xref: /freebsd/usr.bin/top/machine.c (revision 3fc36ee018bb836bd1796067cf4ef8683f166ebc)
1 /*
2  * top - a top users display for Unix
3  *
4  * SYNOPSIS:  For FreeBSD-2.x and later
5  *
6  * DESCRIPTION:
7  * Originally written for BSD4.4 system by Christos Zoulas.
8  * Ported to FreeBSD 2.x by Steven Wallace && Wolfram Schneider
9  * Order support hacked in from top-3.5beta6/machine/m_aix41.c
10  *   by Monte Mitzelfelt (for latest top see http://www.groupsys.com/topinfo/)
11  *
12  * This is the machine-dependent module for FreeBSD 2.2
13  * Works for:
14  *	FreeBSD 2.2.x, 3.x, 4.x, and probably FreeBSD 2.1.x
15  *
16  * LIBS: -lkvm
17  *
18  * AUTHOR:  Christos Zoulas <christos@ee.cornell.edu>
19  *          Steven Wallace  <swallace@freebsd.org>
20  *          Wolfram Schneider <wosch@FreeBSD.org>
21  *          Thomas Moestl <tmoestl@gmx.net>
22  *
23  * $FreeBSD$
24  */
25 
26 #include <sys/param.h>
27 #include <sys/errno.h>
28 #include <sys/file.h>
29 #include <sys/proc.h>
30 #include <sys/resource.h>
31 #include <sys/rtprio.h>
32 #include <sys/signal.h>
33 #include <sys/sysctl.h>
34 #include <sys/time.h>
35 #include <sys/user.h>
36 #include <sys/vmmeter.h>
37 
38 #include <err.h>
39 #include <kvm.h>
40 #include <math.h>
41 #include <nlist.h>
42 #include <paths.h>
43 #include <pwd.h>
44 #include <stdio.h>
45 #include <stdlib.h>
46 #include <string.h>
47 #include <strings.h>
48 #include <unistd.h>
49 #include <vis.h>
50 
51 #include "top.h"
52 #include "machine.h"
53 #include "screen.h"
54 #include "utils.h"
55 #include "layout.h"
56 
57 #define GETSYSCTL(name, var) getsysctl(name, &(var), sizeof(var))
58 #define	SMPUNAMELEN	13
59 #define	UPUNAMELEN	15
60 
61 extern struct process_select ps;
62 extern char* printable(char *);
63 static int smpmode;
64 enum displaymodes displaymode;
65 #ifdef TOP_USERNAME_LEN
66 static int namelength = TOP_USERNAME_LEN;
67 #else
68 static int namelength = 8;
69 #endif
70 /* TOP_JID_LEN based on max of 999999 */
71 #define TOP_JID_LEN 7
72 #define TOP_SWAP_LEN 6
73 static int jidlength;
74 static int swaplength;
75 static int cmdlengthdelta;
76 
77 /* Prototypes for top internals */
78 void quit(int);
79 
80 /* get_process_info passes back a handle.  This is what it looks like: */
81 
82 struct handle {
83 	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
84 	int remaining;			/* number of pointers remaining */
85 };
86 
87 /* declarations for load_avg */
88 #include "loadavg.h"
89 
90 /* define what weighted cpu is.  */
91 #define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
92 			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
93 
94 /* what we consider to be process size: */
95 #define PROCSIZE(pp) ((pp)->ki_size / 1024)
96 
97 #define RU(pp)	(&(pp)->ki_rusage)
98 #define RUTOT(pp) \
99 	(RU(pp)->ru_inblock + RU(pp)->ru_oublock + RU(pp)->ru_majflt)
100 
101 #define	PCTCPU(pp) (pcpu[pp - pbase])
102 
103 /* definitions for indices in the nlist array */
104 
105 /*
106  *  These definitions control the format of the per-process area
107  */
108 
109 static char io_header[] =
110     "  PID%*s %-*.*s   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND";
111 
112 #define io_Proc_format \
113     "%5d%*s %-*.*s %6ld %6ld %6ld %6ld %6ld %6ld %6.2f%% %.*s"
114 
115 static char smp_header_thr[] =
116     "  PID%*s %-*.*s  THR PRI NICE   SIZE    RES%*s STATE   C   TIME %7s COMMAND";
117 static char smp_header[] =
118     "  PID%*s %-*.*s "   "PRI NICE   SIZE    RES%*s STATE   C   TIME %7s COMMAND";
119 
120 #define smp_Proc_format \
121     "%5d%*s %-*.*s %s%3d %4s%7s %6s%*.*s %-6.6s %2d%7s %6.2f%% %.*s"
122 
123 static char up_header_thr[] =
124     "  PID%*s %-*.*s  THR PRI NICE   SIZE    RES%*s STATE    TIME %7s COMMAND";
125 static char up_header[] =
126     "  PID%*s %-*.*s "   "PRI NICE   SIZE    RES%*s STATE    TIME %7s COMMAND";
127 
128 #define up_Proc_format \
129     "%5d%*s %-*.*s %s%3d %4s%7s %6s%*.*s %-6.6s%.0d%7s %6.2f%% %.*s"
130 
131 
132 /* process state names for the "STATE" column of the display */
133 /* the extra nulls in the string "run" are for adding a slash and
134    the processor number when needed */
135 
136 char *state_abbrev[] = {
137 	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
138 };
139 
140 
141 static kvm_t *kd;
142 
143 /* values that we stash away in _init and use in later routines */
144 
145 static double logcpu;
146 
147 /* these are retrieved from the kernel in _init */
148 
149 static load_avg  ccpu;
150 
151 /* these are used in the get_ functions */
152 
153 static int lastpid;
154 
155 /* these are for calculating cpu state percentages */
156 
157 static long cp_time[CPUSTATES];
158 static long cp_old[CPUSTATES];
159 static long cp_diff[CPUSTATES];
160 
161 /* these are for detailing the process states */
162 
163 int process_states[8];
164 char *procstatenames[] = {
165 	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
166 	" zombie, ", " waiting, ", " lock, ",
167 	NULL
168 };
169 
170 /* these are for detailing the cpu states */
171 
172 int cpu_states[CPUSTATES];
173 char *cpustatenames[] = {
174 	"user", "nice", "system", "interrupt", "idle", NULL
175 };
176 
177 /* these are for detailing the memory statistics */
178 
179 int memory_stats[7];
180 char *memorynames[] = {
181 	"K Active, ", "K Inact, ", "K Wired, ", "K Cache, ", "K Buf, ",
182 	"K Free", NULL
183 };
184 
185 int arc_stats[7];
186 char *arcnames[] = {
187 	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
188 	NULL
189 };
190 
191 int swap_stats[7];
192 char *swapnames[] = {
193 	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
194 	NULL
195 };
196 
197 
198 /* these are for keeping track of the proc array */
199 
200 static int nproc;
201 static int onproc = -1;
202 static int pref_len;
203 static struct kinfo_proc *pbase;
204 static struct kinfo_proc **pref;
205 static struct kinfo_proc *previous_procs;
206 static struct kinfo_proc **previous_pref;
207 static int previous_proc_count = 0;
208 static int previous_proc_count_max = 0;
209 static int previous_thread;
210 
211 /* data used for recalculating pctcpu */
212 static double *pcpu;
213 static struct timespec proc_uptime;
214 static struct timeval proc_wall_time;
215 static struct timeval previous_wall_time;
216 static uint64_t previous_interval = 0;
217 
218 /* total number of io operations */
219 static long total_inblock;
220 static long total_oublock;
221 static long total_majflt;
222 
223 /* these are for getting the memory statistics */
224 
225 static int arc_enabled;
226 static int pageshift;		/* log base 2 of the pagesize */
227 
228 /* define pagetok in terms of pageshift */
229 
230 #define pagetok(size) ((size) << pageshift)
231 
232 /* swap usage */
233 #define ki_swap(kip) \
234     ((kip)->ki_swrss > (kip)->ki_rssize ? (kip)->ki_swrss - (kip)->ki_rssize : 0)
235 
236 /* useful externals */
237 long percentages();
238 
239 #ifdef ORDER
240 /*
241  * Sorting orders.  The first element is the default.
242  */
243 char *ordernames[] = {
244 	"cpu", "size", "res", "time", "pri", "threads",
245 	"total", "read", "write", "fault", "vcsw", "ivcsw",
246 	"jid", "swap", "pid", NULL
247 };
248 #endif
249 
250 /* Per-cpu time states */
251 static int maxcpu;
252 static int maxid;
253 static int ncpus;
254 static u_long cpumask;
255 static long *times;
256 static long *pcpu_cp_time;
257 static long *pcpu_cp_old;
258 static long *pcpu_cp_diff;
259 static int *pcpu_cpu_states;
260 
261 static int compare_swap(const void *a, const void *b);
262 static int compare_jid(const void *a, const void *b);
263 static int compare_pid(const void *a, const void *b);
264 static int compare_tid(const void *a, const void *b);
265 static const char *format_nice(const struct kinfo_proc *pp);
266 static void getsysctl(const char *name, void *ptr, size_t len);
267 static int swapmode(int *retavail, int *retfree);
268 static void update_layout(void);
269 
270 void
271 toggle_pcpustats(void)
272 {
273 
274 	if (ncpus == 1)
275 		return;
276 	update_layout();
277 }
278 
279 /* Adjust display based on ncpus and the ARC state. */
280 static void
281 update_layout(void)
282 {
283 
284 	y_mem = 3;
285 	y_arc = 4;
286 	y_swap = 4 + arc_enabled;
287 	y_idlecursor = 5 + arc_enabled;
288 	y_message = 5 + arc_enabled;
289 	y_header = 6 + arc_enabled;
290 	y_procs = 7 + arc_enabled;
291 	Header_lines = 7 + arc_enabled;
292 
293 	if (pcpu_stats) {
294 		y_mem += ncpus - 1;
295 		y_arc += ncpus - 1;
296 		y_swap += ncpus - 1;
297 		y_idlecursor += ncpus - 1;
298 		y_message += ncpus - 1;
299 		y_header += ncpus - 1;
300 		y_procs += ncpus - 1;
301 		Header_lines += ncpus - 1;
302 	}
303 }
304 
305 int
306 machine_init(struct statics *statics, char do_unames)
307 {
308 	int i, j, empty, pagesize;
309 	uint64_t arc_size;
310 	size_t size;
311 	struct passwd *pw;
312 
313 	size = sizeof(smpmode);
314 	if ((sysctlbyname("machdep.smp_active", &smpmode, &size,
315 	    NULL, 0) != 0 &&
316 	    sysctlbyname("kern.smp.active", &smpmode, &size,
317 	    NULL, 0) != 0) ||
318 	    size != sizeof(smpmode))
319 		smpmode = 0;
320 
321 	size = sizeof(arc_size);
322 	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
323 	    NULL, 0) == 0 && arc_size != 0)
324 		arc_enabled = 1;
325 
326 	if (do_unames) {
327 	    while ((pw = getpwent()) != NULL) {
328 		if (strlen(pw->pw_name) > namelength)
329 			namelength = strlen(pw->pw_name);
330 	    }
331 	}
332 	if (smpmode && namelength > SMPUNAMELEN)
333 		namelength = SMPUNAMELEN;
334 	else if (namelength > UPUNAMELEN)
335 		namelength = UPUNAMELEN;
336 
337 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
338 	if (kd == NULL)
339 		return (-1);
340 
341 	GETSYSCTL("kern.ccpu", ccpu);
342 
343 	/* this is used in calculating WCPU -- calculate it ahead of time */
344 	logcpu = log(loaddouble(ccpu));
345 
346 	pbase = NULL;
347 	pref = NULL;
348 	pcpu = NULL;
349 	nproc = 0;
350 	onproc = -1;
351 
352 	/* get the page size and calculate pageshift from it */
353 	pagesize = getpagesize();
354 	pageshift = 0;
355 	while (pagesize > 1) {
356 		pageshift++;
357 		pagesize >>= 1;
358 	}
359 
360 	/* we only need the amount of log(2)1024 for our conversion */
361 	pageshift -= LOG1024;
362 
363 	/* fill in the statics information */
364 	statics->procstate_names = procstatenames;
365 	statics->cpustate_names = cpustatenames;
366 	statics->memory_names = memorynames;
367 	if (arc_enabled)
368 		statics->arc_names = arcnames;
369 	else
370 		statics->arc_names = NULL;
371 	statics->swap_names = swapnames;
372 #ifdef ORDER
373 	statics->order_names = ordernames;
374 #endif
375 
376 	/* Allocate state for per-CPU stats. */
377 	cpumask = 0;
378 	ncpus = 0;
379 	GETSYSCTL("kern.smp.maxcpus", maxcpu);
380 	size = sizeof(long) * maxcpu * CPUSTATES;
381 	times = malloc(size);
382 	if (times == NULL)
383 		err(1, "malloc %zu bytes", size);
384 	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
385 		err(1, "sysctlbyname kern.cp_times");
386 	pcpu_cp_time = calloc(1, size);
387 	maxid = (size / CPUSTATES / sizeof(long)) - 1;
388 	for (i = 0; i <= maxid; i++) {
389 		empty = 1;
390 		for (j = 0; empty && j < CPUSTATES; j++) {
391 			if (times[i * CPUSTATES + j] != 0)
392 				empty = 0;
393 		}
394 		if (!empty) {
395 			cpumask |= (1ul << i);
396 			ncpus++;
397 		}
398 	}
399 	size = sizeof(long) * ncpus * CPUSTATES;
400 	pcpu_cp_old = calloc(1, size);
401 	pcpu_cp_diff = calloc(1, size);
402 	pcpu_cpu_states = calloc(1, size);
403 	statics->ncpus = ncpus;
404 
405 	update_layout();
406 
407 	/* all done! */
408 	return (0);
409 }
410 
411 char *
412 format_header(char *uname_field)
413 {
414 	static char Header[128];
415 	const char *prehead;
416 
417 	if (ps.jail)
418 		jidlength = TOP_JID_LEN + 1;	/* +1 for extra left space. */
419 	else
420 		jidlength = 0;
421 
422 	if (ps.swap)
423 		swaplength = TOP_SWAP_LEN + 1;  /* +1 for extra left space */
424 	else
425 		swaplength = 0;
426 
427 	switch (displaymode) {
428 	case DISP_CPU:
429 		/*
430 		 * The logic of picking the right header format seems reverse
431 		 * here because we only want to display a THR column when
432 		 * "thread mode" is off (and threads are not listed as
433 		 * separate lines).
434 		 */
435 		prehead = smpmode ?
436 		    (ps.thread ? smp_header : smp_header_thr) :
437 		    (ps.thread ? up_header : up_header_thr);
438 		snprintf(Header, sizeof(Header), prehead,
439 		    jidlength, ps.jail ? " JID" : "",
440 		    namelength, namelength, uname_field,
441 		    swaplength, ps.swap ? " SWAP" : "",
442 		    ps.wcpu ? "WCPU" : "CPU");
443 		break;
444 	case DISP_IO:
445 		prehead = io_header;
446 		snprintf(Header, sizeof(Header), prehead,
447 		    jidlength, ps.jail ? " JID" : "",
448 		    namelength, namelength, uname_field);
449 		break;
450 	}
451 	cmdlengthdelta = strlen(Header) - 7;
452 	return (Header);
453 }
454 
455 static int swappgsin = -1;
456 static int swappgsout = -1;
457 extern struct timeval timeout;
458 
459 
460 void
461 get_system_info(struct system_info *si)
462 {
463 	long total;
464 	struct loadavg sysload;
465 	int mib[2];
466 	struct timeval boottime;
467 	uint64_t arc_stat, arc_stat2;
468 	int i, j;
469 	size_t size;
470 
471 	/* get the CPU stats */
472 	size = (maxid + 1) * CPUSTATES * sizeof(long);
473 	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
474 		err(1, "sysctlbyname kern.cp_times");
475 	GETSYSCTL("kern.cp_time", cp_time);
476 	GETSYSCTL("vm.loadavg", sysload);
477 	GETSYSCTL("kern.lastpid", lastpid);
478 
479 	/* convert load averages to doubles */
480 	for (i = 0; i < 3; i++)
481 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
482 
483 	/* convert cp_time counts to percentages */
484 	for (i = j = 0; i <= maxid; i++) {
485 		if ((cpumask & (1ul << i)) == 0)
486 			continue;
487 		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
488 		    &pcpu_cp_time[j * CPUSTATES],
489 		    &pcpu_cp_old[j * CPUSTATES],
490 		    &pcpu_cp_diff[j * CPUSTATES]);
491 		j++;
492 	}
493 	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
494 
495 	/* sum memory & swap statistics */
496 	{
497 		static unsigned int swap_delay = 0;
498 		static int swapavail = 0;
499 		static int swapfree = 0;
500 		static long bufspace = 0;
501 		static int nspgsin, nspgsout;
502 
503 		GETSYSCTL("vfs.bufspace", bufspace);
504 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
505 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
506 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[2]);
507 		GETSYSCTL("vm.stats.vm.v_cache_count", memory_stats[3]);
508 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
509 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
510 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
511 		/* convert memory stats to Kbytes */
512 		memory_stats[0] = pagetok(memory_stats[0]);
513 		memory_stats[1] = pagetok(memory_stats[1]);
514 		memory_stats[2] = pagetok(memory_stats[2]);
515 		memory_stats[3] = pagetok(memory_stats[3]);
516 		memory_stats[4] = bufspace / 1024;
517 		memory_stats[5] = pagetok(memory_stats[5]);
518 		memory_stats[6] = -1;
519 
520 		/* first interval */
521 		if (swappgsin < 0) {
522 			swap_stats[4] = 0;
523 			swap_stats[5] = 0;
524 		}
525 
526 		/* compute differences between old and new swap statistic */
527 		else {
528 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
529 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
530 		}
531 
532 		swappgsin = nspgsin;
533 		swappgsout = nspgsout;
534 
535 		/* call CPU heavy swapmode() only for changes */
536 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
537 			swap_stats[3] = swapmode(&swapavail, &swapfree);
538 			swap_stats[0] = swapavail;
539 			swap_stats[1] = swapavail - swapfree;
540 			swap_stats[2] = swapfree;
541 		}
542 		swap_delay = 1;
543 		swap_stats[6] = -1;
544 	}
545 
546 	if (arc_enabled) {
547 		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
548 		arc_stats[0] = arc_stat >> 10;
549 		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
550 		arc_stats[1] = arc_stat >> 10;
551 		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
552 		arc_stats[2] = arc_stat >> 10;
553 		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
554 		arc_stats[3] = arc_stat >> 10;
555 		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
556 		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
557 		arc_stats[4] = arc_stat + arc_stat2 >> 10;
558 		GETSYSCTL("kstat.zfs.misc.arcstats.other_size", arc_stat);
559 		arc_stats[5] = arc_stat >> 10;
560 		si->arc = arc_stats;
561 	}
562 
563 	/* set arrays and strings */
564 	if (pcpu_stats) {
565 		si->cpustates = pcpu_cpu_states;
566 		si->ncpus = ncpus;
567 	} else {
568 		si->cpustates = cpu_states;
569 		si->ncpus = 1;
570 	}
571 	si->memory = memory_stats;
572 	si->swap = swap_stats;
573 
574 
575 	if (lastpid > 0) {
576 		si->last_pid = lastpid;
577 	} else {
578 		si->last_pid = -1;
579 	}
580 
581 	/*
582 	 * Print how long system has been up.
583 	 * (Found by looking getting "boottime" from the kernel)
584 	 */
585 	mib[0] = CTL_KERN;
586 	mib[1] = KERN_BOOTTIME;
587 	size = sizeof(boottime);
588 	if (sysctl(mib, 2, &boottime, &size, NULL, 0) != -1 &&
589 	    boottime.tv_sec != 0) {
590 		si->boottime = boottime;
591 	} else {
592 		si->boottime.tv_sec = -1;
593 	}
594 }
595 
596 #define NOPROC	((void *)-1)
597 
598 /*
599  * We need to compare data from the old process entry with the new
600  * process entry.
601  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
602  * structure to cache the mapping.  We also use a negative cache pointer
603  * of NOPROC to avoid duplicate lookups.
604  * XXX: this could be done when the actual processes are fetched, we do
605  * it here out of laziness.
