2 * Copyright (c) 1983 Regents of the University of California.
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.
22 "@(#) Copyright (c) 1983 Regents of the University of California.\n\
23 All rights reserved.\n";
27 static char sccsid[] = "@(#)gprof.c 5.6 (Berkeley) 6/1/90";
37 * things which get -E excluded by default.
39 char *defaultEs[] = { "mcount" , "__mcleanup" , 0 };
41 int discard_underscores = 1; /* Should we discard initial underscores? */
42 int bsd_style_output = 0; /* As opposed to FSF style output */
56 while ( *argv != 0 && **argv == '-' ) {
71 debug |= atoi( *argv );
74 printf("[main] debug = %d\n", debug);
76 printf("%s: -d ignored\n", whoami);
81 addlist( Elist , *argv );
83 addlist( elist , *argv );
87 addlist( elist , *++argv );
92 addlist( Flist , *argv );
94 addlist( flist , *argv );
98 addlist( flist , *++argv );
102 addlist( kfromlist , *++argv );
103 addlist( ktolist , *++argv );
109 case 'T': /* "Traditional" output format */
110 bsd_style_output = 1;
117 Usage: %s [-a] [-b] [-c] [-d[num]] [-E function-name] [-e function-name]\n\
118 [-F function-name] [-f function-name] [-k from to] [-s] [-T] [-z]\n\
119 [image-file] [profile-file...]\n", whoami);
128 a_outname = A_OUTNAME;
137 * turn off default functions
139 for ( sp = &defaultEs[0] ; *sp ; sp++ ) {
141 addlist( Elist , *sp );
143 addlist( elist , *sp );
146 * how many ticks per second?
147 * if we can't tell, report time in ticks.
152 fprintf(stderr, "time is in ticks, not seconds\n");
155 * get information about a.out file.
159 * get information about mon.out file(s).
162 getpfile( gmonname );
166 } while ( *argv++ != 0 );
168 * dump out a gmon.sum file if requested
174 * assign samples to procedures
178 * assemble the dynamic profile
180 timesortnlp = doarcs();
182 if (bsd_style_output) {
183 printgprof( timesortnlp ); /* print the dynamic profile */
184 printprof(); /* print the flat profile */
186 printprof(); /* print the flat profile */
187 printgprof( timesortnlp ); /* print the dynamic profile */
197 * Set up string and symbol tables from a.out.
198 * and optionally the text space.
199 * On return symbol table is sorted by value.
205 abfd = bfd_openr (a_outname, NULL);
212 if (!bfd_check_format (abfd, bfd_object)) {
213 fprintf (stderr, "%s: %s: bad format\n", whoami, a_outname);
217 /* getstrtab(nfile); */
219 gettextspace( abfd );
220 qsort(nl, nname, sizeof(nltype), valcmp);
223 if ( debug & AOUTDEBUG ) {
226 for (j = 0; j < nname; j++){
227 printf("[getnfile] 0X%08x\t%s\n", nl[j].value, nl[j].name);
234 * Read in symbol table
243 i = get_symtab_upper_bound (abfd); /* This will probably give us more
244 * than we need, but that's ok.
246 syms = (asymbol**)xmalloc (i);
247 nosyms = bfd_canonicalize_symtab (abfd, syms);
250 for (i = 0; i < nosyms; i++) {
251 if (!funcsymbol (syms[i]))
257 fprintf(stderr, "%s: %s: no symbols\n", whoami , a_outname );
261 nl = (nltype *) calloc( askfor , sizeof(nltype) );
263 fprintf(stderr, "%s: No room for %d bytes of symbol table\n",
264 whoami, askfor * sizeof(nltype) );
268 /* pass2 - read symbols */
271 for (i = 0; i < nosyms; i++) {
272 if (!funcsymbol (syms[i])) {
274 if ( debug & AOUTDEBUG ) {
275 printf( "[getsymtab] rejecting: 0x%x %s\n" ,
276 syms[i]->value, syms[i]->name);
281 /* Symbol offsets are always section-relative. */
282 npe->value = syms[i]->value + syms[i]->section->vma;
283 npe->name = syms[i]->name;
285 /* If we see "main" without an initial '_', we assume
286 names are *not* prefixed by '_'. */
287 if (npe->name[0] == 'm' && discard_underscores
288 && strcmp(npe->name, "main") == 0)
289 discard_underscores = 0;
292 if ( debug & AOUTDEBUG ) {
293 printf( "[getsymtab] %d %s 0x%08x\n" ,
294 nname , npe -> name , npe -> value );
304 * read in the text space of an a.out file
315 texsec = bfd_get_section_by_name (abfd, ".text");
316 if (texsec == NULL) {
320 textspace = (u_char *) malloc( texsec->_cooked_size );
322 if ( textspace == 0 ) {
323 fprintf( stderr , "%s: ran out room for %d bytes of text space: " ,
324 whoami , texsec->_cooked_size);
325 fprintf( stderr , "can't do -c\n" );
328 bfd_get_section_contents (abfd, texsec, textspace, texsec->filepos,
329 texsec->_cooked_size);
332 * information from a gmon.out file is in two parts:
333 * an array of sampling hits within pc ranges,
342 struct veryrawarc rawarc;
344 pfile = openpfile(filename);
347 * the rest of the file consists of
348 * a bunch of <from,self,count> tuples.
