1/*-
2 * Copyright (c) 2003-2008, Joseph Koshy
3 * Copyright (c) 2007 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by A. Joseph Koshy under
7 * sponsorship from the FreeBSD Foundation and Google, Inc.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD$");
33
34#include <sys/param.h>
35#include <sys/cpuset.h>
36#include <sys/event.h>
37#include <sys/queue.h>
38#include <sys/socket.h>
39#include <sys/stat.h>
40#include <sys/sysctl.h>
41#include <sys/time.h>
42#include <sys/ttycom.h>
43#include <sys/user.h>
44#include <sys/wait.h>
45
46#include <assert.h>
47#include <curses.h>
48#include <err.h>
49#include <errno.h>
50#include <fcntl.h>
51#include <kvm.h>
52#include <libgen.h>
53#include <limits.h>
54#include <math.h>
55#include <pmc.h>
56#include <pmclog.h>
57#include <regex.h>
58#include <signal.h>
59#include <stdarg.h>
60#include <stdint.h>
61#include <stdio.h>
62#include <stdlib.h>
63#include <string.h>
64#include <sysexits.h>
65#include <unistd.h>
66
67#include "pmcstat.h"
68
69/*
70 * A given invocation of pmcstat(8) can manage multiple PMCs of both
71 * the system-wide and per-process variety.  Each of these could be in
72 * 'counting mode' or in 'sampling mode'.
73 *
74 * For 'counting mode' PMCs, pmcstat(8) will periodically issue a
75 * pmc_read() at the configured time interval and print out the value
76 * of the requested PMCs.
77 *
78 * For 'sampling mode' PMCs it can log to a file for offline analysis,
79 * or can analyse sampling data "on the fly", either by converting
80 * samples to printed textual form or by creating gprof(1) compatible
81 * profiles, one per program executed.  When creating gprof(1)
82 * profiles it can optionally merge entries from multiple processes
83 * for a given executable into a single profile file.
84 *
85 * pmcstat(8) can also execute a command line and attach PMCs to the
86 * resulting child process.  The protocol used is as follows:
87 *
88 * - parent creates a socketpair for two way communication and
89 *   fork()s.
90 * - subsequently:
91 *
92 *   /Parent/				/Child/
93 *
94 *   - Wait for childs token.
95 *					- Sends token.
96 *					- Awaits signal to start.
97 *  - Attaches PMCs to the child's pid
98 *    and starts them. Sets up
99 *    monitoring for the child.
100 *  - Signals child to start.
101 *					- Recieves signal, attempts exec().
102 *
103 * After this point normal processing can happen.
104 */
105
106/* Globals */
107
108int	pmcstat_interrupt = 0;
109int	pmcstat_displayheight = DEFAULT_DISPLAY_HEIGHT;
110int	pmcstat_displaywidth  = DEFAULT_DISPLAY_WIDTH;
111int	pmcstat_sockpair[NSOCKPAIRFD];
112int	pmcstat_kq;
113kvm_t	*pmcstat_kvm;
114struct kinfo_proc *pmcstat_plist;
115struct pmcstat_args args;
116
117static void
118pmcstat_clone_event_descriptor(struct pmcstat_ev *ev, const cpuset_t *cpumask)
119{
120	int cpu, mcpu;
121	struct pmcstat_ev *ev_clone;
122
123	mcpu = sizeof(*cpumask) * NBBY;
124	for (cpu = 0; cpu < mcpu; cpu++) {
125		if (!CPU_ISSET(cpu, cpumask))
126			continue;
127
128		if ((ev_clone = malloc(sizeof(*ev_clone))) == NULL)
129			errx(EX_SOFTWARE, "ERROR: Out of memory");
130		(void) memset(ev_clone, 0, sizeof(*ev_clone));
131
132		ev_clone->ev_count = ev->ev_count;
133		ev_clone->ev_cpu   = cpu;
134		ev_clone->ev_cumulative = ev->ev_cumulative;
135		ev_clone->ev_flags = ev->ev_flags;
136		ev_clone->ev_mode  = ev->ev_mode;
137		ev_clone->ev_name  = strdup(ev->ev_name);
138		ev_clone->ev_pmcid = ev->ev_pmcid;
139		ev_clone->ev_saved = ev->ev_saved;
140		ev_clone->ev_spec  = strdup(ev->ev_spec);
141
142		STAILQ_INSERT_TAIL(&args.pa_events, ev_clone, ev_next);
143	}
144}
145
146static void
147pmcstat_get_cpumask(const char *cpuspec, cpuset_t *cpumask)
148{
149	int cpu;
150	const char *s;
151	char *end;
152
153	CPU_ZERO(cpumask);
154	s = cpuspec;
155
156	do {
157		cpu = strtol(s, &end, 0);
158		if (cpu < 0 || end == s)
159			errx(EX_USAGE,
160			    "ERROR: Illegal CPU specification \"%s\".",
161			    cpuspec);
162		CPU_SET(cpu, cpumask);
163		s = end + strspn(end, ", \t");
164	} while (*s);
165}
166
167void
168pmcstat_attach_pmcs(void)
169{
170	struct pmcstat_ev *ev;
171	struct pmcstat_target *pt;
172	int count;
173
174	/* Attach all process PMCs to target processes. */
175	count = 0;
176	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
177		if (PMC_IS_SYSTEM_MODE(ev->ev_mode))
178			continue;
179		SLIST_FOREACH(pt, &args.pa_targets, pt_next)
180			if (pmc_attach(ev->ev_pmcid, pt->pt_pid) == 0)
181				count++;
182			else if (errno != ESRCH)
183				err(EX_OSERR,
184"ERROR: cannot attach pmc \"%s\" to process %d",
185				    ev->ev_name, (int)pt->pt_pid);
186	}
187
188	if (count == 0)
189		errx(EX_DATAERR, "ERROR: No processes were attached to.");
190}
191
192
193void
194pmcstat_cleanup(void)
195{
196	struct pmcstat_ev *ev, *tmp;
197
198	/* release allocated PMCs. */
199	STAILQ_FOREACH_SAFE(ev, &args.pa_events, ev_next, tmp)
200	    if (ev->ev_pmcid != PMC_ID_INVALID) {
201		if (pmc_stop(ev->ev_pmcid) < 0)
202			err(EX_OSERR, "ERROR: cannot stop pmc 0x%x \"%s\"",
203			    ev->ev_pmcid, ev->ev_name);
204		if (pmc_release(ev->ev_pmcid) < 0)
205			err(EX_OSERR, "ERROR: cannot release pmc 0x%x \"%s\"",
206			    ev->ev_pmcid, ev->ev_name);
207		free(ev->ev_name);
208		free(ev->ev_spec);
209		STAILQ_REMOVE(&args.pa_events, ev, pmcstat_ev, ev_next);
210		free(ev);
211	    }
212
213	/* de-configure the log file if present. */
214	if (args.pa_flags & (FLAG_HAS_PIPE | FLAG_HAS_OUTPUT_LOGFILE))
215		(void) pmc_configure_logfile(-1);
216
217	if (args.pa_logparser) {
218		pmclog_close(args.pa_logparser);
219		args.pa_logparser = NULL;
220	}
221
222	pmcstat_shutdown_logging();
223}
224
225void
226pmcstat_create_process(void)
227{
228	char token;
229	pid_t pid;
230	struct kevent kev;
231	struct pmcstat_target *pt;
232
233	if (socketpair(AF_UNIX, SOCK_STREAM, 0, pmcstat_sockpair) < 0)
234		err(EX_OSERR, "ERROR: cannot create socket pair");
235
236	switch (pid = fork()) {
237	case -1:
238		err(EX_OSERR, "ERROR: cannot fork");
239		/*NOTREACHED*/
240
241	case 0:		/* child */
242		(void) close(pmcstat_sockpair[PARENTSOCKET]);
243
244		/* Write a token to tell our parent we've started executing. */
245		if (write(pmcstat_sockpair[CHILDSOCKET], "+", 1) != 1)
