. */ if (strcmp (bfd_get_section_name (abfd, o), ".debug") != 0) o->size = 0; o->lineno_count = 0; if ((o->flags & SEC_RELOC) != 0) { bfd_size_type i; struct internal_reloc *rel; asection **rel_csect; rel = reloc_info[o->target_index].relocs; rel_csect = reloc_info[o->target_index].csects; for (i = 0; i < o->reloc_count; i++, rel++, rel_csect++) { if (*rel_csect == NULL) { (*_bfd_error_handler) (_("%B: reloc %s:%d not in csect"), abfd, o->name, i); bfd_set_error (bfd_error_bad_value); goto error_return; } /* We identify all function symbols that are the target of a relocation, so that we can create glue code for functions imported from dynamic objects. */ if (info->output_bfd->xvec == abfd->xvec && *rel_csect != bfd_und_section_ptr && obj_xcoff_sym_hashes (abfd)[rel->r_symndx] != NULL) { struct xcoff_link_hash_entry *h; h = obj_xcoff_sym_hashes (abfd)[rel->r_symndx]; /* If the symbol name starts with a period, it is the code of a function. If the symbol is currently undefined, then add an undefined symbol for the function descriptor. This should do no harm, because any regular object that defines the function should also define the function descriptor. It helps, because it means that we will identify the function descriptor with a dynamic object if a dynamic object defines it. */ if (h->root.root.string[0] == '.' && h->descriptor == NULL) { struct xcoff_link_hash_entry *hds; struct bfd_link_hash_entry *bh; hds = xcoff_link_hash_lookup (xcoff_hash_table (info), h->root.root.string + 1, TRUE, FALSE, TRUE); if (hds == NULL) goto error_return; if (hds->root.type == bfd_link_hash_new) { bh = &hds->root; if (! (_bfd_generic_link_add_one_symbol (info, abfd, hds->root.root.string, (flagword) 0, bfd_und_section_ptr, (bfd_vma) 0, NULL, FALSE, TRUE, &bh))) goto error_return; hds = (struct xcoff_link_hash_entry *) bh; } hds->flags |= XCOFF_DESCRIPTOR; BFD_ASSERT ((h->flags & XCOFF_DESCRIPTOR) == 0); hds->descriptor = h; h->descriptor = hds; } if (h->root.root.string[0] == '.') h->flags |= XCOFF_CALLED; } } free (reloc_info[o->target_index].csects); reloc_info[o->target_index].csects = NULL; /* Reset SEC_RELOC and the reloc_count, since the reloc information is now attached to the csects. */ o->flags &=~ SEC_RELOC; o->reloc_count = 0; /* If we are not keeping memory, free the reloc information. */ if (! info->keep_memory && coff_section_data (abfd, o) != NULL && coff_section_data (abfd, o)->relocs != NULL && ! coff_section_data (abfd, o)->keep_relocs) { free (coff_section_data (abfd, o)->relocs); coff_section_data (abfd, o)->relocs = NULL; } } /* Free up the line numbers. FIXME: We could cache these somewhere for the final link, to avoid reading them again. */ if (reloc_info[o->target_index].linenos != NULL) { free (reloc_info[o->target_index].linenos); reloc_info[o->target_index].linenos = NULL; } } free (reloc_info); obj_coff_keep_syms (abfd) = keep_syms; return TRUE; error_return: if (reloc_info != NULL) { for (o = abfd->sections; o != NULL; o = o->next) { if (reloc_info[o->target_index].csects != NULL) free (reloc_info[o->target_index].csects); if (reloc_info[o->target_index].linenos != NULL) free (reloc_info[o->target_index].linenos); } free (reloc_info); } obj_coff_keep_syms (abfd) = keep_syms; return FALSE; } #undef N_TMASK #undef N_BTSHFT /* Add symbols from an XCOFF object file. */ static bfd_boolean xcoff_link_add_object_symbols (bfd *abfd, struct bfd_link_info *info) { if (! _bfd_coff_get_external_symbols (abfd)) return FALSE; if (! xcoff_link_add_symbols (abfd, info)) return FALSE; if (! info->keep_memory) { if (! _bfd_coff_free_symbols (abfd)) return FALSE; } return TRUE; } /* Look through the loader symbols to see if this dynamic object should be included in the link. The native linker uses the loader symbols, not the normal symbol table, so we do too. */ static bfd_boolean xcoff_link_check_dynamic_ar_symbols (bfd *abfd, struct bfd_link_info *info, bfd_boolean *pneeded, bfd **subsbfd) { asection *lsec; bfd_byte *contents; struct internal_ldhdr ldhdr; const char *strings; bfd_byte *elsym, *elsymend; *pneeded = FALSE; lsec = bfd_get_section_by_name (abfd, ".loader"); if (lsec == NULL) /* There are no symbols, so don't try to include it. */ return TRUE; if (! xcoff_get_section_contents (abfd, lsec)) return FALSE; contents = coff_section_data (abfd, lsec)->contents; bfd_xcoff_swap_ldhdr_in (abfd, contents, &ldhdr); strings = (char *) contents + ldhdr.l_stoff; elsym = contents + bfd_xcoff_loader_symbol_offset (abfd, &ldhdr); elsymend = elsym + ldhdr.l_nsyms * bfd_xcoff_ldsymsz (abfd); for (; elsym < elsymend; elsym += bfd_xcoff_ldsymsz (abfd)) { struct internal_ldsym ldsym; char nambuf[SYMNMLEN + 1]; const char *name; struct bfd_link_hash_entry *h; bfd_xcoff_swap_ldsym_in (abfd, elsym, &ldsym); /* We are only interested in exported symbols. */ if ((ldsym.l_smtype & L_EXPORT) == 0) continue; if (ldsym._l._l_l._l_zeroes == 0) name = strings + ldsym._l._l_l._l_offset; else { memcpy (nambuf, ldsym._l._l_name, SYMNMLEN); nambuf[SYMNMLEN] = '\0'; name = nambuf; } h = bfd_link_hash_lookup (info->hash, name, FALSE, FALSE, TRUE); /* We are only interested in symbols that are currently undefined. At this point we know that we are using an XCOFF hash table. */ if (h != NULL && h->type == bfd_link_hash_undefined && (((struct xcoff_link_hash_entry *) h)->flags & XCOFF_DEF_DYNAMIC) == 0) { if (!(*info->callbacks ->add_archive_element) (info, abfd, name, subsbfd)) return FALSE; *pneeded = TRUE; return TRUE; } } /* We do not need this shared object. */ if (contents != NULL && ! coff_section_data (abfd, lsec)->keep_contents) { free (coff_section_data (abfd, lsec)->contents); coff_section_data (abfd, lsec)->contents = NULL; } return TRUE; } /* Look through the symbols to see if this object file should be included in the link. */ static bfd_boolean xcoff_link_check_ar_symbols (bfd *abfd, struct bfd_link_info *info, bfd_boolean *pneeded, bfd **subsbfd) { bfd_size_type symesz; bfd_byte *esym; bfd_byte *esym_end; *pneeded = FALSE; if ((abfd->flags & DYNAMIC) != 0 && ! info->static_link && info->output_bfd->xvec == abfd->xvec) return xcoff_link_check_dynamic_ar_symbols (abfd, info, pneeded, subsbfd); symesz = bfd_coff_symesz (abfd); esym = (bfd_byte *) obj_coff_external_syms (abfd); esym_end = esym + obj_raw_syment_count (abfd) * symesz; while (esym < esym_end) { struct internal_syment sym; bfd_coff_swap_sym_in (abfd, (void *) esym, (void *) &sym); if (EXTERN_SYM_P (sym.n_sclass) && sym.n_scnum != N_UNDEF) { const char *name; char buf[SYMNMLEN + 1]; struct bfd_link_hash_entry *h; /* This symbol is externally visible, and is defined by this object file. */ name = _bfd_coff_internal_syment_name (abfd, &sym, buf); if (name == NULL) return FALSE; h = bfd_link_hash_lookup (info->hash, name, FALSE, FALSE, TRUE); /* We are only interested in symbols that are currently undefined. If a symbol is currently known to be common, XCOFF linkers do not bring in an object file which defines it. We also don't bring in symbols to satisfy undefined references in shared objects. */ if (h != NULL && h->type == bfd_link_hash_undefined && (info->output_bfd->xvec != abfd->xvec || (((struct xcoff_link_hash_entry *) h)->flags & XCOFF_DEF_DYNAMIC) == 0)) { if (!(*info->callbacks ->add_archive_element) (info, abfd, name, subsbfd)) return FALSE; *pneeded = TRUE; return TRUE; } } esym += (sym.n_numaux + 1) * symesz; } /* We do not need this object file. */ return TRUE; } /* Check a single archive element to see if we need to include it in the link. *PNEEDED is set according to whether this element is needed in the link or not. This is called via _bfd_generic_link_add_archive_symbols. */ static bfd_boolean xcoff_link_check_archive_element (bfd *abfd, struct bfd_link_info *info, bfd_boolean *pneeded) { bfd_boolean keep_syms_p; bfd *oldbfd; keep_syms_p = (obj_coff_external_syms (abfd) != NULL); if (!_bfd_coff_get_external_symbols (abfd)) return FALSE; oldbfd = abfd; if (!xcoff_link_check_ar_symbols (abfd, info, pneeded, &abfd)) return FALSE; if (*pneeded) { /* Potentially, the add_archive_element hook may have set a substitute BFD for us. */ if (abfd != oldbfd) { if (!keep_syms_p && !_bfd_coff_free_symbols (oldbfd)) return FALSE; keep_syms_p = (obj_coff_external_syms (abfd) != NULL); if (!_bfd_coff_get_external_symbols (abfd)) return FALSE; } if (!xcoff_link_add_symbols (abfd, info)) return FALSE; if (info->keep_memory) keep_syms_p = TRUE; } if (!keep_syms_p) { if (!_bfd_coff_free_symbols (abfd)) return FALSE; } return TRUE; } /* Given an XCOFF BFD, add symbols to the global hash table as appropriate. */ bfd_boolean _bfd_xcoff_bfd_link_add_symbols (bfd *abfd, struct bfd_link_info *info) { switch (bfd_get_format (abfd)) { case bfd_object: return xcoff_link_add_object_symbols (abfd, info); case bfd_archive: /* If the archive has a map, do the usual search. We then need to check the archive for dynamic objects, because they may not appear in the archive map even though they should, perhaps, be included. If the archive has no map, we just consider each object file in turn, since that apparently is what the AIX native linker does. */ if (bfd_has_map (abfd)) { if (! (_bfd_generic_link_add_archive_symbols (abfd, info, xcoff_link_check_archive_element))) return FALSE; } { bfd *member; member = bfd_openr_next_archived_file (abfd, NULL); while (member != NULL) { if (bfd_check_format (member, bfd_object) && (info->output_bfd->xvec == member->xvec) && (! bfd_has_map (abfd) || (member->flags & DYNAMIC) != 0)) { bfd_boolean needed; if (! xcoff_link_check_archive_element (member, info, &needed)) return FALSE; if (needed) member->archive_pass = -1; } member = bfd_openr_next_archived_file (abfd, member); } } return TRUE; default: bfd_set_error (bfd_error_wrong_format); return FALSE; } } bfd_boolean _bfd_xcoff_define_common_symbol (bfd *output_bfd ATTRIBUTE_UNUSED, struct bfd_link_info *info ATTRIBUTE_UNUSED, struct bfd_link_hash_entry *harg) { struct xcoff_link_hash_entry *h; if (!bfd_generic_define_common_symbol (output_bfd, info, harg)) return FALSE; h = (struct xcoff_link_hash_entry *) harg; h->flags |= XCOFF_DEF_REGULAR; return TRUE; } /* If symbol H has not been interpreted as a function descriptor, see whether it should be. Set up its descriptor information if so. */ static bfd_boolean xcoff_find_function (struct bfd_link_info *info, struct xcoff_link_hash_entry *h) { if ((h->flags & XCOFF_DESCRIPTOR) == 0 && h->root.root.string[0] != '.') { char *fnname; struct xcoff_link_hash_entry *hfn; bfd_size_type amt; amt = strlen (h->root.root.string) + 2; fnname = bfd_malloc (amt); if (fnname == NULL) return FALSE; fnname[0] = '.'; strcpy (fnname + 1, h->root.root.string); hfn = xcoff_link_hash_lookup (xcoff_hash_table (info), fnname, FALSE, FALSE, TRUE); free (fnname); if (hfn != NULL && hfn->smclas == XMC_PR && (hfn->root.type == bfd_link_hash_defined || hfn->root.type == bfd_link_hash_defweak)) { h->flags |= XCOFF_DESCRIPTOR; h->descriptor = hfn; hfn->descriptor = h; } } return TRUE; } /* Return true if the given bfd contains at least one shared object. */ static bfd_boolean xcoff_archive_contains_shared_object_p (struct bfd_link_info *info, bfd *archive) { struct xcoff_archive_info *archive_info; bfd *member; archive_info = xcoff_get_archive_info (info, archive); if (!archive_info->know_contains_shared_object_p) { member = bfd_openr_next_archived_file (archive, NULL); while (member != NULL && (member->flags & DYNAMIC) == 0) member = bfd_openr_next_archived_file (archive, member); archive_info->contains_shared_object_p = (member != NULL); archive_info->know_contains_shared_object_p = 1; } return archive_info->contains_shared_object_p; } /* Symbol H qualifies for export by -bexpfull. Return true if it also qualifies for export by -bexpall. */ static bfd_boolean xcoff_covered_by_expall_p (struct xcoff_link_hash_entry *h) { /* Exclude symbols beginning with '_'. */ if (h->root.root.string[0] == '_') return FALSE; /* Exclude archive members that would otherwise be unreferenced. */ if ((h->flags & XCOFF_MARK) == 0 && (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) && h->root.u.def.section->owner != NULL && h->root.u.def.section->owner->my_archive != NULL) return FALSE; return TRUE; } /* Return true if symbol H qualifies for the forms of automatic export specified by AUTO_EXPORT_FLAGS. */ static bfd_boolean xcoff_auto_export_p (struct bfd_link_info *info, struct xcoff_link_hash_entry *h, unsigned int auto_export_flags) { /* Don't automatically export things that were explicitly exported. */ if ((h->flags & XCOFF_EXPORT) != 0) return FALSE; /* Don't export things that we don't define. */ if ((h->flags & XCOFF_DEF_REGULAR) == 0) return FALSE; /* Don't export functions; export their descriptors instead. */ if (h->root.root.string[0] == '.') return FALSE; /* We don't export a symbol which is being defined by an object included from an archive which contains a shared object. The rationale is that if an archive contains both an unshared and a shared object, then there must be some reason that the unshared object is unshared, and we don't want to start providing a shared version of it. In particular, this solves a bug involving the _savefNN set of functions. gcc will call those functions without providing a slot to restore the TOC, so it is essential that these functions be linked in directly and not from a shared object, which means that a shared object which also happens to link them in must not export them. This is confusing, but I haven't been able to think of a different approach. Note that the symbols can, of course, be exported explicitly. */ if (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) { bfd *owner; owner = h->root.u.def.section->owner; if (owner != NULL && owner->my_archive != NULL && xcoff_archive_contains_shared_object_p (info, owner->my_archive)) return FALSE; } /* Otherwise, all symbols are exported by -bexpfull. */ if ((auto_export_flags & XCOFF_EXPFULL) != 0) return TRUE; /* Despite its name, -bexpall exports most but not all symbols. */ if ((auto_export_flags & XCOFF_EXPALL) != 0 && xcoff_covered_by_expall_p (h)) return TRUE; return FALSE; } /* Return true if relocation REL needs to be copied to the .loader section. If REL is against a global symbol, H is that symbol, otherwise it is null. */ static bfd_boolean xcoff_need_ldrel_p (struct bfd_link_info *info, struct internal_reloc *rel, struct xcoff_link_hash_entry *h) { if (!xcoff_hash_table (info)->loader_section) return FALSE; switch (rel->r_type) { case R_TOC: case R_GL: case R_TCL: case R_TRL: case R_TRLA: /* We should never need a .loader reloc for a TOC-relative reloc. */ return FALSE; default: /* In this case, relocations against defined symbols can be resolved statically. */ if (h == NULL || h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak || h->root.type == bfd_link_hash_common) return FALSE; /* We will always provide a local definition of function symbols, even if we don't have one yet. */ if ((h->flags & XCOFF_CALLED) != 0) return FALSE; return TRUE; case R_POS: case R_NEG: case R_RL: case R_RLA: /* Absolute relocations against absolute symbols can be resolved statically. */ if (h != NULL && (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) && bfd_is_abs_section (h->root.u.def.section)) return FALSE; return TRUE; } } /* Mark a symbol as not being garbage, including the section in which it is defined. */ static inline bfd_boolean xcoff_mark_symbol (struct bfd_link_info *info, struct xcoff_link_hash_entry *h) { if ((h->flags & XCOFF_MARK) != 0) return TRUE; h->flags |= XCOFF_MARK; /* If we're marking an undefined symbol, try find some way of defining it. */ if (!info->relocatable && (h->flags & XCOFF_IMPORT) == 0 && (h->flags & XCOFF_DEF_REGULAR) == 0 && (h->root.type == bfd_link_hash_undefined || h->root.type == bfd_link_hash_undefweak)) { /* First check whether this symbol can be interpreted as an undefined function descriptor for a defined function symbol. */ if (!xcoff_find_function (info, h)) return FALSE; if ((h->flags & XCOFF_DESCRIPTOR) != 0 && (h->descriptor->root.type == bfd_link_hash_defined || h->descriptor->root.type == bfd_link_hash_defweak)) { /* This is a descriptor for a defined symbol, but the input objects have not defined the descriptor itself. Fill in the definition automatically. Note that we do this even if we found a dynamic definition of H. The local function definition logically overrides the dynamic one. */ asection *sec; sec = xcoff_hash_table (info)->descriptor_section; h->root.type = bfd_link_hash_defined; h->root.u.def.section = sec; h->root.u.def.value = sec->size; h->smclas = XMC_DS; h->flags |= XCOFF_DEF_REGULAR; /* The size of the function descriptor depends on whether this is xcoff32 (12) or xcoff64 (24). */ sec->size += bfd_xcoff_function_descriptor_size (sec->owner); /* A function descriptor uses two relocs: one for the associated code, and one for the TOC address. */ xcoff_hash_table (info)->ldrel_count += 2; sec->reloc_count += 2; /* Mark the function itself. */ if (!xcoff_mark_symbol (info, h->descriptor)) return FALSE; /* Mark the TOC section, so that we get an anchor to relocate against. */ if (!xcoff_mark (info, xcoff_hash_table (info)->toc_section)) return FALSE; /* We handle writing out the contents of the descriptor in xcoff_write_global_symbol. */ } else if (info->static_link) /* We can't get a symbol value dynamically, so just assume that it's undefined. */ h->flags |= XCOFF_WAS_UNDEFINED; else if ((h->flags & XCOFF_CALLED) != 0) { /* This is a function symbol for which we need to create linkage code. */ asection *sec; struct xcoff_link_hash_entry *hds; /* Mark the descriptor (and its TOC section). */ hds = h->descriptor; BFD_ASSERT ((hds->root.type == bfd_link_hash_undefined || hds->root.type == bfd_link_hash_undefweak) && (hds->flags & XCOFF_DEF_REGULAR) == 0); if (!xcoff_mark_symbol (info, hds)) return FALSE; /* Treat this symbol as undefined if the descriptor was. */ if ((hds->flags & XCOFF_WAS_UNDEFINED) != 0) h->flags |= XCOFF_WAS_UNDEFINED; /* Allocate room for the global linkage code itself. */ sec = xcoff_hash_table (info)->linkage_section; h->root.type = bfd_link_hash_defined; h->root.u.def.section = sec; h->root.u.def.value = sec->size; h->smclas = XMC_GL; h->flags |= XCOFF_DEF_REGULAR; sec->size += bfd_xcoff_glink_code_size (info->output_bfd); /* The global linkage code requires a TOC entry for the descriptor. */ if (hds->toc_section == NULL) { int byte_size; /* 32 vs 64 xcoff32 uses 4 bytes in the toc. xcoff64 uses 8 bytes in the toc. */ if (bfd_xcoff_is_xcoff64 (info->output_bfd)) byte_size = 8; else if (bfd_xcoff_is_xcoff32 (info->output_bfd)) byte_size = 4; else return FALSE; /* Allocate room in the fallback TOC section. */ hds->toc_section = xcoff_hash_table (info)->toc_section; hds->u.toc_offset = hds->toc_section->size; hds->toc_section->size += byte_size; if (!xcoff_mark (info, hds->toc_section)) return FALSE; /* Allocate room for a static and dynamic R_TOC relocation. */ ++xcoff_hash_table (info)->ldrel_count; ++hds->toc_section->reloc_count; /* Set the index to -2 to force this symbol to get written out. */ hds->indx = -2; hds->flags |= XCOFF_SET_TOC | XCOFF_LDREL; } } else if ((h->flags & XCOFF_DEF_DYNAMIC) == 0) { /* Record that the symbol was undefined, then import it. -brtl links use a special fake import file. */ h->flags |= XCOFF_WAS_UNDEFINED | XCOFF_IMPORT; if (xcoff_hash_table (info)->rtld) { if (!xcoff_set_import_path (info, h, "", "..", "")) return FALSE; } else { if (!xcoff_set_import_path (info, h, NULL, NULL, NULL)) return FALSE; } } } if (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) { asection *hsec; hsec = h->root.u.def.section; if (! bfd_is_abs_section (hsec) && (hsec->flags & SEC_MARK) == 0) { if (! xcoff_mark (info, hsec)) return FALSE; } } if (h->toc_section != NULL && (h->toc_section->flags & SEC_MARK) == 0) { if (! xcoff_mark (info, h->toc_section)) return FALSE; } return TRUE; } /* Look for a symbol called NAME. If the symbol is defined, mark it. If the symbol exists, set FLAGS. */ static bfd_boolean xcoff_mark_symbol_by_name (struct bfd_link_info *info, const char *name, unsigned int flags) { struct xcoff_link_hash_entry *h; h = xcoff_link_hash_lookup (xcoff_hash_table (info), name, FALSE, FALSE, TRUE); if (h != NULL) { h->flags |= flags; if (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) { if (!xcoff_mark (info, h->root.u.def.section)) return FALSE; } } return TRUE; } /* The mark phase of garbage collection. For a given section, mark it, and all the sections which define symbols to which it refers. Because this function needs to look at the relocs, we also count the number of relocs which need to be copied into the .loader section. */ static bfd_boolean xcoff_mark (struct bfd_link_info *info, asection *sec) { if (bfd_is_abs_section (sec) || (sec->flags & SEC_MARK) != 0) return TRUE; sec->flags |= SEC_MARK; if (sec->owner->xvec == info->output_bfd->xvec && coff_section_data (sec->owner, sec) != NULL && xcoff_section_data (sec->owner, sec) != NULL) { struct xcoff_link_hash_entry **syms; struct internal_reloc *rel, *relend; asection **csects; unsigned long i, first, last; /* Mark all the symbols in this section. */ syms = obj_xcoff_sym_hashes (sec->owner); csects = xcoff_data (sec->owner)->csects; first = xcoff_section_data (sec->owner, sec)->first_symndx; last = xcoff_section_data (sec->owner, sec)->last_symndx; for (i = first; i <= last; i++) if (csects[i] == sec && syms[i] != NULL && (syms[i]->flags & XCOFF_MARK) == 0) { if (!xcoff_mark_symbol (info, syms[i])) return FALSE; } /* Look through the section relocs. */ if ((sec->flags & SEC_RELOC) != 0 && sec->reloc_count > 0) { rel = xcoff_read_internal_relocs (sec->owner, sec, TRUE, NULL, FALSE, NULL); if (rel == NULL) return FALSE; relend = rel + sec->reloc_count; for (; rel < relend; rel++) { struct xcoff_link_hash_entry *h; if ((unsigned int) rel->r_symndx > obj_raw_syment_count (sec->owner)) continue; h = obj_xcoff_sym_hashes (sec->owner)[rel->r_symndx]; if (h != NULL) { if ((h->flags & XCOFF_MARK) == 0) { if (!xcoff_mark_symbol (info, h)) return FALSE; } } else { asection *rsec; rsec = xcoff_data (sec->owner)->csects[rel->r_symndx]; if (rsec != NULL && (rsec->flags & SEC_MARK) == 0) { if (!xcoff_mark (info, rsec)) return FALSE; } } /* See if this reloc needs to be copied into the .loader section. */ if (xcoff_need_ldrel_p (info, rel, h)) { ++xcoff_hash_table (info)->ldrel_count; if (h != NULL) h->flags |= XCOFF_LDREL; } } if (! info->keep_memory && coff_section_data (sec->owner, sec) != NULL && coff_section_data (sec->owner, sec)->relocs != NULL && ! coff_section_data (sec->owner, sec)->keep_relocs) { free (coff_section_data (sec->owner, sec)->relocs); coff_section_data (sec->owner, sec)->relocs = NULL; } } } return TRUE; } /* Routines that are called after all the input files have been handled, but before the sections are laid out in memory. */ /* The sweep phase of garbage collection. Remove all garbage sections. */ static void xcoff_sweep (struct bfd_link_info *info) { bfd *sub; for (sub = info->input_bfds; sub != NULL; sub = sub->link_next) { asection *o; for (o = sub->sections; o != NULL; o = o->next) { if ((o->flags & SEC_MARK) == 0) { /* Keep all sections from non-XCOFF input files. Keep special sections. Keep .debug sections for the moment. */ if (sub->xvec != info->output_bfd->xvec || o == xcoff_hash_table (info)->debug_section || o == xcoff_hash_table (info)->loader_section || o == xcoff_hash_table (info)->linkage_section || o == xcoff_hash_table (info)->descriptor_section || (bfd_get_section_flags (sub, o) & SEC_DEBUGGING) || strcmp (o->name, ".debug") == 0) o->flags |= SEC_MARK; else { o->size = 0; o->reloc_count = 0; } } } } } /* Record the number of elements in a set. This is used to output the correct csect length. */ bfd_boolean bfd_xcoff_link_record_set (bfd *output_bfd, struct bfd_link_info *info, struct bfd_link_hash_entry *harg, bfd_size_type size) { struct xcoff_link_hash_entry *h = (struct xcoff_link_hash_entry *) harg; struct xcoff_link_size_list *n; bfd_size_type amt; if (bfd_get_flavour (output_bfd) != bfd_target_xcoff_flavour) return TRUE; /* This will hardly ever be called. I don't want to burn four bytes per global symbol, so instead the size is kept on a linked list attached to the hash table. */ amt = sizeof (* n); n = bfd_alloc (output_bfd, amt); if (n == NULL) return FALSE; n->next = xcoff_hash_table (info)->size_list; n->h = h; n->size = size; xcoff_hash_table (info)->size_list = n; h->flags |= XCOFF_HAS_SIZE; return TRUE; } /* Import a symbol. */ bfd_boolean bfd_xcoff_import_symbol (bfd *output_bfd, struct bfd_link_info *info, struct bfd_link_hash_entry *harg, bfd_vma val, const char *imppath, const char *impfile, const char *impmember, unsigned int syscall_flag) { struct xcoff_link_hash_entry *h = (struct xcoff_link_hash_entry *) harg; if (bfd_get_flavour (output_bfd) != bfd_target_xcoff_flavour) return TRUE; /* A symbol name which starts with a period is the code for a function. If the symbol is undefined, then add an undefined symbol for the function descriptor, and import that instead. */ if (h->root.root.string[0] == '.' && h->root.type == bfd_link_hash_undefined && val == (bfd_vma) -1) { struct xcoff_link_hash_entry *hds; hds = h->descriptor; if (hds == NULL) { hds = xcoff_link_hash_lookup (xcoff_hash_table (info), h->root.root.string + 1, TRUE, FALSE, TRUE); if (hds == NULL) return FALSE; if (hds->root.type == bfd_link_hash_new) { hds->root.type = bfd_link_hash_undefined; hds->root.u.undef.abfd = h->root.u.undef.abfd; } hds->flags |= XCOFF_DESCRIPTOR; BFD_ASSERT ((h->flags & XCOFF_DESCRIPTOR) == 0); hds->descriptor = h; h->descriptor = hds; } /* Now, if the descriptor is undefined, import the descriptor rather than the symbol we were told to import. FIXME: Is this correct in all cases? */ if (hds->root.type == bfd_link_hash_undefined) h = hds; } h->flags |= (XCOFF_IMPORT | syscall_flag); if (val != (bfd_vma) -1) { if (h->root.type == bfd_link_hash_defined && (! bfd_is_abs_section (h->root.u.def.section) || h->root.u.def.value != val)) { if (! ((*info->callbacks->multiple_definition) (info, &h->root, output_bfd, bfd_abs_section_ptr, val))) return FALSE; } h->root.type = bfd_link_hash_defined; h->root.u.def.section = bfd_abs_section_ptr; h->root.u.def.value = val; h->smclas = XMC_XO; } if (!xcoff_set_import_path (info, h, imppath, impfile, impmember)) return FALSE; return TRUE; } /* Export a symbol. */ bfd_boolean bfd_xcoff_export_symbol (bfd *output_bfd, struct bfd_link_info *info, struct bfd_link_hash_entry *harg) { struct xcoff_link_hash_entry *h = (struct xcoff_link_hash_entry *) harg; if (bfd_get_flavour (output_bfd) != bfd_target_xcoff_flavour) return TRUE; h->flags |= XCOFF_EXPORT; /* FIXME: I'm not at all sure what syscall is supposed to mean, so I'm just going to ignore it until somebody explains it. */ /* Make sure we don't garbage collect this symbol. */ if (! xcoff_mark_symbol (info, h)) return FALSE; /* If this is a function descriptor, make sure we don't garbage collect the associated function code. We normally don't have to worry about this, because the descriptor will be attached to a section with relocs, but if we are creating the descriptor ourselves those relocs will not be visible to the mark code. */ if ((h->flags & XCOFF_DESCRIPTOR) != 0) { if (! xcoff_mark_symbol (info, h->descriptor)) return FALSE; } return TRUE; } /* Count a reloc against a symbol. This is called for relocs generated by the linker script, typically for