350 while ( fread( &rawarc , sizeof rawarc , 1 , pfile ) == 1 ) {
351 arc.raw_frompc = bfd_get_32 (abfd, (bfd_byte *) rawarc.raw_frompc);
352 arc.raw_selfpc = bfd_get_32 (abfd, (bfd_byte *) rawarc.raw_selfpc);
353 arc.raw_count = bfd_get_32 (abfd, (bfd_byte *) rawarc.raw_count);
355 if ( debug & SAMPLEDEBUG ) {
356 printf( "[getpfile] frompc 0x%x selfpc 0x%x count %d\n" ,
357 arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
376 if((pfile = fopen(filename, "r")) == NULL) {
380 if (sizeof(struct rawhdr) != fread(&raw, 1, sizeof(struct rawhdr), pfile))
382 fprintf(stderr, "%s: file too short to be a gmon file\n", filename);
385 tmp.lowpc = (UNIT *)bfd_get_32 (abfd, (bfd_byte *) &raw.lowpc[0]);
386 tmp.highpc = (UNIT *)bfd_get_32 (abfd, (bfd_byte *) &raw.highpc[0]);
387 tmp.ncnt = bfd_get_32 (abfd, (bfd_byte *) &raw.ncnt[0]);
389 if ( s_highpc != 0 && ( tmp.lowpc != h.lowpc ||
390 tmp.highpc != h.highpc || tmp.ncnt != h.ncnt ) ) {
391 fprintf(stderr, "%s: incompatible with first gmon file\n", filename);
395 s_lowpc = (unsigned long) h.lowpc;
396 s_highpc = (unsigned long) h.highpc;
397 lowpc = (unsigned long)h.lowpc / sizeof(UNIT);
398 highpc = (unsigned long)h.highpc / sizeof(UNIT);
399 sampbytes = h.ncnt - sizeof(struct rawhdr);
400 nsamples = sampbytes / sizeof (UNIT);
402 if ( debug & SAMPLEDEBUG ) {
403 printf( "[openpfile] hdr.lowpc 0x%x hdr.highpc 0x%x hdr.ncnt %d\n",
404 h.lowpc , h.highpc , h.ncnt );
405 printf( "[openpfile] s_lowpc 0x%x s_highpc 0x%x\n" ,
406 s_lowpc , s_highpc );
407 printf( "[openpfile] lowpc 0x%x highpc 0x%x\n" ,
409 printf( "[openpfile] sampbytes %d nsamples %d\n" ,
410 sampbytes , nsamples );
422 parentp = nllookup( rawp -> raw_frompc );
423 childp = nllookup( rawp -> raw_selfpc );
425 && onlist( kfromlist , parentp -> name )
426 && onlist( ktolist , childp -> name ) ) {
429 childp -> ncall += rawp -> raw_count;
431 if ( debug & TALLYDEBUG ) {
432 printf( "[tally] arc from %s to %s traversed %d times\n" ,
433 parentp -> name , childp -> name , rawp -> raw_count );
436 addarc( parentp , childp , rawp -> raw_count );
440 * dump out the gmon.sum file
445 register nltype *nlp;
446 register arctype *arcp;
450 if ( ( sfile = fopen ( sumfile , "w" ) ) == NULL ) {
455 * dump the header; use the last header read in
457 if ( fwrite( &h , sizeof h , 1 , sfile ) != 1 ) {
464 if (fwrite(samples, sizeof (UNIT), nsamples, sfile) != nsamples) {
469 * dump the normalized raw arc information
471 for ( nlp = nl ; nlp < npe ; nlp++ ) {
472 for ( arcp = nlp -> children ; arcp ; arcp = arcp -> arc_childlist ) {
473 arc.raw_frompc = arcp -> arc_parentp -> value;
474 arc.raw_selfpc = arcp -> arc_childp -> value;
475 arc.raw_count = arcp -> arc_count;
476 if ( fwrite ( &arc , sizeof arc , 1 , sfile ) != 1 ) {
481 if ( debug & SAMPLEDEBUG ) {
482 printf( "[dumpsum] frompc 0x%x selfpc 0x%x count %d\n" ,
483 arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
494 if ( p1 -> value < p2 -> value ) {
497 if ( p1 -> value > p2 -> value ) {
510 samples = (int *) calloc (nsamples, sizeof(int));
512 fprintf( stderr , "%s: No room for %d sample pc's\n",
517 for (i = 0; i < nsamples; i++) {
521 fread(raw, sizeof (raw), 1, pfile);
522 value = bfd_get_16 (abfd, (bfd_byte *) raw);
529 "%s: unexpected EOF after reading %d/%d samples\n",
530 whoami , --i , nsamples );
536 * Assign samples to the procedures to which they belong.
