Commit | Line | Data |
---|---|---|
73fbec60 FW |
1 | #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> | |
abf917cd | 6 | #include <linux/context_tracking.h> |
73fbec60 FW |
7 | #include "sched.h" |
8 | ||
9 | ||
10 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
11 | ||
12 | /* | |
13 | * There are no locks covering percpu hardirq/softirq time. | |
bf9fae9f | 14 | * They are only modified in vtime_account, on corresponding CPU |
73fbec60 FW |
15 | * 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 | |
bf9fae9f | 18 | * race with irq/vtime_account on this CPU. We would either get old |
73fbec60 FW |
19 | * 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 | */ | |
23 | DEFINE_PER_CPU(u64, cpu_hardirq_time); | |
24 | DEFINE_PER_CPU(u64, cpu_softirq_time); | |
25 | ||
26 | static DEFINE_PER_CPU(u64, irq_start_time); | |
27 | static int sched_clock_irqtime; | |
28 | ||
29 | void enable_sched_clock_irqtime(void) | |
30 | { | |
31 | sched_clock_irqtime = 1; | |
32 | } | |
33 | ||
34 | void disable_sched_clock_irqtime(void) | |
35 | { | |
36 | sched_clock_irqtime = 0; | |
37 | } | |
38 | ||
39 | #ifndef CONFIG_64BIT | |
40 | DEFINE_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 | */ | |
3e1df4f5 | 47 | void irqtime_account_irq(struct task_struct *curr) |
73fbec60 FW |
48 | { |
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 | } | |
3e1df4f5 | 77 | EXPORT_SYMBOL_GPL(irqtime_account_irq); |
73fbec60 FW |
78 | |
79 | static 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 | ||
94 | static 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 | ||
115 | static inline void task_group_account_field(struct task_struct *p, int index, | |
116 | u64 tmp) | |
117 | { | |
118 | #ifdef CONFIG_CGROUP_CPUACCT | |
119 | struct kernel_cpustat *kcpustat; | |
120 | struct cpuacct *ca; | |
121 | #endif | |
122 | /* | |
123 | * Since all updates are sure to touch the root cgroup, we | |
124 | * get ourselves ahead and touch it first. If the root cgroup | |
125 | * is the only cgroup, then nothing else should be necessary. | |
126 | * | |
127 | */ | |
128 | __get_cpu_var(kernel_cpustat).cpustat[index] += tmp; | |
129 | ||
130 | #ifdef CONFIG_CGROUP_CPUACCT | |
131 | if (unlikely(!cpuacct_subsys.active)) | |
132 | return; | |
133 | ||
134 | rcu_read_lock(); | |
135 | ca = task_ca(p); | |
136 | while (ca && (ca != &root_cpuacct)) { | |
137 | kcpustat = this_cpu_ptr(ca->cpustat); | |
138 | kcpustat->cpustat[index] += tmp; | |
139 | ca = parent_ca(ca); | |
140 | } | |
141 | rcu_read_unlock(); | |
142 | #endif | |
143 | } | |
144 | ||
145 | /* | |
146 | * Account user cpu time to a process. | |
147 | * @p: the process that the cpu time gets accounted to | |
148 | * @cputime: the cpu time spent in user space since the last update | |
149 | * @cputime_scaled: cputime scaled by cpu frequency | |
150 | */ | |
151 | void account_user_time(struct task_struct *p, cputime_t cputime, | |
152 | cputime_t cputime_scaled) | |
153 | { | |
154 | int index; | |
155 | ||
156 | /* Add user time to process. */ | |
157 | p->utime += cputime; | |
158 | p->utimescaled += cputime_scaled; | |
159 | account_group_user_time(p, cputime); | |
160 | ||
161 | index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; | |
162 | ||
163 | /* Add user time to cpustat. */ | |
164 | task_group_account_field(p, index, (__force u64) cputime); | |
165 | ||
166 | /* Account for user time used */ | |
6fac4829 | 167 | acct_account_cputime(p); |
73fbec60 FW |
168 | } |
169 | ||
170 | /* | |
171 | * Account guest cpu time to a process. | |
172 | * @p: the process that the cpu time gets accounted to | |
173 | * @cputime: the cpu time spent in virtual machine since the last update | |
174 | * @cputime_scaled: cputime scaled by cpu frequency | |
