xref: /freebsd/usr.sbin/pmcstat/pmcstat.c (revision 2b833162)
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 <stdbool.h>
63 #include <stdint.h>
64 #include <stdio.h>
65 #include <stdlib.h>
66 #include <string.h>
67 #include <sysexits.h>
68 #include <unistd.h>
69 
70 #include <libpmcstat.h>
71 
72 #include "pmcstat.h"
73 
74 /*
75  * A given invocation of pmcstat(8) can manage multiple PMCs of both
76  * the system-wide and per-process variety.  Each of these could be in
77  * 'counting mode' or in 'sampling mode'.
78  *
79  * For 'counting mode' PMCs, pmcstat(8) will periodically issue a
80  * pmc_read() at the configured time interval and print out the value
81  * of the requested PMCs.
82  *
83  * For 'sampling mode' PMCs it can log to a file for offline analysis,
84  * or can analyse sampling data "on the fly", either by converting
85  * samples to printed textual form or by creating gprof(1) compatible
86  * profiles, one per program executed.  When creating gprof(1)
87  * profiles it can optionally merge entries from multiple processes
88  * for a given executable into a single profile file.
89  *
90  * pmcstat(8) can also execute a command line and attach PMCs to the
91  * resulting child process.  The protocol used is as follows:
92  *
93  * - parent creates a socketpair for two way communication and
94  *   fork()s.
95  * - subsequently:
96  *
97  *   /Parent/				/Child/
98  *
99  *   - Wait for childs token.
100  *					- Sends token.
101  *					- Awaits signal to start.
102  *  - Attaches PMCs to the child's pid
103  *    and starts them. Sets up
104  *    monitoring for the child.
105  *  - Signals child to start.
106  *					- Receives signal, attempts exec().
107  *
108  * After this point normal processing can happen.
109  */
110 
111 /* Globals */
112 
113 int		pmcstat_displayheight = DEFAULT_DISPLAY_HEIGHT;
114 int		pmcstat_displaywidth  = DEFAULT_DISPLAY_WIDTH;
115 static int	pmcstat_sockpair[NSOCKPAIRFD];
116 static int	pmcstat_kq;
117 static kvm_t	*pmcstat_kvm;
118 static struct kinfo_proc *pmcstat_plist;
119 struct pmcstat_args args;
120 static bool	libpmc_initialized = false;
121 
122 static void
123 pmcstat_get_cpumask(const char *cpuspec, cpuset_t *cpumask)
124 {
125 	int cpu;
126 	const char *s;
127 	char *end;
128 
129 	CPU_ZERO(cpumask);
130 	s = cpuspec;
131 
132 	do {
133 		cpu = strtol(s, &end, 0);
134 		if (cpu < 0 || end == s)
135 			errx(EX_USAGE,
136 			    "ERROR: Illegal CPU specification \"%s\".",
137 			    cpuspec);
138 		CPU_SET(cpu, cpumask);
139 		s = end + strspn(end, ", \t");
140 	} while (*s);
141 	assert(!CPU_EMPTY(cpumask));
142 }
143 
144 void
145 pmcstat_cleanup(void)
146 {
147 	struct pmcstat_ev *ev;
148 
149 	/* release allocated PMCs. */
150 	STAILQ_FOREACH(ev, &args.pa_events, ev_next)
151 		if (ev->ev_pmcid != PMC_ID_INVALID) {
152 			if (pmc_stop(ev->ev_pmcid) < 0)
153 				err(EX_OSERR,
154 				    "ERROR: cannot stop pmc 0x%x \"%s\"",
155 				    ev->ev_pmcid, ev->ev_name);
156 			if (pmc_release(ev->ev_pmcid) < 0)
157 				err(EX_OSERR,
158 				    "ERROR: cannot release pmc 0x%x \"%s\"",
159 				    ev->ev_pmcid, ev->ev_name);
160 		}
161 
162 	/* de-configure the log file if present. */
163 	if (args.pa_flags & (FLAG_HAS_PIPE | FLAG_HAS_OUTPUT_LOGFILE))
164 		(void) pmc_configure_logfile(-1);
165 
166 	if (args.pa_logparser) {
167 		pmclog_close(args.pa_logparser);
168 		args.pa_logparser = NULL;
169 	}
170 
171 	pmcstat_log_shutdown_logging();
172 }
173 
174 void
175 pmcstat_find_targets(const char *spec)
176 {
177 	int n, nproc, pid, rv;
178 	struct pmcstat_target *pt;
179 	char errbuf[_POSIX2_LINE_MAX], *end;
180 	static struct kinfo_proc *kp;
181 	regex_t reg;
182 	regmatch_t regmatch;
183 
184 	/* First check if we've been given a process id. */
185       	pid = strtol(spec, &end, 0);
186 	if (end != spec && pid >= 0) {
187 		if ((pt = malloc(sizeof(*pt))) == NULL)
188 			goto outofmemory;
189 		pt->pt_pid = pid;
190 		SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
191 		return;
192 	}
193 
194 	/* Otherwise treat arg as a regular expression naming processes. */
195 	if (pmcstat_kvm == NULL) {
196 		if ((pmcstat_kvm = kvm_openfiles(NULL, "/dev/null", NULL, 0,
197 		    errbuf)) == NULL)
198 			err(EX_OSERR, "ERROR: Cannot open kernel \"%s\"",
199 			    errbuf);
200 		if ((pmcstat_plist = kvm_getprocs(pmcstat_kvm, KERN_PROC_PROC,
201 		    0, &nproc)) == NULL)
202 			err(EX_OSERR, "ERROR: Cannot get process list: %s",
203 			    kvm_geterr(pmcstat_kvm));
204 	} else
205 		nproc = 0;
206 
207 	if ((rv = regcomp(&reg, spec, REG_EXTENDED|REG_NOSUB)) != 0) {
208 		regerror(rv, &reg, errbuf, sizeof(errbuf));
209 		err(EX_DATAERR, "ERROR: Failed to compile regex \"%s\": %s",
210 		    spec, errbuf);
211 	}
212 
213 	for (n = 0, kp = pmcstat_plist; n < nproc; n++, kp++) {
214 		if ((rv = regexec(&reg, kp->ki_comm, 1, &regmatch, 0)) == 0) {
215 			if ((pt = malloc(sizeof(*pt))) == NULL)
216 				goto outofmemory;
217 			pt->pt_pid = kp->ki_pid;
218 			SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
219 		} else if (rv != REG_NOMATCH) {
220 			regerror(rv, &reg, errbuf, sizeof(errbuf));
221 			errx(EX_SOFTWARE, "ERROR: Regex evalation failed: %s",
222 			    errbuf);
223 		}
224 	}
225 
226 	regfree(&reg);
227 
228 	return;
229 
230  outofmemory:
231 	errx(EX_SOFTWARE, "Out of memory.");
232 	/*NOTREACHED*/
233 }
234 
235 void
236 pmcstat_kill_process(void)
237 {
238 	struct pmcstat_target *pt;
239 
240 	assert(args.pa_flags & FLAG_HAS_COMMANDLINE);
241 
242 	/*
243 	 * If a command line was specified, it would be the very first
244 	 * in the list, before any other processes specified by -t.
