xref: /netbsd/usr.sbin/lockstat/main.c (revision 1661f738)
1 /*	$NetBSD: main.c,v 1.21 2021/12/12 22:20:53 andvar Exp $	*/
2 
3 /*-
4  * Copyright (c) 2006, 2007, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 #ifndef lint
34 __RCSID("$NetBSD: main.c,v 1.21 2021/12/12 22:20:53 andvar Exp $");
35 #endif /* not lint */
36 
37 #include <sys/types.h>
38 #include <sys/param.h>
39 #include <sys/time.h>
40 #include <sys/fcntl.h>
41 #include <sys/ioctl.h>
42 #include <sys/wait.h>
43 #include <sys/signal.h>
44 #include <sys/sysctl.h>
45 
46 #include <dev/lockstat.h>
47 
48 #include <stdio.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <limits.h>
52 #include <unistd.h>
53 #include <err.h>
54 #include <paths.h>
55 #include <util.h>
56 #include <ctype.h>
57 #include <errno.h>
58 #include <stdbool.h>
59 
60 #include "extern.h"
61 
62 #define	_PATH_DEV_LOCKSTAT	"/dev/lockstat"
63 
64 #define	MILLI	1000.0
65 #define	MICRO	1000000.0
66 #define	NANO	1000000000.0
67 #define	PICO	1000000000000.0
68 
69 TAILQ_HEAD(lock_head, lockstruct);
70 typedef struct lock_head locklist_t;
71 TAILQ_HEAD(buf_head, lsbuf);
72 typedef struct buf_head buflist_t;
73 SLIST_HEAD(bucket, lockstruct);
74 typedef struct bucket bucket_t;
75 
76 typedef struct lockstruct {
77 	TAILQ_ENTRY(lockstruct)	chain;
78 	SLIST_ENTRY(lockstruct)	bucket;
79 	buflist_t		bufs;
80 	buflist_t		tosort;
81 	uintptr_t		lock;
82  	double			time;
83 	uint32_t		count;
84 	u_int			flags;
85 	u_int			nbufs;
86 	char			name[NAME_SIZE];
87 } lock_t;
88 
89 typedef struct name {
90 	const char	*name;
91 	int		mask;
92 } name_t;
93 
94 static const name_t locknames[] = {
95 	{ "adaptive_mutex", LB_ADAPTIVE_MUTEX },
96 	{ "spin_mutex", LB_SPIN_MUTEX },
97 	{ "rwlock", LB_RWLOCK },
98 	{ "kernel_lock", LB_KERNEL_LOCK },
99 	{ "preemption", LB_NOPREEMPT },
100 	{ "misc", LB_MISC },
101 	{ NULL, 0 }
102 };
103 
104 static const name_t eventnames[] = {
105 	{ "spin", LB_SPIN },
106 	{ "sleep_exclusive", LB_SLEEP1 },
107 	{ "sleep_shared", LB_SLEEP2 },
108 	{ NULL, 0 },
109 };
110 
111 static const name_t alltypes[] = {
112 	{ "Adaptive mutex spin", LB_ADAPTIVE_MUTEX | LB_SPIN },
113 	{ "Adaptive mutex sleep", LB_ADAPTIVE_MUTEX | LB_SLEEP1 },
114 	{ "Spin mutex spin", LB_SPIN_MUTEX | LB_SPIN },
115 	{ "RW lock sleep (writer)", LB_RWLOCK | LB_SLEEP1 },
116 	{ "RW lock sleep (reader)", LB_RWLOCK | LB_SLEEP2 },
117 	{ "RW lock spin", LB_RWLOCK | LB_SPIN },
118 	{ "Kernel lock spin", LB_KERNEL_LOCK | LB_SPIN },
119 	{ "Kernel preemption defer", LB_NOPREEMPT | LB_SPIN },
120 	{ "Miscellaneous wait", LB_MISC | LB_SPIN },
121 	{ NULL, 0 }
122 };
123 
124 static const name_t xtypes[] = {
125 	{ "Spin", LB_SPIN },
126 	{ "Sleep (writer)", LB_SLEEP1 },
127 	{ "Sleep (reader)", LB_SLEEP2 },
128 	{ NULL, 0 }
129 };
130 
131 static locklist_t	locklist;
132 static locklist_t	freelist;
133 static locklist_t	sortlist;
134 static bucket_t		bucket[256];
135 
136 #define	HASH(a)		(&bucket[((a) >> 6) & (__arraycount(bucket) - 1)])
137 
138 static lsbuf_t		*bufs;
139 static lsdisable_t	ld;
140 static bool		lflag;
141 static bool		fflag;
142 static int		nbufs;
143 static bool		cflag;
144 static bool		dflag;
145 static bool		xflag;
146 static int		lsfd;
147 static int		displayed;
148 static int		bin64;
149 static double		tscale;
150 static double		cscale;
151 static double		cpuscale[sizeof(ld.ld_freq) / sizeof(ld.ld_freq[0])];
152 static FILE		*outfp;
153 
154 static void	findsym(findsym_t, char *, uintptr_t *, uintptr_t *, bool);
155 static void	spawn(int, char **);
156 static void	display(int, const char *name);
157 __dead static void	listnames(const name_t *);
158 static void	collapse(bool, bool);
159 static int	matchname(const name_t *, char *);
160 static void	makelists(int, int);
161 static void	nullsig(int);
162 __dead static void	usage(void);
163 static int	ncpu(void);
164 static lock_t	*morelocks(void);
165 
166 int
main(int argc,char ** argv)167 main(int argc, char **argv)
168 {
169 	int eventtype, locktype, ch, nlfd, fd;
170 	size_t i;
171 	bool sflag, pflag, mflag, Mflag;
172 	const char *nlistf, *outf;
173 	char *lockname, *funcname;
174 	const name_t *name;
175 	lsenable_t le;
176 	double ms;
177 	char *p;
178 
179 	nlistf = NULL;
180 	outf = NULL;
181 	lockname = NULL;
182 	funcname = NULL;
183 	eventtype = -1;
184 	locktype = -1;
185 	nbufs = 0;
186 	sflag = false;
187 	pflag = false;
188 	mflag = false;
189 	Mflag = false;
190 
191 	while ((ch = getopt(argc, argv, "E:F:L:MN:T:b:cdeflmo:pstx")) != -1)
192 		switch (ch) {
193 		case 'E':
194 			eventtype = matchname(eventnames, optarg);
195 			break;
196 		case 'F':
197 			funcname = optarg;
198 			break;
199 		case 'L':
200 			lockname = optarg;
201 			break;
202 		case 'N':
203 			nlistf = optarg;
204 			break;
205 		case 'T':
206 			locktype = matchname(locknames, optarg);
207 			break;
208 		case 'b':
209 			nbufs = (int)strtol(optarg, &p, 0);
210 			if (!isdigit((u_int)*optarg) || *p != '\0')
211 				usage();
212 			break;
213 		case 'c':
214 			cflag = true;
215 			break;
216 		case 'd':
217 			dflag = true;
218 			break;
219 		case 'e':
220 			listnames(eventnames);
221 			break;
222 		case 'f':
223 			fflag = true;
224 			break;
225 		case 'l':
226 			lflag = true;
227 			break;
228 		case 'm':
229 			mflag = true;
230 			break;
231 		case 'M':
232 			Mflag = true;
233 			break;
234 		case 'o':
235 			outf = optarg;
236 			break;
237 		case 'p':
238 			pflag = true;
239 			break;
240 		case 's':
241 			sflag = true;
242 			break;
243 		case 't':
244 			listnames(locknames);
245 			break;
246 		case 'x':
247 			xflag = true;
248 			break;
249 		default:
250 			usage();
251 		}
252 	argc -= optind;
253 	argv += optind;
254 
255 	if (*argv == NULL && !dflag)
256 		usage();
257 
258 	if (outf) {
259 		fd = open(outf, O_WRONLY | O_CREAT | O_TRUNC, 0600);
260 		if (fd == -1)
261 			err(EXIT_FAILURE, "opening %s", outf);
262 		outfp = fdopen(fd, "w");
263 	} else
264 		outfp = stdout;
265 
266 	/*
267 	 * Find the name list for resolving symbol names, and load it into
268 	 * memory.
269 	 */
270 	if (nlistf == NULL) {
271 		nlfd = open(_PATH_KSYMS, O_RDONLY);
272 		nlistf = getbootfile();
273 	} else
274 		nlfd = -1;
275 	if (nlfd == -1) {
276 		if ((nlfd = open(nlistf, O_RDONLY)) < 0)
277 			err(EXIT_FAILURE, "cannot open " _PATH_KSYMS " or %s",
278 			    nlistf);
279 	}
280 	if (loadsym32(nlfd) != 0) {
281 		if (loadsym64(nlfd) != 0)
282 			errx(EXIT_FAILURE, "unable to load symbol table");
283 		bin64 = 1;
284 	}
285 	close(nlfd);
286 
287 	memset(&le, 0, sizeof(le));
288 	le.le_nbufs = nbufs;
289 
290 	/*
291 	 * Set up initial filtering.
292 	 */
293 	if (lockname != NULL) {
294 		findsym(LOCK_BYNAME, lockname, &le.le_lockstart,
295 		    &le.le_lockend, true);
296 		le.le_flags |= LE_ONE_LOCK;
297 	}
298 	if (!lflag)
299 		le.le_flags |= LE_CALLSITE;
300 	if (!fflag)
301 		le.le_flags |= LE_LOCK;
302 	if (funcname != NULL) {
303 		if (lflag)
304 			usage();
305 		findsym(FUNC_BYNAME, funcname, &le.le_csstart, &le.le_csend, true);
306 		le.le_flags |= LE_ONE_CALLSITE;
307 	}
308 	le.le_mask = (eventtype & LB_EVENT_MASK) | (locktype & LB_LOCK_MASK);
309 
310 	/*
311 	 * Start tracing.
312 	 */
313 	if ((lsfd = open(_PATH_DEV_LOCKSTAT, O_RDONLY)) < 0)
314 		err(EXIT_FAILURE, "cannot open " _PATH_DEV_LOCKSTAT);
315 	if (ioctl(lsfd, IOC_LOCKSTAT_GVERSION, &ch) < 0)
316 		err(EXIT_FAILURE, "ioctl");
317 	if (ch != LS_VERSION)
318 		errx(EXIT_FAILURE,
319 		    "incompatible lockstat interface version (%d, kernel %d)",
320 			LS_VERSION, ch);
321 	if (dflag) {
322 		goto disable;
323 	}
324 	if (ioctl(lsfd, IOC_LOCKSTAT_ENABLE, &le))
325 		err(EXIT_FAILURE, "cannot enable tracing");
326 
327 	/*
328 	 * Execute the traced program.
329 	 */
330 	spawn(argc, argv);
331 
332 disable:
333 	/*
334 	 * Stop tracing, and read the trace buffers from the kernel.
335 	 */
336 	if (ioctl(lsfd, IOC_LOCKSTAT_DISABLE, &ld) == -1) {
337 		if (errno == EOVERFLOW) {
338 			warnx("overflowed available kernel trace buffers");
339 			exit(EXIT_FAILURE);
340 		}
341 		err(EXIT_FAILURE, "cannot disable tracing");
342 	}
343 	if ((bufs = malloc(ld.ld_size)) == NULL)
344 		err(EXIT_FAILURE, "cannot allocate memory for user buffers");
345 	if ((size_t)read(lsfd, bufs, ld.ld_size) != ld.ld_size)
346 		err(EXIT_FAILURE, "reading from " _PATH_DEV_LOCKSTAT);
347 	if (close(lsfd))
348 		err(EXIT_FAILURE, "close(" _PATH_DEV_LOCKSTAT ")");
349 
350 	/*
351 	 * Figure out how to scale the results.  For internal use we convert
352 	 * all times from CPU frequency based to picoseconds, and values are
353 	 * eventually displayed in ms.
354 	 */
355 	for (i = 0; i < sizeof(ld.ld_freq) / sizeof(ld.ld_freq[0]); i++)
356 		if (ld.ld_freq[i] != 0)
357 			cpuscale[i] = PICO / ld.ld_freq[i];
358 	ms = ld.ld_time.tv_sec * MILLI + ld.ld_time.tv_nsec / MICRO;
359 	if (pflag)
360 		cscale = 1.0 / ncpu();
361 	else
362 		cscale = 1.0;
363 	cscale *= (sflag ? MILLI / ms : 1.0);
364 	tscale = cscale / NANO;
365 	nbufs = (int)(ld.ld_size / sizeof(lsbuf_t));
366 
367 	TAILQ_INIT(&locklist);
368 	TAILQ_INIT(&sortlist);
369 	TAILQ_INIT(&freelist);
370 
371 	if ((mflag | Mflag) != 0)
372 		collapse(mflag, Mflag);
373 
374 	/*
375 	 * Display the results.
376 	 */
377 	fprintf(outfp, "Elapsed time: %.2f seconds.", ms / MILLI);
378 	if (sflag || pflag) {
379 		fprintf(outfp, " Displaying ");
380 		if (pflag)
381 			fprintf(outfp, "per-CPU ");
382 		if (sflag)
383 			fprintf(outfp, "per-second ");
384 		fprintf(outfp, "averages.");
385 	}
386 	putc('\n', outfp);
387 
388 	for (name = xflag ? xtypes : alltypes; name->name != NULL; name++) {
389 		if (eventtype != -1 &&
390 		    (name->mask & LB_EVENT_MASK) != eventtype)
391 			continue;
392 		if (locktype != -1 &&
393 		    (name->mask & LB_LOCK_MASK) != locktype)
394 			continue;
395 		display(name->mask, name->name);
396 	}
397 
398 	if (displayed == 0)
399 		fprintf(outfp, "None of the selected events were recorded.\n");
400 	exit(EXIT_SUCCESS);
401 }
402 
403 static void
usage(void)404 usage(void)
405 {
406 
407 	fprintf(stderr,
408 	    "%s: usage:\n"
409 	    "%s [options] <command>\n\n"
410 	    "-b nbuf\t\tset number of event buffers to allocate\n"
411 	    "-c\t\treport percentage of total events by count, not time\n"
412 	    "-d\t\tdisable lockstat\n"
413 	    "-E event\tdisplay only one type of event\n"
414 	    "-e\t\tlist event types\n"
415 	    "-F func\t\tlimit trace to one function\n"
416 	    "-f\t\ttrace only by function\n"
417 	    "-L lock\t\tlimit trace to one lock (name, or address)\n"
418 	    "-l\t\ttrace only by lock\n"
419 	    "-M\t\tmerge lock addresses within unique objects\n"
420 	    "-m\t\tmerge call sites within unique functions\n"
421 	    "-N nlist\tspecify name list file\n"
422 	    "-o file\t\tsend output to named file, not stdout\n"
423 	    "-p\t\tshow average count/time per CPU, not total\n"
424 	    "-s\t\tshow average count/time per second, not total\n"
425 	    "-T type\t\tdisplay only one type of lock\n"
426 	    "-t\t\tlist lock types\n"
427 	    "-x\t\tdon't differentiate event types\n",
428 	    getprogname(), getprogname());
429 
430 	exit(EXIT_FAILURE);
431 }
432 
433 static void
nullsig(int junk)434 nullsig(int junk)
435 {
436 
437 	(void)junk;
438 }
439 
440 static void
listnames(const name_t * name)441 listnames(const name_t *name)
442 {
443 
444 	for (; name->name != NULL; name++)
445 		printf("%s\n", name->name);
446 
447 	exit(EXIT_SUCCESS);
448 }
449 
450 static int
matchname(const name_t * name,char * string)451 matchname(const name_t *name, char *string)
452 {
453 	int empty, mask;
454 	char *sp;
455 
456 	empty = 1;
457 	mask = 0;
458 
459 	while ((sp = strsep(&string, ",")) != NULL) {
460 		if (*sp == '\0')
461 			usage();
462 
463 		for (; name->name != NULL; name++) {
464 			if (strcasecmp(name->name, sp) == 0) {
465 				mask |= name->mask;
466 				break;
467 			}
468 		}
469 		if (name->name == NULL)
470 			errx(EXIT_FAILURE, "unknown identifier `%s'", sp);
471 		empty = 0;
472 	}
473 
474 	if (empty)
475 		usage();
476 
477 	return mask;
478 }
479 
480 /*
481  * Return the number of CPUs in the running system.
