blob: 223a35efa0a6bb4abbcc55ff627a5257716d1e75 [file] [log] [blame]
Frederic Weisbecker73fbec62012-06-16 15:57:37 +02001#include <linux/export.h>
2#include <linux/sched.h>
3#include <linux/tsacct_kern.h>
4#include <linux/kernel_stat.h>
5#include <linux/static_key.h>
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +02006#include <linux/context_tracking.h>
Frederic Weisbecker73fbec62012-06-16 15:57:37 +02007#include "sched.h"
8
9
10#ifdef CONFIG_IRQ_TIME_ACCOUNTING
11
12/*
13 * There are no locks covering percpu hardirq/softirq time.
Frederic Weisbeckerbf9fae92012-09-08 15:23:11 +020014 * They are only modified in vtime_account, on corresponding CPU
Frederic Weisbecker73fbec62012-06-16 15:57:37 +020015 * with interrupts disabled. So, writes are safe.
16 * They are read and saved off onto struct rq in update_rq_clock().
17 * This may result in other CPU reading this CPU's irq time and can
Frederic Weisbeckerbf9fae92012-09-08 15:23:11 +020018 * race with irq/vtime_account on this CPU. We would either get old
Frederic Weisbecker73fbec62012-06-16 15:57:37 +020019 * or new value with a side effect of accounting a slice of irq time to wrong
20 * task when irq is in progress while we read rq->clock. That is a worthy
21 * compromise in place of having locks on each irq in account_system_time.
22 */
23DEFINE_PER_CPU(u64, cpu_hardirq_time);
24DEFINE_PER_CPU(u64, cpu_softirq_time);
25
26static DEFINE_PER_CPU(u64, irq_start_time);
27static int sched_clock_irqtime;
28
29void enable_sched_clock_irqtime(void)
30{
31 sched_clock_irqtime = 1;
32}
33
34void disable_sched_clock_irqtime(void)
35{
36 sched_clock_irqtime = 0;
37}
38
39#ifndef CONFIG_64BIT
40DEFINE_PER_CPU(seqcount_t, irq_time_seq);
41#endif /* CONFIG_64BIT */
42
43/*
44 * Called before incrementing preempt_count on {soft,}irq_enter
45 * and before decrementing preempt_count on {soft,}irq_exit.
46 */
Frederic Weisbecker3e1df4f52012-10-06 05:23:22 +020047void irqtime_account_irq(struct task_struct *curr)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +020048{
49 unsigned long flags;
50 s64 delta;
51 int cpu;
52
53 if (!sched_clock_irqtime)
54 return;
55
56 local_irq_save(flags);
57
58 cpu = smp_processor_id();
59 delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time);
60 __this_cpu_add(irq_start_time, delta);
61
62 irq_time_write_begin();
63 /*
64 * We do not account for softirq time from ksoftirqd here.
65 * We want to continue accounting softirq time to ksoftirqd thread
66 * in that case, so as not to confuse scheduler with a special task
67 * that do not consume any time, but still wants to run.
68 */
69 if (hardirq_count())
70 __this_cpu_add(cpu_hardirq_time, delta);
71 else if (in_serving_softirq() && curr != this_cpu_ksoftirqd())
72 __this_cpu_add(cpu_softirq_time, delta);
73
74 irq_time_write_end();
75 local_irq_restore(flags);
76}
Frederic Weisbecker3e1df4f52012-10-06 05:23:22 +020077EXPORT_SYMBOL_GPL(irqtime_account_irq);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +020078
79static int irqtime_account_hi_update(void)
80{
81 u64 *cpustat = kcpustat_this_cpu->cpustat;
82 unsigned long flags;
83 u64 latest_ns;
84 int ret = 0;
85
86 local_irq_save(flags);
87 latest_ns = this_cpu_read(cpu_hardirq_time);
88 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ])
89 ret = 1;
90 local_irq_restore(flags);
91 return ret;
92}
93
94static int irqtime_account_si_update(void)
95{
96 u64 *cpustat = kcpustat_this_cpu->cpustat;
97 unsigned long flags;
98 u64 latest_ns;
99 int ret = 0;
100
101 local_irq_save(flags);
102 latest_ns = this_cpu_read(cpu_softirq_time);
103 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ])
104 ret = 1;
105 local_irq_restore(flags);
106 return ret;
107}
108
109#else /* CONFIG_IRQ_TIME_ACCOUNTING */
110
111#define sched_clock_irqtime (0)
112
113#endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
114
115static inline void task_group_account_field(struct task_struct *p, int index,
116 u64 tmp)
117{
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200118 /*
119 * Since all updates are sure to touch the root cgroup, we
120 * get ourselves ahead and touch it first. If the root cgroup
121 * is the only cgroup, then nothing else should be necessary.