175 | */ | |
176 | static void account_guest_time(struct task_struct *p, cputime_t cputime, | |
177 | cputime_t cputime_scaled) | |
178 | { | |
179 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
180 | ||
181 | /* Add guest time to process. */ | |
182 | p->utime += cputime; | |
183 | p->utimescaled += cputime_scaled; | |
184 | account_group_user_time(p, cputime); | |
185 | p->gtime += cputime; | |
186 | ||
187 | /* Add guest time to cpustat. */ | |
188 | if (TASK_NICE(p) > 0) { | |
189 | cpustat[CPUTIME_NICE] += (__force u64) cputime; | |
190 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
191 | } else { | |
192 | cpustat[CPUTIME_USER] += (__force u64) cputime; | |
193 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
194 | } | |
195 | } | |
196 | ||
197 | /* | |
198 | * Account system cpu time to a process and desired cpustat field | |
199 | * @p: the process that the cpu time gets accounted to | |
200 | * @cputime: the cpu time spent in kernel space since the last update | |
201 | * @cputime_scaled: cputime scaled by cpu frequency | |
202 | * @target_cputime64: pointer to cpustat field that has to be updated | |
203 | */ | |
204 | static inline | |
205 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
206 | cputime_t cputime_scaled, int index) | |
207 | { | |
208 | /* Add system time to process. */ | |
209 | p->stime += cputime; | |
210 | p->stimescaled += cputime_scaled; | |
211 | account_group_system_time(p, cputime); | |
212 | ||
213 | /* Add system time to cpustat. */ | |
214 | task_group_account_field(p, index, (__force u64) cputime); | |
215 | ||
216 | /* Account for system time used */ | |
6fac4829 | 217 | acct_account_cputime(p); |
73fbec60 FW |
218 | } |
219 | ||
220 | /* | |
221 | * Account system cpu time to a process. | |
222 | * @p: the process that the cpu time gets accounted to | |
223 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
224 | * @cputime: the cpu time spent in kernel space since the last update | |
225 | * @cputime_scaled: cputime scaled by cpu frequency | |
226 | */ | |
227 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
228 | cputime_t cputime, cputime_t cputime_scaled) | |
229 | { | |
230 | int index; | |
231 | ||
232 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { | |
233 | account_guest_time(p, cputime, cputime_scaled); | |
234 | return; | |
235 | } | |
236 | ||
237 | if (hardirq_count() - hardirq_offset) | |
238 | index = CPUTIME_IRQ; | |
239 | else if (in_serving_softirq()) | |
240 | index = CPUTIME_SOFTIRQ; | |
241 | else | |
242 | index = CPUTIME_SYSTEM; | |
243 | ||
244 | __account_system_time(p, cputime, cputime_scaled, index); | |
245 | } | |
246 | ||
247 | /* | |
248 | * Account for involuntary wait time. | |
249 | * @cputime: the cpu time spent in involuntary wait | |
250 | */ | |
251 | void account_steal_time(cputime_t cputime) | |
252 | { | |
253 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
254 | ||
255 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; | |
256 | } | |
257 | ||
258 | /* | |
259 | * Account for idle time. | |
260 | * @cputime: the cpu time spent in idle wait | |
261 | */ | |
262 | void account_idle_time(cputime_t cputime) | |
263 | { | |
264 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
265 | struct rq *rq = this_rq(); | |
266 | ||
267 | if (atomic_read(&rq->nr_iowait) > 0) | |
268 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; | |
269 | else | |
270 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; | |
271 | } | |
272 | ||
273 | static __always_inline bool steal_account_process_tick(void) | |
274 | { | |
275 | #ifdef CONFIG_PARAVIRT | |
276 | if (static_key_false(¶virt_steal_enabled)) { | |
277 | u64 steal, st = 0; | |
278 | ||
279 | steal = paravirt_steal_clock(smp_processor_id()); | |