246			err(EX_OSERR, "ERROR (child): cannot write token");
247
248		/* Wait for our parent to signal us to start. */
249		if (read(pmcstat_sockpair[CHILDSOCKET], &token, 1) < 0)
250			err(EX_OSERR, "ERROR (child): cannot read token");
251		(void) close(pmcstat_sockpair[CHILDSOCKET]);
252
253		/* exec() the program requested */
254		execvp(*args.pa_argv, args.pa_argv);
255		/* and if that fails, notify the parent */
256		kill(getppid(), SIGCHLD);
257		err(EX_OSERR, "ERROR: execvp \"%s\" failed", *args.pa_argv);
258		/*NOTREACHED*/
259
260	default:	/* parent */
261		(void) close(pmcstat_sockpair[CHILDSOCKET]);
262		break;
263	}
264
265	/* Ask to be notified via a kevent when the target process exits. */
266	EV_SET(&kev, pid, EVFILT_PROC, EV_ADD|EV_ONESHOT, NOTE_EXIT, 0,
267	    NULL);
268	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
269		err(EX_OSERR, "ERROR: cannot monitor child process %d", pid);
270
271	if ((pt = malloc(sizeof(*pt))) == NULL)
272		errx(EX_SOFTWARE, "ERROR: Out of memory.");
273
274	pt->pt_pid = pid;
275	SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
276
277	/* Wait for the child to signal that its ready to go. */
278	if (read(pmcstat_sockpair[PARENTSOCKET], &token, 1) < 0)
279		err(EX_OSERR, "ERROR (parent): cannot read token");
280
281	return;
282}
283
284void
285pmcstat_find_targets(const char *spec)
286{
287	int n, nproc, pid, rv;
288	struct pmcstat_target *pt;
289	char errbuf[_POSIX2_LINE_MAX], *end;
290	static struct kinfo_proc *kp;
291	regex_t reg;
292	regmatch_t regmatch;
293
294	/* First check if we've been given a process id. */
295      	pid = strtol(spec, &end, 0);
296	if (end != spec && pid >= 0) {
297		if ((pt = malloc(sizeof(*pt))) == NULL)
298			goto outofmemory;
299		pt->pt_pid = pid;
300		SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
301		return;
302	}
303
304	/* Otherwise treat arg as a regular expression naming processes. */
305	if (pmcstat_kvm == NULL) {
306		if ((pmcstat_kvm = kvm_openfiles(NULL, "/dev/null", NULL, 0,
307		    errbuf)) == NULL)
308			err(EX_OSERR, "ERROR: Cannot open kernel \"%s\"",
309			    errbuf);
310		if ((pmcstat_plist = kvm_getprocs(pmcstat_kvm, KERN_PROC_PROC,
311		    0, &nproc)) == NULL)
312			err(EX_OSERR, "ERROR: Cannot get process list: %s",
313			    kvm_geterr(pmcstat_kvm));
314	} else
315		nproc = 0;
316
317	if ((rv = regcomp(&reg, spec, REG_EXTENDED|REG_NOSUB)) != 0) {
318		regerror(rv, &reg, errbuf, sizeof(errbuf));
319		err(EX_DATAERR, "ERROR: Failed to compile regex \"%s\": %s",
320		    spec, errbuf);
321	}
322
323	for (n = 0, kp = pmcstat_plist; n < nproc; n++, kp++) {
324		if ((rv = regexec(&reg, kp->ki_comm, 1, &regmatch, 0)) == 0) {
325			if ((pt = malloc(sizeof(*pt))) == NULL)
326				goto outofmemory;
327			pt->pt_pid = kp->ki_pid;
328			SLIST_INSERT_HEAD(&args.pa_targets, pt, pt_next);
329		} else if (rv != REG_NOMATCH) {
330			regerror(rv, &reg, errbuf, sizeof(errbuf));
331			errx(EX_SOFTWARE, "ERROR: Regex evalation failed: %s",
332			    errbuf);
333		}
334	}
335
336	regfree(&reg);
337
338	return;
339
340 outofmemory:
341	errx(EX_SOFTWARE, "Out of memory.");
342	/*NOTREACHED*/
343}
344
345void
346pmcstat_kill_process(void)
347{
348	struct pmcstat_target *pt;
349
350	assert(args.pa_flags & FLAG_HAS_COMMANDLINE);
351
352	/*
353	 * If a command line was specified, it would be the very first
354	 * in the list, before any other processes specified by -t.
355	 */
356	pt = SLIST_FIRST(&args.pa_targets);
357	assert(pt != NULL);
358
359	if (kill(pt->pt_pid, SIGINT) != 0)
360		err(EX_OSERR, "ERROR: cannot signal child process");
361}
362
363void
364pmcstat_start_pmcs(void)
365{
366	struct pmcstat_ev *ev;
367
368	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
369
370	    assert(ev->ev_pmcid != PMC_ID_INVALID);
371
372	    if (pmc_start(ev->ev_pmcid) < 0) {
373	        warn("ERROR: Cannot start pmc 0x%x \"%s\"",
374		    ev->ev_pmcid, ev->ev_name);
375		pmcstat_cleanup();
376		exit(EX_OSERR);
377	    }
378	}
379
380}
381
382void
383pmcstat_print_headers(void)
384{
385	struct pmcstat_ev *ev;
386	int c, w;
387
388	(void) fprintf(args.pa_printfile, PRINT_HEADER_PREFIX);
389
390	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
391		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
392			continue;
393
394		c = PMC_IS_SYSTEM_MODE(ev->ev_mode) ? 's' : 'p';
395
396		if (ev->ev_fieldskip != 0)
397			(void) fprintf(args.pa_printfile, "%*s",
398			    ev->ev_fieldskip, "");
399		w = ev->ev_fieldwidth - ev->ev_fieldskip - 2;
400
401		if (c == 's')
402			(void) fprintf(args.pa_printfile, "s/%02d/%-*s ",
403			    ev->ev_cpu, w-3, ev->ev_name);
404		else
405			(void) fprintf(args.pa_printfile, "p/%*s ", w,
406			    ev->ev_name);
407	}
408
409	(void) fflush(args.pa_printfile);
410}
411
412void
413pmcstat_print_counters(void)
414{
415	int extra_width;
416	struct pmcstat_ev *ev;
417	pmc_value_t value;
418
419	extra_width = sizeof(PRINT_HEADER_PREFIX) - 1;
420
421	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
422
423		/* skip sampling mode counters */
424		if (PMC_IS_SAMPLING_MODE(ev->ev_mode))
425			continue;
426
427		if (pmc_read(ev->ev_pmcid, &value) < 0)
428			err(EX_OSERR, "ERROR: Cannot read pmc \"%s\"",
429			    ev->ev_name);
430
431		(void) fprintf(args.pa_printfile, "%*ju ",
432		    ev->ev_fieldwidth + extra_width,
433		    (uintmax_t) ev->ev_cumulative ? value :
434		    (value - ev->ev_saved));
435
436		if (ev->ev_cumulative == 0)
437			ev->ev_saved = value;
438		extra_width = 0;
439	}
440
441	(void) fflush(args.pa_printfile);
442}
443
444/*
445 * Print output
446 */
447
448void
449pmcstat_print_pmcs(void)
450{
451	static int linecount = 0;
452
453	/* check if we need to print a header line */
454	if (++linecount > pmcstat_displayheight) {
455		(void) fprintf(args.pa_printfile, "\n");
456		linecount = 1;
457	}
458	if (linecount == 1)
459		pmcstat_print_headers();
460	(void) fprintf(args.pa_printfile, "\n");
461
462	pmcstat_print_counters();
463
464	return;
465}
466
467/*
468 * Do process profiling
469 *
470 * If a pid was specified, attach each allocated PMC to the target
471 * process.  Otherwise, fork a child and attach the PMCs to the child,
472 * and have the child exec() the target program.
473 */
474
475void
476pmcstat_start_process(void)
477{
478	/* Signal the child to proceed. */
479	if (write(pmcstat_sockpair[PARENTSOCKET], "!", 1) != 1)
480		err(EX_OSERR, "ERROR (parent): write of token failed");
481
482	(void) close(pmcstat_sockpair[PARENTSOCKET]);
483}
484
485void
486pmcstat_show_usage(void)
487{
488	errx(EX_USAGE,
489	    "[options] [commandline]\n"
490	    "\t Measure process and/or system performance using hardware\n"
491	    "\t performance monitoring counters.\n"
492	    "\t Options include:\n"
493	    "\t -C\t\t (toggle) show cumulative counts\n"
494	    "\t -D path\t create profiles in directory \"path\"\n"
495	    "\t -E\t\t (toggle) show counts at process exit\n"
496	    "\t -F file\t write a system-wide callgraph (Kcachegrind format)"
497		" to \"file\"\n"
498	    "\t -G file\t write a system-wide callgraph to \"file\"\n"
499	    "\t -M file\t print executable/gmon file map to \"file\"\n"
500	    "\t -N\t\t (toggle) capture callchains\n"
501	    "\t -O file\t send log output to \"file\"\n"
502	    "\t -P spec\t allocate a process-private sampling PMC\n"
503	    "\t -R file\t read events from \"file\"\n"
504	    "\t -S spec\t allocate a system-wide sampling PMC\n"
505	    "\t -T\t\t start in top mode\n"
506	    "\t -W\t\t (toggle) show counts per context switch\n"
507	    "\t -c cpu-list\t set cpus for subsequent system-wide PMCs\n"
508	    "\t -d\t\t (toggle) track descendants\n"
509	    "\t -f spec\t pass \"spec\" to as plugin option\n"
510	    "\t -g\t\t produce gprof(1) compatible profiles\n"
511	    "\t -k dir\t\t set the path to the kernel\n"
512	    "\t -l secs\t set duration time\n"
513	    "\t -m file\t print sampled PCs to \"file\"\n"
514	    "\t -n rate\t set sampling rate\n"
515	    "\t -o file\t send print output to \"file\"\n"
516	    "\t -p spec\t allocate a process-private counting PMC\n"
517	    "\t -q\t\t suppress verbosity\n"
518	    "\t -r fsroot\t specify FS root directory\n"
519	    "\t -s spec\t allocate a system-wide counting PMC\n"
520	    "\t -t process-spec attach to running processes matching "
521		"\"process-spec\"\n"
522	    "\t -v\t\t increase verbosity\n"
523	    "\t -w secs\t set printing time interval\n"
524	    "\t -z depth\t limit callchain display depth"
525	);
526}
527
528/*
529 * At exit handler for top mode
530 */
531
532void
533pmcstat_topexit(void)
534{
535	if (!args.pa_toptty)
536		return;
537
538	/*
539	 * Shutdown ncurses.
540	 */
541	clrtoeol();
542	refresh();
543	endwin();
544}
545
546/*
547 * Main
548 */
549
550int
551main(int argc, char **argv)
552{
553	cpuset_t cpumask;
554	double interval;
555	double duration;
556	int hcpu, option, npmc, ncpu;
557	int c, check_driver_stats, current_cpu, current_sampling_count;
558	int do_callchain, do_descendants, do_logproccsw, do_logprocexit;
559	int do_print, do_read;
560	size_t dummy;
561	int graphdepth;
562	int pipefd[2], rfd;
563	int use_cumulative_counts;
564	short cf, cb;
565	char *end, *tmp;
566	const char *errmsg, *graphfilename;
567	enum pmcstat_state runstate;
568	struct pmc_driverstats ds_start, ds_end;
569	struct pmcstat_ev *ev;
570	struct sigaction sa;
571	struct kevent kev;
572	struct winsize ws;
573	struct stat sb;
574	char buffer[PATH_MAX];
575
576	check_driver_stats      = 0;
577	current_cpu 		= 0;
578	current_sampling_count  = DEFAULT_SAMPLE_COUNT;
579	do_callchain		= 1;
580	do_descendants          = 0;
581	do_logproccsw           = 0;
582	do_logprocexit          = 0;
583	use_cumulative_counts   = 0;
584	graphfilename		= "-";
585	args.pa_required	= 0;
586	args.pa_flags		= 0;
587	args.pa_verbosity	= 1;
588	args.pa_logfd		= -1;
589	args.pa_fsroot		= "";
590	args.pa_kernel		= strdup("/boot/kernel");
591	args.pa_samplesdir	= ".";
592	args.pa_printfile	= stderr;
593	args.pa_graphdepth	= DEFAULT_CALLGRAPH_DEPTH;
594	args.pa_graphfile	= NULL;
595	args.pa_interval	= DEFAULT_WAIT_INTERVAL;
596	args.pa_mapfilename	= NULL;
597	args.pa_inputpath	= NULL;
598	args.pa_outputpath	= NULL;
599	args.pa_pplugin		= PMCSTAT_PL_NONE;
600	args.pa_plugin		= PMCSTAT_PL_NONE;
601	args.pa_ctdumpinstr	= 1;
602	args.pa_topmode		= PMCSTAT_TOP_DELTA;
603	args.pa_toptty		= 0;
604	args.pa_topcolor	= 0;
605	args.pa_mergepmc	= 0;
606	args.pa_duration	= 0.0;
607	STAILQ_INIT(&args.pa_events);
608	SLIST_INIT(&args.pa_targets);
609	bzero(&ds_start, sizeof(ds_start));
610	bzero(&ds_end, sizeof(ds_end));
611	ev = NULL;
612	CPU_ZERO(&cpumask);
613
614	/*
615	 * The initial CPU mask specifies all non-halted CPUS in the
616	 * system.