606  */
607 const struct kinfo_proc *
608 get_old_proc(struct kinfo_proc *pp)
609 {
610 	struct kinfo_proc **oldpp, *oldp;
611 
612 	/*
613 	 * If this is the first fetch of the kinfo_procs then we don't have
614 	 * any previous entries.
615 	 */
616 	if (previous_proc_count == 0)
617 		return (NULL);
618 	/* negative cache? */
619 	if (pp->ki_udata == NOPROC)
620 		return (NULL);
621 	/* cached? */
622 	if (pp->ki_udata != NULL)
623 		return (pp->ki_udata);
624 	/*
625 	 * Not cached,
626 	 * 1) look up based on pid.
627 	 * 2) compare process start.
628 	 * If we fail here, then setup a negative cache entry, otherwise
629 	 * cache it.
630 	 */
631 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
632 	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
633 	if (oldpp == NULL) {
634 		pp->ki_udata = NOPROC;
635 		return (NULL);
636 	}
637 	oldp = *oldpp;
638 	if (bcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
639 		pp->ki_udata = NOPROC;
640 		return (NULL);
641 	}
642 	pp->ki_udata = oldp;
643 	return (oldp);
644 }
645 
646 /*
647  * Return the total amount of IO done in blocks in/out and faults.
648  * store the values individually in the pointers passed in.
649  */
650 long
651 get_io_stats(struct kinfo_proc *pp, long *inp, long *oup, long *flp,
652     long *vcsw, long *ivcsw)
653 {
654 	const struct kinfo_proc *oldp;
655 	static struct kinfo_proc dummy;
656 	long ret;
657 
658 	oldp = get_old_proc(pp);
659 	if (oldp == NULL) {
660 		bzero(&dummy, sizeof(dummy));
661 		oldp = &dummy;
662 	}
663 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
664 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
665 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
666 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
667 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
668 	ret =
669 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
670 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
671 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
672 	return (ret);
673 }
674 
675 /*
676  * If there was a previous update, use the delta in ki_runtime over
677  * the previous interval to calculate pctcpu.  Otherwise, fall back
678  * to using the kernel's ki_pctcpu.
679  */
680 static double
681 proc_calc_pctcpu(struct kinfo_proc *pp)
682 {
683 	const struct kinfo_proc *oldp;
684 
685 	if (previous_interval != 0) {
686 		oldp = get_old_proc(pp);
687 		if (oldp != NULL)
688 			return ((double)(pp->ki_runtime - oldp->ki_runtime)
689 			    / previous_interval);
690 
691 		/*
692 		 * If this process/thread was created during the previous
693 		 * interval, charge it's total runtime to the previous
694 		 * interval.
695 		 */
696 		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
697 		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
698 		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
699 			return ((double)pp->ki_runtime / previous_interval);
700 	}
701 	return (pctdouble(pp->ki_pctcpu));
702 }
703 
704 /*
705  * Return true if this process has used any CPU time since the
706  * previous update.
707  */
708 static int
709 proc_used_cpu(struct kinfo_proc *pp)
710 {
711 	const struct kinfo_proc *oldp;
712 
713 	oldp = get_old_proc(pp);
714 	if (oldp == NULL)
715 		return (PCTCPU(pp) != 0);
716 	return (pp->ki_runtime != oldp->ki_runtime ||
717 	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
718 	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
719 }
720 
721 /*
722  * Return the total number of block in/out and faults by a process.
723  */
724 long
725 get_io_total(struct kinfo_proc *pp)
726 {
727 	long dummy;
728 
729 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
730 }
731 
732 static struct handle handle;
733 
734 caddr_t
735 get_process_info(struct system_info *si, struct process_select *sel,
736     int (*compare)(const void *, const void *))
737 {
738 	int i;
739 	int total_procs;
740 	long p_io;
741 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
742 	long nsec;
743 	int active_procs;
744 	struct kinfo_proc **prefp;
745 	struct kinfo_proc *pp;
746 	struct timespec previous_proc_uptime;
747 
748 	/* these are copied out of sel for speed */
749 	int show_idle;
750 	int show_jid;
751 	int show_self;
752 	int show_system;
753 	int show_uid;
754 	int show_command;
755 	int show_kidle;
756 
757 	/*
758 	 * If thread state was toggled, don't cache the previous processes.
759 	 */
760 	if (previous_thread != sel->thread)
761 		nproc = 0;
762 	previous_thread = sel->thread;
763 
764 	/*
765 	 * Save the previous process info.
766 	 */
767 	if (previous_proc_count_max < nproc) {
768 		free(previous_procs);
769 		previous_procs = malloc(nproc * sizeof(*previous_procs));
770 		free(previous_pref);
771 		previous_pref = malloc(nproc * sizeof(*previous_pref));
772 		if (previous_procs == NULL || previous_pref == NULL) {
773 			(void) fprintf(stderr, "top: Out of memory.\n");
774 			quit(23);
775 		}
776 		previous_proc_count_max = nproc;
777 	}
778 	if (nproc) {
779 		for (i = 0; i < nproc; i++)
780 			previous_pref[i] = &previous_procs[i];
781 		bcopy(pbase, previous_procs, nproc * sizeof(*previous_procs));
782 		qsort(previous_pref, nproc, sizeof(*previous_pref),
783 		    ps.thread ? compare_tid : compare_pid);
784 	}
785 	previous_proc_count = nproc;
786 	previous_proc_uptime = proc_uptime;
787 	previous_wall_time = proc_wall_time;
788 	previous_interval = 0;
789 
790 	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
791 	    0, &nproc);
792 	(void)gettimeofday(&proc_wall_time, NULL);
793 	if (clock_gettime(CLOCK_UPTIME, &proc_uptime) != 0)
794 		memset(&proc_uptime, 0, sizeof(proc_uptime));
795 	else if (previous_proc_uptime.tv_sec != 0 &&
796 	    previous_proc_uptime.tv_nsec != 0) {
797 		previous_interval = (proc_uptime.tv_sec -
798 		    previous_proc_uptime.tv_sec) * 1000000;
799 		nsec = proc_uptime.tv_nsec - previous_proc_uptime.tv_nsec;
800 		if (nsec < 0) {
801 			previous_interval -= 1000000;
802 			nsec += 1000000000;
803 		}
804 		previous_interval += nsec / 1000;
805 	}
806 	if (nproc > onproc) {
807 		pref = realloc(pref, sizeof(*pref) * nproc);
808 		pcpu = realloc(pcpu, sizeof(*pcpu) * nproc);
809 		onproc = nproc;
810 	}
811 	if (pref == NULL || pbase == NULL || pcpu == NULL) {
812 		(void) fprintf(stderr, "top: Out of memory.\n");
813 		quit(23);
814 	}
815 	/* get a pointer to the states summary array */
816 	si->procstates = process_states;
817 
818 	/* set up flags which define what we are going to select */
819 	show_idle = sel->idle;
820 	show_jid = sel->jid != -1;
821 	show_self = sel->self == -1;
822 	show_system = sel->system;
823 	show_uid = sel->uid != -1;
824 	show_command = sel->command != NULL;
825 	show_kidle = sel->kidle;
826 
827 	/* count up process states and get pointers to interesting procs */
828 	total_procs = 0;
829 	active_procs = 0;
830 	total_inblock = 0;
831 	total_oublock = 0;
832 	total_majflt = 0;
833 	memset((char *)process_states, 0, sizeof(process_states));
834 	prefp = pref;
835 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
836 
837 		if (pp->ki_stat == 0)
838 			/* not in use */
839 			continue;
840 
841 		if (!show_self && pp->ki_pid == sel->self)
842 			/* skip self */
843 			continue;
844 
845 		if (!show_system && (pp->ki_flag & P_SYSTEM))
846 			/* skip system process */
847 			continue;
848 
849 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
850 		    &p_vcsw, &p_ivcsw);
851 		total_inblock += p_inblock;
852 		total_oublock += p_oublock;
853 		total_majflt += p_majflt;
854 		total_procs++;
855 		process_states[pp->ki_stat]++;
856 
857 		if (pp->ki_stat == SZOMB)
858 			/* skip zombies */
859 			continue;
860 
861 		if (!show_kidle && pp->ki_tdflags & TDF_IDLETD)
862 			/* skip kernel idle process */
863 			continue;
864 
865 		PCTCPU(pp) = proc_calc_pctcpu(pp);
866 		if (sel->thread && PCTCPU(pp) > 1.0)
867 			PCTCPU(pp) = 1.0;
868 		if (displaymode == DISP_CPU && !show_idle &&
869 		    (!proc_used_cpu(pp) ||
870 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
871 			/* skip idle or non-running processes */
872 			continue;
873 
874 		if (displaymode == DISP_IO && !show_idle && p_io == 0)
875 			/* skip processes that aren't doing I/O */
876 			continue;
877 
878 		if (show_jid && pp->ki_jid != sel->jid)
879 			/* skip proc. that don't belong to the selected JID */
880 			continue;
881 
882 		if (show_uid && pp->ki_ruid != (uid_t)sel->uid)
883 			/* skip proc. that don't belong to the selected UID */
884 			continue;
885 
886 		*prefp++ = pp;
887 		active_procs++;
888 	}
889 
890 	/* if requested, sort the "interesting" processes */
891 	if (compare != NULL)
892 		qsort(pref, active_procs, sizeof(*pref), compare);
893 
894 	/* remember active and total counts */
895 	si->p_total = total_procs;
896 	si->p_active = pref_len = active_procs;
897 
898 	/* pass back a handle */
899 	handle.next_proc = pref;
900 	handle.remaining = active_procs;
901 	return ((caddr_t)&handle);
902 }
903 
904 static char fmt[512];	/* static area where result is built */
905 
906 char *
907 format_next_process(caddr_t handle, char *(*get_userid)(int), int flags)
908 {
909 	struct kinfo_proc *pp;
910 	const struct kinfo_proc *oldp;
911 	long cputime;
912 	double pct;
913 	struct handle *hp;
914 	char status[16];
915 	int cpu, state;
916 	struct rusage ru, *rup;
917 	long p_tot, s_tot;
918 	char *proc_fmt, thr_buf[6];
919 	char jid_buf[TOP_JID_LEN + 1], swap_buf[TOP_SWAP_LEN + 1];
920 	char *cmdbuf = NULL;
921 	char **args;
922 	const int cmdlen = 128;
923 
924 	/* find and remember the next proc structure */
925 	hp = (struct handle *)handle;
926 	pp = *(hp->next_proc++);
927 	hp->remaining--;
928 
929 	/* get the process's command name */
930 	if ((pp->ki_flag & P_INMEM) == 0) {
931 		/*
932 		 * Print swapped processes as <pname>
933 		 */
934 		size_t len;
935 
936 		len = strlen(pp->ki_comm);
937 		if (len > sizeof(pp->ki_comm) - 3)
938 			len = sizeof(pp->ki_comm) - 3;
939 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
940 		pp->ki_comm[0] = '<';
941 		pp->ki_comm[len + 1] = '>';
942 		pp->ki_comm[len + 2] = '\0';
943 	}
944 
945 	/*
946 	 * Convert the process's runtime from microseconds to seconds.  This
947 	 * time includes the interrupt time although that is not wanted here.
948 	 * ps(1) is similarly sloppy.
949 	 */
950 	cputime = (pp->ki_runtime + 500000) / 1000000;
951 
952 	/* calculate the base for cpu percentages */
953 	pct = PCTCPU(pp);
954 
955 	/* generate "STATE" field */
956 	switch (state = pp->ki_stat) {
957 	case SRUN:
958 		if (smpmode && pp->ki_oncpu != NOCPU)
959 			sprintf(status, "CPU%d", pp->ki_oncpu);
960 		else
961 			strcpy(status, "RUN");
962 		break;
963 	case SLOCK:
964 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
965 			sprintf(status, "*%.6s", pp->ki_lockname);
966 			break;
967 		}
968 		/* fall through */
969 	case SSLEEP:
970 		if (pp->ki_wmesg != NULL) {
971 			sprintf(status, "%.6s", pp->ki_wmesg);
972 			break;
973 		}
974 		/* FALLTHROUGH */
975 	default:
976 
977 		if (state >= 0 &&
978 		    state < sizeof(state_abbrev) / sizeof(*state_abbrev))
979 			sprintf(status, "%.6s", state_abbrev[state]);
980 		else
981 			sprintf(status, "?%5d", state);
982 		break;
983 	}
984 
985 	cmdbuf = (char *)malloc(cmdlen + 1);
986 	if (cmdbuf == NULL) {
987 		warn("malloc(%d)", cmdlen + 1);
988 		return NULL;
989 	}
990 
991 	if (!(flags & FMT_SHOWARGS)) {
992 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
993 		    pp->ki_tdname[0]) {
994 			snprintf(cmdbuf, cmdlen, "%s{%s}", pp->ki_comm,
995 			    pp->ki_tdname);
996 		} else {
997 			snprintf(cmdbuf, cmdlen, "%s", pp->ki_comm);
998 		}
999 	} else {
1000 		if (pp->ki_flag & P_SYSTEM ||
1001 		    pp->ki_args == NULL ||
1002 		    (args = kvm_getargv(kd, pp, cmdlen)) == NULL ||
1003 		    !(*args)) {
1004 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1005 		    	    pp->ki_tdname[0]) {
1006 				snprintf(cmdbuf, cmdlen,
1007 				    "[%s{%s}]", pp->ki_comm, pp->ki_tdname);
1008 			} else {
1009 				snprintf(cmdbuf, cmdlen,
1010 				    "[%s]", pp->ki_comm);
1011 			}
1012 		} else {
1013 			char *src, *dst, *argbuf;
1014 			char *cmd;
1015 			size_t argbuflen;
1016 			size_t len;
1017 
1018 			argbuflen = cmdlen * 4;
1019 			argbuf = (char *)malloc(argbuflen + 1);
1020 			if (argbuf == NULL) {
1021 				warn("malloc(%zu)", argbuflen + 1);
1022 				free(cmdbuf);
1023 				return NULL;
1024 			}
1025 
1026 			dst = argbuf;
1027 
1028 			/* Extract cmd name from argv */
1029 			cmd = strrchr(*args, '/');
1030 			if (cmd == NULL)
1031 				cmd = *args;
1032 			else
1033 				cmd++;
1034 
1035 			for (; (src = *args++) != NULL; ) {
1036 				if (*src == '\0')
1037 					continue;
1038 				len = (argbuflen - (dst - argbuf) - 1) / 4;
1039 				strvisx(dst, src,
1040 				    MIN(strlen(src), len),
1041 				    VIS_NL | VIS_CSTYLE);
1042 				while (*dst != '\0')
1043 					dst++;
1044 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
1045 					*dst++ = ' '; /* add delimiting space */
1046 			}
1047 			if (dst != argbuf && dst[-1] == ' ')
1048 				dst--;
1049 			*dst = '\0';
1050 
1051 			if (strcmp(cmd, pp->ki_comm) != 0) {
1052 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1053 				    pp->ki_tdname[0])
1054 					snprintf(cmdbuf, cmdlen,
1055 					    "%s (%s){%s}", argbuf, pp->ki_comm,
1056 					    pp->ki_tdname);
1057 				else
1058 					snprintf(cmdbuf, cmdlen,
1059 					    "%s (%s)", argbuf, pp->ki_comm);
1060 			} else {
1061 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1062 				    pp->ki_tdname[0])
1063 					snprintf(cmdbuf, cmdlen,
1064 					    "%s{%s}", argbuf, pp->ki_tdname);
1065 				else
1066 					strlcpy(cmdbuf, argbuf, cmdlen);
1067 			}
1068 			free(argbuf);
1069 		}
1070 	}
1071 
1072 	if (ps.jail == 0)
1073 		jid_buf[0] = '\0';
1074 	else
1075 		snprintf(jid_buf, sizeof(jid_buf), "%*d",
1076 		    jidlength - 1, pp->ki_jid);
1077 
1078 	if (ps.swap == 0)
1079 		swap_buf[0] = '\0';
1080 	else
1081 		snprintf(swap_buf, sizeof(swap_buf), "%*s",
1082 		    swaplength - 1,
1083 		    format_k2(pagetok(ki_swap(pp)))); /* XXX */
1084 
1085 	if (displaymode == DISP_IO) {
1086 		oldp = get_old_proc(pp);
1087 		if (oldp != NULL) {
1088 			ru.ru_inblock = RU(pp)->ru_inblock -
1089 			    RU(oldp)->ru_inblock;
1090 			ru.ru_oublock = RU(pp)->ru_oublock -
1091 			    RU(oldp)->ru_oublock;
1092 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1093 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1094 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1095 			rup = &ru;
1096 		} else {
1097 			rup = RU(pp);
1098 		}
1099 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1100 		s_tot = total_inblock + total_oublock + total_majflt;
1101 
1102 		snprintf(fmt, sizeof(fmt), io_Proc_format,
1103 		    pp->ki_pid,
1104 		    jidlength, jid_buf,
1105 		    namelength, namelength, (*get_userid)(pp->ki_ruid),
1106 		    rup->ru_nvcsw,
1107 		    rup->ru_nivcsw,
1108 		    rup->ru_inblock,
1109 		    rup->ru_oublock,
1110 		    rup->ru_majflt,
1111 		    p_tot,
1112 		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
1113 		    screen_width > cmdlengthdelta ?
1114 		    screen_width - cmdlengthdelta : 0,
1115 		    printable(cmdbuf));
1116 
1117 		free(cmdbuf);
1118 
1119 		return (fmt);
1120 	}
1121 
1122 	/* format this entry */
1123 	if (smpmode) {
1124 		if (state == SRUN && pp->ki_oncpu != NOCPU)
1125 			cpu = pp->ki_oncpu;
1126 		else
1127 			cpu = pp->ki_lastcpu;
1128 	} else
1129 		cpu = 0;
1130 	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
1131 	if (ps.thread != 0)
1132 		thr_buf[0] = '\0';
1133 	else
1134 		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
1135 		    (int)(sizeof(thr_buf) - 2), pp->ki_numthreads);
1136 
1137 	snprintf(fmt, sizeof(fmt), proc_fmt,
1138 	    pp->ki_pid,
1139 	    jidlength, jid_buf,
1140 	    namelength, namelength, (*get_userid)(pp->ki_ruid),
1141 	    thr_buf,
1142 	    pp->ki_pri.pri_level - PZERO,
1143 	    format_nice(pp),
1144 	    format_k2(PROCSIZE(pp)),
1145 	    format_k2(pagetok(pp->ki_rssize)),
1146 	    swaplength, swaplength, swap_buf,
1147 	    status,
1148 	    cpu,
1149 	    format_time(cputime),
1150 	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
1151 	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
1152 	    printable(cmdbuf));
1153 
1154 	free(cmdbuf);
1155 
1156 	/* return the result */
1157 	return (fmt);
1158 }
1159 
1160 static void
1161 getsysctl(const char *name, void *ptr, size_t len)
1162 {
1163 	size_t nlen = len;
1164 
1165 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1166 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1167 		    strerror(errno));
1168 		quit(23);
1169 	}
1170 	if (nlen != len) {
1171 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1172 		    name, (unsigned long)len, (unsigned long)nlen);
1173 		quit(23);
1174 	}
1175 }
1176 
1177 static const char *
1178 format_nice(const struct kinfo_proc *pp)
1179 {
1180 	const char *fifo, *kproc;
1181 	int rtpri;
1182 	static char nicebuf[4 + 1];
1183 
1184 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1185 	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1186 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1187 	case PRI_ITHD:
1188 		return ("-");
1189 	case PRI_REALTIME:
1190 		/*
1191 		 * XXX: the kernel doesn't tell us the original rtprio and
1192 		 * doesn't really know what it was, so to recover it we
1193 		 * must be more chummy with the implementation than the
1194 		 * implementation is with itself.  pri_user gives a
1195 		 * constant "base" priority, but is only initialized
1196 		 * properly for user threads.  pri_native gives what the
1197 		 * kernel calls the "base" priority, but it isn't constant
1198 		 * since it is changed by priority propagation.  pri_native
1199 		 * also isn't properly initialized for all threads, but it
1200 		 * is properly initialized for kernel realtime and idletime
1201 		 * threads.  Thus we use pri_user for the base priority of
1202 		 * user threads (it is always correct) and pri_native for
1203 		 * the base priority of kernel realtime and idletime threads
1204 		 * (there is nothing better, and it is usually correct).
1205 		 *
1206 		 * The field width and thus the buffer are too small for
1207 		 * values like "kr31F", but such values shouldn't occur,
1208 		 * and if they do then the tailing "F" is not displayed.
1209 		 */
1210 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1211 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1212 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1213 		    kproc, rtpri, fifo);
1214 		break;
1215 	case PRI_TIMESHARE:
1216 		if (pp->ki_flag & P_KPROC)
1217 			return ("-");
1218 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1219 		break;
1220 	case PRI_IDLE:
1221 		/* XXX: as above. */
1222 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1223 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1224 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1225 		    kproc, rtpri, fifo);
1226 		break;
1227 	default:
1228 		return ("?");
1229 	}
1230 	return (nicebuf);
1231 }
1232 
1233 /* comparison routines for qsort */
1234 
1235 static int
1236 compare_pid(const void *p1, const void *p2)
1237 {
1238 	const struct kinfo_proc * const *pp1 = p1;
1239 	const struct kinfo_proc * const *pp2 = p2;
1240 
1241 	if ((*pp2)->ki_pid < 0 || (*pp1)->ki_pid < 0)
1242 		abort();
1243 
1244 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1245 }
1246 
1247 static int
1248 compare_tid(const void *p1, const void *p2)
1249 {
1250 	const struct kinfo_proc * const *pp1 = p1;
1251 	const struct kinfo_proc * const *pp2 = p2;
1252 
1253 	if ((*pp2)->ki_tid < 0 || (*pp1)->ki_tid < 0)
1254 		abort();
1255 
1256 	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1257 }
1258 
1259 /*
1260  *  proc_compare - comparison function for "qsort"
1261  *	Compares the resource consumption of two processes using five
1262  *	distinct keys.  The keys (in descending order of importance) are:
1263  *	percent cpu, cpu ticks, state, resident set size, total virtual
1264  *	memory usage.  The process states are ordered as follows (from least
1265  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1266  *	array declaration below maps a process state index into a number
1267  *	that reflects this ordering.
1268  */
1269 
1270 static int sorted_state[] = {
1271 	0,	/* not used		*/
1272 	3,	/* sleep		*/
1273 	1,	/* ABANDONED (WAIT)	*/
1274 	6,	/* run			*/
1275 	5,	/* start		*/
1276 	2,	/* zombie		*/
1277 	4	/* stop			*/
1278 };
1279 
1280 
1281 #define ORDERKEY_PCTCPU(a, b) do { \
1282 	double diff; \
1283 	if (ps.wcpu) \
1284 		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1285 		    weighted_cpu(PCTCPU((a)), (a)); \
1286 	else \
1287 		diff = PCTCPU((b)) - PCTCPU((a)); \
1288 	if (diff != 0) \
1289 		return (diff > 0 ? 1 : -1); \
1290 } while (0)
1291 
1292 #define ORDERKEY_CPTICKS(a, b) do { \
1293 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1294 	if (diff != 0) \
1295 		return (diff > 0 ? 1 : -1); \
1296 } while (0)
1297 
1298 #define ORDERKEY_STATE(a, b) do { \
1299 	int diff = sorted_state[(b)->ki_stat] - sorted_state[(a)->ki_stat]; \
1300 	if (diff != 0) \
1301 		return (diff > 0 ? 1 : -1); \
1302 } while (0)
1303 
1304 #define ORDERKEY_PRIO(a, b) do { \
1305 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1306 	if (diff != 0) \
1307 		return (diff > 0 ? 1 : -1); \
1308 } while (0)
1309 
1310 #define	ORDERKEY_THREADS(a, b) do { \
1311 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1312 	if (diff != 0) \
1313 		return (diff > 0 ? 1 : -1); \
1314 } while (0)
1315 
1316 #define ORDERKEY_RSSIZE(a, b) do { \
1317 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1318 	if (diff != 0) \
1319 		return (diff > 0 ? 1 : -1); \
1320 } while (0)
1321 
1322 #define ORDERKEY_MEM(a, b) do { \
1323 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1324 	if (diff != 0) \
1325 		return (diff > 0 ? 1 : -1); \
1326 } while (0)
1327 
1328 #define ORDERKEY_JID(a, b) do { \
1329 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1330 	if (diff != 0) \
1331 		return (diff > 0 ? 1 : -1); \
1332 } while (0)
1333 
1334 #define ORDERKEY_SWAP(a, b) do { \
1335 	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1336 	if (diff != 0) \
1337 		return (diff > 0 ? 1 : -1); \
1338 } while (0)
1339 
1340 /* compare_cpu - the comparison function for sorting by cpu percentage */
1341 
1342 int
1343 #ifdef ORDER
1344 compare_cpu(void *arg1, void *arg2)
1345 #else
1346 proc_compare(void *arg1, void *arg2)
1347 #endif
1348 {
1349 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1350 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1351 
1352 	ORDERKEY_PCTCPU(p1, p2);
1353 	ORDERKEY_CPTICKS(p1, p2);
1354 	ORDERKEY_STATE(p1, p2);
1355 	ORDERKEY_PRIO(p1, p2);
1356 	ORDERKEY_RSSIZE(p1, p2);
1357 	ORDERKEY_MEM(p1, p2);
1358 
1359 	return (0);
1360 }
1361 
1362 #ifdef ORDER
1363 /* "cpu" compare routines */
1364 int compare_size(), compare_res(), compare_time(), compare_prio(),
1365     compare_threads();
1366 
1367 /*
1368  * "io" compare routines.  Context switches aren't i/o, but are displayed
1369  * on the "io" display.
1370  */
1371 int compare_iototal(), compare_ioread(), compare_iowrite(), compare_iofault(),
1372     compare_vcsw(), compare_ivcsw();
1373 
1374 int (*compares[])() = {
1375 	compare_cpu,
1376 	compare_size,
1377 	compare_res,
1378 	compare_time,
1379 	compare_prio,
1380 	compare_threads,
1381 	compare_iototal,
1382 	compare_ioread,
1383 	compare_iowrite,
1384 	compare_iofault,
1385 	compare_vcsw,
1386 	compare_ivcsw,
1387 	compare_jid,
1388 	compare_swap,
1389 	NULL
1390 };
1391 
1392 /* compare_size - the comparison function for sorting by total memory usage */
1393 
1394 int
1395 compare_size(void *arg1, void *arg2)
1396 {
1397 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1398 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1399 
1400 	ORDERKEY_MEM(p1, p2);
1401 	ORDERKEY_RSSIZE(p1, p2);
1402 	ORDERKEY_PCTCPU(p1, p2);
1403 	ORDERKEY_CPTICKS(p1, p2);
1404 	ORDERKEY_STATE(p1, p2);
1405 	ORDERKEY_PRIO(p1, p2);
1406 
1407 	return (0);
1408 }
1409 
1410 /* compare_res - the comparison function for sorting by resident set size */
1411 
1412 int
1413 compare_res(void *arg1, void *arg2)
1414 {
1415 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1416 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1417 
1418 	ORDERKEY_RSSIZE(p1, p2);
1419 	ORDERKEY_MEM(p1, p2);
1420 	ORDERKEY_PCTCPU(p1, p2);
1421 	ORDERKEY_CPTICKS(p1, p2);
1422 	ORDERKEY_STATE(p1, p2);
1423 	ORDERKEY_PRIO(p1, p2);
1424 
1425 	return (0);
1426 }
1427 
1428 /* compare_time - the comparison function for sorting by total cpu time */
1429 
1430 int
1431 compare_time(void *arg1, void *arg2)
1432 {
1433 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1434 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1435 
1436 	ORDERKEY_CPTICKS(p1, p2);
1437 	ORDERKEY_PCTCPU(p1, p2);
1438 	ORDERKEY_STATE(p1, p2);
1439 	ORDERKEY_PRIO(p1, p2);
1440 	ORDERKEY_RSSIZE(p1, p2);
1441 	ORDERKEY_MEM(p1, p2);
1442 
1443 	return (0);
1444 }
1445 
1446 /* compare_prio - the comparison function for sorting by priority */
1447 
1448 int
1449 compare_prio(void *arg1, void *arg2)
1450 {
1451 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1452 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1453 
1454 	ORDERKEY_PRIO(p1, p2);
1455 	ORDERKEY_CPTICKS(p1, p2);
1456 	ORDERKEY_PCTCPU(p1, p2);
1457 	ORDERKEY_STATE(p1, p2);
1458 	ORDERKEY_RSSIZE(p1, p2);
1459 	ORDERKEY_MEM(p1, p2);
1460 
1461 	return (0);
1462 }
1463 
1464 /* compare_threads - the comparison function for sorting by threads */
1465 int
1466 compare_threads(void *arg1, void *arg2)
1467 {
1468 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1469 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1470 
1471 	ORDERKEY_THREADS(p1, p2);
1472 	ORDERKEY_PCTCPU(p1, p2);
1473 	ORDERKEY_CPTICKS(p1, p2);
1474 	ORDERKEY_STATE(p1, p2);
1475 	ORDERKEY_PRIO(p1, p2);
1476 	ORDERKEY_RSSIZE(p1, p2);
1477 	ORDERKEY_MEM(p1, p2);
1478 
1479 	return (0);
1480 }
1481 
1482 /* compare_jid - the comparison function for sorting by jid */
1483 static int
1484 compare_jid(const void *arg1, const void *arg2)
1485 {
1486 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1487 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1488 
1489 	ORDERKEY_JID(p1, p2);
1490 	ORDERKEY_PCTCPU(p1, p2);
1491 	ORDERKEY_CPTICKS(p1, p2);
1492 	ORDERKEY_STATE(p1, p2);
1493 	ORDERKEY_PRIO(p1, p2);
1494 	ORDERKEY_RSSIZE(p1, p2);
1495 	ORDERKEY_MEM(p1, p2);
1496 
1497 	return (0);
1498 }
1499 
1500 /* compare_swap - the comparison function for sorting by swap */
1501 static int
1502 compare_swap(const void *arg1, const void *arg2)
1503 {
1504 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1505 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1506 
1507 	ORDERKEY_SWAP(p1, p2);
1508 	ORDERKEY_PCTCPU(p1, p2);
1509 	ORDERKEY_CPTICKS(p1, p2);
1510 	ORDERKEY_STATE(p1, p2);
1511 	ORDERKEY_PRIO(p1, p2);
1512 	ORDERKEY_RSSIZE(p1, p2);
1513 	ORDERKEY_MEM(p1, p2);
1514 
1515 	return (0);
1516 }
1517 #endif /* ORDER */
1518 
1519 /* assorted comparison functions for sorting by i/o */
1520 
1521 int
1522 #ifdef ORDER
1523 compare_iototal(void *arg1, void *arg2)
1524 #else
1525 io_compare(void *arg1, void *arg2)
1526 #endif
1527 {
1528 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1529 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1530 
1531 	return (get_io_total(p2) - get_io_total(p1));
1532 }
1533 
1534 #ifdef ORDER
1535 int
1536 compare_ioread(void *arg1, void *arg2)
1537 {
1538 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1539 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1540 	long dummy, inp1, inp2;
1541 
1542 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1543 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1544 
1545 	return (inp2 - inp1);
1546 }
1547 
1548 int
1549 compare_iowrite(void *arg1, void *arg2)
1550 {
1551 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1552 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1553 	long dummy, oup1, oup2;
1554 
1555 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1556 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1557 
1558 	return (oup2 - oup1);
1559 }
1560 
1561 int
1562 compare_iofault(void *arg1, void *arg2)
1563 {
1564 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1565 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1566 	long dummy, flp1, flp2;
1567 
1568 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1569 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1570 
1571 	return (flp2 - flp1);
1572 }
1573 
1574 int
1575 compare_vcsw(void *arg1, void *arg2)
1576 {
1577 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1578 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1579 	long dummy, flp1, flp2;
1580 
1581 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1582 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1583 
1584 	return (flp2 - flp1);
1585 }
1586 
1587 int
1588 compare_ivcsw(void *arg1, void *arg2)
1589 {
1590 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1591 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1592 	long dummy, flp1, flp2;
1593 
1594 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1595 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1596 
1597 	return (flp2 - flp1);
1598 }
1599 #endif /* ORDER */
1600 
1601 /*
1602  * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1603  *		the process does not exist.
1604  *		It is EXTREMELY IMPORTANT that this function work correctly.
1605  *		If top runs setuid root (as in SVR4), then this function
1606  *		is the only thing that stands in the way of a serious
1607  *		security problem.  It validates requests for the "kill"
1608  *		and "renice" commands.
1609  */
1610 
1611 int
1612 proc_owner(int pid)
1613 {
1614 	int cnt;
1615 	struct kinfo_proc **prefp;
1616 	struct kinfo_proc *pp;
1617 
1618 	prefp = pref;
1619 	cnt = pref_len;
1620 	while (--cnt >= 0) {
1621 		pp = *prefp++;
1622 		if (pp->ki_pid == (pid_t)pid)
1623 			return ((int)pp->ki_ruid);
1624 	}
1625 	return (-1);
1626 }
1627 
1628 static int
1629 swapmode(int *retavail, int *retfree)
1630 {
1631 	int n;
1632 	int pagesize = getpagesize();
1633 	struct kvm_swap swapary[1];
1634 
1635 	*retavail = 0;
1636 	*retfree = 0;
1637 
1638 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1639 
1640 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1641 	if (n < 0 || swapary[0].ksw_total == 0)
1642 		return (0);
1643 
1644 	*retavail = CONVERT(swapary[0].ksw_total);
1645 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1646 
1647 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1648 	return (n);
1649 }
1650