taskstats: add context-switch counters
[linux-2.6.git] / fs / proc / array.c
1 /*
2  *  linux/fs/proc/array.c
3  *
4  *  Copyright (C) 1992  by Linus Torvalds
5  *  based on ideas by Darren Senn
6  *
7  * Fixes:
8  * Michael. K. Johnson: stat,statm extensions.
9  *                      <johnsonm@stolaf.edu>
10  *
11  * Pauline Middelink :  Made cmdline,envline only break at '\0's, to
12  *                      make sure SET_PROCTITLE works. Also removed
13  *                      bad '!' which forced address recalculation for
14  *                      EVERY character on the current page.
15  *                      <middelin@polyware.iaf.nl>
16  *
17  * Danny ter Haar    :  added cpuinfo
18  *                      <dth@cistron.nl>
19  *
20  * Alessandro Rubini :  profile extension.
21  *                      <rubini@ipvvis.unipv.it>
22  *
23  * Jeff Tranter      :  added BogoMips field to cpuinfo
24  *                      <Jeff_Tranter@Mitel.COM>
25  *
26  * Bruno Haible      :  remove 4K limit for the maps file
27  *                      <haible@ma2s2.mathematik.uni-karlsruhe.de>
28  *
29  * Yves Arrouye      :  remove removal of trailing spaces in get_array.
30  *                      <Yves.Arrouye@marin.fdn.fr>
31  *
32  * Jerome Forissier  :  added per-CPU time information to /proc/stat
33  *                      and /proc/<pid>/cpu extension
34  *                      <forissier@isia.cma.fr>
35  *                      - Incorporation and non-SMP safe operation
36  *                      of forissier patch in 2.1.78 by
37  *                      Hans Marcus <crowbar@concepts.nl>
38  *
39  * aeb@cwi.nl        :  /proc/partitions
40  *
41  *
42  * Alan Cox          :  security fixes.
43  *                      <Alan.Cox@linux.org>
44  *
45  * Al Viro           :  safe handling of mm_struct
46  *
47  * Gerhard Wichert   :  added BIGMEM support
48  * Siemens AG           <Gerhard.Wichert@pdb.siemens.de>
49  *
50  * Al Viro & Jeff Garzik :  moved most of the thing into base.c and
51  *                       :  proc_misc.c. The rest may eventually go into
52  *                       :  base.c too.
53  */
54
55 #include <linux/types.h>
56 #include <linux/errno.h>
57 #include <linux/time.h>
58 #include <linux/kernel.h>
59 #include <linux/kernel_stat.h>
60 #include <linux/tty.h>
61 #include <linux/string.h>
62 #include <linux/mman.h>
63 #include <linux/proc_fs.h>
64 #include <linux/ioport.h>
65 #include <linux/mm.h>
66 #include <linux/hugetlb.h>
67 #include <linux/pagemap.h>
68 #include <linux/swap.h>
69 #include <linux/slab.h>
70 #include <linux/smp.h>
71 #include <linux/signal.h>
72 #include <linux/highmem.h>
73 #include <linux/file.h>
74 #include <linux/times.h>
75 #include <linux/cpuset.h>
76 #include <linux/rcupdate.h>
77 #include <linux/delayacct.h>
78
79 #include <asm/uaccess.h>
80 #include <asm/pgtable.h>
81 #include <asm/io.h>
82 #include <asm/processor.h>
83 #include "internal.h"
84
85 /* Gcc optimizes away "strlen(x)" for constant x */
86 #define ADDBUF(buffer, string) \
87 do { memcpy(buffer, string, strlen(string)); \
88      buffer += strlen(string); } while (0)
89
90 static inline char * task_name(struct task_struct *p, char * buf)
91 {
92         int i;
93         char * name;
94         char tcomm[sizeof(p->comm)];
95
96         get_task_comm(tcomm, p);
97
98         ADDBUF(buf, "Name:\t");
99         name = tcomm;
100         i = sizeof(tcomm);
101         do {
102                 unsigned char c = *name;
103                 name++;
104                 i--;
105                 *buf = c;
106                 if (!c)
107                         break;
108                 if (c == '\\') {
109                         buf[1] = c;
110                         buf += 2;
111                         continue;
112                 }
113                 if (c == '\n') {
114                         buf[0] = '\\';
115                         buf[1] = 'n';
116                         buf += 2;
117                         continue;
118                 }
119                 buf++;
120         } while (i);
121         *buf = '\n';
122         return buf+1;
123 }
124
125 /*
126  * The task state array is a strange "bitmap" of
127  * reasons to sleep. Thus "running" is zero, and
128  * you can test for combinations of others with
129  * simple bit tests.
130  */
131 static const char *task_state_array[] = {
132         "R (running)",          /*  0 */
133         "S (sleeping)",         /*  1 */
134         "D (disk sleep)",       /*  2 */
135         "T (stopped)",          /*  4 */
136         "T (tracing stop)",     /*  8 */
137         "Z (zombie)",           /* 16 */
138         "X (dead)"              /* 32 */
139 };
140
141 static inline const char * get_task_state(struct task_struct *tsk)
142 {
143         unsigned int state = (tsk->state & (TASK_RUNNING |
144                                             TASK_INTERRUPTIBLE |
145                                             TASK_UNINTERRUPTIBLE |
146                                             TASK_STOPPED |
147                                             TASK_TRACED)) |
148                         (tsk->exit_state & (EXIT_ZOMBIE |
149                                             EXIT_DEAD));
150         const char **p = &task_state_array[0];
151
152         while (state) {
153                 p++;
154                 state >>= 1;
155         }
156         return *p;
157 }
158
159 static inline char * task_state(struct task_struct *p, char *buffer)
160 {
161         struct group_info *group_info;
162         int g;
163         struct fdtable *fdt = NULL;
164
165         rcu_read_lock();
166         buffer += sprintf(buffer,
167                 "State:\t%s\n"
168                 "Tgid:\t%d\n"
169                 "Pid:\t%d\n"
170                 "PPid:\t%d\n"
171                 "TracerPid:\t%d\n"
172                 "Uid:\t%d\t%d\t%d\t%d\n"
173                 "Gid:\t%d\t%d\t%d\t%d\n",
174                 get_task_state(p),
175                 p->tgid, p->pid,
176                 pid_alive(p) ? rcu_dereference(p->real_parent)->tgid : 0,
177                 pid_alive(p) && p->ptrace ? rcu_dereference(p->parent)->pid : 0,
178                 p->uid, p->euid, p->suid, p->fsuid,
179                 p->gid, p->egid, p->sgid, p->fsgid);
180
181         task_lock(p);
182         if (p->files)
183                 fdt = files_fdtable(p->files);
184         buffer += sprintf(buffer,
185                 "FDSize:\t%d\n"
186                 "Groups:\t",
187                 fdt ? fdt->max_fds : 0);
188         rcu_read_unlock();
189
190         group_info = p->group_info;
191         get_group_info(group_info);
192         task_unlock(p);
193
194         for (g = 0; g < min(group_info->ngroups,NGROUPS_SMALL); g++)
195                 buffer += sprintf(buffer, "%d ", GROUP_AT(group_info,g));
196         put_group_info(group_info);
197
198         buffer += sprintf(buffer, "\n");
199         return buffer;
200 }
201
202 static char * render_sigset_t(const char *header, sigset_t *set, char *buffer)
203 {
204         int i, len;
205
206         len = strlen(header);
207         memcpy(buffer, header, len);
208         buffer += len;
209
210         i = _NSIG;
211         do {
212                 int x = 0;
213
214                 i -= 4;
215                 if (sigismember(set, i+1)) x |= 1;
216                 if (sigismember(set, i+2)) x |= 2;
217                 if (sigismember(set, i+3)) x |= 4;
218                 if (sigismember(set, i+4)) x |= 8;
219                 *buffer++ = (x < 10 ? '0' : 'a' - 10) + x;
220         } while (i >= 4);
221
222         *buffer++ = '\n';
223         *buffer = 0;
224         return buffer;
225 }
226
227 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign,
228                                     sigset_t *catch)
229 {
230         struct k_sigaction *k;
231         int i;
232
233         k = p->sighand->action;
234         for (i = 1; i <= _NSIG; ++i, ++k) {
235                 if (k->sa.sa_handler == SIG_IGN)
236                         sigaddset(ign, i);
237                 else if (k->sa.sa_handler != SIG_DFL)
238                         sigaddset(catch, i);
239         }
240 }
241
242 static inline char * task_sig(struct task_struct *p, char *buffer)
243 {
244         unsigned long flags;
245         sigset_t pending, shpending, blocked, ignored, caught;
246         int num_threads = 0;
247         unsigned long qsize = 0;
248         unsigned long qlim = 0;
249
250         sigemptyset(&pending);
251         sigemptyset(&shpending);
252         sigemptyset(&blocked);
253         sigemptyset(&ignored);
254         sigemptyset(&caught);
255
256         rcu_read_lock();
257         if (lock_task_sighand(p, &flags)) {
258                 pending = p->pending.signal;
259                 shpending = p->signal->shared_pending.signal;
260                 blocked = p->blocked;
261                 collect_sigign_sigcatch(p, &ignored, &caught);
262                 num_threads = atomic_read(&p->signal->count);
263                 qsize = atomic_read(&p->user->sigpending);
264                 qlim = p->signal->rlim[RLIMIT_SIGPENDING].rlim_cur;
265                 unlock_task_sighand(p, &flags);
266         }
267         rcu_read_unlock();
268
269         buffer += sprintf(buffer, "Threads:\t%d\n", num_threads);
270         buffer += sprintf(buffer, "SigQ:\t%lu/%lu\n", qsize, qlim);
271
272         /* render them all */
273         buffer = render_sigset_t("SigPnd:\t", &pending, buffer);
274         buffer = render_sigset_t("ShdPnd:\t", &shpending, buffer);
275         buffer = render_sigset_t("SigBlk:\t", &blocked, buffer);
276         buffer = render_sigset_t("SigIgn:\t", &ignored, buffer);
277         buffer = render_sigset_t("SigCgt:\t", &caught, buffer);
278
279         return buffer;
280 }
281
282 static inline char *task_cap(struct task_struct *p, char *buffer)
283 {
284     return buffer + sprintf(buffer, "CapInh:\t%016x\n"
285                             "CapPrm:\t%016x\n"
286                             "CapEff:\t%016x\n",
287                             cap_t(p->cap_inheritable),
288                             cap_t(p->cap_permitted),
289                             cap_t(p->cap_effective));
290 }
291
292 static inline char *task_context_switch_counts(struct task_struct *p,
293                                                 char *buffer)
294 {
295         return buffer + sprintf(buffer, "voluntary_ctxt_switches:\t%lu\n"
296                             "nonvoluntary_ctxt_switches:\t%lu\n",
297                             p->nvcsw,
298                             p->nivcsw);
299 }
300
301 int proc_pid_status(struct task_struct *task, char * buffer)
302 {
303         char * orig = buffer;
304         struct mm_struct *mm = get_task_mm(task);
305
306         buffer = task_name(task, buffer);
307         buffer = task_state(task, buffer);
308  
309         if (mm) {
310                 buffer = task_mem(mm, buffer);
311                 mmput(mm);
312         }
313         buffer = task_sig(task, buffer);
314         buffer = task_cap(task, buffer);
315         buffer = cpuset_task_status_allowed(task, buffer);
316 #if defined(CONFIG_S390)
317         buffer = task_show_regs(task, buffer);
318 #endif
319         buffer = task_context_switch_counts(task, buffer);
320         return buffer - orig;
321 }
322
323 static clock_t task_utime(struct task_struct *p)
324 {
325         clock_t utime = cputime_to_clock_t(p->utime),
326                 total = utime + cputime_to_clock_t(p->stime);
327         u64 temp;
328
329         /*
330          * Use CFS's precise accounting:
331          */
332         temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime);
333
334         if (total) {
335                 temp *= utime;
336                 do_div(temp, total);
337         }
338         utime = (clock_t)temp;
339
340         return utime;
341 }
342
343 static clock_t task_stime(struct task_struct *p)
344 {
345         clock_t stime = cputime_to_clock_t(p->stime);
346
347         /*
348          * Use CFS's precise accounting. (we subtract utime from
349          * the total, to make sure the total observed by userspace
350          * grows monotonically - apps rely on that):
351          */
352         stime = nsec_to_clock_t(p->se.sum_exec_runtime) - task_utime(p);
353
354         return stime;
355 }
356
357
358 static int do_task_stat(struct task_struct *task, char * buffer, int whole)
359 {
360         unsigned long vsize, eip, esp, wchan = ~0UL;
361         long priority, nice;
362         int tty_pgrp = -1, tty_nr = 0;
363         sigset_t sigign, sigcatch;
364         char state;
365         int res;
366         pid_t ppid = 0, pgid = -1, sid = -1;
367         int num_threads = 0;
368         struct mm_struct *mm;
369         unsigned long long start_time;
370         unsigned long cmin_flt = 0, cmaj_flt = 0;
371         unsigned long  min_flt = 0,  maj_flt = 0;
372         cputime_t cutime, cstime;
373         clock_t utime, stime;
374         unsigned long rsslim = 0;
375         char tcomm[sizeof(task->comm)];
376         unsigned long flags;
377
378         state = *get_task_state(task);
379         vsize = eip = esp = 0;
380         mm = get_task_mm(task);
381         if (mm) {
382                 vsize = task_vsize(mm);
383                 eip = KSTK_EIP(task);
384                 esp = KSTK_ESP(task);
385         }
386
387         get_task_comm(tcomm, task);
388
389         sigemptyset(&sigign);
390         sigemptyset(&sigcatch);
391         cutime = cstime = cputime_zero;
392         utime = stime = 0;
393
394         rcu_read_lock();
395         if (lock_task_sighand(task, &flags)) {
396                 struct signal_struct *sig = task->signal;
397
398                 if (sig->tty) {
399                         tty_pgrp = pid_nr(sig->tty->pgrp);
400                         tty_nr = new_encode_dev(tty_devnum(sig->tty));
401                 }
402
403                 num_threads = atomic_read(&sig->count);
404                 collect_sigign_sigcatch(task, &sigign, &sigcatch);
405
406                 cmin_flt = sig->cmin_flt;
407                 cmaj_flt = sig->cmaj_flt;
408                 cutime = sig->cutime;
409                 cstime = sig->cstime;
410                 rsslim = sig->rlim[RLIMIT_RSS].rlim_cur;
411
412                 /* add up live thread stats at the group level */
413                 if (whole) {
414                         struct task_struct *t = task;
415                         do {
416                                 min_flt += t->min_flt;
417                                 maj_flt += t->maj_flt;
418                                 utime += task_utime(t);
419                                 stime += task_stime(t);
420                                 t = next_thread(t);
421                         } while (t != task);
422
423                         min_flt += sig->min_flt;
424                         maj_flt += sig->maj_flt;
425                         utime += cputime_to_clock_t(sig->utime);
426                         stime += cputime_to_clock_t(sig->stime);
427                 }
428
429                 sid = signal_session(sig);
430                 pgid = process_group(task);
431                 ppid = rcu_dereference(task->real_parent)->tgid;
432
433                 unlock_task_sighand(task, &flags);
434         }
435         rcu_read_unlock();
436
437         if (!whole || num_threads<2)
438                 wchan = get_wchan(task);
439         if (!whole) {
440                 min_flt = task->min_flt;
441                 maj_flt = task->maj_flt;
442                 utime = task_utime(task);
443                 stime = task_stime(task);
444         }
445
446         /* scale priority and nice values from timeslices to -20..20 */
447         /* to make it look like a "normal" Unix priority/nice value  */
448         priority = task_prio(task);
449         nice = task_nice(task);
450
451         /* Temporary variable needed for gcc-2.96 */
452         /* convert timespec -> nsec*/
453         start_time =
454                 (unsigned long long)task->real_start_time.tv_sec * NSEC_PER_SEC
455                                 + task->real_start_time.tv_nsec;
456         /* convert nsec -> ticks */
457         start_time = nsec_to_clock_t(start_time);
458
459         res = sprintf(buffer,"%d (%s) %c %d %d %d %d %d %u %lu \
460 %lu %lu %lu %lu %lu %ld %ld %ld %ld %d 0 %llu %lu %ld %lu %lu %lu %lu %lu \
461 %lu %lu %lu %lu %lu %lu %lu %lu %d %d %u %u %llu\n",
462                 task->pid,
463                 tcomm,
464                 state,
465                 ppid,
466                 pgid,
467                 sid,
468                 tty_nr,
469                 tty_pgrp,
470                 task->flags,
471                 min_flt,
472                 cmin_flt,
473                 maj_flt,
474                 cmaj_flt,
475                 utime,
476                 stime,
477                 cputime_to_clock_t(cutime),
478                 cputime_to_clock_t(cstime),
479                 priority,
480                 nice,
481                 num_threads,
482                 start_time,
483                 vsize,
484                 mm ? get_mm_rss(mm) : 0,
485                 rsslim,
486                 mm ? mm->start_code : 0,
487                 mm ? mm->end_code : 0,
488                 mm ? mm->start_stack : 0,
489                 esp,
490                 eip,
491                 /* The signal information here is obsolete.
492                  * It must be decimal for Linux 2.0 compatibility.
493                  * Use /proc/#/status for real-time signals.
494                  */
495                 task->pending.signal.sig[0] & 0x7fffffffUL,
496                 task->blocked.sig[0] & 0x7fffffffUL,
497                 sigign      .sig[0] & 0x7fffffffUL,
498                 sigcatch    .sig[0] & 0x7fffffffUL,
499                 wchan,
500                 0UL,
501                 0UL,
502                 task->exit_signal,
503                 task_cpu(task),
504                 task->rt_priority,
505                 task->policy,
506                 (unsigned long long)delayacct_blkio_ticks(task));
507         if(mm)
508                 mmput(mm);
509         return res;
510 }
511
512 int proc_tid_stat(struct task_struct *task, char * buffer)
513 {
514         return do_task_stat(task, buffer, 0);
515 }
516
517 int proc_tgid_stat(struct task_struct *task, char * buffer)
518 {
519         return do_task_stat(task, buffer, 1);
520 }
521
522 int proc_pid_statm(struct task_struct *task, char *buffer)
523 {
524         int size = 0, resident = 0, shared = 0, text = 0, lib = 0, data = 0;
525         struct mm_struct *mm = get_task_mm(task);
526         
527         if (mm) {
528                 size = task_statm(mm, &shared, &text, &data, &resident);
529                 mmput(mm);
530         }
531
532         return sprintf(buffer,"%d %d %d %d %d %d %d\n",
533                        size, resident, shared, text, lib, data, 0);
534 }