538 * There are three cases as to where pcl and pch can be
539 * with respect to the routine entry addresses svalue0 and svalue1
540 * as shown in the following diagram. overlap computes the
541 * distance between the arrows, the fraction of the sample
542 * that is to be credited to the routine which starts at svalue0.
548 * +-----------------------------------------------+
550 * | ->| |<- ->| |<- ->| |<- |
552 * +---------+ +---------+ +---------+
556 * pcl pch pcl pch pcl pch
558 * For the vax we assert that samples will never fall in the first
559 * two bytes of any routine, since that is the entry mask,
560 * thus we give call alignentries() to adjust the entry points if
561 * the entry mask falls in one bucket but the code for the routine
562 * doesn't start until the next bucket. In conjunction with the
563 * alignment of routine addresses, this should allow us to have
564 * only one sample for every four bytes of text space and never
565 * have any overlap (the two end cases, above).
572 unsigned long pcl, pch;
574 unsigned long overlap;
575 unsigned long svalue0, svalue1;
577 /* read samples and assign to namelist symbols */
578 scale = highpc - lowpc;
579 scale /= nsamples - 1;
581 for (i = 0, j = 1; i < nsamples; i++) {
585 pcl = lowpc + scale * i;
586 pch = lowpc + scale * (i + 1);
589 if ( debug & SAMPLEDEBUG ) {
590 printf( "[asgnsamples] pcl 0x%x pch 0x%x ccnt %d\n" ,
595 for (j = j - 1; j < nname; j++) {
596 svalue0 = nl[j].svalue;
597 svalue1 = nl[j+1].svalue;
599 * if high end of tick is below entry address,
605 * if low end of tick into next routine,
606 * go for next routine.
610 overlap = min(pch, svalue1) - max(pcl, svalue0);
613 if (debug & SAMPLEDEBUG) {
614 printf("[asgnsamples] (0x%x->0x%x-0x%x) %s gets %f ticks %d overlap\n",
615 nl[j].value/sizeof(UNIT), svalue0, svalue1,
617 overlap * time / scale, overlap);
620 nl[j].time += overlap * time / scale;
625 if (debug & SAMPLEDEBUG) {
626 printf("[asgnsamples] totime %f\n", totime);
651 * calculate scaled entry point addresses (to save time in asgnsamples),
652 * and possibly push the scaled entry points over the entry mask,
653 * if it turns out that the entry point is in one bucket and the code
654 * for a routine is in the next bucket.
658 register struct nl *nlp;
659 unsigned long bucket_of_entry;
660 unsigned long bucket_of_code;
662 for (nlp = nl; nlp < npe; nlp++) {
663 nlp -> svalue = nlp -> value / sizeof(UNIT);
664 bucket_of_entry = (nlp->svalue - lowpc) / scale;
665 bucket_of_code = (nlp->svalue + UNITS_TO_CODE - lowpc) / scale;
666 if (bucket_of_entry < bucket_of_code) {
668 if (debug & SAMPLEDEBUG) {
669 printf("[alignentries] pushing svalue 0x%x to 0x%x\n",
670 nlp->svalue, nlp->svalue + UNITS_TO_CODE);
673 nlp->svalue += UNITS_TO_CODE;
682 extern char *strtab; /* string table from a.out */
683 extern int aflag; /* if static functions aren't desired */
689 * must be a text symbol,
690 * and static text symbols don't qualify if aflag set.
697 if (aflag && (symp->flags&BSF_LOCAL)) {
699 fprintf (stderr, "%s(%d): %s: not a function\n", __FILE__, __LINE__, symp->name);
704 symprefix = bfd_get_symbol_leading_char (abfd);
705 i = bfd_decode_symclass (symp);
706 #if defined(DEBUG) && 0
707 if (i != 'T' && i != 't')
708 fprintf (stderr, "%s(%d): %s is of class %c\n", __FILE__, __LINE__, symp->name, i);
712 * Any external text symbol should be okay. (Only problem would be
713 * variables in the text section.)
720 * 't' is static text; -a says to ignore it. So if it's not
721 * a static text symbol, *or* it is and the user gave -a, we
725 if (i != 't' || aflag)
729 * can't have any `funny' characters in name,
730 * where `funny' includes `.', .o file names
731 * and `$', pascal labels.
736 for (name = symp->name; *name; name++) {
737 if ( *name == '.' || *name == '$' ) {
742 /* On systems where the C compiler adds an underscore to all names,
743 * static names without underscores seem usually to be labels in
744 * hand written assembler in the library. We don't want these
745 * names. This is certainly necessary on a Sparc running SunOS 4.1
746 * (try profiling a program that does a lot of division). I don't
747 * know whether it has harmful side effects on other systems.
748 * Perhaps it should be made configurable.
751 if (symprefix && symprefix != *symp->name)