122 *
123 */
124 __get_cpu_var(kernel_cpustat).cpustat[index] += tmp;
125
Li Zefan1966aaf2013-03-29 14:37:06 +0800126 cpuacct_account_field(p, index, tmp);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200127}
128
129/*
130 * Account user cpu time to a process.
131 * @p: the process that the cpu time gets accounted to
132 * @cputime: the cpu time spent in user space since the last update
133 * @cputime_scaled: cputime scaled by cpu frequency
134 */
135void account_user_time(struct task_struct *p, cputime_t cputime,
136 cputime_t cputime_scaled)
137{
138 int index;
139
140 /* Add user time to process. */
141 p->utime += cputime;
142 p->utimescaled += cputime_scaled;
143 account_group_user_time(p, cputime);
144
145 index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER;
146
147 /* Add user time to cpustat. */
148 task_group_account_field(p, index, (__force u64) cputime);
149
150 /* Account for user time used */
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100151 acct_account_cputime(p);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200152}
153
154/*
155 * Account guest cpu time to a process.
156 * @p: the process that the cpu time gets accounted to
157 * @cputime: the cpu time spent in virtual machine since the last update
158 * @cputime_scaled: cputime scaled by cpu frequency
159 */
160static void account_guest_time(struct task_struct *p, cputime_t cputime,
161 cputime_t cputime_scaled)
162{
163 u64 *cpustat = kcpustat_this_cpu->cpustat;
164
165 /* Add guest time to process. */
166 p->utime += cputime;
167 p->utimescaled += cputime_scaled;
168 account_group_user_time(p, cputime);
169 p->gtime += cputime;
170
171 /* Add guest time to cpustat. */
172 if (TASK_NICE(p) > 0) {
173 cpustat[CPUTIME_NICE] += (__force u64) cputime;
174 cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime;
175 } else {
176 cpustat[CPUTIME_USER] += (__force u64) cputime;
177 cpustat[CPUTIME_GUEST] += (__force u64) cputime;
178 }
179}
180
181/*
182 * Account system cpu time to a process and desired cpustat field
183 * @p: the process that the cpu time gets accounted to
184 * @cputime: the cpu time spent in kernel space since the last update
185 * @cputime_scaled: cputime scaled by cpu frequency
186 * @target_cputime64: pointer to cpustat field that has to be updated
187 */
188static inline
189void __account_system_time(struct task_struct *p, cputime_t cputime,
190 cputime_t cputime_scaled, int index)
191{
192 /* Add system time to process. */
193 p->stime += cputime;
194 p->stimescaled += cputime_scaled;
195 account_group_system_time(p, cputime);
196
197 /* Add system time to cpustat. */
198 task_group_account_field(p, index, (__force u64) cputime);
199
200 /* Account for system time used */
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100201 acct_account_cputime(p);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200202}
203
204/*
205 * Account system cpu time to a process.
206 * @p: the process that the cpu time gets accounted to
207 * @hardirq_offset: the offset to subtract from hardirq_count()
208 * @cputime: the cpu time spent in kernel space since the last update
209 * @cputime_scaled: cputime scaled by cpu frequency
210 */
211void account_system_time(struct task_struct *p, int hardirq_offset,
212 cputime_t cputime, cputime_t cputime_scaled)
213{
214 int index;
215
216 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
217 account_guest_time(p, cputime, cputime_scaled);
218 return;
219 }
220
221 if (hardirq_count() - hardirq_offset)
222 index = CPUTIME_IRQ;
223 else if (in_serving_softirq())
224 index = CPUTIME_SOFTIRQ;
225 else
226 index = CPUTIME_SYSTEM;
227
228 __account_system_time(p, cputime, cputime_scaled, index);
229}
230
231/*
232 * Account for involuntary wait time.
233 * @cputime: the cpu time spent in involuntary wait
234 */
235void account_steal_time(cputime_t cputime)
236{
237 u64 *cpustat = kcpustat_this_cpu->cpustat;
238
239 cpustat[CPUTIME_STEAL] += (__force u64) cputime;
240}
241
242/*
243 * Account for idle time.
244 * @cputime: the cpu time spent in idle wait
245 */
246void account_idle_time(cputime_t cputime)
247{
248 u64 *cpustat = kcpustat_this_cpu->cpustat;
249 struct rq *rq = this_rq();
250
251 if (atomic_read(&rq->nr_iowait) > 0)
252 cpustat[CPUTIME_IOWAIT] += (__force u64) cputime;
253 else
254 cpustat[CPUTIME_IDLE] += (__force u64) cputime;
255}
256
257static __always_inline bool steal_account_process_tick(void)
258{
259#ifdef CONFIG_PARAVIRT
260 if (static_key_false(&paravirt_steal_enabled)) {
261 u64 steal, st = 0;
262
263 steal = paravirt_steal_clock(smp_processor_id());
264 steal -= this_rq()->prev_steal_time;
265
266 st = steal_ticks(steal);
267 this_rq()->prev_steal_time += st * TICK_NSEC;
268
269 account_steal_time(st);
270 return st;
271 }
272#endif
273 return false;
274}
275
Frederic Weisbeckera634f932012-11-21 15:55:59 +0100276/*
277 * Accumulate raw cputime values of dead tasks (sig->[us]time) and live
278 * tasks (sum on group iteration) belonging to @tsk's group.
279 */
280void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
281{
282 struct signal_struct *sig = tsk->signal;
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100283 cputime_t utime, stime;
Frederic Weisbeckera634f932012-11-21 15:55:59 +0100284 struct task_struct *t;
285
286 times->utime = sig->utime;
287 times->stime = sig->stime;
288 times->sum_exec_runtime = sig->sum_sched_runtime;
289
290 rcu_read_lock();
291 /* make sure we can trust tsk->thread_group list */
292 if (!likely(pid_alive(tsk)))
293 goto out;
294
295 t = tsk;
296 do {
Stanislaw Gruszkae614b332013-04-04 10:57:48 +0200297 task_cputime(t, &utime, &stime);
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100298 times->utime += utime;
299 times->stime += stime;
Frederic Weisbeckera634f932012-11-21 15:55:59 +0100300 times->sum_exec_runtime += task_sched_runtime(t);
301 } while_each_thread(tsk, t);
302out:
303 rcu_read_unlock();
304}
305
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200306#ifdef CONFIG_IRQ_TIME_ACCOUNTING
307/*
308 * Account a tick to a process and cpustat
309 * @p: the process that the cpu time gets accounted to
310 * @user_tick: is the tick from userspace
311 * @rq: the pointer to rq
312 *
313 * Tick demultiplexing follows the order
314 * - pending hardirq update
315 * - pending softirq update
316 * - user_time
317 * - idle_time
318 * - system time
319 * - check for guest_time
320 * - else account as system_time
321 *
322 * Check for hardirq is done both for system and user time as there is
323 * no timer going off while we are on hardirq and hence we may never get an
324 * opportunity to update it solely in system time.
325 * p->stime and friends are only updated on system time and not on irq
326 * softirq as those do not count in task exec_runtime any more.
327 */
328static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
329 struct rq *rq)
330{
331 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
332 u64 *cpustat = kcpustat_this_cpu->cpustat;
333
334 if (steal_account_process_tick())
335 return;
336
337 if (irqtime_account_hi_update()) {
338 cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy;
339 } else if (irqtime_account_si_update()) {
340 cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy;
341 } else if (this_cpu_ksoftirqd() == p) {
342 /*
343 * ksoftirqd time do not get accounted in cpu_softirq_time.
344 * So, we have to handle it separately here.
345 * Also, p->stime needs to be updated for ksoftirqd.
346 */
347 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
348 CPUTIME_SOFTIRQ);
349 } else if (user_tick) {
350 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
351 } else if (p == rq->idle) {
352 account_idle_time(cputime_one_jiffy);
353 } else if (p->flags & PF_VCPU) { /* System time or guest time */
354 account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled);
355 } else {
356 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
357 CPUTIME_SYSTEM);
358 }
359}
360
361static void irqtime_account_idle_ticks(int ticks)
362{
363 int i;
364 struct rq *rq = this_rq();
365
366 for (i = 0; i < ticks; i++)
367 irqtime_account_process_tick(current, 0, rq);
368}
369#else /* CONFIG_IRQ_TIME_ACCOUNTING */
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200370static inline void irqtime_account_idle_ticks(int ticks) {}
371static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick,
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200372 struct rq *rq) {}
373#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
374
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200375/*
376 * Use precise platform statistics if available:
377 */
378#ifdef CONFIG_VIRT_CPU_ACCOUNTING
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200379
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100380#ifndef __ARCH_HAS_VTIME_TASK_SWITCH
381void vtime_task_switch(struct task_struct *prev)
382{
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200383 if (!vtime_accounting_enabled())
384 return;
385
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100386 if (is_idle_task(prev))
387 vtime_account_idle(prev);
388 else
389 vtime_account_system(prev);
390
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200391#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100392 vtime_account_user(prev);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200393#endif
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100394 arch_vtime_task_switch(prev);
395}
396#endif
Frederic Weisbecker11113332012-10-24 18:05:51 +0200397
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200398/*
399 * Archs that account the whole time spent in the idle task
400 * (outside irq) as idle time can rely on this and just implement
Frederic Weisbeckerfd25b4c2012-11-13 18:21:22 +0100401 * vtime_account_system() and vtime_account_idle(). Archs that
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200402 * have other meaning of the idle time (s390 only includes the
403 * time spent by the CPU when it's in low power mode) must override
404 * vtime_account().
405 */
406#ifndef __ARCH_HAS_VTIME_ACCOUNT
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100407void vtime_account_irq_enter(struct task_struct *tsk)
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200408{
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200409 if (!vtime_accounting_enabled())
410 return;
411
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200412 if (!in_interrupt()) {
413 /*
414 * If we interrupted user, context_tracking_in_user()
415 * is 1 because the context tracking don't hook
416 * on irq entry/exit. This way we know if
417 * we need to flush user time on kernel entry.
418 */
419 if (context_tracking_in_user()) {
420 vtime_account_user(tsk);
421 return;
422 }
423
424 if (is_idle_task(tsk)) {
425 vtime_account_idle(tsk);
426 return;
427 }
428 }
429 vtime_account_system(tsk);
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200430}
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100431EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200432#endif /* __ARCH_HAS_VTIME_ACCOUNT */
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100433#endif /* CONFIG_VIRT_CPU_ACCOUNTING */
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200434
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200435
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100436#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
437void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200438{
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100439 *ut = p->utime;
440 *st = p->stime;
441}
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200442
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100443void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
444{
445 struct task_cputime cputime;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200446
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100447 thread_group_cputime(p, &cputime);
448
449 *ut = cputime.utime;
450 *st = cputime.stime;
451}
452#else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
453/*
454 * Account a single tick of cpu time.
455 * @p: the process that the cpu time gets accounted to
456 * @user_tick: indicates if the tick is a user or a system tick
457 */
458void account_process_tick(struct task_struct *p, int user_tick)
459{
460 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
461 struct rq *rq = this_rq();
462
463 if (vtime_accounting_enabled())
464 return;
465
466 if (sched_clock_irqtime) {
467 irqtime_account_process_tick(p, user_tick, rq);
468 return;
469 }
470
471 if (steal_account_process_tick())
472 return;
473
474 if (user_tick)
475 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
476 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
477 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
478 one_jiffy_scaled);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200479 else
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100480 account_idle_time(cputime_one_jiffy);
481}
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200482
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100483/*
484 * Account multiple ticks of steal time.
485 * @p: the process from which the cpu time has been stolen
486 * @ticks: number of stolen ticks
487 */
488void account_steal_ticks(unsigned long ticks)
489{
490 account_steal_time(jiffies_to_cputime(ticks));
491}
492
493/*
494 * Account multiple ticks of idle time.
495 * @ticks: number of stolen ticks
496 */
497void account_idle_ticks(unsigned long ticks)
498{
499
500 if (sched_clock_irqtime) {
501 irqtime_account_idle_ticks(ticks);
502 return;
503 }
504
505 account_idle_time(jiffies_to_cputime(ticks));
506}
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200507
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100508/*
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200509 * Perform (stime * rtime) / total, but avoid multiplication overflow by
510 * loosing precision when the numbers are big.
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100511 */
512static cputime_t scale_stime(u64 stime, u64 rtime, u64 total)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200513{
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200514 u64 scaled;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200515
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200516 for (;;) {
517 /* Make sure "rtime" is the bigger of stime/rtime */
Stanislaw Gruszka84f9f3a2013-05-02 15:34:33 +0200518 if (stime > rtime)
519 swap(rtime, stime);
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200520
521 /* Make sure 'total' fits in 32 bits */
522 if (total >> 32)
523 goto drop_precision;
524
525 /* Does rtime (and thus stime) fit in 32 bits? */
526 if (!(rtime >> 32))
527 break;
528
529 /* Can we just balance rtime/stime rather than dropping bits? */
530 if (stime >> 31)
531 goto drop_precision;
532
533 /* We can grow stime and shrink rtime and try to make them both fit */
534 stime <<= 1;
535 rtime >>= 1;
536 continue;
537
538drop_precision:
539 /* We drop from rtime, it has more bits than stime */
540 rtime >>= 1;
541 total >>= 1;
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100542 }
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200543
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200544 /*
545 * Make sure gcc understands that this is a 32x32->64 multiply,
546 * followed by a 64/32->64 divide.
547 */
548 scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total);
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100549 return (__force cputime_t) scaled;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200550}
551
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100552/*
553 * Adjust tick based cputime random precision against scheduler
554 * runtime accounting.
555 */
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100556static void cputime_adjust(struct task_cputime *curr,
557 struct cputime *prev,
558 cputime_t *ut, cputime_t *st)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200559{
Stanislaw Gruszka68aa8ef2013-04-30 11:35:06 +0200560 cputime_t rtime, stime, utime, total;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200561
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100562 if (vtime_accounting_enabled()) {
563 *ut = curr->utime;
564 *st = curr->stime;
565 return;
566 }
567
Frederic Weisbecker62188452013-01-26 17:19:42 +0100568 stime = curr->stime;
569 total = stime + curr->utime;
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100570
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200571 /*
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100572 * Tick based cputime accounting depend on random scheduling
573 * timeslices of a task to be interrupted or not by the timer.
574 * Depending on these circumstances, the number of these interrupts
575 * may be over or under-optimistic, matching the real user and system
576 * cputime with a variable precision.
577 *
578 * Fix this by scaling these tick based values against the total
579 * runtime accounted by the CFS scheduler.
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200580 */
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100581 rtime = nsecs_to_cputime(curr->sum_exec_runtime);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200582
Stanislaw Gruszka772c8082013-04-30 11:35:05 +0200583 /*
584 * Update userspace visible utime/stime values only if actual execution
585 * time is bigger than already exported. Note that can happen, that we
586 * provided bigger values due to scaling inaccuracy on big numbers.
587 */
588 if (prev->stime + prev->utime >= rtime)
589 goto out;
590
Stanislaw Gruszka68aa8ef2013-04-30 11:35:06 +0200591 if (total) {
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100592 stime = scale_stime((__force u64)stime,
593 (__force u64)rtime, (__force u64)total);
Stanislaw Gruszka68aa8ef2013-04-30 11:35:06 +0200594 utime = rtime - stime;
595 } else {
596 stime = rtime;
597 utime = 0;
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100598 }
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200599
600 /*
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100601 * If the tick based count grows faster than the scheduler one,
602 * the result of the scaling may go backward.
603 * Let's enforce monotonicity.
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200604 */
Frederic Weisbecker62188452013-01-26 17:19:42 +0100605 prev->stime = max(prev->stime, stime);
Stanislaw Gruszka68aa8ef2013-04-30 11:35:06 +0200606 prev->utime = max(prev->utime, utime);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200607
Stanislaw Gruszka772c8082013-04-30 11:35:05 +0200608out:
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100609 *ut = prev->utime;
610 *st = prev->stime;
611}
612
613void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
614{
615 struct task_cputime cputime = {
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100616 .sum_exec_runtime = p->se.sum_exec_runtime,
617 };
618
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100619 task_cputime(p, &cputime.utime, &cputime.stime);
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100620 cputime_adjust(&cputime, &p->prev_cputime, ut, st);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200621}
622
623/*
624 * Must be called with siglock held.
625 */
Frederic Weisbeckere80d0a1a2012-11-21 16:26:44 +0100626void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200627{
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200628 struct task_cputime cputime;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200629
630 thread_group_cputime(p, &cputime);
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100631 cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200632}
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100633#endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200634
635#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100636static unsigned long long vtime_delta(struct task_struct *tsk)
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200637{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100638 unsigned long long clock;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200639
Frederic Weisbecker7f6575f2013-02-23 17:28:45 +0100640 clock = local_clock();
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100641 if (clock < tsk->vtime_snap)
642 return 0;
643
644 return clock - tsk->vtime_snap;
645}
646
647static cputime_t get_vtime_delta(struct task_struct *tsk)
648{
649 unsigned long long delta = vtime_delta(tsk);
650
651 WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_SLEEPING);
652 tsk->vtime_snap += delta;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200653
654 /* CHECKME: always safe to convert nsecs to cputime? */
655 return nsecs_to_cputime(delta);
656}
657
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100658static void __vtime_account_system(struct task_struct *tsk)
659{
660 cputime_t delta_cpu = get_vtime_delta(tsk);
661
662 account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu));
663}
664
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200665void vtime_account_system(struct task_struct *tsk)
666{
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200667 if (!vtime_accounting_enabled())
668 return;
669
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100670 write_seqlock(&tsk->vtime_seqlock);
671 __vtime_account_system(tsk);
672 write_sequnlock(&tsk->vtime_seqlock);
673}
674
675void vtime_account_irq_exit(struct task_struct *tsk)
676{
677 if (!vtime_accounting_enabled())
678 return;
679
680 write_seqlock(&tsk->vtime_seqlock);
681 if (context_tracking_in_user())
682 tsk->vtime_snap_whence = VTIME_USER;
683 __vtime_account_system(tsk);
684 write_sequnlock(&tsk->vtime_seqlock);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200685}
686
687void vtime_account_user(struct task_struct *tsk)
688{
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200689 cputime_t delta_cpu;
690
691 if (!vtime_accounting_enabled())
692 return;
693
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100694 delta_cpu = get_vtime_delta(tsk);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200695
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100696 write_seqlock(&tsk->vtime_seqlock);
697 tsk->vtime_snap_whence = VTIME_SYS;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200698 account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100699 write_sequnlock(&tsk->vtime_seqlock);
700}
701
702void vtime_user_enter(struct task_struct *tsk)
703{
704 if (!vtime_accounting_enabled())
705 return;
706
707 write_seqlock(&tsk->vtime_seqlock);
708 tsk->vtime_snap_whence = VTIME_USER;
709 __vtime_account_system(tsk);
710 write_sequnlock(&tsk->vtime_seqlock);
711}
712
713void vtime_guest_enter(struct task_struct *tsk)
714{
Frederic Weisbecker5b206d42013-07-12 19:05:14 +0200715 /*
716 * The flags must be updated under the lock with
717 * the vtime_snap flush and update.
718 * That enforces a right ordering and update sequence
719 * synchronization against the reader (task_gtime())
720 * that can thus safely catch up with a tickless delta.
721 */
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100722 write_seqlock(&tsk->vtime_seqlock);
723 __vtime_account_system(tsk);
724 current->flags |= PF_VCPU;
725 write_sequnlock(&tsk->vtime_seqlock);
726}
727
728void vtime_guest_exit(struct task_struct *tsk)
729{
730 write_seqlock(&tsk->vtime_seqlock);
731 __vtime_account_system(tsk);
732 current->flags &= ~PF_VCPU;
733 write_sequnlock(&tsk->vtime_seqlock);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200734}
735
736void vtime_account_idle(struct task_struct *tsk)
737{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100738 cputime_t delta_cpu = get_vtime_delta(tsk);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200739
740 account_idle_time(delta_cpu);
741}
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200742
743bool vtime_accounting_enabled(void)
744{
745 return context_tracking_active();
746}
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100747
748void arch_vtime_task_switch(struct task_struct *prev)
749{
750 write_seqlock(&prev->vtime_seqlock);
751 prev->vtime_snap_whence = VTIME_SLEEPING;
752 write_sequnlock(&prev->vtime_seqlock);
753
754 write_seqlock(&current->vtime_seqlock);
755 current->vtime_snap_whence = VTIME_SYS;
Frederic Weisbecker45eacc62013-05-15 22:16:32 +0200756 current->vtime_snap = sched_clock_cpu(smp_processor_id());
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100757 write_sequnlock(&current->vtime_seqlock);
758}
759
Frederic Weisbecker45eacc62013-05-15 22:16:32 +0200760void vtime_init_idle(struct task_struct *t, int cpu)
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100761{
762 unsigned long flags;
763
764 write_seqlock_irqsave(&t->vtime_seqlock, flags);
765 t->vtime_snap_whence = VTIME_SYS;
Frederic Weisbecker45eacc62013-05-15 22:16:32 +0200766 t->vtime_snap = sched_clock_cpu(cpu);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100767 write_sequnlock_irqrestore(&t->vtime_seqlock, flags);
768}
769
770cputime_t task_gtime(struct task_struct *t)
771{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100772 unsigned int seq;
773 cputime_t gtime;
774
775 do {
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100776 seq = read_seqbegin(&t->vtime_seqlock);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100777
778 gtime = t->gtime;
779 if (t->flags & PF_VCPU)
780 gtime += vtime_delta(t);
781
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100782 } while (read_seqretry(&t->vtime_seqlock, seq));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100783
784 return gtime;
785}
786
787/*
788 * Fetch cputime raw values from fields of task_struct and
789 * add up the pending nohz execution time since the last
790 * cputime snapshot.
791 */
792static void
793fetch_task_cputime(struct task_struct *t,
794 cputime_t *u_dst, cputime_t *s_dst,
795 cputime_t *u_src, cputime_t *s_src,
796 cputime_t *udelta, cputime_t *sdelta)
797{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100798 unsigned int seq;
799 unsigned long long delta;
800
801 do {
802 *udelta = 0;
803 *sdelta = 0;
804
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100805 seq = read_seqbegin(&t->vtime_seqlock);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100806
807 if (u_dst)
808 *u_dst = *u_src;
809 if (s_dst)
810 *s_dst = *s_src;
811
812 /* Task is sleeping, nothing to add */
813 if (t->vtime_snap_whence == VTIME_SLEEPING ||
814 is_idle_task(t))
815 continue;
816
817 delta = vtime_delta(t);
818
819 /*
820 * Task runs either in user or kernel space, add pending nohz time to
821 * the right place.
822 */
823 if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) {
824 *udelta = delta;
825 } else {
826 if (t->vtime_snap_whence == VTIME_SYS)
827 *sdelta = delta;
828 }
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100829 } while (read_seqretry(&t->vtime_seqlock, seq));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100830}
831
832
833void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime)
834{
835 cputime_t udelta, sdelta;
836
837 fetch_task_cputime(t, utime, stime, &t->utime,
838 &t->stime, &udelta, &sdelta);
839 if (utime)
840 *utime += udelta;
841 if (stime)
842 *stime += sdelta;
843}
844
845void task_cputime_scaled(struct task_struct *t,
846 cputime_t *utimescaled, cputime_t *stimescaled)
847{
848 cputime_t udelta, sdelta;
849
850 fetch_task_cputime(t, utimescaled, stimescaled,
851 &t->utimescaled, &t->stimescaled, &udelta, &sdelta);
852 if (utimescaled)
853 *utimescaled += cputime_to_scaled(udelta);
854 if (stimescaled)
855 *stimescaled += cputime_to_scaled(sdelta);
856}
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200857#endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */