1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1983, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #ifndef lint 33 static const char copyright[] = 34 "@(#) Copyright (c) 1983, 1993\n\ 35 The Regents of the University of California. All rights reserved.\n"; 36 #endif /* not lint */ 37 38 #if 0 39 #ifndef lint 40 static char sccsid[] = "@(#)gprof.c 8.1 (Berkeley) 6/6/93"; 41 #endif /* not lint */ 42 #endif 43 44 #include <sys/cdefs.h> 45 __FBSDID("$FreeBSD$"); 46 47 #include <err.h> 48 #include <limits.h> 49 #include <stdint.h> 50 #include <string.h> 51 52 #include "gprof.h" 53 54 static int valcmp(const void *, const void *); 55 56 static struct gmonhdr gmonhdr; 57 static int lflag; 58 static int Lflag; 59 60 int 61 main(int argc, char **argv) 62 { 63 char **sp; 64 nltype **timesortnlp; 65 char **defaultEs; 66 67 --argc; 68 argv++; 69 debug = 0; 70 bflag = TRUE; 71 while ( *argv != 0 && **argv == '-' ) { 72 (*argv)++; 73 switch ( **argv ) { 74 case 'a': 75 aflag = TRUE; 76 break; 77 case 'b': 78 bflag = FALSE; 79 break; 80 case 'C': 81 Cflag = TRUE; 82 cyclethreshold = atoi( *++argv ); 83 break; 84 case 'd': 85 dflag = TRUE; 86 setlinebuf(stdout); 87 debug |= atoi( *++argv ); 88 debug |= ANYDEBUG; 89 # ifdef DEBUG 90 printf("[main] debug = %d\n", debug); 91 # else /* not DEBUG */ 92 printf("gprof: -d ignored\n"); 93 # endif /* DEBUG */ 94 break; 95 case 'E': 96 ++argv; 97 addlist( Elist , *argv ); 98 Eflag = TRUE; 99 addlist( elist , *argv ); 100 eflag = TRUE; 101 break; 102 case 'e': 103 addlist( elist , *++argv ); 104 eflag = TRUE; 105 break; 106 case 'F': 107 ++argv; 108 addlist( Flist , *argv ); 109 Fflag = TRUE; 110 addlist( flist , *argv ); 111 fflag = TRUE; 112 break; 113 case 'f': 114 addlist( flist , *++argv ); 115 fflag = TRUE; 116 break; 117 case 'k': 118 addlist( kfromlist , *++argv ); 119 addlist( ktolist , *++argv ); 120 kflag = TRUE; 121 break; 122 case 'K': 123 Kflag = TRUE; 124 break; 125 case 'l': 126 lflag = 1; 127 Lflag = 0; 128 break; 129 case 'L': 130 Lflag = 1; 131 lflag = 0; 132 break; 133 case 's': 134 sflag = TRUE; 135 break; 136 case 'u': 137 uflag = TRUE; 138 break; 139 case 'z': 140 zflag = TRUE; 141 break; 142 } 143 argv++; 144 } 145 if ( *argv != 0 ) { 146 a_outname = *argv; 147 argv++; 148 } else { 149 a_outname = A_OUTNAME; 150 } 151 if ( *argv != 0 ) { 152 gmonname = *argv; 153 argv++; 154 } else { 155 gmonname = (char *) malloc(strlen(a_outname)+6); 156 strcpy(gmonname, a_outname); 157 strcat(gmonname, ".gmon"); 158 } 159 /* 160 * get information from the executable file. 161 */ 162 if ((Kflag && kernel_getnfile(a_outname, &defaultEs) == -1) || 163 (!Kflag && elf_getnfile(a_outname, &defaultEs) == -1 && 164 aout_getnfile(a_outname, &defaultEs) == -1)) 165 errx(1, "%s: bad format", a_outname); 166 /* 167 * sort symbol table. 168 */ 169 qsort(nl, nname, sizeof(nltype), valcmp); 170 /* 171 * turn off default functions 172 */ 173 for ( sp = defaultEs ; *sp ; sp++ ) { 174 Eflag = TRUE; 175 addlist( Elist , *sp ); 176 eflag = TRUE; 177 addlist( elist , *sp ); 178 } 179 /* 180 * get information about mon.out file(s). 181 */ 182 do { 183 getpfile( gmonname ); 184 if ( *argv != 0 ) { 185 gmonname = *argv; 186 } 187 } while ( *argv++ != 0 ); 188 /* 189 * how many ticks per second? 190 * if we can't tell, report time in ticks. 191 */ 192 if (hz == 0) { 193 hz = 1; 194 fprintf(stderr, "time is in ticks, not seconds\n"); 195 } 196 /* 197 * dump out a gmon.sum file if requested 198 */ 199 if ( sflag ) { 200 dumpsum( GMONSUM ); 201 } 202 /* 203 * assign samples to procedures 204 */ 205 asgnsamples(); 206 /* 207 * assemble the dynamic profile 208 */ 209 timesortnlp = doarcs(); 210 /* 211 * print the dynamic profile 212 */ 213 if(!lflag) { 214 printgprof( timesortnlp ); 215 } 216 /* 217 * print the flat profile 218 */ 219 if(!Lflag) { 220 printprof(); 221 } 222 /* 223 * print the index 224 */ 225 printindex(); 226 exit(0); 227 } 228 229 /* 230 * information from a gmon.out file is in two parts: 231 * an array of sampling hits within pc ranges, 232 * and the arcs. 233 */ 234 void 235 getpfile(char *filename) 236 { 237 FILE *pfile; 238 struct rawarc arc; 239 240 pfile = openpfile(filename); 241 readsamples(pfile); 242 /* 243 * the rest of the file consists of 244 * a bunch of <from,self,count> tuples. 245 */ 246 while ( fread( &arc , sizeof arc , 1 , pfile ) == 1 ) { 247 # ifdef DEBUG 248 if ( debug & SAMPLEDEBUG ) { 249 printf( "[getpfile] frompc 0x%lx selfpc 0x%lx count %ld\n" , 250 arc.raw_frompc , arc.raw_selfpc , arc.raw_count ); 251 } 252 # endif /* DEBUG */ 253 /* 254 * add this arc 255 */ 256 tally( &arc ); 257 } 258 fclose(pfile); 259 } 260 261 FILE * 262 openpfile(char *filename) 263 { 264 struct gmonhdr tmp; 265 FILE *pfile; 266 int size; 267 int rate; 268 269 if((pfile = fopen(filename, "r")) == NULL) 270 err(1, "%s", filename); 271 fread(&tmp, sizeof(struct gmonhdr), 1, pfile); 272 if ( s_highpc != 0 && ( tmp.lpc != gmonhdr.lpc || 273 tmp.hpc != gmonhdr.hpc || tmp.ncnt != gmonhdr.ncnt ) ) 274 errx(1, "%s: incompatible with first gmon file", filename); 275 gmonhdr = tmp; 276 if ( gmonhdr.version == GMONVERSION ) { 277 rate = gmonhdr.profrate; 278 size = sizeof(struct gmonhdr); 279 } else { 280 fseek(pfile, sizeof(struct ophdr), SEEK_SET); 281 size = sizeof(struct ophdr); 282 gmonhdr.profrate = rate = hertz(); 283 gmonhdr.version = GMONVERSION; 284 } 285 if (hz == 0) { 286 hz = rate; 287 } else if (hz != rate) 288 errx(0, "%s: profile clock rate (%d) %s (%ld) in first gmon file", 289 filename, rate, "incompatible with clock rate", hz); 290 if ( gmonhdr.histcounter_type == 0 ) { 291 /* Historical case. The type was u_short (2 bytes in practice). */ 292 histcounter_type = 16; 293 histcounter_size = 2; 294 } else { 295 histcounter_type = gmonhdr.histcounter_type; 296 histcounter_size = abs(histcounter_type) / CHAR_BIT; 297 } 298 s_lowpc = (unsigned long) gmonhdr.lpc; 299 s_highpc = (unsigned long) gmonhdr.hpc; 300 lowpc = (unsigned long)gmonhdr.lpc / HISTORICAL_SCALE_2; 301 highpc = (unsigned long)gmonhdr.hpc / HISTORICAL_SCALE_2; 302 sampbytes = gmonhdr.ncnt - size; 303 nsamples = sampbytes / histcounter_size; 304 # ifdef DEBUG 305 if ( debug & SAMPLEDEBUG ) { 306 printf( "[openpfile] hdr.lpc 0x%lx hdr.hpc 0x%lx hdr.ncnt %d\n", 307 gmonhdr.lpc , gmonhdr.hpc , gmonhdr.ncnt ); 308 printf( "[openpfile] s_lowpc 0x%lx s_highpc 0x%lx\n" , 309 s_lowpc , s_highpc ); 310 printf( "[openpfile] lowpc 0x%lx highpc 0x%lx\n" , 311 lowpc , highpc ); 312 printf( "[openpfile] sampbytes %d nsamples %d\n" , 313 sampbytes , nsamples ); 314 printf( "[openpfile] sample rate %ld\n" , hz ); 315 } 316 # endif /* DEBUG */ 317 return(pfile); 318 } 319 320 void 321 tally(struct rawarc *rawp) 322 { 323 nltype *parentp; 324 nltype *childp; 325 326 parentp = nllookup( rawp -> raw_frompc ); 327 childp = nllookup( rawp -> raw_selfpc ); 328 if ( parentp == 0 || childp == 0 ) 329 return; 330 if ( kflag 331 && onlist( kfromlist , parentp -> name ) 332 && onlist( ktolist , childp -> name ) ) { 333 return; 334 } 335 childp -> ncall += rawp -> raw_count; 336 # ifdef DEBUG 337 if ( debug & TALLYDEBUG ) { 338 printf( "[tally] arc from %s to %s traversed %ld times\n" , 339 parentp -> name , childp -> name , rawp -> raw_count ); 340 } 341 # endif /* DEBUG */ 342 addarc( parentp , childp , rawp -> raw_count ); 343 } 344 345 /* 346 * dump out the gmon.sum file 347 */ 348 void 349 dumpsum(const char *sumfile) 350 { 351 register nltype *nlp; 352 register arctype *arcp; 353 struct rawarc arc; 354 FILE *sfile; 355 356 if ( ( sfile = fopen ( sumfile , "w" ) ) == NULL ) 357 err( 1 , "%s" , sumfile ); 358 /* 359 * dump the header; use the last header read in 360 */ 361 if ( fwrite( &gmonhdr , sizeof gmonhdr , 1 , sfile ) != 1 ) 362 err( 1 , "%s" , sumfile ); 363 /* 364 * dump the samples 365 */ 366 if (fwrite(samples, histcounter_size, nsamples, sfile) != nsamples) 367 err( 1 , "%s" , sumfile ); 368 /* 369 * dump the normalized raw arc information 370 */ 371 for ( nlp = nl ; nlp < npe ; nlp++ ) { 372 for ( arcp = nlp -> children ; arcp ; arcp = arcp -> arc_childlist ) { 373 arc.raw_frompc = arcp -> arc_parentp -> value; 374 arc.raw_selfpc = arcp -> arc_childp -> value; 375 arc.raw_count = arcp -> arc_count; 376 if ( fwrite ( &arc , sizeof arc , 1 , sfile ) != 1 ) 377 err( 1 , "%s" , sumfile ); 378 # ifdef DEBUG 379 if ( debug & SAMPLEDEBUG ) { 380 printf( "[dumpsum] frompc 0x%lx selfpc 0x%lx count %ld\n" , 381 arc.raw_frompc , arc.raw_selfpc , arc.raw_count ); 382 } 383 # endif /* DEBUG */ 384 } 385 } 386 fclose( sfile ); 387 } 388 389 static int 390 valcmp(const void *v1, const void *v2) 391 { 392 const nltype *p1 = (const nltype *)v1; 393 const nltype *p2 = (const nltype *)v2; 394 395 if ( p1 -> value < p2 -> value ) { 396 return LESSTHAN; 397 } 398 if ( p1 -> value > p2 -> value ) { 399 return GREATERTHAN; 400 } 401 return EQUALTO; 402 } 403 404 void 405 readsamples(FILE *pfile) 406 { 407 int i; 408 intmax_t sample; 409 410 if (samples == 0) { 411 samples = (double *) calloc(nsamples, sizeof(double)); 412 if (samples == NULL) 413 errx(0, "no room for %d sample pc's", nsamples); 414 } 415 for (i = 0; i < nsamples; i++) { 416 fread(&sample, histcounter_size, 1, pfile); 417 if (feof(pfile)) 418 break; 419 switch ( histcounter_type ) { 420 case -8: 421 samples[i] += *(int8_t *)&sample; 422 break; 423 case 8: 424 samples[i] += *(u_int8_t *)&sample; 425 break; 426 case -16: 427 samples[i] += *(int16_t *)&sample; 428 break; 429 case 16: 430 samples[i] += *(u_int16_t *)&sample; 431 break; 432 case -32: 433 samples[i] += *(int32_t *)&sample; 434 break; 435 case 32: 436 samples[i] += *(u_int32_t *)&sample; 437 break; 438 case -64: 439 samples[i] += *(int64_t *)&sample; 440 break; 441 case 64: 442 samples[i] += *(u_int64_t *)&sample; 443 break; 444 default: 445 err(1, "unsupported histogram counter type %d", histcounter_type); 446 } 447 } 448 if (i != nsamples) 449 errx(1, "unexpected EOF after reading %d/%d samples", --i , nsamples ); 450 } 451 452 /* 453 * Assign samples to the procedures to which they belong. 454 * 455 * There are three cases as to where pcl and pch can be 456 * with respect to the routine entry addresses svalue0 and svalue1 457 * as shown in the following diagram. overlap computes the 458 * distance between the arrows, the fraction of the sample 459 * that is to be credited to the routine which starts at svalue0. 460 * 461 * svalue0 svalue1 462 * | | 463 * v v 464 * 465 * +-----------------------------------------------+ 466 * | | 467 * | ->| |<- ->| |<- ->| |<- | 468 * | | | | | | 469 * +---------+ +---------+ +---------+ 470 * 471 * ^ ^ ^ ^ ^ ^ 472 * | | | | | | 473 * pcl pch pcl pch pcl pch 474 * 475 * For the vax we assert that samples will never fall in the first 476 * two bytes of any routine, since that is the entry mask, 477 * thus we give call alignentries() to adjust the entry points if 478 * the entry mask falls in one bucket but the code for the routine 479 * doesn't start until the next bucket. In conjunction with the 480 * alignment of routine addresses, this should allow us to have 481 * only one sample for every four bytes of text space and never 482 * have any overlap (the two end cases, above). 483 */ 484 void 485 asgnsamples(void) 486 { 487 register int j; 488 double ccnt; 489 double thetime; 490 unsigned long pcl, pch; 491 register int i; 492 unsigned long overlap; 493 unsigned long svalue0, svalue1; 494 495 /* read samples and assign to namelist symbols */ 496 scale = highpc - lowpc; 497 scale /= nsamples; 498 alignentries(); 499 for (i = 0, j = 1; i < nsamples; i++) { 500 ccnt = samples[i]; 501 if (ccnt == 0) 502 continue; 503 pcl = lowpc + (unsigned long)(scale * i); 504 pch = lowpc + (unsigned long)(scale * (i + 1)); 505 thetime = ccnt; 506 # ifdef DEBUG 507 if ( debug & SAMPLEDEBUG ) { 508 printf( "[asgnsamples] pcl 0x%lx pch 0x%lx ccnt %.0f\n" , 509 pcl , pch , ccnt ); 510 } 511 # endif /* DEBUG */ 512 totime += thetime; 513 for (j = j - 1; j < nname; j++) { 514 svalue0 = nl[j].svalue; 515 svalue1 = nl[j+1].svalue; 516 /* 517 * if high end of tick is below entry address, 518 * go for next tick. 519 */ 520 if (pch < svalue0) 521 break; 522 /* 523 * if low end of tick into next routine, 524 * go for next routine. 525 */ 526 if (pcl >= svalue1) 527 continue; 528 overlap = min(pch, svalue1) - max(pcl, svalue0); 529 if (overlap > 0) { 530 # ifdef DEBUG 531 if (debug & SAMPLEDEBUG) { 532 printf("[asgnsamples] (0x%lx->0x%lx-0x%lx) %s gets %f ticks %lu overlap\n", 533 nl[j].value / HISTORICAL_SCALE_2, 534 svalue0, svalue1, nl[j].name, 535 overlap * thetime / scale, overlap); 536 } 537 # endif /* DEBUG */ 538 nl[j].time += overlap * thetime / scale; 539 } 540 } 541 } 542 # ifdef DEBUG 543 if (debug & SAMPLEDEBUG) { 544 printf("[asgnsamples] totime %f\n", totime); 545 } 546 # endif /* DEBUG */ 547 } 548 549 550 unsigned long 551 min(unsigned long a, unsigned long b) 552 { 553 if (a<b) 554 return(a); 555 return(b); 556 } 557 558 unsigned long 559 max(unsigned long a, unsigned long b) 560 { 561 if (a>b) 562 return(a); 563 return(b); 564 } 565 566 /* 567 * calculate scaled entry point addresses (to save time in asgnsamples), 568 * and possibly push the scaled entry points over the entry mask, 569 * if it turns out that the entry point is in one bucket and the code 570 * for a routine is in the next bucket. 571 */ 572 void 573 alignentries(void) 574 { 575 register struct nl *nlp; 576 unsigned long bucket_of_entry; 577 unsigned long bucket_of_code; 578 579 for (nlp = nl; nlp < npe; nlp++) { 580 nlp -> svalue = nlp -> value / HISTORICAL_SCALE_2; 581 bucket_of_entry = (nlp->svalue - lowpc) / scale; 582 bucket_of_code = (nlp->svalue + OFFSET_OF_CODE / HISTORICAL_SCALE_2 - 583 lowpc) / scale; 584 if (bucket_of_entry < bucket_of_code) { 585 # ifdef DEBUG 586 if (debug & SAMPLEDEBUG) { 587 printf("[alignentries] pushing svalue 0x%lx to 0x%lx\n", 588 nlp->svalue, 589 nlp->svalue + OFFSET_OF_CODE / HISTORICAL_SCALE_2); 590 } 591 # endif /* DEBUG */ 592 nlp->svalue += OFFSET_OF_CODE / HISTORICAL_SCALE_2; 593 } 594 } 595 } 596