617	 */
618	dummy = sizeof(int);
619	if (sysctlbyname("hw.ncpu", &ncpu, &dummy, NULL, 0) < 0)
620		err(EX_OSERR, "ERROR: Cannot determine the number of CPUs");
621	for (hcpu = 0; hcpu < ncpu; hcpu++)
622		CPU_SET(hcpu, &cpumask);
623
624	while ((option = getopt(argc, argv,
625	    "CD:EF:G:M:NO:P:R:S:TWc:df:gk:l:m:n:o:p:qr:s:t:vw:z:")) != -1)
626		switch (option) {
627		case 'C':	/* cumulative values */
628			use_cumulative_counts = !use_cumulative_counts;
629			args.pa_required |= FLAG_HAS_COUNTING_PMCS;
630			break;
631
632		case 'c':	/* CPU */
633
634			if (optarg[0] == '*' && optarg[1] == '\0') {
635				for (hcpu = 0; hcpu < ncpu; hcpu++)
636					CPU_SET(hcpu, &cpumask);
637			} else
638				pmcstat_get_cpumask(optarg, &cpumask);
639
640			args.pa_flags	 |= FLAGS_HAS_CPUMASK;
641			args.pa_required |= FLAG_HAS_SYSTEM_PMCS;
642			break;
643
644		case 'D':
645			if (stat(optarg, &sb) < 0)
646				err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
647				    optarg);
648			if (!S_ISDIR(sb.st_mode))
649				errx(EX_USAGE,
650				    "ERROR: \"%s\" is not a directory.",
651				    optarg);
652			args.pa_samplesdir = optarg;
653			args.pa_flags     |= FLAG_HAS_SAMPLESDIR;
654			args.pa_required  |= FLAG_DO_GPROF;
655			break;
656
657		case 'd':	/* toggle descendents */
658			do_descendants = !do_descendants;
659			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
660			break;
661
662		case 'F':	/* produce a system-wide calltree */
663			args.pa_flags |= FLAG_DO_CALLGRAPHS;
664			args.pa_plugin = PMCSTAT_PL_CALLTREE;
665			graphfilename = optarg;
666			break;
667
668		case 'f':	/* plugins options */
669			if (args.pa_plugin == PMCSTAT_PL_NONE)
670				err(EX_USAGE, "ERROR: Need -g/-G/-m/-T.");
671			pmcstat_pluginconfigure_log(optarg);
672			break;
673
674		case 'G':	/* produce a system-wide callgraph */
675			args.pa_flags |= FLAG_DO_CALLGRAPHS;
676			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
677			graphfilename = optarg;
678			break;
679
680		case 'g':	/* produce gprof compatible profiles */
681			args.pa_flags |= FLAG_DO_GPROF;
682			args.pa_pplugin = PMCSTAT_PL_CALLGRAPH;
683			args.pa_plugin	= PMCSTAT_PL_GPROF;
684			break;
685
686		case 'k':	/* pathname to the kernel */
687			free(args.pa_kernel);
688			args.pa_kernel = strdup(optarg);
689			args.pa_required |= FLAG_DO_ANALYSIS;
690			args.pa_flags    |= FLAG_HAS_KERNELPATH;
691			break;
692
693		case 'l':	/* time duration in seconds */
694			duration = strtod(optarg, &end);
695			if (*end != '\0' || duration <= 0)
696				errx(EX_USAGE, "ERROR: Illegal duration time "
697				    "value \"%s\".", optarg);
698			args.pa_flags |= FLAG_HAS_DURATION;
699			args.pa_duration = duration;
700			break;
701
702		case 'm':
703			args.pa_flags |= FLAG_DO_ANNOTATE;
704			args.pa_plugin = PMCSTAT_PL_ANNOTATE;
705			graphfilename  = optarg;
706			break;
707
708		case 'E':	/* log process exit */
709			do_logprocexit = !do_logprocexit;
710			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
711			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
712			break;
713
714		case 'M':	/* mapfile */
715			args.pa_mapfilename = optarg;
716			break;
717
718		case 'N':
719			do_callchain = !do_callchain;
720			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
721			break;
722
723		case 'p':	/* process virtual counting PMC */
724		case 's':	/* system-wide counting PMC */
725		case 'P':	/* process virtual sampling PMC */
726		case 'S':	/* system-wide sampling PMC */
727			if ((ev = malloc(sizeof(*ev))) == NULL)
728				errx(EX_SOFTWARE, "ERROR: Out of memory.");
729
730			switch (option) {
731			case 'p': ev->ev_mode = PMC_MODE_TC; break;
732			case 's': ev->ev_mode = PMC_MODE_SC; break;
733			case 'P': ev->ev_mode = PMC_MODE_TS; break;
734			case 'S': ev->ev_mode = PMC_MODE_SS; break;
735			}
736
737			if (option == 'P' || option == 'p') {
738				args.pa_flags |= FLAG_HAS_PROCESS_PMCS;
739				args.pa_required |= (FLAG_HAS_COMMANDLINE |
740				    FLAG_HAS_TARGET);
741			}
742
743			if (option == 'P' || option == 'S') {
744				args.pa_flags |= FLAG_HAS_SAMPLING_PMCS;
745				args.pa_required |= (FLAG_HAS_PIPE |
746				    FLAG_HAS_OUTPUT_LOGFILE);
747			}
748
749			if (option == 'p' || option == 's')
750				args.pa_flags |= FLAG_HAS_COUNTING_PMCS;
751
752			if (option == 's' || option == 'S')
753				args.pa_flags |= FLAG_HAS_SYSTEM_PMCS;
754
755			ev->ev_spec  = strdup(optarg);
756
757			if (option == 'S' || option == 'P')
758				ev->ev_count = current_sampling_count;
759			else
760				ev->ev_count = -1;
761
762			if (option == 'S' || option == 's') {
763				hcpu = sizeof(cpumask) * NBBY;
764				for (hcpu--; hcpu >= 0; hcpu--)
765					if (CPU_ISSET(hcpu, &cpumask))
766						break;
767				ev->ev_cpu = hcpu;
768			} else
769				ev->ev_cpu = PMC_CPU_ANY;
770
771			ev->ev_flags = 0;
772			if (do_callchain)
773				ev->ev_flags |= PMC_F_CALLCHAIN;
774			if (do_descendants)
775				ev->ev_flags |= PMC_F_DESCENDANTS;
776			if (do_logprocexit)
777				ev->ev_flags |= PMC_F_LOG_PROCEXIT;
778			if (do_logproccsw)
779				ev->ev_flags |= PMC_F_LOG_PROCCSW;
780
781			ev->ev_cumulative  = use_cumulative_counts;
782
783			ev->ev_saved = 0LL;
784			ev->ev_pmcid = PMC_ID_INVALID;
785
786			/* extract event name */
787			c = strcspn(optarg, ", \t");
788			ev->ev_name = malloc(c + 1);
789			(void) strncpy(ev->ev_name, optarg, c);
790			*(ev->ev_name + c) = '\0';
791
792			STAILQ_INSERT_TAIL(&args.pa_events, ev, ev_next);
793
794			if (option == 's' || option == 'S') {
795				hcpu = CPU_ISSET(ev->ev_cpu, &cpumask);
796				CPU_CLR(ev->ev_cpu, &cpumask);
797				pmcstat_clone_event_descriptor(ev, &cpumask);
798				if (hcpu != 0)
799					CPU_SET(ev->ev_cpu, &cpumask);
800			}
801
802			break;
803
804		case 'n':	/* sampling count */
805			current_sampling_count = strtol(optarg, &end, 0);
806			if (*end != '\0' || current_sampling_count <= 0)
807				errx(EX_USAGE,
808				    "ERROR: Illegal count value \"%s\".",
809				    optarg);
810			args.pa_required |= FLAG_HAS_SAMPLING_PMCS;
811			break;
812
813		case 'o':	/* outputfile */
814			if (args.pa_printfile != NULL &&
815			    args.pa_printfile != stdout &&
816			    args.pa_printfile != stderr)
817				(void) fclose(args.pa_printfile);
818			if ((args.pa_printfile = fopen(optarg, "w")) == NULL)
819				errx(EX_OSERR,
820				    "ERROR: cannot open \"%s\" for writing.",
821				    optarg);
822			args.pa_flags |= FLAG_DO_PRINT;
823			break;
824
825		case 'O':	/* sampling output */
826			if (args.pa_outputpath)
827				errx(EX_USAGE,
828"ERROR: option -O may only be specified once.");
829			args.pa_outputpath = optarg;
830			args.pa_flags |= FLAG_HAS_OUTPUT_LOGFILE;
831			break;
832
833		case 'q':	/* quiet mode */
834			args.pa_verbosity = 0;
835			break;
836
837		case 'r':	/* root FS path */
838			args.pa_fsroot = optarg;
839			break;
840
841		case 'R':	/* read an existing log file */
842			if (args.pa_inputpath != NULL)
843				errx(EX_USAGE,
844"ERROR: option -R may only be specified once.");
845			args.pa_inputpath = optarg;
846			if (args.pa_printfile == stderr)
847				args.pa_printfile = stdout;
848			args.pa_flags |= FLAG_READ_LOGFILE;
849			break;
850
851		case 't':	/* target pid or process name */
852			pmcstat_find_targets(optarg);
853
854			args.pa_flags |= FLAG_HAS_TARGET;
855			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
856			break;
857
858		case 'T':	/* top mode */
859			args.pa_flags |= FLAG_DO_TOP;
860			args.pa_plugin = PMCSTAT_PL_CALLGRAPH;
861			args.pa_ctdumpinstr = 0;
862			args.pa_mergepmc = 1;
863			if (args.pa_printfile == stderr)
864				args.pa_printfile = stdout;
865			break;
866
867		case 'v':	/* verbose */
868			args.pa_verbosity++;
869			break;
870
871		case 'w':	/* wait interval */
872			interval = strtod(optarg, &end);
873			if (*end != '\0' || interval <= 0)
874				errx(EX_USAGE,
875"ERROR: Illegal wait interval value \"%s\".",
876				    optarg);
877			args.pa_flags |= FLAG_HAS_WAIT_INTERVAL;
878			args.pa_interval = interval;
879			break;
880
881		case 'W':	/* toggle LOG_CSW */
882			do_logproccsw = !do_logproccsw;
883			args.pa_required |= (FLAG_HAS_PROCESS_PMCS |
884			    FLAG_HAS_COUNTING_PMCS | FLAG_HAS_OUTPUT_LOGFILE);
885			break;
886
887		case 'z':
888			graphdepth = strtod(optarg, &end);
889			if (*end != '\0' || graphdepth <= 0)
890				errx(EX_USAGE,
891				    "ERROR: Illegal callchain depth \"%s\".",
892				    optarg);
893			args.pa_graphdepth = graphdepth;
894			args.pa_required |= FLAG_DO_CALLGRAPHS;
895			break;
896
897		case '?':
898		default:
899			pmcstat_show_usage();
900			break;
901
902		}
903
904	args.pa_argc = (argc -= optind);
905	args.pa_argv = (argv += optind);
906
907	/* If we read from logfile and no specified CPU mask use
908	 * the maximum CPU count.
909	 */
910	if ((args.pa_flags & FLAG_READ_LOGFILE) &&
911	    (args.pa_flags & FLAGS_HAS_CPUMASK) == 0)
912		CPU_FILL(&cpumask);
913
914	args.pa_cpumask = cpumask; /* For selecting CPUs using -R. */
915
916	if (argc)	/* command line present */
917		args.pa_flags |= FLAG_HAS_COMMANDLINE;
918
919	if (args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS |
920	    FLAG_DO_ANNOTATE | FLAG_DO_TOP))
921		args.pa_flags |= FLAG_DO_ANALYSIS;
922
923	/*
924	 * Check invocation syntax.
925	 */
926
927	/* disallow -O and -R together */
928	if (args.pa_outputpath && args.pa_inputpath)
929		errx(EX_USAGE,
930		    "ERROR: options -O and -R are mutually exclusive.");
931
932	/* disallow -T and -l together */
933	if ((args.pa_flags & FLAG_HAS_DURATION) &&
934	    (args.pa_flags & FLAG_DO_TOP))
935		errx(EX_USAGE, "ERROR: options -T and -l are mutually "
936		    "exclusive.");
937
938	/* -m option is allowed with -R only. */
939	if (args.pa_flags & FLAG_DO_ANNOTATE && args.pa_inputpath == NULL)
940		errx(EX_USAGE, "ERROR: option -m requires an input file");
941
942	/* -m option is not allowed combined with -g or -G. */
943	if (args.pa_flags & FLAG_DO_ANNOTATE &&
944	    args.pa_flags & (FLAG_DO_GPROF | FLAG_DO_CALLGRAPHS))
945		errx(EX_USAGE,
946		    "ERROR: option -m and -g | -G are mutually exclusive");
947
948	if (args.pa_flags & FLAG_READ_LOGFILE) {
949		errmsg = NULL;
950		if (args.pa_flags & FLAG_HAS_COMMANDLINE)
951			errmsg = "a command line specification";
952		else if (args.pa_flags & FLAG_HAS_TARGET)
953			errmsg = "option -t";
954		else if (!STAILQ_EMPTY(&args.pa_events))
955			errmsg = "a PMC event specification";
956		if (errmsg)
957			errx(EX_USAGE,
958			    "ERROR: option -R may not be used with %s.",
959			    errmsg);
960	} else if (STAILQ_EMPTY(&args.pa_events))
961		/* All other uses require a PMC spec. */
962		pmcstat_show_usage();
963
964	/* check for -t pid without a process PMC spec */
965	if ((args.pa_required & FLAG_HAS_TARGET) &&
966	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
967		errx(EX_USAGE,
968"ERROR: option -t requires a process mode PMC to be specified."
969		    );
970
971	/* check for process-mode options without a command or -t pid */
972	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
973	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
974		errx(EX_USAGE,
975"ERROR: options -d, -E, -p, -P, and -W require a command line or target process."
976		    );
977
978	/* check for -p | -P without a target process of some sort */
979	if ((args.pa_required & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) &&
980	    (args.pa_flags & (FLAG_HAS_COMMANDLINE | FLAG_HAS_TARGET)) == 0)
981		errx(EX_USAGE,
982"ERROR: options -P and -p require a target process or a command line."
983		    );
984
985	/* check for process-mode options without a process-mode PMC */
986	if ((args.pa_required & FLAG_HAS_PROCESS_PMCS) &&
987	    (args.pa_flags & FLAG_HAS_PROCESS_PMCS) == 0)
988		errx(EX_USAGE,
989"ERROR: options -d, -E, and -W require a process mode PMC to be specified."
