xref: /freebsd/usr.bin/top/machine.c (revision 2282756519d9337dec783509a432fd6ddfd06683)
1 /*
2  * top - a top users display for Unix
3  *
4  * DESCRIPTION:
5  * Originally written for BSD4.4 system by Christos Zoulas.
6  * Ported to FreeBSD 2.x by Steven Wallace && Wolfram Schneider
7  * Order support hacked in from top-3.5beta6/machine/m_aix41.c
8  *   by Monte Mitzelfelt (for latest top see http://www.groupsys.com/topinfo/)
9  *
10  * AUTHOR:  Christos Zoulas <christos@ee.cornell.edu>
11  *          Steven Wallace  <swallace@FreeBSD.org>
12  *          Wolfram Schneider <wosch@FreeBSD.org>
13  *          Thomas Moestl <tmoestl@gmx.net>
14  *          Eitan Adler <eadler@FreeBSD.org>
15  *
16  * $FreeBSD$
17  */
18 
19 #include <sys/errno.h>
20 #include <sys/fcntl.h>
21 #include <sys/param.h>
22 #include <sys/priority.h>
23 #include <sys/proc.h>
24 #include <sys/resource.h>
25 #include <sys/sbuf.h>
26 #include <sys/sysctl.h>
27 #include <sys/time.h>
28 #include <sys/user.h>
29 
30 #include <assert.h>
31 #include <err.h>
32 #include <libgen.h>
33 #include <kvm.h>
34 #include <math.h>
35 #include <paths.h>
36 #include <stdio.h>
37 #include <stdbool.h>
38 #include <stdint.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <time.h>
42 #include <unistd.h>
43 #include <vis.h>
44 
45 #include "top.h"
46 #include "display.h"
47 #include "machine.h"
48 #include "loadavg.h"
49 #include "screen.h"
50 #include "utils.h"
51 #include "layout.h"
52 
53 #define GETSYSCTL(name, var) getsysctl(name, &(var), sizeof(var))
54 
55 extern struct timeval timeout;
56 static int smpmode;
57 enum displaymodes displaymode;
58 static const int namelength = 10;
59 /* TOP_JID_LEN based on max of 999999 */
60 #define TOP_JID_LEN 6
61 #define TOP_SWAP_LEN 5
62 
63 /* get_process_info passes back a handle.  This is what it looks like: */
64 
65 struct handle {
66 	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
67 	int remaining;			/* number of pointers remaining */
68 };
69 
70 
71 /* define what weighted cpu is.  */
72 #define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
73 			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
74 
75 /* what we consider to be process size: */
76 #define PROCSIZE(pp) ((pp)->ki_size / 1024)
77 
78 #define RU(pp)	(&(pp)->ki_rusage)
79 
80 #define	PCTCPU(pp) (pcpu[pp - pbase])
81 
82 /* process state names for the "STATE" column of the display */
83 /* the extra nulls in the string "run" are for adding a slash and
84    the processor number when needed */
85 
86 static const char *state_abbrev[] = {
87 	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
88 };
89 
90 
91 static kvm_t *kd;
92 
93 /* values that we stash away in _init and use in later routines */
94 
95 static double logcpu;
96 
97 /* these are retrieved from the kernel in _init */
98 
99 static load_avg  ccpu;
100 
101 /* these are used in the get_ functions */
102 
103 static int lastpid;
104 
105 /* these are for calculating cpu state percentages */
106 
107 static long cp_time[CPUSTATES];
108 static long cp_old[CPUSTATES];
109 static long cp_diff[CPUSTATES];
110 
111 /* these are for detailing the process states */
112 
113 static const char *procstatenames[] = {
114 	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
115 	" zombie, ", " waiting, ", " lock, ",
116 	NULL
117 };
118 static int process_states[nitems(procstatenames)];
119 
120 /* these are for detailing the cpu states */
121 
122 static int cpu_states[CPUSTATES];
123 static const char *cpustatenames[] = {
124 	"user", "nice", "system", "interrupt", "idle", NULL
125 };
126 
127 /* these are for detailing the memory statistics */
128 
129 static const char *memorynames[] = {
130 	"K Active, ", "K Inact, ", "K Laundry, ", "K Wired, ", "K Buf, ",
131 	"K Free", NULL
132 };
133 static int memory_stats[nitems(memorynames)];
134 
135 static const char *arcnames[] = {
136 	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
137 	NULL
138 };
139 static int arc_stats[nitems(arcnames)];
140 
141 static const char *carcnames[] = {
142 	"K Compressed, ", "K Uncompressed, ", ":1 Ratio, ",
143 	NULL
144 };
145 static int carc_stats[nitems(carcnames)];
146 
147 static const char *swapnames[] = {
148 	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
149 	NULL
150 };
151 static int swap_stats[nitems(swapnames)];
152 
153 
154 /* these are for keeping track of the proc array */
155 
156 static int nproc;
157 static int onproc = -1;
158 static int pref_len;
159 static struct kinfo_proc *pbase;
160 static struct kinfo_proc **pref;
161 static struct kinfo_proc *previous_procs;
162 static struct kinfo_proc **previous_pref;
163 static int previous_proc_count = 0;
164 static int previous_proc_count_max = 0;
165 static int previous_thread;
166 
167 /* data used for recalculating pctcpu */
168 static double *pcpu;
169 static struct timespec proc_uptime;
170 static struct timeval proc_wall_time;
171 static struct timeval previous_wall_time;
172 static uint64_t previous_interval = 0;
173 
174 /* total number of io operations */
175 static long total_inblock;
176 static long total_oublock;
177 static long total_majflt;
178 
179 /* these are for getting the memory statistics */
180 
181 static int arc_enabled;
182 static int carc_enabled;
183 static int pageshift;		/* log base 2 of the pagesize */
184 
185 /* define pagetok in terms of pageshift */
186 
187 #define pagetok(size) ((size) << pageshift)
188 
189 /* swap usage */
190 #define ki_swap(kip) \
191     ((kip)->ki_swrss > (kip)->ki_rssize ? (kip)->ki_swrss - (kip)->ki_rssize : 0)
192 
193 /*
194  * Sorting orders.  The first element is the default.
