2 * top - a top users display for Unix
4 * SYNOPSIS: For DragonFly 2.x and later
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/)
12 * This is the machine-dependent module for DragonFly 2.5.1
14 * DragonFly 2.x and above
18 * AUTHOR: Jan Lentfer <Jan.Lentfer@web.de>
19 * This module has been put together from different sources and is based on the
20 * work of many other people, e.g. Matthew Dillon, Simon Schubert, Jordan Gordeev.
22 * $FreeBSD: src/usr.bin/top/machine.c,v 1.29.2.2 2001/07/31 20:27:05 tmm Exp $
23 * $DragonFly: src/usr.bin/top/machine.c,v 1.26 2008/10/16 01:52:33 swildner Exp $
27 #include <sys/types.h>
28 #include <sys/signal.h>
29 #include <sys/param.h>
38 #include <sys/errno.h>
39 #include <sys/sysctl.h>
43 #include <sys/vmmeter.h>
44 #include <sys/resource.h>
45 #include <sys/rtprio.h>
52 #include <osreldate.h> /* for changes in kernel structures */
54 #include <sys/kinfo.h>
62 int swapmode(int *retavail, int *retfree);
64 static int namelength;
66 static int show_fullcmd;
71 * needs to be a global symbol, so wrapper can be modified accordingly.
73 static int show_threads = 0;
75 /* get_process_info passes back a handle. This is what it looks like: */
78 struct kinfo_proc **next_proc; /* points to next valid proc pointer */
79 int remaining; /* number of pointers remaining */
82 /* declarations for load_avg */
85 #define PP(pp, field) ((pp)->kp_ ## field)
86 #define LP(pp, field) ((pp)->kp_lwp.kl_ ## field)
87 #define VP(pp, field) ((pp)->kp_vm_ ## field)
89 /* define what weighted cpu is. */
90 #define weighted_cpu(pct, pp) (PP((pp), swtime) == 0 ? 0.0 : \
91 ((pct) / (1.0 - exp(PP((pp), swtime) * logcpu))))
93 /* what we consider to be process size: */
94 #define PROCSIZE(pp) (VP((pp), map_size) / 1024)
97 * These definitions control the format of the per-process area
100 static char smp_header[] =
101 " PID %-*.*s PRI NICE SIZE RES STATE C TIME WCPU CPU COMMAND";
103 #define smp_Proc_format \
104 "%5d %-*.*s %3d %3d%7s %6s %-6.6s %1x%7s %5.2f%% %5.2f%% %.*s"
106 static char up_header[] =
107 " PID %-*.*s PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND";
109 #define up_Proc_format \
110 "%5d %-*.*s %3d %3d%7s %6s %-6.6s%.0d%7s %5.2f%% %5.2f%% %.*s"
114 /* process state names for the "STATE" column of the display */
116 * the extra nulls in the string "run" are for adding a slash and the
117 * processor number when needed
120 const char *state_abbrev[] = {
121 "", "RUN\0\0\0", "STOP", "SLEEP",
127 /* values that we stash away in _init and use in later routines */
129 static double logcpu;
134 /* these are for calculating cpu state percentages */
136 static struct kinfo_cputime *cp_time, *cp_old;
138 /* these are for detailing the process states */
140 int process_states[6];
141 char *procstatenames[] = {
142 "", " starting, ", " running, ", " sleeping, ", " stopped, ",
147 /* these are for detailing the cpu states */
150 char *cpustatenames[CPU_STATES + 1] = {
151 "user", "nice", "system", "interrupt", "idle", NULL
154 /* these are for detailing the memory statistics */
156 long memory_stats[7];