280 | steal -= this_rq()->prev_steal_time; | |
281 | ||
282 | st = steal_ticks(steal); | |
283 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
284 | ||
285 | account_steal_time(st); | |
286 | return st; | |
287 | } | |
288 | #endif | |
289 | return false; | |
290 | } | |
291 | ||
a634f933 FW |
292 | /* |
293 | * Accumulate raw cputime values of dead tasks (sig->[us]time) and live | |
294 | * tasks (sum on group iteration) belonging to @tsk's group. | |
295 | */ | |
296 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) | |
297 | { | |
298 | struct signal_struct *sig = tsk->signal; | |
6fac4829 | 299 | cputime_t utime, stime; |
a634f933 FW |
300 | struct task_struct *t; |
301 | ||
302 | times->utime = sig->utime; | |
303 | times->stime = sig->stime; | |
304 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
305 | ||
306 | rcu_read_lock(); | |
307 | /* make sure we can trust tsk->thread_group list */ | |
308 | if (!likely(pid_alive(tsk))) | |
309 | goto out; | |
310 | ||
311 | t = tsk; | |
312 | do { | |
6fac4829 FW |
313 | task_cputime(tsk, &utime, &stime); |
314 | times->utime += utime; | |
315 | times->stime += stime; | |
a634f933 FW |
316 | times->sum_exec_runtime += task_sched_runtime(t); |
317 | } while_each_thread(tsk, t); | |
318 | out: | |
319 | rcu_read_unlock(); | |
320 | } | |
321 | ||
73fbec60 FW |
322 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
323 | /* | |
324 | * Account a tick to a process and cpustat | |
325 | * @p: the process that the cpu time gets accounted to | |
326 | * @user_tick: is the tick from userspace | |
327 | * @rq: the pointer to rq | |
328 | * | |
329 | * Tick demultiplexing follows the order | |
330 | * - pending hardirq update | |
331 | * - pending softirq update | |
332 | * - user_time | |
333 | * - idle_time | |
334 | * - system time | |
335 | * - check for guest_time | |
336 | * - else account as system_time | |
337 | * | |
338 | * Check for hardirq is done both for system and user time as there is | |
339 | * no timer going off while we are on hardirq and hence we may never get an | |
340 | * opportunity to update it solely in system time. | |
341 | * p->stime and friends are only updated on system time and not on irq | |
342 | * softirq as those do not count in task exec_runtime any more. | |
343 | */ | |
344 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
345 | struct rq *rq) | |
346 | { | |
347 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
348 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
349 | ||
350 | if (steal_account_process_tick()) | |
351 | return; | |
352 | ||
353 | if (irqtime_account_hi_update()) { | |
354 | cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy; | |
355 | } else if (irqtime_account_si_update()) { | |
356 | cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy; | |
357 | } else if (this_cpu_ksoftirqd() == p) { | |
358 | /* | |
359 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
360 | * So, we have to handle it separately here. | |
361 | * Also, p->stime needs to be updated for ksoftirqd. | |
362 | */ | |
363 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
364 | CPUTIME_SOFTIRQ); | |
365 | } else if (user_tick) { | |
366 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
367 | } else if (p == rq->idle) { | |
368 | account_idle_time(cputime_one_jiffy); | |
369 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
370 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
371 | } else { | |
372 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
373 | CPUTIME_SYSTEM); | |
374 | } | |
375 | } | |
376 | ||
377 | static void irqtime_account_idle_ticks(int ticks) | |
378 | { | |
379 | int i; | |
380 | struct rq *rq = this_rq(); | |
381 | ||
382 | for (i = 0; i < ticks; i++) | |