195  */
196 static const char *ordernames[] = {
197 	"cpu", "size", "res", "time", "pri", "threads",
198 	"total", "read", "write", "fault", "vcsw", "ivcsw",
199 	"jid", "swap", "pid", NULL
200 };
201 
202 /* Per-cpu time states */
203 static int maxcpu;
204 static int maxid;
205 static int ncpus;
206 static unsigned long cpumask;
207 static long *times;
208 static long *pcpu_cp_time;
209 static long *pcpu_cp_old;
210 static long *pcpu_cp_diff;
211 static int *pcpu_cpu_states;
212 
213 static int compare_swap(const void *a, const void *b);
214 static int compare_jid(const void *a, const void *b);
215 static int compare_pid(const void *a, const void *b);
216 static int compare_tid(const void *a, const void *b);
217 static const char *format_nice(const struct kinfo_proc *pp);
218 static void getsysctl(const char *name, void *ptr, size_t len);
219 static int swapmode(int *retavail, int *retfree);
220 static void update_layout(void);
221 static int find_uid(uid_t needle, int *haystack);
222 
223 static int
224 find_uid(uid_t needle, int *haystack)
225 {
226 	size_t i = 0;
227 
228 	for (; i < TOP_MAX_UIDS; ++i)
229 		if ((uid_t)haystack[i] == needle)
230 			return 1;
231 	return (0);
232 }
233 
234 void
235 toggle_pcpustats(void)
236 {
237 
238 	if (ncpus == 1)
239 		return;
240 	update_layout();
241 }
242 
243 /* Adjust display based on ncpus and the ARC state. */
244 static void
245 update_layout(void)
246 {
247 
248 	y_mem = 3;
249 	y_arc = 4;
250 	y_carc = 5;
251 	y_swap = 4 + arc_enabled + carc_enabled;
252 	y_idlecursor = 5 + arc_enabled + carc_enabled;
253 	y_message = 5 + arc_enabled + carc_enabled;
254 	y_header = 6 + arc_enabled + carc_enabled;
255 	y_procs = 7 + arc_enabled + carc_enabled;
256 	Header_lines = 7 + arc_enabled + carc_enabled;
257 
258 	if (pcpu_stats) {
259 		y_mem += ncpus - 1;
260 		y_arc += ncpus - 1;
261 		y_carc += ncpus - 1;
262 		y_swap += ncpus - 1;
263 		y_idlecursor += ncpus - 1;
264 		y_message += ncpus - 1;
265 		y_header += ncpus - 1;
266 		y_procs += ncpus - 1;
267 		Header_lines += ncpus - 1;
268 	}
269 }
270 
271 int
272 machine_init(struct statics *statics)
273 {
274 	int i, j, empty, pagesize;
275 	uint64_t arc_size;
276 	int carc_en;
277 	size_t size;
278 
279 	size = sizeof(smpmode);
280 	if ((sysctlbyname("machdep.smp_active", &smpmode, &size,
281 	    NULL, 0) != 0 &&
282 	    sysctlbyname("kern.smp.active", &smpmode, &size,
283 	    NULL, 0) != 0) ||
284 	    size != sizeof(smpmode))
285 		smpmode = 0;
286 
287 	size = sizeof(arc_size);
288 	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
289 	    NULL, 0) == 0 && arc_size != 0)
290 		arc_enabled = 1;
291 	size = sizeof(carc_en);
292 	if (arc_enabled &&
293 	    sysctlbyname("vfs.zfs.compressed_arc_enabled", &carc_en, &size,
294 	    NULL, 0) == 0 && carc_en == 1)
295 		carc_enabled = 1;
296 
297 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
298 	if (kd == NULL)
299 		return (-1);
300 
301 	GETSYSCTL("kern.ccpu", ccpu);
302 
303 	/* this is used in calculating WCPU -- calculate it ahead of time */
304 	logcpu = log(loaddouble(ccpu));
305 
306 	pbase = NULL;
307 	pref = NULL;
308 	pcpu = NULL;
309 	nproc = 0;
310 	onproc = -1;
311 
312 	/* get the page size and calculate pageshift from it */
313 	pagesize = getpagesize();
314 	pageshift = 0;
315 	while (pagesize > 1) {
316 		pageshift++;
317 		pagesize >>= 1;
318 	}
319 
320 	/* we only need the amount of log(2)1024 for our conversion */
321 	pageshift -= LOG1024;
322 
323 	/* fill in the statics information */
324 	statics->procstate_names = procstatenames;
325 	statics->cpustate_names = cpustatenames;
326 	statics->memory_names = memorynames;
327 	if (arc_enabled)
328 		statics->arc_names = arcnames;
329 	else
330 		statics->arc_names = NULL;
331 	if (carc_enabled)
332 		statics->carc_names = carcnames;
333 	else
334 		statics->carc_names = NULL;
335 	statics->swap_names = swapnames;
336 	statics->order_names = ordernames;
337 
338 	/* Allocate state for per-CPU stats. */
339 	cpumask = 0;
340 	ncpus = 0;
341 	GETSYSCTL("kern.smp.maxcpus", maxcpu);
342 	times = calloc(maxcpu * CPUSTATES, sizeof(long));
343 	if (times == NULL)
344 		err(1, "calloc for kern.smp.maxcpus");
345 	size = sizeof(long) * maxcpu * CPUSTATES;
346 	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
347 		err(1, "sysctlbyname kern.cp_times");
348 	pcpu_cp_time = calloc(1, size);
349 	maxid = (size / CPUSTATES / sizeof(long)) - 1;
350 	for (i = 0; i <= maxid; i++) {
351 		empty = 1;
352 		for (j = 0; empty && j < CPUSTATES; j++) {
353 			if (times[i * CPUSTATES + j] != 0)
354 				empty = 0;
355 		}
356 		if (!empty) {
357 			cpumask |= (1ul << i);
358 			ncpus++;
359 		}
360 	}
361 	assert(ncpus > 0);
362 	pcpu_cp_old = calloc(ncpus * CPUSTATES, sizeof(long));
363 	pcpu_cp_diff = calloc(ncpus * CPUSTATES, sizeof(long));
364 	pcpu_cpu_states = calloc(ncpus * CPUSTATES, sizeof(int));
365 	statics->ncpus = ncpus;
366 
367 	update_layout();
368 
369 	/* all done! */
370 	return (0);
371 }
372 
373 char *
374 format_header(const char *uname_field)
375 {
376 	static struct sbuf* header = NULL;
377 
378 	/* clean up from last time. */
379 	if (header != NULL) {
380 		sbuf_clear(header);
381 	} else {
382 		header = sbuf_new_auto();
383 	}
384 
385 	switch (displaymode) {
386 	case DISP_CPU: {
387 		sbuf_printf(header, "  %s", ps.thread_id ? " THR" : "PID");
388 		sbuf_printf(header, "%*s", ps.jail ? TOP_JID_LEN : 0,
389 									ps.jail ? " JID" : "");
390 		sbuf_printf(header, " %-*.*s  ", namelength, namelength, uname_field);
391 		if (!ps.thread) {
392 			sbuf_cat(header, "THR ");
393 		}
394 		sbuf_cat(header, "PRI NICE   SIZE    RES ");
395 		if (ps.swap) {
396 			sbuf_printf(header, "%*s ", TOP_SWAP_LEN - 1, "SWAP");
397 		}
398 		sbuf_cat(header, "STATE    ");
