gprof/gprof.c

   1 /*
   2  * Copyright (c) 1983 Regents of the University of California.
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms are permitted
   6  * provided that: (1) source distributions retain this entire copyright
   7  * notice and comment, and (2) distributions including binaries display
   8  * the following acknowledgement:  ``This product includes software
   9  * developed by the University of California, Berkeley and its contributors''
  10  * in the documentation or other materials provided with the distribution
  11  * and in all advertising materials mentioning features or use of this
  12  * software. Neither the name of the University nor the names of its
  13  * contributors may be used to endorse or promote products derived
  14  * from this software without specific prior written permission.
  15  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
  16  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
  17  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
  18  */
  19
  20 #ifndef lint
  21 char copyright[] =
  22 "@(#) Copyright (c) 1983 Regents of the University of California.\n\
  23  All rights reserved.\n";
  24 #endif /* not lint */
  25
  26 #ifndef lint
  27 static char sccsid[] = "@(#)gprof.c     5.6 (Berkeley) 6/1/90";
  28 #endif /* not lint */
  29
  30 #include "gprof.h"
  31
  32 bfd     *abfd;
  33
  34 char    *whoami = "gprof";
  35
  36     /*
  37      *  things which get -E excluded by default.
  38      */
  39 char    *defaultEs[] = { "mcount" , "__mcleanup" , 0 };
  40
  41 main(argc, argv)
  42     int argc;
  43     char **argv;
  44 {
  45     char        **sp;
  46     nltype      **timesortnlp;
  47
  48     --argc;
  49     argv++;
  50     debug = 0;
  51     bflag = TRUE;
  52     while ( *argv != 0 && **argv == '-' ) {
  53         (*argv)++;
  54         switch ( **argv ) {
  55         case 'a':
  56             aflag = TRUE;
  57             break;
  58         case 'b':
  59             bflag = FALSE;
  60             break;
  61         case 'c':
  62             cflag = TRUE;
  63             break;
  64         case 'd':
  65             dflag = TRUE;
  66             (*argv)++;
  67             debug |= atoi( *argv );
  68             debug |= ANYDEBUG;
  69 #           ifdef DEBUG
  70                 printf("[main] debug = %d\n", debug);
  71 #           else not DEBUG
  72                 printf("%s: -d ignored\n", whoami);
  73 #           endif DEBUG
  74             break;
  75         case 'E':
  76             ++argv;
  77             addlist( Elist , *argv );
  78             Eflag = TRUE;
  79             addlist( elist , *argv );
  80             eflag = TRUE;
  81             break;
  82         case 'e':
  83             addlist( elist , *++argv );
  84             eflag = TRUE;
  85             break;
  86         case 'F':
  87             ++argv;
  88             addlist( Flist , *argv );
  89             Fflag = TRUE;
  90             addlist( flist , *argv );
  91             fflag = TRUE;
  92             break;
  93         case 'f':
  94             addlist( flist , *++argv );
  95             fflag = TRUE;
  96             break;
  97         case 'k':
  98             addlist( kfromlist , *++argv );
  99             addlist( ktolist , *++argv );
 100             kflag = TRUE;
 101             break;
 102         case 's':
 103             sflag = TRUE;
 104             break;
 105         case 'z':
 106             zflag = TRUE;
 107             break;
 108         }
 109         argv++;
 110     }
 111     if ( *argv != 0 ) {
 112         a_outname  = *argv;
 113         argv++;
 114     } else {
 115         a_outname  = A_OUTNAME;
 116     }
 117     if ( *argv != 0 ) {
 118         gmonname = *argv;
 119         argv++;
 120     } else {
 121         gmonname = GMONNAME;
 122     }
 123         /*
 124          *      turn off default functions
 125          */
 126     for ( sp = &defaultEs[0] ; *sp ; sp++ ) {
 127         Eflag = TRUE;
 128         addlist( Elist , *sp );
 129         eflag = TRUE;
 130         addlist( elist , *sp );
 131     }
 132         /*
 133          *      how many ticks per second?