990		    );
991
992	/* check for -c cpu with no system mode PMCs or logfile. */
993	if ((args.pa_required & FLAG_HAS_SYSTEM_PMCS) &&
994	    (args.pa_flags & FLAG_HAS_SYSTEM_PMCS) == 0 &&
995	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
996		errx(EX_USAGE,
997"ERROR: option -c requires at least one system mode PMC to be specified."
998		    );
999
1000	/* check for counting mode options without a counting PMC */
1001	if ((args.pa_required & FLAG_HAS_COUNTING_PMCS) &&
1002	    (args.pa_flags & FLAG_HAS_COUNTING_PMCS) == 0)
1003		errx(EX_USAGE,
1004"ERROR: options -C, -W and -o require at least one counting mode PMC to be specified."
1005		    );
1006
1007	/* check for sampling mode options without a sampling PMC spec */
1008	if ((args.pa_required & FLAG_HAS_SAMPLING_PMCS) &&
1009	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) == 0)
1010		errx(EX_USAGE,
1011"ERROR: options -N, -n and -O require at least one sampling mode PMC to be specified."
1012		    );
1013
1014	/* check if -g/-G/-m/-T are being used correctly */
1015	if ((args.pa_flags & FLAG_DO_ANALYSIS) &&
1016	    !(args.pa_flags & (FLAG_HAS_SAMPLING_PMCS|FLAG_READ_LOGFILE)))
1017		errx(EX_USAGE,
1018"ERROR: options -g/-G/-m/-T require sampling PMCs or -R to be specified."
1019		    );
1020
1021	/* check if -O was spuriously specified */
1022	if ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) &&
1023	    (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0)
1024		errx(EX_USAGE,
1025"ERROR: option -O is used only with options -E, -P, -S and -W."
1026		    );
1027
1028	/* -k kernel path require -g/-G/-m/-T or -R */
1029	if ((args.pa_flags & FLAG_HAS_KERNELPATH) &&
1030	    (args.pa_flags & FLAG_DO_ANALYSIS) == 0 &&
1031	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
1032	    errx(EX_USAGE, "ERROR: option -k is only used with -g/-R/-m/-T.");
1033
1034	/* -D only applies to gprof output mode (-g) */
1035	if ((args.pa_flags & FLAG_HAS_SAMPLESDIR) &&
1036	    (args.pa_flags & FLAG_DO_GPROF) == 0)
1037	    errx(EX_USAGE, "ERROR: option -D is only used with -g.");
1038
1039	/* -M mapfile requires -g or -R */
1040	if (args.pa_mapfilename != NULL &&
1041	    (args.pa_flags & FLAG_DO_GPROF) == 0 &&
1042	    (args.pa_flags & FLAG_READ_LOGFILE) == 0)
1043	    errx(EX_USAGE, "ERROR: option -M is only used with -g/-R.");
1044
1045	/*
1046	 * Disallow textual output of sampling PMCs if counting PMCs
1047	 * have also been asked for, mostly because the combined output
1048	 * is difficult to make sense of.
1049	 */
1050	if ((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1051	    (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
1052	    ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) == 0))
1053		errx(EX_USAGE,
1054"ERROR: option -O is required if counting and sampling PMCs are specified together."
1055		    );
1056
1057	/*
1058	 * Check if "-k kerneldir" was specified, and if whether
1059	 * 'kerneldir' actually refers to a file.  If so, use
1060	 * `dirname path` to determine the kernel directory.
1061	 */
1062	if (args.pa_flags & FLAG_HAS_KERNELPATH) {
1063		(void) snprintf(buffer, sizeof(buffer), "%s%s", args.pa_fsroot,
1064		    args.pa_kernel);
1065		if (stat(buffer, &sb) < 0)
1066			err(EX_OSERR, "ERROR: Cannot locate kernel \"%s\"",
1067			    buffer);
1068		if (!S_ISREG(sb.st_mode) && !S_ISDIR(sb.st_mode))
1069			errx(EX_USAGE, "ERROR: \"%s\": Unsupported file type.",
1070			    buffer);
1071		if (!S_ISDIR(sb.st_mode)) {
1072			tmp = args.pa_kernel;
1073			args.pa_kernel = strdup(dirname(args.pa_kernel));
1074			free(tmp);
1075			(void) snprintf(buffer, sizeof(buffer), "%s%s",
1076			    args.pa_fsroot, args.pa_kernel);
1077			if (stat(buffer, &sb) < 0)
1078				err(EX_OSERR, "ERROR: Cannot stat \"%s\"",
1079				    buffer);
1080			if (!S_ISDIR(sb.st_mode))
1081				errx(EX_USAGE,
1082				    "ERROR: \"%s\" is not a directory.",
1083				    buffer);
1084		}
1085	}
1086
1087	/*
1088	 * If we have a callgraph be created, select the outputfile.
1089	 */
1090	if (args.pa_flags & FLAG_DO_CALLGRAPHS) {
1091		if (strcmp(graphfilename, "-") == 0)
1092		    args.pa_graphfile = args.pa_printfile;
1093		else {
1094			args.pa_graphfile = fopen(graphfilename, "w");
1095			if (args.pa_graphfile == NULL)
1096				err(EX_OSERR,
1097				    "ERROR: cannot open \"%s\" for writing",
1098				    graphfilename);
1099		}
1100	}
1101	if (args.pa_flags & FLAG_DO_ANNOTATE) {
1102		args.pa_graphfile = fopen(graphfilename, "w");
1103		if (args.pa_graphfile == NULL)
1104			err(EX_OSERR, "ERROR: cannot open \"%s\" for writing",
1105			    graphfilename);
1106	}
1107
1108	/* if we've been asked to process a log file, skip init */
1109	if ((args.pa_flags & FLAG_READ_LOGFILE) == 0) {
1110		if (pmc_init() < 0)
1111			err(EX_UNAVAILABLE,
1112			    "ERROR: Initialization of the pmc(3) library failed"
1113			    );
1114
1115		if ((npmc = pmc_npmc(0)) < 0) /* assume all CPUs are identical */
1116			err(EX_OSERR,
1117"ERROR: Cannot determine the number of PMCs on CPU %d",
1118			    0);
1119	}
1120
1121	/* Allocate a kqueue */
1122	if ((pmcstat_kq = kqueue()) < 0)
1123		err(EX_OSERR, "ERROR: Cannot allocate kqueue");
1124
1125	/* Setup the logfile as the source. */
1126	if (args.pa_flags & FLAG_READ_LOGFILE) {
1127		/*
1128		 * Print the log in textual form if we haven't been
1129		 * asked to generate profiling information.
1130		 */
1131		if ((args.pa_flags & FLAG_DO_ANALYSIS) == 0)
1132			args.pa_flags |= FLAG_DO_PRINT;
1133
1134		pmcstat_initialize_logging();
1135		rfd = pmcstat_open_log(args.pa_inputpath,
1136		    PMCSTAT_OPEN_FOR_READ);
1137		if ((args.pa_logparser = pmclog_open(rfd)) == NULL)
1138			err(EX_OSERR, "ERROR: Cannot create parser");
1139		if (fcntl(rfd, F_SETFL, O_NONBLOCK) < 0)
1140			err(EX_OSERR, "ERROR: fcntl(2) failed");
1141		EV_SET(&kev, rfd, EVFILT_READ, EV_ADD,
1142		    0, 0, NULL);
1143		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1144			err(EX_OSERR, "ERROR: Cannot register kevent");
1145	}
1146	/*
1147	 * Configure the specified log file or setup a default log
1148	 * consumer via a pipe.
1149	 */
1150	if (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) {
1151		if (args.pa_outputpath)
1152			args.pa_logfd = pmcstat_open_log(args.pa_outputpath,
1153			    PMCSTAT_OPEN_FOR_WRITE);
1154		else {
1155			/*
1156			 * process the log on the fly by reading it in
1157			 * through a pipe.
1158			 */
1159			if (pipe(pipefd) < 0)
1160				err(EX_OSERR, "ERROR: pipe(2) failed");
1161
1162			if (fcntl(pipefd[READPIPEFD], F_SETFL, O_NONBLOCK) < 0)
1163				err(EX_OSERR, "ERROR: fcntl(2) failed");
1164
1165			EV_SET(&kev, pipefd[READPIPEFD], EVFILT_READ, EV_ADD,
1166			    0, 0, NULL);
1167
1168			if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1169				err(EX_OSERR, "ERROR: Cannot register kevent");
1170
1171			args.pa_logfd = pipefd[WRITEPIPEFD];
1172
1173			args.pa_flags |= FLAG_HAS_PIPE;
1174			if ((args.pa_flags & FLAG_DO_TOP) == 0)
1175				args.pa_flags |= FLAG_DO_PRINT;
1176			args.pa_logparser = pmclog_open(pipefd[READPIPEFD]);
1177		}
1178
1179		if (pmc_configure_logfile(args.pa_logfd) < 0)
1180			err(EX_OSERR, "ERROR: Cannot configure log file");
1181	}
1182
1183	/* remember to check for driver errors if we are sampling or logging */
1184	check_driver_stats = (args.pa_flags & FLAG_HAS_SAMPLING_PMCS) ||
1185	    (args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE);
1186
1187	/*
1188	if (args.pa_flags & FLAG_READ_LOGFILE) {
1189	 * Allocate PMCs.
1190	 */
1191
1192	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1193		if (pmc_allocate(ev->ev_spec, ev->ev_mode,
1194		    ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid) < 0)
1195			err(EX_OSERR,
1196"ERROR: Cannot allocate %s-mode pmc with specification \"%s\"",
1197			    PMC_IS_SYSTEM_MODE(ev->ev_mode) ?
1198			    "system" : "process", ev->ev_spec);
1199
1200		if (PMC_IS_SAMPLING_MODE(ev->ev_mode) &&
1201		    pmc_set(ev->ev_pmcid, ev->ev_count) < 0)
1202			err(EX_OSERR,
1203			    "ERROR: Cannot set sampling count for PMC \"%s\"",
1204			    ev->ev_name);
1205	}
1206
1207	/* compute printout widths */
1208	STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
1209		int counter_width;
1210		int display_width;
1211		int header_width;
1212
1213		(void) pmc_width(ev->ev_pmcid, &counter_width);
1214		header_width = strlen(ev->ev_name) + 2; /* prefix '%c/' */
1215		display_width = (int) floor(counter_width / 3.32193) + 1;
1216
1217		if (PMC_IS_SYSTEM_MODE(ev->ev_mode))
1218			header_width += 3; /* 2 digit CPU number + '/' */
1219
1220		if (header_width > display_width) {
1221			ev->ev_fieldskip = 0;
1222			ev->ev_fieldwidth = header_width;
1223		} else {
1224			ev->ev_fieldskip = display_width -
1225			    header_width;
1226			ev->ev_fieldwidth = display_width;
1227		}
1228	}
1229
1230	/*
1231	 * If our output is being set to a terminal, register a handler
1232	 * for window size changes.
1233	 */
1234
1235	if (isatty(fileno(args.pa_printfile))) {
1236
1237		if (ioctl(fileno(args.pa_printfile), TIOCGWINSZ, &ws) < 0)
1238			err(EX_OSERR, "ERROR: Cannot determine window size");
1239
1240		pmcstat_displayheight = ws.ws_row - 1;
1241		pmcstat_displaywidth  = ws.ws_col - 1;
1242
1243		EV_SET(&kev, SIGWINCH, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1244
1245		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1246			err(EX_OSERR,
1247			    "ERROR: Cannot register kevent for SIGWINCH");
1248
1249		args.pa_toptty = 1;
1250	}
1251
1252	/*
1253	 * Listen to key input in top mode.
1254	 */
1255	if (args.pa_flags & FLAG_DO_TOP) {
1256		EV_SET(&kev, fileno(stdin), EVFILT_READ, EV_ADD, 0, 0, NULL);
1257		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1258			err(EX_OSERR, "ERROR: Cannot register kevent");
1259	}
1260
1261	EV_SET(&kev, SIGINT, 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 SIGINT");
1264
1265	EV_SET(&kev, SIGIO, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1266	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1267		err(EX_OSERR, "ERROR: Cannot register kevent for SIGIO");
1268
1269	/*
1270	 * An exec() failure of a forked child is signalled by the
1271	 * child sending the parent a SIGCHLD.  We don't register an
1272	 * actual signal handler for SIGCHLD, but instead use our
1273	 * kqueue to pick up the signal.
1274	 */
1275	EV_SET(&kev, SIGCHLD, EVFILT_SIGNAL, EV_ADD, 0, 0, NULL);
1276	if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1277		err(EX_OSERR, "ERROR: Cannot register kevent for SIGCHLD");
1278
1279	/*
1280	 * Setup a timer if we have counting mode PMCs needing to be printed or
1281	 * top mode plugin is active.
1282	 */
1283	if (((args.pa_flags & FLAG_HAS_COUNTING_PMCS) &&
1284	     (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) ||
1285	    (args.pa_flags & FLAG_DO_TOP)) {
1286		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1287		    args.pa_interval * 1000, NULL);
1288
1289		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1290			err(EX_OSERR,
1291			    "ERROR: Cannot register kevent for timer");
1292	}
1293
1294	/*
1295	 * Setup a duration timer if we have sampling mode PMCs and
1296	 * a duration time is set
1297	 */
1298	if ((args.pa_flags & FLAG_HAS_SAMPLING_PMCS) &&
1299	    (args.pa_flags & FLAG_HAS_DURATION)) {
1300		EV_SET(&kev, 0, EVFILT_TIMER, EV_ADD, 0,
1301		    args.pa_duration * 1000, NULL);
1302
1303		if (kevent(pmcstat_kq, &kev, 1, NULL, 0, NULL) < 0)
1304			err(EX_OSERR, "ERROR: Cannot register kevent for "
1305			    "time duration");
1306	}
1307
1308	/* attach PMCs to the target process, starting it if specified */
1309	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1310		pmcstat_create_process();
1311
1312	if (check_driver_stats && pmc_get_driver_stats(&ds_start) < 0)
1313		err(EX_OSERR, "ERROR: Cannot retrieve driver statistics");
1314
1315	/* Attach process pmcs to the target process. */
1316	if (args.pa_flags & (FLAG_HAS_TARGET | FLAG_HAS_COMMANDLINE)) {
1317		if (SLIST_EMPTY(&args.pa_targets))
1318			errx(EX_DATAERR,
1319			    "ERROR: No matching target processes.");
1320		if (args.pa_flags & FLAG_HAS_PROCESS_PMCS)
1321			pmcstat_attach_pmcs();
1322
1323		if (pmcstat_kvm) {
1324			kvm_close(pmcstat_kvm);
1325			pmcstat_kvm = NULL;
1326		}
1327	}
1328
1329	/* start the pmcs */
1330	pmcstat_start_pmcs();
1331
1332	/* start the (commandline) process if needed */
1333	if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1334		pmcstat_start_process();
1335
1336	/* initialize logging */
1337	pmcstat_initialize_logging();
1338
1339	/* Handle SIGINT using the kqueue loop */
1340	sa.sa_handler = SIG_IGN;
1341	sa.sa_flags   = 0;
1342	(void) sigemptyset(&sa.sa_mask);
1343
1344	if (sigaction(SIGINT, &sa, NULL) < 0)
1345		err(EX_OSERR, "ERROR: Cannot install signal handler");
1346
1347	/*
1348	 * Setup the top mode display.
