1 /* Copyright (c) 1980 Regents of the University of California */ 2 static char sccsid[] = "@(#)asmain.c 4.4 08/16/80"; 3 #include <stdio.h> 4 #include <ctype.h> 5 #include <signal.h> 6 7 #include "as.h" 8 #include "assyms.h" 9 #include "asexpr.h" 10 #include "asscan.h" 11 12 #ifdef UNIX 13 #define unix_lang_name "VAX/UNIX Assembler Vasmain.c" 14 #endif 15 16 #ifdef VMS 17 #define vms_lang_name "VAX/VMS C Assembler V1.00" 18 #endif VMS 19 20 /* 21 * variables to manage reading the assembly source files 22 */ 23 char *dotsname; /*the current file name; managed by the parser*/ 24 int lineno; /*current line number; managed by the parser*/ 25 char *innames[32]; /*names of the files being assembled*/ 26 int ninfiles; /*how many interesting files there are*/ 27 /* 28 * Flags settable from the argv process argument list 29 */ 30 int silent = 0; /*don't complain about any errors*/ 31 int savelabels = 0; /*write the labels to the a.out file*/ 32 int d124 = 4; /*default allocate 4 bytes for unknown pointers*/ 33 int anyerrs = 0; /*no errors yet*/ 34 int orgwarn = 0; /*Bad origins*/ 35 int passno = 1; /* current pass*/ 36 int jxxxJUMP = 0; /* in jxxxes that branch too far, use jmp instead of brw */ 37 int readonlydata = 0; /* initialzed data -> text space */ 38 39 #ifdef DEBUG 40 int debug = 0; 41 int toktrace = 0; 42 #endif 43 44 int useVM = /*put the temp file in virtual memory*/ 45 #ifdef VMS 46 1; /*VMS has virtual memory (duh)*/ 47 #endif VMS 48 #ifdef UNIX 49 0; 50 #endif 51 52 char *endcore; /*where to get more symbol space*/ 53 54 /* 55 * Managers of the a.out file. 56 */ 57 struct exec hdr; 58 #define MAGIC 0407 59 u_long tsize; /* total text size */ 60 u_long dsize; /* total data size */ 61 u_long datbase; /* base of the data segment */ 62 u_long trsize; /* total text relocation size */ 63 u_long drsize; /* total data relocation size */ 64 65 /* 66 * Information about the current segment is accumulated in 67 * usedot; the most important information stored is the 68 * accumulated size of each of the text and data segments 69 * 70 * dotp points to the correct usedot expression for the current segment 71 */ 72 struct exp usedot[NLOC+NLOC]; /* info about all segments */ 73 struct exp *dotp; /* data/text location pointer */ 74 /* 75 * The inter pass temporary file is opened and closed by stdio, but 76 * is written to using direct read/write, as the temporary file 77 * is composed of buffers exactly BUFSIZ long. 78 */ 79 FILE *tmpfil; /* interpass communication file */ 80 /* 81 * a.out is created during the second pass. 82 * It is opened by stdio, but is filled with the parallel 83 * block I/O library 84 */ 85 char *outfile = "a.out"; 86 FILE *a_out_file; 87 off_t a_out_off; /* cumulative offsets for segments */ 88 /* 89 * The logical files containing the assembled data for each of 90 * the text and data segments are 91 * managed by the parallel block I/O library. 92 * a.out is logically opened in many places at once to 93 * receive the assembled data from the various segments as 94 * it all trickles in, but is physically opened only once 95 * to minimize file overhead. 96 */ 97 BFILE *usefile[NLOC+NLOC]; /* text/data files */ 98 BFILE *txtfil; /* current text/data file */ 99 /* 100 * Relocation information is accumulated seperately for each 101 * segment. This is required by the old loader (from BTL), 102 * but not by the new loader (Bill Joy). 103 * 104 * However, the size of the relocation information can not be computed 105 * during or after the 1st pass because the ''absoluteness' of values 106 * is unknown until all locally declared symbols have been seen. 107 * Thus, the size of the relocation information is only 108 * known after the second pass is finished. 109 * This obviates the use of the block I/O 110 * library, which requires knowing the exact offsets in a.out. 111 * 112 * So, we save the relocation information internally (we don't 113 * go to internal files to minimize overhead). 114 * 115 * Empirically, we studied 259 files composing the system, 116 * two compilers and a compiler generator: (all of which have 117 * fairly large source files) 118 * 119 * Number of files = 259 120 * Number of non zero text reloc files: 233 121 * Number of non zero data reloc files: 53 122 * Average text relocation = 889 123 * Average data relocation = 346 124 * Number of files > BUFSIZ text relocation = 71 125 * Number of files > BUFSIZ data relocation = 6 126 * 127 * For compiled C code, there is usually one text segment and two 128 * data segments; we see that allocating our own buffers and 129 * doing our internal handling of relocation information will, 130 * on the average, not use more memory than taken up by the buffers 131 * allocated for doing file I/O in parallel to a number of file. 132 * 133 * If we are assembling with the -V option, we 134 * use the left over token buffers from the 2nd pass, 135 * otherwise, we create our own. 136 * 137 * When the 2nd pass is complete, closeoutrel flushes the token 138 * buffers out to a BFILE. 139 * 140 * The internals to relbufdesc are known only in assyms.c 141 * 142 * outrel constructs the relocation information. 143 * closeoutrel flushes the relocation information to relfil. 144 */ 145 struct relbufdesc *rusefile[NLOC+NLOC]; 146 struct relbufdesc *relfil; /* un concatnated relocation info */ 147 BFILE *relocfile; /* concatnated relocation info */ 148 /* 149 * Once the relocation information has been written, 150 * we can write out the symbol table using the Block I/O 151 * mechanisms, as we once again know the offsets into 152 * the a.out file. 153 * 154 * We use relfil to output the symbol table information. 155 */ 156 157 char *tmpdirprefix = 158 #ifdef UNIX 159 "/tmp/"; 160 #else VMS 161 "/usr/tmp/"; 162 #endif 163 164 #define TMP_SUFFIX "asXXXXXX" 165 char tmpn1[TNAMESIZE]; 166 167 int delexit(); 168 169 main(argc, argv) 170 int argc; 171 char **argv; 172 { 173 174 tmpn1[0] = 0; 175 endcore = (char *)sbrk(0); 176 177 argprocess(argc, argv); /* process argument lists */ 178 if (anyerrs) exit(1); 179 180 initialize(); 181 zeroorigins(); /* set origins to zero */ 182 zerolocals(); /* fix local label counters */ 183 184 i_pass1(); /* open temp files, etc */ 185 pass1(); /* first pass through .s files */ 186 testlocals(); /* check for undefined locals */ 187 if (anyerrs) delexit(); 188 189 pass1_5(); /* resolve jxxx */ 190 if (anyerrs) delexit(); 191 192 open_a_out(); /* open a.out */ 193 roundsegments(); /* round segments to FW */ 194 build_hdr(); /* build initial header, and output */ 195 196 i_pass2(); /* reopen temporary file, etc */ 197 pass2(); /* second pass through the virtual .s */ 198 if (anyerrs) delexit(); 199 200 fillsegments(); /* fill segments with 0 to FW */ 201 reloc_syms(); /* dump relocation and symbol table */ 202 203 delete(); /* remove tmp file */ 204 bflush(); /* close off block I/O view of a.out */ 205 fix_a_out(); /* add in text and data reloc counts */ 206 207 if (anyerrs == 0 && orgwarn) 208 yyerror("Caution: absolute origins.\n"); 209 exit(anyerrs != 0); 210 } /*end of UNIX main*/ 211 212 argprocess(argc, argv) 213 int argc; 214 char *argv[]; 215 { 216 register char *cp; 217 218 ninfiles = 0; 219 silent = 0; 220 #ifdef DEBUG 221 debug = 0; 222 #endif 223 dotsname = "<argv error>"; 224 while (argc > 1) { 225 if (argv[1][0] == '-'){ 226 cp = argv[1] + 1; 227 /* 228 * We can throw away single minus signs, so 229 * that make scripts for the PDP 11 assembler work 230 * on this assembler too 231 */ 232 while (*cp){ 233 switch(*cp++){ 234 default: 235 yyerror("Unknown flag: %c", *--cp); 236 cp++; 237 break; 238 case 'd': 239 d124 = *cp++ - '0'; 240 if ( (d124 != 1) && (d124 != 2) && 241 (d124 != 4)){ 242 yyerror("-d[124] only"); 243 exit(1); 244 } 245 break; 246 case 'o': 247 if (argc < 3){ 248 yyerror("-o what???"); 249 exit(1); 250 } 251 outfile = argv[2]; 252 bumpone: 253 argc -= 2; 254 argv += 2; 255 goto nextarg; 256 257 case 't': 258 if (argc < 3){ 259 yyerror("-t what???"); 260 exit(1); 261 } 262 tmpdirprefix = argv[2]; 263 goto bumpone; 264 265 case 'V': 266 useVM = 1; 267 break; 268 case 'W': 269 silent = 1; 270 break; 271 case 'L': 272 savelabels = 1; 273 break; 274 case 'J': 275 jxxxJUMP = 1; 276 break; 277 #ifdef DEBUG 278 case 'D': 279 debug = 1; 280 break; 281 case 'T': 282 toktrace = 1; 283 break; 284 #endif 285 case 'R': 286 readonlydata = 1; 287 break; 288 } /*end of the switch*/ 289 } /*end of pulling out all arguments*/ 290 } /*end of a flag argument*/ 291 else { /*file name*/ 292 if (ninfiles > 32){ 293 yyerror("More than 32 file names"); 294 exit(3); 295 } 296 innames[ninfiles++] = argv[1]; 297 } 298 --argc; ++argv; 299 nextarg:; 300 } 301 } 302 303 initialize() 304 { 305 if (signal(SIGINT, SIG_IGN) != SIG_IGN) 306 signal(SIGINT, delexit); 307 /* 308 * Install symbols in the table 309 */ 310 symtabinit(); 311 syminstall(); 312 /* 313 * Build the expression parser accelerator token sets 314 */ 315 buildtokensets(); 316 } 317 318 zeroorigins() 319 { 320 register int locindex; 321 /* 322 * Mark usedot: the first NLOC slots are for named text segments, 323 * the next for named data segments. 324 */ 325 for (locindex = 0; locindex < NLOC; locindex++){ 326 usedot[locindex].e_xtype = XTEXT; 327 usedot[NLOC + locindex].e_xtype = XDATA; 328 usedot[locindex].e_xvalue = 0; 329 usedot[NLOC + locindex].e_xvalue = 0; 330 usedot[locindex].e_yvalue = 0; 331 usedot[NLOC + locindex].e_yvalue = 0; 332 } 333 } 334 335 zerolocals() 336 { 337 register int i; 338 339 for (i = 0; i <= 9; i++) { 340 lgensym[i] = 1; 341 genref[i] = 0; 342 } 343 } 344 345 i_pass1() 346 { 347 if (useVM == 0){ 348 strcat(tmpn1, tmpdirprefix); 349 strcat(tmpn1, TMP_SUFFIX); 350 mktemp(tmpn1); 351 tmpfil = fopen(tmpn1, "w"); 352 if (tmpfil==NULL) { 353 yyerror("Bad pass 1 temporary file for writing %s", tmpn1); 354 delexit(); 355 } 356 } 357 358 inittmpfile(); 359 initijxxx(); 360 } 361 362 pass1() 363 { 364 register int i; 365 366 passno = 1; 367 dotp = &usedot[0]; 368 txtfil = (BFILE *)0; 369 relfil = (struct relbufdesc *)0; 370 371 if (ninfiles == 0){ /*take the input from stdin directly*/ 372 lineno = 1; 373 dotsname = "<stdin>"; 374 375 yyparse(); 376 } else { /*we have the names tanked*/ 377 for (i = 0; i < ninfiles; i++){ 378 new_dot_s(innames[i]); 379 if (freopen(innames[i], "r", stdin) == NULL) { 380 yyerror( "Can't open source file %s\n", 381 innames[i]); 382 exit(2); 383 } 384 /* stdio is NOT used to read the input characters */ 385 /* we use read directly, into our own buffers */ 386 yyparse(); 387 } 388 } 389 390 closetmpfile(); /*kick out the last buffered intermediate text*/ 391 } 392 393 testlocals() 394 { 395 register int i; 396 for (i = 0; i <= 9; i++) { 397 if (genref[i]) 398 yyerror("Reference to undefined local label %df", i); 399 lgensym[i] = 1; 400 genref[i] = 0; 401 } 402 } 403 404 pass1_5() 405 { 406 sortsymtab(); 407 #ifdef DEBUG 408 if (debug) dumpsymtab(); 409 #endif 410 jxxxfix(); 411 #ifdef DEBUG 412 if (debug) dumpsymtab(); 413 #endif 414 } 415 416 open_a_out() 417 { 418 /* 419 * Open up the a.out file now, and get set to build 420 * up offsets into it for all of the various text,data 421 * text relocation and data relocation segments. 422 */ 423 a_out_file = fopen(outfile, "w"); 424 if (a_out_file == NULL) { 425 yyerror("Cannot create %s", outfile); 426 delexit(); 427 } 428 biofd = a_out_file->_file; 429 a_out_off = 0; 430 } 431 432 roundsegments() 433 { 434 register int locindex; 435 register long v; 436 /* 437 * round and assign text segment origins 438 * the exec header always goes in usefile[0] 439 */ 440 tsize = 0; 441 for (locindex=0; locindex<NLOC; locindex++) { 442 v = round(usedot[locindex].e_xvalue, FW); 443 usedot[locindex].e_xvalue = tsize; 444 if ((locindex == 0) || (v != 0) ){ 445 usefile[locindex] = (BFILE *)Calloc(1, sizeof(BFILE)); 446 bopen(usefile[locindex], a_out_off); 447 if (locindex == 0) 448 a_out_off = sizeof (struct exec); 449 } else { 450 usefile[locindex] = (BFILE *)-1; 451 } 452 tsize += v; 453 a_out_off += v; 454 } 455 /* 456 * Round and assign data segment origins. 457 */ 458 datbase = round(tsize, FW); 459 for (locindex=0; locindex<NLOC; locindex++) { 460 v = round(usedot[NLOC+locindex].e_xvalue, FW); 461 usedot[NLOC+locindex].e_xvalue = datbase + dsize; 462 if (v != 0){ 463 usefile[NLOC + locindex] = (BFILE *)Calloc(1,sizeof(BFILE)); 464 bopen(usefile[NLOC + locindex], a_out_off); 465 } else { 466 usefile[NLOC + locindex] = (BFILE *)-1; 467 } 468 dsize += v; 469 a_out_off += v; 470 } 471 /* 472 * Assign final values to symbols 473 */ 474 hdr.a_bss = dsize; 475 freezesymtab(); /* this touches hdr.a_bss */ 476 stabfix(); 477 /* 478 * Set up the relocation information "files" to 479 * be zero; outrel takes care of the rest 480 */ 481 for (locindex = 0; locindex < NLOC + NLOC; locindex++){ 482 rusefile[locindex] = (struct relbufdesc *)0; 483 } 484 } 485 486 build_hdr() 487 { 488 /* 489 * Except for the text and data relocation sizes, 490 * calculate the final values for the header 491 * 492 * Write out the initial copy; we to come 493 * back later and patch up a_trsize and a_drsize, 494 * and overwrite this first version of the header. 495 */ 496 hdr.a_magic = MAGIC; 497 hdr.a_text = tsize; 498 hdr.a_data = dsize; 499 hdr.a_bss -= dsize; 500 hdr.a_syms = sizesymtab(); /* Does not include string pool length */ 501 hdr.a_entry = 0; 502 hdr.a_trsize = 0; 503 hdr.a_drsize = 0; 504 505 bwrite((char *)&hdr, sizeof(hdr), usefile[0]); 506 } 507 508 i_pass2() 509 { 510 if (useVM == 0) { 511 fclose(tmpfil); 512 tmpfil = fopen(tmpn1, "r"); 513 if (tmpfil==NULL) { 514 yyerror("Bad pass 2 temporary file for reading %s", tmpn1); 515 delexit(); 516 } 517 } 518 } 519 520 pass2() 521 { 522 #ifdef DEBUG 523 if (debug) 524 printf("\n\n\n\t\tPASS 2\n\n\n\n"); 525 #endif DEBUG 526 passno = 2; 527 lineno = 1; 528 dotp = &usedot[0]; 529 txtfil = usefile[0]; /* already opened (always!) */ 530 relfil = 0; /* outrel takes care of the rest */ 531 initoutrel(); 532 533 inittmpfile(); 534 535 yyparse(); 536 537 closetmpfile(); 538 } 539 540 fillsegments() 541 { 542 int locindex; 543 /* 544 * Round text and data segments to FW by appending zeros 545 */ 546 for (locindex = 0; locindex < NLOC + NLOC; locindex++) { 547 if (usefile[locindex]) { 548 txtfil = usefile[locindex]; 549 dotp = &usedot[locindex]; 550 while (usedot[locindex].e_xvalue & FW) 551 outb(0); 552 } 553 } 554 } 555 556 reloc_syms() 557 { 558 u_long closerelfil(); 559 /* 560 * Move the relocation information to a.out 561 * a_out_off is the offset so far: 562 * exec + text segments + data segments 563 */ 564 relocfile = (BFILE *)Calloc(1,sizeof(BFILE)); 565 bopen(relocfile, a_out_off); 566 a_out_off += closeoutrel(relocfile); 567 568 hdr.a_trsize = trsize; 569 hdr.a_drsize = drsize; 570 if (readonlydata) { 571 hdr.a_text += hdr.a_data; 572 hdr.a_data = 0; 573 hdr.a_trsize += hdr.a_drsize; 574 hdr.a_drsize = 0; 575 } 576 /* 577 * Output the symbol table 578 * and if FLEXNAMES is set, the string pool 579 */ 580 symwrite(relocfile); 581 } 582 583 fix_a_out() 584 { 585 if (lseek(a_out_file->_file, 0L, 0) < 0) 586 yyerror("Reposition for header rewrite fails"); 587 if (write(a_out_file->_file, (char *)&hdr, sizeof (struct exec)) < 0) 588 yyerror("Rewrite of header fails"); 589 } 590 591 delexit() 592 { 593 delete(); 594 if (passno == 2){ 595 unlink(outfile); 596 } 597 exit(1); 598 } 599 600 delete() 601 { 602 if (useVM == 0 || tmpn1[0]) 603 unlink(tmpn1); 604 } 605 606 sawabort() 607 { 608 char *fillinbuffer(); 609 while (fillinbuffer() != (char *)0) 610 continue; 611 delete(); 612 exit(1); /*although the previous pass will also exit non zero*/ 613 } 614 615 panic(fmt, a1, a2, a3, a4) 616 char *fmt; 617 /*VARARGS 1*/ 618 { 619 yyerror("Assembler panic: bad internal data structure."); 620 yyerror(fmt, a1, a2, a3, a4); 621 delete(); 622 abort(); 623 } 624