global constructors and destructors. */ bfd_boolean bfd_xcoff_link_count_reloc (bfd *output_bfd, struct bfd_link_info *info, const char *name) { struct xcoff_link_hash_entry *h; if (bfd_get_flavour (output_bfd) != bfd_target_xcoff_flavour) return TRUE; h = ((struct xcoff_link_hash_entry *) bfd_wrapped_link_hash_lookup (output_bfd, info, name, FALSE, FALSE, FALSE)); if (h == NULL) { (*_bfd_error_handler) (_("%s: no such symbol"), name); bfd_set_error (bfd_error_no_symbols); return FALSE; } h->flags |= XCOFF_REF_REGULAR; if (xcoff_hash_table (info)->loader_section) { h->flags |= XCOFF_LDREL; ++xcoff_hash_table (info)->ldrel_count; } /* Mark the symbol to avoid garbage collection. */ if (! xcoff_mark_symbol (info, h)) return FALSE; return TRUE; } /* This function is called for each symbol to which the linker script assigns a value. */ bfd_boolean bfd_xcoff_record_link_assignment (bfd *output_bfd, struct bfd_link_info *info, const char *name) { struct xcoff_link_hash_entry *h; if (bfd_get_flavour (output_bfd) != bfd_target_xcoff_flavour) return TRUE; h = xcoff_link_hash_lookup (xcoff_hash_table (info), name, TRUE, TRUE, FALSE); if (h == NULL) return FALSE; h->flags |= XCOFF_DEF_REGULAR; return TRUE; } /* An xcoff_link_hash_traverse callback for which DATA points to an xcoff_loader_info. Mark all symbols that should be automatically exported. */ static bfd_boolean xcoff_mark_auto_exports (struct xcoff_link_hash_entry *h, void *data) { struct xcoff_loader_info *ldinfo; ldinfo = (struct xcoff_loader_info *) data; if (xcoff_auto_export_p (ldinfo->info, h, ldinfo->auto_export_flags)) { if (!xcoff_mark_symbol (ldinfo->info, h)) ldinfo->failed = TRUE; } return TRUE; } /* Add a symbol to the .loader symbols, if necessary. */ /* INPUT_BFD has an external symbol associated with hash table entry H and csect CSECT. Return true if INPUT_BFD defines H. */ static bfd_boolean xcoff_final_definition_p (bfd *input_bfd, struct xcoff_link_hash_entry *h, asection *csect) { switch (h->root.type) { case bfd_link_hash_defined: case bfd_link_hash_defweak: /* No input bfd owns absolute symbols. They are written by xcoff_write_global_symbol instead. */ return (!bfd_is_abs_section (csect) && h->root.u.def.section == csect); case bfd_link_hash_common: return h->root.u.c.p->section->owner == input_bfd; case bfd_link_hash_undefined: case bfd_link_hash_undefweak: /* We can't treat undef.abfd as the owner because that bfd might be a dynamic object. Allow any bfd to claim it. */ return TRUE; default: abort (); } } /* See if H should have a loader symbol associated with it. */ static bfd_boolean xcoff_build_ldsym (struct xcoff_loader_info *ldinfo, struct xcoff_link_hash_entry *h) { bfd_size_type amt; /* Warn if this symbol is exported but not defined. */ if ((h->flags & XCOFF_EXPORT) != 0 && (h->flags & XCOFF_WAS_UNDEFINED) != 0) { (*_bfd_error_handler) (_("warning: attempt to export undefined symbol `%s'"), h->root.root.string); return TRUE; } /* We need to add a symbol to the .loader section if it is mentioned in a reloc which we are copying to the .loader section and it was not defined or common, or if it is the entry point, or if it is being exported. */ if (((h->flags & XCOFF_LDREL) == 0 || h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak || h->root.type == bfd_link_hash_common) && (h->flags & XCOFF_ENTRY) == 0 && (h->flags & XCOFF_EXPORT) == 0) return TRUE; /* We need to add this symbol to the .loader symbols. */ BFD_ASSERT (h->ldsym == NULL); amt = sizeof (struct internal_ldsym); h->ldsym = bfd_zalloc (ldinfo->output_bfd, amt); if (h->ldsym == NULL) { ldinfo->failed = TRUE; return FALSE; } if ((h->flags & XCOFF_IMPORT) != 0) { /* Give imported descriptors class XMC_DS rather than XMC_UA. */ if ((h->flags & XCOFF_DESCRIPTOR) != 0) h->smclas = XMC_DS; h->ldsym->l_ifile = h->ldindx; } /* The first 3 symbol table indices are reserved to indicate the data, text and bss sections. */ h->ldindx = ldinfo->ldsym_count + 3; ++ldinfo->ldsym_count; if (! bfd_xcoff_put_ldsymbol_name (ldinfo->output_bfd, ldinfo, h->ldsym, h->root.root.string)) return FALSE; h->flags |= XCOFF_BUILT_LDSYM; return TRUE; } /* An xcoff_htab_traverse callback that is called for each symbol once garbage collection is complete. */ static bfd_boolean xcoff_post_gc_symbol (struct xcoff_link_hash_entry *h, void * p) { struct xcoff_loader_info *ldinfo = (struct xcoff_loader_info *) p; if (h->root.type == bfd_link_hash_warning) h = (struct xcoff_link_hash_entry *) h->root.u.i.link; /* __rtinit, this symbol has special handling. */ if (h->flags & XCOFF_RTINIT) return TRUE; /* We don't want to garbage collect symbols which are not defined in XCOFF files. This is a convenient place to mark them. */ if (xcoff_hash_table (ldinfo->info)->gc && (h->flags & XCOFF_MARK) == 0 && (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) && (h->root.u.def.section->owner == NULL || (h->root.u.def.section->owner->xvec != ldinfo->info->output_bfd->xvec))) h->flags |= XCOFF_MARK; /* Skip discarded symbols. */ if (xcoff_hash_table (ldinfo->info)->gc && (h->flags & XCOFF_MARK) == 0) return TRUE; /* If this is still a common symbol, and it wasn't garbage collected, we need to actually allocate space for it in the .bss section. */ if (h->root.type == bfd_link_hash_common && h->root.u.c.p->section->size == 0) { BFD_ASSERT (bfd_is_com_section (h->root.u.c.p->section)); h->root.u.c.p->section->size = h->root.u.c.size; } if (xcoff_hash_table (ldinfo->info)->loader_section) { if (xcoff_auto_export_p (ldinfo->info, h, ldinfo->auto_export_flags)) h->flags |= XCOFF_EXPORT; if (!xcoff_build_ldsym (ldinfo, h)) return FALSE; } return TRUE; } /* INPUT_BFD includes XCOFF symbol ISYM, which is associated with linker hash table entry H and csect CSECT. AUX contains ISYM's auxillary csect information, if any. NAME is the function's name if the name is stored in the .debug section, otherwise it is null. Return 1 if we should include an appropriately-adjusted ISYM in the output file, 0 if we should discard ISYM, or -1 if an error occured. */ static int xcoff_keep_symbol_p (struct bfd_link_info *info, bfd *input_bfd, struct internal_syment *isym, union internal_auxent *aux, struct xcoff_link_hash_entry *h, asection *csect, const char *name) { int smtyp; /* If we are skipping this csect, we want to strip the symbol too. */ if (csect == NULL) return 0; /* Likewise if we garbage-collected the csect. */ if (xcoff_hash_table (info)->gc && !bfd_is_abs_section (csect) && !bfd_is_und_section (csect) && (csect->flags & SEC_MARK) == 0) return 0; /* An XCOFF linker always removes C_STAT symbols. */ if (isym->n_sclass == C_STAT) return 0; /* We generate the TOC anchor separately. */ if (isym->n_sclass == C_HIDEXT && aux->x_csect.x_smclas == XMC_TC0) return 0; /* If we are stripping all symbols, we want to discard this one. */ if (info->strip == strip_all) return 0; /* Discard symbols that are defined elsewhere. */ if (EXTERN_SYM_P (isym->n_sclass)) { if ((h->flags & XCOFF_ALLOCATED) != 0) return 0; if (!xcoff_final_definition_p (input_bfd, h, csect)) return 0; } /* If we're discarding local symbols, check whether ISYM is local. */ smtyp = SMTYP_SMTYP (aux->x_csect.x_smtyp); if (info->discard == discard_all && !EXTERN_SYM_P (isym->n_sclass) && (isym->n_sclass != C_HIDEXT || smtyp != XTY_SD)) return 0; /* If we're stripping debugging symbols, check whether ISYM is one. */ if (info->strip == strip_debugger && isym->n_scnum == N_DEBUG) return 0; /* If we are stripping symbols based on name, check how ISYM's name should be handled. */ if (info->strip == strip_some || info->discard == discard_l) { char buf[SYMNMLEN + 1]; if (name == NULL) { name = _bfd_coff_internal_syment_name (input_bfd, isym, buf); if (name == NULL) return -1; } if (info->strip == strip_some && bfd_hash_lookup (info->keep_hash, name, FALSE, FALSE) == NULL) return 0; if (info->discard == discard_l && !EXTERN_SYM_P (isym->n_sclass) && (isym->n_sclass != C_HIDEXT || smtyp != XTY_SD) && bfd_is_local_label_name (input_bfd, name)) return 0; } return 1; } /* Lay out the .loader section, filling in the header and the import paths. LIBPATH is as for bfd_xcoff_size_dynamic_sections. */ static bfd_boolean xcoff_build_loader_section (struct xcoff_loader_info *ldinfo, const char *libpath) { bfd *output_bfd; struct xcoff_link_hash_table *htab; struct internal_ldhdr *ldhdr; struct xcoff_import_file *fl; bfd_size_type stoff; size_t impsize, impcount; asection *lsec; char *out; /* Work out the size of the import file names. Each import file ID consists of three null terminated strings: the path, the file name, and the archive member name. The first entry in the list of names is the path to use to find objects, which the linker has passed in as the libpath argument. For some reason, the path entry in the other import file names appears to always be empty. */ output_bfd = ldinfo->output_bfd; htab = xcoff_hash_table (ldinfo->info); impsize = strlen (libpath) + 3; impcount = 1; for (fl = htab->imports; fl != NULL; fl = fl->next) { ++impcount; impsize += (strlen (fl->path) + strlen (fl->file) + strlen (fl->member) + 3); } /* Set up the .loader section header. */ ldhdr = &htab->ldhdr; ldhdr->l_version = bfd_xcoff_ldhdr_version(output_bfd); ldhdr->l_nsyms = ldinfo->ldsym_count; ldhdr->l_nreloc = htab->ldrel_count; ldhdr->l_istlen = impsize; ldhdr->l_nimpid = impcount; ldhdr->l_impoff = (bfd_xcoff_ldhdrsz (output_bfd) + ldhdr->l_nsyms * bfd_xcoff_ldsymsz (output_bfd) + ldhdr->l_nreloc * bfd_xcoff_ldrelsz (output_bfd)); ldhdr->l_stlen = ldinfo->string_size; stoff = ldhdr->l_impoff + impsize; if (ldinfo->string_size == 0) ldhdr->l_stoff = 0; else ldhdr->l_stoff = stoff; /* 64 bit elements to ldhdr The swap out routine for 32 bit will ignore them. Nothing fancy, symbols come after the header and relocs come after symbols. */ ldhdr->l_symoff = bfd_xcoff_ldhdrsz (output_bfd); ldhdr->l_rldoff = (bfd_xcoff_ldhdrsz (output_bfd) + ldhdr->l_nsyms * bfd_xcoff_ldsymsz (output_bfd)); /* We now know the final size of the .loader section. Allocate space for it. */ lsec = htab->loader_section; lsec->size = stoff + ldhdr->l_stlen; lsec->contents = bfd_zalloc (output_bfd, lsec->size); if (lsec->contents == NULL) return FALSE; /* Set up the header. */ bfd_xcoff_swap_ldhdr_out (output_bfd, ldhdr, lsec->contents); /* Set up the import file names. */ out = (char *) lsec->contents + ldhdr->l_impoff; strcpy (out, libpath); out += strlen (libpath) + 1; *out++ = '\0'; *out++ = '\0'; for (fl = htab->imports; fl != NULL; fl = fl->next) { const char *s; s = fl->path; while ((*out++ = *s++) != '\0') ; s = fl->file; while ((*out++ = *s++) != '\0') ; s = fl->member; while ((*out++ = *s++) != '\0') ; } BFD_ASSERT ((bfd_size_type) ((bfd_byte *) out - lsec->contents) == stoff); /* Set up the symbol string table. */ if (ldinfo->string_size > 0) { memcpy (out, ldinfo->strings, ldinfo->string_size); free (ldinfo->strings); ldinfo->strings = NULL; } /* We can't set up the symbol table or the relocs yet, because we don't yet know the final position of the various sections. The .loader symbols are written out when the corresponding normal symbols are written out in xcoff_link_input_bfd or xcoff_write_global_symbol. The .loader relocs are written out when the corresponding normal relocs are handled in xcoff_link_input_bfd. */ return TRUE; } /* Build the .loader section. This is called by the XCOFF linker emulation before_allocation routine. We must set the size of the .loader section before the linker lays out the output file. LIBPATH is the library path to search for shared objects; this is normally built from the -L arguments passed to the linker. ENTRY is the name of the entry point symbol (the -e linker option). FILE_ALIGN is the alignment to use for sections within the file (the -H linker option). MAXSTACK is the maximum stack size (the -bmaxstack linker option). MAXDATA is the maximum data size (the -bmaxdata linker option). GC is whether to do garbage collection (the -bgc linker option). MODTYPE is the module type (the -bmodtype linker option). TEXTRO is whether the text section must be read only (the -btextro linker option). AUTO_EXPORT_FLAGS is a mask of XCOFF_EXPALL and XCOFF_EXPFULL. SPECIAL_SECTIONS is set by this routine to csects with magic names like _end. */ bfd_boolean bfd_xcoff_size_dynamic_sections (bfd *output_bfd, struct bfd_link_info *info, const char *libpath, const char *entry, unsigned long file_align, unsigned long maxstack, unsigned long maxdata, bfd_boolean gc, int modtype, bfd_boolean textro, unsigned int auto_export_flags, asection **special_sections, bfd_boolean rtld) { struct xcoff_loader_info ldinfo; int i; asection *sec; bfd *sub; struct bfd_strtab_hash *debug_strtab; bfd_byte *debug_contents = NULL; bfd_size_type amt; if (bfd_get_flavour (output_bfd) != bfd_target_xcoff_flavour) { for (i = 0; i < XCOFF_NUMBER_OF_SPECIAL_SECTIONS; i++) special_sections[i] = NULL; return TRUE; } ldinfo.failed = FALSE; ldinfo.output_bfd = output_bfd; ldinfo.info = info; ldinfo.auto_export_flags = auto_export_flags; ldinfo.ldsym_count = 0; ldinfo.string_size = 0; ldinfo.strings = NULL; ldinfo.string_alc = 0; xcoff_data (output_bfd)->maxstack = maxstack; xcoff_data (output_bfd)->maxdata = maxdata; xcoff_data (output_bfd)->modtype = modtype; xcoff_hash_table (info)->file_align = file_align; xcoff_hash_table (info)->textro = textro; xcoff_hash_table (info)->rtld = rtld; /* __rtinit */ if (xcoff_hash_table (info)->loader_section && (info->init_function || info->fini_function || rtld)) { struct xcoff_link_hash_entry *hsym; struct internal_ldsym *ldsym; hsym = xcoff_link_hash_lookup (xcoff_hash_table (info), "__rtinit", FALSE, FALSE, TRUE); if (hsym == NULL) { (*_bfd_error_handler) (_("error: undefined symbol __rtinit")); return FALSE; } xcoff_mark_symbol (info, hsym); hsym->flags |= (XCOFF_DEF_REGULAR | XCOFF_RTINIT); /* __rtinit initialized. */ amt = sizeof (* ldsym); ldsym = bfd_malloc (amt); ldsym->l_value = 0; /* Will be filled in later. */ ldsym->l_scnum = 2; /* Data section. */ ldsym->l_smtype = XTY_SD; /* Csect section definition. */ ldsym->l_smclas = 5; /* .rw. */ ldsym->l_ifile = 0; /* Special system loader symbol. */ ldsym->l_parm = 0; /* NA. */ /* Force __rtinit to be the first symbol in the loader symbol table See xcoff_build_ldsyms The first 3 symbol table indices are reserved to indicate the data, text and bss sections. */ BFD_ASSERT (0 == ldinfo.ldsym_count); hsym->ldindx = 3; ldinfo.ldsym_count = 1; hsym->ldsym = ldsym; if (! bfd_xcoff_put_ldsymbol_name (ldinfo.output_bfd, &ldinfo, hsym->ldsym, hsym->root.root.string)) return FALSE; /* This symbol is written out by xcoff_write_global_symbol Set stuff up so xcoff_write_global_symbol logic works. */ hsym->flags |= XCOFF_DEF_REGULAR | XCOFF_MARK; hsym->root.type = bfd_link_hash_defined; hsym->root.u.def.value = 0; } /* Garbage collect unused sections. */ if (info->relocatable || !gc) { gc = FALSE; xcoff_hash_table (info)->gc = FALSE; /* We still need to call xcoff_mark, in order to set ldrel_count correctly. */ for (sub = info->input_bfds; sub != NULL; sub = sub->link_next) { asection *o; for (o = sub->sections; o != NULL; o = o->next) { /* We shouldn't unconditionaly mark the TOC section. The output file should only have a TOC if either (a) one of the input files did or (b) we end up creating TOC references as part of the link process. */ if (o != xcoff_hash_table (info)->toc_section && (o->flags & SEC_MARK) == 0) { if (! xcoff_mark (info, o)) goto error_return; } } } } else { if (entry != NULL && !xcoff_mark_symbol_by_name (info, entry, XCOFF_ENTRY)) goto error_return; if (info->init_function != NULL && !xcoff_mark_symbol_by_name (info, info->init_function, 0)) goto error_return; if (info->fini_function != NULL && !xcoff_mark_symbol_by_name (info, info->fini_function, 0)) goto error_return; if (auto_export_flags != 0) { xcoff_link_hash_traverse (xcoff_hash_table (info), xcoff_mark_auto_exports, &ldinfo); if (ldinfo.failed) goto error_return; } xcoff_sweep (info); xcoff_hash_table (info)->gc = TRUE; } /* Return special sections to the caller. */ for (i = 0; i < XCOFF_NUMBER_OF_SPECIAL_SECTIONS; i++) { sec = xcoff_hash_table (info)->special_sections[i]; if (sec != NULL && gc && (sec->flags & SEC_MARK) == 0) sec = NULL; special_sections[i] = sec; } if (info->input_bfds == NULL) /* I'm not sure what to do in this bizarre case. */ return TRUE; xcoff_link_hash_traverse (xcoff_hash_table (info), xcoff_post_gc_symbol, (void *) &ldinfo); if (ldinfo.failed) goto error_return; if (xcoff_hash_table (info)->loader_section && !xcoff_build_loader_section (&ldinfo, libpath)) goto error_return; /* Allocate space for the magic sections. */ sec = xcoff_hash_table (info)->linkage_section; if (sec->size > 0) { sec->contents = bfd_zalloc (output_bfd, sec->size); if (sec->contents == NULL) goto error_return; } sec = xcoff_hash_table (info)->toc_section; if (sec->size > 0) { sec->contents = bfd_zalloc (output_bfd, sec->size); if (sec->contents == NULL) goto error_return; } sec = xcoff_hash_table (info)->descriptor_section; if (sec->size > 0) { sec->contents = bfd_zalloc (output_bfd, sec->size); if (sec->contents == NULL) goto error_return; } /* Now that we've done garbage collection, decide which symbols to keep, and figure out the contents of the .debug section. */ debug_strtab = xcoff_hash_table (info)->debug_strtab; for (sub = info->input_bfds; sub != NULL; sub = sub->link_next) { asection *subdeb; bfd_size_type symcount; long *debug_index; asection **csectpp; unsigned int *lineno_counts; struct xcoff_link_hash_entry **sym_hash; bfd_byte *esym, *esymend; bfd_size_type symesz; if (sub->xvec != info->output_bfd->xvec) continue; if ((sub->flags & DYNAMIC) != 0 && !info->static_link) continue; if (! _bfd_coff_get_external_symbols (sub)) goto error_return; symcount = obj_raw_syment_count (sub); debug_index = bfd_zalloc (sub, symcount * sizeof (long)); if (debug_index == NULL) goto error_return; xcoff_data (sub)->debug_indices = debug_index; if (info->strip == strip_all || info->strip == strip_debugger || info->discard == discard_all) /* We're stripping all debugging information, so there's no need to read SUB's .debug section. */ subdeb = NULL; else { /* Grab the contents of SUB's .debug section, if any. */ subdeb = bfd_get_section_by_name (sub, ".debug"); if (subdeb != NULL && subdeb->size > 0) { /* We use malloc and copy the names into the debug stringtab, rather than bfd_alloc, because I expect that, when linking many files together, many of the strings will be the same. Storing the strings in the hash table should save space in this case. */ if (!bfd_malloc_and_get_section (sub, subdeb, &debug_contents)) goto error_return; } } csectpp = xcoff_data (sub)->csects; lineno_counts = xcoff_data (sub)->lineno_counts; sym_hash = obj_xcoff_sym_hashes (sub); symesz = bfd_coff_symesz (sub); esym = (bfd_byte *) obj_coff_external_syms (sub); esymend = esym + symcount * symesz; while (esym < esymend) { struct internal_syment sym; union internal_auxent aux; asection *csect; const char *name; int keep_p; bfd_coff_swap_sym_in (sub, esym, &sym); /* Read in the csect information, if any. */ if (CSECT_SYM_P (sym.n_sclass)) { BFD_ASSERT (sym.n_numaux > 0); bfd_coff_swap_aux_in (sub, esym + symesz * sym.n_numaux, sym.n_type, sym.n_sclass, sym.n_numaux - 1, sym.n_numaux, &aux); } /* If this symbol's name is stored in the debug section, get a pointer to it. */ if (debug_contents != NULL && sym._n._n_n._n_zeroes == 0 && bfd_coff_symname_in_debug (sub, &sym)) name = (const char *) debug_contents + sym._n._n_n._n_offset; else name = NULL; /* Decide whether to copy this symbol to the output file. */ csect = *csectpp; keep_p = xcoff_keep_symbol_p (info, sub, &sym, &aux, *sym_hash, csect, name); if (keep_p < 0) return FALSE; if (!keep_p) /* Use a debug_index of -2 to record that a symbol should be stripped. */ *debug_index = -2; else { /* See whether we should store the symbol name in the output .debug section. */ if (name != NULL) { bfd_size_type indx; indx = _bfd_stringtab_add (debug_strtab, name, TRUE, TRUE); if (indx == (bfd_size_type) -1) goto error_return; *debug_index = indx; } else *debug_index = -1; if (*sym_hash != 0) (*sym_hash)->flags |= XCOFF_ALLOCATED; if (*lineno_counts > 0) csect->output_section->lineno_count += *lineno_counts; } esym += (sym.n_numaux + 1) * symesz; csectpp += sym.n_numaux + 1; sym_hash += sym.n_numaux + 1; lineno_counts += sym.n_numaux + 1; debug_index += sym.n_numaux + 1; } if (debug_contents) { free (debug_contents); debug_contents = NULL; /* Clear the size of subdeb, so that it is not included directly in the output file. */ subdeb->size = 0; } if (! info->keep_memory) { if (! _bfd_coff_free_symbols (sub)) goto error_return; } } if (info->strip != strip_all) xcoff_hash_table (info)->debug_section->size = _bfd_stringtab_size (debug_strtab); return TRUE; error_return: if (ldinfo.strings != NULL) free (ldinfo.strings); if (debug_contents != NULL) free (debug_contents); return FALSE; } bfd_boolean bfd_xcoff_link_generate_rtinit (bfd *abfd, const char *init, const char *fini, bfd_boolean rtld) { struct bfd_in_memory *bim; bim = bfd_malloc ((bfd_size_type) sizeof (* bim)); if (bim == NULL) return FALSE; bim->size = 0; bim->buffer = 0; abfd->link_next = 0; abfd->format = bfd_object; abfd->iostream = (void *) bim; abfd->flags = BFD_IN_MEMORY; abfd->iovec = &_bfd_memory_iovec; abfd->direction = write_direction; abfd->origin = 0; abfd->where = 0; if (! bfd_xcoff_generate_rtinit (abfd, init, fini, rtld)) return FALSE; /* need to reset to unknown or it will not be read back in correctly */ abfd->format = bfd_unknown; abfd->direction = read_direction; abfd->where = 0; return TRUE; } /* Return the section that defines H. Return null if no section does. */ static asection * xcoff_symbol_section (struct xcoff_link_hash_entry *h) { switch (h->root.type) { case bfd_link_hash_defined: case bfd_link_hash_defweak: return h->root.u.def.section; case bfd_link_hash_common: return h->root.u.c.p->section; default: return NULL; } } /* Add a .loader relocation for input relocation IREL. If the loader relocation should be against an output section, HSEC points to the input section that IREL is against, otherwise HSEC is null. H is the symbol that IREL is against, or null if it isn't against a global symbol. REFERENCE_BFD is the bfd to use in error messages about the relocation. */ static bfd_boolean xcoff_create_ldrel (bfd *output_bfd, struct xcoff_final_link_info *flinfo, asection *output_section, bfd *reference_bfd, struct internal_reloc *irel, asection *hsec, struct xcoff_link_hash_entry *h) { struct internal_ldrel ldrel; ldrel.l_vaddr = irel->r_vaddr; if (hsec != NULL) { const char *secname; secname = hsec->output_section->name; if (strcmp (secname, ".text") == 0) ldrel.l_symndx = 0; else if (strcmp (secname, ".data") == 0) ldrel.l_symndx = 1; else if (strcmp (secname, ".bss") == 0) ldrel.l_symndx = 2; else { (*_bfd_error_handler) (_("%B: loader reloc in unrecognized section `%s'"), reference_bfd, secname); bfd_set_error (bfd_error_nonrepresentable_section); return FALSE; } } else if (h != NULL) { if (h->ldindx < 0) { (*_bfd_error_handler) (_("%B: `%s' in loader reloc but not loader sym"), reference_bfd, h->root.root.string); bfd_set_error (bfd_error_bad_value); return FALSE; } ldrel.l_symndx = h->ldindx; } else ldrel.l_symndx = -(bfd_size_type) 1; ldrel.l_rtype = (irel->r_size << 8) | irel->r_type; ldrel.l_rsecnm = output_section->target_index; if (xcoff_hash_table (flinfo->info)->textro && strcmp (output_section->name, ".text") == 0) { (*_bfd_error_handler) (_("%B: loader reloc in read-only section %A"), reference_bfd, output_section); bfd_set_error (bfd_error_invalid_operation); return FALSE; } bfd_xcoff_swap_ldrel_out (output_bfd, &ldrel, flinfo->ldrel); flinfo->ldrel += bfd_xcoff_ldrelsz (output_bfd); return TRUE; } /* Link an input file into the linker output file. This function handles all the sections and relocations of the input file at once. */ static bfd_boolean xcoff_link_input_bfd (struct xcoff_final_link_info *flinfo, bfd *input_bfd) { bfd *output_bfd; const char *strings; bfd_size_type syment_base; unsigned int n_tmask; unsigned int n_btshft; bfd_boolean copy, hash; bfd_size_type isymesz; bfd_size_type osymesz; bfd_size_type linesz; bfd_byte *esym; bfd_byte *esym_end; struct xcoff_link_hash_entry **sym_hash; struct internal_syment *isymp; asection **csectpp; unsigned int *lineno_counts; long *debug_index; long *indexp; unsigned long output_index; bfd_byte *outsym; unsigned int incls; asection *oline; bfd_boolean keep_syms; asection *o; /* We can just skip DYNAMIC files, unless this is a static link. */ if ((input_bfd->flags & DYNAMIC) != 0 && ! flinfo->info->static_link) return TRUE; /* Move all the symbols to the output file. */ output_bfd = flinfo->output_bfd; strings = NULL; syment_base = obj_raw_syment_count (output_bfd); isymesz = bfd_coff_symesz (input_bfd); osymesz = bfd_coff_symesz (output_bfd); linesz = bfd_coff_linesz (input_bfd); BFD_ASSERT (linesz == bfd_coff_linesz (output_bfd)); n_tmask = coff_data (input_bfd)->local_n_tmask; n_btshft = coff_data (input_bfd)->local_n_btshft; /* Define macros so that ISFCN, et. al., macros work correctly. */ #define N_TMASK n_tmask #define N_BTSHFT n_btshft copy = FALSE; if (! flinfo->info->keep_memory) copy = TRUE; hash = TRUE; if ((output_bfd->flags & BFD_TRADITIONAL_FORMAT) != 0) hash = FALSE; if (! _bfd_coff_get_external_symbols (input_bfd)) return FALSE; /* Make one pass over the symbols and assign indices to symbols that we have decided to keep. Also use create .loader symbol information and update information in hash table entries. */ esym = (bfd_byte *) obj_coff_external_syms (input_bfd); esym_end = esym + obj_raw_syment_count (input_bfd) * isymesz; sym_hash = obj_xcoff_sym_hashes (input_bfd); csectpp = xcoff_data (input_bfd)->csects; debug_index = xcoff_data (input_bfd)->debug_indices; isymp = flinfo->internal_syms; indexp = flinfo->sym_indices; output_index = syment_base; while (esym < esym_end) { union internal_auxent aux; int smtyp = 0; int add; bfd_coff_swap_sym_in (input_bfd, (void *) esym, (void *) isymp); /* Read in the csect information, if any. */ if (CSECT_SYM_P (isymp->n_sclass)) { BFD_ASSERT (isymp->n_numaux > 0); bfd_coff_swap_aux_in (input_bfd, (void *) (esym + isymesz * isymp->n_numaux), isymp->n_type, isymp->n_sclass, isymp->n_numaux - 1, isymp->n_numaux, (void *) &aux); smtyp = SMTYP_SMTYP (aux.x_csect.x_smtyp); } /* If this symbol is in the .loader section, swap out the .loader symbol information. If this is an external symbol reference to a defined symbol, though, then wait until we get to the definition. */ if (EXTERN_SYM_P (isymp->n_sclass) && *sym_hash != NULL && (*sym_hash)->ldsym != NULL && xcoff_final_definition_p (input_bfd, *sym_hash, *csectpp)) { struct xcoff_link_hash_entry *h; struct internal_ldsym *ldsym; h = *sym_hash; ldsym = h->ldsym; if (isymp->n_scnum > 0) { ldsym->l_scnum = (*csectpp)->output_section->target_index; ldsym->l_value = (isymp->n_value + (*csectpp)->output_section->vma + (*csectpp)->output_offset - (*csectpp)->vma); } else { ldsym->l_scnum = isymp->n_scnum; ldsym->l_value = isymp->n_value; } ldsym->l_smtype = smtyp; if (((h->flags & XCOFF_DEF_REGULAR) == 0 && (h->flags & XCOFF_DEF_DYNAMIC) != 0) || (h->flags & XCOFF_IMPORT) != 0) ldsym->l_smtype |= L_IMPORT; if (((h->flags & XCOFF_DEF_REGULAR) != 0 && (h->flags & XCOFF_DEF_DYNAMIC) != 0) || (h->flags & XCOFF_EXPORT) != 0) ldsym->l_smtype |= L_EXPORT; if ((h->flags & XCOFF_ENTRY) != 