383 | irqtime_account_process_tick(current, 0, rq); | |
384 | } | |
385 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
3f4724ea FW |
386 | static inline void irqtime_account_idle_ticks(int ticks) {} |
387 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
73fbec60 FW |
388 | struct rq *rq) {} |
389 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
390 | ||
3f4724ea | 391 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
73fbec60 FW |
392 | /* |
393 | * Account a single tick of cpu time. | |
394 | * @p: the process that the cpu time gets accounted to | |
395 | * @user_tick: indicates if the tick is a user or a system tick | |
396 | */ | |
397 | void account_process_tick(struct task_struct *p, int user_tick) | |
398 | { | |
399 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
400 | struct rq *rq = this_rq(); | |
401 | ||
3f4724ea FW |
402 | if (vtime_accounting_enabled()) |
403 | return; | |
404 | ||
73fbec60 FW |
405 | if (sched_clock_irqtime) { |
406 | irqtime_account_process_tick(p, user_tick, rq); | |
407 | return; | |
408 | } | |
409 | ||
410 | if (steal_account_process_tick()) | |
411 | return; | |
412 | ||
413 | if (user_tick) | |
414 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
415 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) | |
416 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, | |
417 | one_jiffy_scaled); | |
418 | else | |
419 | account_idle_time(cputime_one_jiffy); | |
420 | } | |
421 | ||
422 | /* | |
423 | * Account multiple ticks of steal time. | |
424 | * @p: the process from which the cpu time has been stolen | |
425 | * @ticks: number of stolen ticks | |
426 | */ | |
427 | void account_steal_ticks(unsigned long ticks) | |
428 | { | |
429 | account_steal_time(jiffies_to_cputime(ticks)); | |
430 | } | |
431 | ||
432 | /* | |
433 | * Account multiple ticks of idle time. | |
434 | * @ticks: number of stolen ticks | |
435 | */ | |
436 | void account_idle_ticks(unsigned long ticks) | |
437 | { | |
438 | ||
439 | if (sched_clock_irqtime) { | |
440 | irqtime_account_idle_ticks(ticks); | |
441 | return; | |
442 | } | |
443 | ||
444 | account_idle_time(jiffies_to_cputime(ticks)); | |
445 | } | |
3f4724ea | 446 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ |
73fbec60 FW |
447 | |
448 | /* | |
449 | * Use precise platform statistics if available: | |
450 | */ | |
451 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
e80d0a1a | 452 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 FW |
453 | { |
454 | *ut = p->utime; | |
455 | *st = p->stime; | |
456 | } | |
457 | ||
e80d0a1a | 458 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 FW |
459 | { |
460 | struct task_cputime cputime; | |
461 | ||
462 | thread_group_cputime(p, &cputime); | |
463 | ||
464 | *ut = cputime.utime; | |
465 | *st = cputime.stime; | |
466 | } | |
a7e1a9e3 | 467 | |
e3942ba0 FW |
468 | #ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
469 | void vtime_task_switch(struct task_struct *prev) | |
470 | { | |
3f4724ea FW |
471 | if (!vtime_accounting_enabled()) |
472 | return; | |
473 | ||
e3942ba0 FW |
474 | if (is_idle_task(prev)) |
475 | vtime_account_idle(prev); | |
476 | else | |
477 | vtime_account_system(prev); | |
478 | ||
abf917cd | 479 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
e3942ba0 | 480 | vtime_account_user(prev); |
abf917cd | 481 | #endif |
e3942ba0 FW |
482 | arch_vtime_task_switch(prev); |
483 | } | |
484 | #endif | |
11113334 | 485 | |
a7e1a9e3 FW |
486 | /* |
487 | * Archs that account the whole time spent in the idle task | |
488 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 489 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
490 | * have other meaning of the idle time (s390 only includes the |