245 	 */
246 	pt = SLIST_FIRST(&args.pa_targets);
247 	assert(pt != NULL);
248 
249 	if (kill(pt->pt_pid, SIGINT) != 0)
250 		err(EX_OSERR, "ERROR: cannot signal child process");
251 }
252 
253 void
254 pmcstat_start_pmcs(void)
255 {
256 	struct pmcstat_ev *ev;
257 
258 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
259 
260 	    assert(ev->ev_pmcid != PMC_ID_INVALID);
261 
262 	    if (pmc_start(ev->ev_pmcid) < 0) {
263 	        warn("ERROR: Cannot start pmc 0x%x \"%s\"",
264 		    ev->ev_pmcid, ev->ev_name);
265 		pmcstat_cleanup();
266 		exit(EX_OSERR);
267 	    }
268 	}
269 }
270 
271 void
272 pmcstat_print_headers(void)
273 {
274 	struct pmcstat_ev *ev;
275 	int c, w;
276 
277 	(void) fprintf(args.pa_printfile, PRINT_HEADER_PREFIX);
278 
279 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
280 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
281 			continue;
282 
283 		c = PMC_IS_SYSTEM_MODE(ev->ev_mode) ? 's' : 'p';
284 
285 		if (ev->ev_fieldskip != 0)
286 			(void) fprintf(args.pa_printfile, "%*s",
287 			    ev->ev_fieldskip, "");
288 		w = ev->ev_fieldwidth - ev->ev_fieldskip - 2;
289 
290 		if (c == 's')
291 			(void) fprintf(args.pa_printfile, "s/%02d/%-*s ",
292 			    ev->ev_cpu, w-3, ev->ev_name);
293 		else
294 			(void) fprintf(args.pa_printfile, "p/%*s ", w,
295 			    ev->ev_name);
296 	}
297 
298 	(void) fflush(args.pa_printfile);
299 }
300 
301 void
302 pmcstat_print_counters(void)
303 {
304 	int extra_width;
305 	struct pmcstat_ev *ev;
306 	pmc_value_t value;
307 
308 	extra_width = sizeof(PRINT_HEADER_PREFIX) - 1;
309 
310 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
311 
312 		/* skip sampling mode counters */
313 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
314 			continue;
315 
316 		if (pmc_read(ev->ev_pmcid, &value) < 0)
317 			err(EX_OSERR, "ERROR: Cannot read pmc \"%s\"",
318 			    ev->ev_name);
319 
320 		(void) fprintf(args.pa_printfile, "%*ju ",
321 		    ev->ev_fieldwidth + extra_width,
322 		    (uintmax_t) ev->ev_cumulative ? value :
323 		    (value - ev->ev_saved));
324 
325 		if (ev->ev_cumulative == 0)
326 			ev->ev_saved = value;
327 		extra_width = 0;
328 	}
329 
330 	(void) fflush(args.pa_printfile);
331 }
332 
333 /*
334  * Print output
335  */
336 
337 void
338 pmcstat_print_pmcs(void)
339 {
340 	static int linecount = 0;
341 
342 	/* check if we need to print a header line */
343 	if (++linecount > pmcstat_displayheight) {
344 		(void) fprintf(args.pa_printfile, "\n");
345 		linecount = 1;
346 	}
347 	if (linecount == 1)
348 		pmcstat_print_headers();
349 	(void) fprintf(args.pa_printfile, "\n");
350 
351 	pmcstat_print_counters();
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\t 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 static inline void
423 libpmc_initialize(int *npmc)
424 {
425 
426 	if (libpmc_initialized)
427 		return;
428 	if (pmc_init() < 0)
429 		err(EX_UNAVAILABLE, "ERROR: Initialization of the pmc(3)"
430 		    " library failed");
431 
432 	/* assume all CPUs are identical */
433 	if ((*npmc = pmc_npmc(0)) < 0)
434 		err(EX_OSERR, "ERROR: Cannot determine the number of PMCs on "
435 		    "CPU %d", 0);
436 	libpmc_initialized = true;
437 }
438 /*
439  * Main
440  */
441 
442 int
443 main(int argc, char **argv)
444 {
445 	cpuset_t cpumask, dommask, rootmask;
446 	double interval;
447 	double duration;
448 	int option, npmc;
449 	int c, check_driver_stats;
450 	int do_callchain, do_descendants, do_logproccsw, do_logprocexit;
451 	int do_print, do_read, do_listcounters, do_descr, domains;
452 	int do_userspace, i;
453 	size_t len;
454 	int graphdepth;
455 	int pipefd[2], rfd;
456 	int use_cumulative_counts;
457 	short cf, cb;
458 	uint64_t current_sampling_count;
459 	char *end, *tmp, *event;
460 	const char *errmsg, *graphfilename;
461 	enum pmcstat_state runstate;
462 	struct pmc_driverstats ds_start, ds_end;
463 	struct pmcstat_ev *ev;
464 	struct sigaction sa;
465 	struct kevent kev;
466 	struct winsize ws;
467 	struct stat sb;
468 	char buffer[PATH_MAX];
469 	uint32_t caps;
470 
471 	check_driver_stats      = 0;
472 	current_sampling_count  = 0;
473 	do_callchain		= 1;
474 	do_descr                = 0;
475 	do_descendants          = 0;
476 	do_userspace            = 0;
477 	do_logproccsw           = 0;
478 	do_logprocexit          = 0;
479 	do_listcounters         = 0;
480 	domains			= 0;
481 	use_cumulative_counts   = 0;
482 	graphfilename		= "-";
483 	args.pa_required	= 0;
484 	args.pa_flags		= 0;
485 	args.pa_verbosity	= 1;
486 	args.pa_logfd		= -1;
487 	args.pa_fsroot		= "";
488 	args.pa_samplesdir	= ".";
489 	args.pa_printfile	= stderr;
490 	args.pa_graphdepth	= DEFAULT_CALLGRAPH_DEPTH;
491 	args.pa_graphfile	= NULL;
492 	args.pa_interval	= DEFAULT_WAIT_INTERVAL;
493 	args.pa_mapfilename	= NULL;
494 	args.pa_inputpath	= NULL;
495 	args.pa_outputpath	= NULL;
496 	args.pa_pplugin		= PMCSTAT_PL_NONE;
497 	args.pa_plugin		= PMCSTAT_PL_NONE;
498 	args.pa_ctdumpinstr	= 1;
499 	args.pa_topmode		= PMCSTAT_TOP_DELTA;
500 	args.pa_toptty		= 0;
501 	args.pa_topcolor	= 0;
502 	args.pa_mergepmc	= 0;
503 	args.pa_duration	= 0.0;
504 	STAILQ_INIT(&args.pa_events);
505 	SLIST_INIT(&args.pa_targets);
506 	bzero(&ds_start, sizeof(ds_start));
507 	bzero(&ds_end, sizeof(ds_end));
508 	ev = NULL;
509 	event = NULL;
510 	caps = 0;
511 	CPU_ZERO(&cpumask);
512 
513 
514 	/* Default to using the running system kernel. */
515 	len = 0;
516 	if (sysctlbyname("kern.bootfile", NULL, &len, NULL, 0) == -1)
517 		err(EX_OSERR, "ERROR: Cannot determine path of running kernel");
518 	args.pa_kernel = malloc(len);
519 	if (args.pa_kernel == NULL)
520 		errx(EX_SOFTWARE, "ERROR: Out of memory.");
521 	if (sysctlbyname("kern.bootfile", args.pa_kernel, &len, NULL, 0) == -1)
522 		err(EX_OSERR, "ERROR: Cannot determine path of running kernel");
523 	len = sizeof(domains);
524 	if (sysctlbyname("vm.ndomains", &domains, &len, NULL, 0) == -1)
525 		err(EX_OSERR, "ERROR: Cannot get number of domains");
526 
527 	/*
528 	 * The initial CPU mask specifies the root mask of this process
529 	 * which is usually all CPUs in the system.