1349	 */
1350	if (args.pa_flags & FLAG_DO_TOP) {
1351		args.pa_flags &= ~FLAG_DO_PRINT;
1352
1353		if (args.pa_toptty) {
1354			/*
1355			 * Init ncurses.
1356			 */
1357			initscr();
1358			if(has_colors() == TRUE) {
1359				args.pa_topcolor = 1;
1360				start_color();
1361				use_default_colors();
1362				pair_content(0, &cf, &cb);
1363				init_pair(1, COLOR_RED, cb);
1364				init_pair(2, COLOR_YELLOW, cb);
1365				init_pair(3, COLOR_GREEN, cb);
1366			}
1367			cbreak();
1368			noecho();
1369			nonl();
1370			nodelay(stdscr, 1);
1371			intrflush(stdscr, FALSE);
1372			keypad(stdscr, TRUE);
1373			clear();
1374			/* Get terminal width / height with ncurses. */
1375			getmaxyx(stdscr,
1376			    pmcstat_displayheight, pmcstat_displaywidth);
1377			pmcstat_displayheight--; pmcstat_displaywidth--;
1378			atexit(pmcstat_topexit);
1379		}
1380	}
1381
1382	/*
1383	 * loop till either the target process (if any) exits, or we
1384	 * are killed by a SIGINT or we reached the time duration.
1385	 */
1386	runstate = PMCSTAT_RUNNING;
1387	do_print = do_read = 0;
1388	do {
1389		if ((c = kevent(pmcstat_kq, NULL, 0, &kev, 1, NULL)) <= 0) {
1390			if (errno != EINTR)
1391				err(EX_OSERR, "ERROR: kevent failed");
1392			else
1393				continue;
1394		}
1395
1396		if (kev.flags & EV_ERROR)
1397			errc(EX_OSERR, kev.data, "ERROR: kevent failed");
1398
1399		switch (kev.filter) {
1400		case EVFILT_PROC:  /* target has exited */
1401			runstate = pmcstat_close_log();
1402			do_print = 1;
1403			break;
1404
1405		case EVFILT_READ:  /* log file data is present */
1406			if (kev.ident == (unsigned)fileno(stdin) &&
1407			    (args.pa_flags & FLAG_DO_TOP)) {
1408				if (pmcstat_keypress_log())
1409					runstate = pmcstat_close_log();
1410			} else {
1411				do_read = 0;
1412				runstate = pmcstat_process_log();
1413			}
1414			break;
1415
1416		case EVFILT_SIGNAL:
1417			if (kev.ident == SIGCHLD) {
1418				/*
1419				 * The child process sends us a
1420				 * SIGCHLD if its exec() failed.  We
1421				 * wait for it to exit and then exit
1422				 * ourselves.
1423				 */
1424				(void) wait(&c);
1425				runstate = PMCSTAT_FINISHED;
1426			} else if (kev.ident == SIGIO) {
1427				/*
1428				 * We get a SIGIO if a PMC loses all
1429				 * of its targets, or if logfile
1430				 * writes encounter an error.
1431				 */
1432				runstate = pmcstat_close_log();
1433				do_print = 1; /* print PMCs at exit */
1434			} else if (kev.ident == SIGINT) {
1435				/* Kill the child process if we started it */
1436				if (args.pa_flags & FLAG_HAS_COMMANDLINE)
1437					pmcstat_kill_process();
1438				runstate = pmcstat_close_log();
1439			} else if (kev.ident == SIGWINCH) {
1440				if (ioctl(fileno(args.pa_printfile),
1441					TIOCGWINSZ, &ws) < 0)
1442				    err(EX_OSERR,
1443				        "ERROR: Cannot determine window size");
1444				pmcstat_displayheight = ws.ws_row - 1;
1445				pmcstat_displaywidth  = ws.ws_col - 1;
1446			} else
1447				assert(0);
1448
1449			break;
1450
1451		case EVFILT_TIMER:
1452			/* time duration reached, exit */
1453			if (args.pa_flags & FLAG_HAS_DURATION) {
1454				runstate = PMCSTAT_FINISHED;
1455				break;
1456			}
1457			/* print out counting PMCs */
1458			if ((args.pa_flags & FLAG_DO_TOP) &&
1459			     pmc_flush_logfile() == 0)
1460				do_read = 1;
1461			do_print = 1;
1462			break;
1463
1464		}
1465
1466		if (do_print && !do_read) {
1467			if ((args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0) {
1468				pmcstat_print_pmcs();
1469				if (runstate == PMCSTAT_FINISHED &&
1470				    /* final newline */
1471				    (args.pa_flags & FLAG_DO_PRINT) == 0)
1472					(void) fprintf(args.pa_printfile, "\n");
1473			}
1474			if (args.pa_flags & FLAG_DO_TOP)
1475				pmcstat_display_log();
1476			do_print = 0;
1477		}
1478
1479	} while (runstate != PMCSTAT_FINISHED);
1480
1481	if ((args.pa_flags & FLAG_DO_TOP) && args.pa_toptty) {
1482		pmcstat_topexit();
1483		args.pa_toptty = 0;
1484	}
1485
1486	/* flush any pending log entries */
1487	if (args.pa_flags & (FLAG_HAS_OUTPUT_LOGFILE | FLAG_HAS_PIPE))
1488		pmc_close_logfile();
1489
1490	pmcstat_cleanup();
1491
1492	free(args.pa_kernel);
1493
1494	/* check if the driver lost any samples or events */
1495	if (check_driver_stats) {
1496		if (pmc_get_driver_stats(&ds_end) < 0)
1497			err(EX_OSERR,
1498			    "ERROR: Cannot retrieve driver statistics");
1499		if (ds_start.pm_intr_bufferfull != ds_end.pm_intr_bufferfull &&
1500		    args.pa_verbosity > 0)
1501			warnx("WARNING: some samples were dropped.\n"
1502"Please consider tuning the \"kern.hwpmc.nsamples\" tunable."
1503			    );
1504		if (ds_start.pm_buffer_requests_failed !=
1505		    ds_end.pm_buffer_requests_failed &&
1506		    args.pa_verbosity > 0)
1507			warnx("WARNING: some events were discarded.\n"
1508"Please consider tuning the \"kern.hwpmc.nbuffers\" tunable."
1509			    );
1510	}
1511
1512	exit(EX_OK);
1513}
1514