399 		if (smpmode) {
400 			sbuf_cat(header, "C   ");
401 		}
402 		sbuf_cat(header, "TIME ");
403 		sbuf_printf(header, " %6s ", ps.wcpu ? "WCPU" : "CPU");
404 		sbuf_cat(header, "COMMAND");
405 		sbuf_finish(header);
406 		break;
407 	}
408 	case DISP_IO: {
409 		sbuf_printf(header, "  %s%*s %-*.*s",
410 			ps.thread_id ? " THR" : "PID",
411 		    ps.jail ? TOP_JID_LEN : 0, ps.jail ? " JID" : "",
412 		    namelength, namelength, uname_field);
413 		sbuf_cat(header, "   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND");
414 		sbuf_finish(header);
415 		break;
416 	}
417 	case DISP_MAX:
418 		assert("displaymode must not be set to DISP_MAX");
419 	}
420 
421 	return sbuf_data(header);
422 }
423 
424 static int swappgsin = -1;
425 static int swappgsout = -1;
426 
427 
428 void
429 get_system_info(struct system_info *si)
430 {
431 	struct loadavg sysload;
432 	int mib[2];
433 	struct timeval boottime;
434 	uint64_t arc_stat, arc_stat2;
435 	int i, j;
436 	size_t size;
437 
438 	/* get the CPU stats */
439 	size = (maxid + 1) * CPUSTATES * sizeof(long);
440 	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
441 		err(1, "sysctlbyname kern.cp_times");
442 	GETSYSCTL("kern.cp_time", cp_time);
443 	GETSYSCTL("vm.loadavg", sysload);
444 	GETSYSCTL("kern.lastpid", lastpid);
445 
446 	/* convert load averages to doubles */
447 	for (i = 0; i < 3; i++)
448 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
449 
450 	/* convert cp_time counts to percentages */
451 	for (i = j = 0; i <= maxid; i++) {
452 		if ((cpumask & (1ul << i)) == 0)
453 			continue;
454 		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
455 		    &pcpu_cp_time[j * CPUSTATES],
456 		    &pcpu_cp_old[j * CPUSTATES],
457 		    &pcpu_cp_diff[j * CPUSTATES]);
458 		j++;
459 	}
460 	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
461 
462 	/* sum memory & swap statistics */
463 	{
464 		static unsigned int swap_delay = 0;
465 		static int swapavail = 0;
466 		static int swapfree = 0;
467 		static long bufspace = 0;
468 		static uint64_t nspgsin, nspgsout;
469 
470 		GETSYSCTL("vfs.bufspace", bufspace);
471 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
472 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
473 		GETSYSCTL("vm.stats.vm.v_laundry_count", memory_stats[2]);
474 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[3]);
475 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
476 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
477 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
478 		/* convert memory stats to Kbytes */
479 		memory_stats[0] = pagetok(memory_stats[0]);
480 		memory_stats[1] = pagetok(memory_stats[1]);
481 		memory_stats[2] = pagetok(memory_stats[2]);
482 		memory_stats[3] = pagetok(memory_stats[3]);
483 		memory_stats[4] = bufspace / 1024;
484 		memory_stats[5] = pagetok(memory_stats[5]);
485 		memory_stats[6] = -1;
486 
487 		/* first interval */
488 		if (swappgsin < 0) {
489 			swap_stats[4] = 0;
490 			swap_stats[5] = 0;
491 		}
492 
493 		/* compute differences between old and new swap statistic */
494 		else {
495 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
496 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
497 		}
498 
499 		swappgsin = nspgsin;
500 		swappgsout = nspgsout;
501 
502 		/* call CPU heavy swapmode() only for changes */
503 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
504 			swap_stats[3] = swapmode(&swapavail, &swapfree);
505 			swap_stats[0] = swapavail;
506 			swap_stats[1] = swapavail - swapfree;
507 			swap_stats[2] = swapfree;
508 		}
509 		swap_delay = 1;
510 		swap_stats[6] = -1;
511 	}
512 
513 	if (arc_enabled) {
514 		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
515 		arc_stats[0] = arc_stat >> 10;
516 		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
517 		arc_stats[1] = arc_stat >> 10;
518 		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
519 		arc_stats[2] = arc_stat >> 10;
520 		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
521 		arc_stats[3] = arc_stat >> 10;
522 		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
523 		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
524 		arc_stats[4] = (arc_stat + arc_stat2) >> 10;
525 		GETSYSCTL("kstat.zfs.misc.arcstats.other_size", arc_stat);
526 		arc_stats[5] = arc_stat >> 10;
527 		si->arc = arc_stats;
528 	}
529 	if (carc_enabled) {
530 		GETSYSCTL("kstat.zfs.misc.arcstats.compressed_size", arc_stat);
531 		carc_stats[0] = arc_stat >> 10;
532 		carc_stats[2] = arc_stat >> 10; /* For ratio */
533 		GETSYSCTL("kstat.zfs.misc.arcstats.uncompressed_size", arc_stat);
534 		carc_stats[1] = arc_stat >> 10;
535 		si->carc = carc_stats;
536 	}
537 
538 	/* set arrays and strings */
539 	if (pcpu_stats) {
540 		si->cpustates = pcpu_cpu_states;
541 		si->ncpus = ncpus;
542 	} else {
543 		si->cpustates = cpu_states;
544 		si->ncpus = 1;
545 	}
546 	si->memory = memory_stats;
547 	si->swap = swap_stats;
548 
549 
550 	if (lastpid > 0) {
551 		si->last_pid = lastpid;
552 	} else {
553 		si->last_pid = -1;
554 	}
555 
556 	/*
557 	 * Print how long system has been up.
558 	 * (Found by looking getting "boottime" from the kernel)
559 	 */
560 	mib[0] = CTL_KERN;
561 	mib[1] = KERN_BOOTTIME;
562 	size = sizeof(boottime);
563 	if (sysctl(mib, nitems(mib), &boottime, &size, NULL, 0) != -1 &&
564 	    boottime.tv_sec != 0) {
565 		si->boottime = boottime;
566 	} else {
567 		si->boottime.tv_sec = -1;
568 	}
569 }
570 
571 #define NOPROC	((void *)-1)
572 
573 /*
574  * We need to compare data from the old process entry with the new
575  * process entry.
576  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
577  * structure to cache the mapping.  We also use a negative cache pointer
578  * of NOPROC to avoid duplicate lookups.
579  * XXX: this could be done when the actual processes are fetched, we do
580  * it here out of laziness.
581  */
582 static const struct kinfo_proc *
583 get_old_proc(struct kinfo_proc *pp)
584 {
585 	const struct kinfo_proc * const *oldpp, *oldp;
586 
587 	/*
588 	 * If this is the first fetch of the kinfo_procs then we don't have
589 	 * any previous entries.
590 	 */
591 	if (previous_proc_count == 0)
592 		return (NULL);
593 	/* negative cache? */
594 	if (pp->ki_udata == NOPROC)
595 		return (NULL);
596 	/* cached? */
597 	if (pp->ki_udata != NULL)
598 		return (pp->ki_udata);
599 	/*
600 	 * Not cached,
601 	 * 1) look up based on pid.
602 	 * 2) compare process start.
603 	 * If we fail here, then setup a negative cache entry, otherwise
604 	 * cache it.
605 	 */
606 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
607 	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
608 	if (oldpp == NULL) {
609 		pp->ki_udata = NOPROC;
610 		return (NULL);
611 	}
612 	oldp = *oldpp;
613 	if (memcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
614 		pp->ki_udata = NOPROC;
615 		return (NULL);
616 	}
617 	pp->ki_udata = oldp;
618 	return (oldp);
619 }
620 
621 /*
622  * Return the total amount of IO done in blocks in/out and faults.
623  * store the values individually in the pointers passed in.
624  */
625 static long
626 get_io_stats(const struct kinfo_proc *pp, long *inp, long *oup, long *flp,
627     long *vcsw, long *ivcsw)
628 {
629 	const struct kinfo_proc *oldp;
630 	static struct kinfo_proc dummy;
631 	long ret;
632 
633 	oldp = get_old_proc(pp);
634 	if (oldp == NULL) {
635 		memset(&dummy, 0, sizeof(dummy));
636 		oldp = &dummy;
637 	}
638 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
639 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
640 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
641 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
642 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
643 	ret =
644 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
645 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
646 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
647 	return (ret);
648 }
649 
650 /*
651  * If there was a previous update, use the delta in ki_runtime over
652  * the previous interval to calculate pctcpu.  Otherwise, fall back
653  * to using the kernel's ki_pctcpu.
654  */
655 static double
656 proc_calc_pctcpu(struct kinfo_proc *pp)
657 {
658 	const struct kinfo_proc *oldp;
659 
660 	if (previous_interval != 0) {
661 		oldp = get_old_proc(pp);
662 		if (oldp != NULL)
663 			return ((double)(pp->ki_runtime - oldp->ki_runtime)
664 			    / previous_interval);
665 
666 		/*
667 		 * If this process/thread was created during the previous
668 		 * interval, charge it's total runtime to the previous
669 		 * interval.
670 		 */
671 		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
672 		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
673 		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
674 			return ((double)pp->ki_runtime / previous_interval);
675 	}
676 	return (pctdouble(pp->ki_pctcpu));
677 }
678 
679 /*
680  * Return true if this process has used any CPU time since the
681  * previous update.
682  */
683 static int
684 proc_used_cpu(struct kinfo_proc *pp)
685 {
686 	const struct kinfo_proc *oldp;
687 
688 	oldp = get_old_proc(pp);
689 	if (oldp == NULL)
690 		return (PCTCPU(pp) != 0);
691 	return (pp->ki_runtime != oldp->ki_runtime ||
692 	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
693 	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
694 }
695 
696 /*
697  * Return the total number of block in/out and faults by a process.
698  */
699 static long
700 get_io_total(const struct kinfo_proc *pp)
701 {
702 	long dummy;
703 
704 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
705 }
706 
707 static struct handle handle;
708 
709 void *
710 get_process_info(struct system_info *si, struct process_select *sel,
711     int (*compare)(const void *, const void *))
712 {
713 	int i;
714 	int total_procs;
715 	long p_io;
716 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
717 	long nsec;
718 	int active_procs;
719 	struct kinfo_proc **prefp;
720 	struct kinfo_proc *pp;
721 	struct timespec previous_proc_uptime;
722 
723 	/*
724 	 * If thread state was toggled, don't cache the previous processes.
725 	 */
726 	if (previous_thread != sel->thread)
727 		nproc = 0;
728 	previous_thread = sel->thread;
729 
730 	/*
731 	 * Save the previous process info.
732 	 */
733 	if (previous_proc_count_max < nproc) {
734 		free(previous_procs);
735 		previous_procs = calloc(nproc, sizeof(*previous_procs));
736 		free(previous_pref);
737 		previous_pref = calloc(nproc, sizeof(*previous_pref));
738 		if (previous_procs == NULL || previous_pref == NULL) {
739 			fprintf(stderr, "top: Out of memory.\n");
740 			quit(TOP_EX_SYS_ERROR);
741 		}
742 		previous_proc_count_max = nproc;
743 	}
744 	if (nproc) {
745 		for (i = 0; i < nproc; i++)
746 			previous_pref[i] = &previous_procs[i];
747 		memcpy(previous_procs, pbase, nproc * sizeof(*previous_procs));
748 		qsort(previous_pref, nproc, sizeof(*previous_pref),
749 		    ps.thread ? compare_tid : compare_pid);
750 	}
751 	previous_proc_count = nproc;
752 	previous_proc_uptime = proc_uptime;
753 	previous_wall_time = proc_wall_time;
754 	previous_interval = 0;
755 
756 	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
757 	    0, &nproc);
758 	gettimeofday(&proc_wall_time, NULL);
759 	if (clock_gettime(CLOCK_UPTIME, &proc_uptime) != 0)
760 		memset(&proc_uptime, 0, sizeof(proc_uptime));
761 	else if (previous_proc_uptime.tv_sec != 0 &&
762 	    previous_proc_uptime.tv_nsec != 0) {
763 		previous_interval = (proc_uptime.tv_sec -
764 		    previous_proc_uptime.tv_sec) * 1000000;
765 		nsec = proc_uptime.tv_nsec - previous_proc_uptime.tv_nsec;
766 		if (nsec < 0) {
767 			previous_interval -= 1000000;
768 			nsec += 1000000000;
769 		}
770 		previous_interval += nsec / 1000;
771 	}
772 	if (nproc > onproc) {
773 		pref = realloc(pref, sizeof(*pref) * nproc);
774 		pcpu = realloc(pcpu, sizeof(*pcpu) * nproc);
775 		onproc = nproc;
776 	}
777 	if (pref == NULL || pbase == NULL || pcpu == NULL) {
778 		fprintf(stderr, "top: Out of memory.\n");
779 		quit(TOP_EX_SYS_ERROR);
780 	}
781 	/* get a pointer to the states summary array */
782 	si->procstates = process_states;
783 
784 	/* count up process states and get pointers to interesting procs */
785 	total_procs = 0;
786 	active_procs = 0;
787 	total_inblock = 0;
788 	total_oublock = 0;
789 	total_majflt = 0;
790 	memset(process_states, 0, sizeof(process_states));
791 	prefp = pref;
792 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
793 
794 		if (pp->ki_stat == 0)
795 			/* not in use */
796 			continue;
797 
798 		if (!sel->self && pp->ki_pid == mypid && sel->pid == -1)
799 			/* skip self */
800 			continue;
801 
802 		if (!sel->system && (pp->ki_flag & P_SYSTEM) && sel->pid == -1)
803 			/* skip system process */
804 			continue;
805 
806 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
807 		    &p_vcsw, &p_ivcsw);
808 		total_inblock += p_inblock;
809 		total_oublock += p_oublock;
810 		total_majflt += p_majflt;
811 		total_procs++;
812 		process_states[(unsigned char)pp->ki_stat]++;
813 
814 		if (pp->ki_stat == SZOMB)
815 			/* skip zombies */
816 			continue;
817 
818 		if (!sel->kidle && pp->ki_tdflags & TDF_IDLETD && sel->pid == -1)
819 			/* skip kernel idle process */
820 			continue;
821 
822 		PCTCPU(pp) = proc_calc_pctcpu(pp);
823 		if (sel->thread && PCTCPU(pp) > 1.0)
824 			PCTCPU(pp) = 1.0;
825 		if (displaymode == DISP_CPU && !sel->idle &&
826 		    (!proc_used_cpu(pp) ||
827 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
828 			/* skip idle or non-running processes */
829 			continue;
830 
831 		if (displaymode == DISP_IO && !sel->idle && p_io == 0)
832 			/* skip processes that aren't doing I/O */
833 			continue;
834 
835 		if (sel->jid != -1 && pp->ki_jid != sel->jid)
836 			/* skip proc. that don't belong to the selected JID */
837 			continue;
838 
839 		if (sel->uid[0] != -1 && !find_uid(pp->ki_ruid, sel->uid))
840 			/* skip proc. that don't belong to the selected UID */
841 			continue;
842 
843 		if (sel->pid != -1 && pp->ki_pid != sel->pid)
844 			continue;
845 
846 		*prefp++ = pp;
847 		active_procs++;
848 	}
849 
850 	/* if requested, sort the "interesting" processes */
851 	if (compare != NULL)
852 		qsort(pref, active_procs, sizeof(*pref), compare);
853 
854 	/* remember active and total counts */
855 	si->p_total = total_procs;
856 	si->p_pactive = pref_len = active_procs;
857 
858 	/* pass back a handle */
859 	handle.next_proc = pref;
860 	handle.remaining = active_procs;
861 	return (&handle);
862 }
863 
864 char *
865 format_next_process(struct handle * xhandle, char *(*get_userid)(int), int flags)
866 {
867 	struct kinfo_proc *pp;
868 	const struct kinfo_proc *oldp;
869 	long cputime;
870 	char status[22];
871 	size_t state;
872 	struct rusage ru, *rup;
873 	long p_tot, s_tot;
874 	char *cmdbuf = NULL;
875 	char **args;
876 	static struct sbuf* procbuf = NULL;
877 
878 	/* clean up from last time. */
879 	if (procbuf != NULL) {
880 		sbuf_clear(procbuf);
881 	} else {
882 		procbuf = sbuf_new_auto();
883 	}
884 
885 
886 	/* find and remember the next proc structure */
887 	pp = *(xhandle->next_proc++);
888 	xhandle->remaining--;
889 
890 	/* get the process's command name */
891 	if ((pp->ki_flag & P_INMEM) == 0) {
892 		/*
893 		 * Print swapped processes as <pname>
894 		 */
895 		size_t len;
896 
897 		len = strlen(pp->ki_comm);
898 		if (len > sizeof(pp->ki_comm) - 3)
899 			len = sizeof(pp->ki_comm) - 3;
900 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
901 		pp->ki_comm[0] = '<';
902 		pp->ki_comm[len + 1] = '>';
903 		pp->ki_comm[len + 2] = '\0';
904 	}
905 
906 	/*
907 	 * Convert the process's runtime from microseconds to seconds.  This
908 	 * time includes the interrupt time although that is not wanted here.
909 	 * ps(1) is similarly sloppy.
910 	 */
911 	cputime = (pp->ki_runtime + 500000) / 1000000;
912 
913 	/* generate "STATE" field */
914 	switch (state = pp->ki_stat) {
915 	case SRUN:
916 		if (smpmode && pp->ki_oncpu != NOCPU)
917 			sprintf(status, "CPU%d", pp->ki_oncpu);
918 		else
919 			strcpy(status, "RUN");
920 		break;
921 	case SLOCK:
922 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
923 			sprintf(status, "*%.6s", pp->ki_lockname);
924 			break;
925 		}
926 		/* fall through */
927 	case SSLEEP:
928 		sprintf(status, "%.6s", pp->ki_wmesg);
929 		break;
930 	default:
931 
932 		if (state < nitems(state_abbrev)) {
933 			sprintf(status, "%.6s", state_abbrev[state]);
934 		} else {
935 			sprintf(status, "?%5zu", state);
936 		}
937 		break;
938 	}
939 
940 	cmdbuf = calloc(screen_width + 1, 1);
941 	if (cmdbuf == NULL) {
942 		warn("calloc(%d)", screen_width + 1);
943 		return NULL;
944 	}
945 
946 	if (!(flags & FMT_SHOWARGS)) {
947 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
948 		    pp->ki_tdname[0]) {
949 			snprintf(cmdbuf, screen_width, "%s{%s%s}", pp->ki_comm,
950 			    pp->ki_tdname, pp->ki_moretdname);
951 		} else {
952 			snprintf(cmdbuf, screen_width, "%s", pp->ki_comm);
953 		}
954 	} else {
955 		if (pp->ki_flag & P_SYSTEM ||
956 		    (args = kvm_getargv(kd, pp, screen_width)) == NULL ||
957 		    !(*args)) {
958 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
959 		    	    pp->ki_tdname[0]) {
960 				snprintf(cmdbuf, screen_width,
961 				    "[%s{%s%s}]", pp->ki_comm, pp->ki_tdname,
962 				    pp->ki_moretdname);
963 			} else {
964 				snprintf(cmdbuf, screen_width,
965 				    "[%s]", pp->ki_comm);
966 			}
967 		} else {
968 			const char *src;
969 			char *dst, *argbuf;
970 			const char *cmd;
971 			size_t argbuflen;
972 			size_t len;
973 
