gprof.c revision 299877 
1/*
2 * Copyright (c) 1983, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 *    may be used to endorse or promote products derived from this software
15 *    without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30#ifndef lint
31static const char copyright[] =
32"@(#) Copyright (c) 1983, 1993\n\
33	The Regents of the University of California.  All rights reserved.\n";
34#endif /* not lint */
35
36#if 0
37#ifndef lint
38static char sccsid[] = "@(#)gprof.c	8.1 (Berkeley) 6/6/93";
39#endif /* not lint */
40#endif
41
42#include <sys/cdefs.h>
43__FBSDID("$FreeBSD: head/usr.bin/gprof/gprof.c 299877 2016-05-16 01:11:02Z araujo $");
44
45#include <err.h>
46#include <limits.h>
47#include <stdint.h>
48#include <string.h>
49
50#include "gprof.h"
51
52static int valcmp(const void *, const void *);
53
54static struct gmonhdr	gmonhdr;
55static int lflag;
56static int Lflag;
57
58int
59main(int argc, char **argv)
60{
61    char	**sp;
62    nltype	**timesortnlp;
63    char	**defaultEs;
64
65    --argc;
66    argv++;
67    debug = 0;
68    bflag = TRUE;
69    while ( *argv != 0 && **argv == '-' ) {
70	(*argv)++;
71	switch ( **argv ) {
72	case 'a':
73	    aflag = TRUE;
74	    break;
75	case 'b':
76	    bflag = FALSE;
77	    break;
78	case 'C':
79	    Cflag = TRUE;
80	    cyclethreshold = atoi( *++argv );
81	    break;
82	case 'd':
83	    dflag = TRUE;
84	    setlinebuf(stdout);
85	    debug |= atoi( *++argv );
86	    debug |= ANYDEBUG;
87#	    ifdef DEBUG
88		printf("[main] debug = %d\n", debug);
89#	    else /* not DEBUG */
90		printf("gprof: -d ignored\n");
91#	    endif /* DEBUG */
92	    break;
93	case 'E':
94	    ++argv;
95	    addlist( Elist , *argv );
96	    Eflag = TRUE;
97	    addlist( elist , *argv );
98	    eflag = TRUE;
99	    break;
100	case 'e':
101	    addlist( elist , *++argv );
102	    eflag = TRUE;
103	    break;
104	case 'F':
105	    ++argv;
106	    addlist( Flist , *argv );
107	    Fflag = TRUE;
108	    addlist( flist , *argv );
109	    fflag = TRUE;
110	    break;
111	case 'f':
112	    addlist( flist , *++argv );
113	    fflag = TRUE;
114	    break;
115	case 'k':
116	    addlist( kfromlist , *++argv );
117	    addlist( ktolist , *++argv );
118	    kflag = TRUE;
119	    break;
120	case 'K':
121	    Kflag = TRUE;
122	    break;
123	case 'l':
124	    lflag = 1;
125	    Lflag = 0;
126	    break;
127	case 'L':
128	    Lflag = 1;
129	    lflag = 0;
130	    break;
131	case 's':
132	    sflag = TRUE;
133	    break;
134	case 'u':
135	    uflag = TRUE;
136	    break;
137	case 'z':
138	    zflag = TRUE;
139	    break;
140	}
141	argv++;
142    }
143    if ( *argv != 0 ) {
144	a_outname  = *argv;
145	argv++;
146    } else {
147	a_outname  = A_OUTNAME;
148    }
149    if ( *argv != 0 ) {
150	gmonname = *argv;
151	argv++;
152    } else {
153	gmonname = (char *) malloc(strlen(a_outname)+6);
154	strcpy(gmonname, a_outname);
155	strcat(gmonname, ".gmon");
156    }
157	/*
158	 *	get information from the executable file.
159	 */
160    if ((Kflag && kernel_getnfile(a_outname, &defaultEs) == -1) ||
161      (!Kflag && elf_getnfile(a_outname, &defaultEs) == -1 &&
162      aout_getnfile(a_outname, &defaultEs) == -1))
163	errx(1, "%s: bad format", a_outname);
164	/*
165	 *	sort symbol table.
166	 */
167    qsort(nl, nname, sizeof(nltype), valcmp);
168	/*
169	 *	turn off default functions
170	 */
171    for ( sp = defaultEs ; *sp ; sp++ ) {
172	Eflag = TRUE;
173	addlist( Elist , *sp );
174	eflag = TRUE;
175	addlist( elist , *sp );
176    }
177	/*
178	 *	get information about mon.out file(s).
179	 */
180    do	{
181	getpfile( gmonname );
182	if ( *argv != 0 ) {
183	    gmonname = *argv;
184	}
185    } while ( *argv++ != 0 );
186	/*
187	 *	how many ticks per second?
188	 *	if we can't tell, report time in ticks.
189	 */
190    if (hz == 0) {
191	hz = 1;
192	fprintf(stderr, "time is in ticks, not seconds\n");
193    }
194	/*
195	 *	dump out a gmon.sum file if requested
196	 */
197    if ( sflag ) {
198	dumpsum( GMONSUM );
199    }
200	/*
201	 *	assign samples to procedures
202	 */
203    asgnsamples();
204	/*
205	 *	assemble the dynamic profile
206	 */
207    timesortnlp = doarcs();
208	/*
209	 *	print the dynamic profile
210	 */
211    if(!lflag) {
212	    printgprof( timesortnlp );
213    }
214	/*
215	 *	print the flat profile
216	 */
217    if(!Lflag) {
218	    printprof();
219    }
220	/*
221	 *	print the index
222	 */
223    printindex();
224    exit(0);
225}
226
227    /*
228     *	information from a gmon.out file is in two parts:
229     *	an array of sampling hits within pc ranges,
230     *	and the arcs.
231     */
232void
233getpfile(char *filename)
234{
235    FILE		*pfile;
236    struct rawarc	arc;
237
238    pfile = openpfile(filename);
239    readsamples(pfile);
240	/*
241	 *	the rest of the file consists of
242	 *	a bunch of <from,self,count> tuples.
243	 */
244    while ( fread( &arc , sizeof arc , 1 , pfile ) == 1 ) {
245#	ifdef DEBUG
246	    if ( debug & SAMPLEDEBUG ) {
247		printf( "[getpfile] frompc 0x%lx selfpc 0x%lx count %ld\n" ,
248			arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
249	    }
250#	endif /* DEBUG */
251	    /*
252	     *	add this arc
253	     */
254	tally( &arc );
255    }
256    fclose(pfile);
257}
258
259FILE *
260openpfile(char *filename)
261{
262    struct gmonhdr	tmp;
263    FILE		*pfile;
264    int			size;
265    int			rate;
266
267    if((pfile = fopen(filename, "r")) == NULL)
268	err(1, "%s", filename);
269    fread(&tmp, sizeof(struct gmonhdr), 1, pfile);
270    if ( s_highpc != 0 && ( tmp.lpc != gmonhdr.lpc ||
271	 tmp.hpc != gmonhdr.hpc || tmp.ncnt != gmonhdr.ncnt ) )
272	errx(1, "%s: incompatible with first gmon file", filename);
273    gmonhdr = tmp;
274    if ( gmonhdr.version == GMONVERSION ) {
275	rate = gmonhdr.profrate;
276	size = sizeof(struct gmonhdr);
277    } else {
278	fseek(pfile, sizeof(struct ophdr), SEEK_SET);
279	size = sizeof(struct ophdr);
280	gmonhdr.profrate = rate = hertz();
281	gmonhdr.version = GMONVERSION;
282    }
283    if (hz == 0) {
284	hz = rate;
285    } else if (hz != rate)
286	errx(0, "%s: profile clock rate (%d) %s (%ld) in first gmon file",
287	    filename, rate, "incompatible with clock rate", hz);
288    if ( gmonhdr.histcounter_type == 0 ) {
289	/* Historical case.  The type was u_short (2 bytes in practice). */
290	histcounter_type = 16;
291	histcounter_size = 2;
292    } else {
293	histcounter_type = gmonhdr.histcounter_type;
294	histcounter_size = abs(histcounter_type) / CHAR_BIT;
295    }
296    s_lowpc = (unsigned long) gmonhdr.lpc;
297    s_highpc = (unsigned long) gmonhdr.hpc;
298    lowpc = (unsigned long)gmonhdr.lpc / HISTORICAL_SCALE_2;
299    highpc = (unsigned long)gmonhdr.hpc / HISTORICAL_SCALE_2;
300    sampbytes = gmonhdr.ncnt - size;
301    nsamples = sampbytes / histcounter_size;
302#   ifdef DEBUG
303	if ( debug & SAMPLEDEBUG ) {
304	    printf( "[openpfile] hdr.lpc 0x%lx hdr.hpc 0x%lx hdr.ncnt %d\n",
305		gmonhdr.lpc , gmonhdr.hpc , gmonhdr.ncnt );
306	    printf( "[openpfile]   s_lowpc 0x%lx   s_highpc 0x%lx\n" ,
307		s_lowpc , s_highpc );
308	    printf( "[openpfile]     lowpc 0x%lx     highpc 0x%lx\n" ,
309		lowpc , highpc );
310	    printf( "[openpfile] sampbytes %d nsamples %d\n" ,
311		sampbytes , nsamples );
312	    printf( "[openpfile] sample rate %ld\n" , hz );
313	}
314#   endif /* DEBUG */
315    return(pfile);
316}
317
318void
319tally(struct rawarc *rawp)
320{
321    nltype		*parentp;
322    nltype		*childp;
323
324    parentp = nllookup( rawp -> raw_frompc );
325    childp = nllookup( rawp -> raw_selfpc );
326    if ( parentp == 0 || childp == 0 )
327	return;
328    if ( kflag
329	 && onlist( kfromlist , parentp -> name )
330	 && onlist( ktolist , childp -> name ) ) {
331	return;
332    }
333    childp -> ncall += rawp -> raw_count;
334#   ifdef DEBUG
335	if ( debug & TALLYDEBUG ) {
336	    printf( "[tally] arc from %s to %s traversed %ld times\n" ,
337		    parentp -> name , childp -> name , rawp -> raw_count );
338	}
339#   endif /* DEBUG */
340    addarc( parentp , childp , rawp -> raw_count );
341}
342
343/*
344 * dump out the gmon.sum file
345 */
346void
347dumpsum(const char *sumfile)
348{
349    register nltype *nlp;
350    register arctype *arcp;
351    struct rawarc arc;
352    FILE *sfile;
353
354    if ( ( sfile = fopen ( sumfile , "w" ) ) == NULL )
355	err( 1 , "%s" , sumfile );
356    /*
357     * dump the header; use the last header read in
358     */
359    if ( fwrite( &gmonhdr , sizeof gmonhdr , 1 , sfile ) != 1 )
360	err( 1 , "%s" , sumfile );
361    /*
362     * dump the samples
363     */
364    if (fwrite(samples, histcounter_size, nsamples, sfile) != nsamples)
365	err( 1 , "%s" , sumfile );
366    /*
367     * dump the normalized raw arc information
368     */
369    for ( nlp = nl ; nlp < npe ; nlp++ ) {
370	for ( arcp = nlp -> children ; arcp ; arcp = arcp -> arc_childlist ) {
371	    arc.raw_frompc = arcp -> arc_parentp -> value;
372	    arc.raw_selfpc = arcp -> arc_childp -> value;
373	    arc.raw_count = arcp -> arc_count;
374	    if ( fwrite ( &arc , sizeof arc , 1 , sfile ) != 1 )
375		err( 1 , "%s" , sumfile );
376#	    ifdef DEBUG
377		if ( debug & SAMPLEDEBUG ) {
378		    printf( "[dumpsum] frompc 0x%lx selfpc 0x%lx count %ld\n" ,
379			    arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
380		}
381#	    endif /* DEBUG */
382	}
383    }
384    fclose( sfile );
385}
386
387static int
388valcmp(const void *v1, const void *v2)
389{
390    const nltype *p1 = (const nltype *)v1;
391    const nltype *p2 = (const nltype *)v2;
392
393    if ( p1 -> value < p2 -> value ) {
394	return LESSTHAN;
395    }
396    if ( p1 -> value > p2 -> value ) {
397	return GREATERTHAN;
398    }
399    return EQUALTO;
400}
401
402void
403readsamples(FILE *pfile)
404{
405    int		i;
406    intmax_t	sample;
407
408    if (samples == 0) {
409	samples = (double *) calloc(nsamples, sizeof(double));
410	if (samples == NULL)
411	    errx(0, "no room for %d sample pc's", nsamples);
412    }
413    for (i = 0; i < nsamples; i++) {
414	fread(&sample, histcounter_size, 1, pfile);
415	if (feof(pfile))
416		break;
417	switch ( histcounter_type ) {
418	case -8:
419	    samples[i] += *(int8_t *)&sample;
420	    break;
421	case 8:
422	    samples[i] += *(u_int8_t *)&sample;
423	    break;
424	case -16:
425	    samples[i] += *(int16_t *)&sample;
426	    break;
427	case 16:
428	    samples[i] += *(u_int16_t *)&sample;
429	    break;
430	case -32:
431	    samples[i] += *(int32_t *)&sample;
432	    break;
433	case 32:
434	    samples[i] += *(u_int32_t *)&sample;
435	    break;
436	case -64:
437	    samples[i] += *(int64_t *)&sample;
438	    break;
439	case 64:
440	    samples[i] += *(u_int64_t *)&sample;
441	    break;
442	default:
443	    err(1, "unsupported histogram counter type %d", histcounter_type);
444	}
445    }
446    if (i != nsamples)
447	errx(1, "unexpected EOF after reading %d/%d samples", --i , nsamples );
448}
449
450/*
451 *	Assign samples to the procedures to which they belong.
452 *
453 *	There are three cases as to where pcl and pch can be
454 *	with respect to the routine entry addresses svalue0 and svalue1
455 *	as shown in the following diagram.  overlap computes the
456 *	distance between the arrows, the fraction of the sample
457 *	that is to be credited to the routine which starts at svalue0.
458 *
459 *	    svalue0                                         svalue1
460 *	       |                                               |
461 *	       v                                               v
462 *
463 *	       +-----------------------------------------------+
464 *	       |					       |
465 *	  |  ->|    |<-		->|         |<-		->|    |<-  |
466 *	  |         |		  |         |		  |         |
467 *	  +---------+		  +---------+		  +---------+
468 *
469 *	  ^         ^		  ^         ^		  ^         ^
470 *	  |         |		  |         |		  |         |
471 *	 pcl       pch		 pcl       pch		 pcl       pch
472 *
473 *	For the vax we assert that samples will never fall in the first
474 *	two bytes of any routine, since that is the entry mask,
475 *	thus we give call alignentries() to adjust the entry points if
476 *	the entry mask falls in one bucket but the code for the routine
477 *	doesn't start until the next bucket.  In conjunction with the
478 *	alignment of routine addresses, this should allow us to have
479 *	only one sample for every four bytes of text space and never
480 *	have any overlap (the two end cases, above).
481 */
482void
483asgnsamples(void)
484{
485    register int	j;
486    double		ccnt;
487    double		thetime;
488    unsigned long	pcl, pch;
489    register int	i;
490    unsigned long	overlap;
491    unsigned long	svalue0, svalue1;
492
493    /* read samples and assign to namelist symbols */
494    scale = highpc - lowpc;
495    scale /= nsamples;
496    alignentries();
497    for (i = 0, j = 1; i < nsamples; i++) {
498	ccnt = samples[i];
499	if (ccnt == 0)
500		continue;
501	pcl = lowpc + (unsigned long)(scale * i);
502	pch = lowpc + (unsigned long)(scale * (i + 1));
503	thetime = ccnt;
504#	ifdef DEBUG
505	    if ( debug & SAMPLEDEBUG ) {
506		printf( "[asgnsamples] pcl 0x%lx pch 0x%lx ccnt %.0f\n" ,
507			pcl , pch , ccnt );
508	    }
509#	endif /* DEBUG */
510	totime += thetime;
511	for (j = j - 1; j < nname; j++) {
512	    svalue0 = nl[j].svalue;
513	    svalue1 = nl[j+1].svalue;
514		/*
515		 *	if high end of tick is below entry address,
516		 *	go for next tick.
517		 */
518	    if (pch < svalue0)
519		    break;
520		/*
521		 *	if low end of tick into next routine,
522		 *	go for next routine.
523		 */
524	    if (pcl >= svalue1)
525		    continue;
526	    overlap = min(pch, svalue1) - max(pcl, svalue0);
527	    if (overlap > 0) {
528#		ifdef DEBUG
529		    if (debug & SAMPLEDEBUG) {
530			printf("[asgnsamples] (0x%lx->0x%lx-0x%lx) %s gets %f ticks %lu overlap\n",
531				nl[j].value / HISTORICAL_SCALE_2,
532				svalue0, svalue1, nl[j].name,
533				overlap * thetime / scale, overlap);
534		    }
535#		endif /* DEBUG */
536		nl[j].time += overlap * thetime / scale;
537	    }
538	}
539    }
540#   ifdef DEBUG
541	if (debug & SAMPLEDEBUG) {
542	    printf("[asgnsamples] totime %f\n", totime);
543	}
544#   endif /* DEBUG */
545}
546
547
548unsigned long
549min(unsigned long a, unsigned long b)
550{
551    if (a<b)
552	return(a);
553    return(b);
554}
555
556unsigned long
557max(unsigned long a, unsigned long b)
558{
559    if (a>b)
560	return(a);
561    return(b);
562}
563
564    /*
565     *	calculate scaled entry point addresses (to save time in asgnsamples),
566     *	and possibly push the scaled entry points over the entry mask,
567     *	if it turns out that the entry point is in one bucket and the code
568     *	for a routine is in the next bucket.
569     */
570void
571alignentries(void)
572{
573    register struct nl	*nlp;
574    unsigned long	bucket_of_entry;
575    unsigned long	bucket_of_code;
576
577    for (nlp = nl; nlp < npe; nlp++) {
578	nlp -> svalue = nlp -> value / HISTORICAL_SCALE_2;
579	bucket_of_entry = (nlp->svalue - lowpc) / scale;
580	bucket_of_code = (nlp->svalue + OFFSET_OF_CODE / HISTORICAL_SCALE_2 -
581	  lowpc) / scale;
582	if (bucket_of_entry < bucket_of_code) {
583#	    ifdef DEBUG
584		if (debug & SAMPLEDEBUG) {
585		    printf("[alignentries] pushing svalue 0x%lx to 0x%lx\n",
586			    nlp->svalue,
587			    nlp->svalue + OFFSET_OF_CODE / HISTORICAL_SCALE_2);
588		}
589#	    endif /* DEBUG */
590	    nlp->svalue += OFFSET_OF_CODE / HISTORICAL_SCALE_2;
591	}
592    }
593}
594