xref: /freebsd/usr.sbin/pmcstat/pmcstat.c (revision 4ed925457ab06e83238a5db33e89ccc94b99a713)
1 /*-
2  * Copyright (c) 2003-2008, Joseph Koshy
3  * Copyright (c) 2007 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by A. Joseph Koshy under
7  * sponsorship from the FreeBSD Foundation and Google, Inc.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/types.h>
35 #include <sys/event.h>
36 #include <sys/param.h>
37 #include <sys/queue.h>
38 #include <sys/socket.h>
39 #include <sys/stat.h>
40 #include <sys/sysctl.h>
41 #include <sys/time.h>
42 #include <sys/ttycom.h>
43 #include <sys/user.h>
44 #include <sys/wait.h>
45 
46 #include <assert.h>
47 #include <curses.h>
48 #include <err.h>
49 #include <errno.h>
50 #include <fcntl.h>
51 #include <kvm.h>
52 #include <libgen.h>
53 #include <limits.h>
54 #include <math.h>
55 #include <pmc.h>
56 #include <pmclog.h>
57 #include <regex.h>
58 #include <signal.h>
59 #include <stdarg.h>
60 #include <stdint.h>
61 #include <stdio.h>
62 #include <stdlib.h>
63 #include <string.h>
64 #include <sysexits.h>
65 #include <unistd.h>
66 
67 #include "pmcstat.h"
68 
69 /*
70  * A given invocation of pmcstat(8) can manage multiple PMCs of both
71  * the system-wide and per-process variety.  Each of these could be in
72  * 'counting mode' or in 'sampling mode'.
73  *
74  * For 'counting mode' PMCs, pmcstat(8) will periodically issue a
75  * pmc_read() at the configured time interval and print out the value
76  * of the requested PMCs.
77  *
78  * For 'sampling mode' PMCs it can log to a file for offline analysis,
79  * or can analyse sampling data "on the fly", either by converting
80  * samples to printed textual form or by creating gprof(1) compatible
81  * profiles, one per program executed.  When creating gprof(1)
82  * profiles it can optionally merge entries from multiple processes
83  * for a given executable into a single profile file.
84  *
85  * pmcstat(8) can also execute a command line and attach PMCs to the
86  * resulting child process.  The protocol used is as follows:
87  *
88  * - parent creates a socketpair for two way communication and
89  *   fork()s.
90  * - subsequently:
91  *
92  *   /Parent/				/Child/
93  *
94  *   - Wait for childs token.
95  *					- Sends token.
96  *					- Awaits signal to start.
97  *  - Attaches PMCs to the child's pid
98  *    and starts them. Sets up
99  *    monitoring for the child.
100  *  - Signals child to start.
101  *					- Recieves signal, attempts exec().
102  *
103  * After this point normal processing can happen.
104  */
105 
106 /* Globals */
107 
108 int	pmcstat_interrupt = 0;
109 int	pmcstat_displayheight = DEFAULT_DISPLAY_HEIGHT;
110 int	pmcstat_displaywidth  = DEFAULT_DISPLAY_WIDTH;
111 int	pmcstat_sockpair[NSOCKPAIRFD];
112 int	pmcstat_kq;
113 kvm_t	*pmcstat_kvm;
114 struct kinfo_proc *pmcstat_plist;
115 struct pmcstat_args args;
116 
117 void
118 pmcstat_attach_pmcs(void)
119 {
120 	struct pmcstat_ev *ev;
121 	struct pmcstat_target *pt;
122 	int count;
123 
124 	/* Attach all process PMCs to target processes. */
125 	count = 0;
126 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
127 		if (PMC_IS_SYSTEM_MODE(ev->ev_mode))
128 			continue;
129 		SLIST_FOREACH(pt, &args.pa_targets, pt_next)
130 			if (pmc_attach(ev->ev_pmcid, pt->pt_pid) == 0)
131 				count++;
132 			else if (errno != ESRCH)
133 				err(EX_OSERR, "ERROR: cannot attach pmc "
134 				    "\"%s\" to process %d", ev->ev_name,
135 				    (int) pt->pt_pid);
136 	}
137 
138 	if (count == 0)
139 		errx(EX_DATAERR, "ERROR: No processes were attached to.");
140 }
141 
142 
143 void
144 pmcstat_cleanup(void)
145 {
146 	struct pmcstat_ev *ev, *tmp;
147 
148 	/* release allocated PMCs. */
149 	STAILQ_FOREACH_SAFE(ev, &args.pa_events, ev_next, tmp)
150 	    if (ev->ev_pmcid != PMC_ID_INVALID) {
151 		if (pmc_stop(ev->ev_pmcid) < 0)
152 			err(EX_OSERR, "ERROR: cannot stop pmc 0x%x "
153 			    "\"%s\"", ev->ev_pmcid, ev->ev_name);
154 		if (pmc_release(ev->ev_pmcid) < 0)
155 			err(EX_OSERR, "ERROR: cannot release pmc "
156 			    "0x%x \"%s\"", ev->ev_pmcid, ev->ev_name);
157 		free(ev->ev_name);
158 		free(ev->ev_spec);
159 		STAILQ_REMOVE(&args.pa_events, ev, pmcstat_ev, ev_next);
160 		free(ev);
161 	    }
162 
163 	/* de-configure the log file if present. */
164 	if (args.pa_flags & (FLAG_HAS_PIPE | FLAG_HAS_OUTPUT_LOGFILE))
165 		(void) pmc_configure_logfile(-1);
166 
167 	if (args.pa_logparser) {
168 		pmclog_close(args.pa_logparser);
169 		args.pa_logparser = NULL;
170 	}
171 
172 	if (args.pa_flags & (FLAG_HAS_PIPE | FLAG_HAS_OUTPUT_LOGFILE))
173 		pmcstat_shutdown_logging();
174 }
175 
176 void
177 pmcstat_clone_event_descriptor(struct pmcstat_ev *ev,
178     uint32_t cpumask)
179 {
180 	int cpu;
181 	struct pmcstat_ev *ev_clone;
182 
183 	while ((cpu = ffs(cpumask)) > 0) {
184 		cpu--;
185 
186 		if ((ev_clone = malloc(sizeof(*ev_clone))) == NULL)
187 			errx(EX_SOFTWARE, "ERROR: Out of memory");
188 		(void) memset(ev_clone, 0, sizeof(*ev_clone));
189 
190 		ev_clone->ev_count = ev->ev_count;
191 		ev_clone->ev_cpu   = cpu;
192 		ev_clone->ev_cumulative = ev->ev_cumulative;
193 		ev_clone->ev_flags = ev->ev_flags;
194 		ev_clone->ev_mode  = ev->ev_mode;
195 		ev_clone->ev_name  = strdup(ev->ev_name);
196 		ev_clone->ev_pmcid = ev->ev_pmcid;
197 		ev_clone->ev_saved = ev->ev_saved;
198 		ev_clone->ev_spec  = strdup(ev->ev_spec);
199 
200 		STAILQ_INSERT_TAIL(&args.pa_events, ev_clone, ev_next);
201 
202 		cpumask &= ~(1 << cpu);
203 	}
204 }
205 
206 void
207 pmcstat_create_process(void)
208 {
209 	char token;
210 	pid_t pid;
211 	struct kevent kev;
212 	struct pmcstat_target *pt;
213 
214 	if (socketpair(AF_UNIX, SOCK_STREAM, 0, pmcstat_sockpair) < 0)
215 		err(EX_OSERR, "ERROR: cannot create socket pair");
216 
217 	switch (pid = fork()) {
218 	case -1:
219 		err(EX_OSERR, "ERROR: cannot fork");
220 		/*NOTREACHED*/
221 
222 	case 0:		/* child */
223 		(void) close(pmcstat_sockpair[PARENTSOCKET]);
224 
225 		/* Write a token to tell our parent we've started executing. */
226 		if (write(pmcstat_sockpair[CHILDSOCKET], "+", 1) != 1)
227 			err(EX_OSERR, "ERROR (child): cannot write token");
228 
229 		/* Wait for our parent to signal us to start. */
230 		if (read(pmcstat_sockpair[CHILDSOCKET], &token, 1) < 0)
231 			err(EX_OSERR, "ERROR (child): cannot read token");
232 		(void) close(pmcstat_sockpair[CHILDSOCKET]);
233 
234 		/* exec() the program requested */
235 		execvp(*args.pa_argv, args.pa_argv);
236 		/* and if that fails, notify the parent */
237 		kill(getppid(), SIGCHLD);
238 		err(EX_OSERR, "ERROR: execvp \"%s\" failed", *args.pa_argv);
239 		/*NOTREACHED*/
240 
241 	default:	/* parent */
242 		(void) close(pmcstat_sockpair[CHILDSOCKET]);
243 		break;
244 	}
245 
246 	/* Ask to be notified via a kevent when the target process exits. */
247 	EV_SET(&kev, pid, EVFILT_PROC, EV_ADD|EV_ONESHOT, NOTE_EXIT, 0,
248 	    NULL);
249 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
250 		err(EX_OSERR, "ERROR: cannot monitor child process %d", pid);
251 
252 	if ((pt = malloc(sizeof(*pt))) == NULL)
253 		errx(EX_SOFTWARE, "ERROR: Out of memory.");
254 
255 	pt->pt_pid = pid;
256 	SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
257 
258 	/* Wait for the child to signal that its ready to go. */
259 	if (read(pmcstat_sockpair[PARENTSOCKET], &token, 1) < 0)
260 		err(EX_OSERR, "ERROR (parent): cannot read token");
261 
262 	return;
263 }
264 
265 void
266 pmcstat_find_targets(const char *spec)
267 {
268 	int n, nproc, pid, rv;
269 	struct pmcstat_target *pt;
270 	char errbuf[_POSIX2_LINE_MAX], *end;
271 	static struct kinfo_proc *kp;
272 	regex_t reg;
273 	regmatch_t regmatch;
274 
275 	/* First check if we've been given a process id. */
276       	pid = strtol(spec, &end, 0);
277 	if (end != spec && pid >= 0) {
278 		if ((pt = malloc(sizeof(*pt))) == NULL)
279 			goto outofmemory;
280 		pt->pt_pid = pid;
281 		SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
282 		return;
283 	}
284 
285 	/* Otherwise treat arg as a regular expression naming processes. */
286 	if (pmcstat_kvm == NULL) {
287 		if ((pmcstat_kvm = kvm_openfiles(NULL, "/dev/null", NULL, 0,
288 		    errbuf)) == NULL)
289 			err(EX_OSERR, "ERROR: Cannot open kernel \"%s\"",
290 			    errbuf);
291 		if ((pmcstat_plist = kvm_getprocs(pmcstat_kvm, KERN_PROC_PROC,
292 		    0, &nproc)) == NULL)
293 			err(EX_OSERR, "ERROR: Cannot get process list: %s",
294 			    kvm_geterr(pmcstat_kvm));
295 	}
296 
297 	if ((rv = regcomp(&reg, spec, REG_EXTENDED|REG_NOSUB)) != 0) {
298 		regerror(rv, &reg, errbuf, sizeof(errbuf));
299 		err(EX_DATAERR, "ERROR: Failed to compile regex \"%s\": %s",
300 		    spec, errbuf);
301 	}
302 
303 	for (n = 0, kp = pmcstat_plist; n < nproc; n++, kp++) {
304 		if ((rv = regexec(&reg, kp->ki_comm, 1, &regmatch, 0)) == 0) {
305 			if ((pt = malloc(sizeof(*pt))) == NULL)
306 				goto outofmemory;
307 			pt->pt_pid = kp->ki_pid;
308 			SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
309 		} else if (rv != REG_NOMATCH) {
310 			regerror(rv, &reg, errbuf, sizeof(errbuf));
311 			errx(EX_SOFTWARE, "ERROR: Regex evalation failed: %s",
312 			    errbuf);
313 		}
314 	}
315 
316 	regfree(&reg);
317 
318 	return;
319 
320  outofmemory:
321 	errx(EX_SOFTWARE, "Out of memory.");
322 	/*NOTREACHED*/
323 }
324 
325 uint32_t
326 pmcstat_get_cpumask(const char *cpuspec)
327 {
328 	uint32_t cpumask;
329 	int cpu;
330 	const char *s;
331 	char *end;
332 
333 	s = cpuspec;
334 	cpumask = 0ULL;
335 
336 	do {
337 		cpu = strtol(s, &end, 0);
338 		if (cpu < 0 || end == s)
339 			errx(EX_USAGE, "ERROR: Illegal CPU specification "
340 			    "\"%s\".", cpuspec);
341 		cpumask |= (1 << cpu);
342 		s = end + strspn(end, ", \t");
343 	} while (*s);
344 
345 	return (cpumask);
346 }
347 
348 void
349 pmcstat_kill_process(void)
350 {
351 	struct pmcstat_target *pt;
352 
353 	assert(args.pa_flags & FLAG_HAS_COMMANDLINE);
354 
355 	/*
356 	 * If a command line was specified, it would be the very first
357 	 * in the list, before any other processes specified by -t.
358 	 */
359 	pt = SLIST_FIRST(&args.pa_targets);
360 	assert(pt != NULL);
361 
362 	if (kill(pt->pt_pid, SIGINT) != 0)
363 		err(EX_OSERR, "ERROR: cannot signal child process");
364 }
365 
366 void
367 pmcstat_start_pmcs(void)
368 {
369 	struct pmcstat_ev *ev;
370 
371 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
372 
373 	    assert(ev->ev_pmcid != PMC_ID_INVALID);
374 
375 	    if (pmc_start(ev->ev_pmcid) < 0) {
376 	        warn("ERROR: Cannot start pmc 0x%x \"%s\"",
377 		    ev->ev_pmcid, ev->ev_name);
378 		pmcstat_cleanup();
379 		exit(EX_OSERR);
380 	    }
381 	}
382 
383 }
384 
385 void
386 pmcstat_print_headers(void)
387 {
388 	struct pmcstat_ev *ev;
389 	int c, w;
390 
391 	(void) fprintf(args.pa_printfile, PRINT_HEADER_PREFIX);
392 
393 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
394 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
395 			continue;
396 
397 		c = PMC_IS_SYSTEM_MODE(ev->ev_mode) ? 's' : 'p';
398 
399 		if (ev->ev_fieldskip != 0)
400 			(void) fprintf(args.pa_printfile, "%*s",
401 			    ev->ev_fieldskip, "");
402 		w = ev->ev_fieldwidth - ev->ev_fieldskip - 2;
403 
404 		if (c == 's')
405 			(void) fprintf(args.pa_printfile, "s/%02d/%-*s ",
406 			    ev->ev_cpu, w-3, ev->ev_name);
407 		else
408 			(void) fprintf(args.pa_printfile, "p/%*s ", w,
409 			    ev->ev_name);
410 	}
411 
412 	(void) fflush(args.pa_printfile);
413 }
414 
415 void
416 pmcstat_print_counters(void)
417 {
418 	int extra_width;
419 	struct pmcstat_ev *ev;
420 	pmc_value_t value;
421 
422 	extra_width = sizeof(PRINT_HEADER_PREFIX) - 1;
423 
424 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
425 
426 		/* skip sampling mode counters */
427 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
428 			continue;
429 
430 		if (pmc_read(ev->ev_pmcid, &value) < 0)
431 			err(EX_OSERR, "ERROR: Cannot read pmc "
432 			    "\"%s\"", ev->ev_name);
433 
434 		(void) fprintf(args.pa_printfile, "%*ju ",
435 		    ev->ev_fieldwidth + extra_width,
436 		    (uintmax_t) ev->ev_cumulative ? value :
437 		    (value - ev->ev_saved));
438 
439 		if (ev->ev_cumulative == 0)
440 			ev->ev_saved = value;
441 		extra_width = 0;
442 	}
443 
444 	(void) fflush(args.pa_printfile);
445 }
446 
447 /*
448  * Print output
449  */
450 
451 void
452 pmcstat_print_pmcs(void)
453 {
454 	static int linecount = 0;
455 
456 	/* check if we need to print a header line */
457 	if (++linecount > pmcstat_displayheight) {
458 		(void) fprintf(args.pa_printfile, "\n");
459 		linecount = 1;
460 	}
461 	if (linecount == 1)
462 		pmcstat_print_headers();
463 	(void) fprintf(args.pa_printfile, "\n");
464 
465 	pmcstat_print_counters();
466 
467 	return;
468 }
469 
470 /*
471  * Do process profiling
472  *
473  * If a pid was specified, attach each allocated PMC to the target
474  * process.  Otherwise, fork a child and attach the PMCs to the child,
475  * and have the child exec() the target program.
476  */
477 
478 void
479 pmcstat_start_process(void)
480 {
481 	/* Signal the child to proceed. */
482 	if (write(pmcstat_sockpair[PARENTSOCKET], "!", 1) != 1)
483 		err(EX_OSERR, "ERROR (parent): write of token failed");
484 
485 	(void) close(pmcstat_sockpair[PARENTSOCKET]);
486 }
487 
488 void
489 pmcstat_show_usage(void)
490 {
491 	errx(EX_USAGE,
492 	    "[options] [commandline]\n"
493 	    "\t Measure process and/or system performance using hardware\n"
494 	    "\t performance monitoring counters.\n"
495 	    "\t Options include:\n"
496 	    "\t -C\t\t (toggle) show cumulative counts\n"
497 	    "\t -D path\t create profiles in directory \"path\"\n"
498 	    "\t -E\t\t (toggle) show counts at process exit\n"
499 	    "\t -F file\t write a system-wide callgraph (Kcachegrind format)"
500 		" to \"file\"\n"
501 	    "\t -G file\t write a system-wide callgraph to \"file\"\n"
502 	    "\t -M file\t print executable/gmon file map to \"file\"\n"
503 	    "\t -N\t\t (toggle) capture callchains\n"
504 	    "\t -O file\t send log output to \"file\"\n"
505 	    "\t -P spec\t allocate a process-private sampling PMC\n"
506 	    "\t -R file\t read events from \"file\"\n"
507 	    "\t -S spec\t allocate a system-wide sampling PMC\n"
508 	    "\t -T\t\t start in top mode\n"
509 	    "\t -W\t\t (toggle) show counts per context switch\n"
510 	    "\t -c cpu-list\t set cpus for subsequent system-wide PMCs\n"
511 	    "\t -d\t\t (toggle) track descendants\n"
512 	    "\t -f spec\t pass \"spec\" to as plugin option\n"
513 	    "\t -g\t\t produce gprof(1) compatible profiles\n"
514 	    "\t -k dir\t\t set the path to the kernel\n"
515 	    "\t -n rate\t set sampling rate\n"
516 	    "\t -o file\t send print output to \"file\"\n"
517 	    "\t -p spec\t allocate a process-private counting PMC\n"
518 	    "\t -q\t\t suppress verbosity\n"
519 	    "\t -r fsroot\t specify FS root directory\n"
520 	    "\t -s spec\t allocate a system-wide counting PMC\n"
521 	    "\t -t process-spec attach to running processes matching "
522 		"\"process-spec\"\n"
523 	    "\t -v\t\t increase verbosity\n"
524 	    "\t -w secs\t set printing time interval\n"
525 	    "\t -z depth\t limit callchain display depth"
526 	);
527 }
528 
529 /*
530  * At exit handler for top mode
531  */
532 
533 void
534 pmcstat_topexit(void)
535 {
536 	if (!args.pa_toptty)
537 		return;
538 
539 	/*
540 	 * Shutdown ncurses.
541 	 */
542 	clrtoeol();
543 	refresh();
544 	endwin();
545 }
546 
547 /*
548  * Main
549  */
550 
551 int
552 main(int argc, char **argv)
553 {
554 	double interval;
555 	int option, npmc, ncpu, haltedcpus;
556 	int c, check_driver_stats, current_cpu, current_sampling_count;
557 	int do_callchain, do_descendants, do_logproccsw, do_logprocexit;
558 	int do_print;
559 	size_t dummy;
560 	int graphdepth;
561 	int pipefd[2];
562 	int use_cumulative_counts;
563 	short cf, cb;
564 	uint32_t cpumask;
565 	char *end, *tmp;
566 	const char *errmsg, *graphfilename;
567 	enum pmcstat_state runstate;
568 	struct pmc_driverstats ds_start, ds_end;
569 	struct pmcstat_ev *ev;
570 	struct sigaction sa;
571 	struct kevent kev;
572 	struct winsize ws;
573 	struct stat sb;
574 	char buffer[PATH_MAX];
575 
576 	check_driver_stats      = 0;
577 	current_cpu 		= 0;
578 	current_sampling_count  = DEFAULT_SAMPLE_COUNT;
579 	do_callchain		= 1;
580 	do_descendants          = 0;
581 	do_logproccsw           = 0;
582 	do_logprocexit          = 0;
583 	use_cumulative_counts   = 0;
584 	graphfilename		= "-";
585 	args.pa_required	= 0;
586 	args.pa_flags		= 0;
587 	args.pa_verbosity	= 1;
588 	args.pa_logfd		= -1;
589 	args.pa_fsroot		= "";
590 	args.pa_kernel		= strdup("/boot/kernel");
591 	args.pa_samplesdir	= ".";
592 	args.pa_printfile	= stderr;
593 	args.pa_graphdepth	= DEFAULT_CALLGRAPH_DEPTH;
594 	args.pa_graphfile	= NULL;
595 	args.pa_interval	= DEFAULT_WAIT_INTERVAL;
596 	args.pa_mapfilename	= NULL;
597 	args.pa_inputpath	= NULL;
598 	args.pa_outputpath	= NULL;
599 	args.pa_pplugin		= PMCSTAT_PL_NONE;
600 	args.pa_plugin		= PMCSTAT_PL_NONE;
601 	args.pa_ctdumpinstr	= 1;
602 	args.pa_topmode		= PMCSTAT_TOP_DELTA;
603 	args.pa_toptty		= 0;
604 	args.pa_topcolor	= 0;
605 	args.pa_mergepmc	= 0;
606 	STAILQ_INIT(&args.pa_events);
607 	SLIST_INIT(&args.pa_targets);
608 	bzero(&ds_start, sizeof(ds_start));
609 	bzero(&ds_end, sizeof(ds_end));
610 	ev = NULL;
611 
612 	/*
613 	 * The initial CPU mask specifies all non-halted CPUS in the
614 	 * system.
615 	 */
616 	dummy = sizeof(int);
617 	if (sysctlbyname("hw.ncpu", &ncpu, &dummy, NULL, 0) < 0)
618 		err(EX_OSERR, "ERROR: Cannot determine the number of CPUs");
619 	cpumask = (1 << ncpu) - 1;
620 	haltedcpus = 0;
621 	if (ncpu > 1) {
622 		if (sysctlbyname("machdep.hlt_cpus", &haltedcpus, &dummy,
623 		    NULL, 0) < 0)
624 			err(EX_OSERR, "ERROR: Cannot determine which CPUs are "
625 			    "halted");
626 		cpumask &= ~haltedcpus;
627 	}
628 
629 	while ((option = getopt(argc, argv,
630 	    "CD:EF:G:M:NO:P:R:S:TWc:df:gk:m:n:o:p:qr:s:t:vw:z:")) != -1)
631 		switch (option) {
632 		case 'C':	/* cumulative values */
633 			use_cumulative_counts = !use_cumulative_counts;
634 			args.pa_required |= FLAG_HAS_COUNTING_PMCS;
635 			break;
636 
637 		case 'c':	/* CPU */
638 
639 			if (optarg[0] == '*' && optarg[1] == '\0')
640 				cpumask = ((1 << ncpu) - 1) & ~haltedcpus;
641 			else
642 				cpumask = pmcstat_get_cpumask(optarg);
643 
644 			args.pa_required |= FLAG_HAS_SYSTEM_PMCS;
645 			break;
646 
647 		case 'D':
648 			if (stat(optarg, &sb) < 0)
649 				err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
650 				    optarg);
651 			if (!S_ISDIR(sb.st_mode))
652 				errx(EX_USAGE, "ERROR: \"%s\" is not a "
653 				    "directory.", optarg);
654 			args.pa_samplesdir = optarg;
655 			args.pa_flags     |= FLAG_HAS_SAMPLESDIR;
656 			args.pa_required  |= FLAG_DO_GPROF;
657 			break;
658 
659 		case 'd':	/* toggle descendents */
660 			do_descendants = !do_descendants;
661 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
662 			break;
663 
664 		case 'F':	/* produce a system-wide calltree */
665 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
666 			args.pa_plugin = PMCSTAT_PL_CALLTREE;
667 			graphfilename = optarg;
668 			break;
669 
670 		case 'f':	/* plugins options */
671 			if (args.pa_plugin == PMCSTAT_PL_NONE)
672 				err(EX_USAGE, "ERROR: Need -g/-G/-m/-T.");
673 			pmcstat_pluginconfigure_log(optarg);
674 			break;
675 
676 		case 'G':	/* produce a system-wide callgraph */
677 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
678 			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
679 			graphfilename = optarg;
680 			break;
681 
682 		case 'g':	/* produce gprof compatible profiles */
683 			args.pa_flags |= FLAG_DO_GPROF;
684 			args.pa_pplugin = PMCSTAT_PL_CALLGRAPH;
685 			args.pa_plugin	= PMCSTAT_PL_GPROF;
686 			break;
687 
688 		case 'k':	/* pathname to the kernel */
689 			free(args.pa_kernel);
690 			args.pa_kernel = strdup(optarg);
691 			args.pa_required |= FLAG_DO_ANALYSIS;
692 			args.pa_flags    |= FLAG_HAS_KERNELPATH;
693 			break;
694 
695 		case 'm':
696 			args.pa_flags |= FLAG_DO_ANNOTATE;
697 			args.pa_plugin = PMCSTAT_PL_ANNOTATE;
698 			graphfilename  = optarg;
699 			break;
700 
701 		case 'E':	/* log process exit */
702 			do_logprocexit = !do_logprocexit;
703 			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
704 			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
705 			break;
706 
707 		case 'M':	/* mapfile */
708 			args.pa_mapfilename = optarg;
709 			break;
710 
711 		case 'N':
712 			do_callchain = !do_callchain;
713 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
714 			break;
715 
716 		case 'p':	/* process virtual counting PMC */
717 		case 's':	/* system-wide counting PMC */
718 		case 'P':	/* process virtual sampling PMC */
719 		case 'S':	/* system-wide sampling PMC */
720 			if ((ev = malloc(sizeof(*ev))) == NULL)
721 				errx(EX_SOFTWARE, "ERROR: Out of memory.");
722 
723 			switch (option) {
724 			case 'p': ev->ev_mode = PMC_MODE_TC; break;
725 			case 's': ev->ev_mode = PMC_MODE_SC; break;
726 			case 'P': ev->ev_mode = PMC_MODE_TS; break;
727 			case 'S': ev->ev_mode = PMC_MODE_SS; break;
728 			}
729 
730 			if (option == 'P' || option == 'p') {
731 				args.pa_flags |= FLAG_HAS_PROCESS_PMCS;
732 				args.pa_required |= (FLAG_HAS_COMMANDLINE |
733 				    FLAG_HAS_TARGET);
734 			}
735 
736 			if (option == 'P' || option == 'S') {
737 				args.pa_flags |= FLAG_HAS_SAMPLING_PMCS;
738 				args.pa_required |= (FLAG_HAS_PIPE |
739 				    FLAG_HAS_OUTPUT_LOGFILE);
740 			}
741 
742 			if (option == 'p' || option == 's')
743 				args.pa_flags |= FLAG_HAS_COUNTING_PMCS;
744 
745 			if (option == 's' || option == 'S')
746 				args.pa_flags |= FLAG_HAS_SYSTEM_PMCS;
747 
748 			ev->ev_spec  = strdup(optarg);
749 
750 			if (option == 'S' || option == 'P')
751 				ev->ev_count = current_sampling_count;
752 			else
753 				ev->ev_count = -1;
754 
755 			if (option == 'S' || option == 's')
756 				ev->ev_cpu = ffs(cpumask) - 1;
757 			else
758 				ev->ev_cpu = PMC_CPU_ANY;
759 
760 			ev->ev_flags = 0;
761 			if (do_callchain)
762 				ev->ev_flags |= PMC_F_CALLCHAIN;
763 			if (do_descendants)
764 				ev->ev_flags |= PMC_F_DESCENDANTS;
765 			if (do_logprocexit)
766 				ev->ev_flags |= PMC_F_LOG_PROCEXIT;
767 			if (do_logproccsw)
768 				ev->ev_flags |= PMC_F_LOG_PROCCSW;
769 
770 			ev->ev_cumulative  = use_cumulative_counts;
771 
772 			ev->ev_saved = 0LL;
773 			ev->ev_pmcid = PMC_ID_INVALID;
774 
775 			/* extract event name */
776 			c = strcspn(optarg, ", \t");
777 			ev->ev_name = malloc(c + 1);
778 			(void) strncpy(ev->ev_name, optarg, c);
779 			*(ev->ev_name + c) = '\0';
780 
781 			STAILQ_INSERT_TAIL(&args.pa_events, ev, ev_next);
782 
783 			if (option == 's' || option == 'S')
784 				pmcstat_clone_event_descriptor(ev,
785 				    cpumask & ~(1 << ev->ev_cpu));
786 
787 			break;
788 
789 		case 'n':	/* sampling count */
790 			current_sampling_count = strtol(optarg, &end, 0);
791 			if (*end != '\0' || current_sampling_count <= 0)
792 				errx(EX_USAGE,
793 				    "ERROR: Illegal count value \"%s\".",
794 				    optarg);
795 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
796 			break;
797 
798 		case 'o':	/* outputfile */
799 			if (args.pa_printfile != NULL)
800 				(void) fclose(args.pa_printfile);
801 			if ((args.pa_printfile = fopen(optarg, "w")) == NULL)
802 				errx(EX_OSERR, "ERROR: cannot open \"%s\" for "
803 				    "writing.", optarg);
804 			args.pa_flags |= FLAG_DO_PRINT;
805 			break;
806 
807 		case 'O':	/* sampling output */
808 			if (args.pa_outputpath)
809 				errx(EX_USAGE, "ERROR: option -O may only be "
810 				    "specified once.");
811 			args.pa_outputpath = optarg;
812 			args.pa_flags |= FLAG_HAS_OUTPUT_LOGFILE;
813 			break;
814 
815 		case 'q':	/* quiet mode */
816 			args.pa_verbosity = 0;
817 			break;
818 
819 		case 'r':	/* root FS path */
820 			args.pa_fsroot = optarg;
821 			break;
822 
823 		case 'R':	/* read an existing log file */
824 			if (args.pa_inputpath != NULL)
825 				errx(EX_USAGE, "ERROR: option -R may only be "
826 				    "specified once.");
827 			args.pa_inputpath = optarg;
828 			if (args.pa_printfile == stderr)
829 				args.pa_printfile = stdout;
830 			args.pa_flags |= FLAG_READ_LOGFILE;
831 			break;
832 
833 		case 't':	/* target pid or process name */
834 			pmcstat_find_targets(optarg);
835 
836 			args.pa_flags |= FLAG_HAS_TARGET;
837 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
838 			break;
839 
840 		case 'T':	/* top mode */
841 			args.pa_flags |= FLAG_DO_TOP;
842 			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
843 			args.pa_ctdumpinstr = 0;
844 			args.pa_mergepmc = 1;
845 			if (args.pa_printfile == stderr)
846 				args.pa_printfile = stdout;
847 			break;
848 
849 		case 'v':	/* verbose */
850 			args.pa_verbosity++;
851 			break;
852 
853 		case 'w':	/* wait interval */
854 			interval = strtod(optarg, &end);
855 			if (*end != '\0' || interval <= 0)
856 				errx(EX_USAGE, "ERROR: Illegal wait interval "
857 				    "value \"%s\".", optarg);
858 			args.pa_flags |= FLAG_HAS_WAIT_INTERVAL;
859 			args.pa_interval = interval;
860 			break;
861 
862 		case 'W':	/* toggle LOG_CSW */
863 			do_logproccsw = !do_logproccsw;
864 			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
865 			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
866 			break;
867 
868 		case 'z':
869 			graphdepth = strtod(optarg, &end);
870 			if (*end != '\0' || graphdepth <= 0)
871 				errx(EX_USAGE, "ERROR: Illegal callchain "
872 				    "depth \"%s\".", optarg);
873 			args.pa_graphdepth = graphdepth;
874 			args.pa_required |= FLAG_DO_CALLGRAPHS;
875 			break;
876 
877 		case '?':
878 		default:
879 			pmcstat_show_usage();
880 			break;
881 
882 		}
883 
884 	args.pa_argc = (argc -= optind);
885 	args.pa_argv = (argv += optind);
886 
887 	args.pa_cpumask = cpumask; /* For selecting CPUs using -R. */
888 
889 	if (argc)	/* command line present */
890 		args.pa_flags |= FLAG_HAS_COMMANDLINE;
891 
892 	if (args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS |
893 	    FLAG_DO_ANNOTATE | FLAG_DO_TOP))
894 		args.pa_flags |= FLAG_DO_ANALYSIS;
895 
896 	/*
897 	 * Check invocation syntax.
898 	 */
899 
900 	/* disallow -O and -R together */
901 	if (args.pa_outputpath && args.pa_inputpath)
902 		errx(EX_USAGE, "ERROR: options -O and -R are mutually "
903 		    "exclusive.");
904 
905 	/* -m option is allowed with -R only. */
906 	if (args.pa_flags & FLAG_DO_ANNOTATE && args.pa_inputpath == NULL)
907 		errx(EX_USAGE, "ERROR: option -m requires an input file");
908 
909 	/* -m option is not allowed combined with -g or -G. */
910 	if (args.pa_flags & FLAG_DO_ANNOTATE &&
911 	    args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS))
912 		errx(EX_USAGE, "ERROR: option -m and -g | -G are mutually "
913 		    "exclusive");
914 
915 	if (args.pa_flags & FLAG_READ_LOGFILE) {
916 		errmsg = NULL;
917 		if (args.pa_flags & FLAG_HAS_COMMANDLINE)
918 			errmsg = "a command line specification";
919 		else if (args.pa_flags & FLAG_HAS_TARGET)
920 			errmsg = "option -t";
921 		else if (!STAILQ_EMPTY(&args.pa_events))
922 			errmsg = "a PMC event specification";
923 		if (errmsg)
924 			errx(EX_USAGE, "ERROR: option -R may not be used with "
925 			    "%s.", errmsg);
926 	} else if (STAILQ_EMPTY(&args.pa_events))
927 		/* All other uses require a PMC spec. */
928 		pmcstat_show_usage();
929 
930 	/* check for -t pid without a process PMC spec */
931 	if ((args.pa_required & FLAG_HAS_TARGET) &&
932 	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
933 		errx(EX_USAGE, "ERROR: option -t requires a process mode PMC "
934 		    "to be specified.");
935 
936 	/* check for process-mode options without a command or -t pid */
937 	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
938 	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
939 		errx(EX_USAGE, "ERROR: options -d, -E, -p, -P, and -W require "
940 		    "a command line or target process.");
941 
942 	/* check for -p | -P without a target process of some sort */
943 	if ((args.pa_required & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) &&
944 	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
945 		errx(EX_USAGE, "ERROR: options -P and -p require a "
946 		    "target process or a command line.");
947 
948 	/* check for process-mode options without a process-mode PMC */
949 	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
950 	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
951 		errx(EX_USAGE, "ERROR: options -d, -E, and -W require a "
952 		    "process mode PMC to be specified.");
953 
954 	/* check for -c cpu with no system mode PMCs or logfile. */
955 	if ((args.pa_required & FLAG_HAS_SYSTEM_PMCS) &&
956 	    (args.pa_flags & FLAG_HAS_SYSTEM_PMCS) == 0 &&
957 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
958 		errx(EX_USAGE, "ERROR: option -c requires at least one "
959 		    "system mode PMC to be specified.");
960 
961 	/* check for counting mode options without a counting PMC */
962 	if ((args.pa_required & FLAG_HAS_COUNTING_PMCS) &&
963 	    (args.pa_flags & FLAG_HAS_COUNTING_PMCS) == 0)
964 		errx(EX_USAGE, "ERROR: options -C, -W and -o require at "
965 		    "least one counting mode PMC to be specified.");
966 
967 	/* check for sampling mode options without a sampling PMC spec */
968 	if ((args.pa_required & FLAG_HAS_SAMPLING_PMCS) &&
969 	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) == 0)
970 		errx(EX_USAGE, "ERROR: options -N, -n and -O require at "
971 		    "least one sampling mode PMC to be specified.");
972 
973 	/* check if -g/-G/-m/-T are being used correctly */
974 	if ((args.pa_flags & FLAG_DO_ANALYSIS) &&
975 	    !(args.pa_flags & (FLAG_HAS_SAMPLING_PMCS|FLAG_READ_LOGFILE)))
976 		errx(EX_USAGE, "ERROR: options -g/-G/-m/-T require sampling PMCs "
977 		    "or -R to be specified.");
978 
979 	/* check if -O was spuriously specified */
980 	if ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) &&
981 	    (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0)
982 		errx(EX_USAGE,
983 		    "ERROR: option -O is used only with options "
984 		    "-E, -P, -S and -W.");
985 
986 	/* -k kernel path require -g/-G/-m/-T or -R */
987 	if ((args.pa_flags & FLAG_HAS_KERNELPATH) &&
988 	    (args.pa_flags & FLAG_DO_ANALYSIS) == 0 &&
989 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
990 	    errx(EX_USAGE, "ERROR: option -k is only used with -g/-R/-m/-T.");
991 
992 	/* -D only applies to gprof output mode (-g) */
993 	if ((args.pa_flags & FLAG_HAS_SAMPLESDIR) &&
994 	    (args.pa_flags & FLAG_DO_GPROF) == 0)
995 	    errx(EX_USAGE, "ERROR: option -D is only used with -g.");
996 
997 	/* -M mapfile requires -g or -R */
998 	if (args.pa_mapfilename != NULL &&
999 	    (args.pa_flags & FLAG_DO_GPROF) == 0 &&
1000 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
1001 	    errx(EX_USAGE, "ERROR: option -M is only used with -g/-R.");
1002 
1003 	/* -T is incompatible with -R (replay logfile is a TODO) */
1004 	if ((args.pa_flags & FLAG_DO_TOP) &&
1005 	    (args.pa_flags & FLAG_READ_LOGFILE))
1006 		errx(EX_USAGE, "ERROR: option -T is incompatible with -R.");
1007 
1008 	/*
1009 	 * Disallow textual output of sampling PMCs if counting PMCs
1010 	 * have also been asked for, mostly because the combined output
1011 	 * is difficult to make sense of.
1012 	 */
1013 	if ((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1014 	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
1015 	    ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) == 0))
1016 		errx(EX_USAGE, "ERROR: option -O is required if counting and "
1017 		    "sampling PMCs are specified together.");
1018 
1019 	/*
1020 	 * Check if "-k kerneldir" was specified, and if whether
1021 	 * 'kerneldir' actually refers to a a file.  If so, use
1022 	 * `dirname path` to determine the kernel directory.
1023 	 */
1024 	if (args.pa_flags & FLAG_HAS_KERNELPATH) {
1025 		(void) snprintf(buffer, sizeof(buffer), "%s%s", args.pa_fsroot,
1026 		    args.pa_kernel);
1027 		if (stat(buffer, &sb) < 0)
1028 			err(EX_OSERR, "ERROR: Cannot locate kernel \"%s\"",
1029 			    buffer);
1030 		if (!S_ISREG(sb.st_mode) && !S_ISDIR(sb.st_mode))
1031 			errx(EX_USAGE, "ERROR: \"%s\": Unsupported file type.",
1032 			    buffer);
1033 		if (!S_ISDIR(sb.st_mode)) {
1034 			tmp = args.pa_kernel;
1035 			args.pa_kernel = strdup(dirname(args.pa_kernel));
1036 			free(tmp);
1037 			(void) snprintf(buffer, sizeof(buffer), "%s%s",
1038 			    args.pa_fsroot, args.pa_kernel);
1039 			if (stat(buffer, &sb) < 0)
1040 				err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
1041 				    buffer);
1042 			if (!S_ISDIR(sb.st_mode))
1043 				errx(EX_USAGE, "ERROR: \"%s\" is not a "
1044 				    "directory.", buffer);
1045 		}
1046 	}
1047 
1048 	/*
1049 	 * If we have a callgraph be created, select the outputfile.
1050 	 */
1051 	if (args.pa_flags & FLAG_DO_CALLGRAPHS) {
1052 		if (strcmp(graphfilename, "-") == 0)
1053 		    args.pa_graphfile = args.pa_printfile;
1054 		else {
1055 			args.pa_graphfile = fopen(graphfilename, "w");
1056 			if (args.pa_graphfile == NULL)
1057 				err(EX_OSERR, "ERROR: cannot open \"%s\" "
1058 				    "for writing", graphfilename);
1059 		}
1060 	}
1061 	if (args.pa_flags & FLAG_DO_ANNOTATE) {
1062 		args.pa_graphfile = fopen(graphfilename, "w");
1063 		if (args.pa_graphfile == NULL)
1064 			err(EX_OSERR, "ERROR: cannot open \"%s\" for writing",
1065 			    graphfilename);
1066 	}
1067 
1068 	/* if we've been asked to process a log file, do that and exit */
1069 	if (args.pa_flags & FLAG_READ_LOGFILE) {
1070 		/*
1071 		 * Print the log in textual form if we haven't been
1072 		 * asked to generate profiling information.
1073 		 */
1074 		if ((args.pa_flags & FLAG_DO_ANALYSIS) == 0)
1075 			args.pa_flags |= FLAG_DO_PRINT;
1076 
1077 		pmcstat_initialize_logging();
1078 		args.pa_logfd = pmcstat_open_log(args.pa_inputpath,
1079 		    PMCSTAT_OPEN_FOR_READ);
1080 		if ((args.pa_logparser = pmclog_open(args.pa_logfd)) == NULL)
1081 			err(EX_OSERR, "ERROR: Cannot create parser");
1082 		pmcstat_process_log();
1083 		pmcstat_shutdown_logging();
1084 		exit(EX_OK);
1085 	}
1086 
1087 	/* otherwise, we've been asked to collect data */
1088 	if (pmc_init() < 0)
1089 		err(EX_UNAVAILABLE,
1090 		    "ERROR: Initialization of the pmc(3) library failed");
1091 
1092 	if ((npmc = pmc_npmc(0)) < 0) /* assume all CPUs are identical */
1093 		err(EX_OSERR, "ERROR: Cannot determine the number of PMCs "
1094 		    "on CPU %d", 0);
1095 
1096 	/* Allocate a kqueue */
1097 	if ((pmcstat_kq = kqueue()) < 0)
1098 		err(EX_OSERR, "ERROR: Cannot allocate kqueue");
1099 
1100 	/*
1101 	 * Configure the specified log file or setup a default log
1102 	 * consumer via a pipe.
1103 	 */
1104 	if (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) {
1105 		if (args.pa_outputpath)
1106 			args.pa_logfd = pmcstat_open_log(args.pa_outputpath,
1107 			    PMCSTAT_OPEN_FOR_WRITE);
1108 		else {
1109 			/*
1110 			 * process the log on the fly by reading it in
1111 			 * through a pipe.
1112 			 */
1113 			if (pipe(pipefd) < 0)
1114 				err(EX_OSERR, "ERROR: pipe(2) failed");
1115 
1116 			if (fcntl(pipefd[READPIPEFD], F_SETFL, O_NONBLOCK) < 0)
1117 				err(EX_OSERR, "ERROR: fcntl(2) failed");
1118 
1119 			EV_SET(&kev, pipefd[READPIPEFD], EVFILT_READ, EV_ADD,
1120 			    0, 0, NULL);
1121 
1122 			if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1123 				err(EX_OSERR, "ERROR: Cannot register kevent");
1124 
1125 			args.pa_logfd = pipefd[WRITEPIPEFD];
1126 
1127 			args.pa_flags |= FLAG_HAS_PIPE;
1128 			if ((args.pa_flags & FLAG_DO_TOP) == 0)
1129 				args.pa_flags |= FLAG_DO_PRINT;
1130 			args.pa_logparser = pmclog_open(pipefd[READPIPEFD]);
1131 		}
1132 
1133 		if (pmc_configure_logfile(args.pa_logfd) < 0)
1134 			err(EX_OSERR, "ERROR: Cannot configure log file");
1135 	}
1136 
1137 	/* remember to check for driver errors if we are sampling or logging */
1138 	check_driver_stats = (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) ||
1139 	    (args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE);
1140 
1141 	/*
1142 	 * Allocate PMCs.
1143 	 */
1144 
1145 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1146 	    if (pmc_allocate(ev->ev_spec, ev->ev_mode,
1147 		    ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid) < 0)
1148 		    err(EX_OSERR, "ERROR: Cannot allocate %s-mode pmc with "
1149 			"specification \"%s\"",
1150 			PMC_IS_SYSTEM_MODE(ev->ev_mode) ? "system" : "process",
1151 			ev->ev_spec);
1152 
1153 	    if (PMC_IS_SAMPLING_MODE(ev->ev_mode) &&
1154 		pmc_set(ev->ev_pmcid, ev->ev_count) < 0)
1155 		    err(EX_OSERR, "ERROR: Cannot set sampling count "
1156 			"for PMC \"%s\"", ev->ev_name);
1157 	}
1158 
1159 	/* compute printout widths */
1160 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1161 		int counter_width;
1162 		int display_width;
1163 		int header_width;
1164 
1165 		(void) pmc_width(ev->ev_pmcid, &counter_width);
1166 		header_width = strlen(ev->ev_name) + 2; /* prefix '%c/' */
1167 		display_width = (int) floor(counter_width / 3.32193) + 1;
1168 
1169 		if (PMC_IS_SYSTEM_MODE(ev->ev_mode))
1170 			header_width += 3; /* 2 digit CPU number + '/' */
1171 
1172 		if (header_width > display_width) {
1173 			ev->ev_fieldskip = 0;
1174 			ev->ev_fieldwidth = header_width;
1175 		} else {
1176 			ev->ev_fieldskip = display_width -
1177 			    header_width;
1178 			ev->ev_fieldwidth = display_width;
1179 		}
1180 	}
1181 
1182 	/*
1183 	 * If our output is being set to a terminal, register a handler
1184 	 * for window size changes.
1185 	 */
1186 
1187 	if (isatty(fileno(args.pa_printfile))) {
1188 
1189 		if (ioctl(fileno(args.pa_printfile), TIOCGWINSZ, &ws) < 0)
1190 			err(EX_OSERR, "ERROR: Cannot determine window size");
1191 
1192 		pmcstat_displayheight = ws.ws_row - 1;
1193 		pmcstat_displaywidth  = ws.ws_col - 1;
1194 
1195 		EV_SET(&kev, SIGWINCH, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1196 
1197 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1198 			err(EX_OSERR, "ERROR: Cannot register kevent for "
1199 			    "SIGWINCH");
1200 
1201 		args.pa_toptty = 1;
1202 	}
1203 
1204 	/*
1205 	 * Listen to key input in top mode.
1206 	 */
1207 	if (args.pa_flags & FLAG_DO_TOP) {
1208 		EV_SET(&kev, fileno(stdin), EVFILT_READ, EV_ADD, 0, 0, NULL);
1209 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1210 			err(EX_OSERR, "ERROR: Cannot register kevent");
1211 	}
1212 
1213 	EV_SET(&kev, SIGINT, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1214 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1215 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGINT");
1216 
1217 	EV_SET(&kev, SIGIO, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1218 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1219 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGIO");
1220 
1221 	/*
1222 	 * An exec() failure of a forked child is signalled by the
1223 	 * child sending the parent a SIGCHLD.  We don't register an
1224 	 * actual signal handler for SIGCHLD, but instead use our
1225 	 * kqueue to pick up the signal.
1226 	 */
1227 	EV_SET(&kev, SIGCHLD, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1228 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1229 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGCHLD");
1230 
1231 	/*
1232 	 * Setup a timer if we have counting mode PMCs needing to be printed or
1233 	 * top mode plugin is active.
1234 	 */
1235 	if (((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1236 	     (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) ||
1237 	    (args.pa_flags & FLAG_DO_TOP)) {
1238 		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1239 		    args.pa_interval * 1000, NULL);
1240 
1241 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1242 			err(EX_OSERR, "ERROR: Cannot register kevent for "
1243 			    "timer");
1244 	}
1245 
1246 	/* attach PMCs to the target process, starting it if specified */
1247 	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1248 		pmcstat_create_process();
1249 
1250 	if (check_driver_stats && pmc_get_driver_stats(&ds_start) < 0)
1251 		err(EX_OSERR, "ERROR: Cannot retrieve driver statistics");
1252 
1253 	/* Attach process pmcs to the target process. */
1254 	if (args.pa_flags & (FLAG_HAS_TARGET | FLAG_HAS_COMMANDLINE)) {
1255 		if (SLIST_EMPTY(&args.pa_targets))
1256 			errx(EX_DATAERR, "ERROR: No matching target "
1257 			    "processes.");
1258 		if (args.pa_flags & FLAG_HAS_PROCESS_PMCS)
1259 			pmcstat_attach_pmcs();
1260 
1261 		if (pmcstat_kvm) {
1262 			kvm_close(pmcstat_kvm);
1263 			pmcstat_kvm = NULL;
1264 		}
1265 	}
1266 
1267 	/* start the pmcs */
1268 	pmcstat_start_pmcs();
1269 
1270 	/* start the (commandline) process if needed */
1271 	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1272 		pmcstat_start_process();
1273 
1274 	/* initialize logging if printing the configured log */
1275 	if ((args.pa_flags & (FLAG_DO_PRINT | FLAG_DO_TOP)) &&
1276 	    (args.pa_flags & (FLAG_HAS_PIPE | FLAG_HAS_OUTPUT_LOGFILE)))
1277 		pmcstat_initialize_logging();
1278 
1279 	/* Handle SIGINT using the kqueue loop */
1280 	sa.sa_handler = SIG_IGN;
1281 	sa.sa_flags   = 0;
1282 	(void) sigemptyset(&sa.sa_mask);
1283 
1284 	if (sigaction(SIGINT, &sa, NULL) < 0)
1285 		err(EX_OSERR, "ERROR: Cannot install signal handler");
1286 
1287 	/*
1288 	 * Setup the top mode display.
1289 	 */
1290 	if (args.pa_flags & FLAG_DO_TOP) {
1291 		args.pa_flags &= ~FLAG_DO_PRINT;
1292 
1293 		if (args.pa_toptty) {
1294 			/*
1295 			 * Init ncurses.
1296 			 */
1297 			initscr();
1298 			if(has_colors() == TRUE) {
1299 				args.pa_topcolor = 1;
1300 				start_color();
1301 				use_default_colors();
1302 				pair_content(0, &cf, &cb);
1303 				init_pair(1, COLOR_RED, cb);
1304 				init_pair(2, COLOR_YELLOW, cb);
1305 				init_pair(3, COLOR_GREEN, cb);
1306 			}
1307 			cbreak();
1308 			noecho();
1309 			nonl();
1310 			nodelay(stdscr, 1);
1311 			intrflush(stdscr, FALSE);
1312 			keypad(stdscr, TRUE);
1313 			clear();
1314 			atexit(pmcstat_topexit);
1315 		}
1316 	}
1317 
1318 	/*
1319 	 * loop till either the target process (if any) exits, or we
1320 	 * are killed by a SIGINT.
1321 	 */
1322 	runstate = PMCSTAT_RUNNING;
1323 	do_print = 0;
1324 	do {
1325 		if ((c = kevent(pmcstat_kq, NULL, 0, &kev, 1, NULL)) <= 0) {
1326 			if (errno != EINTR)
1327 				err(EX_OSERR, "ERROR: kevent failed");
1328 			else
1329 				continue;
1330 		}
1331 
1332 		if (kev.flags & EV_ERROR)
1333 			errc(EX_OSERR, kev.data, "ERROR: kevent failed");
1334 
1335 		switch (kev.filter) {
1336 		case EVFILT_PROC:  /* target has exited */
1337 			if (args.pa_flags & (FLAG_HAS_OUTPUT_LOGFILE |
1338 				FLAG_HAS_PIPE))
1339 				runstate = pmcstat_close_log();
1340 			else
1341 				runstate = PMCSTAT_FINISHED;
1342 			do_print = 1;
1343 			break;
1344 
1345 		case EVFILT_READ:  /* log file data is present */
1346 			if (kev.ident == (unsigned)fileno(stdin)) {
1347 				if (pmcstat_keypress_log())
1348 					runstate = pmcstat_close_log();
1349 			} else
1350 				runstate = pmcstat_process_log();
1351 			break;
1352 
1353 		case EVFILT_SIGNAL:
1354 			if (kev.ident == SIGCHLD) {
1355 				/*
1356 				 * The child process sends us a
1357 				 * SIGCHLD if its exec() failed.  We
1358 				 * wait for it to exit and then exit
1359 				 * ourselves.
1360 				 */
1361 				(void) wait(&c);
1362 				runstate = PMCSTAT_FINISHED;
1363 			} else if (kev.ident == SIGIO) {
1364 				/*
1365 				 * We get a SIGIO if a PMC loses all
1366 				 * of its targets, or if logfile
1367 				 * writes encounter an error.
1368 				 */
1369 				if (args.pa_flags & (FLAG_HAS_OUTPUT_LOGFILE |
1370 				    FLAG_HAS_PIPE)) {
1371 					runstate = pmcstat_close_log();
1372 					if (args.pa_flags &
1373 					    (FLAG_DO_PRINT|FLAG_DO_ANALYSIS))
1374 						pmcstat_process_log();
1375 				}
1376 				do_print = 1; /* print PMCs at exit */
1377 				runstate = PMCSTAT_FINISHED;
1378 			} else if (kev.ident == SIGINT) {
1379 				/* Kill the child process if we started it */
1380 				if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1381 					pmcstat_kill_process();
1382 				/* Close the pipe to self, if present. */
1383 				if (args.pa_flags & FLAG_HAS_PIPE)
1384 					(void) close(pipefd[READPIPEFD]);
1385 				runstate = PMCSTAT_FINISHED;
1386 			} else if (kev.ident == SIGWINCH) {
1387 				if (ioctl(fileno(args.pa_printfile),
1388 					TIOCGWINSZ, &ws) < 0)
1389 				    err(EX_OSERR, "ERROR: Cannot determine "
1390 					"window size");
1391 				pmcstat_displayheight = ws.ws_row - 1;
1392 				pmcstat_displaywidth  = ws.ws_col - 1;
1393 			} else
1394 				assert(0);
1395 
1396 			break;
1397 
1398 		case EVFILT_TIMER: /* print out counting PMCs */
1399 			do_print = 1;
1400 			break;
1401 
1402 		}
1403 
1404 		if (do_print) {
1405 			if ((args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) {
1406 				pmcstat_print_pmcs();
1407 				if (runstate == PMCSTAT_FINISHED && /* final newline */
1408 				    (args.pa_flags & FLAG_DO_PRINT) == 0)
1409 					(void) fprintf(args.pa_printfile, "\n");
1410 			}
1411 			if (args.pa_flags & FLAG_DO_TOP)
1412 				pmcstat_display_log();
1413 			do_print = 0;
1414 		}
1415 
1416 	} while (runstate != PMCSTAT_FINISHED);
1417 
1418 	if ((args.pa_flags & FLAG_DO_TOP) && args.pa_toptty) {
1419 		pmcstat_topexit();
1420 		args.pa_toptty = 0;
1421 	}
1422 
1423 	/* flush any pending log entries */
1424 	if (args.pa_flags & (FLAG_HAS_OUTPUT_LOGFILE | FLAG_HAS_PIPE))
1425 		pmc_flush_logfile();
1426 
1427 	pmcstat_cleanup();
1428 
1429 	free(args.pa_kernel);
1430 
1431 	/* check if the driver lost any samples or events */
1432 	if (check_driver_stats) {
1433 		if (pmc_get_driver_stats(&ds_end) < 0)
1434 			err(EX_OSERR, "ERROR: Cannot retrieve driver "
1435 			    "statistics");
1436 		if (ds_start.pm_intr_bufferfull != ds_end.pm_intr_bufferfull &&
1437 		    args.pa_verbosity > 0)
1438 			warnx("WARNING: some samples were dropped.  Please "
1439 			    "consider tuning the \"kern.hwpmc.nsamples\" "
1440 			    "tunable.");
1441 		if (ds_start.pm_buffer_requests_failed !=
1442 		    ds_end.pm_buffer_requests_failed &&
1443 		    args.pa_verbosity > 0)
1444 			warnx("WARNING: some events were discarded.  Please "
1445 			    "consider tuning the \"kern.hwpmc.nbuffers\" "
1446 			    "tunable.");
1447 	}
1448 
1449 	exit(EX_OK);
1450 }
1451