 134          *      if we can't tell, report time in ticks.
 135          */
 136     hz = hertz();
 137     if (hz == 0) {
 138         hz = 1;
 139         fprintf(stderr, "time is in ticks, not seconds\n");
 140     }
 141         /*
 142          *      get information about a.out file.
 143          */
 144     getnfile();
 145         /*
 146          *      get information about mon.out file(s).
 147          */
 148     do  {
 149         getpfile( gmonname );
 150         if ( *argv != 0 ) {
 151             gmonname = *argv;
 152         }
 153     } while ( *argv++ != 0 );
 154         /*
 155          *      dump out a gmon.sum file if requested
 156          */
 157     if ( sflag ) {
 158         dumpsum( GMONSUM );
 159     }
 160         /*
 161          *      assign samples to procedures
 162          */
 163     asgnsamples();
 164         /*
 165          *      assemble the dynamic profile
 166          */
 167     timesortnlp = doarcs();
 168         /*
 169          *      print the dynamic profile
 170          */
 171     printgprof( timesortnlp );
 172         /*
 173          *      print the flat profile
 174          */
 175     printprof();
 176         /*
 177          *      print the index
 178          */
 179     printindex();
 180     done();
 181 }
 182
 183     /*
 184      * Set up string and symbol tables from a.out.
 185      *  and optionally the text space.
 186      * On return symbol table is sorted by value.
 187      */
 188 getnfile()
 189 {
 190   int           valcmp();
 191
 192   abfd = bfd_openr (a_outname, NULL);
 193
 194   if (abfd == NULL) {
 195     perror (a_outname);
 196     done();
 197   }
 198
 199   if (!bfd_check_format (abfd, bfd_object)) {
 200     fprintf (stderr, "%s: %s: bad format\n", whoami, a_outname);
 201     done();
 202   }
 203
 204 /*  getstrtab(nfile); */
 205   getsymtab(abfd);
 206   gettextspace( abfd );
 207   qsort(nl, nname, sizeof(nltype), valcmp);
 208
 209 #   ifdef DEBUG
 210   if ( debug & AOUTDEBUG ) {
 211     register int j;
 212
 213     for (j = 0; j < nname; j++){
 214       printf("[getnfile] 0X%08x\t%s\n", nl[j].value, nl[j].name);
 215     }
 216   }
 217 #   endif DEBUG
 218 }
 219
 220 /*
 221  * Read in symbol table
 222  */
 223 getsymtab(abfd)
 224 bfd     *abfd;
 225 {
 226     register long       i;
 227     int                 askfor;
 228     int                 nosyms;
 229     asymbol             **syms;
 230     i = get_symtab_upper_bound (abfd);  /* This will probably give us more
 231                                          * than we need, but that's ok.
 232                                          */
 233     syms = (asymbol**)malloc (i);
 234     nosyms = bfd_canonicalize_symtab (abfd, syms);
 235
 236     nname = 0;
 237     for (i = 0; i < nosyms; i++) {
 238       if (!funcsymbol (syms[i]))
 239         continue;
 240       nname++;
 241     }
 242
 243     if (nname == 0) {
 244       fprintf(stderr, "%s: %s: no symbols\n", whoami , a_outname );
 245       done();
 246     }
 247     askfor = nname + 1;
 248     nl = (nltype *) calloc( askfor , sizeof(nltype) );
 249     if (nl == 0) {
 250         fprintf(stderr, "%s: No room for %d bytes of symbol table\n",
 251                 whoami, askfor * sizeof(nltype) );
 252         done();
 253     }
 254
 255     /* pass2 - read symbols */
 256     npe = nl;
 257     nname = 0;
 258     for (i = 0; i < nosyms; i++) {
 259       if (!funcsymbol (syms[i])) {
 260 #           ifdef DEBUG
 261         if ( debug & AOUTDEBUG ) {
 262           printf( "[getsymtab] rejecting: 0x%x %s\n" ,
 263                  syms[i]->value, syms[i]->name);
 264         }
 265 #           endif DEBUG
 266         continue;
 267       }
 268 #if 0
 269       npe->value = syms[i]->value + syms[i]->section->vma;
 270 #else
 271       npe->value = syms[i]->value;
