blob: c1d7493825ae515d35e57deab35c8f8c8c2f8c10 [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
Frederic Weisbeckerb0493402013-07-12 03:10:15 +0200381void vtime_common_task_switch(struct task_struct *prev)
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100382{
383 if (is_idle_task(prev))
384 vtime_account_idle(prev);
385 else
386 vtime_account_system(prev);
387
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200388#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100389 vtime_account_user(prev);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200390#endif
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100391 arch_vtime_task_switch(prev);
392}
393#endif
Frederic Weisbecker11113332012-10-24 18:05:51 +0200394
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200395/*
396 * Archs that account the whole time spent in the idle task
397 * (outside irq) as idle time can rely on this and just implement
Frederic Weisbeckerfd25b4c2012-11-13 18:21:22 +0100398 * vtime_account_system() and vtime_account_idle(). Archs that
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200399 * have other meaning of the idle time (s390 only includes the
400 * time spent by the CPU when it's in low power mode) must override
401 * vtime_account().
402 */
403#ifndef __ARCH_HAS_VTIME_ACCOUNT
Frederic Weisbeckerb0493402013-07-12 03:10:15 +0200404void vtime_common_account_irq_enter(struct task_struct *tsk)
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200405{
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200406 if (!in_interrupt()) {
407 /*
408 * If we interrupted user, context_tracking_in_user()
409 * is 1 because the context tracking don't hook
410 * on irq entry/exit. This way we know if
411 * we need to flush user time on kernel entry.
412 */
413 if (context_tracking_in_user()) {
414 vtime_account_user(tsk);
415 return;
416 }
417
418 if (is_idle_task(tsk)) {
419 vtime_account_idle(tsk);
420 return;
421 }
422 }
423 vtime_account_system(tsk);
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200424}
Frederic Weisbeckerb0493402013-07-12 03:10:15 +0200425EXPORT_SYMBOL_GPL(vtime_common_account_irq_enter);
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200426#endif /* __ARCH_HAS_VTIME_ACCOUNT */
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100427#endif /* CONFIG_VIRT_CPU_ACCOUNTING */
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200428
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200429
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100430#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
431void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200432{
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100433 *ut = p->utime;
434 *st = p->stime;
435}
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200436
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100437void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
438{
439 struct task_cputime cputime;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200440
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100441 thread_group_cputime(p, &cputime);
442
443 *ut = cputime.utime;
444 *st = cputime.stime;
445}
446#else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
447/*
448 * Account a single tick of cpu time.
449 * @p: the process that the cpu time gets accounted to
450 * @user_tick: indicates if the tick is a user or a system tick
451 */
452void account_process_tick(struct task_struct *p, int user_tick)
453{
454 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
455 struct rq *rq = this_rq();
456
457 if (vtime_accounting_enabled())
458 return;
459
460 if (sched_clock_irqtime) {
461 irqtime_account_process_tick(p, user_tick, rq);
462 return;
463 }
464
465 if (steal_account_process_tick())
466 return;
467
468 if (user_tick)
469 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
470 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
471 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
472 one_jiffy_scaled);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200473 else
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100474 account_idle_time(cputime_one_jiffy);
475}
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200476
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100477/*
478 * Account multiple ticks of steal time.
479 * @p: the process from which the cpu time has been stolen
480 * @ticks: number of stolen ticks
481 */
482void account_steal_ticks(unsigned long ticks)
483{
484 account_steal_time(jiffies_to_cputime(ticks));
485}
486
487/*
488 * Account multiple ticks of idle time.
489 * @ticks: number of stolen ticks
490 */
491void account_idle_ticks(unsigned long ticks)
492{
493
494 if (sched_clock_irqtime) {
495 irqtime_account_idle_ticks(ticks);
496 return;
497 }
498
499 account_idle_time(jiffies_to_cputime(ticks));
500}
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200501
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100502/*
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200503 * Perform (stime * rtime) / total, but avoid multiplication overflow by
504 * loosing precision when the numbers are big.
