diff options
Diffstat (limited to 'bfd/coff-alpha.c')
| -rw-r--r-- | bfd/coff-alpha.c | 360 |
1 files changed, 324 insertions, 36 deletions
diff --git a/bfd/coff-alpha.c b/bfd/coff-alpha.c index 4ad7077..1bbec4c 100644 --- a/bfd/coff-alpha.c +++ b/bfd/coff-alpha.c @@ -28,6 +28,7 @@ Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ #include "coff/symconst.h" #include "coff/ecoff.h" #include "coff/alpha.h" +#include "aout/ar.h" #include "libcoff.h" #include "libecoff.h" @@ -57,6 +58,10 @@ static boolean alpha_relocate_section PARAMS ((bfd *, struct bfd_link_info *, bfd_byte *, PTR)); static boolean alpha_adjust_headers PARAMS ((bfd *, struct internal_filehdr *, struct internal_aouthdr *)); +static PTR alpha_ecoff_read_ar_hdr PARAMS ((bfd *)); +static bfd *alpha_ecoff_get_elt_at_filepos PARAMS ((bfd *, file_ptr)); +static bfd *alpha_ecoff_openr_next_archived_file PARAMS ((bfd *, bfd *)); +static bfd *alpha_ecoff_get_elt_at_index PARAMS ((bfd *, symindex)); /* ECOFF has COFF sections, but the debugging information is stored in a completely different format. ECOFF targets use some of the @@ -525,7 +530,7 @@ alpha_ecoff_swap_reloc_in (abfd, ext_ptr, intern) intern->r_vaddr = bfd_h_get_64 (abfd, (bfd_byte *) ext->r_vaddr); intern->r_symndx = bfd_h_get_32 (abfd, (bfd_byte *) ext->r_symndx); - BFD_ASSERT (abfd->xvec->header_byteorder_big_p == false); + BFD_ASSERT (bfd_header_little_endian (abfd)); intern->r_type = ((ext->r_bits[0] & RELOC_BITS0_TYPE_LITTLE) >> RELOC_BITS0_TYPE_SH_LITTLE); @@ -598,7 +603,7 @@ alpha_ecoff_swap_reloc_out (abfd, intern, dst) bfd_h_put_64 (abfd, intern->r_vaddr, (bfd_byte *) ext->r_vaddr); bfd_h_put_32 (abfd, symndx, (bfd_byte *) ext->r_symndx); - BFD_ASSERT (abfd->xvec->header_byteorder_big_p == false); + BFD_ASSERT (bfd_header_little_endian (abfd)); ext->r_bits[0] = ((intern->r_type << RELOC_BITS0_TYPE_SH_LITTLE) & RELOC_BITS0_TYPE_LITTLE); @@ -914,16 +919,12 @@ alpha_ecoff_get_relocated_section_contents (abfd, link_info, link_order, /* This marks the ldah of an ldah/lda pair which loads the gp register with the difference of the gp value and the current location. The second of the pair is r_size bytes - ahead, and is marked with an ALPHA_R_IGNORE reloc. */ + ahead; it used to be marked with an ALPHA_R_IGNORE reloc, + but that no longer happens in OSF/1 3.2. */ { unsigned long insn1, insn2; bfd_vma addend; - BFD_ASSERT (reloc_vector[1] != NULL - && reloc_vector[1]->howto->type == ALPHA_R_IGNORE - && (rel->address + rel->addend - == reloc_vector[1]->address)); - /* Get the two instructions. */ insn1 = bfd_get_32 (input_bfd, data + rel->address); insn2 = bfd_get_32 (input_bfd, data + rel->address + rel->addend); @@ -945,7 +946,7 @@ alpha_ecoff_get_relocated_section_contents (abfd, link_info, link_order, /* The existing addend includes the different between the gp of the input BFD and the address in the input BFD. Subtract this out. */ - addend -= (reloc_vector[1]->addend + addend -= (ecoff_data (input_bfd)->gp - (input_section->vma + rel->address)); /* Now add in the final gp value, and subtract out the @@ -1364,7 +1365,7 @@ alpha_relocate_section (output_bfd, info, input_bfd, input_section, bfd_byte *contents; PTR external_relocs; { - asection **symndx_to_section; + asection **symndx_to_section, *lita_sec; struct ecoff_link_hash_entry **sym_hashes; bfd_vma gp; boolean gp_undefined; @@ -1422,14 +1423,79 @@ alpha_relocate_section (output_bfd, info, input_bfd, input_section, sym_hashes = ecoff_data (input_bfd)->sym_hashes; + /* On the Alpha, the .lita section must be addressable by the global + pointer. To support large programs, we need to allow multiple + global pointers. This works as long as each input .lita section + is <64KB big. This implies that when producing relocatable + output, the .lita section is limited to 64KB. . */ + + lita_sec = symndx_to_section[RELOC_SECTION_LITA]; gp = ecoff_data (output_bfd)->gp; - if (gp == 0) - gp_undefined = true; - else - gp_undefined = false; + if (! info->relocateable && lita_sec != NULL) + { + struct ecoff_section_tdata *lita_sec_data; + + /* Make sure we have a section data structure to which we can + hang on to the gp value we pick for the section. */ + lita_sec_data = ecoff_section_data (input_bfd, lita_sec); + if (lita_sec_data == NULL) + { + lita_sec_data = ((struct ecoff_section_tdata *) + bfd_zalloc (input_bfd, + sizeof (struct ecoff_section_tdata))); + ecoff_section_data (input_bfd, lita_sec) = lita_sec_data; + } + + if (lita_sec_data->gp != 0) + { + /* If we already assigned a gp to this section, we better + stick with that value. */ + gp = lita_sec_data->gp; + } + else + { + bfd_vma lita_vma; + bfd_size_type lita_size; + + lita_vma = lita_sec->output_offset + lita_sec->output_section->vma; + lita_size = lita_sec->_cooked_size; + if (lita_size == 0) + lita_size = lita_sec->_raw_size; + + if (gp == 0 + || lita_vma < gp - 0x8000 + || lita_vma + lita_size >= gp + 0x8000) + { + /* Either gp hasn't been set at all or the current gp + cannot address this .lita section. In both cases we + reset the gp to point into the "middle" of the + current input .lita section. For now, we issue a + warning when redefining the gp value (probably should + be made optional). */ + if (gp && !ecoff_data (output_bfd)->issued_multiple_gp_warning) + { + (*_bfd_error_handler) + ("%s: warning: using multiple gp values", + bfd_get_filename (output_bfd)); + ecoff_data (output_bfd)->issued_multiple_gp_warning = true; + } + if (lita_vma < gp - 0x8000) + gp = lita_vma + lita_size - 0x8000; + else + gp = lita_vma + 0x8000; + + } + + lita_sec_data->gp = gp; + } + + ecoff_data (output_bfd)->gp = gp; + } + + gp_undefined = (gp == 0); - BFD_ASSERT (output_bfd->xvec->header_byteorder_big_p == false); - BFD_ASSERT (input_bfd->xvec->header_byteorder_big_p == false); + BFD_ASSERT (bfd_header_little_endian (output_bfd)); + BFD_ASSERT (bfd_header_little_endian (input_bfd)); ext_rel = (struct external_reloc *) external_relocs; ext_rel_end = ext_rel + input_section->reloc_count; @@ -1469,12 +1535,12 @@ alpha_relocate_section (output_bfd, info, input_bfd, input_section, abort (); case ALPHA_R_IGNORE: - /* This reloc appears after a GPDISP reloc. It marks the - position of the second instruction to be altered by the - GPDISP reloc, but is not otherwise used for anything. - For some reason, the address of the relocation does not - appear to include the section VMA, unlike the other - relocation types. */ + /* This reloc appears after a GPDISP reloc. On earlier + versions of OSF/1, It marked the position of the second + instruction to be altered by the GPDISP reloc, but it is + not otherwise used for anything. For some reason, the + address of the relocation does not appear to include the + section VMA, unlike the other relocation types. */ if (info->relocateable) bfd_h_put_64 (input_bfd, input_section->output_offset + r_vaddr, @@ -1544,19 +1610,11 @@ alpha_relocate_section (output_bfd, info, input_bfd, input_section, /* This marks the ldah of an ldah/lda pair which loads the gp register with the difference of the gp value and the current location. The second of the pair is r_symndx - bytes ahead, and is also marked with an ALPHA_R_IGNORE - reloc. */ + bytes ahead. It used to be marked with an ALPHA_R_IGNORE + reloc, but OSF/1 3.2 no longer does that. */ { unsigned long insn1, insn2; - BFD_ASSERT (ext_rel + 1 < ext_rel_end - && (((ext_rel + 1)->r_bits[0] - & RELOC_BITS0_TYPE_LITTLE) - >> RELOC_BITS0_TYPE_SH_LITTLE) == ALPHA_R_IGNORE - && (bfd_h_get_64 (input_bfd, - (bfd_byte *) (ext_rel + 1)->r_vaddr) - == r_vaddr - input_section->vma + r_symndx)); - /* Get the two instructions. */ insn1 = bfd_get_32 (input_bfd, contents + r_vaddr - input_section->vma); @@ -1949,6 +2007,234 @@ alpha_adjust_headers (abfd, fhdr, ahdr) return true; } +/* Archive handling. In OSF/1 (or Digital Unix) v3.2, Digital + introduced archive packing, in which the elements in an archive are + optionally compressed using a simple dictionary scheme. We know + how to read such archives, but we don't write them. */ + +#define alpha_ecoff_slurp_armap _bfd_ecoff_slurp_armap +#define alpha_ecoff_slurp_extended_name_table \ + _bfd_ecoff_slurp_extended_name_table +#define alpha_ecoff_construct_extended_name_table \ + _bfd_ecoff_construct_extended_name_table +#define alpha_ecoff_truncate_arname _bfd_ecoff_truncate_arname +#define alpha_ecoff_write_armap _bfd_ecoff_write_armap +#define alpha_ecoff_generic_stat_arch_elt _bfd_ecoff_generic_stat_arch_elt +#define alpha_ecoff_update_armap_timestamp _bfd_ecoff_update_armap_timestamp + +/* A compressed file uses this instead of ARFMAG. */ + +#define ARFZMAG "Z\012" + +/* Read an archive header. This is like the standard routine, but it + also accepts ARFZMAG. */ + +static PTR +alpha_ecoff_read_ar_hdr (abfd) + bfd *abfd; +{ + struct areltdata *ret; + struct ar_hdr *h; + + ret = (struct areltdata *) _bfd_generic_read_ar_hdr_mag (abfd, ARFZMAG); + if (ret == NULL) + return NULL; + + h = (struct ar_hdr *) ret->arch_header; + if (strncmp (h->ar_fmag, ARFZMAG, 2) == 0) + { + bfd_byte ab[8]; + + /* This is a compressed file. We must set the size correctly. + The size is the eight bytes after the dummy file header. */ + if (bfd_seek (abfd, FILHSZ, SEEK_CUR) != 0 + || bfd_read (ab, 1, 8, abfd) != 8 + || bfd_seek (abfd, - (FILHSZ + 8), SEEK_CUR) != 0) + return NULL; + + ret->parsed_size = bfd_h_get_64 (abfd, ab); + } + + return (PTR) ret; +} + +/* Get an archive element at a specified file position. This is where + we uncompress the archive element if necessary. */ + +static bfd * +alpha_ecoff_get_elt_at_filepos (archive, filepos) + bfd *archive; + file_ptr filepos; +{ + bfd *nbfd = NULL; + struct areltdata *tdata; + struct ar_hdr *hdr; + bfd_byte ab[8]; + bfd_size_type size; + bfd_byte *buf, *p; + struct bfd_in_memory *bim; + + nbfd = _bfd_get_elt_at_filepos (archive, filepos); + if (nbfd == NULL) + goto error_return; + + if ((nbfd->flags & BFD_IN_MEMORY) != 0) + { + /* We have already expanded this BFD. */ + return nbfd; + } + + tdata = (struct areltdata *) nbfd->arelt_data; + hdr = (struct ar_hdr *) tdata->arch_header; + if (strncmp (hdr->ar_fmag, ARFZMAG, 2) != 0) + return nbfd; + + /* We must uncompress this element. We do this by copying it into a + memory buffer, and making bfd_read and bfd_seek use that buffer. + This can use a lot of memory, but it's simpler than getting a + temporary file, making that work with the file descriptor caching + code, and making sure that it is deleted at all appropriate + times. It can be changed if it ever becomes important. */ + + /* The compressed file starts with a dummy ECOFF file header. */ + if (bfd_seek (nbfd, FILHSZ, SEEK_SET) != 0) + goto error_return; + + /* The next eight bytes are the real file size. */ + if (bfd_read (ab, 1, 8, nbfd) != 8) + goto error_return; + size = bfd_h_get_64 (nbfd, ab); + + if (size == 0) + buf = NULL; + else + { + bfd_size_type left; + bfd_byte dict[4096]; + unsigned int h; + bfd_byte b; + + buf = (bfd_byte *) bfd_alloc (nbfd, size); + if (buf == NULL) + goto error_return; + p = buf; + + left = size; + + /* I don't know what the next eight bytes are for. */ + if (bfd_read (ab, 1, 