 272 #endif
 273       npe->name = syms[i]->name;
 274 #       ifdef DEBUG
 275       if ( debug & AOUTDEBUG ) {
 276         printf( "[getsymtab] %d %s 0x%08x\n" ,
 277                nname , npe -> name , npe -> value );
 278       }
 279 #       endif DEBUG
 280       npe++;
 281       nname++;
 282     }
 283     npe->value = -1;
 284 }
 285
 286 /*
 287  *      read in the text space of an a.out file
 288  */
 289 gettextspace( abfd )
 290      bfd        *abfd;
 291 {
 292   asection      *texsec;
 293
 294   if ( cflag == 0 ) {
 295     return;
 296   }
 297
 298   texsec = bfd_get_section_by_name (abfd, ".text");
 299   if (texsec == NULL) {
 300     return;
 301   }
 302
 303   textspace = (u_char *) malloc( texsec->_cooked_size );
 304
 305   if ( textspace == 0 ) {
 306     fprintf( stderr , "%s: ran out room for %d bytes of text space:  " ,
 307             whoami , texsec->_cooked_size);
 308     fprintf( stderr , "can't do -c\n" );
 309     return;
 310   }
 311   bfd_get_section_contents (abfd, texsec, textspace, texsec->filepos,
 312                             texsec->_cooked_size);
 313 }
 314 /*
 315  *      information from a gmon.out file is in two parts:
 316  *      an array of sampling hits within pc ranges,
 317  *      and the arcs.
 318  */
 319 getpfile(filename)
 320     char *filename;
 321 {
 322     FILE                *pfile;
 323     FILE                *openpfile();
 324     struct rawarc       arc;
 325
 326     pfile = openpfile(filename);
 327     readsamples(pfile);
 328         /*
 329          *      the rest of the file consists of
 330          *      a bunch of <from,self,count> tuples.
 331          */
 332     while ( fread( &arc , sizeof arc , 1 , pfile ) == 1 ) {
 333       arc.raw_frompc = bfd_get_32 (abfd, &arc.raw_frompc);
 334       arc.raw_selfpc = bfd_get_32 (abfd, &arc.raw_selfpc);
 335       arc.raw_count = bfd_get_32 (abfd, &arc.raw_count);
 336 #       ifdef DEBUG
 337             if ( debug & SAMPLEDEBUG ) {
 338                 printf( "[getpfile] frompc 0x%x selfpc 0x%x count %d\n" ,
 339                         arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
 340             }
 341 #       endif DEBUG
 342             /*
 343              *  add this arc
 344              */
 345         tally( &arc );
 346     }
 347     fclose(pfile);
 348 }
 349
 350 FILE *
 351 openpfile(filename)
 352     char *filename;
 353 {
 354     struct hdr  tmp;
 355     FILE        *pfile;
 356
 357     if((pfile = fopen(filename, "r")) == NULL) {
 358         perror(filename);
 359         done();
 360     }
 361     fread(&tmp, sizeof(struct hdr), 1, pfile);
 362     tmp.lowpc = (char*)bfd_get_32 (abfd, &tmp.lowpc);
 363     tmp.highpc = (char*)bfd_get_32 (abfd, &tmp.highpc);
 364     tmp.ncnt = bfd_get_32 (abfd, &tmp.ncnt);
 365
 366     if ( s_highpc != 0 && ( tmp.lowpc != h.lowpc ||
 367          tmp.highpc != h.highpc || tmp.ncnt != h.ncnt ) ) {
 368         fprintf(stderr, "%s: incompatible with first gmon file\n", filename);
 369         done();
 370     }
 371     h = tmp;
 372     s_lowpc = (unsigned long) h.lowpc;
 373     s_highpc = (unsigned long) h.highpc;
 374     lowpc = (unsigned long)h.lowpc / sizeof(UNIT);
 375     highpc = (unsigned long)h.highpc / sizeof(UNIT);
 376     sampbytes = h.ncnt - sizeof(struct hdr);
 377     nsamples = sampbytes / sizeof (UNIT);
 378 #   ifdef DEBUG
 379         if ( debug & SAMPLEDEBUG ) {
 380             printf( "[openpfile] hdr.lowpc 0x%x hdr.highpc 0x%x hdr.ncnt %d\n",
 381                 h.lowpc , h.highpc , h.ncnt );
 382             printf( "[openpfile]   s_lowpc 0x%x   s_highpc 0x%x\n" ,
 383                 s_lowpc , s_highpc );
 384             printf( "[openpfile]     lowpc 0x%x     highpc 0x%x\n" ,
 385                 lowpc , highpc );
 386             printf( "[openpfile] sampbytes %d nsamples %d\n" ,