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100505 */
506static cputime_t scale_stime(u64 stime, u64 rtime, u64 total)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200507{
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200508 u64 scaled;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200509
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200510 for (;;) {
511 /* Make sure "rtime" is the bigger of stime/rtime */
Stanislaw Gruszka84f9f3a2013-05-02 15:34:33 +0200512 if (stime > rtime)
513 swap(rtime, stime);
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200514
515 /* Make sure 'total' fits in 32 bits */
516 if (total >> 32)
517 goto drop_precision;
518
519 /* Does rtime (and thus stime) fit in 32 bits? */
520 if (!(rtime >> 32))
521 break;
522
523 /* Can we just balance rtime/stime rather than dropping bits? */
524 if (stime >> 31)
525 goto drop_precision;
526
527 /* We can grow stime and shrink rtime and try to make them both fit */
528 stime <<= 1;
529 rtime >>= 1;
530 continue;
531
532drop_precision:
533 /* We drop from rtime, it has more bits than stime */
534 rtime >>= 1;
535 total >>= 1;
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100536 }
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200537
Stanislaw Gruszka55eaa7c2013-04-30 17:14:42 +0200538 /*
539 * Make sure gcc understands that this is a 32x32->64 multiply,
540 * followed by a 64/32->64 divide.
541 */
542 scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total);
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100543 return (__force cputime_t) scaled;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200544}
545
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100546/*
547 * Adjust tick based cputime random precision against scheduler
548 * runtime accounting.
549 */
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100550static void cputime_adjust(struct task_cputime *curr,
551 struct cputime *prev,
552 cputime_t *ut, cputime_t *st)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200553{
Stanislaw Gruszka68aa8ef2013-04-30 11:35:06 +0200554 cputime_t rtime, stime, utime, total;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200555
Frederic Weisbecker62188452013-01-26 17:19:42 +0100556 stime = curr->stime;
557 total = stime + curr->utime;
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100558
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200559 /*
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100560 * Tick based cputime accounting depend on random scheduling
561 * timeslices of a task to be interrupted or not by the timer.
562 * Depending on these circumstances, the number of these interrupts
563 * may be over or under-optimistic, matching the real user and system
564 * cputime with a variable precision.
565 *
566 * Fix this by scaling these tick based values against the total
567 * runtime accounted by the CFS scheduler.
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200568 */
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100569 rtime = nsecs_to_cputime(curr->sum_exec_runtime);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200570
Stanislaw Gruszka772c8082013-04-30 11:35:05 +0200571 /*
572 * Update userspace visible utime/stime values only if actual execution
573 * time is bigger than already exported. Note that can happen, that we
574 * provided bigger values due to scaling inaccuracy on big numbers.
575 */
576 if (prev->stime + prev->utime >= rtime)
577 goto out;
578
Stanislaw Gruszka68aa8ef2013-04-30 11:35:06 +0200579 if (total) {
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100580 stime = scale_stime((__force u64)stime,
581 (__force u64)rtime, (__force u64)total);
Stanislaw Gruszka68aa8ef2013-04-30 11:35:06 +0200582 utime = rtime - stime;
583 } else {
584 stime = rtime;
585 utime = 0;
Frederic Weisbeckerd9a3c982013-02-20 18:54:55 +0100586 }
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200587
588 /*
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100589 * If the tick based count grows faster than the scheduler one,
590 * the result of the scaling may go backward.
591 * Let's enforce monotonicity.
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200592 */
Frederic Weisbecker62188452013-01-26 17:19:42 +0100593 prev->stime = max(prev->stime, stime);
Stanislaw Gruszka68aa8ef2013-04-30 11:35:06 +0200594 prev->utime = max(prev->utime, utime);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200595
Stanislaw Gruszka772c8082013-04-30 11:35:05 +0200596out:
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100597 *ut = prev->utime;
598 *st = prev->stime;
599}
600
601void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
602{
603 struct task_cputime cputime = {
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100604 .sum_exec_runtime = p->se.sum_exec_runtime,
605 };
606
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100607 task_cputime(p, &cputime.utime, &cputime.stime);
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100608 cputime_adjust(&cputime, &p->prev_cputime, ut, st);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200609}
610
611/*
612 * Must be called with siglock held.
613 */
Frederic Weisbeckere80d0a1a2012-11-21 16:26:44 +0100614void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200615{
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200616 struct task_cputime cputime;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200617
618 thread_group_cputime(p, &cputime);
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100619 cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200620}
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100621#endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200622
623#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100624static unsigned long long vtime_delta(struct task_struct *tsk)
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200625{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100626 unsigned long long clock;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200627
Frederic Weisbecker7f6575f2013-02-23 17:28:45 +0100628 clock = local_clock();
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100629 if (clock < tsk->vtime_snap)
630 return 0;
631
632 return clock - tsk->vtime_snap;
633}
634
635static cputime_t get_vtime_delta(struct task_struct *tsk)
636{
637 unsigned long long delta = vtime_delta(tsk);
638
639 WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_SLEEPING);
640 tsk->vtime_snap += delta;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200641
642 /* CHECKME: always safe to convert nsecs to cputime? */
643 return nsecs_to_cputime(delta);
644}
645
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100646static void __vtime_account_system(struct task_struct *tsk)
647{
648 cputime_t delta_cpu = get_vtime_delta(tsk);
649
650 account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu));
651}
652
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200653void vtime_account_system(struct task_struct *tsk)
654{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100655 write_seqlock(&tsk->vtime_seqlock);
656 __vtime_account_system(tsk);
657 write_sequnlock(&tsk->vtime_seqlock);
658}
659
Frederic Weisbeckerb0493402013-07-12 03:10:15 +0200660void vtime_gen_account_irq_exit(struct task_struct *tsk)
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100661{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100662 write_seqlock(&tsk->vtime_seqlock);
Frederic Weisbeckeraf2350b2013-07-15 16:35:55 +0200663 __vtime_account_system(tsk);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100664 if (context_tracking_in_user())
665 tsk->vtime_snap_whence = VTIME_USER;
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100666 write_sequnlock(&tsk->vtime_seqlock);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200667}
668
669void vtime_account_user(struct task_struct *tsk)
670{
Frederic Weisbecker54461562013-07-13 17:10:18 +0200671 cputime_t delta_cpu;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200672
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100673 write_seqlock(&tsk->vtime_seqlock);
Frederic Weisbecker54461562013-07-13 17:10:18 +0200674 delta_cpu = get_vtime_delta(tsk);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100675 tsk->vtime_snap_whence = VTIME_SYS;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200676 account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100677 write_sequnlock(&tsk->vtime_seqlock);
678}
679
680void vtime_user_enter(struct task_struct *tsk)
681{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100682 write_seqlock(&tsk->vtime_seqlock);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100683 __vtime_account_system(tsk);
Frederic Weisbeckeraf2350b2013-07-15 16:35:55 +0200684 tsk->vtime_snap_whence = VTIME_USER;
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100685 write_sequnlock(&tsk->vtime_seqlock);
686}
687
688void vtime_guest_enter(struct task_struct *tsk)
689{
Frederic Weisbecker5b206d42013-07-12 19:05:14 +0200690 /*
691 * The flags must be updated under the lock with
692 * the vtime_snap flush and update.
693 * That enforces a right ordering and update sequence
694 * synchronization against the reader (task_gtime())
695 * that can thus safely catch up with a tickless delta.
696 */
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100697 write_seqlock(&tsk->vtime_seqlock);
698 __vtime_account_system(tsk);
699 current->flags |= PF_VCPU;
700 write_sequnlock(&tsk->vtime_seqlock);
701}
Frederic Weisbecker48d6a812013-07-10 02:44:35 +0200702EXPORT_SYMBOL_GPL(vtime_guest_enter);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100703
704void vtime_guest_exit(struct task_struct *tsk)
705{
706 write_seqlock(&tsk->vtime_seqlock);
707 __vtime_account_system(tsk);
708 current->flags &= ~PF_VCPU;
709 write_sequnlock(&tsk->vtime_seqlock);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200710}
Frederic Weisbecker48d6a812013-07-10 02:44:35 +0200711EXPORT_SYMBOL_GPL(vtime_guest_exit);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200712
713void vtime_account_idle(struct task_struct *tsk)
714{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100715 cputime_t delta_cpu = get_vtime_delta(tsk);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200716
717 account_idle_time(delta_cpu);
718}
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200719
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100720void arch_vtime_task_switch(struct task_struct *prev)
721{
722 write_seqlock(&prev->vtime_seqlock);
723 prev->vtime_snap_whence = VTIME_SLEEPING;
724 write_sequnlock(&prev->vtime_seqlock);
725
726 write_seqlock(&current->vtime_seqlock);
727 current->vtime_snap_whence = VTIME_SYS;
Frederic Weisbecker45eacc62013-05-15 22:16:32 +0200728 current->vtime_snap = sched_clock_cpu(smp_processor_id());
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100729 write_sequnlock(&current->vtime_seqlock);
730}
731
Frederic Weisbecker45eacc62013-05-15 22:16:32 +0200732void vtime_init_idle(struct task_struct *t, int cpu)
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100733{
734 unsigned long flags;
735
736 write_seqlock_irqsave(&t->vtime_seqlock, flags);
737 t->vtime_snap_whence = VTIME_SYS;
Frederic Weisbecker45eacc62013-05-15 22:16:32 +0200738 t->vtime_snap = sched_clock_cpu(cpu);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100739 write_sequnlock_irqrestore(&t->vtime_seqlock, flags);
740}
741
742cputime_t task_gtime(struct task_struct *t)
743{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100744 unsigned int seq;
745 cputime_t gtime;
746
747 do {
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100748 seq = read_seqbegin(&t->vtime_seqlock);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100749
750 gtime = t->gtime;
751 if (t->flags & PF_VCPU)
752 gtime += vtime_delta(t);
753
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100754 } while (read_seqretry(&t->vtime_seqlock, seq));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100755
756 return gtime;
757}
758
759/*
760 * Fetch cputime raw values from fields of task_struct and
761 * add up the pending nohz execution time since the last
762 * cputime snapshot.
763 */
764static void
765fetch_task_cputime(struct task_struct *t,
766 cputime_t *u_dst, cputime_t *s_dst,
767 cputime_t *u_src, cputime_t *s_src,
768 cputime_t *udelta, cputime_t *sdelta)
769{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100770 unsigned int seq;
771 unsigned long long delta;
772
773 do {
774 *udelta = 0;
775 *sdelta = 0;
776
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100777 seq = read_seqbegin(&t->vtime_seqlock);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100778
779 if (u_dst)
780 *u_dst = *u_src;
781 if (s_dst)
782 *s_dst = *s_src;
783
784 /* Task is sleeping, nothing to add */
785 if (t->vtime_snap_whence == VTIME_SLEEPING ||
786 is_idle_task(t))
787 continue;
788
789 delta = vtime_delta(t);
790
791 /*
792 * Task runs either in user or kernel space, add pending nohz time to
793 * the right place.
794 */
795 if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) {
796 *udelta = delta;
797 } else {
798 if (t->vtime_snap_whence == VTIME_SYS)
799 *sdelta = delta;
800 }
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100801 } while (read_seqretry(&t->vtime_seqlock, seq));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100802}
803
804
805void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime)
806{
807 cputime_t udelta, sdelta;
808
809 fetch_task_cputime(t, utime, stime, &t->utime,
810 &t->stime, &udelta, &sdelta);
811 if (utime)
812 *utime += udelta;
813 if (stime)
814 *stime += sdelta;
815}
816
817void task_cputime_scaled(struct task_struct *t,
818 cputime_t *utimescaled, cputime_t *stimescaled)
819{
820 cputime_t udelta, sdelta;
821
822 fetch_task_cputime(t, utimescaled, stimescaled,
823 &t->utimescaled, &t->stimescaled, &udelta, &sdelta);
824 if (utimescaled)
825 *utimescaled += cputime_to_scaled(udelta);
826 if (stimescaled)
827 *stimescaled += cputime_to_scaled(sdelta);
828}
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200829#endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */