xref: /freebsd/usr.bin/top/machine.c (revision 499fe48de8938d4c7b0a91e20eb6c16db9d55633)
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 static int has_swap;
154 
155 /* these are for keeping track of the proc array */
156 
157 static int nproc;
158 static int onproc = -1;
159 static int pref_len;
160 static struct kinfo_proc *pbase;
161 static struct kinfo_proc **pref;
162 static struct kinfo_proc *previous_procs;
163 static struct kinfo_proc **previous_pref;
164 static int previous_proc_count = 0;
165 static int previous_proc_count_max = 0;
166 static int previous_thread;
167 
168 /* data used for recalculating pctcpu */
169 static double *pcpu;
170 static struct timespec proc_uptime;
171 static struct timeval proc_wall_time;
172 static struct timeval previous_wall_time;
173 static uint64_t previous_interval = 0;
174 
175 /* total number of io operations */
176 static long total_inblock;
177 static long total_oublock;
178 static long total_majflt;
179 
180 /* these are for getting the memory statistics */
181 
182 static int arc_enabled;
183 static int carc_enabled;
184 static int pageshift;		/* log base 2 of the pagesize */
185 
186 /* define pagetok in terms of pageshift */
187 
188 #define pagetok(size) ((size) << pageshift)
189 
190 /* swap usage */
191 #define ki_swap(kip) \
192     ((kip)->ki_swrss > (kip)->ki_rssize ? (kip)->ki_swrss - (kip)->ki_rssize : 0)
193 
194 /*
195  * Sorting orders.  The first element is the default.
196  */
197 static const char *ordernames[] = {
198 	"cpu", "size", "res", "time", "pri", "threads",
199 	"total", "read", "write", "fault", "vcsw", "ivcsw",
200 	"jid", "swap", "pid", NULL
201 };
202 
203 /* Per-cpu time states */
204 static int maxcpu;
205 static int maxid;
206 static int ncpus;
207 static unsigned long cpumask;
208 static long *times;
209 static long *pcpu_cp_time;
210 static long *pcpu_cp_old;
211 static long *pcpu_cp_diff;
212 static int *pcpu_cpu_states;
213 
214 static int compare_swap(const void *a, const void *b);
215 static int compare_jid(const void *a, const void *b);
216 static int compare_pid(const void *a, const void *b);
217 static int compare_tid(const void *a, const void *b);
218 static const char *format_nice(const struct kinfo_proc *pp);
219 static void getsysctl(const char *name, void *ptr, size_t len);
220 static int swapmode(int *retavail, int *retfree);
221 static void update_layout(void);
222 static int find_uid(uid_t needle, int *haystack);
223 
224 static int
225 find_uid(uid_t needle, int *haystack)
226 {
227 	size_t i = 0;
228 
229 	for (; i < TOP_MAX_UIDS; ++i)
230 		if ((uid_t)haystack[i] == needle)
231 			return 1;
232 	return (0);
233 }
234 
235 void
236 toggle_pcpustats(void)
237 {
238 
239 	if (ncpus == 1)
240 		return;
241 	update_layout();
242 }
243 
244 /* Adjust display based on ncpus and the ARC state. */
245 static void
246 update_layout(void)
247 {
248 
249 	y_mem = 3;
250 	y_arc = 4;
251 	y_carc = 5;
252 	y_swap = 3 + arc_enabled + carc_enabled + has_swap;
253 	y_idlecursor = 4 + arc_enabled + carc_enabled + has_swap;
254 	y_message = 4 + arc_enabled + carc_enabled + has_swap;
255 	y_header = 5 + arc_enabled + carc_enabled + has_swap;
256 	y_procs = 6 + arc_enabled + carc_enabled + has_swap;
257 	Header_lines = 6 + arc_enabled + carc_enabled + has_swap;
258 
259 	if (pcpu_stats) {
260 		y_mem += ncpus - 1;
261 		y_arc += ncpus - 1;
262 		y_carc += ncpus - 1;
263 		y_swap += ncpus - 1;
264 		y_idlecursor += ncpus - 1;
265 		y_message += ncpus - 1;
266 		y_header += ncpus - 1;
267 		y_procs += ncpus - 1;
268 		Header_lines += ncpus - 1;
269 	}
270 }
271 
272 int
273 machine_init(struct statics *statics)
274 {
275 	int i, j, empty, pagesize;
276 	uint64_t arc_size;
277 	int carc_en, nswapdev;
278 	size_t size;
279 
280 	size = sizeof(smpmode);
281 	if ((sysctlbyname("machdep.smp_active", &smpmode, &size,
282 	    NULL, 0) != 0 &&
283 	    sysctlbyname("kern.smp.active", &smpmode, &size,
284 	    NULL, 0) != 0) ||
285 	    size != sizeof(smpmode))
286 		smpmode = 0;
287 
288 	size = sizeof(arc_size);
289 	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
290 	    NULL, 0) == 0 && arc_size != 0)
291 		arc_enabled = 1;
292 	size = sizeof(carc_en);
293 	if (arc_enabled &&
294 	    sysctlbyname("vfs.zfs.compressed_arc_enabled", &carc_en, &size,
295 	    NULL, 0) == 0 && carc_en == 1)
296 		carc_enabled = 1;
297 
298 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
299 	if (kd == NULL)
300 		return (-1);
301 
302 	size = sizeof(nswapdev);
303 	if (sysctlbyname("vm.nswapdev", &nswapdev, &size, NULL,
304 		0) == 0 && nswapdev != 0)
305 			has_swap = 1;
306 
307 	GETSYSCTL("kern.ccpu", ccpu);
308 
309 	/* this is used in calculating WCPU -- calculate it ahead of time */
310 	logcpu = log(loaddouble(ccpu));
311 
312 	pbase = NULL;
313 	pref = NULL;
314 	pcpu = NULL;
315 	nproc = 0;
316 	onproc = -1;
317 
318 	/* get the page size and calculate pageshift from it */
319 	pagesize = getpagesize();
320 	pageshift = 0;
321 	while (pagesize > 1) {
322 		pageshift++;
323 		pagesize >>= 1;
324 	}
325 
326 	/* we only need the amount of log(2)1024 for our conversion */
327 	pageshift -= LOG1024;
328 
329 	/* fill in the statics information */
330 	statics->procstate_names = procstatenames;
331 	statics->cpustate_names = cpustatenames;
332 	statics->memory_names = memorynames;
333 	if (arc_enabled)
334 		statics->arc_names = arcnames;
335 	else
336 		statics->arc_names = NULL;
337 	if (carc_enabled)
338 		statics->carc_names = carcnames;
339 	else
340 		statics->carc_names = NULL;
341 	if (has_swap)
342 		statics->swap_names = swapnames;
343 	else
344 		statics->swap_names = NULL;
345 	statics->order_names = ordernames;
346 
347 	/* Allocate state for per-CPU stats. */
348 	cpumask = 0;
349 	ncpus = 0;
350 	GETSYSCTL("kern.smp.maxcpus", maxcpu);
351 	times = calloc(maxcpu * CPUSTATES, sizeof(long));
352 	if (times == NULL)
353 		err(1, "calloc for kern.smp.maxcpus");
354 	size = sizeof(long) * maxcpu * CPUSTATES;
355 	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
356 		err(1, "sysctlbyname kern.cp_times");
357 	pcpu_cp_time = calloc(1, size);
358 	maxid = (size / CPUSTATES / sizeof(long)) - 1;
359 	for (i = 0; i <= maxid; i++) {
360 		empty = 1;
361 		for (j = 0; empty && j < CPUSTATES; j++) {
362 			if (times[i * CPUSTATES + j] != 0)
363 				empty = 0;
364 		}
365 		if (!empty) {
366 			cpumask |= (1ul << i);
367 			ncpus++;
368 		}
369 	}
370 	assert(ncpus > 0);
371 	pcpu_cp_old = calloc(ncpus * CPUSTATES, sizeof(long));
372 	pcpu_cp_diff = calloc(ncpus * CPUSTATES, sizeof(long));
373 	pcpu_cpu_states = calloc(ncpus * CPUSTATES, sizeof(int));
374 	statics->ncpus = ncpus;
375 
376 	update_layout();
377 
378 	/* all done! */
379 	return (0);
380 }
381 
382 char *
383 format_header(const char *uname_field)
384 {
385 	static struct sbuf* header = NULL;
386 
387 	/* clean up from last time. */
388 	if (header != NULL) {
389 		sbuf_clear(header);
390 	} else {
391 		header = sbuf_new_auto();
392 	}
393 
394 	switch (displaymode) {
395 	case DISP_CPU: {
396 		sbuf_printf(header, "  %s", ps.thread_id ? " THR" : "PID");
397 		sbuf_printf(header, "%*s", ps.jail ? TOP_JID_LEN : 0,
398 									ps.jail ? " JID" : "");
399 		sbuf_printf(header, " %-*.*s  ", namelength, namelength, uname_field);
400 		if (!ps.thread) {
401 			sbuf_cat(header, "THR ");
402 		}
403 		sbuf_cat(header, "PRI NICE   SIZE    RES ");
404 		if (ps.swap) {
405 			sbuf_printf(header, "%*s ", TOP_SWAP_LEN - 1, "SWAP");
406 		}
407 		sbuf_cat(header, "STATE    ");
408 		if (smpmode) {
409 			sbuf_cat(header, "C   ");
410 		}
411 		sbuf_cat(header, "TIME ");
412 		sbuf_printf(header, " %6s ", ps.wcpu ? "WCPU" : "CPU");
413 		sbuf_cat(header, "COMMAND");
414 		sbuf_finish(header);
415 		break;
416 	}
417 	case DISP_IO: {
418 		sbuf_printf(header, "  %s%*s %-*.*s",
419 			ps.thread_id ? " THR" : "PID",
420 		    ps.jail ? TOP_JID_LEN : 0, ps.jail ? " JID" : "",
421 		    namelength, namelength, uname_field);
422 		sbuf_cat(header, "   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND");
423 		sbuf_finish(header);
424 		break;
425 	}
426 	case DISP_MAX:
427 		assert("displaymode must not be set to DISP_MAX");
428 	}
429 
430 	return sbuf_data(header);
431 }
432 
433 static int swappgsin = -1;
434 static int swappgsout = -1;
435 
436 
437 void
438 get_system_info(struct system_info *si)
439 {
440 	struct loadavg sysload;
441 	int mib[2];
442 	struct timeval boottime;
443 	uint64_t arc_stat, arc_stat2;
444 	int i, j;
445 	size_t size;
446 
447 	/* get the CPU stats */
448 	size = (maxid + 1) * CPUSTATES * sizeof(long);
449 	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
450 		err(1, "sysctlbyname kern.cp_times");
451 	GETSYSCTL("kern.cp_time", cp_time);
452 	GETSYSCTL("vm.loadavg", sysload);
453 	GETSYSCTL("kern.lastpid", lastpid);
454 
455 	/* convert load averages to doubles */
456 	for (i = 0; i < 3; i++)
457 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
458 
459 	/* convert cp_time counts to percentages */
460 	for (i = j = 0; i <= maxid; i++) {
461 		if ((cpumask & (1ul << i)) == 0)
462 			continue;
463 		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
464 		    &pcpu_cp_time[j * CPUSTATES],
465 		    &pcpu_cp_old[j * CPUSTATES],
466 		    &pcpu_cp_diff[j * CPUSTATES]);
467 		j++;
468 	}
469 	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
470 
471 	/* sum memory & swap statistics */
472 	{
473 		static unsigned int swap_delay = 0;
474 		static int swapavail = 0;
475 		static int swapfree = 0;
476 		static long bufspace = 0;
477 		static uint64_t nspgsin, nspgsout;
478 
479 		GETSYSCTL("vfs.bufspace", bufspace);
480 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
481 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
482 		GETSYSCTL("vm.stats.vm.v_laundry_count", memory_stats[2]);
483 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[3]);
484 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
485 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
486 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
487 		/* convert memory stats to Kbytes */
488 		memory_stats[0] = pagetok(memory_stats[0]);
489 		memory_stats[1] = pagetok(memory_stats[1]);
490 		memory_stats[2] = pagetok(memory_stats[2]);
491 		memory_stats[3] = pagetok(memory_stats[3]);
492 		memory_stats[4] = bufspace / 1024;
493 		memory_stats[5] = pagetok(memory_stats[5]);
494 		memory_stats[6] = -1;
495 
496 		/* first interval */
497 		if (swappgsin < 0) {
498 			swap_stats[4] = 0;
499 			swap_stats[5] = 0;
500 		}
501 
502 		/* compute differences between old and new swap statistic */
503 		else {
504 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
505 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
506 		}
507 
508 		swappgsin = nspgsin;
509 		swappgsout = nspgsout;
510 
511 		/* call CPU heavy swapmode() only for changes */
512 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
513 			swap_stats[3] = swapmode(&swapavail, &swapfree);
514 			swap_stats[0] = swapavail;
515 			swap_stats[1] = swapavail - swapfree;
516 			swap_stats[2] = swapfree;
517 		}
518 		swap_delay = 1;
519 		swap_stats[6] = -1;
520 	}
521 
522 	if (arc_enabled) {
523 		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
524 		arc_stats[0] = arc_stat >> 10;
525 		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
526 		arc_stats[1] = arc_stat >> 10;
527 		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
528 		arc_stats[2] = arc_stat >> 10;
529 		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
530 		arc_stats[3] = arc_stat >> 10;
531 		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
532 		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
533 		arc_stats[4] = (arc_stat + arc_stat2) >> 10;
534 		GETSYSCTL("kstat.zfs.misc.arcstats.bonus_size", arc_stat);
535 		arc_stats[5] = arc_stat >> 10;
536 		GETSYSCTL("kstat.zfs.misc.arcstats.dnode_size", arc_stat);
537 		arc_stats[5] += arc_stat >> 10;
538 		GETSYSCTL("kstat.zfs.misc.arcstats.dbuf_size", arc_stat);
539 		arc_stats[5] += arc_stat >> 10;
540 		si->arc = arc_stats;
541 	}
542 	if (carc_enabled) {
543 		GETSYSCTL("kstat.zfs.misc.arcstats.compressed_size", arc_stat);
544 		carc_stats[0] = arc_stat >> 10;
545 		carc_stats[2] = arc_stat >> 10; /* For ratio */
546 		GETSYSCTL("kstat.zfs.misc.arcstats.uncompressed_size", arc_stat);
547 		carc_stats[1] = arc_stat >> 10;
548 		si->carc = carc_stats;
549 	}
550 
551 	/* set arrays and strings */
552 	if (pcpu_stats) {
553 		si->cpustates = pcpu_cpu_states;
554 		si->ncpus = ncpus;
555 	} else {
556 		si->cpustates = cpu_states;
557 		si->ncpus = 1;
558 	}
559 	si->memory = memory_stats;
560 	si->swap = swap_stats;
561 
562 
563 	if (lastpid > 0) {
564 		si->last_pid = lastpid;
565 	} else {
566 		si->last_pid = -1;
567 	}
568 
569 	/*
570 	 * Print how long system has been up.
571 	 * (Found by looking getting "boottime" from the kernel)
572 	 */
573 	mib[0] = CTL_KERN;
574 	mib[1] = KERN_BOOTTIME;
575 	size = sizeof(boottime);
576 	if (sysctl(mib, nitems(mib), &boottime, &size, NULL, 0) != -1 &&
577 	    boottime.tv_sec != 0) {
578 		si->boottime = boottime;
579 	} else {
580 		si->boottime.tv_sec = -1;
581 	}
582 }
583 
584 #define NOPROC	((void *)-1)
585 
586 /*
587  * We need to compare data from the old process entry with the new
588  * process entry.
589  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
590  * structure to cache the mapping.  We also use a negative cache pointer
591  * of NOPROC to avoid duplicate lookups.
592  * XXX: this could be done when the actual processes are fetched, we do
593  * it here out of laziness.
594  */
595 static const struct kinfo_proc *
596 get_old_proc(struct kinfo_proc *pp)
597 {
598 	const struct kinfo_proc * const *oldpp, *oldp;
599 
600 	/*
601 	 * If this is the first fetch of the kinfo_procs then we don't have
602 	 * any previous entries.
603 	 */
604 	if (previous_proc_count == 0)
605 		return (NULL);
606 	/* negative cache? */
607 	if (pp->ki_udata == NOPROC)
608 		return (NULL);
609 	/* cached? */
610 	if (pp->ki_udata != NULL)
611 		return (pp->ki_udata);
612 	/*
613 	 * Not cached,
614 	 * 1) look up based on pid.
615 	 * 2) compare process start.
616 	 * If we fail here, then setup a negative cache entry, otherwise
617 	 * cache it.
618 	 */
619 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
620 	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
621 	if (oldpp == NULL) {
622 		pp->ki_udata = NOPROC;
623 		return (NULL);
624 	}
625 	oldp = *oldpp;
626 	if (memcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
627 		pp->ki_udata = NOPROC;
628 		return (NULL);
629 	}
630 	pp->ki_udata = __DECONST(void *, oldp);
631 	return (oldp);
632 }
633 
634 /*
635  * Return the total amount of IO done in blocks in/out and faults.
636  * store the values individually in the pointers passed in.
637  */
638 static long
639 get_io_stats(const struct kinfo_proc *pp, long *inp, long *oup, long *flp,
640     long *vcsw, long *ivcsw)
641 {
642 	const struct kinfo_proc *oldp;
643 	static struct kinfo_proc dummy;
644 	long ret;
645 
646 	oldp = get_old_proc(__DECONST(struct kinfo_proc *, pp));
647 	if (oldp == NULL) {
648 		memset(&dummy, 0, sizeof(dummy));
649 		oldp = &dummy;
650 	}
651 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
652 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
653 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
654 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
655 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
656 	ret =
657 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
658 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
659 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
660 	return (ret);
661 }
662 
663 /*
664  * If there was a previous update, use the delta in ki_runtime over
665  * the previous interval to calculate pctcpu.  Otherwise, fall back
666  * to using the kernel's ki_pctcpu.
667  */
668 static double
669 proc_calc_pctcpu(struct kinfo_proc *pp)
670 {
671 	const struct kinfo_proc *oldp;
672 
673 	if (previous_interval != 0) {
674 		oldp = get_old_proc(pp);
675 		if (oldp != NULL)
676 			return ((double)(pp->ki_runtime - oldp->ki_runtime)
677 			    / previous_interval);
678 
679 		/*
680 		 * If this process/thread was created during the previous
681 		 * interval, charge it's total runtime to the previous
682 		 * interval.
683 		 */
684 		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
685 		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
686 		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
687 			return ((double)pp->ki_runtime / previous_interval);
688 	}
689 	return (pctdouble(pp->ki_pctcpu));
690 }
691 
692 /*
693  * Return true if this process has used any CPU time since the
694  * previous update.
695  */
696 static int
697 proc_used_cpu(struct kinfo_proc *pp)
698 {
699 	const struct kinfo_proc *oldp;
700 
701 	oldp = get_old_proc(pp);
702 	if (oldp == NULL)
703 		return (PCTCPU(pp) != 0);
704 	return (pp->ki_runtime != oldp->ki_runtime ||
705 	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
706 	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
707 }
708 
709 /*
710  * Return the total number of block in/out and faults by a process.
711  */
712 static long
713 get_io_total(const struct kinfo_proc *pp)
714 {
715 	long dummy;
716 
717 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
718 }
719 
720 static struct handle handle;
721 
722 void *
723 get_process_info(struct system_info *si, struct process_select *sel,
724     int (*compare)(const void *, const void *))
725 {
726 	int i;
727 	int total_procs;
728 	long p_io;
729 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
730 	long nsec;
731 	int active_procs;
732 	struct kinfo_proc **prefp;
733 	struct kinfo_proc *pp;
734 	struct timespec previous_proc_uptime;
735 
736 	/*
737 	 * If thread state was toggled, don't cache the previous processes.
738 	 */
739 	if (previous_thread != sel->thread)
740 		nproc = 0;
741 	previous_thread = sel->thread;
742 
743 	/*
744 	 * Save the previous process info.
745 	 */
746 	if (previous_proc_count_max < nproc) {
747 		free(previous_procs);
748 		previous_procs = calloc(nproc, sizeof(*previous_procs));
749 		free(previous_pref);
750 		previous_pref = calloc(nproc, sizeof(*previous_pref));
751 		if (previous_procs == NULL || previous_pref == NULL) {
752 			fprintf(stderr, "top: Out of memory.\n");
753 			quit(TOP_EX_SYS_ERROR);
754 		}
755 		previous_proc_count_max = nproc;
756 	}
757 	if (nproc) {
758 		for (i = 0; i < nproc; i++)
759 			previous_pref[i] = &previous_procs[i];
760 		memcpy(previous_procs, pbase, nproc * sizeof(*previous_procs));
761 		qsort(previous_pref, nproc, sizeof(*previous_pref),
762 		    ps.thread ? compare_tid : compare_pid);
763 	}
764 	previous_proc_count = nproc;
765 	previous_proc_uptime = proc_uptime;
766 	previous_wall_time = proc_wall_time;
767 	previous_interval = 0;
768 
769 	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
770 	    0, &nproc);
771 	gettimeofday(&proc_wall_time, NULL);
772 	if (clock_gettime(CLOCK_UPTIME, &proc_uptime) != 0)
773 		memset(&proc_uptime, 0, sizeof(proc_uptime));
774 	else if (previous_proc_uptime.tv_sec != 0 &&
775 	    previous_proc_uptime.tv_nsec != 0) {
776 		previous_interval = (proc_uptime.tv_sec -
777 		    previous_proc_uptime.tv_sec) * 1000000;
778 		nsec = proc_uptime.tv_nsec - previous_proc_uptime.tv_nsec;
779 		if (nsec < 0) {
780 			previous_interval -= 1000000;
781 			nsec += 1000000000;
782 		}
783 		previous_interval += nsec / 1000;
784 	}
785 	if (nproc > onproc) {
786 		pref = realloc(pref, sizeof(*pref) * nproc);
787 		pcpu = realloc(pcpu, sizeof(*pcpu) * nproc);
788 		onproc = nproc;
789 	}
790 	if (pref == NULL || pbase == NULL || pcpu == NULL) {
791 		fprintf(stderr, "top: Out of memory.\n");
792 		quit(TOP_EX_SYS_ERROR);
793 	}
794 	/* get a pointer to the states summary array */
795 	si->procstates = process_states;
796 
797 	/* count up process states and get pointers to interesting procs */
798 	total_procs = 0;
799 	active_procs = 0;
800 	total_inblock = 0;
801 	total_oublock = 0;
802 	total_majflt = 0;
803 	memset(process_states, 0, sizeof(process_states));
804 	prefp = pref;
805 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
806 
807 		if (pp->ki_stat == 0)
808 			/* not in use */
809 			continue;
810 
811 		if (!sel->self && pp->ki_pid == mypid && sel->pid == -1)
812 			/* skip self */
813 			continue;
814 
815 		if (!sel->system && (pp->ki_flag & P_SYSTEM) && sel->pid == -1)
816 			/* skip system process */
817 			continue;
818 
819 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
820 		    &p_vcsw, &p_ivcsw);
821 		total_inblock += p_inblock;
822 		total_oublock += p_oublock;
823 		total_majflt += p_majflt;
824 		total_procs++;
825 		process_states[(unsigned char)pp->ki_stat]++;
826 
827 		if (pp->ki_stat == SZOMB)
828 			/* skip zombies */
829 			continue;
830 
831 		if (!sel->kidle && pp->ki_tdflags & TDF_IDLETD && sel->pid == -1)
832 			/* skip kernel idle process */
833 			continue;
834 
835 		PCTCPU(pp) = proc_calc_pctcpu(pp);
836 		if (sel->thread && PCTCPU(pp) > 1.0)
837 			PCTCPU(pp) = 1.0;
838 		if (displaymode == DISP_CPU && !sel->idle &&
839 		    (!proc_used_cpu(pp) ||
840 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
841 			/* skip idle or non-running processes */
842 			continue;
843 
844 		if (displaymode == DISP_IO && !sel->idle && p_io == 0)
845 			/* skip processes that aren't doing I/O */
846 			continue;
847 
848 		if (sel->jid != -1 && pp->ki_jid != sel->jid)
849 			/* skip proc. that don't belong to the selected JID */
850 			continue;
851 
852 		if (sel->uid[0] != -1 && !find_uid(pp->ki_ruid, sel->uid))
853 			/* skip proc. that don't belong to the selected UID */
854 			continue;
855 
856 		if (sel->pid != -1 && pp->ki_pid != sel->pid)
857 			continue;
858 
859 		*prefp++ = pp;
860 		active_procs++;
861 	}
862 
863 	/* if requested, sort the "interesting" processes */
864 	if (compare != NULL)
865 		qsort(pref, active_procs, sizeof(*pref), compare);
866 
867 	/* remember active and total counts */
868 	si->p_total = total_procs;
869 	si->p_pactive = pref_len = active_procs;
870 
871 	/* pass back a handle */
872 	handle.next_proc = pref;
873 	handle.remaining = active_procs;
874 	return (&handle);
875 }
876 
877 char *
878 format_next_process(struct handle * xhandle, char *(*get_userid)(int), int flags)
879 {
880 	struct kinfo_proc *pp;
881 	const struct kinfo_proc *oldp;
882 	long cputime;
883 	char status[22];
884 	size_t state;
885 	struct rusage ru, *rup;
886 	long p_tot, s_tot;
887 	char *cmdbuf = NULL;
888 	char **args;
889 	static struct sbuf* procbuf = NULL;
890 
891 	/* clean up from last time. */
892 	if (procbuf != NULL) {
893 		sbuf_clear(procbuf);
894 	} else {
895 		procbuf = sbuf_new_auto();
896 	}
897 
898 
899 	/* find and remember the next proc structure */
900 	pp = *(xhandle->next_proc++);
901 	xhandle->remaining--;
902 
903 	/* get the process's command name */
904 	if ((pp->ki_flag & P_INMEM) == 0) {
905 		/*
906 		 * Print swapped processes as <pname>
907 		 */
908 		size_t len;
909 
910 		len = strlen(pp->ki_comm);
911 		if (len > sizeof(pp->ki_comm) - 3)
912 			len = sizeof(pp->ki_comm) - 3;
913 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
914 		pp->ki_comm[0] = '<';
915 		pp->ki_comm[len + 1] = '>';
916 		pp->ki_comm[len + 2] = '\0';
917 	}
918 
919 	/*
920 	 * Convert the process's runtime from microseconds to seconds.  This
921 	 * time includes the interrupt time although that is not wanted here.
922 	 * ps(1) is similarly sloppy.
923 	 */
924 	cputime = (pp->ki_runtime + 500000) / 1000000;
925 
926 	/* generate "STATE" field */
927 	switch (state = pp->ki_stat) {
928 	case SRUN:
929 		if (smpmode && pp->ki_oncpu != NOCPU)
930 			sprintf(status, "CPU%d", pp->ki_oncpu);
931 		else
932 			strcpy(status, "RUN");
933 		break;
934 	case SLOCK:
935 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
936 			sprintf(status, "*%.6s", pp->ki_lockname);
937 			break;
938 		}
939 		/* fall through */
940 	case SSLEEP:
941 		sprintf(status, "%.6s", pp->ki_wmesg);
942 		break;
943 	default:
944 
945 		if (state < nitems(state_abbrev)) {
946 			sprintf(status, "%.6s", state_abbrev[state]);
947 		} else {
948 			sprintf(status, "?%5zu", state);
949 		}
950 		break;
951 	}
952 
953 	cmdbuf = calloc(screen_width + 1, 1);
954 	if (cmdbuf == NULL) {
955 		warn("calloc(%d)", screen_width + 1);
956 		return NULL;
957 	}
958 
959 	if (!(flags & FMT_SHOWARGS)) {
960 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
961 		    pp->ki_tdname[0]) {
962 			snprintf(cmdbuf, screen_width, "%s{%s%s}", pp->ki_comm,
963 			    pp->ki_tdname, pp->ki_moretdname);
964 		} else {
965 			snprintf(cmdbuf, screen_width, "%s", pp->ki_comm);
966 		}
967 	} else {
968 		if (pp->ki_flag & P_SYSTEM ||
969 		    (args = kvm_getargv(kd, pp, screen_width)) == NULL ||
970 		    !(*args)) {
971 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
972 		    	    pp->ki_tdname[0]) {
973 				snprintf(cmdbuf, screen_width,
974 				    "[%s{%s%s}]", pp->ki_comm, pp->ki_tdname,
975 				    pp->ki_moretdname);
976 			} else {
977 				snprintf(cmdbuf, screen_width,
978 				    "[%s]", pp->ki_comm);
979 			}
980 		} else {
981 			const char *src;
982 			char *dst, *argbuf;
983 			const char *cmd;
984 			size_t argbuflen;
985 			size_t len;
986 
987 			argbuflen = screen_width * 4;
988 			argbuf = calloc(argbuflen + 1, 1);
989 			if (argbuf == NULL) {
990 				warn("calloc(%zu)", argbuflen + 1);
991 				free(cmdbuf);
992 				return NULL;
993 			}
994 
995 			dst = argbuf;
996 
997 			/* Extract cmd name from argv */
998 			cmd = basename(*args);
999 
1000 			for (; (src = *args++) != NULL; ) {
1001 				if (*src == '\0')
1002 					continue;
1003 				len = (argbuflen - (dst - argbuf) - 1) / 4;
1004 				strvisx(dst, src,
1005 				    MIN(strlen(src), len),
1006 				    VIS_NL | VIS_CSTYLE | VIS_OCTAL | VIS_SAFE);
1007 				while (*dst != '\0')
1008 					dst++;
1009 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
1010 					*dst++ = ' '; /* add delimiting space */
1011 			}
1012 			if (dst != argbuf && dst[-1] == ' ')
1013 				dst--;
1014 			*dst = '\0';
1015 
1016 			if (strcmp(cmd, pp->ki_comm) != 0) {
1017 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1018 				    pp->ki_tdname[0])
1019 					snprintf(cmdbuf, screen_width,
1020 					    "%s (%s){%s%s}", argbuf,
1021 					    pp->ki_comm, pp->ki_tdname,
1022 					    pp->ki_moretdname);
1023 				else
1024 					snprintf(cmdbuf, screen_width,
1025 					    "%s (%s)", argbuf, pp->ki_comm);
1026 			} else {
1027 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1028 				    pp->ki_tdname[0])
1029 					snprintf(cmdbuf, screen_width,
1030 					    "%s{%s%s}", argbuf, pp->ki_tdname,
1031 					    pp->ki_moretdname);
1032 				else
1033 					strlcpy(cmdbuf, argbuf, screen_width);
1034 			}
1035 			free(argbuf);
1036 		}
1037 	}
1038 
1039 	if (displaymode == DISP_IO) {
1040 		oldp = get_old_proc(pp);
1041 		if (oldp != NULL) {
1042 			ru.ru_inblock = RU(pp)->ru_inblock -
1043 			    RU(oldp)->ru_inblock;
1044 			ru.ru_oublock = RU(pp)->ru_oublock -
1045 			    RU(oldp)->ru_oublock;
1046 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1047 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1048 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1049 			rup = &ru;
1050 		} else {
1051 			rup = RU(pp);
1052 		}
1053 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1054 		s_tot = total_inblock + total_oublock + total_majflt;
1055 
1056 		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1057 
1058 		if (ps.jail) {
1059 			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1060 		}
1061 		sbuf_printf(procbuf, "%-*.*s", namelength, namelength, (*get_userid)(pp->ki_ruid));
1062 		sbuf_printf(procbuf, "%6ld ", rup->ru_nvcsw);
1063 		sbuf_printf(procbuf, "%6ld ", rup->ru_nivcsw);
1064 		sbuf_printf(procbuf, "%6ld ", rup->ru_inblock);
1065 		sbuf_printf(procbuf, "%6ld ", rup->ru_oublock);
1066 		sbuf_printf(procbuf, "%6ld ", rup->ru_majflt);
1067 		sbuf_printf(procbuf, "%6ld ", p_tot);
1068 		sbuf_printf(procbuf, "%6.2f%% ", s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot));
1069 
1070 	} else {
1071 		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1072 		if (ps.jail) {
1073 			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1074 		}
1075 		sbuf_printf(procbuf, "%-*.*s ", namelength, namelength, (*get_userid)(pp->ki_ruid));
1076 
1077 		if (!ps.thread) {
1078 			sbuf_printf(procbuf, "%4d ", pp->ki_numthreads);
1079 		} else {
1080 			sbuf_printf(procbuf, " ");
1081 		}
1082 
1083 		sbuf_printf(procbuf, "%3d ", pp->ki_pri.pri_level - PZERO);
1084 		sbuf_printf(procbuf, "%4s", format_nice(pp));
1085 		sbuf_printf(procbuf, "%7s ", format_k(PROCSIZE(pp)));
1086 		sbuf_printf(procbuf, "%6s ", format_k(pagetok(pp->ki_rssize)));
1087 		if (ps.swap) {
1088 			sbuf_printf(procbuf, "%*s ",
1089 				TOP_SWAP_LEN - 1,
1090 				format_k(pagetok(ki_swap(pp))));
1091 		}
1092 		sbuf_printf(procbuf, "%-6.6s ", status);
1093 		if (smpmode) {
1094 			int cpu;
1095 			if (state == SRUN && pp->ki_oncpu != NOCPU) {
1096 				cpu = pp->ki_oncpu;
1097 			} else {
1098 				cpu = pp->ki_lastcpu;
1099 			}
1100 			sbuf_printf(procbuf, "%3d ", cpu);
1101 		}
1102 		sbuf_printf(procbuf, "%6s ", format_time(cputime));
1103 		sbuf_printf(procbuf, "%6.2f%% ", ps.wcpu ? 100.0 * weighted_cpu(PCTCPU(pp), pp) : 100.0 * PCTCPU(pp));
1104 	}
1105 	sbuf_printf(procbuf, "%s", cmdbuf);
1106 	free(cmdbuf);
1107 	return (sbuf_data(procbuf));
1108 }
1109 
1110 static void
1111 getsysctl(const char *name, void *ptr, size_t len)
1112 {
1113 	size_t nlen = len;
1114 
1115 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1116 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1117 		    strerror(errno));
1118 		quit(TOP_EX_SYS_ERROR);
1119 	}
1120 	if (nlen != len) {
1121 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1122 		    name, (unsigned long)len, (unsigned long)nlen);
1123 		quit(TOP_EX_SYS_ERROR);
1124 	}
1125 }
1126 
1127 static const char *
1128 format_nice(const struct kinfo_proc *pp)
1129 {
1130 	const char *fifo, *kproc;
1131 	int rtpri;
1132 	static char nicebuf[4 + 1];
1133 
1134 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1135 	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1136 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1137 	case PRI_ITHD:
1138 		return ("-");
1139 	case PRI_REALTIME:
1140 		/*
1141 		 * XXX: the kernel doesn't tell us the original rtprio and
1142 		 * doesn't really know what it was, so to recover it we
1143 		 * must be more chummy with the implementation than the
1144 		 * implementation is with itself.  pri_user gives a
1145 		 * constant "base" priority, but is only initialized
1146 		 * properly for user threads.  pri_native gives what the
1147 		 * kernel calls the "base" priority, but it isn't constant
1148 		 * since it is changed by priority propagation.  pri_native
1149 		 * also isn't properly initialized for all threads, but it
1150 		 * is properly initialized for kernel realtime and idletime
1151 		 * threads.  Thus we use pri_user for the base priority of
1152 		 * user threads (it is always correct) and pri_native for
1153 		 * the base priority of kernel realtime and idletime threads
1154 		 * (there is nothing better, and it is usually correct).
1155 		 *
1156 		 * The field width and thus the buffer are too small for
1157 		 * values like "kr31F", but such values shouldn't occur,
1158 		 * and if they do then the tailing "F" is not displayed.
1159 		 */
1160 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1161 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1162 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1163 		    kproc, rtpri, fifo);
1164 		break;
1165 	case PRI_TIMESHARE:
1166 		if (pp->ki_flag & P_KPROC)
1167 			return ("-");
1168 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1169 		break;
1170 	case PRI_IDLE:
1171 		/* XXX: as above. */
1172 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1173 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1174 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1175 		    kproc, rtpri, fifo);
1176 		break;
1177 	default:
1178 		return ("?");
1179 	}
1180 	return (nicebuf);
1181 }
1182 
1183 /* comparison routines for qsort */
1184 
1185 static int
1186 compare_pid(const void *p1, const void *p2)
1187 {
1188 	const struct kinfo_proc * const *pp1 = p1;
1189 	const struct kinfo_proc * const *pp2 = p2;
1190 
1191 	assert((*pp2)->ki_pid >= 0 && (*pp1)->ki_pid >= 0);
1192 
1193 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1194 }
1195 
1196 static int
1197 compare_tid(const void *p1, const void *p2)
1198 {
1199 	const struct kinfo_proc * const *pp1 = p1;
1200 	const struct kinfo_proc * const *pp2 = p2;
1201 
1202 	assert((*pp2)->ki_tid >= 0 && (*pp1)->ki_tid >= 0);
1203 
1204 	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1205 }
1206 
1207 /*
1208  *  proc_compare - comparison function for "qsort"
1209  *	Compares the resource consumption of two processes using five
1210  *	distinct keys.  The keys (in descending order of importance) are:
1211  *	percent cpu, cpu ticks, state, resident set size, total virtual
1212  *	memory usage.  The process states are ordered as follows (from least
1213  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1214  *	array declaration below maps a process state index into a number
1215  *	that reflects this ordering.
1216  */
1217 
1218 static int sorted_state[] = {
1219 	0,	/* not used		*/
1220 	3,	/* sleep		*/
1221 	1,	/* ABANDONED (WAIT)	*/
1222 	6,	/* run			*/
1223 	5,	/* start		*/
1224 	2,	/* zombie		*/
1225 	4	/* stop			*/
1226 };
1227 
1228 
1229 #define ORDERKEY_PCTCPU(a, b) do { \
1230 	double diff; \
1231 	if (ps.wcpu) \
1232 		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1233 		    weighted_cpu(PCTCPU((a)), (a)); \
1234 	else \
1235 		diff = PCTCPU((b)) - PCTCPU((a)); \
1236 	if (diff != 0) \
1237 		return (diff > 0 ? 1 : -1); \
1238 } while (0)
1239 
1240 #define ORDERKEY_CPTICKS(a, b) do { \
1241 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1242 	if (diff != 0) \
1243 		return (diff > 0 ? 1 : -1); \
1244 } while (0)
1245 
1246 #define ORDERKEY_STATE(a, b) do { \
1247 	int diff = sorted_state[(unsigned char)(b)->ki_stat] - sorted_state[(unsigned char)(a)->ki_stat]; \
1248 	if (diff != 0) \
1249 		return (diff > 0 ? 1 : -1); \
1250 } while (0)
1251 
1252 #define ORDERKEY_PRIO(a, b) do { \
1253 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1254 	if (diff != 0) \
1255 		return (diff > 0 ? 1 : -1); \
1256 } while (0)
1257 
1258 #define	ORDERKEY_THREADS(a, b) do { \
1259 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1260 	if (diff != 0) \
1261 		return (diff > 0 ? 1 : -1); \
1262 } while (0)
1263 
1264 #define ORDERKEY_RSSIZE(a, b) do { \
1265 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1266 	if (diff != 0) \
1267 		return (diff > 0 ? 1 : -1); \
1268 } while (0)
1269 
1270 #define ORDERKEY_MEM(a, b) do { \
1271 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1272 	if (diff != 0) \
1273 		return (diff > 0 ? 1 : -1); \
1274 } while (0)
1275 
1276 #define ORDERKEY_JID(a, b) do { \
1277 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1278 	if (diff != 0) \
1279 		return (diff > 0 ? 1 : -1); \
1280 } while (0)
1281 
1282 #define ORDERKEY_SWAP(a, b) do { \
1283 	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1284 	if (diff != 0) \
1285 		return (diff > 0 ? 1 : -1); \
1286 } while (0)
1287 
1288 /* compare_cpu - the comparison function for sorting by cpu percentage */
1289 
1290 static int
1291 compare_cpu(const void *arg1, const void *arg2)
1292 {
1293 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1294 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1295 
1296 	ORDERKEY_PCTCPU(p1, p2);
1297 	ORDERKEY_CPTICKS(p1, p2);
1298 	ORDERKEY_STATE(p1, p2);
1299 	ORDERKEY_PRIO(p1, p2);
1300 	ORDERKEY_RSSIZE(p1, p2);
1301 	ORDERKEY_MEM(p1, p2);
1302 
1303 	return (0);
1304 }
1305 
1306 /* compare_size - the comparison function for sorting by total memory usage */
1307 
1308 static int
1309 compare_size(const void *arg1, const void *arg2)
1310 {
1311 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1312 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1313 
1314 	ORDERKEY_MEM(p1, p2);
1315 	ORDERKEY_RSSIZE(p1, p2);
1316 	ORDERKEY_PCTCPU(p1, p2);
1317 	ORDERKEY_CPTICKS(p1, p2);
1318 	ORDERKEY_STATE(p1, p2);
1319 	ORDERKEY_PRIO(p1, p2);
1320 
1321 	return (0);
1322 }
1323 
1324 /* compare_res - the comparison function for sorting by resident set size */
1325 
1326 static int
1327 compare_res(const void *arg1, const void *arg2)
1328 {
1329 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1330 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1331 
1332 	ORDERKEY_RSSIZE(p1, p2);
1333 	ORDERKEY_MEM(p1, p2);
1334 	ORDERKEY_PCTCPU(p1, p2);
1335 	ORDERKEY_CPTICKS(p1, p2);
1336 	ORDERKEY_STATE(p1, p2);
1337 	ORDERKEY_PRIO(p1, p2);
1338 
1339 	return (0);
1340 }
1341 
1342 /* compare_time - the comparison function for sorting by total cpu time */
1343 
1344 static int
1345 compare_time(const void *arg1, const void *arg2)
1346 {
1347 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const  *)arg1;
1348 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *) arg2;
1349 
1350 	ORDERKEY_CPTICKS(p1, p2);
1351 	ORDERKEY_PCTCPU(p1, p2);
1352 	ORDERKEY_STATE(p1, p2);
1353 	ORDERKEY_PRIO(p1, p2);
1354 	ORDERKEY_RSSIZE(p1, p2);
1355 	ORDERKEY_MEM(p1, p2);
1356 
1357 	return (0);
1358 }
1359 
1360 /* compare_prio - the comparison function for sorting by priority */
1361 
1362 static int
1363 compare_prio(const void *arg1, const void *arg2)
1364 {
1365 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1366 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1367 
1368 	ORDERKEY_PRIO(p1, p2);
1369 	ORDERKEY_CPTICKS(p1, p2);
1370 	ORDERKEY_PCTCPU(p1, p2);
1371 	ORDERKEY_STATE(p1, p2);
1372 	ORDERKEY_RSSIZE(p1, p2);
1373 	ORDERKEY_MEM(p1, p2);
1374 
1375 	return (0);
1376 }
1377 
1378 /* compare_threads - the comparison function for sorting by threads */
1379 static int
1380 compare_threads(const void *arg1, const void *arg2)
1381 {
1382 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1383 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1384 
1385 	ORDERKEY_THREADS(p1, p2);
1386 	ORDERKEY_PCTCPU(p1, p2);
1387 	ORDERKEY_CPTICKS(p1, p2);
1388 	ORDERKEY_STATE(p1, p2);
1389 	ORDERKEY_PRIO(p1, p2);
1390 	ORDERKEY_RSSIZE(p1, p2);
1391 	ORDERKEY_MEM(p1, p2);
1392 
1393 	return (0);
1394 }
1395 
1396 /* compare_jid - the comparison function for sorting by jid */
1397 static int
1398 compare_jid(const void *arg1, const void *arg2)
1399 {
1400 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1401 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1402 
1403 	ORDERKEY_JID(p1, p2);
1404 	ORDERKEY_PCTCPU(p1, p2);
1405 	ORDERKEY_CPTICKS(p1, p2);
1406 	ORDERKEY_STATE(p1, p2);
1407 	ORDERKEY_PRIO(p1, p2);
1408 	ORDERKEY_RSSIZE(p1, p2);
1409 	ORDERKEY_MEM(p1, p2);
1410 
1411 	return (0);
1412 }
1413 
1414 /* compare_swap - the comparison function for sorting by swap */
1415 static int
1416 compare_swap(const void *arg1, const void *arg2)
1417 {
1418 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1419 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1420 
1421 	ORDERKEY_SWAP(p1, p2);
1422 	ORDERKEY_PCTCPU(p1, p2);
1423 	ORDERKEY_CPTICKS(p1, p2);
1424 	ORDERKEY_STATE(p1, p2);
1425 	ORDERKEY_PRIO(p1, p2);
1426 	ORDERKEY_RSSIZE(p1, p2);
1427 	ORDERKEY_MEM(p1, p2);
1428 
1429 	return (0);
1430 }
1431 
1432 /* assorted comparison functions for sorting by i/o */
1433 
1434 static int
1435 compare_iototal(const void *arg1, const void *arg2)
1436 {
1437 	const struct kinfo_proc * const p1 = *(const struct kinfo_proc * const *)arg1;
1438 	const struct kinfo_proc * const p2 = *(const struct kinfo_proc * const *)arg2;
1439 
1440 	return (get_io_total(p2) - get_io_total(p1));
1441 }
1442 
1443 static int
1444 compare_ioread(const void *arg1, const void *arg2)
1445 {
1446 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1447 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1448 	long dummy, inp1, inp2;
1449 
1450 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1451 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1452 
1453 	return (inp2 - inp1);
1454 }
1455 
1456 static int
1457 compare_iowrite(const void *arg1, const void *arg2)
1458 {
1459 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1460 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1461 	long dummy, oup1, oup2;
1462 
1463 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1464 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1465 
1466 	return (oup2 - oup1);
1467 }
1468 
1469 static int
1470 compare_iofault(const void *arg1, const void *arg2)
1471 {
1472 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1473 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1474 	long dummy, flp1, flp2;
1475 
1476 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1477 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1478 
1479 	return (flp2 - flp1);
1480 }
1481 
1482 static int
1483 compare_vcsw(const void *arg1, const void *arg2)
1484 {
1485 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1486 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1487 	long dummy, flp1, flp2;
1488 
1489 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1490 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1491 
1492 	return (flp2 - flp1);
1493 }
1494 
1495 static int
1496 compare_ivcsw(const void *arg1, const void *arg2)
1497 {
1498 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1499 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1500 	long dummy, flp1, flp2;
1501 
1502 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1503 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1504 
1505 	return (flp2 - flp1);
1506 }
1507 
1508 int (*compares[])(const void *arg1, const void *arg2) = {
1509 	compare_cpu,
1510 	compare_size,
1511 	compare_res,
1512 	compare_time,
1513 	compare_prio,
1514 	compare_threads,
1515 	compare_iototal,
1516 	compare_ioread,
1517 	compare_iowrite,
1518 	compare_iofault,
1519 	compare_vcsw,
1520 	compare_ivcsw,
1521 	compare_jid,
1522 	compare_swap,
1523 	NULL
1524 };
1525 
1526 
1527 static int
1528 swapmode(int *retavail, int *retfree)
1529 {
1530 	int n;
1531 	struct kvm_swap swapary[1];
1532 	static int pagesize = 0;
1533 	static unsigned long swap_maxpages = 0;
1534 
1535 	*retavail = 0;
1536 	*retfree = 0;
1537 
1538 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1539 
1540 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1541 	if (n < 0 || swapary[0].ksw_total == 0)
1542 		return (0);
1543 
1544 	if (pagesize == 0)
1545 		pagesize = getpagesize();
1546 	if (swap_maxpages == 0)
1547 		GETSYSCTL("vm.swap_maxpages", swap_maxpages);
1548 
1549 	/* ksw_total contains the total size of swap all devices which may
1550 	   exceed the maximum swap size allocatable in the system */
1551 	if ( swapary[0].ksw_total > swap_maxpages )
1552 		swapary[0].ksw_total = swap_maxpages;
1553 
1554 	*retavail = CONVERT(swapary[0].ksw_total);
1555 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1556 
1557 #undef CONVERT
1558 
1559 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1560 	return (n);
1561 }
1562