8, nbfd) != 8) + goto error_return; + + /* This is the uncompression algorithm. It's a simple + dictionary based scheme in which each character is predicted + by a hash of the previous three characters. A control byte + indicates whether the character is predicted or whether it + appears in the input stream; each control byte manages the + next eight bytes in the output stream. */ + memset (dict, 0, sizeof dict); + h = 0; + while (bfd_read (&b, 1, 1, nbfd) == 1) + { + unsigned int i; + + for (i = 0; i < 8; i++, b >>= 1) + { + bfd_byte n; + + if ((b & 1) == 0) + n = dict[h]; + else + { + if (! bfd_read (&n, 1, 1, nbfd)) + goto error_return; + dict[h] = n; + } + + *p++ = n; + + --left; + if (left == 0) + break; + + h <<= 4; + h ^= n; + h &= sizeof dict - 1; + } + + if (left == 0) + break; + } + } + + /* Now the uncompressed file contents are in buf. */ + bim = ((struct bfd_in_memory *) + bfd_alloc (nbfd, sizeof (struct bfd_in_memory))); + if (bim == NULL) + goto error_return; + bim->size = size; + bim->buffer = buf; + + nbfd->mtime_set = true; + nbfd->mtime = strtol (hdr->ar_date, (char **) NULL, 10); + + nbfd->flags |= BFD_IN_MEMORY; + nbfd->iostream = (PTR) bim; + BFD_ASSERT (! nbfd->cacheable); + + return nbfd; + + error_return: + if (nbfd != NULL) + bfd_close (nbfd); + return NULL; +} + +/* Open the next archived file. */ + +static bfd * +alpha_ecoff_openr_next_archived_file (archive, last_file) + bfd *archive; + bfd *last_file; +{ + file_ptr filestart; + + if (last_file == NULL) + filestart = bfd_ardata (archive)->first_file_filepos; + else + { + struct areltdata *t; + struct ar_hdr *h; + bfd_size_type size; + + /* We can't use arelt_size here, because that uses parsed_size, + which is the uncompressed size. We need the compressed size. */ + t = (struct areltdata *) last_file->arelt_data; + h = (struct ar_hdr *) t->arch_header; + size = strtol (h->ar_size, (char **) NULL, 10); + + /* Pad to an even boundary... + Note that last_file->origin can be odd in the case of + BSD-4.4-style element with a long odd size. */ + filestart = last_file->origin + size; + filestart += filestart % 2; + } + + return alpha_ecoff_get_elt_at_filepos (archive, filestart); +} + +/* Open the archive file given an index into the armap. */ + +static bfd * +alpha_ecoff_get_elt_at_index (abfd, index) + bfd *abfd; + symindex index; +{ + carsym *entry; + + entry = bfd_ardata (abfd)->symdefs + index; + return alpha_ecoff_get_elt_at_filepos (abfd, entry->file_offset); +} + /* This is the ECOFF backend structure. The backend field of the target vector points to this. */ @@ -2038,7 +2324,9 @@ static const struct ecoff_backend_data alpha_ecoff_backend_data = /* Relocate section contents while linking. */ alpha_relocate_section, /* Do final adjustments to filehdr and aouthdr. */ - alpha_adjust_headers + alpha_adjust_headers, + /* Read an element from an archive at a given file position. */ + alpha_ecoff_get_elt_at_filepos }; /* Looking up a reloc type is Alpha specific. */ @@ -2059,8 +2347,8 @@ const bfd_target ecoffalpha_little_vec = { "ecoff-littlealpha", /* name */ bfd_target_ecoff_flavour, - false, /* data byte order is little */ - false, /* header byte order is little */ + BFD_ENDIAN_LITTLE, /* data byte order is little */ + BFD_ENDIAN_LITTLE, /* header byte order is little */ (HAS_RELOC | EXEC_P | /* object flags */ HAS_LINENO | HAS_DEBUG | @@ -2087,7 +2375,7 @@ const bfd_target ecoffalpha_little_vec = BFD_JUMP_TABLE_GENERIC (_bfd_ecoff), BFD_JUMP_TABLE_COPY (_bfd_ecoff), BFD_JUMP_TABLE_CORE (_bfd_nocore), - BFD_JUMP_TABLE_ARCHIVE (_bfd_ecoff), + BFD_JUMP_TABLE_ARCHIVE (alpha_ecoff), BFD_JUMP_TABLE_SYMBOLS (_bfd_ecoff), BFD_JUMP_TABLE_RELOCS (_bfd_ecoff), BFD_JUMP_TABLE_WRITE (_bfd_ecoff), |