157 char *memorynames[] = {
158 "K Active, ", "K Inact, ", "K Wired, ", "K Cache, ", "K Buf, ", "K Free",
163 char *swapnames[] = {
165 "K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
170 /* these are for keeping track of the proc array */
173 static int onproc = -1;
175 static struct kinfo_proc *pbase;
176 static struct kinfo_proc **pref;
178 /* these are for getting the memory statistics */
180 static int pageshift; /* log base 2 of the pagesize */
182 /* define pagetok in terms of pageshift */
184 #define pagetok(size) ((size) << pageshift)
186 /* sorting orders. first is default */
187 char *ordernames[] = {
188 "cpu", "size", "res", "time", "pri", "thr", "pid", NULL
191 /* compare routines */
192 int proc_compare (struct kinfo_proc **, struct kinfo_proc **);
193 int compare_size (struct kinfo_proc **, struct kinfo_proc **);
194 int compare_res (struct kinfo_proc **, struct kinfo_proc **);
195 int compare_time (struct kinfo_proc **, struct kinfo_proc **);
196 int compare_prio(struct kinfo_proc **, struct kinfo_proc **);
197 int compare_thr (struct kinfo_proc **, struct kinfo_proc **);
198 int compare_pid (struct kinfo_proc **, struct kinfo_proc **);
200 int (*proc_compares[]) (struct kinfo_proc **,struct kinfo_proc **) = {
212 cputime_percentages(int out[CPU_STATES], struct kinfo_cputime *new,
213 struct kinfo_cputime *old)
215 struct kinfo_cputime diffs;
216 uint64_t total_change, half_total;
221 diffs.cp_user = new->cp_user - old->cp_user;
222 diffs.cp_nice = new->cp_nice - old->cp_nice;
223 diffs.cp_sys = new->cp_sys - old->cp_sys;
224 diffs.cp_intr = new->cp_intr - old->cp_intr;
225 diffs.cp_idle = new->cp_idle - old->cp_idle;
226 total_change = diffs.cp_user + diffs.cp_nice + diffs.cp_sys +
227 diffs.cp_intr + diffs.cp_idle;
228 old->cp_user = new->cp_user;
229 old->cp_nice = new->cp_nice;
230 old->cp_sys = new->cp_sys;
231 old->cp_intr = new->cp_intr;
232 old->cp_idle = new->cp_idle;
234 /* avoid divide by zero potential */
235 if (total_change == 0)
238 /* calculate percentages based on overall change, rounding up */
239 half_total = total_change >> 1;
241 out[0] = ((diffs.cp_user * 1000LL + half_total) / total_change);
242 out[1] = ((diffs.cp_nice * 1000LL + half_total) / total_change);
243 out[2] = ((diffs.cp_sys * 1000LL + half_total) / total_change);
244 out[3] = ((diffs.cp_intr * 1000LL + half_total) / total_change);
245 out[4] = ((diffs.cp_idle * 1000LL + half_total) / total_change);
249 machine_init(struct statics *statics)
254 struct timeval boottime;
257 if (kinfo_get_cpus(&n_cpus))
258 err(1, "kinfo_get_cpus failed");
261 modelen = sizeof(boottime);
262 if (sysctlbyname("kern.boottime", &boottime, &modelen, NULL, 0) == -1) {
263 /* we have no boottime to report */
264 boottime.tv_sec = -1;
266 modelen = sizeof(smpmode);
267 if ((sysctlbyname("machdep.smp_active", &smpmode, &modelen, NULL, 0) < 0 &&
268 sysctlbyname("smp.smp_active", &smpmode, &modelen, NULL, 0) < 0) ||
269 modelen != sizeof(smpmode))
272 while ((pw = getpwent()) != NULL) {
273 if ((int)strlen(pw->pw_name) > namelength)
274 namelength = strlen(pw->pw_name);
278 if (smpmode && namelength > 13)
280 else if (namelength > 15)
283 if ((kd = kvm_open(NULL, NULL, NULL, O_RDONLY, NULL)) == NULL)
286 if (kinfo_get_sched_ccpu(&ccpu)) {
287 fprintf(stderr, "top: kinfo_get_sched_ccpu failed\n");
290 /* this is used in calculating WCPU -- calculate it ahead of time */
291 logcpu = log(loaddouble(ccpu));
298 * get the page size with "getpagesize" and calculate pageshift from
301 pagesize = getpagesize();
303 while (pagesize > 1) {
308 /* we only need the amount of log(2)1024 for our conversion */
309 pageshift -= LOG1024;
311 /* fill in the statics information */
312 statics->procstate_names = procstatenames;
313 statics->cpustate_names = cpustatenames;
314 statics->memory_names = memorynames;
315 statics->boottime = boottime.tv_sec;
316 statics->swap_names = swapnames;
317 statics->order_names = ordernames;
318 /* we need kvm descriptor in order to show full commands */
319 statics->flags.fullcmds = kd != NULL;
326 format_header(char *uname_field)
328 static char Header[128];
330 snprintf(Header, sizeof(Header), smpmode ? smp_header : up_header,
331 namelength, namelength, uname_field);
333 if (screen_width <= 79)
336 cmdlength = screen_width;
338 cmdlength = cmdlength - strlen(Header) + 6;
343 static int swappgsin = -1;
344 static int swappgsout = -1;
345 extern struct timeval timeout;
348 get_system_info(struct system_info *si)
353 if (cpu_states == NULL) {
354 cpu_states = malloc(sizeof(*cpu_states) * CPU_STATES * n_cpus);
355 if (cpu_states == NULL)
357 bzero(cpu_states, sizeof(*cpu_states) * CPU_STATES * n_cpus);
359 if (cp_time == NULL) {
360 cp_time = malloc(2 * n_cpus * sizeof(cp_time[0]));
363 cp_old = cp_time + n_cpus;
365 len = n_cpus * sizeof(cp_old[0]);
367 if (sysctlbyname("kern.cputime", cp_old, &len, NULL, 0))
368 err(1, "kern.cputime");
370 len = n_cpus * sizeof(cp_time[0]);
372 if (sysctlbyname("kern.cputime", cp_time, &len, NULL, 0))
373 err(1, "kern.cputime");
375 getloadavg(si->load_avg, 3);
379 /* convert cp_time counts to percentages */
380 for (cpu = 0; cpu < n_cpus; ++cpu) {
381 cputime_percentages(cpu_states + cpu * CPU_STATES,
382 &cp_time[cpu], &cp_old[cpu]);
385 /* sum memory & swap statistics */
389 size_t vms_size = sizeof(vms);
390 size_t vmm_size = sizeof(vmm);
391 static unsigned int swap_delay = 0;
392 static int swapavail = 0;
393 static int swapfree = 0;
394 static int bufspace = 0;
396 if (sysctlbyname("vm.vmstats", &vms, &vms_size, NULL, 0))
397 err(1, "sysctlbyname: vm.vmstats");
399 if (sysctlbyname("vm.vmmeter", &vmm, &vmm_size, NULL, 0))
400 err(1, "sysctlbyname: vm.vmmeter");
402 if (kinfo_get_vfs_bufspace(&bufspace))
403 err(1, "kinfo_get_vfs_bufspace");
405 /* convert memory stats to Kbytes */
406 memory_stats[0] = pagetok(vms.v_active_count);
407 memory_stats[1] = pagetok(vms.v_inactive_count);
408 memory_stats[2] = pagetok(vms.v_wire_count);
409 memory_stats[3] = pagetok(vms.v_cache_count);
410 memory_stats[4] = bufspace / 1024;
411 memory_stats[5] = pagetok(vms.v_free_count);
412 memory_stats[6] = -1;
419 /* compute differences between old and new swap statistic */
421 swap_stats[4] = pagetok(((vmm.v_swappgsin - swappgsin)));
422 swap_stats[5] = pagetok(((vmm.v_swappgsout - swappgsout)));
425 swappgsin = vmm.v_swappgsin;
426 swappgsout = vmm.v_swappgsout;
428 /* call CPU heavy swapmode() only for changes */
429 if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
430 swap_stats[3] = swapmode(&swapavail, &swapfree);
431 swap_stats[0] = swapavail;
432 swap_stats[1] = swapavail - swapfree;
433 swap_stats[2] = swapfree;
439 /* set arrays and strings */
440 si->cpustates = cpu_states;
441 si->memory = memory_stats;
442 si->swap = swap_stats;
446 si->last_pid = lastpid;
453 static struct handle handle;
456 get_process_info(struct system_info *si, struct process_select *sel,
462 struct kinfo_proc **prefp;
463 struct kinfo_proc *pp;
465 /* these are copied out of sel for speed */
471 pbase = kvm_getprocs(kd, KERN_PROC_ALL, 0, &nproc);
473 pref = (struct kinfo_proc **)realloc(pref, sizeof(struct kinfo_proc *)
475 if (pref == NULL || pbase == NULL) {
476 (void)fprintf(stderr, "top: Out of memory.\n");
479 /* get a pointer to the states summary array */
480 si->procstates = process_states;
482 /* set up flags which define what we are going to select */
483 show_idle = sel->idle;
484 show_system = sel->system;
485 show_uid = sel->uid != -1;
486 show_fullcmd = sel->fullcmd;
488 /* count up process states and get pointers to interesting procs */
491 memset((char *)process_states, 0, sizeof(process_states));
493 for (pp = pbase, i = 0; i < nproc; pp++, i++) {
495 * Place pointers to each valid proc structure in pref[].
496 * Process slots that are actually in use have a non-zero
497 * status field. Processes with P_SYSTEM set are system
498 * processes---these get ignored unless show_sysprocs is set.
500 if ((show_threads && (LP(pp, pid) == -1)) ||
501 (show_system || ((PP(pp, flags) & P_SYSTEM) == 0))) {
503 process_states[(unsigned char)PP(pp, stat)]++;
504 if ((show_threads && (LP(pp, pid) == -1)) ||
505 (show_idle || (LP(pp, pctcpu) != 0) ||
506 (LP(pp, stat) == LSRUN)) &&
507 (!show_uid || PP(pp, ruid) == (uid_t) sel->uid)) {
514 qsort((char *)pref, active_procs, sizeof(struct kinfo_proc *),
515 (int (*)(const void *, const void *))proc_compares[compare_index]);
517 /* remember active and total counts */
518 si->p_total = total_procs;
519 si->p_active = pref_len = active_procs;
521 /* pass back a handle */
522 handle.next_proc = pref;
523 handle.remaining = active_procs;
524 return ((caddr_t) & handle);
527 char fmt[128]; /* static area where result is built */
530 format_next_process(caddr_t xhandle, char *(*get_userid) (int))
532 struct kinfo_proc *pp;
542 /* find and remember the next proc structure */
543 hp = (struct handle *)xhandle;
544 pp = *(hp->next_proc++);
547 /* get the process's command name */
549 if ((comm_full = kvm_getargv(kd, pp, 0)) == NULL) {
558 * Convert the process's runtime from microseconds to seconds. This
559 * time includes the interrupt time although that is not wanted here.
560 * ps(1) is similarly sloppy.
562 cputime = (LP(pp, uticks) + LP(pp, sticks)) / 1000000;
564 /* calculate the base for cpu percentages */
565 pct = pctdouble(LP(pp, pctcpu));
567 /* generate "STATE" field */
568 switch (state = LP(pp, stat)) {
570 if (smpmode && LP(pp, tdflags) & TDF_RUNNING)
571 sprintf(status, "CPU%d", LP(pp, cpuid));
573 strcpy(status, "RUN");
576 if (LP(pp, wmesg) != NULL) {
577 sprintf(status, "%.6s", LP(pp, wmesg));
584 (unsigned)state < sizeof(state_abbrev) / sizeof(*state_abbrev))
585 sprintf(status, "%.6s", state_abbrev[(unsigned char)state]);
587 sprintf(status, "?%5d", state);
591 if (PP(pp, stat) == SZOMB)
592 strcpy(status, "ZOMB");
595 * idle time 0 - 31 -> nice value +21 - +52 normal time -> nice
596 * value -20 - +20 real time 0 - 31 -> nice value -52 - -21 thread
597 * 0 - 31 -> nice value -53 -
599 switch (LP(pp, rtprio.type)) {
600 case RTP_PRIO_REALTIME:
601 xnice = PRIO_MIN - 1 - RTP_PRIO_MAX + LP(pp, rtprio.prio);
604 xnice = PRIO_MAX + 1 + LP(pp, rtprio.prio);
606 case RTP_PRIO_THREAD:
607 xnice = PRIO_MIN - 1 - RTP_PRIO_MAX - LP(pp, rtprio.prio);
610 xnice = PP(pp, nice);
614 /* format this entry */
615 snprintf(fmt, sizeof(fmt),
616 smpmode ? smp_Proc_format : up_Proc_format,
618 namelength, namelength,
619 get_userid(PP(pp, ruid)),
620 (int)((show_threads && (LP(pp, pid) == -1)) ?
621 LP(pp, tdprio) : LP(pp, prio)),
623 format_k(PROCSIZE(pp)),
624 format_k(pagetok(VP(pp, rssize))),
626 (int)(smpmode ? LP(pp, cpuid) : 0),
627 format_time(cputime),
628 100.0 * weighted_cpu(pct, pp),
631 show_fullcmd ? *comm_full : comm);
633 /* return the result */
637 /* comparison routines for qsort */
640 * proc_compare - comparison function for "qsort"
641 * Compares the resource consumption of two processes using five
642 * distinct keys. The keys (in descending order of importance) are:
643 * percent cpu, cpu ticks, state, resident set size, total virtual
644 * memory usage. The process states are ordered as follows (from least
645 * to most important): WAIT, zombie, sleep, stop, start, run. The
646 * array declaration below maps a process state index into a number
647 * that reflects this ordering.
650 static unsigned char sorted_state[] =
654 1, /* ABANDONED (WAIT) */
662 #define ORDERKEY_PCTCPU \
663 if (lresult = (long) LP(p2, pctcpu) - (long) LP(p1, pctcpu), \
664 (result = lresult > 0 ? 1 : lresult < 0 ? -1 : 0) == 0)
666 #define CPTICKS(p) (LP(p, uticks) + LP(p, sticks))
668 #define ORDERKEY_CPTICKS \
669 if ((result = CPTICKS(p2) > CPTICKS(p1) ? 1 : \
670 CPTICKS(p2) < CPTICKS(p1) ? -1 : 0) == 0)
672 #define ORDERKEY_STATE \
673 if ((result = sorted_state[(unsigned char) PP(p2, stat)] - \
674 sorted_state[(unsigned char) PP(p1, stat)]) == 0)
676 #define ORDERKEY_PRIO \
677 if ((result = LP(p2, prio) - LP(p1, prio)) == 0)
679 #define ORDERKEY_KTHREADS \
680 if ((result = (LP(p1, pid) == 0) - (LP(p2, pid) == 0)) == 0)
682 #define ORDERKEY_KTHREADS_PRIO \
683 if ((result = LP(p2, tdprio) - LP(p1, tdprio)) == 0)
685 #define ORDERKEY_RSSIZE \
686 if ((result = VP(p2, rssize) - VP(p1, rssize)) == 0)
688 #define ORDERKEY_MEM \
689 if ( (result = PROCSIZE(p2) - PROCSIZE(p1)) == 0 )
691 #define ORDERKEY_PID \
692 if ( (result = PP(p1, pid) - PP(p2, pid)) == 0)
694 /* compare_cpu - the comparison function for sorting by cpu percentage */
697 proc_compare(struct kinfo_proc **pp1, struct kinfo_proc **pp2)
699 struct kinfo_proc *p1;
700 struct kinfo_proc *p2;
704 /* remove one level of indirection */
705 p1 = *(struct kinfo_proc **) pp1;
706 p2 = *(struct kinfo_proc **) pp2;
719 /* compare_size - the comparison function for sorting by total memory usage */
722 compare_size(struct kinfo_proc **pp1, struct kinfo_proc **pp2)
724 struct kinfo_proc *p1;
725 struct kinfo_proc *p2;
729 /* remove one level of indirection */
730 p1 = *(struct kinfo_proc **) pp1;
731 p2 = *(struct kinfo_proc **) pp2;
744 /* compare_res - the comparison function for sorting by resident set size */
747 compare_res(struct kinfo_proc **pp1, struct kinfo_proc **pp2)
749 struct kinfo_proc *p1;
750 struct kinfo_proc *p2;
754 /* remove one level of indirection */
755 p1 = *(struct kinfo_proc **) pp1;
756 p2 = *(struct kinfo_proc **) pp2;
769 /* compare_time - the comparison function for sorting by total cpu time */
772 compare_time(struct kinfo_proc **pp1, struct kinfo_proc **pp2)
774 struct kinfo_proc *p1;
775 struct kinfo_proc *p2;
779 /* remove one level of indirection */
780 p1 = *(struct kinfo_proc **) pp1;
781 p2 = *(struct kinfo_proc **) pp2;
786 ORDERKEY_KTHREADS_PRIO
796 /* compare_prio - the comparison function for sorting by cpu percentage */
799 compare_prio(struct kinfo_proc **pp1, struct kinfo_proc **pp2)
801 struct kinfo_proc *p1;
802 struct kinfo_proc *p2;
806 /* remove one level of indirection */
807 p1 = *(struct kinfo_proc **) pp1;
808 p2 = *(struct kinfo_proc **) pp2;
811 ORDERKEY_KTHREADS_PRIO
824 compare_thr(struct kinfo_proc **pp1, struct kinfo_proc **pp2)
826 struct kinfo_proc *p1;
827 struct kinfo_proc *p2;
831 /* remove one level of indirection */
832 p1 = *(struct kinfo_proc **)pp1;
833 p2 = *(struct kinfo_proc **)pp2;
836 ORDERKEY_KTHREADS_PRIO
847 /* compare_pid - the comparison function for sorting by process id */
850 compare_pid(struct kinfo_proc **pp1, struct kinfo_proc **pp2)
852 struct kinfo_proc *p1;
853 struct kinfo_proc *p2;
856 /* remove one level of indirection */
857 p1 = *(struct kinfo_proc **) pp1;
858 p2 = *(struct kinfo_proc **) pp2;
867 * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
868 * the process does not exist.
869 * It is EXTREMLY IMPORTANT that this function work correctly.
870 * If top runs setuid root (as in SVR4), then this function
871 * is the only thing that stands in the way of a serious
872 * security problem. It validates requests for the "kill"
873 * and "renice" commands.
880 struct kinfo_proc **prefp;
881 struct kinfo_proc *pp;
885 while (--xcnt >= 0) {
887 if (PP(pp, pid) == (pid_t) pid) {
888 return ((int)PP(pp, ruid));
896 * swapmode is based on a program called swapinfo written
897 * by Kevin Lahey <kml@rokkaku.atl.ga.us>.
900 swapmode(int *retavail, int *retfree)
903 int pagesize = getpagesize();
904 struct kvm_swap swapary[1];
909 #define CONVERT(v) ((quad_t)(v) * pagesize / 1024)
911 n = kvm_getswapinfo(kd, swapary, 1, 0);
912 if (n < 0 || swapary[0].ksw_total == 0)
915 *retavail = CONVERT(swapary[0].ksw_total);
916 *retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
918 n = (int)((double)swapary[0].ksw_used * 100.0 /
919 (double)swapary[0].ksw_total);