 387                 sampbytes , nsamples );
 388         }
 389 #   endif DEBUG
 390     return(pfile);
 391 }
 392
 393 tally( rawp )
 394     struct rawarc       *rawp;
 395 {
 396     nltype              *parentp;
 397     nltype              *childp;
 398
 399     parentp = nllookup( rawp -> raw_frompc );
 400     childp = nllookup( rawp -> raw_selfpc );
 401     if ( kflag
 402          && onlist( kfromlist , parentp -> name )
 403          && onlist( ktolist , childp -> name ) ) {
 404         return;
 405     }
 406     childp -> ncall += rawp -> raw_count;
 407 #   ifdef DEBUG
 408         if ( debug & TALLYDEBUG ) {
 409             printf( "[tally] arc from %s to %s traversed %d times\n" ,
 410                     parentp -> name , childp -> name , rawp -> raw_count );
 411         }
 412 #   endif DEBUG
 413     addarc( parentp , childp , rawp -> raw_count );
 414 }
 415
 416 /*
 417  * dump out the gmon.sum file
 418  */
 419 dumpsum( sumfile )
 420     char *sumfile;
 421 {
 422     register nltype *nlp;
 423     register arctype *arcp;
 424     struct rawarc arc;
 425     FILE *sfile;
 426
 427     if ( ( sfile = fopen ( sumfile , "w" ) ) == NULL ) {
 428         perror( sumfile );
 429         done();
 430     }
 431     /*
 432      * dump the header; use the last header read in
 433      */
 434     if ( fwrite( &h , sizeof h , 1 , sfile ) != 1 ) {
 435         perror( sumfile );
 436         done();
 437     }
 438     /*
 439      * dump the samples
 440      */
 441     if (fwrite(samples, sizeof (UNIT), nsamples, sfile) != nsamples) {
 442         perror( sumfile );
 443         done();
 444     }
 445     /*
 446      * dump the normalized raw arc information
 447      */
 448     for ( nlp = nl ; nlp < npe ; nlp++ ) {
 449         for ( arcp = nlp -> children ; arcp ; arcp = arcp -> arc_childlist ) {
 450             arc.raw_frompc = arcp -> arc_parentp -> value;
 451             arc.raw_selfpc = arcp -> arc_childp -> value;
 452             arc.raw_count = arcp -> arc_count;
 453             if ( fwrite ( &arc , sizeof arc , 1 , sfile ) != 1 ) {
 454                 perror( sumfile );
 455                 done();
 456             }
 457 #           ifdef DEBUG
 458                 if ( debug & SAMPLEDEBUG ) {
 459                     printf( "[dumpsum] frompc 0x%x selfpc 0x%x count %d\n" ,
 460                             arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
 461                 }
 462 #           endif DEBUG
 463         }
 464     }
 465     fclose( sfile );
 466 }
 467
 468 valcmp(p1, p2)
 469     nltype *p1, *p2;
 470 {
 471     if ( p1 -> value < p2 -> value ) {
 472         return LESSTHAN;
 473     }
 474     if ( p1 -> value > p2 -> value ) {
 475         return GREATERTHAN;
 476     }
 477     return EQUALTO;
 478 }
 479
 480 readsamples(pfile)
 481     FILE        *pfile;
 482 {
 483     register i;
 484     UNIT        sample;
 485
 486     if (samples == 0) {
 487         samples = (UNIT *) malloc (sampbytes * sizeof(UNIT));
 488         if (samples == 0) {
 489             fprintf( stderr , "%s: No room for %d sample pc's\n",
 490                 whoami , sampbytes / sizeof (UNIT));
 491             done();
 492         }
 493         memset (samples, 0, sampbytes * sizeof(UNIT));
 494     }
 495     for (i = 0; i < nsamples; i++) {
 496         fread(&sample, sizeof (UNIT), 1, pfile);
 497         sample = bfd_get_16 (abfd, &sample);
 498         if (feof(pfile))
 499                 break;
 500         samples[i] += sample;
 501     }
 502     if (i != nsamples) {
 503         fprintf(stderr,
 504             "%s: unexpected EOF after reading %d/%d samples\n",
 505                 whoami , --i , nsamples );
 506         done();
 507     }
 508 }
 509
 510 /*
 511  *      Assign samples to the procedures to which they belong.
 512  *
 513  *      There are three cases as to where pcl and pch can be
 514  *      with respect to the routine entry addresses svalue0 and svalue1
 515  *      as shown in the following diagram.  overlap computes the
 516  *      distance between the arrows, the fraction of the sample
 517  *      that is to be credited to the routine which starts at svalue0.
 518  *
 519  *          svalue0                                         svalue1
 520  *             |                                               |
 521  *             v                                               v
 522  *
 523  *             +-----------------------------------------------+
 524  *             |                                               |
 525  *        |  ->|    |<-         ->|         |<-         ->|    |<-  |
 526  *        |         |             |         |             |         |
 527  *        +---------+             +---------+             +---------+
 528  *
 529  *        ^         ^             ^         ^             ^         ^
 530  *        |         |             |         |             |         |
 531  *       pcl       pch           pcl       pch           pcl       pch
 532  *
 533  *      For the vax we assert that samples will never fall in the first
 534  *      two bytes of any routine, since that is the entry mask,
 535  *      thus we give call alignentries() to adjust the entry points if
 536  *      the entry mask falls in one bucket but the code for the routine
 537  *      doesn't start until the next bucket.  In conjunction with the
 538  *      alignment of routine addresses, this should allow us to have
 539  *      only one sample for every four bytes of text space and never
 540  *      have any overlap (the two end cases, above).
 541  */
 542 asgnsamples()
 543 {
 544     register int        j;
 545     UNIT                ccnt;
 546     double              time;
 547     unsigned long       pcl, pch;
 548     register int        i;
 549     unsigned long       overlap;
 550     unsigned long       svalue0, svalue1;
 551
 552     /* read samples and assign to namelist symbols */
 553     scale = highpc - lowpc;
 554     scale /= nsamples;
 555     alignentries();
 556     for (i = 0, j = 1; i < nsamples; i++) {
 557         ccnt = samples[i];
 558         if (ccnt == 0)
 559                 continue;
 560         pcl = lowpc + scale * i;
 561         pch = lowpc + scale * (i + 1);
 562         time = ccnt;
 563 #       ifdef DEBUG
 564             if ( debug & SAMPLEDEBUG ) {
 565                 printf( "[asgnsamples] pcl 0x%x pch 0x%x ccnt %d\n" ,
 566                         pcl , pch , ccnt );
 567             }
 568 #       endif DEBUG
 569         totime += time;
 570         for (j = j - 1; j < nname; j++) {
 571             svalue0 = nl[j].svalue;
 572             svalue1 = nl[j+1].svalue;
 573                 /*
 574                  *      if high end of tick is below entry address,
 575                  *      go for next tick.
 576                  */
 577             if (pch < svalue0)
 578                     break;
 579                 /*
 580                  *      if low end of tick into next routine,
 581                  *      go for next routine.
 582                  */
 583             if (pcl >= svalue1)
 584                     continue;
 585             overlap = min(pch, svalue1) - max(pcl, svalue0);
 586             if (overlap > 0) {
 587 #               ifdef DEBUG
 588                     if (debug & SAMPLEDEBUG) {
 589                         printf("[asgnsamples] (0x%x->0x%x-0x%x) %s gets %f ticks %d overlap\n",
 590                                 nl[j].value/sizeof(UNIT), svalue0, svalue1,
 591                                 nl[j].name,
 592                                 overlap * time / scale, overlap);
 593                     }
 594 #               endif DEBUG
 595                 nl[j].time += overlap * time / scale;
 596             }
 597         }
 598     }
 599 #   ifdef DEBUG
 600         if (debug & SAMPLEDEBUG) {
 601             printf("[asgnsamples] totime %f\n", totime);
 602         }
 603 #   endif DEBUG
 604 }
 605
 606
 607 unsigned long
 608 min(a, b)
 609     unsigned long a,b;
 610 {
 611     if (a<b)
 612         return(a);
 613     return(b);
 614 }
 615
 616 unsigned long
 617 max(a, b)
 618     unsigned long a,b;
 619 {
 620     if (a>b)
 621         return(a);
 622     return(b);
 623 }
 624
 625     /*
 626      *  calculate scaled entry point addresses (to save time in asgnsamples),
 627      *  and possibly push the scaled entry points over the entry mask,
 628      *  if it turns out that the entry point is in one bucket and the code
 629      *  for a routine is in the next bucket.
 630      */
 631 alignentries()
 632 {
 633     register struct nl  *nlp;
 634     unsigned long       bucket_of_entry;
 635     unsigned long       bucket_of_code;
 636
 637     for (nlp = nl; nlp < npe; nlp++) {
 638         nlp -> svalue = nlp -> value / sizeof(UNIT);
 639         bucket_of_entry = (nlp->svalue - lowpc) / scale;
 640         bucket_of_code = (nlp->svalue + UNITS_TO_CODE - lowpc) / scale;
 641         if (bucket_of_entry < bucket_of_code) {
 642 #           ifdef DEBUG
 643                 if (debug & SAMPLEDEBUG) {
 644                     printf("[alignentries] pushing svalue 0x%x to 0x%x\n",
 645                             nlp->svalue, nlp->svalue + UNITS_TO_CODE);
 646                 }
 647 #           endif DEBUG
 648             nlp->svalue += UNITS_TO_CODE;
 649         }
 650     }
 651 }
 652
 653 bool
 654 funcsymbol( symp )
 655      asymbol *symp;
 656 {
 657   extern char   *strtab;        /* string table from a.out */
 658   extern int    aflag;          /* if static functions aren't desired */
 659   char  *name;
 660   int i;
 661
 662   /*
 663    *    must be a text symbol,
 664    *    and static text symbols don't qualify if aflag set.
 665    */
 666
 667
 668   if (!symp->section)
 669     return FALSE;
 670
 671 #if 0
 672   if (!aflag && (symp->flags&BSF_LOCAL)) {
 673 #if defined(DEBUG)
 674     fprintf (stderr, "%s(%d):  %s:  not a function\n", __FILE__, __LINE__, symp->name);
 675 #endif
 676     return FALSE;
 677   }
 678 #endif  /* 0 */
 679   /*
 680    *    can't have any `funny' characters in name,
 681    *    where `funny' includes  `.', .o file names
 682    *                    and     `$', pascal labels.
 683    */
 684   if (!symp->name)
 685     return FALSE;
 686
 687   for (name = symp->name; *name; name++) {
 688     if ( *name == '.' || *name == '$' ) {
 689       return FALSE;
 690     }
 691   }
 692
 693   i = bfd_decode_symclass (symp);
 694 #if defined(DEBUG) && 0
 695   if (i != 'T' && i != 't')
 696     fprintf (stderr, "%s(%d):  %s is of class %c\n", __FILE__, __LINE__, symp->name, i);
 697 #endif
 698
 699   return (i == 'T' || i == 't');
 700 }
 701
 702 done()
 703 {
 704
 705     exit(0);
 706 }