482  */
483 static int
ncpu(void)484 ncpu(void)
485 {
486 	int rv, mib[2];
487 	size_t varlen;
488 
489 	mib[0] = CTL_HW;
490 	mib[1] = HW_NCPU;
491 	varlen = sizeof(rv);
492 	if (sysctl(mib, 2, &rv, &varlen, NULL, (size_t)0) < 0)
493 		rv = 1;
494 
495 	return (rv);
496 }
497 
498 /*
499  * Call into the ELF parser and look up a symbol by name or by address.
500  */
501 static void
findsym(findsym_t find,char * name,uintptr_t * start,uintptr_t * end,bool chg)502 findsym(findsym_t find, char *name, uintptr_t *start, uintptr_t *end, bool chg)
503 {
504 	uintptr_t tend, sa, ea;
505 	char *p;
506 	int rv;
507 
508 	if (!chg) {
509 		sa = *start;
510 		start = &sa;
511 		end = &ea;
512 	}
513 
514 	if (end == NULL)
515 		end = &tend;
516 
517 	if (find == LOCK_BYNAME) {
518 		if (isdigit((u_int)name[0])) {
519 			*start = (uintptr_t)strtoul(name, &p, 0);
520 			if (*p == '\0')
521 				return;
522 		}
523 	}
524 
525 	if (bin64)
526 		rv = findsym64(find, name, start, end);
527 	else
528 		rv = findsym32(find, name, start, end);
529 
530 	if (find == FUNC_BYNAME || find == LOCK_BYNAME) {
531 		if (rv == -1)
532 			errx(EXIT_FAILURE, "unable to find symbol `%s'", name);
533 		return;
534 	}
535 
536 	if (rv == -1)
537 		snprintf(name, NAME_SIZE, "%016lx", (long)*start);
538 }
539 
540 /*
541  * Fork off the child process and wait for it to complete.  We trap SIGINT
542  * so that the caller can use Ctrl-C to stop tracing early and still get
543  * useful results.
544  */
545 static void
spawn(int argc,char ** argv)546 spawn(int argc, char **argv)
547 {
548 	pid_t pid;
549 
550 	switch (pid = fork()) {
551 	case 0:
552 		close(lsfd);
553 		if (execvp(argv[0], argv) == -1)
554 			err(EXIT_FAILURE, "cannot exec");
555 		break;
556 	case -1:
557 		err(EXIT_FAILURE, "cannot fork to exec");
558 		break;
559 	default:
560 		signal(SIGINT, nullsig);
561 		wait(NULL);
562 		signal(SIGINT, SIG_DFL);
563 		break;
564 	}
565 }
566 
567 /*
568  * Allocate a new block of lock_t structures.
569  */
570 static lock_t *
morelocks(void)571 morelocks(void)
572 {
573 	const int batch = 32;
574 	lock_t *l, *lp, *max;
575 
576 	l = (lock_t *)malloc(sizeof(*l) * batch);
577 
578 	for (lp = l, max = l + batch; lp < max; lp++)
579 		TAILQ_INSERT_TAIL(&freelist, lp, chain);
580 
581 	return l;
582 }
583 
584 /*
585  * Collapse addresses from unique objects.
586  */
587 static void
collapse(bool func,bool lock)588 collapse(bool func, bool lock)
589 {
590 	lsbuf_t *lb, *max;
591 
592 	for (lb = bufs, max = bufs + nbufs; lb < max; lb++) {
593 		if (func && lb->lb_callsite != 0) {
594 			findsym(FUNC_BYADDR, NULL, &lb->lb_callsite, NULL,
595 			    true);
596 		}
597 		if (lock && lb->lb_lock != 0) {
598 			findsym(LOCK_BYADDR, NULL, &lb->lb_lock, NULL,
599 			    true);
600 		}
601 	}
602 }
603 
604 /*
605  * From the kernel supplied data, construct two dimensional lists of locks
606  * and event buffers, indexed by lock type and sorted by event type.
607  */
608 static void
makelists(int mask,int event)609 makelists(int mask, int event)
610 {
611 	lsbuf_t *lb, *lb2, *max;
612 	lock_t *l, *l2;
613 	bucket_t *bp;
614 	int type;
615 	size_t i;
616 
617 	/*
618 	 * Recycle lock_t structures from the last run.
619 	 */
620 	TAILQ_CONCAT(&freelist, &locklist, chain);
621 	for (i = 0; i < __arraycount(bucket); i++) {
622 		SLIST_INIT(&bucket[i]);
623 	}
624 
625 	type = mask & LB_LOCK_MASK;
626 
627 	for (lb = bufs, max = bufs + nbufs; lb < max; lb++) {
628 		if (!xflag && (lb->lb_flags & LB_LOCK_MASK) != type)
629 			continue;
630 		if (lb->lb_counts[event] == 0)
631 			continue;
632 
633 		/*
634 		 * Look for a record describing this lock, and allocate a
635 		 * new one if needed.
636 		 */
637 		bp = HASH(lb->lb_lock);
638 		SLIST_FOREACH(l, bp, bucket) {
639 			if (l->lock == lb->lb_lock)
640 				break;
641 		}
642 		if (l == NULL) {
643 			if ((l = TAILQ_FIRST(&freelist)) == NULL)
644 				l = morelocks();
645 			TAILQ_REMOVE(&freelist, l, chain);
646 			l->flags = lb->lb_flags;
647 			l->lock = lb->lb_lock;
648 			l->nbufs = 0;
649 			l->name[0] = '\0';
650 			l->count = 0;
651 			l->time = 0;
652 			TAILQ_INIT(&l->tosort);
653 			TAILQ_INIT(&l->bufs);
654 			TAILQ_INSERT_TAIL(&sortlist, l, chain);
655 			SLIST_INSERT_HEAD(bp, l, bucket);
656 		}
657 
658 		/*
659 		 * Scale the time values per buffer and summarise
660 		 * times+counts per lock.
661 		 */
662 		lb->lb_times[event] *= cpuscale[lb->lb_cpu];
663 		l->count += lb->lb_counts[event];
664 		l->time += lb->lb_times[event];
665 
666 		/*
667 		 * Merge same lock+callsite pairs from multiple CPUs
668 		 * together.
669 		 */
670 		TAILQ_FOREACH(lb2, &l->tosort, lb_chain.tailq) {
671 			if (lb->lb_callsite == lb2->lb_callsite)
672 				break;
673 		}
674 		if (lb2 != NULL) {
675 			lb2->lb_counts[event] += lb->lb_counts[event];
676 			lb2->lb_times[event] += lb->lb_times[event];
677 		} else {
678 			TAILQ_INSERT_HEAD(&l->tosort, lb, lb_chain.tailq);
679 			l->nbufs++;
680 		}
681 	}
682 
683 	/*
684 	 * Now sort the lists.
685 	 */
686 	while ((l = TAILQ_FIRST(&sortlist)) != NULL) {
687 		TAILQ_REMOVE(&sortlist, l, chain);
688 
689 		/*
690 		 * Sort the buffers into the per-lock list.
691 		 */
692 		while ((lb = TAILQ_FIRST(&l->tosort)) != NULL) {
693 			TAILQ_REMOVE(&l->tosort, lb, lb_chain.tailq);
694 
695 			lb2 = TAILQ_FIRST(&l->bufs);
696 			while (lb2 != NULL) {
697 				if (cflag) {
698 					if (lb->lb_counts[event] >
699 					    lb2->lb_counts[event])
700 						break;
701 				} else if (lb->lb_times[event] >
702 				    lb2->lb_times[event])
703 					break;
704 				lb2 = TAILQ_NEXT(lb2, lb_chain.tailq);
705 			}
706 			if (lb2 == NULL)
707 				TAILQ_INSERT_TAIL(&l->bufs, lb,
708 				    lb_chain.tailq);
709 			else
710 				TAILQ_INSERT_BEFORE(lb2, lb, lb_chain.tailq);
711 		}
712 
713 		/*
714 		 * Sort this lock into the per-type list, based on the
715 		 * totals per lock.
716 		 */
717 		l2 = TAILQ_FIRST(&locklist);
718 		while (l2 != NULL) {
719 			if (cflag) {
720 				if (l->count > l2->count)
721 					break;
722 			} else if (l->time > l2->time)
723 				break;
724 			l2 = TAILQ_NEXT(l2, chain);
725 		}
726 		if (l2 == NULL)
727 			TAILQ_INSERT_TAIL(&locklist, l, chain);
728 		else
729 			TAILQ_INSERT_BEFORE(l2, l, chain);
730 	}
731 }
732 
733 /*
734  * Display a summary table for one lock type / event type pair.
735  */
736 static void
display(int mask,const char * name)737 display(int mask, const char *name)
738 {
739 	lock_t *l;
740 	lsbuf_t *lb;
741 	double pcscale, metric;
742 	char fname[NAME_SIZE];
743 	int event;
744 
745 	event = (mask & LB_EVENT_MASK) - 1;
746 	makelists(mask, event);
747 
748 	if (TAILQ_EMPTY(&locklist))
749 		return;
750 
751 	fprintf(outfp, "\n-- %s\n\n"
752 	    "Total%%  Count   Time/ms          Lock                       Caller\n"
753 	    "------ ------- --------- ---------------------- ------------------------------\n",
754 	    name);
755 
756 	/*
757 	 * Sum up all events for this type of lock + event.
758 	 */
759 	pcscale = 0;
760 	TAILQ_FOREACH(l, &locklist, chain) {
761 		if (cflag)
762 			pcscale += l->count;
763 		else
764 			pcscale += l->time;
765 		displayed++;
766 	}
767 	if (pcscale == 0)
768 		pcscale = 100;
769 	else
770 		pcscale = (100.0 / pcscale);
771 
772 	/*
773 	 * For each lock, print a summary total, followed by a breakdown by
774 	 * caller.
775 	 */
776 	TAILQ_FOREACH(l, &locklist, chain) {
777 		if (cflag)
778 			metric = l->count;
779 		else
780 			metric = l->time;
781 		metric *= pcscale;
782 
783 		if (l->name[0] == '\0')
784 			findsym(LOCK_BYADDR, l->name, &l->lock, NULL, false);
785 
786 		if (lflag || l->nbufs > 1)
787 			fprintf(outfp, "%6.2f %7d %9.2f %-22s <all>\n",
788 			    metric, (int)(l->count * cscale),
789 			    l->time * tscale, l->name);
790 
791 		if (lflag)
792 			continue;
793 
794 		TAILQ_FOREACH(lb, &l->bufs, lb_chain.tailq) {
795 			if (cflag)
796 				metric = lb->lb_counts[event];
797 			else
798 				metric = lb->lb_times[event];
799 			metric *= pcscale;
800 
801 			findsym(FUNC_BYADDR, fname, &lb->lb_callsite, NULL,
802 			    false);
803 			fprintf(outfp, "%6.2f %7d %9.2f %-22s %s\n",
804 			    metric, (int)(lb->lb_counts[event] * cscale),
805 			    lb->lb_times[event] * tscale, l->name, fname);
806 		}
807 	}
808 }
809