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