0) ldsym->l_smtype |= L_ENTRY; if (isymp->n_sclass == C_AIX_WEAKEXT) ldsym->l_smtype |= L_WEAK; ldsym->l_smclas = aux.x_csect.x_smclas; if (ldsym->l_ifile == (bfd_size_type) -1) ldsym->l_ifile = 0; else if (ldsym->l_ifile == 0) { if ((ldsym->l_smtype & L_IMPORT) == 0) ldsym->l_ifile = 0; else { bfd *impbfd; if (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) impbfd = h->root.u.def.section->owner; else if (h->root.type == bfd_link_hash_undefined || h->root.type == bfd_link_hash_undefweak) impbfd = h->root.u.undef.abfd; else impbfd = NULL; if (impbfd == NULL) ldsym->l_ifile = 0; else { BFD_ASSERT (impbfd->xvec == flinfo->output_bfd->xvec); ldsym->l_ifile = xcoff_data (impbfd)->import_file_id; } } } ldsym->l_parm = 0; BFD_ASSERT (h->ldindx >= 0); bfd_xcoff_swap_ldsym_out (flinfo->output_bfd, ldsym, (flinfo->ldsym + ((h->ldindx - 3) * bfd_xcoff_ldsymsz (flinfo->output_bfd)))); h->ldsym = NULL; /* Fill in snentry now that we know the target_index. */ if ((h->flags & XCOFF_ENTRY) != 0 && (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak)) { xcoff_data (output_bfd)->snentry = h->root.u.def.section->output_section->target_index; } } add = 1 + isymp->n_numaux; if (*debug_index == -2) /* We've decided to strip this symbol. */ *indexp = -1; else { /* Assign the next unused index to this symbol. */ *indexp = output_index; if (EXTERN_SYM_P (isymp->n_sclass)) { BFD_ASSERT (*sym_hash != NULL); (*sym_hash)->indx = output_index; } /* If this is a symbol in the TOC which we may have merged (class XMC_TC), remember the symbol index of the TOC symbol. */ if (isymp->n_sclass == C_HIDEXT && aux.x_csect.x_smclas == XMC_TC && *sym_hash != NULL) { BFD_ASSERT (((*sym_hash)->flags & XCOFF_SET_TOC) == 0); BFD_ASSERT ((*sym_hash)->toc_section != NULL); (*sym_hash)->u.toc_indx = output_index; } output_index += add; } esym += add * isymesz; isymp += add; csectpp += add; sym_hash += add; debug_index += add; ++indexp; for (--add; add > 0; --add) *indexp++ = -1; } /* Now write out the symbols that we decided to keep. */ esym = (bfd_byte *) obj_coff_external_syms (input_bfd); esym_end = esym + obj_raw_syment_count (input_bfd) * isymesz; sym_hash = obj_xcoff_sym_hashes (input_bfd); isymp = flinfo->internal_syms; indexp = flinfo->sym_indices; csectpp = xcoff_data (input_bfd)->csects; lineno_counts = xcoff_data (input_bfd)->lineno_counts; debug_index = xcoff_data (input_bfd)->debug_indices; outsym = flinfo->outsyms; incls = 0; oline = NULL; while (esym < esym_end) { int add; add = 1 + isymp->n_numaux; if (*indexp < 0) esym += add * isymesz; else { struct internal_syment isym; int i; /* Adjust the symbol in order to output it. */ isym = *isymp; if (isym._n._n_n._n_zeroes == 0 && isym._n._n_n._n_offset != 0) { /* This symbol has a long name. Enter it in the string table we are building. If *debug_index != -1, the name has already been entered in the .debug section. */ if (*debug_index >= 0) isym._n._n_n._n_offset = *debug_index; else { const char *name; bfd_size_type indx; name = _bfd_coff_internal_syment_name (input_bfd, &isym, NULL); if (name == NULL) return FALSE; indx = _bfd_stringtab_add (flinfo->strtab, name, hash, copy); if (indx == (bfd_size_type) -1) return FALSE; isym._n._n_n._n_offset = STRING_SIZE_SIZE + indx; } } /* Make __rtinit C_HIDEXT rather than C_EXT. This avoids multiple definition problems when linking a shared object statically. (The native linker doesn't enter __rtinit into the normal table at all, but having a local symbol can make the objdump output easier to read.) */ if (isym.n_sclass == C_EXT && *sym_hash && ((*sym_hash)->flags & XCOFF_RTINIT) != 0) isym.n_sclass = C_HIDEXT; /* The value of a C_FILE symbol is the symbol index of the next C_FILE symbol. The value of the last C_FILE symbol is -1. We try to get this right, below, just before we write the symbols out, but in the general case we may have to write the symbol out twice. */ if (isym.n_sclass == C_FILE) { if (flinfo->last_file_index != -1 && flinfo->last_file.n_value != (bfd_vma) *indexp) { /* We must correct the value of the last C_FILE entry. */ flinfo->last_file.n_value = *indexp; if ((bfd_size_type) flinfo->last_file_index >= syment_base) { /* The last C_FILE symbol is in this input file. */ bfd_coff_swap_sym_out (output_bfd, (void *) &flinfo->last_file, (void *) (flinfo->outsyms + ((flinfo->last_file_index - syment_base) * osymesz))); } else { /* We have already written out the last C_FILE symbol. We need to write it out again. We borrow *outsym temporarily. */ file_ptr pos; bfd_coff_swap_sym_out (output_bfd, (void *) &flinfo->last_file, (void *) outsym); pos = obj_sym_filepos (output_bfd); pos += flinfo->last_file_index * osymesz; if (bfd_seek (output_bfd, pos, SEEK_SET) != 0 || (bfd_bwrite (outsym, osymesz, output_bfd) != osymesz)) return FALSE; } } flinfo->last_file_index = *indexp; flinfo->last_file = isym; } /* The value of a C_BINCL or C_EINCL symbol is a file offset into the line numbers. We update the symbol values when we handle the line numbers. */ if (isym.n_sclass == C_BINCL || isym.n_sclass == C_EINCL) { isym.n_value = flinfo->line_filepos; ++incls; } /* The value of a C_BSTAT symbol is the symbol table index of the containing csect. */ else if (isym.n_sclass == C_BSTAT) { bfd_vma indx; indx = isym.n_value; if (indx < obj_raw_syment_count (input_bfd)) { long symindx; symindx = flinfo->sym_indices[indx]; if (symindx < 0) isym.n_value = 0; else isym.n_value = symindx; } } else if (isym.n_sclass != C_ESTAT && isym.n_sclass != C_DECL && isym.n_scnum > 0) { isym.n_scnum = (*csectpp)->output_section->target_index; isym.n_value += ((*csectpp)->output_section->vma + (*csectpp)->output_offset - (*csectpp)->vma); } /* Output the symbol. */ bfd_coff_swap_sym_out (output_bfd, (void *) &isym, (void *) outsym); esym += isymesz; outsym += osymesz; for (i = 0; i < isymp->n_numaux && esym < esym_end; i++) { union internal_auxent aux; bfd_coff_swap_aux_in (input_bfd, (void *) esym, isymp->n_type, isymp->n_sclass, i, isymp->n_numaux, (void *) &aux); if (isymp->n_sclass == C_FILE) { /* This is the file name (or some comment put in by the compiler). If it is long, we must put it in the string table. */ if (aux.x_file.x_n.x_zeroes == 0 && aux.x_file.x_n.x_offset != 0) { const char *filename; bfd_size_type indx; BFD_ASSERT (aux.x_file.x_n.x_offset >= STRING_SIZE_SIZE); if (strings == NULL) { strings = _bfd_coff_read_string_table (input_bfd); if (strings == NULL) return FALSE; } filename = strings + aux.x_file.x_n.x_offset; indx = _bfd_stringtab_add (flinfo->strtab, filename, hash, copy); if (indx == (bfd_size_type) -1) return FALSE; aux.x_file.x_n.x_offset = STRING_SIZE_SIZE + indx; } } else if (CSECT_SYM_P (isymp->n_sclass) && i + 1 == isymp->n_numaux) { /* We don't support type checking. I don't know if anybody does. */ aux.x_csect.x_parmhash = 0; /* I don't think anybody uses these fields, but we'd better clobber them just in case. */ aux.x_csect.x_stab = 0; aux.x_csect.x_snstab = 0; if (SMTYP_SMTYP (aux.x_csect.x_smtyp) == XTY_LD) { unsigned long indx; indx = aux.x_csect.x_scnlen.l; if (indx < obj_raw_syment_count (input_bfd)) { long symindx; symindx = flinfo->sym_indices[indx]; if (symindx < 0) { aux.x_csect.x_scnlen.l = 0; } else { aux.x_csect.x_scnlen.l = symindx; } } } } else if (isymp->n_sclass != C_STAT || isymp->n_type != T_NULL) { unsigned long indx; if (ISFCN (isymp->n_type) || ISTAG (isymp->n_sclass) || isymp->n_sclass == C_BLOCK || isymp->n_sclass == C_FCN) { indx = aux.x_sym.x_fcnary.x_fcn.x_endndx.l; if (indx > 0 && indx < obj_raw_syment_count (input_bfd)) { /* We look forward through the symbol for the index of the next symbol we are going to include. I don't know if this is entirely right. */ while (flinfo->sym_indices[indx] < 0 && indx < obj_raw_syment_count (input_bfd)) ++indx; if (indx >= obj_raw_syment_count (input_bfd)) indx = output_index; else indx = flinfo->sym_indices[indx]; aux.x_sym.x_fcnary.x_fcn.x_endndx.l = indx; } } indx = aux.x_sym.x_tagndx.l; if (indx > 0 && indx < obj_raw_syment_count (input_bfd)) { long symindx; symindx = flinfo->sym_indices[indx]; if (symindx < 0) aux.x_sym.x_tagndx.l = 0; else aux.x_sym.x_tagndx.l = symindx; } } /* Copy over the line numbers, unless we are stripping them. We do this on a symbol by symbol basis in order to more easily handle garbage collection. */ if (CSECT_SYM_P (isymp->n_sclass) && i == 0 && isymp->n_numaux > 1 && ISFCN (isymp->n_type) && aux.x_sym.x_fcnary.x_fcn.x_lnnoptr != 0) { if (*lineno_counts == 0) aux.x_sym.x_fcnary.x_fcn.x_lnnoptr = 0; else { asection *enclosing; unsigned int enc_count; bfd_signed_vma linoff; struct internal_lineno lin; bfd_byte *linp; bfd_byte *linpend; bfd_vma offset; file_ptr pos; bfd_size_type amt; /* Read in the enclosing section's line-number information, if we haven't already. */ o = *csectpp; enclosing = xcoff_section_data (abfd, o)->enclosing; enc_count = xcoff_section_data (abfd, o)->lineno_count; if (oline != enclosing) { pos = enclosing->line_filepos; amt = linesz * enc_count; if (bfd_seek (input_bfd, pos, SEEK_SET) != 0 || (bfd_bread (flinfo->linenos, amt, input_bfd) != amt)) return FALSE; oline = enclosing; } /* Copy across the first entry, adjusting its symbol index. */ linoff = (aux.x_sym.x_fcnary.x_fcn.x_lnnoptr - enclosing->line_filepos); linp = flinfo->linenos + linoff; bfd_coff_swap_lineno_in (input_bfd, linp, &lin); lin.l_addr.l_symndx = *indexp; bfd_coff_swap_lineno_out (output_bfd, &lin, linp); /* Copy the other entries, adjusting their addresses. */ linpend = linp + *lineno_counts * linesz; offset = (o->output_section->vma + o->output_offset - o->vma); for (linp += linesz; linp < linpend; linp += linesz) { bfd_coff_swap_lineno_in (input_bfd, linp, &lin); lin.l_addr.l_paddr += offset; bfd_coff_swap_lineno_out (output_bfd, &lin, linp); } /* Write out the entries we've just processed. */ pos = (o->output_section->line_filepos + o->output_section->lineno_count * linesz); amt = linesz * *lineno_counts; if (bfd_seek (output_bfd, pos, SEEK_SET) != 0 || bfd_bwrite (flinfo->linenos + linoff, amt, output_bfd) != amt) return FALSE; o->output_section->lineno_count += *lineno_counts; /* Record the offset of the symbol's line numbers in the output file. */ aux.x_sym.x_fcnary.x_fcn.x_lnnoptr = pos; if (incls > 0) { struct internal_syment *iisp, *iispend; long *iindp; bfd_byte *oos; bfd_vma range_start, range_end; int iiadd; /* Update any C_BINCL or C_EINCL symbols that refer to a line number in the range we just output. */ iisp = flinfo->internal_syms; iispend = iisp + obj_raw_syment_count (input_bfd); iindp = flinfo->sym_indices; oos = flinfo->outsyms; range_start = enclosing->line_filepos + linoff; range_end = range_start + *lineno_counts * linesz; while (iisp < iispend) { if (*iindp >= 0 && (iisp->n_sclass == C_BINCL || iisp->n_sclass == C_EINCL) && iisp->n_value >= range_start && iisp->n_value < range_end) { struct internal_syment iis; bfd_coff_swap_sym_in (output_bfd, oos, &iis); iis.n_value = (iisp->n_value - range_start + pos); bfd_coff_swap_sym_out (output_bfd, &iis, oos); --incls; } iiadd = 1 + iisp->n_numaux; if (*iindp >= 0) oos += iiadd * osymesz; iisp += iiadd; iindp += iiadd; } } } } bfd_coff_swap_aux_out (output_bfd, (void *) &aux, isymp->n_type, isymp->n_sclass, i, isymp->n_numaux, (void *) outsym); outsym += osymesz; esym += isymesz; } } sym_hash += add; indexp += add; isymp += add; csectpp += add; lineno_counts += add; debug_index += add; } /* If we swapped out a C_FILE symbol, guess that the next C_FILE symbol will be the first symbol in the next input file. In the normal case, this will save us from writing out the C_FILE symbol again. */ if (flinfo->last_file_index != -1 && (bfd_size_type) flinfo->last_file_index >= syment_base) { flinfo->last_file.n_value = output_index; bfd_coff_swap_sym_out (output_bfd, (void *) &flinfo->last_file, (void *) (flinfo->outsyms + ((flinfo->last_file_index - syment_base) * osymesz))); } /* Write the modified symbols to the output file. */ if (outsym > flinfo->outsyms) { file_ptr pos = obj_sym_filepos (output_bfd) + syment_base * osymesz; bfd_size_type amt = outsym - flinfo->outsyms; if (bfd_seek (output_bfd, pos, SEEK_SET) != 0 || bfd_bwrite (flinfo->outsyms, amt, output_bfd) != amt) return FALSE; BFD_ASSERT ((obj_raw_syment_count (output_bfd) + (outsym - flinfo->outsyms) / osymesz) == output_index); obj_raw_syment_count (output_bfd) = output_index; } /* Don't let the linker relocation routines discard the symbols. */ keep_syms = obj_coff_keep_syms (input_bfd); obj_coff_keep_syms (input_bfd) = TRUE; /* Relocate the contents of each section. */ for (o = input_bfd->sections; o != NULL; o = o->next) { bfd_byte *contents; if (! o->linker_mark) /* This section was omitted from the link. */ continue; if ((o->flags & SEC_HAS_CONTENTS) == 0 || o->size == 0 || (o->flags & SEC_IN_MEMORY) != 0) continue; /* We have set filepos correctly for the sections we created to represent csects, so bfd_get_section_contents should work. */ if (coff_section_data (input_bfd, o) != NULL && coff_section_data (input_bfd, o)->contents != NULL) contents = coff_section_data (input_bfd, o)->contents; else { bfd_size_type sz = o->rawsize ? o->rawsize : o->size; if (!bfd_get_section_contents (input_bfd, o, flinfo->contents, 0, sz)) return FALSE; contents = flinfo->contents; } if ((o->flags & SEC_RELOC) != 0) { int target_index; struct internal_reloc *internal_relocs; struct internal_reloc *irel; bfd_vma offset; struct internal_reloc *irelend; struct xcoff_link_hash_entry **rel_hash; long r_symndx; /* Read in the relocs. */ target_index = o->output_section->target_index; internal_relocs = (xcoff_read_internal_relocs (input_bfd, o, FALSE, flinfo->external_relocs, TRUE, (flinfo->section_info[target_index].relocs + o->output_section->reloc_count))); if (internal_relocs == NULL) return FALSE; /* Call processor specific code to relocate the section contents. */ if (! bfd_coff_relocate_section (output_bfd, flinfo->info, input_bfd, o, contents, internal_relocs, flinfo->internal_syms, xcoff_data (input_bfd)->csects)) return FALSE; offset = o->output_section->vma + o->output_offset - o->vma; irel = internal_relocs; irelend = irel + o->reloc_count; rel_hash = (flinfo->section_info[target_index].rel_hashes + o->output_section->reloc_count); for (; irel < irelend; irel++, rel_hash++) { struct xcoff_link_hash_entry *h = NULL; *rel_hash = NULL; /* Adjust the reloc address and symbol index. */ irel->r_vaddr += offset; r_symndx = irel->r_symndx; if (r_symndx == -1) h = NULL; else h = obj_xcoff_sym_hashes (input_bfd)[r_symndx]; if (r_symndx != -1 && flinfo->info->strip != strip_all) { if (h != NULL && h->smclas != XMC_TD && (irel->r_type == R_TOC || irel->r_type == R_GL || irel->r_type == R_TCL || irel->r_type == R_TRL || irel->r_type == R_TRLA)) { /* This is a TOC relative reloc with a symbol attached. The symbol should be the one which this reloc is for. We want to make this reloc against the TOC address of the symbol, not the symbol itself. */ BFD_ASSERT (h->toc_section != NULL); BFD_ASSERT ((h->flags & XCOFF_SET_TOC) == 0); if (h->u.toc_indx != -1) irel->r_symndx = h->u.toc_indx; else { struct xcoff_toc_rel_hash *n; struct xcoff_link_section_info *si; bfd_size_type amt; amt = sizeof (* n); n = bfd_alloc (flinfo->output_bfd, amt); if (n == NULL) return FALSE; si = flinfo->section_info + target_index; n->next = si->toc_rel_hashes; n->h = h; n->rel = irel; si->toc_rel_hashes = n; } } else if (h != NULL) { /* This is a global symbol. */ if (h->indx >= 0) irel->r_symndx = h->indx; else { /* This symbol is being written at the end of the file, and we do not yet know the symbol index. We save the pointer to the hash table entry in the rel_hash list. We set the indx field to -2 to indicate that this symbol must not be stripped. */ *rel_hash = h; h->indx = -2; } } else { long indx; indx = flinfo->sym_indices[r_symndx]; if (indx == -1) { struct internal_syment *is; /* Relocations against a TC0 TOC anchor are automatically transformed to be against the TOC anchor in the output file. */ is = flinfo->internal_syms + r_symndx; if (is->n_sclass == C_HIDEXT && is->n_numaux > 0) { void * auxptr; union internal_auxent aux; auxptr = ((void *) (((bfd_byte *) obj_coff_external_syms (input_bfd)) + ((r_symndx + is->n_numaux) * isymesz))); bfd_coff_swap_aux_in (input_bfd, auxptr, is->n_type, is->n_sclass, is->n_numaux - 1, is->n_numaux, (void *) &aux); if (SMTYP_SMTYP (aux.x_csect.x_smtyp) == XTY_SD && aux.x_csect.x_smclas == XMC_TC0) indx = flinfo->toc_symindx; } } if (indx != -1) irel->r_symndx = indx; else { struct internal_syment *is; const char *name; char buf[SYMNMLEN + 1]; /* This reloc is against a symbol we are stripping. It would be possible to handle this case, but I don't think it's worth it. */ is = flinfo->internal_syms + r_symndx; if (is->n_sclass != C_DWARF) { name = (_bfd_coff_internal_syment_name (input_bfd, is, buf)); if (name == NULL) return FALSE; if (!(*flinfo->info->callbacks->unattached_reloc) (flinfo->info, name, input_bfd, o, irel->r_vaddr)) return FALSE; } } } } if ((o->flags & SEC_DEBUGGING) == 0 && xcoff_need_ldrel_p (flinfo->info, irel, h)) { asection *sec; if (r_symndx == -1) sec = NULL; else if (h == NULL) sec = xcoff_data (input_bfd)->csects[r_symndx]; else sec = xcoff_symbol_section (h); if (!xcoff_create_ldrel (output_bfd, flinfo, o->output_section, input_bfd, irel, sec, h)) return FALSE; } } o->output_section->reloc_count += o->reloc_count; } /* Write out the modified section contents. */ if (! bfd_set_section_contents (output_bfd, o->output_section, contents, (file_ptr) o->output_offset, o->size)) return FALSE; } obj_coff_keep_syms (input_bfd) = keep_syms; if (! flinfo->info->keep_memory) { if (! _bfd_coff_free_symbols (input_bfd)) return FALSE; } return TRUE; } #undef N_TMASK #undef N_BTSHFT /* Sort relocs by VMA. This is called via qsort. */ static int xcoff_sort_relocs (const void * p1, const void * p2) { const struct internal_reloc *r1 = (const struct internal_reloc *) p1; const struct internal_reloc *r2 = (const struct internal_reloc *) p2; if (r1->r_vaddr > r2->r_vaddr) return 1; else if (r1->r_vaddr < r2->r_vaddr) return -1; else return 0; } /* Return true if section SEC is a TOC section. */ static inline bfd_boolean xcoff_toc_section_p (asection *sec) { const char *name; name = sec->name; if (name[0] == '.' && name[1] == 't') { if (name[2] == 'c') { if (name[3] == '0' && name[4] == 0) return TRUE; if (name[3] == 0) return TRUE; } if (name[2] == 'd' && name[3] == 0) return TRUE; } return FALSE; } /* See if the link requires a TOC (it usually does!). If so, find a good place to put the TOC anchor csect, and write out the associated symbol. */ static bfd_boolean xcoff_find_tc0 (bfd *output_bfd, struct xcoff_final_link_info *flinfo) { bfd_vma toc_start, toc_end, start, end, best_address; asection *sec; bfd *input_bfd; int section_index; struct internal_syment irsym; union internal_auxent iraux; file_ptr pos; size_t size; /* Set [TOC_START, TOC_END) to the range of the TOC. Record the index of a csect at the beginning of the TOC. */ toc_start = ~(bfd_vma) 0; toc_end = 0; section_index = -1; for (input_bfd = flinfo->info->input_bfds; input_bfd != NULL; input_bfd = input_bfd->link_next) for (sec = input_bfd->sections; sec != NULL; sec = sec->next) if ((sec->flags & SEC_MARK) != 0 && xcoff_toc_section_p (sec)) { start = sec->output_section->vma + sec->output_offset; if (toc_start > start) { toc_start = start; section_index = sec->output_section->target_index; } end = start + sec->size; if (toc_end < end) toc_end = end; } /* There's no need for a TC0 symbol if we don't have a TOC. */ if (toc_end < toc_start) { xcoff_data (output_bfd)->toc = toc_start; return TRUE; } if (toc_end - toc_start < 0x8000) /* Every TOC csect can be accessed from TOC_START. */ best_address = toc_start; else { /* Find the lowest TOC csect that is still within range of TOC_END. */ best_address = toc_end; for (input_bfd = flinfo->info->input_bfds; input_bfd != NULL; input_bfd = input_bfd->link_next) for (sec = input_bfd->sections; sec != NULL; sec = sec->next) if ((sec->flags & SEC_MARK) != 0 && xcoff_toc_section_p (sec)) { start = sec->output_section->vma + sec->output_offset; if (start < best_address && start + 0x8000 >= toc_end) { best_address = start; section_index = sec->output_section->target_index; } } /* Make sure that the start of the TOC is also within range. */ if (best_address > toc_start + 0x8000) { (*_bfd_error_handler) (_("TOC overflow: 0x%lx > 0x10000; try -mminimal-toc " "when compiling"), (unsigned long) (toc_end - toc_start)); bfd_set_error (bfd_error_file_too_big); return FALSE; } } /* Record the chosen TOC value. */ flinfo->toc_symindx = obj_raw_syment_count (output_bfd); xcoff_data (output_bfd)->toc = best_address; xcoff_data (output_bfd)->sntoc = section_index; /* Fill out the TC0 symbol. */ if (!bfd_xcoff_put_symbol_name (output_bfd, flinfo->strtab, &irsym, "TOC")) return FALSE; irsym.n_value = best_address; irsym.n_scnum = section_index; irsym.n_sclass = C_HIDEXT; irsym.n_type = T_NULL; irsym.n_numaux = 1; bfd_coff_swap_sym_out (output_bfd, &irsym, flinfo->outsyms); /* Fill out the auxillary csect information. */ memset (&iraux, 0, sizeof iraux); iraux.x_csect.x_smtyp = XTY_SD; iraux.x_csect.x_smclas = XMC_TC0; iraux.x_csect.x_scnlen.l = 0; bfd_coff_swap_aux_out (output_bfd, &iraux, T_NULL, C_HIDEXT, 0, 1, flinfo->outsyms + bfd_coff_symesz (output_bfd)); /* Write the contents to the file. */ pos = obj_sym_filepos (output_bfd); pos += obj_raw_syment_count (output_bfd) * bfd_coff_symesz (output_bfd); size = 2 * bfd_coff_symesz (output_bfd); if (bfd_seek (output_bfd, pos, SEEK_SET) != 0 || bfd_bwrite (flinfo->outsyms, size, output_bfd) != size) return FALSE; obj_raw_syment_count (output_bfd) += 2; return TRUE; } /* Write out a non-XCOFF global symbol. */ static bfd_boolean xcoff_write_global_symbol (struct xcoff_link_hash_entry *h, void * inf) { struct xcoff_final_link_info *flinfo = (struct xcoff_final_link_info *) inf; bfd *output_bfd; bfd_byte *outsym; struct internal_syment isym; union internal_auxent aux; bfd_boolean result; file_ptr pos; bfd_size_type amt; output_bfd = flinfo->output_bfd; outsym = flinfo->outsyms; if (h->root.type == bfd_link_hash_warning) { h = (struct xcoff_link_hash_entry *) h->root.u.i.link; if (h->root.type == bfd_link_hash_new) return TRUE; } /* If this symbol was garbage collected, just skip it. */ if (xcoff_hash_table (flinfo->info)->gc && (h->flags & XCOFF_MARK) == 0) return TRUE; /* If we need a .loader section entry, write it out. */ if (h->ldsym != NULL) { struct internal_ldsym *ldsym; bfd *impbfd; ldsym = h->ldsym; if (h->root.type == bfd_link_hash_undefined || h->root.type == bfd_link_hash_undefweak) { ldsym->l_value = 0; ldsym->l_scnum = N_UNDEF; ldsym->l_smtype = XTY_ER; impbfd = h->root.u.undef.abfd; } else if (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) { asection *sec; sec = h->root.u.def.section; ldsym->l_value = (sec->output_section->vma + sec->output_offset + h->root.u.def.value); ldsym->l_scnum = sec->output_section->target_index; ldsym->l_smtype = XTY_SD; impbfd = sec->owner; } else abort (); if (((h->flags & XCOFF_DEF_REGULAR) == 0 && (h->flags & XCOFF_DEF_DYNAMIC) != 0) || (h->flags & XCOFF_IMPORT) != 0) /* Clear l_smtype Import symbols are defined so the check above will make the l_smtype XTY_SD. But this is not correct, it should be cleared. */ ldsym->l_smtype |= L_IMPORT; if (((h->flags & XCOFF_DEF_REGULAR) != 0 && (h->flags & XCOFF_DEF_DYNAMIC) != 0) || (h->flags & XCOFF_EXPORT) != 0) ldsym->l_smtype |= L_EXPORT; if ((h->flags & XCOFF_ENTRY) != 0) ldsym->l_smtype |= L_ENTRY; if ((h->flags & XCOFF_RTINIT) != 0) ldsym->l_smtype = XTY_SD; ldsym->l_smclas = h->smclas; if (ldsym->l_smtype & L_IMPORT) { if ((h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) && (h->root.u.def.value != 0)) ldsym->l_smclas = XMC_XO; else if ((h->flags & (XCOFF_SYSCALL32 | XCOFF_SYSCALL64)) == (XCOFF_SYSCALL32 | XCOFF_SYSCALL64)) ldsym->l_smclas = XMC_SV3264; else if (h->flags & XCOFF_SYSCALL32) ldsym->l_smclas = XMC_SV; else if (h->flags & XCOFF_SYSCALL64) ldsym->l_smclas = XMC_SV64; } if (ldsym->l_ifile == -(bfd_size_type) 1) { ldsym->l_ifile = 0; } else if (ldsym->l_ifile == 0) { if ((ldsym->l_smtype & L_IMPORT) == 0) ldsym->l_ifile = 0; else if (impbfd == NULL) ldsym->l_ifile = 0; else { BFD_ASSERT (impbfd->xvec == output_bfd->xvec); ldsym->l_ifile = xcoff_data (impbfd)->import_file_id; } } ldsym->l_parm = 0; BFD_ASSERT (h->ldindx >= 0); bfd_xcoff_swap_ldsym_out (output_bfd, ldsym, (flinfo->ldsym + (h->ldindx - 3) * bfd_xcoff_ldsymsz(flinfo->output_bfd))); h->ldsym = NULL; } /* If this symbol needs global linkage code, write it out. */ if (h->root.type == bfd_link_hash_defined && (h->root.u.def.section == xcoff_hash_table (flinfo->info)->linkage_section)) { bfd_byte *p; bfd_vma tocoff; unsigned int i; p = h->root.u.def.section->contents + h->root.u.def.value; /* The first instruction in the global linkage code loads a specific TOC element. */ tocoff = (h->descriptor->toc_section->output_section->vma + h->descriptor->toc_section->output_offset - xcoff_data (output_bfd)->toc); if ((h->descriptor->flags & XCOFF_SET_TOC) != 0) tocoff += h->descriptor->u.toc_offset; /* The first instruction in the glink code needs to be cooked to to hold the correct offset in the toc. The rest are just output raw. */ bfd_put_32 (output_bfd, bfd_xcoff_glink_code(output_bfd, 0) | (tocoff & 0xffff), p); /* Start with i == 1 to get past the first instruction done above The /4 is because the glink code is in bytes and we are going 4 at a pop. */ for (i = 1; i < bfd_xcoff_glink_code_size(output_bfd) / 4; i++) bfd_put_32 (output_bfd, (bfd_vma) bfd_xcoff_glink_code(output_bfd, i), &p[4 * i]); } /* If we created a TOC entry for this symbol, write out the required relocs. */ if ((h->flags & XCOFF_SET_TOC) != 0) { asection *tocsec; asection *osec; int oindx; struct internal_reloc *irel; struct internal_syment irsym; union internal_auxent iraux; tocsec = h->toc_section; osec = tocsec->output_section; oindx = osec->target_index; irel = flinfo->section_info[oindx].relocs + osec->reloc_count; irel->r_vaddr = (osec->vma + tocsec->output_offset + h->u.toc_offset); if (h->indx >= 0) irel->r_symndx = h->indx; else { h->indx = -2; irel->r_symndx = obj_raw_syment_count (output_bfd); } BFD_ASSERT (h->ldindx >= 0); /* Initialize the aux union here instead of closer to when it is written out below because the length of the csect depends on whether the output is 32 or 64 bit. */ memset (&iraux, 0, sizeof iraux); iraux.x_csect.x_smtyp = XTY_SD; /* iraux.x_csect.x_scnlen.l = 4 or 8, see below. */ iraux.x_csect.x_smclas = XMC_TC; /* 32 bit uses a 32 bit R_POS to do the relocations 64 bit uses a 64 bit R_POS to do the relocations Also needs to change the csect size : 4 for 32 bit, 8 for 64 bit Which one is determined by the backend. */ if (bfd_xcoff_is_xcoff64 (output_bfd)) { irel->r_size = 63; iraux.x_csect.x_scnlen.l = 8; } else if (bfd_xcoff_is_xcoff32 (output_bfd)) { irel->r_size = 31; iraux.x_csect.x_scnlen.l = 4; } else return FALSE; irel->r_type = R_POS; flinfo->section_info[oindx].rel_hashes[osec->reloc_count] = NULL; ++osec->reloc_count; if (!xcoff_create_ldrel (output_bfd, flinfo, osec, output_bfd, irel, NULL, h)) return FALSE; /* We need to emit a symbol to define a csect which holds the reloc. */ if (flinfo->info->strip != strip_all) { result = bfd_xcoff_put_symbol_name (output_bfd, flinfo->strtab, &irsym, h->root.root.string); if (!result) return FALSE; irsym.n_value = irel->r_vaddr; irsym.n_scnum = osec->target_index; irsym.n_sclass = C_HIDEXT; irsym.n_type = T_NULL; irsym.n_numaux = 1; bfd_coff_swap_sym_out (output_bfd, (void *) &irsym, (void *) outsym); outsym += bfd_coff_symesz (output_bfd); /* Note : iraux is initialized above. */ bfd_coff_swap_aux_out (output_bfd, (void *) &iraux, T_NULL, C_HIDEXT, 0, 1, (void *) outsym); outsym += bfd_coff_auxesz (output_bfd); if (h->indx >= 0) { /* We aren't going to write out the symbols below, so we need to write them out now. */ pos = obj_sym_filepos (output_bfd); pos += (obj_raw_syment_count (output_bfd) * bfd_coff_symesz (output_bfd)); amt = outsym - flinfo->outsyms; if (bfd_seek (output_bfd, pos, SEEK_SET) != 0 || bfd_bwrite (flinfo->outsyms, amt, output_bfd) != amt) return FALSE; obj_raw_syment_count (output_bfd) += (outsym - flinfo->outsyms) / bfd_coff_symesz (output_bfd); outsym = flinfo->outsyms; } } } /* If this symbol is a specially defined function descriptor, write it out. The first word is the address of the function code itself, the second word is the address of the TOC, and the third word is zero. 32 bit vs 64 bit The addresses for the 32 bit will take 4 bytes and the addresses for 64 bit will take 8 bytes. Similar for the relocs. This type of logic was also done above to create a TOC entry in xcoff_write_global_symbol. */ if ((h->flags & XCOFF_DESCRIPTOR) != 0 && h->root.type == bfd_link_hash_defined && (h->root.u.def.section == xcoff_hash_table (flinfo->info)->descriptor_section)) { asection *sec; asection *osec; int oindx; bfd_byte *p; struct xcoff_link_hash_entry *hentry; asection *esec; struct internal_reloc *irel; asection *tsec; unsigned int reloc_size, byte_size; if (bfd_xcoff_is_xcoff64 (output_bfd)) { reloc_size = 63; byte_size = 8; } else if (bfd_xcoff_is_xcoff32 (output_bfd)) { reloc_size = 31; byte_size = 4; } else return FALSE; sec = h->root.u.def.section; osec = sec->output_section; oindx = osec->target_index; p = sec->contents + h->root.u.def.value; hentry = h->descriptor; BFD_ASSERT (hentry != NULL && (hentry->root.type == bfd_link_hash_defined || hentry->root.type == bfd_link_hash_defweak)); esec = hentry->root.u.def.section; irel = flinfo->section_info[oindx].relocs + osec->reloc_count; irel->r_vaddr = (osec->vma + sec->output_offset + h->root.u.def.value); irel->r_symndx = esec->output_section->target_index; irel->r_type = R_POS; irel->r_size = reloc_size; flinfo->section_info[oindx].rel_hashes[osec->reloc_count] = NULL; ++osec->reloc_count; if (!xcoff_create_ldrel (output_bfd, flinfo, osec, output_bfd, irel, esec, NULL)) return FALSE; /* There are three items to write out, the address of the code the address of the toc anchor the environment pointer. We are ignoring the environment pointer. So set it to zero. */ if (bfd_xcoff_is_xcoff64 (output_bfd)) { bfd_put_64 (output_bfd, (esec->output_section->vma + esec->output_offset + hentry->root.u.def.value), p); bfd_put_64 (output_bfd, xcoff_data (output_bfd)->toc, p + 8); bfd_put_64 (output_bfd, (bfd_vma) 0, p + 16); } else { /* 32 bit backend This logic was already called above so the error case where the backend is neither has already been checked. */ bfd_put_32 (output_bfd, (esec->output_section->vma + esec->output_offset + hentry->root.u.def.value), p); bfd_put_32 (output_bfd, xcoff_data (output_bfd)->toc, p + 4); bfd_put_32 (output_bfd, (bfd_vma) 0, p + 8); } tsec = coff_section_from_bfd_index (output_bfd, xcoff_data (output_bfd)->sntoc); ++irel; irel->r_vaddr = (osec->vma + sec->output_offset + h->root.u.def.value + byte_size); irel->r_symndx = tsec->output_section->target_index; irel->r_type = R_POS; irel->r_size = reloc_size; flinfo->section_info[oindx].rel_hashes[osec->reloc_count] = NULL; ++osec->reloc_count; if (!xcoff_create_ldrel (output_bfd, flinfo, osec, output_bfd, irel, tsec, NULL)) return FALSE; } if (h->indx >= 0 || flinfo->info->strip == strip_all) { BFD_ASSERT (outsym == flinfo->outsyms); return TRUE; } if (h->indx != -2 && (flinfo->info->strip == strip_all || (flinfo->info->strip == strip_some && bfd_hash_lookup (flinfo->info->keep_hash, h->root.root.string, FALSE, FALSE) == NULL))) { BFD_ASSERT (outsym == flinfo->outsyms); return TRUE; } if (h->indx != -2 && (h->flags & (XCOFF_REF_REGULAR | XCOFF_DEF_REGULAR)) == 0) { BFD_ASSERT (outsym == flinfo->outsyms); return TRUE; } memset (&aux, 0, sizeof aux); h->indx = obj_raw_syment_count (output_bfd); result = bfd_xcoff_put_symbol_name (output_bfd, flinfo->strtab, &isym, h->root.root.string); if (!result) return FALSE; if (h->root.type == bfd_link_hash_undefined || h->root.type == bfd_link_hash_undefweak) { isym.n_value = 0; isym.n_scnum = N_UNDEF; if (h->root.type == bfd_link_hash_undefweak && C_WEAKEXT == C_AIX_WEAKEXT) isym.n_sclass = C_WEAKEXT; else isym.n_sclass = C_EXT; aux.x_csect.x_smtyp = XTY_ER; } else if ((h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) && h->smclas == XMC_XO) { BFD_ASSERT (bfd_is_abs_section (h->root.u.def.section)); isym.n_value = h->root.u.def.value; isym.n_scnum = N_UNDEF; if (h->root.type == bfd_link_hash_undefweak && C_WEAKEXT == C_AIX_WEAKEXT) isym.n_sclass = C_WEAKEXT; else isym.n_sclass = C_EXT; aux.x_csect.x_smtyp = XTY_ER; } else if (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) { struct xcoff_link_size_list *l; isym.n_value = (h->root.u.def.section->output_section->vma + h->root.u.def.section->output_offset + h->root.u.def.value); if (bfd_is_abs_section (h->root.u.def.section->output_section)) isym.n_scnum = N_ABS; else isym.n_scnum = h->root.u.def.section->output_section->target_index; isym.n_sclass = C_HIDEXT; aux.x_csect.x_smtyp = XTY_SD; if ((h->flags & XCOFF_HAS_SIZE) != 0) { for (l = xcoff_hash_table (flinfo->info)->size_list; l != NULL; l = l->next) { if (l->h == h) { aux.x_csect.x_scnlen.l = l->size; break; } } } } else if (h->root.type == bfd_link_hash_common) { isym.n_value = (h->root.u.c.p->section->output_section->vma + h->root.u.c.p->section->output_offset); isym.n_scnum = h->root.u.c.p->section->output_section->target_index; isym.n_sclass = C_EXT; aux.x_csect.x_smtyp = XTY_CM; aux.x_csect.x_scnlen.l = h->root.u.c.size; } else abort (); isym.n_type = T_NULL; isym.n_numaux = 1; bfd_coff_swap_sym_out (output_bfd, (void *) &isym, (void *) outsym); outsym += bfd_coff_symesz (output_bfd); aux.x_csect.x_smclas = h->smclas; bfd_coff_swap_aux_out (output_bfd, (void *) &aux, T_NULL, isym.n_sclass, 0, 1, (void *) outsym); outsym += bfd_coff_auxesz (output_bfd); if ((h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) && h->smclas != XMC_XO) { /* We just output an SD symbol. Now output an LD symbol. */ h->indx += 2; if (h->root.type == bfd_link_hash_undefweak && C_WEAKEXT == C_AIX_WEAKEXT) isym.n_sclass = C_WEAKEXT; else isym.n_sclass = C_EXT; bfd_coff_swap_sym_out (output_bfd, (void *) &isym, (void *) outsym); outsym += bfd_coff_symesz (output_bfd); aux.x_csect.x_smtyp = XTY_LD; aux.x_csect.x_scnlen.l = obj_raw_syment_count (output_bfd); bfd_coff_swap_aux_out (output_bfd, (void *) &aux, T_NULL, C_EXT, 0, 1, (void *) outsym); outsym += bfd_coff_auxesz (output_bfd); } pos = obj_sym_filepos (output_bfd); pos += obj_raw_syment_count (output_bfd) * bfd_coff_symesz (output_bfd); amt = outsym - flinfo->outsyms; if (bfd_seek (output_bfd, pos, SEEK_SET) != 0 || bfd_bwrite (flinfo->outsyms, amt, output_bfd) != amt) return FALSE; obj_raw_syment_count (output_bfd) += (outsym - flinfo->outsyms) / bfd_coff_symesz (output_bfd); return TRUE; } /* Handle a link order which is supposed to generate a reloc. */ static bfd_boolean xcoff_reloc_link_order (bfd *output_bfd, struct xcoff_final_link_info *flinfo, asection *output_section, struct bfd_link_order *link_order) { reloc_howto_type *howto; struct xcoff_link_hash_entry *h; asection *hsec; bfd_vma hval; bfd_vma addend; struct internal_reloc *irel; struct xcoff_link_hash_entry **rel_hash_ptr; if (link_order->type == bfd_section_reloc_link_order) /* We need to somehow locate a symbol in the right section. The symbol must either have a value of zero, or we must adjust the addend by the value of the symbol. FIXME: Write this when we need it. The old linker couldn't handle this anyhow. */ abort (); howto = bfd_reloc_type_lookup (output_bfd, link_order->u.reloc.p->reloc); if (howto == NULL) { bfd_set_error (bfd_error_bad_value); return FALSE; } h = ((struct xcoff_link_hash_entry *) bfd_wrapped_link_hash_lookup (output_bfd, flinfo->info, link_order->u.reloc.p->u.name, FALSE, FALSE, TRUE)); if (h == NULL) { if (! ((*flinfo->info->callbacks->unattached_reloc) (flinfo->info, link_order->u.reloc.p->u.name, NULL, NULL, (bfd_vma) 0))) return FALSE; return TRUE; } hsec = xcoff_symbol_section (h); if (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) hval = h->root.u.def.value; else hval = 0; addend = link_order->u.reloc.p->addend; if (hsec != NULL) addend += (hsec->output_section->vma + hsec->output_offset + hval); if (addend != 0) { bfd_size_type size; bfd_byte *buf; bfd_reloc_status_type rstat; bfd_boolean ok; size = bfd_get_reloc_size (howto); buf = bfd_zmalloc (size); if (buf == NULL) return FALSE; rstat = _bfd_relocate_contents (howto, output_bfd, addend, buf); switch (rstat) { case bfd_reloc_ok: break; default: case bfd_reloc_outofrange: abort (); case bfd_reloc_overflow: if (! ((*flinfo->info->callbacks->reloc_overflow) (flinfo->info, NULL, link_order->u.reloc.p->u.name, howto->name, addend, NULL, NULL, (bfd_vma) 0))) { free (buf); return FALSE; } break; } ok = bfd_set_section_contents (output_bfd, output_section, (void *) buf, (file_ptr) link_order->offset, size); free (buf); if (! ok) return FALSE; } /* Store the reloc information in the right place. It will get swapped and written out at the end of the final_link routine. */ irel = (flinfo->section_info[output_section->target_index].relocs + output_section->reloc_count); rel_hash_ptr = (flinfo->section_info[output_section->target_index].rel_hashes + output_section->reloc_count); memset (irel, 0, sizeof (struct internal_reloc)); *rel_hash_ptr = NULL; irel->r_vaddr = output_section->vma + link_order->offset; if (h->indx >= 0) irel->r_symndx = h->indx; else { /* Set the index to -2 to force this symbol to get written out. */ h->indx = -2; *rel_hash_ptr = h; irel->r_symndx = 0; } irel->r_type = howto->type; irel->r_size = howto->bitsize - 1; if (howto->complain_on_overflow == complain_overflow_signed) irel->r_size |= 0x80; ++output_section->reloc_count; /* Now output the reloc to the .loader section. */ if (xcoff_hash_table (flinfo->info)->loader_section) { if (!xcoff_create_ldrel (output_bfd, flinfo, output_section, output_bfd, irel, hsec, h)) return FALSE; } return TRUE; } /* Do the final link step. */ bfd_boolean _bfd_xcoff_bfd_final_link (bfd *abfd, struct bfd_link_info *info) { bfd_size_type symesz; struct xcoff_final_link_info flinfo; asection *o; struct bfd_link_order *p; bfd_size_type max_contents_size; bfd_size_type max_sym_count; bfd_size_type max_lineno_count; bfd_size_type max_reloc_count; bfd_size_type max_output_reloc_count; file_ptr rel_filepos; unsigned int relsz; file_ptr line_filepos; unsigned int linesz; bfd *sub; bfd_byte *external_relocs = NULL; char strbuf[STRING_SIZE_SIZE]; file_ptr pos; bfd_size_type amt; if (info->shared) abfd->flags |= DYNAMIC; symesz = bfd_coff_symesz (abfd); flinfo.info = info; flinfo.output_bfd = abfd; flinfo.strtab = NULL; flinfo.section_info = NULL; flinfo.last_file_index = -1; flinfo.toc_symindx = -1; flinfo.internal_syms = NULL; flinfo.sym_indices = NULL; flinfo.outsyms = NULL; flinfo.linenos = NULL; flinfo.contents = NULL; flinfo.external_relocs = NULL; if (xcoff_hash_table (info)->loader_section) { flinfo.ldsym = (xcoff_hash_table (info)->loader_section->contents + bfd_xcoff_ldhdrsz (abfd)); flinfo.ldrel = (xcoff_hash_table (info)->loader_section->contents + bfd_xcoff_ldhdrsz (abfd) + (xcoff_hash_table (info)->ldhdr.l_nsyms * bfd_xcoff_ldsymsz (abfd))); } else { flinfo.ldsym = NULL; flinfo.ldrel = NULL; } xcoff_data (abfd)->coff.link_info = info; flinfo.strtab = _bfd_stringtab_init (); if (flinfo.strtab == NULL) goto error_return; /* Count the relocation entries required for the output file. (We've already counted the line numbers.) Determine a few maximum sizes. */ max_contents_size = 0; max_lineno_count = 0; max_reloc_count = 0; for (o = abfd->sections; o != NULL; o = o->next) { o->reloc_count = 0; for (p = o->map_head.link_order; p != NULL; p = p->next) { if (p->type == bfd_indirect_link_order) { asection *sec; sec = p->u.indirect.section; /* Mark all sections which are to be included in the link. This will normally be every section. We need to do this so that we can identify any sections which the linker has decided to not include. */ sec->linker_mark = TRUE; o->reloc_count += sec->reloc_count; if ((sec->flags & SEC_IN_MEMORY) == 0) { if (sec->rawsize > max_contents_size) max_contents_size = sec->rawsize; if (sec->size > max_contents_size) max_contents_size = sec->size; } if (coff_section_data (sec->owner, sec) != NULL && xcoff_section_data (sec->owner, sec) != NULL && (xcoff_section_data (sec->owner, sec)->lineno_count > max_lineno_count)) max_lineno_count = xcoff_section_data (sec->owner, sec)->lineno_count; if (sec->reloc_count > max_reloc_count) max_reloc_count = sec->reloc_count; } else if (p->type == bfd_section_reloc_link_order || p->type == bfd_symbol_reloc_link_order) ++o->reloc_count; } } /* Compute the file positions for all the sections. */ if (abfd->output_has_begun) { if (xcoff_hash_table (info)->file_align != 0) abort (); } else { bfd_vma file_align; file_align = xcoff_hash_table (info)->file_align; if (file_align != 0) { bfd_boolean saw_contents; int indx; file_ptr sofar; /* Insert .pad sections before every section which has contents and is loaded, if it is preceded by some other section which has contents and is loaded. */ saw_contents = TRUE; for (o = abfd->sections; o != NULL; o = o->next) { if (strcmp (o->name, ".pad") == 0) saw_contents = FALSE; else if ((o->flags & SEC_HAS_CONTENTS) != 0 && (o->flags & SEC_LOAD) != 0) { if (! saw_contents) saw_contents = TRUE; else { asection *n; /* Create a pad section and place it before the section that needs padding. This requires unlinking and relinking the bfd's section list. */ n = bfd_make_section_anyway_with_flags (abfd, ".pad", SEC_HAS_CONTENTS); n->alignment_power = 0; bfd_section_list_remove (abfd, n); bfd_section_list_insert_before (abfd, o, n); saw_contents = FALSE; } } } /* Reset the section indices after inserting the new sections. */ indx = 0; for (o = abfd->sections; o != NULL; o = o->next) { ++indx; o->target_index = indx; } BFD_ASSERT ((unsigned int) indx == abfd->section_count); /* Work out appropriate sizes for the .pad sections to force each section to land on a page boundary. This bit of code knows what compute_section_file_positions is going to do. */ sofar = bfd_coff_filhsz (abfd); sofar += bfd_coff_aoutsz (abfd); sofar += abfd->section_count * bfd_coff_scnhsz (abfd); for (o = abfd->sections; o != NULL; o = o->next) if ((bfd_xcoff_is_reloc_count_overflow (abfd, (bfd_vma) o->reloc_count)) || (bfd_xcoff_is_lineno_count_overflow (abfd, (bfd_vma) o->lineno_count))) /* 64 does not overflow, need to check if 32 does */ sofar += bfd_coff_scnhsz (abfd); for (o = abfd->sections; o != NULL; o = o->next) { if (strcmp (o->name, ".pad") == 0) { bfd_vma pageoff; BFD_ASSERT (o->size == 0); pageoff = sofar & (file_align - 1); if (pageoff != 0) { o->size = file_align - pageoff; sofar += file_align - pageoff; o->flags |= SEC_HAS_CONTENTS; } } else { if ((o->flags & SEC_HAS_CONTENTS) != 0) sofar += BFD_ALIGN (o->size, 1 << o->alignment_power); } } } if (! bfd_coff_compute_section_file_positions (abfd)) goto error_return; } /* Allocate space for the pointers we need to keep for the relocs. */ { unsigned int i; /* We use section_count + 1, rather than section_count, because the target_index fields are 1 based. */ amt = abfd->section_count + 1; amt *= sizeof (struct xcoff_link_section_info); flinfo.section_info = bfd_malloc (amt); if (flinfo.section_info == NULL) goto error_return; for (i = 0; i <= abfd->section_count; i++) { flinfo.section_info[i].relocs = NULL; flinfo.section_info[i].rel_hashes = NULL; flinfo.section_info[i].toc_rel_hashes = NULL; } } /* Set the file positions for the relocs. */ rel_filepos = obj_relocbase (abfd); relsz = bfd_coff_relsz (abfd); max_output_reloc_count = 0; for (o = abfd->sections; o != NULL; o = o->next) { if (o->reloc_count == 0) o->rel_filepos = 0; else { /* A stripped file has no relocs. However, we still allocate the buffers, so that later code doesn't have to worry about whether we are stripping or not. */ if (info->strip == strip_all) o->rel_filepos = 0; else { o->flags |= SEC_RELOC; o->rel_filepos = rel_filepos; rel_filepos += o->reloc_count * relsz; } /* We don't know the indices of global symbols until we have written out all the local symbols. For each section in the output file, we keep an array of pointers to hash table entries. Each entry in the array corresponds to a reloc. When we find a reloc against a global symbol, we set the corresponding entry in this array so that we can fix up the symbol index after we have written out all the local symbols. Because of this problem, we also keep the relocs in memory until the end of the link. This wastes memory. We could backpatch the file later, I suppose, although it would be slow. */ amt = o->reloc_count; amt *= sizeof (struct internal_reloc); flinfo.section_info[o->target_index].relocs = bfd_malloc (amt); amt = o->reloc_count; amt *= sizeof (struct xcoff_link_hash_entry *); flinfo.section_info[o->target_index].rel_hashes = bfd_malloc (amt); if (flinfo.section_info[o->target_index].relocs == NULL || flinfo.section_info[o->target_index].rel_hashes == NULL) goto error_return; if (o->reloc_count > max_output_reloc_count) max_output_reloc_count = o->reloc_count; } } /* We now know the size of the relocs, so we can determine the file positions of the line numbers. */ line_filepos = rel_filepos; flinfo.line_filepos = line_filepos; linesz = bfd_coff_linesz (abfd); for (o = abfd->sections; o != NULL; o = o->next) { if (o->lineno_count == 0) o->line_filepos = 0; else { o->line_filepos = line_filepos; line_filepos += o->lineno_count * linesz; } /* Reset the reloc and lineno counts, so that we can use them to count the number of entries we have output so far. */ o->reloc_count = 0; o->lineno_count = 0; } obj_sym_filepos (abfd) = line_filepos; /* Figure out the largest number of symbols in an input BFD. Take the opportunity to clear the output_has_begun fields of all the input BFD's. We want at least 6 symbols, since that is the number which xcoff_write_global_symbol may need. */ max_sym_count = 6; for (sub = info->input_bfds; sub != NULL; sub = sub->link_next) { bfd_size_type sz; sub->output_has_begun = FALSE; sz = obj_raw_syment_count (sub); if (sz > max_sym_count) max_sym_count = sz; } /* Allocate some buffers used while linking. */ amt = max_sym_count * sizeof (struct internal_syment); flinfo.internal_syms = bfd_malloc (amt); amt = max_sym_count * sizeof (long); flinfo.sym_indices = bfd_malloc (amt); amt = (max_sym_count + 1) * symesz; flinfo.outsyms = bfd_malloc (amt); amt = max_lineno_count * bfd_coff_linesz (abfd); flinfo.linenos = bfd_malloc (amt); amt = max_contents_size; flinfo.contents = bfd_malloc (amt); amt = max_reloc_count * relsz; flinfo.external_relocs = bfd_malloc (amt); if ((flinfo.internal_syms == NULL && max_sym_count > 0) || (flinfo.sym_indices == NULL && max_sym_count > 0) || flinfo.outsyms == NULL || (flinfo.linenos == NULL && max_lineno_count > 0) || (flinfo.contents == NULL && max_contents_size > 0) || (flinfo.external_relocs == NULL && max_reloc_count > 0)) goto error_return; obj_raw_syment_count (abfd) = 0; /* Find a TOC symbol, if we need one. */ if (!xcoff_find_tc0 (abfd, &flinfo)) goto error_return; /* We now know the position of everything in the file, except that we don't know the size of the symbol table and therefore we don't know where the string table starts. We just build the string table in memory as we go along. We process all the relocations for a single input file at once. */ for (o = abfd->sections; o != NULL; o = o->next) { for (p = o->map_head.link_order; p != NULL; p = p->next) { if (p->type == bfd_indirect_link_order && p->u.indirect.section->owner->xvec == abfd->xvec) { sub = p->u.indirect.section->owner; if (! sub->output_has_begun) { if (! xcoff_link_input_bfd (&flinfo, sub)) goto error_return; sub->output_has_begun = TRUE; } } else if (p->type == bfd_section_reloc_link_order || p->type == bfd_symbol_reloc_link_order) { if (! xcoff_reloc_link_order (abfd, &flinfo, o, p)) goto error_return; } else { if (! _bfd_default_link_order (abfd, info, o, p)) goto error_return; } } } /* Free up the buffers used by xcoff_link_input_bfd. */ if (flinfo.internal_syms != NULL) { free (flinfo.internal_syms); flinfo.internal_syms = NULL; } if (flinfo.sym_indices != NULL) { free (flinfo.sym_indices); flinfo.sym_indices = NULL; } if (flinfo.linenos != NULL) { free (flinfo.linenos); flinfo.linenos = NULL; } if (flinfo.contents != NULL) { free (flinfo.contents); flinfo.contents = NULL; } if (flinfo.external_relocs != NULL) { free (flinfo.external_relocs); flinfo.external_relocs = NULL; } /* The value of the last C_FILE symbol is supposed to be -1. Write it out again. */ if (flinfo.last_file_index != -1) { flinfo.last_file.n_value = -(bfd_vma) 1; bfd_coff_swap_sym_out (abfd, (void *) &flinfo.last_file, (void *) flinfo.outsyms); pos = obj_sym_filepos (abfd) + flinfo.last_file_index * symesz; if (bfd_seek (abfd, pos, SEEK_SET) != 0 || bfd_bwrite (flinfo.outsyms, symesz, abfd) != symesz) goto error_return; } /* Write out all the global symbols which do not come from XCOFF input files. */ xcoff_link_hash_traverse (xcoff_hash_table (info), xcoff_write_global_symbol, (void *) &flinfo); if (flinfo.outsyms != NULL) { free (flinfo.outsyms); flinfo.outsyms = NULL; } /* Now that we have written out all the global symbols, we know the symbol indices to use for relocs against them, and we can finally write out the relocs. */ amt = max_output_reloc_count * relsz; external_relocs = bfd_malloc (amt); if (external_relocs == NULL && max_output_reloc_count != 0) goto error_return; for (o = abfd->sections; o != NULL; o = o->next) { struct internal_reloc *irel; struct internal_reloc *irelend; struct xcoff_link_hash_entry **rel_hash; struct xcoff_toc_rel_hash *toc_rel_hash; bfd_byte *erel; bfd_size_type rel_size; /* A stripped file has no relocs. */ if (info->strip == strip_all) { o->reloc_count = 0; continue; } if (o->reloc_count == 0) continue; irel = flinfo.section_info[o->target_index].relocs; irelend = irel + o->reloc_count; rel_hash = flinfo.section_info[o->target_index].rel_hashes; for (; irel < irelend; irel++, rel_hash++, erel += relsz) { if (*rel_hash != NULL) { if ((*rel_hash)->indx < 0) { if (! ((*info->callbacks->unattached_reloc) (info, (*rel_hash)->root.root.string, NULL, o, irel->r_vaddr))) goto error_return; (*rel_hash)->indx = 0; } irel->r_symndx = (*rel_hash)->indx; } } for (toc_rel_hash = flinfo.section_info[o->target_index].toc_rel_hashes; toc_rel_hash != NULL; toc_rel_hash = toc_rel_hash->next) { if (toc_rel_hash->h->u.toc_indx < 0) { if (! ((*info->callbacks->unattached_reloc) (info, toc_rel_hash->h->root.root.string, NULL, o, toc_rel_hash->rel->r_vaddr))) goto error_return; toc_rel_hash->h->u.toc_indx = 0; } toc_rel_hash->rel->r_symndx = toc_rel_hash->h->u.toc_indx; } /* XCOFF requires that the relocs be sorted by address. We tend to produce them in the order in which their containing csects appear in the symbol table, which is not necessarily by address. So we sort them here. There may be a better way to do this. */ qsort ((void *) flinfo.section_info[o->target_index].relocs, o->reloc_count, sizeof (struct internal_reloc), xcoff_sort_relocs); irel = flinfo.section_info[o->target_index].relocs; irelend = irel + o->reloc_count; erel = external_relocs; for (; irel < irelend; irel++, rel_hash++, erel += relsz) bfd_coff_swap_reloc_out (abfd, (void *) irel, (void *) erel); rel_size = relsz * o->reloc_count; if (bfd_seek (abfd, o->rel_filepos, SEEK_SET) != 0 || bfd_bwrite ((void *) external_relocs, rel_size, abfd) != rel_size) goto error_return; } if (external_relocs != NULL) { free (external_relocs); external_relocs = NULL; } /* Free up the section information. */ if (flinfo.section_info != NULL) { unsigned int i; for (i = 0; i < abfd->section_count; i++) { if (flinfo.section_info[i].relocs != NULL) free (flinfo.section_info[i].relocs); if (flinfo.section_info[i].rel_hashes != NULL) free (flinfo.section_info[i].rel_hashes); } free (flinfo.section_info); flinfo.section_info = NULL; } /* Write out the loader section contents. */ o = xcoff_hash_table (info)->loader_section; if (o) { BFD_ASSERT ((bfd_byte *) flinfo.ldrel == (xcoff_hash_table (info)->loader_section->contents + xcoff_hash_table (info)->ldhdr.l_impoff)); if (!bfd_set_section_contents (abfd, o->output_section, o->contents, (file_ptr) o->output_offset, o->size)) goto error_return; } /* Write out the magic sections. */ o = xcoff_hash_table (info)->linkage_section; if (o->size > 0 && ! bfd_set_section_contents (abfd, o->output_section, o->contents, (file_ptr) o->output_offset, o->size)) goto error_return; o = xcoff_hash_table (info)->toc_section; if (o->size > 0 && ! bfd_set_section_contents (abfd, o->output_section, o->contents, (file_ptr) o->output_offset, o->size)) goto error_return; o = xcoff_hash_table (info)->descriptor_section; if (o->size > 0 && ! bfd_set_section_contents (abfd, o->output_section, o->contents, (file_ptr) o->output_offset, o->size)) goto error_return; /* Write out the string table. */ pos = obj_sym_filepos (abfd) + obj_raw_syment_count (abfd) * symesz; if (bfd_seek (abfd, pos, SEEK_SET) != 0) goto error_return; H_PUT_32 (abfd, _bfd_stringtab_size (flinfo.strtab) + STRING_SIZE_SIZE, strbuf); amt = STRING_SIZE_SIZE; if (bfd_bwrite (strbuf, amt, abfd) != amt) goto error_return; if (! _bfd_stringtab_emit (abfd, flinfo.strtab)) goto error_return; _bfd_stringtab_free (flinfo.strtab); /* Write out the debugging string table. */ o = xcoff_hash_table (info)->debug_section; if (o != NULL) { struct bfd_strtab_hash *debug_strtab; debug_strtab = xcoff_hash_table (info)->debug_strtab; BFD_ASSERT (o->output_section->size - o->output_offset >= _bfd_stringtab_size (debug_strtab)); pos = o->output_section->filepos + o->output_offset; if (bfd_seek (abfd, pos, SEEK_SET) != 0) goto error_return; if (! _bfd_stringtab_emit (abfd, debug_strtab)) goto error_return; } /* Setting bfd_get_symcount to 0 will cause write_object_contents to not try to write out the symbols. */ bfd_get_symcount (abfd) = 0; return TRUE; error_return: if (flinfo.strtab != NULL) _bfd_stringtab_free (flinfo.strtab); if (flinfo.section_info != NULL) { unsigned int i; for (i = 0; i < abfd->section_count; i++) { if (flinfo.section_info[i].relocs != NULL) free (flinfo.section_info[i].relocs); if (flinfo.section_info[i].rel_hashes != NULL) free (flinfo.section_info[i].rel_hashes); } free (flinfo.section_info); } if (flinfo.internal_syms != NULL) free (flinfo.internal_syms); if (flinfo.sym_indices != NULL) free (flinfo.sym_indices); if (flinfo.outsyms != NULL) free (flinfo.outsyms); if (flinfo.linenos != NULL) free (flinfo.linenos); if (flinfo.contents != NULL) free (flinfo.contents); if (flinfo.external_relocs != NULL) free (flinfo.external_relocs); if (external_relocs != NULL) free (external_relocs); return FALSE; }
Go to most recent revision | Compare with Previous | Blame | View Log