974 			argbuflen = screen_width * 4;
975 			argbuf = calloc(argbuflen + 1, 1);
976 			if (argbuf == NULL) {
977 				warn("calloc(%zu)", argbuflen + 1);
978 				free(cmdbuf);
979 				return NULL;
980 			}
981 
982 			dst = argbuf;
983 
984 			/* Extract cmd name from argv */
985 			cmd = basename(*args);
986 
987 			for (; (src = *args++) != NULL; ) {
988 				if (*src == '\0')
989 					continue;
990 				len = (argbuflen - (dst - argbuf) - 1) / 4;
991 				strvisx(dst, src,
992 				    MIN(strlen(src), len),
993 				    VIS_NL | VIS_CSTYLE);
994 				while (*dst != '\0')
995 					dst++;
996 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
997 					*dst++ = ' '; /* add delimiting space */
998 			}
999 			if (dst != argbuf && dst[-1] == ' ')
1000 				dst--;
1001 			*dst = '\0';
1002 
1003 			if (strcmp(cmd, pp->ki_comm) != 0) {
1004 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1005 				    pp->ki_tdname[0])
1006 					snprintf(cmdbuf, screen_width,
1007 					    "%s (%s){%s%s}", argbuf,
1008 					    pp->ki_comm, pp->ki_tdname,
1009 					    pp->ki_moretdname);
1010 				else
1011 					snprintf(cmdbuf, screen_width,
1012 					    "%s (%s)", argbuf, pp->ki_comm);
1013 			} else {
1014 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1015 				    pp->ki_tdname[0])
1016 					snprintf(cmdbuf, screen_width,
1017 					    "%s{%s%s}", argbuf, pp->ki_tdname,
1018 					    pp->ki_moretdname);
1019 				else
1020 					strlcpy(cmdbuf, argbuf, screen_width);
1021 			}
1022 			free(argbuf);
1023 		}
1024 	}
1025 
1026 	if (displaymode == DISP_IO) {
1027 		oldp = get_old_proc(pp);
1028 		if (oldp != NULL) {
1029 			ru.ru_inblock = RU(pp)->ru_inblock -
1030 			    RU(oldp)->ru_inblock;
1031 			ru.ru_oublock = RU(pp)->ru_oublock -
1032 			    RU(oldp)->ru_oublock;
1033 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1034 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1035 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1036 			rup = &ru;
1037 		} else {
1038 			rup = RU(pp);
1039 		}
1040 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1041 		s_tot = total_inblock + total_oublock + total_majflt;
1042 
1043 		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1044 
1045 		if (ps.jail) {
1046 			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1047 		}
1048 		sbuf_printf(procbuf, "%-*.*s", namelength, namelength, (*get_userid)(pp->ki_ruid));
1049 		sbuf_printf(procbuf, "%6ld ", rup->ru_nvcsw);
1050 		sbuf_printf(procbuf, "%6ld ", rup->ru_nivcsw);
1051 		sbuf_printf(procbuf, "%6ld ", rup->ru_inblock);
1052 		sbuf_printf(procbuf, "%6ld ", rup->ru_oublock);
1053 		sbuf_printf(procbuf, "%6ld ", rup->ru_majflt);
1054 		sbuf_printf(procbuf, "%6ld ", p_tot);
1055 		sbuf_printf(procbuf, "%6.2f%% ", s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot));
1056 
1057 	} else {
1058 		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1059 		if (ps.jail) {
1060 			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1061 		}
1062 		sbuf_printf(procbuf, "%-*.*s ", namelength, namelength, (*get_userid)(pp->ki_ruid));
1063 
1064 		if (!ps.thread) {
1065 			sbuf_printf(procbuf, "%4d ", pp->ki_numthreads);
1066 		}
1067 
1068 		sbuf_printf(procbuf, "%3d ", pp->ki_pri.pri_level - PZERO);
1069 		sbuf_printf(procbuf, "%4s", format_nice(pp));
1070 		sbuf_printf(procbuf, "%7s ", format_k(PROCSIZE(pp)));
1071 		sbuf_printf(procbuf, "%6s ", format_k(pagetok(pp->ki_rssize)));
1072 		if (ps.swap) {
1073 			sbuf_printf(procbuf, "%*s ",
1074 				TOP_SWAP_LEN - 1,
1075 				format_k(pagetok(ki_swap(pp))));
1076 		}
1077 		sbuf_printf(procbuf, "%-6.6s ", status);
1078 		if (smpmode) {
1079 			int cpu;
1080 			if (state == SRUN && pp->ki_oncpu != NOCPU) {
1081 				cpu = pp->ki_oncpu;
1082 			} else {
1083 				cpu = pp->ki_lastcpu;
1084 			}
1085 			sbuf_printf(procbuf, "%3d ", cpu);
1086 		}
1087 		sbuf_printf(procbuf, "%6s ", format_time(cputime));
1088 		sbuf_printf(procbuf, "%6.2f%% ", ps.wcpu ? 100.0 * weighted_cpu(PCTCPU(pp), pp) : 100.0 * PCTCPU(pp));
1089 	}
1090 	sbuf_printf(procbuf, "%s", printable(cmdbuf));
1091 	free(cmdbuf);
1092 	return (sbuf_data(procbuf));
1093 }
1094 
1095 static void
1096 getsysctl(const char *name, void *ptr, size_t len)
1097 {
1098 	size_t nlen = len;
1099 
1100 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1101 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1102 		    strerror(errno));
1103 		quit(TOP_EX_SYS_ERROR);
1104 	}
1105 	if (nlen != len) {
1106 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1107 		    name, (unsigned long)len, (unsigned long)nlen);
1108 		quit(TOP_EX_SYS_ERROR);
1109 	}
1110 }
1111 
1112 static const char *
1113 format_nice(const struct kinfo_proc *pp)
1114 {
1115 	const char *fifo, *kproc;
1116 	int rtpri;
1117 	static char nicebuf[4 + 1];
1118 
1119 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1120 	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1121 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1122 	case PRI_ITHD:
1123 		return ("-");
1124 	case PRI_REALTIME:
1125 		/*
1126 		 * XXX: the kernel doesn't tell us the original rtprio and
1127 		 * doesn't really know what it was, so to recover it we
1128 		 * must be more chummy with the implementation than the
1129 		 * implementation is with itself.  pri_user gives a
1130 		 * constant "base" priority, but is only initialized
1131 		 * properly for user threads.  pri_native gives what the
1132 		 * kernel calls the "base" priority, but it isn't constant
1133 		 * since it is changed by priority propagation.  pri_native
1134 		 * also isn't properly initialized for all threads, but it
1135 		 * is properly initialized for kernel realtime and idletime
1136 		 * threads.  Thus we use pri_user for the base priority of
1137 		 * user threads (it is always correct) and pri_native for
1138 		 * the base priority of kernel realtime and idletime threads
1139 		 * (there is nothing better, and it is usually correct).
1140 		 *
1141 		 * The field width and thus the buffer are too small for
1142 		 * values like "kr31F", but such values shouldn't occur,
1143 		 * and if they do then the tailing "F" is not displayed.
1144 		 */
1145 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1146 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1147 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1148 		    kproc, rtpri, fifo);
1149 		break;
1150 	case PRI_TIMESHARE:
1151 		if (pp->ki_flag & P_KPROC)
1152 			return ("-");
1153 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1154 		break;
1155 	case PRI_IDLE:
1156 		/* XXX: as above. */
1157 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1158 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1159 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1160 		    kproc, rtpri, fifo);
1161 		break;
1162 	default:
1163 		return ("?");
1164 	}
1165 	return (nicebuf);
1166 }
1167 
1168 /* comparison routines for qsort */
1169 
1170 static int
1171 compare_pid(const void *p1, const void *p2)
1172 {
1173 	const struct kinfo_proc * const *pp1 = p1;
1174 	const struct kinfo_proc * const *pp2 = p2;
1175 
1176 	assert((*pp2)->ki_pid >= 0 && (*pp1)->ki_pid >= 0);
1177 
1178 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1179 }
1180 
1181 static int
1182 compare_tid(const void *p1, const void *p2)
1183 {
1184 	const struct kinfo_proc * const *pp1 = p1;
1185 	const struct kinfo_proc * const *pp2 = p2;
1186 
1187 	assert((*pp2)->ki_tid >= 0 && (*pp1)->ki_tid >= 0);
1188 
1189 	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1190 }
1191 
1192 /*
1193  *  proc_compare - comparison function for "qsort"
1194  *	Compares the resource consumption of two processes using five
1195  *	distinct keys.  The keys (in descending order of importance) are:
1196  *	percent cpu, cpu ticks, state, resident set size, total virtual
1197  *	memory usage.  The process states are ordered as follows (from least
1198  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1199  *	array declaration below maps a process state index into a number
1200  *	that reflects this ordering.
1201  */
1202 
1203 static int sorted_state[] = {
1204 	0,	/* not used		*/
1205 	3,	/* sleep		*/
1206 	1,	/* ABANDONED (WAIT)	*/
1207 	6,	/* run			*/
1208 	5,	/* start		*/
1209 	2,	/* zombie		*/
1210 	4	/* stop			*/
1211 };
1212 
1213 
1214 #define ORDERKEY_PCTCPU(a, b) do { \
1215 	double diff; \
1216 	if (ps.wcpu) \
1217 		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1218 		    weighted_cpu(PCTCPU((a)), (a)); \
1219 	else \
1220 		diff = PCTCPU((b)) - PCTCPU((a)); \
1221 	if (diff != 0) \
1222 		return (diff > 0 ? 1 : -1); \
1223 } while (0)
1224 
1225 #define ORDERKEY_CPTICKS(a, b) do { \
1226 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1227 	if (diff != 0) \
1228 		return (diff > 0 ? 1 : -1); \
1229 } while (0)
1230 
1231 #define ORDERKEY_STATE(a, b) do { \
1232 	int diff = sorted_state[(unsigned char)(b)->ki_stat] - sorted_state[(unsigned char)(a)->ki_stat]; \
1233 	if (diff != 0) \
1234 		return (diff > 0 ? 1 : -1); \
1235 } while (0)
1236 
1237 #define ORDERKEY_PRIO(a, b) do { \
1238 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1239 	if (diff != 0) \
1240 		return (diff > 0 ? 1 : -1); \
1241 } while (0)
1242 
1243 #define	ORDERKEY_THREADS(a, b) do { \
1244 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1245 	if (diff != 0) \
1246 		return (diff > 0 ? 1 : -1); \
1247 } while (0)
1248 
1249 #define ORDERKEY_RSSIZE(a, b) do { \
1250 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1251 	if (diff != 0) \
1252 		return (diff > 0 ? 1 : -1); \
1253 } while (0)
1254 
1255 #define ORDERKEY_MEM(a, b) do { \
1256 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1257 	if (diff != 0) \
1258 		return (diff > 0 ? 1 : -1); \
1259 } while (0)
1260 
1261 #define ORDERKEY_JID(a, b) do { \
1262 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1263 	if (diff != 0) \
1264 		return (diff > 0 ? 1 : -1); \
1265 } while (0)
1266 
1267 #define ORDERKEY_SWAP(a, b) do { \
1268 	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1269 	if (diff != 0) \
1270 		return (diff > 0 ? 1 : -1); \
1271 } while (0)
1272 
1273 /* compare_cpu - the comparison function for sorting by cpu percentage */
1274 
1275 static int
1276 compare_cpu(const void *arg1, const void *arg2)
1277 {
1278 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1279 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1280 
1281 	ORDERKEY_PCTCPU(p1, p2);
1282 	ORDERKEY_CPTICKS(p1, p2);
1283 	ORDERKEY_STATE(p1, p2);
1284 	ORDERKEY_PRIO(p1, p2);
1285 	ORDERKEY_RSSIZE(p1, p2);
1286 	ORDERKEY_MEM(p1, p2);
1287 
1288 	return (0);
1289 }
1290 
1291 /* compare_size - the comparison function for sorting by total memory usage */
1292 
1293 static int
1294 compare_size(const void *arg1, const void *arg2)
1295 {
1296 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1297 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1298 
1299 	ORDERKEY_MEM(p1, p2);
1300 	ORDERKEY_RSSIZE(p1, p2);
1301 	ORDERKEY_PCTCPU(p1, p2);
1302 	ORDERKEY_CPTICKS(p1, p2);
1303 	ORDERKEY_STATE(p1, p2);
1304 	ORDERKEY_PRIO(p1, p2);
1305 
1306 	return (0);
1307 }
1308 
1309 /* compare_res - the comparison function for sorting by resident set size */
1310 
1311 static int
1312 compare_res(const void *arg1, const void *arg2)
1313 {
1314 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1315 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1316 
1317 	ORDERKEY_RSSIZE(p1, p2);
1318 	ORDERKEY_MEM(p1, p2);
1319 	ORDERKEY_PCTCPU(p1, p2);
1320 	ORDERKEY_CPTICKS(p1, p2);
1321 	ORDERKEY_STATE(p1, p2);
1322 	ORDERKEY_PRIO(p1, p2);
1323 
1324 	return (0);
1325 }
1326 
1327 /* compare_time - the comparison function for sorting by total cpu time */
1328 
1329 static int
1330 compare_time(const void *arg1, const void *arg2)
1331 {
1332 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const  *)arg1;
1333 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *) arg2;
1334 
1335 	ORDERKEY_CPTICKS(p1, p2);
1336 	ORDERKEY_PCTCPU(p1, p2);
1337 	ORDERKEY_STATE(p1, p2);
1338 	ORDERKEY_PRIO(p1, p2);
1339 	ORDERKEY_RSSIZE(p1, p2);
1340 	ORDERKEY_MEM(p1, p2);
1341 
1342 	return (0);
1343 }
1344 
1345 /* compare_prio - the comparison function for sorting by priority */
1346 
1347 static int
1348 compare_prio(const void *arg1, const void *arg2)
1349 {
1350 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1351 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1352 
1353 	ORDERKEY_PRIO(p1, p2);
1354 	ORDERKEY_CPTICKS(p1, p2);
1355 	ORDERKEY_PCTCPU(p1, p2);
1356 	ORDERKEY_STATE(p1, p2);
1357 	ORDERKEY_RSSIZE(p1, p2);
1358 	ORDERKEY_MEM(p1, p2);
1359 
1360 	return (0);
1361 }
1362 
1363 /* compare_threads - the comparison function for sorting by threads */
1364 static int
1365 compare_threads(const void *arg1, const void *arg2)
1366 {
1367 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1368 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1369 
1370 	ORDERKEY_THREADS(p1, p2);
1371 	ORDERKEY_PCTCPU(p1, p2);
1372 	ORDERKEY_CPTICKS(p1, p2);
1373 	ORDERKEY_STATE(p1, p2);
1374 	ORDERKEY_PRIO(p1, p2);
1375 	ORDERKEY_RSSIZE(p1, p2);
1376 	ORDERKEY_MEM(p1, p2);
1377 
1378 	return (0);
1379 }
1380 
1381 /* compare_jid - the comparison function for sorting by jid */
1382 static int
1383 compare_jid(const void *arg1, const void *arg2)
1384 {
1385 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1386 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1387 
1388 	ORDERKEY_JID(p1, p2);
1389 	ORDERKEY_PCTCPU(p1, p2);
1390 	ORDERKEY_CPTICKS(p1, p2);
1391 	ORDERKEY_STATE(p1, p2);
1392 	ORDERKEY_PRIO(p1, p2);
1393 	ORDERKEY_RSSIZE(p1, p2);
1394 	ORDERKEY_MEM(p1, p2);
1395 
1396 	return (0);
1397 }
1398 
1399 /* compare_swap - the comparison function for sorting by swap */
1400 static int
1401 compare_swap(const void *arg1, const void *arg2)
1402 {
1403 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1404 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1405 
1406 	ORDERKEY_SWAP(p1, p2);
1407 	ORDERKEY_PCTCPU(p1, p2);
1408 	ORDERKEY_CPTICKS(p1, p2);
1409 	ORDERKEY_STATE(p1, p2);
1410 	ORDERKEY_PRIO(p1, p2);
1411 	ORDERKEY_RSSIZE(p1, p2);
1412 	ORDERKEY_MEM(p1, p2);
1413 
1414 	return (0);
1415 }
1416 
1417 /* assorted comparison functions for sorting by i/o */
1418 
1419 static int
1420 compare_iototal(const void *arg1, const void *arg2)
1421 {
1422 	const struct kinfo_proc * const p1 = *(const struct kinfo_proc * const *)arg1;
1423 	const struct kinfo_proc * const p2 = *(const struct kinfo_proc * const *)arg2;
1424 
1425 	return (get_io_total(p2) - get_io_total(p1));
1426 }
1427 
1428 static int
1429 compare_ioread(const void *arg1, const void *arg2)
1430 {
1431 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1432 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1433 	long dummy, inp1, inp2;
1434 
1435 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1436 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1437 
1438 	return (inp2 - inp1);
1439 }
1440 
1441 static int
1442 compare_iowrite(const void *arg1, const void *arg2)
1443 {
1444 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1445 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1446 	long dummy, oup1, oup2;
1447 
1448 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1449 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1450 
1451 	return (oup2 - oup1);
1452 }
1453 
1454 static int
1455 compare_iofault(const void *arg1, const void *arg2)
1456 {
1457 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1458 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1459 	long dummy, flp1, flp2;
1460 
1461 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1462 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1463 
1464 	return (flp2 - flp1);
1465 }
1466 
1467 static int
1468 compare_vcsw(const void *arg1, const void *arg2)
1469 {
1470 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1471 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1472 	long dummy, flp1, flp2;
1473 
1474 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1475 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1476 
1477 	return (flp2 - flp1);
1478 }
1479 
1480 static int
1481 compare_ivcsw(const void *arg1, const void *arg2)
1482 {
1483 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1484 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1485 	long dummy, flp1, flp2;
1486 
1487 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1488 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1489 
1490 	return (flp2 - flp1);
1491 }
1492 
1493 int (*compares[])(const void *arg1, const void *arg2) = {
1494 	compare_cpu,
1495 	compare_size,
1496 	compare_res,
1497 	compare_time,
1498 	compare_prio,
1499 	compare_threads,
1500 	compare_iototal,
1501 	compare_ioread,
1502 	compare_iowrite,
1503 	compare_iofault,
1504 	compare_vcsw,
1505 	compare_ivcsw,
1506 	compare_jid,
1507 	compare_swap,
1508 	NULL
1509 };
1510 
1511 
1512 static int
1513 swapmode(int *retavail, int *retfree)
1514 {
1515 	int n;
1516 	struct kvm_swap swapary[1];
1517 	static int pagesize = 0;
1518 	static unsigned long swap_maxpages = 0;
1519 
1520 	*retavail = 0;
1521 	*retfree = 0;
1522 
1523 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1524 
1525 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1526 	if (n < 0 || swapary[0].ksw_total == 0)
1527 		return (0);
1528 
1529 	if (pagesize == 0)
1530 		pagesize = getpagesize();
1531 	if (swap_maxpages == 0)
1532 		GETSYSCTL("vm.swap_maxpages", swap_maxpages);
1533 
1534 	/* ksw_total contains the total size of swap all devices which may
1535 	   exceed the maximum swap size allocatable in the system */
1536 	if ( swapary[0].ksw_total > swap_maxpages )
1537 		swapary[0].ksw_total = swap_maxpages;
1538 
1539 	*retavail = CONVERT(swapary[0].ksw_total);
1540 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1541 
1542 #undef CONVERT
1543 
1544 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1545 	return (n);
1546 }
1547