gprof.c revision 216370 
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 216370 2010-12-11 08:32:16Z joel $");
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
54
55static struct gmonhdr	gmonhdr;
56static int lflag;
57static int Lflag;
58
59int
60main(argc, argv)
61    int argc;
62    char **argv;
63{
64    char	**sp;
65    nltype	**timesortnlp;
66    char	**defaultEs;
67
68    --argc;
69    argv++;
70    debug = 0;
71    bflag = TRUE;
72    while ( *argv != 0 && **argv == '-' ) {
73	(*argv)++;
74	switch ( **argv ) {
75	case 'a':
76	    aflag = TRUE;
77	    break;
78	case 'b':
79	    bflag = FALSE;
80	    break;
81	case 'C':
82	    Cflag = TRUE;
83	    cyclethreshold = atoi( *++argv );
84	    break;
85	case 'd':
86	    dflag = TRUE;
87	    setlinebuf(stdout);
88	    debug |= atoi( *++argv );
89	    debug |= ANYDEBUG;
90#	    ifdef DEBUG
91		printf("[main] debug = %d\n", debug);
92#	    else /* not DEBUG */
93		printf("gprof: -d ignored\n");
94#	    endif /* DEBUG */
95	    break;
96	case 'E':
97	    ++argv;
98	    addlist( Elist , *argv );
99	    Eflag = TRUE;
100	    addlist( elist , *argv );
101	    eflag = TRUE;
102	    break;
103	case 'e':
104	    addlist( elist , *++argv );
105	    eflag = TRUE;
106	    break;
107	case 'F':
108	    ++argv;
109	    addlist( Flist , *argv );
110	    Fflag = TRUE;
111	    addlist( flist , *argv );
112	    fflag = TRUE;
113	    break;
114	case 'f':
115	    addlist( flist , *++argv );
116	    fflag = TRUE;
117	    break;
118	case 'k':
119	    addlist( kfromlist , *++argv );
120	    addlist( ktolist , *++argv );
121	    kflag = TRUE;
122	    break;
123	case 'K':
124	    Kflag = TRUE;
125	    break;
126	case 'l':
127	    lflag = 1;
128	    Lflag = 0;
129	    break;
130	case 'L':
131	    Lflag = 1;
132	    lflag = 0;
133	    break;
134	case 's':
135	    sflag = TRUE;
136	    break;
137	case 'u':
138	    uflag = TRUE;
139	    break;
140	case 'z':
141	    zflag = TRUE;
142	    break;
143	}
144	argv++;
145    }
146    if ( *argv != 0 ) {
147	a_outname  = *argv;
148	argv++;
149    } else {
150	a_outname  = A_OUTNAME;
151    }
152    if ( *argv != 0 ) {
153	gmonname = *argv;
154	argv++;
155    } else {
156	gmonname = (char *) malloc(strlen(a_outname)+6);
157	strcpy(gmonname, a_outname);
158	strcat(gmonname, ".gmon");
159    }
160	/*
161	 *	get information from the executable file.
162	 */
163    if ((Kflag && kernel_getnfile(a_outname, &defaultEs) == -1) ||
164      (!Kflag && elf_getnfile(a_outname, &defaultEs) == -1 &&
165      aout_getnfile(a_outname, &defaultEs) == -1))
166	errx(1, "%s: bad format", a_outname);
167	/*
168	 *	sort symbol table.
169	 */
170    qsort(nl, nname, sizeof(nltype), valcmp);
171	/*
172	 *	turn off default functions
173	 */
174    for ( sp = defaultEs ; *sp ; sp++ ) {
175	Eflag = TRUE;
176	addlist( Elist , *sp );
177	eflag = TRUE;
178	addlist( elist , *sp );
179    }
180	/*
181	 *	get information about mon.out file(s).
182	 */
183    do	{
184	getpfile( gmonname );
185	if ( *argv != 0 ) {
186	    gmonname = *argv;
187	}
188    } while ( *argv++ != 0 );
189	/*
190	 *	how many ticks per second?
191	 *	if we can't tell, report time in ticks.
192	 */
193    if (hz == 0) {
194	hz = 1;
195	fprintf(stderr, "time is in ticks, not seconds\n");
196    }
197	/*
198	 *	dump out a gmon.sum file if requested
199	 */
200    if ( sflag ) {
201	dumpsum( GMONSUM );
202    }
203	/*
204	 *	assign samples to procedures
205	 */
206    asgnsamples();
207	/*
208	 *	assemble the dynamic profile
209	 */
210    timesortnlp = doarcs();
211	/*
212	 *	print the dynamic profile
213	 */
214    if(!lflag) {
215	    printgprof( timesortnlp );
216    }
217	/*
218	 *	print the flat profile
219	 */
220    if(!Lflag) {
221	    printprof();
222    }
223	/*
224	 *	print the index
225	 */
226    printindex();
227    exit(0);
228}
229
230    /*
231     *	information from a gmon.out file is in two parts:
232     *	an array of sampling hits within pc ranges,
233     *	and the arcs.
