xref: /freebsd/usr.sbin/pmcstat/pmcstat.c (revision 389e4940069316fe667ffa263fa7d6390d0a960f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2003-2008, Joseph Koshy
5  * Copyright (c) 2007 The FreeBSD Foundation
6  * All rights reserved.
7  *
8  * Portions of this software were developed by A. Joseph Koshy under
9  * sponsorship from the FreeBSD Foundation and Google, Inc.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include <sys/param.h>
37 #include <sys/cpuset.h>
38 #include <sys/event.h>
39 #include <sys/queue.h>
40 #include <sys/socket.h>
41 #include <sys/stat.h>
42 #include <sys/sysctl.h>
43 #include <sys/time.h>
44 #include <sys/ttycom.h>
45 #include <sys/user.h>
46 #include <sys/wait.h>
47 
48 #include <assert.h>
49 #include <curses.h>
50 #include <err.h>
51 #include <errno.h>
52 #include <fcntl.h>
53 #include <kvm.h>
54 #include <libgen.h>
55 #include <limits.h>
56 #include <math.h>
57 #include <pmc.h>
58 #include <pmclog.h>
59 #include <regex.h>
60 #include <signal.h>
61 #include <stdarg.h>
62 #include <stdint.h>
63 #include <stdio.h>
64 #include <stdlib.h>
65 #include <string.h>
66 #include <sysexits.h>
67 #include <unistd.h>
68 
69 #include <libpmcstat.h>
70 
71 #include "pmcstat.h"
72 
73 /*
74  * A given invocation of pmcstat(8) can manage multiple PMCs of both
75  * the system-wide and per-process variety.  Each of these could be in
76  * 'counting mode' or in 'sampling mode'.
77  *
78  * For 'counting mode' PMCs, pmcstat(8) will periodically issue a
79  * pmc_read() at the configured time interval and print out the value
80  * of the requested PMCs.
81  *
82  * For 'sampling mode' PMCs it can log to a file for offline analysis,
83  * or can analyse sampling data "on the fly", either by converting
84  * samples to printed textual form or by creating gprof(1) compatible
85  * profiles, one per program executed.  When creating gprof(1)
86  * profiles it can optionally merge entries from multiple processes
87  * for a given executable into a single profile file.
88  *
89  * pmcstat(8) can also execute a command line and attach PMCs to the
90  * resulting child process.  The protocol used is as follows:
91  *
92  * - parent creates a socketpair for two way communication and
93  *   fork()s.
94  * - subsequently:
95  *
96  *   /Parent/				/Child/
97  *
98  *   - Wait for childs token.
99  *					- Sends token.
100  *					- Awaits signal to start.
101  *  - Attaches PMCs to the child's pid
102  *    and starts them. Sets up
103  *    monitoring for the child.
104  *  - Signals child to start.
105  *					- Receives signal, attempts exec().
106  *
107  * After this point normal processing can happen.
108  */
109 
110 /* Globals */
111 
112 int		pmcstat_displayheight = DEFAULT_DISPLAY_HEIGHT;
113 int		pmcstat_displaywidth  = DEFAULT_DISPLAY_WIDTH;
114 static int	pmcstat_sockpair[NSOCKPAIRFD];
115 static int	pmcstat_kq;
116 static kvm_t	*pmcstat_kvm;
117 static struct kinfo_proc *pmcstat_plist;
118 struct pmcstat_args args;
119 
120 static void
121 pmcstat_get_cpumask(const char *cpuspec, cpuset_t *cpumask)
122 {
123 	int cpu;
124 	const char *s;
125 	char *end;
126 
127 	CPU_ZERO(cpumask);
128 	s = cpuspec;
129 
130 	do {
131 		cpu = strtol(s, &end, 0);
132 		if (cpu < 0 || end == s)
133 			errx(EX_USAGE,
134 			    "ERROR: Illegal CPU specification \"%s\".",
135 			    cpuspec);
136 		CPU_SET(cpu, cpumask);
137 		s = end + strspn(end, ", \t");
138 	} while (*s);
139 	assert(!CPU_EMPTY(cpumask));
140 }
141 
142 void
143 pmcstat_cleanup(void)
144 {
145 	struct pmcstat_ev *ev;
146 
147 	/* release allocated PMCs. */
148 	STAILQ_FOREACH(ev, &args.pa_events, ev_next)
149 		if (ev->ev_pmcid != PMC_ID_INVALID) {
150 			if (pmc_stop(ev->ev_pmcid) < 0)
151 				err(EX_OSERR,
152 				    "ERROR: cannot stop pmc 0x%x \"%s\"",
153 				    ev->ev_pmcid, ev->ev_name);
154 			if (pmc_release(ev->ev_pmcid) < 0)
155 				err(EX_OSERR,
156 				    "ERROR: cannot release pmc 0x%x \"%s\"",
157 				    ev->ev_pmcid, ev->ev_name);
158 		}
159 
160 	/* de-configure the log file if present. */
161 	if (args.pa_flags & (FLAG_HAS_PIPE | FLAG_HAS_OUTPUT_LOGFILE))
162 		(void) pmc_configure_logfile(-1);
163 
164 	if (args.pa_logparser) {
165 		pmclog_close(args.pa_logparser);
166 		args.pa_logparser = NULL;
167 	}
168 
169 	pmcstat_log_shutdown_logging();
170 }
171 
172 void
173 pmcstat_find_targets(const char *spec)
174 {
175 	int n, nproc, pid, rv;
176 	struct pmcstat_target *pt;
177 	char errbuf[_POSIX2_LINE_MAX], *end;
178 	static struct kinfo_proc *kp;
179 	regex_t reg;
180 	regmatch_t regmatch;
181 
182 	/* First check if we've been given a process id. */
183       	pid = strtol(spec, &end, 0);
184 	if (end != spec && pid >= 0) {
185 		if ((pt = malloc(sizeof(*pt))) == NULL)
186 			goto outofmemory;
187 		pt->pt_pid = pid;
188 		SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
189 		return;
190 	}
191 
192 	/* Otherwise treat arg as a regular expression naming processes. */
193 	if (pmcstat_kvm == NULL) {
194 		if ((pmcstat_kvm = kvm_openfiles(NULL, "/dev/null", NULL, 0,
195 		    errbuf)) == NULL)
196 			err(EX_OSERR, "ERROR: Cannot open kernel \"%s\"",
197 			    errbuf);
198 		if ((pmcstat_plist = kvm_getprocs(pmcstat_kvm, KERN_PROC_PROC,
199 		    0, &nproc)) == NULL)
200 			err(EX_OSERR, "ERROR: Cannot get process list: %s",
201 			    kvm_geterr(pmcstat_kvm));
202 	} else
203 		nproc = 0;
204 
205 	if ((rv = regcomp(&reg, spec, REG_EXTENDED|REG_NOSUB)) != 0) {
206 		regerror(rv, &reg, errbuf, sizeof(errbuf));
207 		err(EX_DATAERR, "ERROR: Failed to compile regex \"%s\": %s",
208 		    spec, errbuf);
209 	}
210 
211 	for (n = 0, kp = pmcstat_plist; n < nproc; n++, kp++) {
212 		if ((rv = regexec(&reg, kp->ki_comm, 1, &regmatch, 0)) == 0) {
213 			if ((pt = malloc(sizeof(*pt))) == NULL)
214 				goto outofmemory;
215 			pt->pt_pid = kp->ki_pid;
216 			SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
217 		} else if (rv != REG_NOMATCH) {
218 			regerror(rv, &reg, errbuf, sizeof(errbuf));
219 			errx(EX_SOFTWARE, "ERROR: Regex evalation failed: %s",
220 			    errbuf);
221 		}
222 	}
223 
224 	regfree(&reg);
225 
226 	return;
227 
228  outofmemory:
229 	errx(EX_SOFTWARE, "Out of memory.");
230 	/*NOTREACHED*/
231 }
232 
233 void
234 pmcstat_kill_process(void)
235 {
236 	struct pmcstat_target *pt;
237 
238 	assert(args.pa_flags & FLAG_HAS_COMMANDLINE);
239 
240 	/*
241 	 * If a command line was specified, it would be the very first
242 	 * in the list, before any other processes specified by -t.
243 	 */
244 	pt = SLIST_FIRST(&args.pa_targets);
245 	assert(pt != NULL);
246 
247 	if (kill(pt->pt_pid, SIGINT) != 0)
248 		err(EX_OSERR, "ERROR: cannot signal child process");
249 }
250 
251 void
252 pmcstat_start_pmcs(void)
253 {
254 	struct pmcstat_ev *ev;
255 
256 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
257 
258 	    assert(ev->ev_pmcid != PMC_ID_INVALID);
259 
260 	    if (pmc_start(ev->ev_pmcid) < 0) {
261 	        warn("ERROR: Cannot start pmc 0x%x \"%s\"",
262 		    ev->ev_pmcid, ev->ev_name);
263 		pmcstat_cleanup();
264 		exit(EX_OSERR);
265 	    }
266 	}
267 }
268 
269 void
270 pmcstat_print_headers(void)
271 {
272 	struct pmcstat_ev *ev;
273 	int c, w;
274 
275 	(void) fprintf(args.pa_printfile, PRINT_HEADER_PREFIX);
276 
277 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
278 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
279 			continue;
280 
281 		c = PMC_IS_SYSTEM_MODE(ev->ev_mode) ? 's' : 'p';
282 
283 		if (ev->ev_fieldskip != 0)
284 			(void) fprintf(args.pa_printfile, "%*s",
285 			    ev->ev_fieldskip, "");
286 		w = ev->ev_fieldwidth - ev->ev_fieldskip - 2;
287 
288 		if (c == 's')
289 			(void) fprintf(args.pa_printfile, "s/%02d/%-*s ",
290 			    ev->ev_cpu, w-3, ev->ev_name);
291 		else
292 			(void) fprintf(args.pa_printfile, "p/%*s ", w,
293 			    ev->ev_name);
294 	}
295 
296 	(void) fflush(args.pa_printfile);
297 }
298 
299 void
300 pmcstat_print_counters(void)
301 {
302 	int extra_width;
303 	struct pmcstat_ev *ev;
304 	pmc_value_t value;
305 
306 	extra_width = sizeof(PRINT_HEADER_PREFIX) - 1;
307 
308 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
309 
310 		/* skip sampling mode counters */
311 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
312 			continue;
313 
314 		if (pmc_read(ev->ev_pmcid, &value) < 0)
315 			err(EX_OSERR, "ERROR: Cannot read pmc \"%s\"",
316 			    ev->ev_name);
317 
318 		(void) fprintf(args.pa_printfile, "%*ju ",
319 		    ev->ev_fieldwidth + extra_width,
320 		    (uintmax_t) ev->ev_cumulative ? value :
321 		    (value - ev->ev_saved));
322 
323 		if (ev->ev_cumulative == 0)
324 			ev->ev_saved = value;
325 		extra_width = 0;
326 	}
327 
328 	(void) fflush(args.pa_printfile);
329 }
330 
331 /*
332  * Print output
333  */
334 
335 void
336 pmcstat_print_pmcs(void)
337 {
338 	static int linecount = 0;
339 
340 	/* check if we need to print a header line */
341 	if (++linecount > pmcstat_displayheight) {
342 		(void) fprintf(args.pa_printfile, "\n");
343 		linecount = 1;
344 	}
345 	if (linecount == 1)
346 		pmcstat_print_headers();
347 	(void) fprintf(args.pa_printfile, "\n");
348 
349 	pmcstat_print_counters();
350 
351 	return;
352 }
353 
354 void
355 pmcstat_show_usage(void)
356 {
357 	errx(EX_USAGE,
358 	    "[options] [commandline]\n"
359 	    "\t Measure process and/or system performance using hardware\n"
360 	    "\t performance monitoring counters.\n"
361 	    "\t Options include:\n"
362 	    "\t -C\t\t (toggle) show cumulative counts\n"
363 	    "\t -D path\t create profiles in directory \"path\"\n"
364 	    "\t -E\t\t (toggle) show counts at process exit\n"
365 	    "\t -F file\t write a system-wide callgraph (Kcachegrind format)"
366 		" to \"file\"\n"
367 	    "\t -G file\t write a system-wide callgraph to \"file\"\n"
368 	    "\t -I\t\t don't resolve leaf function name, show address instead\n"
369 	    "\t -L\t\t list all counters available on this host\n"
370 	    "\t -M file\t print executable/gmon file map to \"file\"\n"
371 	    "\t -N\t\t (toggle) capture callchains\n"
372 	    "\t -O file\t send log output to \"file\"\n"
373 	    "\t -P spec\t allocate a process-private sampling PMC\n"
374 	    "\t -R file\t read events from \"file\"\n"
375 	    "\t -S spec\t allocate a system-wide sampling PMC\n"
376 	    "\t -T\t\t start in top mode\n"
377 	    "\t -U \t\n merged user kernel stack capture\n"
378 	    "\t -W\t\t (toggle) show counts per context switch\n"
379 	    "\t -a file\t print sampled PCs and callgraph to \"file\"\n"
380 	    "\t -c cpu-list\t set cpus for subsequent system-wide PMCs\n"
381 	    "\t -d\t\t (toggle) track descendants\n"
382 	    "\t -e\t\t use wide history counter for gprof(1) output\n"
383 	    "\t -f spec\t pass \"spec\" to as plugin option\n"
384 	    "\t -g\t\t produce gprof(1) compatible profiles\n"
385 	    "\t -i lwp\t\t filter on thread id \"lwp\" in post-processing\n"
386 	    "\t -k dir\t\t set the path to the kernel\n"
387 	    "\t -l secs\t set duration time\n"
388 	    "\t -m file\t print sampled PCs to \"file\"\n"
389 	    "\t -n rate\t set sampling rate\n"
390 	    "\t -o file\t send print output to \"file\"\n"
391 	    "\t -p spec\t allocate a process-private counting PMC\n"
392 	    "\t -q\t\t suppress verbosity\n"
393 	    "\t -r fsroot\t specify FS root directory\n"
394 	    "\t -s spec\t allocate a system-wide counting PMC\n"
395 	    "\t -t process-spec attach to running processes matching "
396 		"\"process-spec\"\n"
397 	    "\t -u spec \t provide short description of counters matching spec\n"
398 	    "\t -v\t\t increase verbosity\n"
399 	    "\t -w secs\t set printing time interval\n"
400 	    "\t -z depth\t limit callchain display depth"
401 	);
402 }
403 
404 /*
405  * At exit handler for top mode
406  */
407 
408 void
409 pmcstat_topexit(void)
410 {
411 	if (!args.pa_toptty)
412 		return;
413 
414 	/*
415 	 * Shutdown ncurses.
416 	 */
417 	clrtoeol();
418 	refresh();
419 	endwin();
420 }
421 
422 /*
423  * Main
424  */
425 
426 int
427 main(int argc, char **argv)
428 {
429 	cpuset_t cpumask, rootmask;
430 	double interval;
431 	double duration;
432 	int option, npmc;
433 	int c, check_driver_stats;
434 	int do_callchain, do_descendants, do_logproccsw, do_logprocexit;
435 	int do_print, do_read, do_listcounters, do_descr;
436 	int do_userspace;
437 	size_t len;
438 	int graphdepth;
439 	int pipefd[2], rfd;
440 	int use_cumulative_counts;
441 	short cf, cb;
442 	uint64_t current_sampling_count;
443 	char *end, *tmp, *event;
444 	const char *errmsg, *graphfilename;
445 	enum pmcstat_state runstate;
446 	struct pmc_driverstats ds_start, ds_end;
447 	struct pmcstat_ev *ev;
448 	struct sigaction sa;
449 	struct kevent kev;
450 	struct winsize ws;
451 	struct stat sb;
452 	char buffer[PATH_MAX];
453 
454 	check_driver_stats      = 0;
455 	current_sampling_count  = 0;
456 	do_callchain		= 1;
457 	do_descr                = 0;
458 	do_descendants          = 0;
459 	do_userspace            = 0;
460 	do_logproccsw           = 0;
461 	do_logprocexit          = 0;
462 	do_listcounters         = 0;
463 	use_cumulative_counts   = 0;
464 	graphfilename		= "-";
465 	args.pa_required	= 0;
466 	args.pa_flags		= 0;
467 	args.pa_verbosity	= 1;
468 	args.pa_logfd		= -1;
469 	args.pa_fsroot		= "";
470 	args.pa_samplesdir	= ".";
471 	args.pa_printfile	= stderr;
472 	args.pa_graphdepth	= DEFAULT_CALLGRAPH_DEPTH;
473 	args.pa_graphfile	= NULL;
474 	args.pa_interval	= DEFAULT_WAIT_INTERVAL;
475 	args.pa_mapfilename	= NULL;
476 	args.pa_inputpath	= NULL;
477 	args.pa_outputpath	= NULL;
478 	args.pa_pplugin		= PMCSTAT_PL_NONE;
479 	args.pa_plugin		= PMCSTAT_PL_NONE;
480 	args.pa_ctdumpinstr	= 1;
481 	args.pa_topmode		= PMCSTAT_TOP_DELTA;
482 	args.pa_toptty		= 0;
483 	args.pa_topcolor	= 0;
484 	args.pa_mergepmc	= 0;
485 	args.pa_duration	= 0.0;
486 	STAILQ_INIT(&args.pa_events);
487 	SLIST_INIT(&args.pa_targets);
488 	bzero(&ds_start, sizeof(ds_start));
489 	bzero(&ds_end, sizeof(ds_end));
490 	ev = NULL;
491 	event = NULL;
492 	CPU_ZERO(&cpumask);
493 
494 	/* Default to using the running system kernel. */
495 	len = 0;
496 	if (sysctlbyname("kern.bootfile", NULL, &len, NULL, 0) == -1)
497 		err(EX_OSERR, "ERROR: Cannot determine path of running kernel");
498 	args.pa_kernel = malloc(len);
499 	if (args.pa_kernel == NULL)
500 		errx(EX_SOFTWARE, "ERROR: Out of memory.");
501 	if (sysctlbyname("kern.bootfile", args.pa_kernel, &len, NULL, 0) == -1)
502 		err(EX_OSERR, "ERROR: Cannot determine path of running kernel");
503 
504 	/*
505 	 * The initial CPU mask specifies the root mask of this process
506 	 * which is usually all CPUs in the system.
507 	 */
508 	if (cpuset_getaffinity(CPU_LEVEL_ROOT, CPU_WHICH_PID, -1,
509 	    sizeof(rootmask), &rootmask) == -1)
510 		err(EX_OSERR, "ERROR: Cannot determine the root set of CPUs");
511 	CPU_COPY(&rootmask, &cpumask);
512 
513 	while ((option = getopt(argc, argv,
514 	    "CD:EF:G:ILM:NO:P:R:S:TUWZa:c:def:gi:k:l:m:n:o:p:qr:s:t:u:vw:z:")) != -1)
515 		switch (option) {
516 		case 'a':	/* Annotate + callgraph */
517 			args.pa_flags |= FLAG_DO_ANNOTATE;
518 			args.pa_plugin = PMCSTAT_PL_ANNOTATE_CG;
519 			graphfilename  = optarg;
520 			break;
521 
522 		case 'C':	/* cumulative values */
523 			use_cumulative_counts = !use_cumulative_counts;
524 			args.pa_required |= FLAG_HAS_COUNTING_PMCS;
525 			break;
526 
527 		case 'c':	/* CPU */
528 			if (optarg[0] == '*' && optarg[1] == '\0')
529 				CPU_COPY(&rootmask, &cpumask);
530 			else
531 				pmcstat_get_cpumask(optarg, &cpumask);
532 
533 			args.pa_flags	 |= FLAGS_HAS_CPUMASK;
534 			args.pa_required |= FLAG_HAS_SYSTEM_PMCS;
535 			break;
536 
537 		case 'D':
538 			if (stat(optarg, &sb) < 0)
539 				err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
540 				    optarg);
541 			if (!S_ISDIR(sb.st_mode))
542 				errx(EX_USAGE,
543 				    "ERROR: \"%s\" is not a directory.",
544 				    optarg);
545 			args.pa_samplesdir = optarg;
546 			args.pa_flags     |= FLAG_HAS_SAMPLESDIR;
547 			args.pa_required  |= FLAG_DO_GPROF;
548 			break;
549 
550 		case 'd':	/* toggle descendents */
551 			do_descendants = !do_descendants;
552 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
553 			break;
554 
555 		case 'E':	/* log process exit */
556 			do_logprocexit = !do_logprocexit;
557 			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
558 			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
559 			break;
560 
561 		case 'e':	/* wide gprof metrics */
562 			args.pa_flags |= FLAG_DO_WIDE_GPROF_HC;
563 			break;
564 
565 		case 'F':	/* produce a system-wide calltree */
566 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
567 			args.pa_plugin = PMCSTAT_PL_CALLTREE;
568 			graphfilename = optarg;
569 			break;
570 
571 		case 'f':	/* plugins options */
572 			if (args.pa_plugin == PMCSTAT_PL_NONE)
573 				err(EX_USAGE, "ERROR: Need -g/-G/-m/-T.");
574 			pmcstat_pluginconfigure_log(optarg);
575 			break;
576 
577 		case 'G':	/* produce a system-wide callgraph */
578 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
579 			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
580 			graphfilename = optarg;
581 			break;
582 
583 		case 'g':	/* produce gprof compatible profiles */
584 			args.pa_flags |= FLAG_DO_GPROF;
585 			args.pa_pplugin = PMCSTAT_PL_CALLGRAPH;
586 			args.pa_plugin	= PMCSTAT_PL_GPROF;
587 			break;
588 
589 		case 'I':
590 			args.pa_flags |= FLAG_SKIP_TOP_FN_RES;
591 			break;
592 		case 'i':
593 			args.pa_flags |= FLAG_FILTER_THREAD_ID;
594 			args.pa_tid = strtol(optarg, &end, 0);
595 			break;
596 
597 		case 'k':	/* pathname to the kernel */
598 			free(args.pa_kernel);
599 			args.pa_kernel = strdup(optarg);
600 			if (args.pa_kernel == NULL)
601 				errx(EX_SOFTWARE, "ERROR: Out of memory");
602 			args.pa_required |= FLAG_DO_ANALYSIS;
603 			args.pa_flags    |= FLAG_HAS_KERNELPATH;
604 			break;
605 
606 		case 'L':
607 			do_listcounters = 1;
608 			break;
609 
610 		case 'l':	/* time duration in seconds */
611 			duration = strtod(optarg, &end);
612 			if (*end != '\0' || duration <= 0)
613 				errx(EX_USAGE, "ERROR: Illegal duration time "
614 				    "value \"%s\".", optarg);
615 			args.pa_flags |= FLAG_HAS_DURATION;
616 			args.pa_duration = duration;
617 			break;
618 
619 		case 'm':
620 			args.pa_flags |= FLAG_DO_ANNOTATE;
621 			args.pa_plugin = PMCSTAT_PL_ANNOTATE;
622 			graphfilename  = optarg;
623 			break;
624 
625 		case 'M':	/* mapfile */
626 			args.pa_mapfilename = optarg;
627 			break;
628 
629 		case 'N':
630 			do_callchain = !do_callchain;
631 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
632 			break;
633 
634 		case 'p':	/* process virtual counting PMC */
635 		case 's':	/* system-wide counting PMC */
636 		case 'P':	/* process virtual sampling PMC */
637 		case 'S':	/* system-wide sampling PMC */
638 			if ((ev = malloc(sizeof(*ev))) == NULL)
639 				errx(EX_SOFTWARE, "ERROR: Out of memory.");
640 
641 			switch (option) {
642 			case 'p': ev->ev_mode = PMC_MODE_TC; break;
643 			case 's': ev->ev_mode = PMC_MODE_SC; break;
644 			case 'P': ev->ev_mode = PMC_MODE_TS; break;
645 			case 'S': ev->ev_mode = PMC_MODE_SS; break;
646 			}
647 
648 			if (option == 'P' || option == 'p') {
649 				args.pa_flags |= FLAG_HAS_PROCESS_PMCS;
650 				args.pa_required |= (FLAG_HAS_COMMANDLINE |
651 				    FLAG_HAS_TARGET);
652 			}
653 
654 			if (option == 'P' || option == 'S') {
655 				args.pa_flags |= FLAG_HAS_SAMPLING_PMCS;
656 				args.pa_required |= (FLAG_HAS_PIPE |
657 				    FLAG_HAS_OUTPUT_LOGFILE);
658 			}
659 
660 			if (option == 'p' || option == 's')
661 				args.pa_flags |= FLAG_HAS_COUNTING_PMCS;
662 
663 			if (option == 's' || option == 'S')
664 				args.pa_flags |= FLAG_HAS_SYSTEM_PMCS;
665 
666 			ev->ev_spec = strdup(optarg);
667 			if (ev->ev_spec == NULL)
668 				errx(EX_SOFTWARE, "ERROR: Out of memory.");
669 
670 			if (option == 'S' || option == 'P')
671 				ev->ev_count = current_sampling_count ? current_sampling_count : pmc_pmu_sample_rate_get(ev->ev_spec);
672 			else
673 				ev->ev_count = -1;
674 
675 			if (option == 'S' || option == 's')
676 				ev->ev_cpu = CPU_FFS(&cpumask) - 1;
677 			else
678 				ev->ev_cpu = PMC_CPU_ANY;
679 
680 			ev->ev_flags = 0;
681 			if (do_callchain) {
682 				ev->ev_flags |= PMC_F_CALLCHAIN;
683 				if (do_userspace)
684 					ev->ev_flags |= PMC_F_USERCALLCHAIN;
685 			}
686 			if (do_descendants)
687 				ev->ev_flags |= PMC_F_DESCENDANTS;
688 			if (do_logprocexit)
689 				ev->ev_flags |= PMC_F_LOG_PROCEXIT;
690 			if (do_logproccsw)
691 				ev->ev_flags |= PMC_F_LOG_PROCCSW;
692 
693 			ev->ev_cumulative  = use_cumulative_counts;
694 
695 			ev->ev_saved = 0LL;
696 			ev->ev_pmcid = PMC_ID_INVALID;
697 
698 			/* extract event name */
699 			c = strcspn(optarg, ", \t");
700 			ev->ev_name = malloc(c + 1);
701 			if (ev->ev_name == NULL)
702 				errx(EX_SOFTWARE, "ERROR: Out of memory.");
703 			(void) strncpy(ev->ev_name, optarg, c);
704 			*(ev->ev_name + c) = '\0';
705 
706 			STAILQ_INSERT_TAIL(&args.pa_events, ev, ev_next);
707 
708 			if (option == 's' || option == 'S') {
709 				CPU_CLR(ev->ev_cpu, &cpumask);
710 				pmcstat_clone_event_descriptor(ev, &cpumask, &args);
711 				CPU_SET(ev->ev_cpu, &cpumask);
712 			}
713 
714 			break;
715 
716 		case 'n':	/* sampling count */
717 			current_sampling_count = strtol(optarg, &end, 0);
718 			if (*end != '\0' || current_sampling_count <= 0)
719 				errx(EX_USAGE,
720 				    "ERROR: Illegal count value \"%s\".",
721 				    optarg);
722 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
723 			break;
724 
725 		case 'o':	/* outputfile */
726 			if (args.pa_printfile != NULL &&
727 			    args.pa_printfile != stdout &&
728 			    args.pa_printfile != stderr)
729 				(void) fclose(args.pa_printfile);
730 			if ((args.pa_printfile = fopen(optarg, "w")) == NULL)
731 				errx(EX_OSERR,
732 				    "ERROR: cannot open \"%s\" for writing.",
733 				    optarg);
734 			args.pa_flags |= FLAG_DO_PRINT;
735 			break;
736 
737 		case 'O':	/* sampling output */
738 			if (args.pa_outputpath)
739 				errx(EX_USAGE,
740 "ERROR: option -O may only be specified once.");
741 			args.pa_outputpath = optarg;
742 			args.pa_flags |= FLAG_HAS_OUTPUT_LOGFILE;
743 			break;
744 
745 		case 'q':	/* quiet mode */
746 			args.pa_verbosity = 0;
747 			break;
748 
749 		case 'r':	/* root FS path */
750 			args.pa_fsroot = optarg;
751 			break;
752 
753 		case 'R':	/* read an existing log file */
754 			if (args.pa_inputpath != NULL)
755 				errx(EX_USAGE,
756 "ERROR: option -R may only be specified once.");
757 			args.pa_inputpath = optarg;
758 			if (args.pa_printfile == stderr)
759 				args.pa_printfile = stdout;
760 			args.pa_flags |= FLAG_READ_LOGFILE;
761 			break;
762 
763 		case 't':	/* target pid or process name */
764 			pmcstat_find_targets(optarg);
765 
766 			args.pa_flags |= FLAG_HAS_TARGET;
767 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
768 			break;
769 
770 		case 'T':	/* top mode */
771 			args.pa_flags |= FLAG_DO_TOP;
772 			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
773 			args.pa_ctdumpinstr = 0;
774 			args.pa_mergepmc = 1;
775 			if (args.pa_printfile == stderr)
776 				args.pa_printfile = stdout;
777 			break;
778 
779 		case 'u':
780 			do_descr = 1;
781 			event = optarg;
782 			break;
783 		case 'U':	/* toggle user-space callchain capture */
784 			do_userspace = !do_userspace;
785 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
786 			break;
787 		case 'v':	/* verbose */
788 			args.pa_verbosity++;
789 			break;
790 
791 		case 'w':	/* wait interval */
792 			interval = strtod(optarg, &end);
793 			if (*end != '\0' || interval <= 0)
794 				errx(EX_USAGE,
795 "ERROR: Illegal wait interval value \"%s\".",
796 				    optarg);
797 			args.pa_flags |= FLAG_HAS_WAIT_INTERVAL;
798 			args.pa_interval = interval;
799 			break;
800 
801 		case 'W':	/* toggle LOG_CSW */
802 			do_logproccsw = !do_logproccsw;
803 			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
804 			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
805 			break;
806 
807 		case 'z':
808 			graphdepth = strtod(optarg, &end);
809 			if (*end != '\0' || graphdepth <= 0)
810 				errx(EX_USAGE,
811 				    "ERROR: Illegal callchain depth \"%s\".",
812 				    optarg);
813 			args.pa_graphdepth = graphdepth;
814 			args.pa_required |= FLAG_DO_CALLGRAPHS;
815 			break;
816 
817 		case '?':
818 		default:
819 			pmcstat_show_usage();
820 			break;
821 
822 		}
823 	if ((do_listcounters | do_descr) &&
824 		pmc_pmu_enabled() == 0)
825 			errx(EX_USAGE, "pmu features not supported on host or hwpmc not loaded");
826 	if (do_listcounters) {
827 		pmc_pmu_print_counters(NULL);
828 	} else if (do_descr) {
829 		pmc_pmu_print_counter_desc(event);
830 	}
831 	if (do_listcounters | do_descr)
832 		exit(0);
833 
834 	args.pa_argc = (argc -= optind);
835 	args.pa_argv = (argv += optind);
836 
837 	/* If we read from logfile and no specified CPU mask use
838 	 * the maximum CPU count.
839 	 */
840 	if ((args.pa_flags & FLAG_READ_LOGFILE) &&
841 	    (args.pa_flags & FLAGS_HAS_CPUMASK) == 0)
842 		CPU_FILL(&cpumask);
843 
844 	args.pa_cpumask = cpumask; /* For selecting CPUs using -R. */
845 
846 	if (argc)	/* command line present */
847 		args.pa_flags |= FLAG_HAS_COMMANDLINE;
848 
849 	if (args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS |
850 	    FLAG_DO_ANNOTATE | FLAG_DO_TOP))
851 		args.pa_flags |= FLAG_DO_ANALYSIS;
852 
853 	/*
854 	 * Check invocation syntax.
855 	 */
856 
857 	/* disallow -O and -R together */
858 	if (args.pa_outputpath && args.pa_inputpath)
859 		errx(EX_USAGE,
860 		    "ERROR: options -O and -R are mutually exclusive.");
861 
862 	/* disallow -T and -l together */
863 	if ((args.pa_flags & FLAG_HAS_DURATION) &&
864 	    (args.pa_flags & FLAG_DO_TOP))
865 		errx(EX_USAGE, "ERROR: options -T and -l are mutually "
866 		    "exclusive.");
867 
868 	/* -a and -m require -R */
869 	if (args.pa_flags & FLAG_DO_ANNOTATE && args.pa_inputpath == NULL)
870 		errx(EX_USAGE, "ERROR: option %s requires an input file",
871 		    args.pa_plugin == PMCSTAT_PL_ANNOTATE ? "-m" : "-a");
872 
873 	/* -m option is not allowed combined with -g or -G. */
874 	if (args.pa_flags & FLAG_DO_ANNOTATE &&
875 	    args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS))
876 		errx(EX_USAGE,
877 		    "ERROR: option -m and -g | -G are mutually exclusive");
878 
879 	if (args.pa_flags & FLAG_READ_LOGFILE) {
880 		errmsg = NULL;
881 		if (args.pa_flags & FLAG_HAS_COMMANDLINE)
882 			errmsg = "a command line specification";
883 		else if (args.pa_flags & FLAG_HAS_TARGET)
884 			errmsg = "option -t";
885 		else if (!STAILQ_EMPTY(&args.pa_events))
886 			errmsg = "a PMC event specification";
887 		if (errmsg)
888 			errx(EX_USAGE,
889 			    "ERROR: option -R may not be used with %s.",
890 			    errmsg);
891 	} else if (STAILQ_EMPTY(&args.pa_events))
892 		/* All other uses require a PMC spec. */
893 		pmcstat_show_usage();
894 
895 	/* check for -t pid without a process PMC spec */
896 	if ((args.pa_required & FLAG_HAS_TARGET) &&
897 	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
898 		errx(EX_USAGE,
899 "ERROR: option -t requires a process mode PMC to be specified."
900 		    );
901 
902 	/* check for process-mode options without a command or -t pid */
903 	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
904 	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
905 		errx(EX_USAGE,
906 "ERROR: options -d, -E, -p, -P, and -W require a command line or target process."
907 		    );
908 
909 	/* check for -p | -P without a target process of some sort */
910 	if ((args.pa_required & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) &&
911 	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
912 		errx(EX_USAGE,
913 "ERROR: options -P and -p require a target process or a command line."
914 		    );
915 
916 	/* check for process-mode options without a process-mode PMC */
917 	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
918 	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
919 		errx(EX_USAGE,
920 "ERROR: options -d, -E, and -W require a process mode PMC to be specified."
921 		    );
922 
923 	/* check for -c cpu with no system mode PMCs or logfile. */
924 	if ((args.pa_required & FLAG_HAS_SYSTEM_PMCS) &&
925 	    (args.pa_flags & FLAG_HAS_SYSTEM_PMCS) == 0 &&
926 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
927 		errx(EX_USAGE,
928 "ERROR: option -c requires at least one system mode PMC to be specified."
929 		    );
930 
931 	/* check for counting mode options without a counting PMC */
932 	if ((args.pa_required & FLAG_HAS_COUNTING_PMCS) &&
933 	    (args.pa_flags & FLAG_HAS_COUNTING_PMCS) == 0)
934 		errx(EX_USAGE,
935 "ERROR: options -C, -W and -o require at least one counting mode PMC to be specified."
936 		    );
937 
938 	/* check for sampling mode options without a sampling PMC spec */
939 	if ((args.pa_required & FLAG_HAS_SAMPLING_PMCS) &&
940 	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) == 0)
941 		errx(EX_USAGE,
942 "ERROR: options -N, -n and -O require at least one sampling mode PMC to be specified."
943 		    );
944 
945 	/* check if -g/-G/-m/-T are being used correctly */
946 	if ((args.pa_flags & FLAG_DO_ANALYSIS) &&
947 	    !(args.pa_flags & (FLAG_HAS_SAMPLING_PMCS|FLAG_READ_LOGFILE)))
948 		errx(EX_USAGE,
949 "ERROR: options -g/-G/-m/-T require sampling PMCs or -R to be specified."
950 		    );
951 
952 	/* check if -e was specified without -g */
953 	if ((args.pa_flags & FLAG_DO_WIDE_GPROF_HC) &&
954 	    !(args.pa_flags & FLAG_DO_GPROF))
955 		errx(EX_USAGE,
956 "ERROR: option -e requires gprof mode to be specified."
957 		    );
958 
959 	/* check if -O was spuriously specified */
960 	if ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) &&
961 	    (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0)
962 		errx(EX_USAGE,
963 "ERROR: option -O is used only with options -E, -P, -S and -W."
964 		    );
965 
966 	/* -k kernel path require -g/-G/-m/-T or -R */
967 	if ((args.pa_flags & FLAG_HAS_KERNELPATH) &&
968 	    (args.pa_flags & FLAG_DO_ANALYSIS) == 0 &&
969 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
970 	    errx(EX_USAGE, "ERROR: option -k is only used with -g/-R/-m/-T.");
971 
972 	/* -D only applies to gprof output mode (-g) */
973 	if ((args.pa_flags & FLAG_HAS_SAMPLESDIR) &&
974 	    (args.pa_flags & FLAG_DO_GPROF) == 0)
975 	    errx(EX_USAGE, "ERROR: option -D is only used with -g.");
976 
977 	/* -M mapfile requires -g or -R */
978 	if (args.pa_mapfilename != NULL &&
979 	    (args.pa_flags & FLAG_DO_GPROF) == 0 &&
980 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
981 	    errx(EX_USAGE, "ERROR: option -M is only used with -g/-R.");
982 
983 	/*
984 	 * Disallow textual output of sampling PMCs if counting PMCs
985 	 * have also been asked for, mostly because the combined output
986 	 * is difficult to make sense of.
987 	 */
988 	if ((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
989 	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
990 	    ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) == 0))
991 		errx(EX_USAGE,
992 "ERROR: option -O is required if counting and sampling PMCs are specified together."
993 		    );
994 
995 	/*
996 	 * Check if 'kerneldir' refers to a file rather than a
997 	 * directory.  If so, use `dirname path` to determine the
998 	 * kernel directory.
999 	 */
1000 	(void) snprintf(buffer, sizeof(buffer), "%s%s", args.pa_fsroot,
1001 	    args.pa_kernel);
1002 	if (stat(buffer, &sb) < 0)
1003 		err(EX_OSERR, "ERROR: Cannot locate kernel \"%s\"",
1004 		    buffer);
1005 	if (!S_ISREG(sb.st_mode) && !S_ISDIR(sb.st_mode))
1006 		errx(EX_USAGE, "ERROR: \"%s\": Unsupported file type.",
1007 		    buffer);
1008 	if (!S_ISDIR(sb.st_mode)) {
1009 		tmp = args.pa_kernel;
1010 		args.pa_kernel = strdup(dirname(args.pa_kernel));
1011 		if (args.pa_kernel == NULL)
1012 			errx(EX_SOFTWARE, "ERROR: Out of memory");
1013 		free(tmp);
1014 		(void) snprintf(buffer, sizeof(buffer), "%s%s",
1015 		    args.pa_fsroot, args.pa_kernel);
1016 		if (stat(buffer, &sb) < 0)
1017 			err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
1018 			    buffer);
1019 		if (!S_ISDIR(sb.st_mode))
1020 			errx(EX_USAGE,
1021 			    "ERROR: \"%s\" is not a directory.",
1022 			    buffer);
1023 	}
1024 
1025 	/*
1026 	 * If we have a callgraph be created, select the outputfile.
1027 	 */
1028 	if (args.pa_flags & FLAG_DO_CALLGRAPHS) {
1029 		if (strcmp(graphfilename, "-") == 0)
1030 		    args.pa_graphfile = args.pa_printfile;
1031 		else {
1032 			args.pa_graphfile = fopen(graphfilename, "w");
1033 			if (args.pa_graphfile == NULL)
1034 				err(EX_OSERR,
1035 				    "ERROR: cannot open \"%s\" for writing",
1036 				    graphfilename);
1037 		}
1038 	}
1039 	if (args.pa_flags & FLAG_DO_ANNOTATE) {
1040 		args.pa_graphfile = fopen(graphfilename, "w");
1041 		if (args.pa_graphfile == NULL)
1042 			err(EX_OSERR, "ERROR: cannot open \"%s\" for writing",
1043 			    graphfilename);
1044 	}
1045 
1046 	/* if we've been asked to process a log file, skip init */
1047 	if ((args.pa_flags & FLAG_READ_LOGFILE) == 0) {
1048 		if (pmc_init() < 0)
1049 			err(EX_UNAVAILABLE,
1050 			    "ERROR: Initialization of the pmc(3) library failed"
1051 			    );
1052 
1053 		if ((npmc = pmc_npmc(0)) < 0) /* assume all CPUs are identical */
1054 			err(EX_OSERR,
1055 "ERROR: Cannot determine the number of PMCs on CPU %d",
1056 			    0);
1057 	}
1058 
1059 	/* Allocate a kqueue */
1060 	if ((pmcstat_kq = kqueue()) < 0)
1061 		err(EX_OSERR, "ERROR: Cannot allocate kqueue");
1062 
1063 	/* Setup the logfile as the source. */
1064 	if (args.pa_flags & FLAG_READ_LOGFILE) {
1065 		/*
1066 		 * Print the log in textual form if we haven't been
1067 		 * asked to generate profiling information.
1068 		 */
1069 		if ((args.pa_flags & FLAG_DO_ANALYSIS) == 0)
1070 			args.pa_flags |= FLAG_DO_PRINT;
1071 
1072 		pmcstat_log_initialize_logging();
1073 		rfd = pmcstat_open_log(args.pa_inputpath,
1074 		    PMCSTAT_OPEN_FOR_READ);
1075 		if ((args.pa_logparser = pmclog_open(rfd)) == NULL)
1076 			err(EX_OSERR, "ERROR: Cannot create parser");
1077 		if (fcntl(rfd, F_SETFL, O_NONBLOCK) < 0)
1078 			err(EX_OSERR, "ERROR: fcntl(2) failed");
1079 		EV_SET(&kev, rfd, EVFILT_READ, EV_ADD,
1080 		    0, 0, NULL);
1081 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1082 			err(EX_OSERR, "ERROR: Cannot register kevent");
1083 	}
1084 	/*
1085 	 * Configure the specified log file or setup a default log
1086 	 * consumer via a pipe.
1087 	 */
1088 	if (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) {
1089 		if (args.pa_outputpath)
1090 			args.pa_logfd = pmcstat_open_log(args.pa_outputpath,
1091 			    PMCSTAT_OPEN_FOR_WRITE);
1092 		else {
1093 			/*
1094 			 * process the log on the fly by reading it in
1095 			 * through a pipe.
1096 			 */
1097 			if (pipe(pipefd) < 0)
1098 				err(EX_OSERR, "ERROR: pipe(2) failed");
1099 
1100 			if (fcntl(pipefd[READPIPEFD], F_SETFL, O_NONBLOCK) < 0)
1101 				err(EX_OSERR, "ERROR: fcntl(2) failed");
1102 
1103 			EV_SET(&kev, pipefd[READPIPEFD], EVFILT_READ, EV_ADD,
1104 			    0, 0, NULL);
1105 
1106 			if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1107 				err(EX_OSERR, "ERROR: Cannot register kevent");
1108 
1109 			args.pa_logfd = pipefd[WRITEPIPEFD];
1110 
1111 			args.pa_flags |= FLAG_HAS_PIPE;
1112 			if ((args.pa_flags & FLAG_DO_TOP) == 0)
1113 				args.pa_flags |= FLAG_DO_PRINT;
1114 			args.pa_logparser = pmclog_open(pipefd[READPIPEFD]);
1115 		}
1116 
1117 		if (pmc_configure_logfile(args.pa_logfd) < 0)
1118 			err(EX_OSERR, "ERROR: Cannot configure log file");
1119 	}
1120 
1121 	/* remember to check for driver errors if we are sampling or logging */
1122 	check_driver_stats = (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) ||
1123 	    (args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE);
1124 
1125 	/*
1126 	if (args.pa_flags & FLAG_READ_LOGFILE) {
1127 	 * Allocate PMCs.
1128 	 */
1129 
1130 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1131 		if (pmc_allocate(ev->ev_spec, ev->ev_mode,
1132 			ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid,
1133 			ev->ev_count) < 0)
1134 			err(EX_OSERR,
1135 "ERROR: Cannot allocate %s-mode pmc with specification \"%s\"",
1136 			    PMC_IS_SYSTEM_MODE(ev->ev_mode) ?
1137 			    "system" : "process", ev->ev_spec);
1138 
1139 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode) &&
1140 		    pmc_set(ev->ev_pmcid, ev->ev_count) < 0)
1141 			err(EX_OSERR,
1142 			    "ERROR: Cannot set sampling count for PMC \"%s\"",
1143 			    ev->ev_name);
1144 	}
1145 
1146 	/* compute printout widths */
1147 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1148 		int counter_width;
1149 		int display_width;
1150 		int header_width;
1151 
1152 		(void) pmc_width(ev->ev_pmcid, &counter_width);
1153 		header_width = strlen(ev->ev_name) + 2; /* prefix '%c/' */
1154 		display_width = (int) floor(counter_width / 3.32193) + 1;
1155 
1156 		if (PMC_IS_SYSTEM_MODE(ev->ev_mode))
1157 			header_width += 3; /* 2 digit CPU number + '/' */
1158 
1159 		if (header_width > display_width) {
1160 			ev->ev_fieldskip = 0;
1161 			ev->ev_fieldwidth = header_width;
1162 		} else {
1163 			ev->ev_fieldskip = display_width -
1164 			    header_width;
1165 			ev->ev_fieldwidth = display_width;
1166 		}
1167 	}
1168 
1169 	/*
1170 	 * If our output is being set to a terminal, register a handler
1171 	 * for window size changes.
1172 	 */
1173 
1174 	if (isatty(fileno(args.pa_printfile))) {
1175 
1176 		if (ioctl(fileno(args.pa_printfile), TIOCGWINSZ, &ws) < 0)
1177 			err(EX_OSERR, "ERROR: Cannot determine window size");
1178 
1179 		pmcstat_displayheight = ws.ws_row - 1;
1180 		pmcstat_displaywidth  = ws.ws_col - 1;
1181 
1182 		EV_SET(&kev, SIGWINCH, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1183 
1184 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1185 			err(EX_OSERR,
1186 			    "ERROR: Cannot register kevent for SIGWINCH");
1187 
1188 		args.pa_toptty = 1;
1189 	}
1190 
1191 	/*
1192 	 * Listen to key input in top mode.
1193 	 */
1194 	if (args.pa_flags & FLAG_DO_TOP) {
1195 		EV_SET(&kev, fileno(stdin), EVFILT_READ, EV_ADD, 0, 0, NULL);
1196 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1197 			err(EX_OSERR, "ERROR: Cannot register kevent");
1198 	}
1199 
1200 	EV_SET(&kev, SIGINT, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1201 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1202 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGINT");
1203 
1204 	EV_SET(&kev, SIGIO, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1205 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1206 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGIO");
1207 
1208 	/*
1209 	 * An exec() failure of a forked child is signalled by the
1210 	 * child sending the parent a SIGCHLD.  We don't register an
1211 	 * actual signal handler for SIGCHLD, but instead use our
1212 	 * kqueue to pick up the signal.
1213 	 */
1214 	EV_SET(&kev, SIGCHLD, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1215 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1216 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGCHLD");
1217 
1218 	/*
1219 	 * Setup a timer if we have counting mode PMCs needing to be printed or
1220 	 * top mode plugin is active.
1221 	 */
1222 	if (((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1223 	     (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) ||
1224 	    (args.pa_flags & FLAG_DO_TOP)) {
1225 		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1226 		    args.pa_interval * 1000, NULL);
1227 
1228 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1229 			err(EX_OSERR,
1230 			    "ERROR: Cannot register kevent for timer");
1231 	}
1232 
1233 	/*
1234 	 * Setup a duration timer if we have sampling mode PMCs and
1235 	 * a duration time is set
1236 	 */
1237 	if ((args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
1238 	    (args.pa_flags & FLAG_HAS_DURATION)) {
1239 		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1240 		    args.pa_duration * 1000, NULL);
1241 
1242 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1243 			err(EX_OSERR, "ERROR: Cannot register kevent for "
1244 			    "time duration");
1245 	}
1246 
1247 	/* attach PMCs to the target process, starting it if specified */
1248 	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1249 		pmcstat_create_process(pmcstat_sockpair, &args, pmcstat_kq);
1250 
1251 	if (check_driver_stats && pmc_get_driver_stats(&ds_start) < 0)
1252 		err(EX_OSERR, "ERROR: Cannot retrieve driver statistics");
1253 
1254 	/* Attach process pmcs to the target process. */
1255 	if (args.pa_flags & (FLAG_HAS_TARGET | FLAG_HAS_COMMANDLINE)) {
1256 		if (SLIST_EMPTY(&args.pa_targets))
1257 			errx(EX_DATAERR,
1258 			    "ERROR: No matching target processes.");
1259 		if (args.pa_flags & FLAG_HAS_PROCESS_PMCS)
1260 			pmcstat_attach_pmcs(&args);
1261 
1262 		if (pmcstat_kvm) {
1263 			kvm_close(pmcstat_kvm);
1264 			pmcstat_kvm = NULL;
1265 		}
1266 	}
1267 
1268 	/* start the pmcs */
1269 	pmcstat_start_pmcs();
1270 
1271 	/* start the (commandline) process if needed */
1272 	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1273 		pmcstat_start_process(pmcstat_sockpair);
1274 
1275 	/* initialize logging */
1276 	pmcstat_log_initialize_logging();
1277 
1278 	/* Handle SIGINT using the kqueue loop */
1279 	sa.sa_handler = SIG_IGN;
1280 	sa.sa_flags   = 0;
1281 	(void) sigemptyset(&sa.sa_mask);
1282 
1283 	if (sigaction(SIGINT, &sa, NULL) < 0)
1284 		err(EX_OSERR, "ERROR: Cannot install signal handler");
1285 
1286 	/*
1287 	 * Setup the top mode display.
1288 	 */
1289 	if (args.pa_flags & FLAG_DO_TOP) {
1290 		args.pa_flags &= ~FLAG_DO_PRINT;
1291 
1292 		if (args.pa_toptty) {
1293 			/*
1294 			 * Init ncurses.
1295 			 */
1296 			initscr();
1297 			if(has_colors() == TRUE) {
1298 				args.pa_topcolor = 1;
1299 				start_color();
1300 				use_default_colors();
1301 				pair_content(0, &cf, &cb);
1302 				init_pair(1, COLOR_RED, cb);
1303 				init_pair(2, COLOR_YELLOW, cb);
1304 				init_pair(3, COLOR_GREEN, cb);
1305 			}
1306 			cbreak();
1307 			noecho();
1308 			nonl();
1309 			nodelay(stdscr, 1);
1310 			intrflush(stdscr, FALSE);
1311 			keypad(stdscr, TRUE);
1312 			clear();
1313 			/* Get terminal width / height with ncurses. */
1314 			getmaxyx(stdscr,
1315 			    pmcstat_displayheight, pmcstat_displaywidth);
1316 			pmcstat_displayheight--; pmcstat_displaywidth--;
1317 			atexit(pmcstat_topexit);
1318 		}
1319 	}
1320 
1321 	/*
1322 	 * loop till either the target process (if any) exits, or we
1323 	 * are killed by a SIGINT or we reached the time duration.
1324 	 */
1325 	runstate = PMCSTAT_RUNNING;
1326 	do_print = do_read = 0;
1327 	do {
1328 		if ((c = kevent(pmcstat_kq, NULL, 0, &kev, 1, NULL)) <= 0) {
1329 			if (errno != EINTR)
1330 				err(EX_OSERR, "ERROR: kevent failed");
1331 			else
1332 				continue;
1333 		}
1334 
1335 		if (kev.flags & EV_ERROR)
1336 			errc(EX_OSERR, kev.data, "ERROR: kevent failed");
1337 
1338 		switch (kev.filter) {
1339 		case EVFILT_PROC:  /* target has exited */
1340 			runstate = pmcstat_close_log(&args);
1341 			do_print = 1;
1342 			break;
1343 
1344 		case EVFILT_READ:  /* log file data is present */
1345 			if (kev.ident == (unsigned)fileno(stdin) &&
1346 			    (args.pa_flags & FLAG_DO_TOP)) {
1347 				if (pmcstat_keypress_log())
1348 					runstate = pmcstat_close_log(&args);
1349 			} else {
1350 				do_read = 0;
1351 				runstate = pmcstat_process_log();
1352 			}
1353 			break;
1354 
1355 		case EVFILT_SIGNAL:
1356 			if (kev.ident == SIGCHLD) {
1357 				/*
1358 				 * The child process sends us a
1359 				 * SIGCHLD if its exec() failed.  We
1360 				 * wait for it to exit and then exit
1361 				 * ourselves.
1362 				 */
1363 				(void) wait(&c);
1364 				runstate = PMCSTAT_FINISHED;
1365 			} else if (kev.ident == SIGIO) {
1366 				/*
1367 				 * We get a SIGIO if a PMC loses all
1368 				 * of its targets, or if logfile
1369 				 * writes encounter an error.
1370 				 */
1371 				runstate = pmcstat_close_log(&args);
1372 				do_print = 1; /* print PMCs at exit */
1373 			} else if (kev.ident == SIGINT) {
1374 				/* Kill the child process if we started it */
1375 				if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1376 					pmcstat_kill_process();
1377 				runstate = pmcstat_close_log(&args);
1378 			} else if (kev.ident == SIGWINCH) {
1379 				if (ioctl(fileno(args.pa_printfile),
1380 					TIOCGWINSZ, &ws) < 0)
1381 				    err(EX_OSERR,
1382 				        "ERROR: Cannot determine window size");
1383 				pmcstat_displayheight = ws.ws_row - 1;
1384 				pmcstat_displaywidth  = ws.ws_col - 1;
1385 			} else
1386 				assert(0);
1387 
1388 			break;
1389 
1390 		case EVFILT_TIMER:
1391 			/* time duration reached, exit */
1392 			if (args.pa_flags & FLAG_HAS_DURATION) {
1393 				runstate = PMCSTAT_FINISHED;
1394 				break;
1395 			}
1396 			/* print out counting PMCs */
1397 			if ((args.pa_flags & FLAG_DO_TOP) &&
1398 			    (args.pa_flags & FLAG_HAS_PIPE) &&
1399 			     pmc_flush_logfile() == 0)
1400 				do_read = 1;
1401 			do_print = 1;
1402 			break;
1403 
1404 		}
1405 
1406 		if (do_print && !do_read) {
1407 			if ((args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) {
1408 				pmcstat_print_pmcs();
1409 				if (runstate == PMCSTAT_FINISHED &&
1410 				    /* final newline */
1411 				    (args.pa_flags & FLAG_DO_PRINT) == 0)
1412 					(void) fprintf(args.pa_printfile, "\n");
1413 			}
1414 			if (args.pa_flags & FLAG_DO_TOP)
1415 				pmcstat_display_log();
1416 			do_print = 0;
1417 		}
1418 
1419 	} while (runstate != PMCSTAT_FINISHED);
1420 
1421 	if ((args.pa_flags & FLAG_DO_TOP) && args.pa_toptty) {
1422 		pmcstat_topexit();
1423 		args.pa_toptty = 0;
1424 	}
1425 
1426 	/* flush any pending log entries */
1427 	if (args.pa_flags & (FLAG_HAS_OUTPUT_LOGFILE | FLAG_HAS_PIPE))
1428 		pmc_close_logfile();
1429 
1430 	pmcstat_cleanup();
1431 
1432 	/* check if the driver lost any samples or events */
1433 	if (check_driver_stats) {
1434 		if (pmc_get_driver_stats(&ds_end) < 0)
1435 			err(EX_OSERR,
1436 			    "ERROR: Cannot retrieve driver statistics");
1437 		if (ds_start.pm_intr_bufferfull != ds_end.pm_intr_bufferfull &&
1438 		    args.pa_verbosity > 0)
1439 			warnx(
1440 "WARNING: sampling was paused at least %u time%s.\n"
1441 "Please consider tuning the \"kern.hwpmc.nsamples\" tunable.",
1442 			    ds_end.pm_intr_bufferfull -
1443 			    ds_start.pm_intr_bufferfull,
1444 			    ((ds_end.pm_intr_bufferfull -
1445 			    ds_start.pm_intr_bufferfull) != 1) ? "s" : ""
1446 			    );
1447 		if (ds_start.pm_buffer_requests_failed !=
1448 		    ds_end.pm_buffer_requests_failed &&
1449 		    args.pa_verbosity > 0)
1450 			warnx(
1451 "WARNING: at least %u event%s were discarded while running.\n"
1452 "Please consider tuning the \"kern.hwpmc.nbuffers\" tunable.",
1453 	 		    ds_end.pm_buffer_requests_failed -
1454 			    ds_start.pm_buffer_requests_failed,
1455 			    ((ds_end.pm_buffer_requests_failed -
1456 			    ds_start.pm_buffer_requests_failed) != 1) ? "s" : ""
1457 			    );
1458 	}
1459 
1460 	exit(EX_OK);
1461 }
1462