530 	 */
531 	if (cpuset_getaffinity(CPU_LEVEL_ROOT, CPU_WHICH_PID, -1,
532 	    sizeof(rootmask), &rootmask) == -1)
533 		err(EX_OSERR, "ERROR: Cannot determine the root set of CPUs");
534 	CPU_COPY(&rootmask, &cpumask);
535 
536 	while ((option = getopt(argc, argv,
537 	    "ACD:EF:G:ILM:NO:P:R:S:TUWZa:c:def:gi:k:l:m:n:o:p:qr:s:t:u:vw:z:")) != -1)
538 		switch (option) {
539 		case 'A':
540 			args.pa_flags |= FLAG_SKIP_TOP_FN_RES;
541 			break;
542 
543 		case 'a':	/* Annotate + callgraph */
544 			args.pa_flags |= FLAG_DO_ANNOTATE;
545 			args.pa_plugin = PMCSTAT_PL_ANNOTATE_CG;
546 			graphfilename  = optarg;
547 			break;
548 
549 		case 'C':	/* cumulative values */
550 			use_cumulative_counts = !use_cumulative_counts;
551 			args.pa_required |= FLAG_HAS_COUNTING_PMCS;
552 			break;
553 
554 		case 'c':	/* CPU */
555 			if (optarg[0] == '*' && optarg[1] == '\0')
556 				CPU_COPY(&rootmask, &cpumask);
557 			else
558 				pmcstat_get_cpumask(optarg, &cpumask);
559 
560 			args.pa_flags	 |= FLAGS_HAS_CPUMASK;
561 			args.pa_required |= FLAG_HAS_SYSTEM_PMCS;
562 			break;
563 
564 		case 'D':
565 			if (stat(optarg, &sb) < 0)
566 				err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
567 				    optarg);
568 			if (!S_ISDIR(sb.st_mode))
569 				errx(EX_USAGE,
570 				    "ERROR: \"%s\" is not a directory.",
571 				    optarg);
572 			args.pa_samplesdir = optarg;
573 			args.pa_flags     |= FLAG_HAS_SAMPLESDIR;
574 			args.pa_required  |= FLAG_DO_GPROF;
575 			break;
576 
577 		case 'd':	/* toggle descendents */
578 			do_descendants = !do_descendants;
579 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
580 			break;
581 
582 		case 'E':	/* log process exit */
583 			do_logprocexit = !do_logprocexit;
584 			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
585 			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
586 			break;
587 
588 		case 'e':	/* wide gprof metrics */
589 			args.pa_flags |= FLAG_DO_WIDE_GPROF_HC;
590 			break;
591 
592 		case 'F':	/* produce a system-wide calltree */
593 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
594 			args.pa_plugin = PMCSTAT_PL_CALLTREE;
595 			graphfilename = optarg;
596 			break;
597 
598 		case 'f':	/* plugins options */
599 			if (args.pa_plugin == PMCSTAT_PL_NONE)
600 				err(EX_USAGE, "ERROR: Need -g/-G/-m/-T.");
601 			pmcstat_pluginconfigure_log(optarg);
602 			break;
603 
604 		case 'G':	/* produce a system-wide callgraph */
605 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
606 			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
607 			graphfilename = optarg;
608 			break;
609 
610 		case 'g':	/* produce gprof compatible profiles */
611 			args.pa_flags |= FLAG_DO_GPROF;
612 			args.pa_pplugin = PMCSTAT_PL_CALLGRAPH;
613 			args.pa_plugin	= PMCSTAT_PL_GPROF;
614 			break;
615 
616 		case 'i':
617 			args.pa_flags |= FLAG_FILTER_THREAD_ID;
618 			args.pa_tid = strtol(optarg, &end, 0);
619 			break;
620 
621 		case 'I':
622 			args.pa_flags |= FLAG_SHOW_OFFSET;
623 			break;
624 
625 		case 'k':	/* pathname to the kernel */
626 			free(args.pa_kernel);
627 			args.pa_kernel = strdup(optarg);
628 			if (args.pa_kernel == NULL)
629 				errx(EX_SOFTWARE, "ERROR: Out of memory");
630 			args.pa_required |= FLAG_DO_ANALYSIS;
631 			args.pa_flags    |= FLAG_HAS_KERNELPATH;
632 			break;
633 
634 		case 'L':
635 			do_listcounters = 1;
636 			break;
637 
638 		case 'l':	/* time duration in seconds */
639 			duration = strtod(optarg, &end);
640 			if (*end != '\0' || duration <= 0)
641 				errx(EX_USAGE, "ERROR: Illegal duration time "
642 				    "value \"%s\".", optarg);
643 			args.pa_flags |= FLAG_HAS_DURATION;
644 			args.pa_duration = duration;
645 			break;
646 
647 		case 'm':
648 			args.pa_flags |= FLAG_DO_ANNOTATE;
649 			args.pa_plugin = PMCSTAT_PL_ANNOTATE;
650 			graphfilename  = optarg;
651 			break;
652 
653 		case 'M':	/* mapfile */
654 			args.pa_mapfilename = optarg;
655 			break;
656 
657 		case 'N':
658 			do_callchain = !do_callchain;
659 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
660 			break;
661 
662 		case 'p':	/* process virtual counting PMC */
663 		case 's':	/* system-wide counting PMC */
664 		case 'P':	/* process virtual sampling PMC */
665 		case 'S':	/* system-wide sampling PMC */
666 			caps = 0;
667 			if ((ev = malloc(sizeof(*ev))) == NULL)
668 				errx(EX_SOFTWARE, "ERROR: Out of memory.");
669 
670 			switch (option) {
671 			case 'p': ev->ev_mode = PMC_MODE_TC; break;
672 			case 's': ev->ev_mode = PMC_MODE_SC; break;
673 			case 'P': ev->ev_mode = PMC_MODE_TS; break;
674 			case 'S': ev->ev_mode = PMC_MODE_SS; break;
675 			}
676 
677 			if (option == 'P' || option == 'p') {
678 				args.pa_flags |= FLAG_HAS_PROCESS_PMCS;
679 				args.pa_required |= (FLAG_HAS_COMMANDLINE |
680 				    FLAG_HAS_TARGET);
681 			}
682 
683 			if (option == 'P' || option == 'S') {
684 				args.pa_flags |= FLAG_HAS_SAMPLING_PMCS;
685 				args.pa_required |= (FLAG_HAS_PIPE |
686 				    FLAG_HAS_OUTPUT_LOGFILE);
687 			}
688 
689 			if (option == 'p' || option == 's')
690 				args.pa_flags |= FLAG_HAS_COUNTING_PMCS;
691 
692 			if (option == 's' || option == 'S')
693 				args.pa_flags |= FLAG_HAS_SYSTEM_PMCS;
694 
695 			ev->ev_spec = strdup(optarg);
696 			if (ev->ev_spec == NULL)
697 				errx(EX_SOFTWARE, "ERROR: Out of memory.");
698 
699 			if (option == 'S' || option == 'P')
700 				ev->ev_count = current_sampling_count ? current_sampling_count : pmc_pmu_sample_rate_get(ev->ev_spec);
701 			else
702 				ev->ev_count = 0;
703 
704 			if (option == 'S' || option == 's')
705 				ev->ev_cpu = CPU_FFS(&cpumask) - 1;
706 			else
707 				ev->ev_cpu = PMC_CPU_ANY;
708 
709 			ev->ev_flags = 0;
710 			if (do_callchain) {
711 				ev->ev_flags |= PMC_F_CALLCHAIN;
712 				if (do_userspace)
713 					ev->ev_flags |= PMC_F_USERCALLCHAIN;
714 			}
715 			if (do_descendants)
716 				ev->ev_flags |= PMC_F_DESCENDANTS;
717 			if (do_logprocexit)
718 				ev->ev_flags |= PMC_F_LOG_PROCEXIT;
719 			if (do_logproccsw)
720 				ev->ev_flags |= PMC_F_LOG_PROCCSW;
721 
722 			ev->ev_cumulative  = use_cumulative_counts;
723 
724 			ev->ev_saved = 0LL;
725 			ev->ev_pmcid = PMC_ID_INVALID;
726 
727 			/* extract event name */
728 			c = strcspn(optarg, ", \t");
729 			ev->ev_name = malloc(c + 1);
730 			if (ev->ev_name == NULL)
731 				errx(EX_SOFTWARE, "ERROR: Out of memory.");
732 			(void) strncpy(ev->ev_name, optarg, c);
733 			*(ev->ev_name + c) = '\0';
734 			libpmc_initialize(&npmc);
735 			if (args.pa_flags & FLAG_HAS_SYSTEM_PMCS) {
736 				if (pmc_allocate(ev->ev_spec, ev->ev_mode,
737 				    ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid,
738 				    ev->ev_count) < 0)
739 					err(EX_OSERR, "ERROR: Cannot allocate "
740 					    "system-mode pmc with specification"
741 					    " \"%s\"", ev->ev_spec);
742 				if (pmc_capabilities(ev->ev_pmcid, &caps)) {
743 					pmc_release(ev->ev_pmcid);
744 					err(EX_OSERR, "ERROR: Cannot get pmc "
745 					    "capabilities");
746 				}
747 			}
748 
749 
750 			STAILQ_INSERT_TAIL(&args.pa_events, ev, ev_next);
751 
752 			if ((caps & PMC_CAP_SYSWIDE) == PMC_CAP_SYSWIDE)
753 				break;
754 			if ((caps & PMC_CAP_DOMWIDE) == PMC_CAP_DOMWIDE) {
755 				CPU_ZERO(&cpumask);
756 				/*
757 				 * Get number of domains and allocate one
758 				 * counter in each.
759 				 * First already allocated.
760 				 */
761 				for (i = 1; i < domains; i++) {
762 					CPU_ZERO(&dommask);
763 					cpuset_getaffinity(CPU_LEVEL_WHICH,
764 					    CPU_WHICH_DOMAIN, i, sizeof(dommask),
765 					    &dommask);
766 					CPU_SET(CPU_FFS(&dommask) - 1, &cpumask);
767 				}
768 				args.pa_flags |= FLAGS_HAS_CPUMASK;
769 			}
770 			if (option == 's' || option == 'S') {
771 				CPU_CLR(ev->ev_cpu, &cpumask);
772 				pmc_id_t saved_pmcid = ev->ev_pmcid;
773 				ev->ev_pmcid = PMC_ID_INVALID;
774 				pmcstat_clone_event_descriptor(ev, &cpumask, &args);
775 				ev->ev_pmcid = saved_pmcid;
776 				CPU_SET(ev->ev_cpu, &cpumask);
777 			}
778 
779 			break;
780 
781 		case 'n':	/* sampling count */
782 			current_sampling_count = strtol(optarg, &end, 0);
783 			if (*end != '\0' || current_sampling_count <= 0)
784 				errx(EX_USAGE,
785 				    "ERROR: Illegal count value \"%s\".",
786 				    optarg);
787 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
788 			break;
789 
790 		case 'o':	/* outputfile */
791 			if (args.pa_printfile != NULL &&
792 			    args.pa_printfile != stdout &&
793 			    args.pa_printfile != stderr)
794 				(void) fclose(args.pa_printfile);
795 			if ((args.pa_printfile = fopen(optarg, "w")) == NULL)
796 				errx(EX_OSERR,
797 				    "ERROR: cannot open \"%s\" for writing.",
798 				    optarg);
799 			args.pa_flags |= FLAG_DO_PRINT;
800 			break;
801 
802 		case 'O':	/* sampling output */
803 			if (args.pa_outputpath)
804 				errx(EX_USAGE,
805 "ERROR: option -O may only be specified once.");
806 			args.pa_outputpath = optarg;
807 			args.pa_flags |= FLAG_HAS_OUTPUT_LOGFILE;
808 			break;
809 
810 		case 'q':	/* quiet mode */
811 			args.pa_verbosity = 0;
812 			break;
813 
814 		case 'r':	/* root FS path */
815 			args.pa_fsroot = optarg;
816 			break;
817 
818 		case 'R':	/* read an existing log file */
819 			if (args.pa_inputpath != NULL)
820 				errx(EX_USAGE,
821 "ERROR: option -R may only be specified once.");
822 			args.pa_inputpath = optarg;
823 			if (args.pa_printfile == stderr)
824 				args.pa_printfile = stdout;
825 			args.pa_flags |= FLAG_READ_LOGFILE;
826 			break;
827 
828 		case 't':	/* target pid or process name */
829 			pmcstat_find_targets(optarg);
830 
831 			args.pa_flags |= FLAG_HAS_TARGET;
832 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
833 			break;
834 
835 		case 'T':	/* top mode */
836 			args.pa_flags |= FLAG_DO_TOP;
837 			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
838 			args.pa_ctdumpinstr = 0;
839 			args.pa_mergepmc = 1;
840 			if (args.pa_printfile == stderr)
841 				args.pa_printfile = stdout;
842 			break;
843 
844 		case 'u':
845 			do_descr = 1;
846 			event = optarg;
847 			break;
848 		case 'U':	/* toggle user-space callchain capture */
849 			do_userspace = !do_userspace;
850 			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
851 			break;
852 		case 'v':	/* verbose */
853 			args.pa_verbosity++;
854 			break;
855 
856 		case 'w':	/* wait interval */
857 			interval = strtod(optarg, &end);
858 			if (*end != '\0' || interval <= 0)
859 				errx(EX_USAGE,
860 "ERROR: Illegal wait interval value \"%s\".",
861 				    optarg);
862 			args.pa_flags |= FLAG_HAS_WAIT_INTERVAL;
863 			args.pa_interval = interval;
864 			break;
865 
866 		case 'W':	/* toggle LOG_CSW */
867 			do_logproccsw = !do_logproccsw;
868 			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
869 			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
870 			break;
871 
872 		case 'z':
873 			graphdepth = strtod(optarg, &end);
874 			if (*end != '\0' || graphdepth <= 0)
875 				errx(EX_USAGE,
876 				    "ERROR: Illegal callchain depth \"%s\".",
877 				    optarg);
878 			args.pa_graphdepth = graphdepth;
879 			args.pa_required |= FLAG_DO_CALLGRAPHS;
880 			break;
881 
882 		case '?':
883 		default:
884 			pmcstat_show_usage();
885 			break;
886 
887 		}
888 	if ((do_listcounters | do_descr) &&
889 		pmc_pmu_enabled() == 0)
890 			errx(EX_USAGE, "pmu features not supported on host or hwpmc not loaded");
891 	if (do_listcounters) {
892 		pmc_pmu_print_counters(NULL);
893 	} else if (do_descr) {
894 		pmc_pmu_print_counter_desc(event);
895 	}
896 	if (do_listcounters | do_descr)
897 		exit(0);
898 
899 	args.pa_argc = (argc -= optind);
900 	args.pa_argv = (argv += optind);
901 
902 	/* If we read from logfile and no specified CPU mask use
903 	 * the maximum CPU count.
904 	 */
905 	if ((args.pa_flags & FLAG_READ_LOGFILE) &&
906 	    (args.pa_flags & FLAGS_HAS_CPUMASK) == 0)
907 		CPU_FILL(&cpumask);
908 
909 	args.pa_cpumask = cpumask; /* For selecting CPUs using -R. */
910 
911 	if (argc)	/* command line present */
912 		args.pa_flags |= FLAG_HAS_COMMANDLINE;
913 
914 	if (args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS |
915 	    FLAG_DO_ANNOTATE | FLAG_DO_TOP))
916 		args.pa_flags |= FLAG_DO_ANALYSIS;
917 
918 	/*
919 	 * Check invocation syntax.
920 	 */
921 
922 	/* disallow -O and -R together */
923 	if (args.pa_outputpath && args.pa_inputpath)
924 		errx(EX_USAGE,
925 		    "ERROR: options -O and -R are mutually exclusive.");
926 
927 	/* disallow -T and -l together */
928 	if ((args.pa_flags & FLAG_HAS_DURATION) &&
929 	    (args.pa_flags & FLAG_DO_TOP))
930 		errx(EX_USAGE, "ERROR: options -T and -l are mutually "
931 		    "exclusive.");
932 
933 	/* -a and -m require -R */
934 	if (args.pa_flags & FLAG_DO_ANNOTATE && args.pa_inputpath == NULL)
935 		errx(EX_USAGE, "ERROR: option %s requires an input file",
936 		    args.pa_plugin == PMCSTAT_PL_ANNOTATE ? "-m" : "-a");
937 
938 	/* -m option is not allowed combined with -g or -G. */
939 	if (args.pa_flags & FLAG_DO_ANNOTATE &&
940 	    args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS))
941 		errx(EX_USAGE,
942 		    "ERROR: option -m and -g | -G are mutually exclusive");
943 
944 	if (args.pa_flags & FLAG_READ_LOGFILE) {
945 		errmsg = NULL;
946 		if (args.pa_flags & FLAG_HAS_COMMANDLINE)
947 			errmsg = "a command line specification";
948 		else if (args.pa_flags & FLAG_HAS_TARGET)
949 			errmsg = "option -t";
950 		else if (!STAILQ_EMPTY(&args.pa_events))
951 			errmsg = "a PMC event specification";
952 		if (errmsg)
953 			errx(EX_USAGE,
954 			    "ERROR: option -R may not be used with %s.",
955 			    errmsg);
956 	} else if (STAILQ_EMPTY(&args.pa_events))
957 		/* All other uses require a PMC spec. */
958 		pmcstat_show_usage();
959 
960 	/* check for -t pid without a process PMC spec */
961 	if ((args.pa_flags & FLAG_HAS_TARGET) &&
962 	    (args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
963 	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
964 		errx(EX_USAGE,
965 "ERROR: option -t requires a process mode PMC to be specified."
966 		    );
967 
968 	/* check for process-mode options without a command or -t pid */
969 	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
970 	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
971 		errx(EX_USAGE,
972 "ERROR: options -d, -E, -p, -P, and -W require a command line or target process."
973 		    );
974 
975 	/* check for -p | -P without a target process of some sort */
976 	if ((args.pa_required & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) &&
977 	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
978 		errx(EX_USAGE,
979 "ERROR: options -P and -p require a target process or a command line."
980 		    );
981 
982 	/* check for process-mode options without a process-mode PMC */
983 	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
984 	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
985 		errx(EX_USAGE,
986 "ERROR: options -d, -E, and -W require a process mode PMC to be specified."
987 		    );
988 
989 	/* check for -c cpu with no system mode PMCs or logfile. */
990 	if ((args.pa_required & FLAG_HAS_SYSTEM_PMCS) &&
991 	    (args.pa_flags & FLAG_HAS_SYSTEM_PMCS) == 0 &&
992 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
993 		errx(EX_USAGE,
994 "ERROR: option -c requires at least one system mode PMC to be specified."
995 		    );
996 
997 	/* check for counting mode options without a counting PMC */
998 	if ((args.pa_required & FLAG_HAS_COUNTING_PMCS) &&
999 	    (args.pa_flags & FLAG_HAS_COUNTING_PMCS) == 0)
1000 		errx(EX_USAGE,
1001 "ERROR: options -C, -W and -o require at least one counting mode PMC to be specified."
1002 		    );
1003 
1004 	/* check for sampling mode options without a sampling PMC spec */
1005 	if ((args.pa_required & FLAG_HAS_SAMPLING_PMCS) &&
1006 	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) == 0)
1007 		errx(EX_USAGE,
1008 "ERROR: options -N, -n and -O require at least one sampling mode PMC to be specified."
1009 		    );
1010 
1011 	/* check if -g/-G/-m/-T are being used correctly */
1012 	if ((args.pa_flags & FLAG_DO_ANALYSIS) &&
1013 	    !(args.pa_flags & (FLAG_HAS_SAMPLING_PMCS|FLAG_READ_LOGFILE)))
1014 		errx(EX_USAGE,
1015 "ERROR: options -g/-G/-m/-T require sampling PMCs or -R to be specified."
1016 		    );
1017 
1018 	/* check if -e was specified without -g */
1019 	if ((args.pa_flags & FLAG_DO_WIDE_GPROF_HC) &&
1020 	    !(args.pa_flags & FLAG_DO_GPROF))
1021 		errx(EX_USAGE,
1022 "ERROR: option -e requires gprof mode to be specified."
1023 		    );
1024 
1025 	/* check if -O was spuriously specified */
1026 	if ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) &&
1027 	    (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0)
1028 		errx(EX_USAGE,
1029 "ERROR: option -O is used only with options -E, -P, -S and -W."
1030 		    );
1031 
1032 	/* -k kernel path require -g/-G/-m/-T or -R */
1033 	if ((args.pa_flags & FLAG_HAS_KERNELPATH) &&
1034 	    (args.pa_flags & FLAG_DO_ANALYSIS) == 0 &&
1035 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
1036 	    errx(EX_USAGE, "ERROR: option -k is only used with -g/-R/-m/-T.");
1037 
1038 	/* -D only applies to gprof output mode (-g) */
1039 	if ((args.pa_flags & FLAG_HAS_SAMPLESDIR) &&
1040 	    (args.pa_flags & FLAG_DO_GPROF) == 0)
1041 	    errx(EX_USAGE, "ERROR: option -D is only used with -g.");
1042 
1043 	/* -M mapfile requires -g or -R */
1044 	if (args.pa_mapfilename != NULL &&
1045 	    (args.pa_flags & FLAG_DO_GPROF) == 0 &&
1046 	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
1047 	    errx(EX_USAGE, "ERROR: option -M is only used with -g/-R.");
1048 
1049 	/*
1050 	 * Disallow textual output of sampling PMCs if counting PMCs
1051 	 * have also been asked for, mostly because the combined output
1052 	 * is difficult to make sense of.
1053 	 */
1054 	if ((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1055 	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
1056 	    ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) == 0))
1057 		errx(EX_USAGE,
1058 "ERROR: option -O is required if counting and sampling PMCs are specified together."
1059 		    );
1060 
1061 	/*
1062 	 * Check if 'kerneldir' refers to a file rather than a
1063 	 * directory.  If so, use `dirname path` to determine the
1064 	 * kernel directory.
1065 	 */
1066 	(void) snprintf(buffer, sizeof(buffer), "%s%s", args.pa_fsroot,
1067 	    args.pa_kernel);
1068 	if (stat(buffer, &sb) < 0)
1069 		err(EX_OSERR, "ERROR: Cannot locate kernel \"%s\"",
1070 		    buffer);
1071 	if (!S_ISREG(sb.st_mode) && !S_ISDIR(sb.st_mode))
1072 		errx(EX_USAGE, "ERROR: \"%s\": Unsupported file type.",
1073 		    buffer);
1074 	if (!S_ISDIR(sb.st_mode)) {
1075 		tmp = args.pa_kernel;
1076 		args.pa_kernel = strdup(dirname(args.pa_kernel));
1077 		if (args.pa_kernel == NULL)
1078 			errx(EX_SOFTWARE, "ERROR: Out of memory");
1079 		free(tmp);
1080 		(void) snprintf(buffer, sizeof(buffer), "%s%s",
1081 		    args.pa_fsroot, args.pa_kernel);
1082 		if (stat(buffer, &sb) < 0)
1083 			err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
1084 			    buffer);
1085 		if (!S_ISDIR(sb.st_mode))
1086 			errx(EX_USAGE,
1087 			    "ERROR: \"%s\" is not a directory.",
1088 			    buffer);
1089 	}
1090 
1091 	/*
1092 	 * If we have a callgraph be created, select the outputfile.
1093 	 */
1094 	if (args.pa_flags & FLAG_DO_CALLGRAPHS) {
1095 		if (strcmp(graphfilename, "-") == 0)
1096 		    args.pa_graphfile = args.pa_printfile;
1097 		else {
1098 			args.pa_graphfile = fopen(graphfilename, "w");
1099 			if (args.pa_graphfile == NULL)
1100 				err(EX_OSERR,
1101 				    "ERROR: cannot open \"%s\" for writing",
1102 				    graphfilename);
1103 		}
1104 	}
1105 	if (args.pa_flags & FLAG_DO_ANNOTATE) {
1106 		args.pa_graphfile = fopen(graphfilename, "w");
1107 		if (args.pa_graphfile == NULL)
1108 			err(EX_OSERR, "ERROR: cannot open \"%s\" for writing",
1109 			    graphfilename);
1110 	}
1111 
1112 	/* if we've been asked to process a log file, skip init */
1113 	if ((args.pa_flags & FLAG_READ_LOGFILE) == 0)
1114 		libpmc_initialize(&npmc);
1115 
1116 	/* Allocate a kqueue */
1117 	if ((pmcstat_kq = kqueue()) < 0)
1118 		err(EX_OSERR, "ERROR: Cannot allocate kqueue");
1119 
1120 	/* Setup the logfile as the source. */
1121 	if (args.pa_flags & FLAG_READ_LOGFILE) {
1122 		/*
1123 		 * Print the log in textual form if we haven't been
1124 		 * asked to generate profiling information.
1125 		 */
1126 		if ((args.pa_flags & FLAG_DO_ANALYSIS) == 0)
1127 			args.pa_flags |= FLAG_DO_PRINT;
1128 
1129 		pmcstat_log_initialize_logging();
1130 		rfd = pmcstat_open_log(args.pa_inputpath,
1131 		    PMCSTAT_OPEN_FOR_READ);
1132 		if ((args.pa_logparser = pmclog_open(rfd)) == NULL)
1133 			err(EX_OSERR, "ERROR: Cannot create parser");
1134 		if (fcntl(rfd, F_SETFL, O_NONBLOCK) < 0)
1135 			err(EX_OSERR, "ERROR: fcntl(2) failed");
1136 		EV_SET(&kev, rfd, EVFILT_READ, EV_ADD,
1137 		    0, 0, NULL);
1138 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1139 			err(EX_OSERR, "ERROR: Cannot register kevent");
1140 	}
1141 	/*
1142 	 * Configure the specified log file or setup a default log
1143 	 * consumer via a pipe.
1144 	 */
1145 	if (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) {
1146 		if (args.pa_outputpath)
1147 			args.pa_logfd = pmcstat_open_log(args.pa_outputpath,
1148 			    PMCSTAT_OPEN_FOR_WRITE);
1149 		else {
1150 			/*
1151 			 * process the log on the fly by reading it in
1152 			 * through a pipe.
1153 			 */
1154 			if (pipe(pipefd) < 0)
1155 				err(EX_OSERR, "ERROR: pipe(2) failed");
1156 
1157 			if (fcntl(pipefd[READPIPEFD], F_SETFL, O_NONBLOCK) < 0)
1158 				err(EX_OSERR, "ERROR: fcntl(2) failed");
1159 
1160 			EV_SET(&kev, pipefd[READPIPEFD], EVFILT_READ, EV_ADD,
1161 			    0, 0, NULL);
1162 
1163 			if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1164 				err(EX_OSERR, "ERROR: Cannot register kevent");
1165 
1166 			args.pa_logfd = pipefd[WRITEPIPEFD];
1167 
1168 			args.pa_flags |= FLAG_HAS_PIPE;
1169 			if ((args.pa_flags & FLAG_DO_TOP) == 0)
1170 				args.pa_flags |= FLAG_DO_PRINT;
1171 			args.pa_logparser = pmclog_open(pipefd[READPIPEFD]);
1172 		}
1173 
1174 		if (pmc_configure_logfile(args.pa_logfd) < 0)
1175 			err(EX_OSERR, "ERROR: Cannot configure log file");
1176 	}
1177 
1178 	/* remember to check for driver errors if we are sampling or logging */
1179 	check_driver_stats = (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) ||
1180 	    (args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE);
1181 
1182 	/*
1183 	if (args.pa_flags & FLAG_READ_LOGFILE) {
1184 	 * Allocate PMCs.
1185 	 */
1186 
1187 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1188 		if (pmc_allocate(ev->ev_spec, ev->ev_mode,
1189 			ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid,
1190 			ev->ev_count) < 0)
1191 			err(EX_OSERR,
1192 "ERROR: Cannot allocate %s-mode pmc with specification \"%s\"",
1193 			    PMC_IS_SYSTEM_MODE(ev->ev_mode) ?
1194 			    "system" : "process", ev->ev_spec);
1195 
1196 		if (PMC_IS_SAMPLING_MODE(ev->ev_mode) &&
1197 		    pmc_set(ev->ev_pmcid, ev->ev_count) < 0)
1198 			err(EX_OSERR,
1199 			    "ERROR: Cannot set sampling count for PMC \"%s\"",
1200 			    ev->ev_name);
1201 	}
1202 
1203 	/* compute printout widths */
1204 	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1205 		int counter_width;
1206 		int display_width;
1207 		int header_width;
1208 
1209 		(void) pmc_width(ev->ev_pmcid, &counter_width);
1210 		header_width = strlen(ev->ev_name) + 2; /* prefix '%c/' */
1211 		display_width = (int) floor(counter_width / 3.32193) + 1;
1212 
1213 		if (PMC_IS_SYSTEM_MODE(ev->ev_mode))
1214 			header_width += 3; /* 2 digit CPU number + '/' */
1215 
1216 		if (header_width > display_width) {
1217 			ev->ev_fieldskip = 0;
1218 			ev->ev_fieldwidth = header_width;
1219 		} else {
1220 			ev->ev_fieldskip = display_width -
1221 			    header_width;
1222 			ev->ev_fieldwidth = display_width;
1223 		}
1224 	}
1225 
1226 	/*
1227 	 * If our output is being set to a terminal, register a handler
1228 	 * for window size changes.
1229 	 */
1230 
1231 	if (isatty(fileno(args.pa_printfile))) {
1232 
1233 		if (ioctl(fileno(args.pa_printfile), TIOCGWINSZ, &ws) < 0)
1234 			err(EX_OSERR, "ERROR: Cannot determine window size");
1235 
1236 		pmcstat_displayheight = ws.ws_row - 1;
1237 		pmcstat_displaywidth  = ws.ws_col - 1;
1238 
1239 		EV_SET(&kev, SIGWINCH, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1240 
1241 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1242 			err(EX_OSERR,
1243 			    "ERROR: Cannot register kevent for SIGWINCH");
1244 
1245 		args.pa_toptty = 1;
1246 	}
1247 
1248 	/*
1249 	 * Listen to key input in top mode.
1250 	 */
1251 	if (args.pa_flags & FLAG_DO_TOP) {
1252 		EV_SET(&kev, fileno(stdin), EVFILT_READ, EV_ADD, 0, 0, NULL);
1253 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1254 			err(EX_OSERR, "ERROR: Cannot register kevent");
1255 	}
1256 
1257 	EV_SET(&kev, SIGINT, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1258 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1259 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGINT");
1260 
1261 	EV_SET(&kev, SIGIO, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1262 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1263 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGIO");
1264 
1265 	/*
1266 	 * An exec() failure of a forked child is signalled by the
1267 	 * child sending the parent a SIGCHLD.  We don't register an
1268 	 * actual signal handler for SIGCHLD, but instead use our
1269 	 * kqueue to pick up the signal.
1270 	 */
1271 	EV_SET(&kev, SIGCHLD, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1272 	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1273 		err(EX_OSERR, "ERROR: Cannot register kevent for SIGCHLD");
1274 
1275 	/*
1276 	 * Setup a timer if we have counting mode PMCs needing to be printed or
1277 	 * top mode plugin is active.
1278 	 */
1279 	if (((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1280 	     (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) ||
1281 	    (args.pa_flags & FLAG_DO_TOP)) {
1282 		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1283 		    args.pa_interval * 1000, NULL);
1284 
1285 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1286 			err(EX_OSERR,
1287 			    "ERROR: Cannot register kevent for timer");
1288 	}
1289 
1290 	/*
1291 	 * Setup a duration timer if we have sampling mode PMCs and
1292 	 * a duration time is set
1293 	 */
1294 	if ((args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
1295 	    (args.pa_flags & FLAG_HAS_DURATION)) {
1296 		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1297 		    args.pa_duration * 1000, NULL);
1298 
1299 		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1300 			err(EX_OSERR, "ERROR: Cannot register kevent for "
1301 			    "time duration");
1302 	}
1303 
1304 	/* attach PMCs to the target process, starting it if specified */
1305 	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1306 		pmcstat_create_process(pmcstat_sockpair, &args, pmcstat_kq);
1307 
1308 	if (check_driver_stats && pmc_get_driver_stats(&ds_start) < 0)
1309 		err(EX_OSERR, "ERROR: Cannot retrieve driver statistics");
1310 
1311 	/* Attach process pmcs to the target process. */
1312 	if (args.pa_flags & (FLAG_HAS_TARGET | FLAG_HAS_COMMANDLINE)) {
1313 		if (SLIST_EMPTY(&args.pa_targets))
1314 			errx(EX_DATAERR,
1315 			    "ERROR: No matching target processes.");
1316 		if (args.pa_flags & FLAG_HAS_PROCESS_PMCS)
1317 			pmcstat_attach_pmcs(&args);
1318 
1319 		if (pmcstat_kvm) {
1320 			kvm_close(pmcstat_kvm);
1321 			pmcstat_kvm = NULL;
1322 		}
1323 	}
1324 
1325 	/* start the pmcs */
1326 	pmcstat_start_pmcs();
1327 
1328 	/* start the (commandline) process if needed */
1329 	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1330 		pmcstat_start_process(pmcstat_sockpair);
1331 
1332 	/* initialize logging */
1333 	pmcstat_log_initialize_logging();
1334 
1335 	/* Handle SIGINT using the kqueue loop */
1336 	sa.sa_handler = SIG_IGN;
1337 	sa.sa_flags   = 0;
1338 	(void) sigemptyset(&sa.sa_mask);
1339 
1340 	if (sigaction(SIGINT, &sa, NULL) < 0)
1341 		err(EX_OSERR, "ERROR: Cannot install signal handler");
1342 
1343 	/*
1344 	 * Setup the top mode display.
1345 	 */
1346 	if (args.pa_flags & FLAG_DO_TOP) {
1347 		args.pa_flags &= ~FLAG_DO_PRINT;
1348 
1349 		if (args.pa_toptty) {
1350 			/*
1351 			 * Init ncurses.
1352 			 */
1353 			initscr();
1354 			if(has_colors() == TRUE) {
1355 				args.pa_topcolor = 1;
1356 				start_color();
1357 				use_default_colors();
1358 				pair_content(0, &cf, &cb);
1359 				init_pair(1, COLOR_RED, cb);
1360 				init_pair(2, COLOR_YELLOW, cb);
1361 				init_pair(3, COLOR_GREEN, cb);
1362 			}
1363 			cbreak();
1364 			noecho();
1365 			nonl();
1366 			nodelay(stdscr, 1);
1367 			intrflush(stdscr, FALSE);
1368 			keypad(stdscr, TRUE);
1369 			clear();
1370 			/* Get terminal width / height with ncurses. */
1371 			getmaxyx(stdscr,
1372 			    pmcstat_displayheight, pmcstat_displaywidth);
1373 			pmcstat_displayheight--; pmcstat_displaywidth--;
1374 			atexit(pmcstat_topexit);
1375 		}
1376 	}
1377 
1378 	/*
1379 	 * loop till either the target process (if any) exits, or we
1380 	 * are killed by a SIGINT or we reached the time duration.
1381 	 */
1382 	runstate = PMCSTAT_RUNNING;
1383 	do_print = do_read = 0;
1384 	do {
1385 		if ((c = kevent(pmcstat_kq, NULL, 0, &kev, 1, NULL)) <= 0) {
1386 			if (errno != EINTR)
1387 				err(EX_OSERR, "ERROR: kevent failed");
1388 			else
1389 				continue;
1390 		}
1391 
1392 		if (kev.flags & EV_ERROR)
1393 			errc(EX_OSERR, kev.data, "ERROR: kevent failed");
1394 
1395 		switch (kev.filter) {
1396 		case EVFILT_PROC:  /* target has exited */
1397 			runstate = pmcstat_close_log(&args);
1398 			do_print = 1;
1399 			break;
1400 
1401 		case EVFILT_READ:  /* log file data is present */
1402 			if (kev.ident == (unsigned)fileno(stdin) &&
1403 			    (args.pa_flags & FLAG_DO_TOP)) {
1404 				if (pmcstat_keypress_log())
1405 					runstate = pmcstat_close_log(&args);
1406 			} else {
1407 				do_read = 0;
1408 				runstate = pmcstat_process_log();
1409 			}
1410 			break;
1411 
1412 		case EVFILT_SIGNAL:
1413 			if (kev.ident == SIGCHLD) {
1414 				/*
1415 				 * The child process sends us a
1416 				 * SIGCHLD if its exec() failed.  We
1417 				 * wait for it to exit and then exit
1418 				 * ourselves.
1419 				 */
1420 				(void) wait(&c);
1421 				runstate = PMCSTAT_FINISHED;
1422 			} else if (kev.ident == SIGIO) {
1423 				/*
1424 				 * We get a SIGIO if a PMC loses all
1425 				 * of its targets, or if logfile
1426 				 * writes encounter an error.
1427 				 */
1428 				runstate = pmcstat_close_log(&args);
1429 				do_print = 1; /* print PMCs at exit */
1430 			} else if (kev.ident == SIGINT) {
1431 				/* Kill the child process if we started it */
1432 				if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1433 					pmcstat_kill_process();
1434 				runstate = pmcstat_close_log(&args);
1435 			} else if (kev.ident == SIGWINCH) {
1436 				if (ioctl(fileno(args.pa_printfile),
1437 					TIOCGWINSZ, &ws) < 0)
1438 				    err(EX_OSERR,
1439 				        "ERROR: Cannot determine window size");
1440 				pmcstat_displayheight = ws.ws_row - 1;
1441 				pmcstat_displaywidth  = ws.ws_col - 1;
1442 			} else
1443 				assert(0);
1444 
1445 			break;
1446 
1447 		case EVFILT_TIMER:
1448 			/* time duration reached, exit */
1449 			if (args.pa_flags & FLAG_HAS_DURATION) {
1450 				runstate = PMCSTAT_FINISHED;
1451 				break;
1452 			}
1453 			/* print out counting PMCs */
1454 			if ((args.pa_flags & FLAG_DO_TOP) &&
1455 			    (args.pa_flags & FLAG_HAS_PIPE) &&
1456 			     pmc_flush_logfile() == 0)
1457 				do_read = 1;
1458 			do_print = 1;
1459 			break;
1460 
1461 		}
1462 
1463 		if (do_print && !do_read) {
1464 			if ((args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) {
1465 				pmcstat_print_pmcs();
1466 				if (runstate == PMCSTAT_FINISHED &&
1467 				    /* final newline */
1468 				    (args.pa_flags & FLAG_DO_PRINT) == 0)
1469 					(void) fprintf(args.pa_printfile, "\n");
1470 			}
1471 			if (args.pa_flags & FLAG_DO_TOP)
1472 				pmcstat_display_log();
1473 			do_print = 0;
1474 		}
1475 
1476 	} while (runstate != PMCSTAT_FINISHED);
1477 
1478 	if ((args.pa_flags & FLAG_DO_TOP) && args.pa_toptty) {
1479 		pmcstat_topexit();
1480 		args.pa_toptty = 0;
1481 	}
1482 
1483 	/* flush any pending log entries */
1484 	if (args.pa_flags & (FLAG_HAS_OUTPUT_LOGFILE | FLAG_HAS_PIPE))
1485 		pmc_close_logfile();
1486 
1487 	pmcstat_cleanup();
1488 
1489 	/* check if the driver lost any samples or events */
1490 	if (check_driver_stats) {
1491 		if (pmc_get_driver_stats(&ds_end) < 0)
1492 			err(EX_OSERR,
1493 			    "ERROR: Cannot retrieve driver statistics");
1494 		if (ds_start.pm_intr_bufferfull != ds_end.pm_intr_bufferfull &&
1495 		    args.pa_verbosity > 0)
1496 			warnx(
1497 "WARNING: sampling was paused at least %u time%s.\n"
1498 "Please consider tuning the \"kern.hwpmc.nsamples\" tunable.",
1499 			    ds_end.pm_intr_bufferfull -
1500 			    ds_start.pm_intr_bufferfull,
1501 			    ((ds_end.pm_intr_bufferfull -
1502 			    ds_start.pm_intr_bufferfull) != 1) ? "s" : ""
1503 			    );
1504 		if (ds_start.pm_buffer_requests_failed !=
1505 		    ds_end.pm_buffer_requests_failed &&
1506 		    args.pa_verbosity > 0)
1507 			warnx(
1508 "WARNING: at least %u event%s were discarded while running.\n"
1509 "Please consider tuning the \"kern.hwpmc.nbuffers_pcpu\" tunable.",
1510 	 		    ds_end.pm_buffer_requests_failed -
1511 			    ds_start.pm_buffer_requests_failed,
1512 			    ((ds_end.pm_buffer_requests_failed -
1513 			    ds_start.pm_buffer_requests_failed) != 1) ? "s" : ""
1514 			    );
1515 	}
1516 
1517 	exit(EX_OK);
1518 }
1519