234     */
235void
236getpfile(filename)
237    char *filename;
238{
239    FILE		*pfile;
240    FILE		*openpfile();
241    struct rawarc	arc;
242
243    pfile = openpfile(filename);
244    readsamples(pfile);
245	/*
246	 *	the rest of the file consists of
247	 *	a bunch of <from,self,count> tuples.
248	 */
249    while ( fread( &arc , sizeof arc , 1 , pfile ) == 1 ) {
250#	ifdef DEBUG
251	    if ( debug & SAMPLEDEBUG ) {
252		printf( "[getpfile] frompc 0x%lx selfpc 0x%lx count %ld\n" ,
253			arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
254	    }
255#	endif /* DEBUG */
256	    /*
257	     *	add this arc
258	     */
259	tally( &arc );
260    }
261    fclose(pfile);
262}
263
264FILE *
265openpfile(filename)
266    char *filename;
267{
268    struct gmonhdr	tmp;
269    FILE		*pfile;
270    int			size;
271    int			rate;
272
273    if((pfile = fopen(filename, "r")) == NULL)
274	err(1, "%s", filename);
275    fread(&tmp, sizeof(struct gmonhdr), 1, pfile);
276    if ( s_highpc != 0 && ( tmp.lpc != gmonhdr.lpc ||
277	 tmp.hpc != gmonhdr.hpc || tmp.ncnt != gmonhdr.ncnt ) )
278	errx(1, "%s: incompatible with first gmon file", filename);
279    gmonhdr = tmp;
280    if ( gmonhdr.version == GMONVERSION ) {
281	rate = gmonhdr.profrate;
282	size = sizeof(struct gmonhdr);
283    } else {
284	fseek(pfile, sizeof(struct ophdr), SEEK_SET);
285	size = sizeof(struct ophdr);
286	gmonhdr.profrate = rate = hertz();
287	gmonhdr.version = GMONVERSION;
288    }
289    if (hz == 0) {
290	hz = rate;
291    } else if (hz != rate)
292	errx(0, "%s: profile clock rate (%d) %s (%ld) in first gmon file",
293	    filename, rate, "incompatible with clock rate", hz);
294    if ( gmonhdr.histcounter_type == 0 ) {
295	/* Historical case.  The type was u_short (2 bytes in practice). */
296	histcounter_type = 16;
297	histcounter_size = 2;
298    } else {
299	histcounter_type = gmonhdr.histcounter_type;
300	histcounter_size = abs(histcounter_type) / CHAR_BIT;
301    }
302    s_lowpc = (unsigned long) gmonhdr.lpc;
303    s_highpc = (unsigned long) gmonhdr.hpc;
304    lowpc = (unsigned long)gmonhdr.lpc / HISTORICAL_SCALE_2;
305    highpc = (unsigned long)gmonhdr.hpc / HISTORICAL_SCALE_2;
306    sampbytes = gmonhdr.ncnt - size;
307    nsamples = sampbytes / histcounter_size;
308#   ifdef DEBUG
309	if ( debug & SAMPLEDEBUG ) {
310	    printf( "[openpfile] hdr.lpc 0x%lx hdr.hpc 0x%lx hdr.ncnt %d\n",
311		gmonhdr.lpc , gmonhdr.hpc , gmonhdr.ncnt );
312	    printf( "[openpfile]   s_lowpc 0x%lx   s_highpc 0x%lx\n" ,
313		s_lowpc , s_highpc );
314	    printf( "[openpfile]     lowpc 0x%lx     highpc 0x%lx\n" ,
315		lowpc , highpc );
316	    printf( "[openpfile] sampbytes %d nsamples %d\n" ,
317		sampbytes , nsamples );
318	    printf( "[openpfile] sample rate %ld\n" , hz );
319	}
320#   endif /* DEBUG */
321    return(pfile);
322}
323
324void
325tally( rawp )
326    struct rawarc	*rawp;
327{
328    nltype		*parentp;
329    nltype		*childp;
330
331    parentp = nllookup( rawp -> raw_frompc );
332    childp = nllookup( rawp -> raw_selfpc );
333    if ( parentp == 0 || childp == 0 )
334	return;
335    if ( kflag
336	 && onlist( kfromlist , parentp -> name )
337	 && onlist( ktolist , childp -> name ) ) {
338	return;
339    }
340    childp -> ncall += rawp -> raw_count;
341#   ifdef DEBUG
342	if ( debug & TALLYDEBUG ) {
343	    printf( "[tally] arc from %s to %s traversed %ld times\n" ,
344		    parentp -> name , childp -> name , rawp -> raw_count );
345	}
346#   endif /* DEBUG */
347    addarc( parentp , childp , rawp -> raw_count );
348}
349
350/*
351 * dump out the gmon.sum file
352 */
353void
354dumpsum( sumfile )
355    char *sumfile;
356{
357    register nltype *nlp;
358    register arctype *arcp;
359    struct rawarc arc;
360    FILE *sfile;
361
362    if ( ( sfile = fopen ( sumfile , "w" ) ) == NULL )
363	err( 1 , "%s" , sumfile );
364    /*
365     * dump the header; use the last header read in
366     */
367    if ( fwrite( &gmonhdr , sizeof gmonhdr , 1 , sfile ) != 1 )
368	err( 1 , "%s" , sumfile );
369    /*
370     * dump the samples
371     */
372    if (fwrite(samples, histcounter_size, nsamples, sfile) != nsamples)
373	err( 1 , "%s" , sumfile );
374    /*
375     * dump the normalized raw arc information
376     */
377    for ( nlp = nl ; nlp < npe ; nlp++ ) {
378	for ( arcp = nlp -> children ; arcp ; arcp = arcp -> arc_childlist ) {
379	    arc.raw_frompc = arcp -> arc_parentp -> value;
380	    arc.raw_selfpc = arcp -> arc_childp -> value;
381	    arc.raw_count = arcp -> arc_count;
382	    if ( fwrite ( &arc , sizeof arc , 1 , sfile ) != 1 )
383		err( 1 , "%s" , sumfile );
384#	    ifdef DEBUG
385		if ( debug & SAMPLEDEBUG ) {
386		    printf( "[dumpsum] frompc 0x%lx selfpc 0x%lx count %ld\n" ,
387			    arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
388		}
389#	    endif /* DEBUG */
390	}
391    }
392    fclose( sfile );
393}
394
395static int
396valcmp(v1, v2)
397    const void *v1;
398    const void *v2;
399{
400    const nltype *p1 = (const nltype *)v1;
401    const nltype *p2 = (const nltype *)v2;
402
403    if ( p1 -> value < p2 -> value ) {
404	return LESSTHAN;
405    }
406    if ( p1 -> value > p2 -> value ) {
407	return GREATERTHAN;
408    }
409    return EQUALTO;
410}
411
412void
413readsamples(pfile)
414    FILE	*pfile;
415{
416    int		i;
417    intmax_t	sample;
418
419    if (samples == 0) {
420	samples = (double *) calloc(nsamples, sizeof(double));
421	if (samples == 0)
422	    errx(0, "no room for %d sample pc's", nsamples);
423    }
424    for (i = 0; i < nsamples; i++) {
425	fread(&sample, histcounter_size, 1, pfile);
426	if (feof(pfile))
427		break;
428	switch ( histcounter_type ) {
429	case -8:
430	    samples[i] += *(int8_t *)&sample;
431	    break;
432	case 8:
433	    samples[i] += *(u_int8_t *)&sample;
434	    break;
435	case -16:
436	    samples[i] += *(int16_t *)&sample;
437	    break;
438	case 16:
439	    samples[i] += *(u_int16_t *)&sample;
440	    break;
441	case -32:
442	    samples[i] += *(int32_t *)&sample;
443	    break;
444	case 32:
445	    samples[i] += *(u_int32_t *)&sample;
446	    break;
447	case -64:
448	    samples[i] += *(int64_t *)&sample;
449	    break;
450	case 64:
451	    samples[i] += *(u_int64_t *)&sample;
452	    break;
453	default:
454	    err(1, "unsupported histogram counter type %d", histcounter_type);
455	}
456    }
457    if (i != nsamples)
458	errx(1, "unexpected EOF after reading %d/%d samples", --i , nsamples );
459}
460
461/*
462 *	Assign samples to the procedures to which they belong.
463 *
464 *	There are three cases as to where pcl and pch can be
465 *	with respect to the routine entry addresses svalue0 and svalue1
466 *	as shown in the following diagram.  overlap computes the
467 *	distance between the arrows, the fraction of the sample
468 *	that is to be credited to the routine which starts at svalue0.
469 *
470 *	    svalue0                                         svalue1
471 *	       |                                               |
472 *	       v                                               v
473 *
474 *	       +-----------------------------------------------+
475 *	       |					       |
476 *	  |  ->|    |<-		->|         |<-		->|    |<-  |
477 *	  |         |		  |         |		  |         |
478 *	  +---------+		  +---------+		  +---------+
479 *
480 *	  ^         ^		  ^         ^		  ^         ^
481 *	  |         |		  |         |		  |         |
482 *	 pcl       pch		 pcl       pch		 pcl       pch
483 *
484 *	For the vax we assert that samples will never fall in the first
485 *	two bytes of any routine, since that is the entry mask,
486 *	thus we give call alignentries() to adjust the entry points if
487 *	the entry mask falls in one bucket but the code for the routine
488 *	doesn't start until the next bucket.  In conjunction with the
489 *	alignment of routine addresses, this should allow us to have
490 *	only one sample for every four bytes of text space and never
491 *	have any overlap (the two end cases, above).
492 */
493void
494asgnsamples()
495{
496    register int	j;
497    double		ccnt;
498    double		time;
499    unsigned long	pcl, pch;
500    register int	i;
501    unsigned long	overlap;
502    unsigned long	svalue0, svalue1;
503
504    /* read samples and assign to namelist symbols */
505    scale = highpc - lowpc;
506    scale /= nsamples;
507    alignentries();
508    for (i = 0, j = 1; i < nsamples; i++) {
509	ccnt = samples[i];
510	if (ccnt == 0)
511		continue;
512	pcl = lowpc + (unsigned long)(scale * i);
513	pch = lowpc + (unsigned long)(scale * (i + 1));
514	time = ccnt;
515#	ifdef DEBUG
516	    if ( debug & SAMPLEDEBUG ) {
517		printf( "[asgnsamples] pcl 0x%lx pch 0x%lx ccnt %.0f\n" ,
518			pcl , pch , ccnt );
519	    }
520#	endif /* DEBUG */
521	totime += time;
522	for (j = j - 1; j < nname; j++) {
523	    svalue0 = nl[j].svalue;
524	    svalue1 = nl[j+1].svalue;
525		/*
526		 *	if high end of tick is below entry address,
527		 *	go for next tick.
528		 */
529	    if (pch < svalue0)
530		    break;
531		/*
532		 *	if low end of tick into next routine,
533		 *	go for next routine.
534		 */
535	    if (pcl >= svalue1)
536		    continue;
537	    overlap = min(pch, svalue1) - max(pcl, svalue0);
538	    if (overlap > 0) {
539#		ifdef DEBUG
540		    if (debug & SAMPLEDEBUG) {
541			printf("[asgnsamples] (0x%lx->0x%lx-0x%lx) %s gets %f ticks %lu overlap\n",
542				nl[j].value / HISTORICAL_SCALE_2,
543				svalue0, svalue1, nl[j].name,
544				overlap * time / scale, overlap);
545		    }
546#		endif /* DEBUG */
547		nl[j].time += overlap * time / scale;
548	    }
549	}
550    }
551#   ifdef DEBUG
552	if (debug & SAMPLEDEBUG) {
553	    printf("[asgnsamples] totime %f\n", totime);
554	}
555#   endif /* DEBUG */
556}
557
558
559unsigned long
560min(a, b)
561    unsigned long a,b;
562{
563    if (a<b)
564	return(a);
565    return(b);
566}
567
568unsigned long
569max(a, b)
570    unsigned long a,b;
571{
572    if (a>b)
573	return(a);
574    return(b);
575}
576
577    /*
578     *	calculate scaled entry point addresses (to save time in asgnsamples),
579     *	and possibly push the scaled entry points over the entry mask,
580     *	if it turns out that the entry point is in one bucket and the code
581     *	for a routine is in the next bucket.
582     */
583void
584alignentries()
585{
586    register struct nl	*nlp;
587    unsigned long	bucket_of_entry;
588    unsigned long	bucket_of_code;
589
590    for (nlp = nl; nlp < npe; nlp++) {
591	nlp -> svalue = nlp -> value / HISTORICAL_SCALE_2;
592	bucket_of_entry = (nlp->svalue - lowpc) / scale;
593	bucket_of_code = (nlp->svalue + OFFSET_OF_CODE / HISTORICAL_SCALE_2 -
594	  lowpc) / scale;
595	if (bucket_of_entry < bucket_of_code) {
596#	    ifdef DEBUG
597		if (debug & SAMPLEDEBUG) {
598		    printf("[alignentries] pushing svalue 0x%lx to 0x%lx\n",
599			    nlp->svalue,
600			    nlp->svalue + OFFSET_OF_CODE / HISTORICAL_SCALE_2);
601		}
602#	    endif /* DEBUG */
603	    nlp->svalue += OFFSET_OF_CODE / HISTORICAL_SCALE_2;
604	}
605    }
606}
607