491 | * time spent by the CPU when it's in low power mode) must override | |
492 | * vtime_account(). | |
493 | */ | |
494 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
6a61671b | 495 | void vtime_account_irq_enter(struct task_struct *tsk) |
a7e1a9e3 | 496 | { |
3f4724ea FW |
497 | if (!vtime_accounting_enabled()) |
498 | return; | |
499 | ||
abf917cd FW |
500 | if (!in_interrupt()) { |
501 | /* | |
502 | * If we interrupted user, context_tracking_in_user() | |
503 | * is 1 because the context tracking don't hook | |
504 | * on irq entry/exit. This way we know if | |
505 | * we need to flush user time on kernel entry. | |
506 | */ | |
507 | if (context_tracking_in_user()) { | |
508 | vtime_account_user(tsk); | |
509 | return; | |
510 | } | |
511 | ||
512 | if (is_idle_task(tsk)) { | |
513 | vtime_account_idle(tsk); | |
514 | return; | |
515 | } | |
516 | } | |
517 | vtime_account_system(tsk); | |
a7e1a9e3 | 518 | } |
6a61671b | 519 | EXPORT_SYMBOL_GPL(vtime_account_irq_enter); |
a7e1a9e3 FW |
520 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ |
521 | ||
3f4724ea | 522 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING */ |
73fbec60 | 523 | |
73fbec60 FW |
524 | static cputime_t scale_utime(cputime_t utime, cputime_t rtime, cputime_t total) |
525 | { | |
526 | u64 temp = (__force u64) rtime; | |
527 | ||
528 | temp *= (__force u64) utime; | |
529 | ||
530 | if (sizeof(cputime_t) == 4) | |
531 | temp = div_u64(temp, (__force u32) total); | |
532 | else | |
533 | temp = div64_u64(temp, (__force u64) total); | |
534 | ||
535 | return (__force cputime_t) temp; | |
536 | } | |
537 | ||
fa092057 FW |
538 | /* |
539 | * Adjust tick based cputime random precision against scheduler | |
540 | * runtime accounting. | |
541 | */ | |
d37f761d FW |
542 | static void cputime_adjust(struct task_cputime *curr, |
543 | struct cputime *prev, | |
544 | cputime_t *ut, cputime_t *st) | |
73fbec60 | 545 | { |
d37f761d | 546 | cputime_t rtime, utime, total; |
73fbec60 | 547 | |
d37f761d FW |
548 | utime = curr->utime; |
549 | total = utime + curr->stime; | |
fa092057 | 550 | |
73fbec60 | 551 | /* |
fa092057 FW |
552 | * Tick based cputime accounting depend on random scheduling |
553 | * timeslices of a task to be interrupted or not by the timer. | |
554 | * Depending on these circumstances, the number of these interrupts | |
555 | * may be over or under-optimistic, matching the real user and system | |
556 | * cputime with a variable precision. | |
557 | * | |
558 | * Fix this by scaling these tick based values against the total | |
559 | * runtime accounted by the CFS scheduler. | |
73fbec60 | 560 | */ |
d37f761d | 561 | rtime = nsecs_to_cputime(curr->sum_exec_runtime); |
73fbec60 FW |
562 | |
563 | if (total) | |
564 | utime = scale_utime(utime, rtime, total); | |
565 | else | |
566 | utime = rtime; | |
567 | ||
568 | /* | |
fa092057 FW |
569 | * If the tick based count grows faster than the scheduler one, |
570 | * the result of the scaling may go backward. | |
571 | * Let's enforce monotonicity. | |
73fbec60 | 572 | */ |
d37f761d FW |
573 | prev->utime = max(prev->utime, utime); |
574 | prev->stime = max(prev->stime, rtime - prev->utime); | |
575 | ||
576 | *ut = prev->utime; | |
577 | *st = prev->stime; | |
578 | } | |
73fbec60 | 579 | |
d37f761d FW |
580 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
581 | { | |
582 | struct task_cputime cputime = { | |
d37f761d FW |
583 | .sum_exec_runtime = p->se.sum_exec_runtime, |
584 | }; | |
585 | ||
6fac4829 | 586 | task_cputime(p, &cputime.utime, &cputime.stime); |
d37f761d | 587 | cputime_adjust(&cputime, &p->prev_cputime, ut, st); |
73fbec60 FW |
588 | } |
589 | ||
590 | /* | |
591 | * Must be called with siglock held. | |
592 | */ | |
e80d0a1a | 593 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 | 594 | { |
73fbec60 | 595 | struct task_cputime cputime; |
73fbec60 FW |
596 | |
597 | thread_group_cputime(p, &cputime); | |
d37f761d | 598 | cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); |
73fbec60 | 599 | } |
3f4724ea | 600 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING */ |
abf917cd FW |
601 | |
602 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN | |
6a61671b FW |
603 | static unsigned long long vtime_delta(struct task_struct *tsk) |
604 | { | |
605 | unsigned long long clock; | |
606 | ||
607 | clock = sched_clock(); | |
608 | if (clock < tsk->vtime_snap) | |
609 | return 0; | |
abf917cd | 610 | |
6a61671b FW |
611 | return clock - tsk->vtime_snap; |
612 | } | |
613 | ||
614 | static cputime_t get_vtime_delta(struct task_struct *tsk) | |
abf917cd | 615 | { |
6a61671b | 616 | unsigned long long delta = vtime_delta(tsk); |
abf917cd | 617 | |
6a61671b FW |
618 | WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_SLEEPING); |
619 | tsk->vtime_snap += delta; | |
abf917cd FW |
620 | |
621 | /* CHECKME: always safe to convert nsecs to cputime? */ | |
622 | return nsecs_to_cputime(delta); | |
623 | } | |
624 | ||
6a61671b FW |
625 | static void __vtime_account_system(struct task_struct *tsk) |
626 | { | |
627 | cputime_t delta_cpu = get_vtime_delta(tsk); | |
628 | ||
629 | account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu)); | |
630 | } | |
631 | ||
abf917cd FW |
632 | void vtime_account_system(struct task_struct *tsk) |
633 | { | |
6a61671b FW |
634 | if (!vtime_accounting_enabled()) |
635 | return; | |
636 | ||
637 | write_seqlock(&tsk->vtime_seqlock); | |
638 | __vtime_account_system(tsk); | |
639 | write_sequnlock(&tsk->vtime_seqlock); | |
640 | } | |
3f4724ea | 641 | |
6a61671b FW |
642 | void vtime_account_irq_exit(struct task_struct *tsk) |
643 | { | |
3f4724ea FW |
644 | if (!vtime_accounting_enabled()) |
645 | return; | |
abf917cd | 646 | |
6a61671b FW |
647 | write_seqlock(&tsk->vtime_seqlock); |
648 | if (context_tracking_in_user()) | |
649 | tsk->vtime_snap_whence = VTIME_USER; | |
650 | __vtime_account_system(tsk); | |
651 | write_sequnlock(&tsk->vtime_seqlock); | |
abf917cd FW |
652 | } |
653 | ||
654 | void vtime_account_user(struct task_struct *tsk) | |
655 | { | |
3f4724ea FW |
656 | cputime_t delta_cpu; |
657 | ||
658 | if (!vtime_accounting_enabled()) | |
659 | return; | |
660 | ||
6a61671b | 661 | delta_cpu = get_vtime_delta(tsk); |
abf917cd | 662 | |
6a61671b FW |
663 | write_seqlock(&tsk->vtime_seqlock); |
664 | tsk->vtime_snap_whence = VTIME_SYS; | |
abf917cd | 665 | account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu)); |
6a61671b FW |
666 | write_sequnlock(&tsk->vtime_seqlock); |
667 | } | |
668 | ||
669 | void vtime_user_enter(struct task_struct *tsk) | |
670 | { | |
671 | if (!vtime_accounting_enabled()) | |
672 | return; | |
673 | ||
674 | write_seqlock(&tsk->vtime_seqlock); | |
675 | tsk->vtime_snap_whence = VTIME_USER; | |
676 | __vtime_account_system(tsk); | |
677 | write_sequnlock(&tsk->vtime_seqlock); | |
678 | } | |
679 | ||
680 | void vtime_guest_enter(struct task_struct *tsk) | |
681 | { | |
682 | write_seqlock(&tsk->vtime_seqlock); | |
683 | __vtime_account_system(tsk); | |
684 | current->flags |= PF_VCPU; | |
685 | write_sequnlock(&tsk->vtime_seqlock); | |
686 | } | |
687 | ||
688 | void vtime_guest_exit(struct task_struct *tsk) | |
689 | { | |
690 | write_seqlock(&tsk->vtime_seqlock); | |
691 | __vtime_account_system(tsk); | |
692 | current->flags &= ~PF_VCPU; | |
693 | write_sequnlock(&tsk->vtime_seqlock); | |
abf917cd FW |
694 | } |
695 | ||
696 | void vtime_account_idle(struct task_struct *tsk) | |
697 | { | |
6a61671b | 698 | cputime_t delta_cpu = get_vtime_delta(tsk); |
abf917cd FW |
699 | |
700 | account_idle_time(delta_cpu); | |
701 | } | |
3f4724ea FW |
702 | |
703 | bool vtime_accounting_enabled(void) | |
704 | { | |
705 | return context_tracking_active(); | |
706 | } | |
6a61671b FW |
707 | |
708 | void arch_vtime_task_switch(struct task_struct *prev) | |
709 | { | |
710 | write_seqlock(&prev->vtime_seqlock); | |
711 | prev->vtime_snap_whence = VTIME_SLEEPING; | |
712 | write_sequnlock(&prev->vtime_seqlock); | |
713 | ||
714 | write_seqlock(¤t->vtime_seqlock); | |
715 | current->vtime_snap_whence = VTIME_SYS; | |
716 | current->vtime_snap = sched_clock(); | |
717 | write_sequnlock(¤t->vtime_seqlock); | |
718 | } | |
719 | ||
720 | void vtime_init_idle(struct task_struct *t) | |
721 | { | |
722 | unsigned long flags; | |
723 | ||
724 | write_seqlock_irqsave(&t->vtime_seqlock, flags); | |
725 | t->vtime_snap_whence = VTIME_SYS; | |
726 | t->vtime_snap = sched_clock(); | |
727 | write_sequnlock_irqrestore(&t->vtime_seqlock, flags); | |
728 | } | |
729 | ||
730 | cputime_t task_gtime(struct task_struct *t) | |
731 | { | |
732 | unsigned long flags; | |
733 | unsigned int seq; | |
734 | cputime_t gtime; | |
735 | ||
736 | do { | |
737 | seq = read_seqbegin_irqsave(&t->vtime_seqlock, flags); | |
738 | ||
739 | gtime = t->gtime; | |
740 | if (t->flags & PF_VCPU) | |
741 | gtime += vtime_delta(t); | |
742 | ||
743 | } while (read_seqretry_irqrestore(&t->vtime_seqlock, seq, flags)); | |
744 | ||
745 | return gtime; | |
746 | } | |
747 | ||
748 | /* | |
749 | * Fetch cputime raw values from fields of task_struct and | |
750 | * add up the pending nohz execution time since the last | |
751 | * cputime snapshot. | |
752 | */ | |
753 | static void | |
754 | fetch_task_cputime(struct task_struct *t, | |
755 | cputime_t *u_dst, cputime_t *s_dst, | |
756 | cputime_t *u_src, cputime_t *s_src, | |
757 | cputime_t *udelta, cputime_t *sdelta) | |
758 | { | |
759 | unsigned long flags; | |
760 | unsigned int seq; | |
761 | unsigned long long delta; | |
762 | ||
763 | do { | |
764 | *udelta = 0; | |
765 | *sdelta = 0; | |
766 | ||
767 | seq = read_seqbegin_irqsave(&t->vtime_seqlock, flags); | |
768 | ||
769 | if (u_dst) | |
770 | *u_dst = *u_src; | |
771 | if (s_dst) | |
772 | *s_dst = *s_src; | |
773 | ||
774 | /* Task is sleeping, nothing to add */ | |
775 | if (t->vtime_snap_whence == VTIME_SLEEPING || | |
776 | is_idle_task(t)) | |
777 | continue; | |
778 | ||
779 | delta = vtime_delta(t); | |
780 | ||
781 | /* | |
782 | * Task runs either in user or kernel space, add pending nohz time to | |
783 | * the right place. | |
784 | */ | |
785 | if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) { | |
786 | *udelta = delta; | |
787 | } else { | |
788 | if (t->vtime_snap_whence == VTIME_SYS) | |
789 | *sdelta = delta; | |
790 | } | |
791 | } while (read_seqretry_irqrestore(&t->vtime_seqlock, seq, flags)); | |
792 | } | |
793 | ||
794 | ||
795 | void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime) | |
796 | { | |
797 | cputime_t udelta, sdelta; | |
798 | ||
799 | fetch_task_cputime(t, utime, stime, &t->utime, | |
800 | &t->stime, &udelta, &sdelta); | |
801 | if (utime) | |
802 | *utime += udelta; | |
803 | if (stime) | |
804 | *stime += sdelta; | |
805 | } | |
806 | ||
807 | void task_cputime_scaled(struct task_struct *t, | |
808 | cputime_t *utimescaled, cputime_t *stimescaled) | |
809 | { | |
810 | cputime_t udelta, sdelta; | |
811 | ||
812 | fetch_task_cputime(t, utimescaled, stimescaled, | |
813 | &t->utimescaled, &t->stimescaled, &udelta, &sdelta); | |
814 | if (utimescaled) | |
815 | *utimescaled += cputime_to_scaled(udelta); | |
816 | if (stimescaled) | |
817 | *stimescaled += cputime_to_scaled(sdelta); | |
818 | } | |
abf917cd | 819 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |