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 | 7 | #include "sched.h" |
1fe7c4ef SS |
8 | #ifdef CONFIG_PARAVIRT |
9 | #include <asm/paravirt.h> | |
10 | #endif | |
73fbec60 FW |
11 | |
12 | ||
13 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
14 | ||
15 | /* | |
16 | * There are no locks covering percpu hardirq/softirq time. | |
bf9fae9f | 17 | * They are only modified in vtime_account, on corresponding CPU |
73fbec60 FW |
18 | * with interrupts disabled. So, writes are safe. |
19 | * They are read and saved off onto struct rq in update_rq_clock(). | |
20 | * This may result in other CPU reading this CPU's irq time and can | |
bf9fae9f | 21 | * race with irq/vtime_account on this CPU. We would either get old |
73fbec60 FW |
22 | * or new value with a side effect of accounting a slice of irq time to wrong |
23 | * task when irq is in progress while we read rq->clock. That is a worthy | |
24 | * compromise in place of having locks on each irq in account_system_time. | |
25 | */ | |
26 | DEFINE_PER_CPU(u64, cpu_hardirq_time); | |
27 | DEFINE_PER_CPU(u64, cpu_softirq_time); | |
28 | ||
29 | static DEFINE_PER_CPU(u64, irq_start_time); | |
30 | static int sched_clock_irqtime; | |
31 | ||
32 | void enable_sched_clock_irqtime(void) | |
33 | { | |
34 | sched_clock_irqtime = 1; | |
35 | } | |
36 | ||
37 | void disable_sched_clock_irqtime(void) | |
38 | { | |
39 | sched_clock_irqtime = 0; | |
40 | } | |
41 | ||
42 | #ifndef CONFIG_64BIT | |
43 | DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
44 | #endif /* CONFIG_64BIT */ | |
45 | ||
46 | /* | |
47 | * Called before incrementing preempt_count on {soft,}irq_enter | |
48 | * and before decrementing preempt_count on {soft,}irq_exit. | |
49 | */ | |
3e1df4f5 | 50 | void irqtime_account_irq(struct task_struct *curr) |
73fbec60 | 51 | { |
73fbec60 FW |
52 | s64 delta; |
53 | int cpu; | |
54 | ||
55 | if (!sched_clock_irqtime) | |
56 | return; | |
57 | ||
73fbec60 FW |
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(); | |
73fbec60 | 75 | } |
3e1df4f5 | 76 | EXPORT_SYMBOL_GPL(irqtime_account_irq); |
73fbec60 | 77 | |
57430218 | 78 | static cputime_t irqtime_account_hi_update(cputime_t maxtime) |
73fbec60 FW |
79 | { |
80 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
81 | unsigned long flags; | |
57430218 | 82 | cputime_t irq_cputime; |
73fbec60 FW |
83 | |
84 | local_irq_save(flags); | |
57430218 RR |
85 | irq_cputime = nsecs_to_cputime64(this_cpu_read(cpu_hardirq_time)) - |
86 | cpustat[CPUTIME_IRQ]; | |
87 | irq_cputime = min(irq_cputime, maxtime); | |
88 | cpustat[CPUTIME_IRQ] += irq_cputime; | |
73fbec60 | 89 | local_irq_restore(flags); |
57430218 | 90 | return irq_cputime; |
73fbec60 FW |
91 | } |
92 | ||
57430218 | 93 | static cputime_t irqtime_account_si_update(cputime_t maxtime) |
73fbec60 FW |
94 | { |
95 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
96 | unsigned long flags; | |
57430218 | 97 | cputime_t softirq_cputime; |
73fbec60 FW |
98 | |
99 | local_irq_save(flags); | |
57430218 RR |
100 | softirq_cputime = nsecs_to_cputime64(this_cpu_read(cpu_softirq_time)) - |
101 | cpustat[CPUTIME_SOFTIRQ]; | |
102 | softirq_cputime = min(softirq_cputime, maxtime); | |
103 | cpustat[CPUTIME_SOFTIRQ] += softirq_cputime; | |
73fbec60 | 104 | local_irq_restore(flags); |
57430218 | 105 | return softirq_cputime; |
73fbec60 FW |
106 | } |
107 | ||
108 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
109 | ||
110 | #define sched_clock_irqtime (0) | |
111 | ||
57430218 RR |
112 | static cputime_t irqtime_account_hi_update(cputime_t dummy) |
113 | { | |
114 | return 0; | |
115 | } | |
116 | ||
117 | static cputime_t irqtime_account_si_update(cputime_t dummy) | |
118 | { | |
119 | return 0; | |
120 | } | |
121 | ||
73fbec60 FW |
122 | #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */ |
123 | ||
124 | static inline void task_group_account_field(struct task_struct *p, int index, | |
125 | u64 tmp) | |
126 | { | |
73fbec60 FW |
127 | /* |
128 | * Since all updates are sure to touch the root cgroup, we | |
129 | * get ourselves ahead and touch it first. If the root cgroup | |
130 | * is the only cgroup, then nothing else should be necessary. | |
131 | * | |
132 | */ | |
a4f61cc0 | 133 | __this_cpu_add(kernel_cpustat.cpustat[index], tmp); |
73fbec60 | 134 | |
1966aaf7 | 135 | cpuacct_account_field(p, index, tmp); |
73fbec60 FW |
136 | } |
137 | ||
138 | /* | |
139 | * Account user cpu time to a process. | |
140 | * @p: the process that the cpu time gets accounted to | |
141 | * @cputime: the cpu time spent in user space since the last update | |
142 | * @cputime_scaled: cputime scaled by cpu frequency | |
143 | */ | |
144 | void account_user_time(struct task_struct *p, cputime_t cputime, | |
145 | cputime_t cputime_scaled) | |
146 | { | |
147 | int index; | |
148 | ||
149 | /* Add user time to process. */ | |
150 | p->utime += cputime; | |
151 | p->utimescaled += cputime_scaled; | |
152 | account_group_user_time(p, cputime); | |
153 | ||
d0ea0268 | 154 | index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
73fbec60 FW |
155 | |
156 | /* Add user time to cpustat. */ | |
157 | task_group_account_field(p, index, (__force u64) cputime); | |
158 | ||
159 | /* Account for user time used */ | |
6fac4829 | 160 | acct_account_cputime(p); |
73fbec60 FW |
161 | } |
162 | ||
163 | /* | |
164 | * Account guest cpu time to a process. | |
165 | * @p: the process that the cpu time gets accounted to | |
166 | * @cputime: the cpu time spent in virtual machine since the last update | |
167 | * @cputime_scaled: cputime scaled by cpu frequency | |
168 | */ | |
169 | static void account_guest_time(struct task_struct *p, cputime_t cputime, | |
170 | cputime_t cputime_scaled) | |
171 | { | |
172 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
173 | ||
174 | /* Add guest time to process. */ | |
175 | p->utime += cputime; | |
176 | p->utimescaled += cputime_scaled; | |
177 | account_group_user_time(p, cputime); | |
178 | p->gtime += cputime; | |
179 | ||
180 | /* Add guest time to cpustat. */ | |
d0ea0268 | 181 | if (task_nice(p) > 0) { |
73fbec60 FW |
182 | cpustat[CPUTIME_NICE] += (__force u64) cputime; |
183 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
184 | } else { | |
185 | cpustat[CPUTIME_USER] += (__force u64) cputime; | |
186 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
187 | } | |
188 | } | |
189 | ||
190 | /* | |
191 | * Account system cpu time to a process and desired cpustat field | |
192 | * @p: the process that the cpu time gets accounted to | |
193 | * @cputime: the cpu time spent in kernel space since the last update | |
194 | * @cputime_scaled: cputime scaled by cpu frequency | |
195 | * @target_cputime64: pointer to cpustat field that has to be updated | |
196 | */ | |
197 | static inline | |
198 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
199 | cputime_t cputime_scaled, int index) | |
200 | { | |
201 | /* Add system time to process. */ | |
202 | p->stime += cputime; | |
203 | p->stimescaled += cputime_scaled; | |
204 | account_group_system_time(p, cputime); | |
205 | ||
206 | /* Add system time to cpustat. */ | |
207 | task_group_account_field(p, index, (__force u64) cputime); | |
208 | ||
209 | /* Account for system time used */ | |
6fac4829 | 210 | acct_account_cputime(p); |
73fbec60 FW |
211 | } |
212 | ||
213 | /* | |
214 | * Account system cpu time to a process. | |
215 | * @p: the process that the cpu time gets accounted to | |
216 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
217 | * @cputime: the cpu time spent in kernel space since the last update | |
218 | * @cputime_scaled: cputime scaled by cpu frequency | |
219 | */ | |
220 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
221 | cputime_t cputime, cputime_t cputime_scaled) | |
222 | { | |
223 | int index; | |
224 | ||
225 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { | |
226 | account_guest_time(p, cputime, cputime_scaled); | |
227 | return; | |
228 | } | |
229 | ||
230 | if (hardirq_count() - hardirq_offset) | |
231 | index = CPUTIME_IRQ; | |
232 | else if (in_serving_softirq()) | |
233 | index = CPUTIME_SOFTIRQ; | |
234 | else | |
235 | index = CPUTIME_SYSTEM; | |
236 | ||
237 | __account_system_time(p, cputime, cputime_scaled, index); | |
238 | } | |
239 | ||
240 | /* | |
241 | * Account for involuntary wait time. | |
242 | * @cputime: the cpu time spent in involuntary wait | |
243 | */ | |
244 | void account_steal_time(cputime_t cputime) | |
245 | { | |
246 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
247 | ||
248 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; | |
249 | } | |
250 | ||
251 | /* | |
252 | * Account for idle time. | |
253 | * @cputime: the cpu time spent in idle wait | |
254 | */ | |
255 | void account_idle_time(cputime_t cputime) | |
256 | { | |
257 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
258 | struct rq *rq = this_rq(); | |
259 | ||
260 | if (atomic_read(&rq->nr_iowait) > 0) | |
261 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; | |
262 | else | |
263 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; | |
264 | } | |
265 | ||
03cbc732 WL |
266 | /* |
267 | * When a guest is interrupted for a longer amount of time, missed clock | |
268 | * ticks are not redelivered later. Due to that, this function may on | |
269 | * occasion account more time than the calling functions think elapsed. | |
270 | */ | |
57430218 | 271 | static __always_inline cputime_t steal_account_process_time(cputime_t maxtime) |
73fbec60 FW |
272 | { |
273 | #ifdef CONFIG_PARAVIRT | |
274 | if (static_key_false(¶virt_steal_enabled)) { | |
57430218 | 275 | cputime_t steal_cputime; |
dee08a72 | 276 | u64 steal; |
73fbec60 FW |
277 | |
278 | steal = paravirt_steal_clock(smp_processor_id()); | |
279 | steal -= this_rq()->prev_steal_time; | |
280 | ||
57430218 RR |
281 | steal_cputime = min(nsecs_to_cputime(steal), maxtime); |
282 | account_steal_time(steal_cputime); | |
283 | this_rq()->prev_steal_time += cputime_to_nsecs(steal_cputime); | |
73fbec60 | 284 | |
57430218 | 285 | return steal_cputime; |
73fbec60 FW |
286 | } |
287 | #endif | |
807e5b80 | 288 | return 0; |
73fbec60 FW |
289 | } |
290 | ||
57430218 RR |
291 | /* |
292 | * Account how much elapsed time was spent in steal, irq, or softirq time. | |
293 | */ | |
294 | static inline cputime_t account_other_time(cputime_t max) | |
295 | { | |
296 | cputime_t accounted; | |
297 | ||
298 | accounted = steal_account_process_time(max); | |
299 | ||
300 | if (accounted < max) | |
301 | accounted += irqtime_account_hi_update(max - accounted); | |
302 | ||
303 | if (accounted < max) | |
304 | accounted += irqtime_account_si_update(max - accounted); | |
305 | ||
306 | return accounted; | |
307 | } | |
308 | ||
a634f933 FW |
309 | /* |
310 | * Accumulate raw cputime values of dead tasks (sig->[us]time) and live | |
311 | * tasks (sum on group iteration) belonging to @tsk's group. | |
312 | */ | |
313 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) | |
314 | { | |
315 | struct signal_struct *sig = tsk->signal; | |
6fac4829 | 316 | cputime_t utime, stime; |
a634f933 | 317 | struct task_struct *t; |
e78c3496 | 318 | unsigned int seq, nextseq; |
9c368b5b | 319 | unsigned long flags; |
a634f933 FW |
320 | |
321 | rcu_read_lock(); | |
e78c3496 RR |
322 | /* Attempt a lockless read on the first round. */ |
323 | nextseq = 0; | |
324 | do { | |
325 | seq = nextseq; | |
9c368b5b | 326 | flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq); |
e78c3496 RR |
327 | times->utime = sig->utime; |
328 | times->stime = sig->stime; | |
329 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
330 | ||
331 | for_each_thread(tsk, t) { | |
332 | task_cputime(t, &utime, &stime); | |
333 | times->utime += utime; | |
334 | times->stime += stime; | |
335 | times->sum_exec_runtime += task_sched_runtime(t); | |
336 | } | |
337 | /* If lockless access failed, take the lock. */ | |
338 | nextseq = 1; | |
339 | } while (need_seqretry(&sig->stats_lock, seq)); | |
9c368b5b | 340 | done_seqretry_irqrestore(&sig->stats_lock, seq, flags); |
a634f933 FW |
341 | rcu_read_unlock(); |
342 | } | |
343 | ||
73fbec60 FW |
344 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
345 | /* | |
346 | * Account a tick to a process and cpustat | |
347 | * @p: the process that the cpu time gets accounted to | |
348 | * @user_tick: is the tick from userspace | |
349 | * @rq: the pointer to rq | |
350 | * | |
351 | * Tick demultiplexing follows the order | |
352 | * - pending hardirq update | |
353 | * - pending softirq update | |
354 | * - user_time | |
355 | * - idle_time | |
356 | * - system time | |
357 | * - check for guest_time | |
358 | * - else account as system_time | |
359 | * | |
360 | * Check for hardirq is done both for system and user time as there is | |
361 | * no timer going off while we are on hardirq and hence we may never get an | |
362 | * opportunity to update it solely in system time. | |
363 | * p->stime and friends are only updated on system time and not on irq | |
364 | * softirq as those do not count in task exec_runtime any more. | |
365 | */ | |
366 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 367 | struct rq *rq, int ticks) |
73fbec60 | 368 | { |
57430218 RR |
369 | u64 cputime = (__force u64) cputime_one_jiffy * ticks; |
370 | cputime_t scaled, other; | |
73fbec60 | 371 | |
57430218 RR |
372 | /* |
373 | * When returning from idle, many ticks can get accounted at | |
374 | * once, including some ticks of steal, irq, and softirq time. | |
375 | * Subtract those ticks from the amount of time accounted to | |
376 | * idle, or potentially user or system time. Due to rounding, | |
377 | * other time can exceed ticks occasionally. | |
378 | */ | |
03cbc732 | 379 | other = account_other_time(ULONG_MAX); |
57430218 | 380 | if (other >= cputime) |
73fbec60 | 381 | return; |
57430218 RR |
382 | cputime -= other; |
383 | scaled = cputime_to_scaled(cputime); | |
73fbec60 | 384 | |
57430218 | 385 | if (this_cpu_ksoftirqd() == p) { |
73fbec60 FW |
386 | /* |
387 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
388 | * So, we have to handle it separately here. | |
389 | * Also, p->stime needs to be updated for ksoftirqd. | |
390 | */ | |
2d513868 | 391 | __account_system_time(p, cputime, scaled, CPUTIME_SOFTIRQ); |
73fbec60 | 392 | } else if (user_tick) { |
2d513868 | 393 | account_user_time(p, cputime, scaled); |
73fbec60 | 394 | } else if (p == rq->idle) { |
2d513868 | 395 | account_idle_time(cputime); |
73fbec60 | 396 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ |
2d513868 | 397 | account_guest_time(p, cputime, scaled); |
73fbec60 | 398 | } else { |
2d513868 | 399 | __account_system_time(p, cputime, scaled, CPUTIME_SYSTEM); |
73fbec60 FW |
400 | } |
401 | } | |
402 | ||
403 | static void irqtime_account_idle_ticks(int ticks) | |
404 | { | |
73fbec60 FW |
405 | struct rq *rq = this_rq(); |
406 | ||
2d513868 | 407 | irqtime_account_process_tick(current, 0, rq, ticks); |
73fbec60 FW |
408 | } |
409 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
3f4724ea FW |
410 | static inline void irqtime_account_idle_ticks(int ticks) {} |
411 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 412 | struct rq *rq, int nr_ticks) {} |
73fbec60 FW |
413 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
414 | ||
73fbec60 FW |
415 | /* |
416 | * Use precise platform statistics if available: | |
417 | */ | |
418 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
a7e1a9e3 | 419 | |
e3942ba0 | 420 | #ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
b0493406 | 421 | void vtime_common_task_switch(struct task_struct *prev) |
e3942ba0 FW |
422 | { |
423 | if (is_idle_task(prev)) | |
424 | vtime_account_idle(prev); | |
425 | else | |
426 | vtime_account_system(prev); | |
427 | ||
abf917cd | 428 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
e3942ba0 | 429 | vtime_account_user(prev); |
abf917cd | 430 | #endif |
e3942ba0 FW |
431 | arch_vtime_task_switch(prev); |
432 | } | |
433 | #endif | |
11113334 | 434 | |
0cfdf9a1 FW |
435 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ |
436 | ||
437 | ||
438 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE | |
a7e1a9e3 FW |
439 | /* |
440 | * Archs that account the whole time spent in the idle task | |
441 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 442 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
443 | * have other meaning of the idle time (s390 only includes the |
444 | * time spent by the CPU when it's in low power mode) must override | |
445 | * vtime_account(). | |
446 | */ | |
447 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
0cfdf9a1 | 448 | void vtime_account_irq_enter(struct task_struct *tsk) |
a7e1a9e3 | 449 | { |
0cfdf9a1 FW |
450 | if (!in_interrupt() && is_idle_task(tsk)) |
451 | vtime_account_idle(tsk); | |
452 | else | |
453 | vtime_account_system(tsk); | |
a7e1a9e3 | 454 | } |
0cfdf9a1 | 455 | EXPORT_SYMBOL_GPL(vtime_account_irq_enter); |
a7e1a9e3 | 456 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ |
9fbc42ea | 457 | |
9fbc42ea FW |
458 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
459 | { | |
460 | *ut = p->utime; | |
461 | *st = p->stime; | |
462 | } | |
9eec50b8 | 463 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
a7e1a9e3 | 464 | |
9fbc42ea FW |
465 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
466 | { | |
467 | struct task_cputime cputime; | |
73fbec60 | 468 | |
9fbc42ea FW |
469 | thread_group_cputime(p, &cputime); |
470 | ||
471 | *ut = cputime.utime; | |
472 | *st = cputime.stime; | |
473 | } | |
474 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ | |
475 | /* | |
476 | * Account a single tick of cpu time. | |
477 | * @p: the process that the cpu time gets accounted to | |
478 | * @user_tick: indicates if the tick is a user or a system tick | |
479 | */ | |
480 | void account_process_tick(struct task_struct *p, int user_tick) | |
73fbec60 | 481 | { |
57430218 | 482 | cputime_t cputime, scaled, steal; |
9fbc42ea | 483 | struct rq *rq = this_rq(); |
73fbec60 | 484 | |
55dbdcfa | 485 | if (vtime_accounting_cpu_enabled()) |
9fbc42ea FW |
486 | return; |
487 | ||
488 | if (sched_clock_irqtime) { | |
2d513868 | 489 | irqtime_account_process_tick(p, user_tick, rq, 1); |
9fbc42ea FW |
490 | return; |
491 | } | |
492 | ||
57430218 | 493 | cputime = cputime_one_jiffy; |
03cbc732 | 494 | steal = steal_account_process_time(ULONG_MAX); |
57430218 RR |
495 | |
496 | if (steal >= cputime) | |
9fbc42ea | 497 | return; |
73fbec60 | 498 | |
57430218 RR |
499 | cputime -= steal; |
500 | scaled = cputime_to_scaled(cputime); | |
501 | ||
9fbc42ea | 502 | if (user_tick) |
57430218 | 503 | account_user_time(p, cputime, scaled); |
9fbc42ea | 504 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
57430218 | 505 | account_system_time(p, HARDIRQ_OFFSET, cputime, scaled); |
73fbec60 | 506 | else |
57430218 | 507 | account_idle_time(cputime); |
9fbc42ea | 508 | } |
73fbec60 | 509 | |
9fbc42ea FW |
510 | /* |
511 | * Account multiple ticks of idle time. | |
512 | * @ticks: number of stolen ticks | |
513 | */ | |
514 | void account_idle_ticks(unsigned long ticks) | |
515 | { | |
f9bcf1e0 | 516 | cputime_t cputime, steal; |
26f2c75c | 517 | |
9fbc42ea FW |
518 | if (sched_clock_irqtime) { |
519 | irqtime_account_idle_ticks(ticks); | |
520 | return; | |
521 | } | |
522 | ||
26f2c75c | 523 | cputime = jiffies_to_cputime(ticks); |
03cbc732 | 524 | steal = steal_account_process_time(ULONG_MAX); |
f9bcf1e0 WL |
525 | |
526 | if (steal >= cputime) | |
527 | return; | |
528 | ||
529 | cputime -= steal; | |
530 | account_idle_time(cputime); | |
9fbc42ea | 531 | } |
73fbec60 | 532 | |
d9a3c982 | 533 | /* |
55eaa7c1 SG |
534 | * Perform (stime * rtime) / total, but avoid multiplication overflow by |
535 | * loosing precision when the numbers are big. | |
d9a3c982 FW |
536 | */ |
537 | static cputime_t scale_stime(u64 stime, u64 rtime, u64 total) | |
73fbec60 | 538 | { |
55eaa7c1 | 539 | u64 scaled; |
73fbec60 | 540 | |
55eaa7c1 SG |
541 | for (;;) { |
542 | /* Make sure "rtime" is the bigger of stime/rtime */ | |
84f9f3a1 SG |
543 | if (stime > rtime) |
544 | swap(rtime, stime); | |
55eaa7c1 SG |
545 | |
546 | /* Make sure 'total' fits in 32 bits */ | |
547 | if (total >> 32) | |
548 | goto drop_precision; | |
549 | ||
550 | /* Does rtime (and thus stime) fit in 32 bits? */ | |
551 | if (!(rtime >> 32)) | |
552 | break; | |
553 | ||
554 | /* Can we just balance rtime/stime rather than dropping bits? */ | |
555 | if (stime >> 31) | |
556 | goto drop_precision; | |
557 | ||
558 | /* We can grow stime and shrink rtime and try to make them both fit */ | |
559 | stime <<= 1; | |
560 | rtime >>= 1; | |
561 | continue; | |
562 | ||
563 | drop_precision: | |
564 | /* We drop from rtime, it has more bits than stime */ | |
565 | rtime >>= 1; | |
566 | total >>= 1; | |
d9a3c982 | 567 | } |
73fbec60 | 568 | |
55eaa7c1 SG |
569 | /* |
570 | * Make sure gcc understands that this is a 32x32->64 multiply, | |
571 | * followed by a 64/32->64 divide. | |
572 | */ | |
573 | scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total); | |
d9a3c982 | 574 | return (__force cputime_t) scaled; |
73fbec60 FW |
575 | } |
576 | ||
347abad9 | 577 | /* |
9d7fb042 PZ |
578 | * Adjust tick based cputime random precision against scheduler runtime |
579 | * accounting. | |
347abad9 | 580 | * |
9d7fb042 PZ |
581 | * Tick based cputime accounting depend on random scheduling timeslices of a |
582 | * task to be interrupted or not by the timer. Depending on these | |
583 | * circumstances, the number of these interrupts may be over or | |
584 | * under-optimistic, matching the real user and system cputime with a variable | |
585 | * precision. | |
586 | * | |
587 | * Fix this by scaling these tick based values against the total runtime | |
588 | * accounted by the CFS scheduler. | |
589 | * | |
590 | * This code provides the following guarantees: | |
591 | * | |
592 | * stime + utime == rtime | |
593 | * stime_i+1 >= stime_i, utime_i+1 >= utime_i | |
594 | * | |
595 | * Assuming that rtime_i+1 >= rtime_i. | |
fa092057 | 596 | */ |
d37f761d | 597 | static void cputime_adjust(struct task_cputime *curr, |
9d7fb042 | 598 | struct prev_cputime *prev, |
d37f761d | 599 | cputime_t *ut, cputime_t *st) |
73fbec60 | 600 | { |
5a8e01f8 | 601 | cputime_t rtime, stime, utime; |
9d7fb042 | 602 | unsigned long flags; |
fa092057 | 603 | |
9d7fb042 PZ |
604 | /* Serialize concurrent callers such that we can honour our guarantees */ |
605 | raw_spin_lock_irqsave(&prev->lock, flags); | |
d37f761d | 606 | rtime = nsecs_to_cputime(curr->sum_exec_runtime); |
73fbec60 | 607 | |
772c808a | 608 | /* |
9d7fb042 PZ |
609 | * This is possible under two circumstances: |
610 | * - rtime isn't monotonic after all (a bug); | |
611 | * - we got reordered by the lock. | |
612 | * | |
613 | * In both cases this acts as a filter such that the rest of the code | |
614 | * can assume it is monotonic regardless of anything else. | |
772c808a SG |
615 | */ |
616 | if (prev->stime + prev->utime >= rtime) | |
617 | goto out; | |
618 | ||
5a8e01f8 SG |
619 | stime = curr->stime; |
620 | utime = curr->utime; | |
621 | ||
173be9a1 PZ |
622 | /* |
623 | * If either stime or both stime and utime are 0, assume all runtime is | |
624 | * userspace. Once a task gets some ticks, the monotonicy code at | |
625 | * 'update' will ensure things converge to the observed ratio. | |
626 | */ | |
627 | if (stime == 0) { | |
628 | utime = rtime; | |
9d7fb042 PZ |
629 | goto update; |
630 | } | |
5a8e01f8 | 631 | |
173be9a1 PZ |
632 | if (utime == 0) { |
633 | stime = rtime; | |
9d7fb042 | 634 | goto update; |
d9a3c982 | 635 | } |
73fbec60 | 636 | |
9d7fb042 PZ |
637 | stime = scale_stime((__force u64)stime, (__force u64)rtime, |
638 | (__force u64)(stime + utime)); | |
639 | ||
173be9a1 | 640 | update: |
9d7fb042 PZ |
641 | /* |
642 | * Make sure stime doesn't go backwards; this preserves monotonicity | |
643 | * for utime because rtime is monotonic. | |
644 | * | |
645 | * utime_i+1 = rtime_i+1 - stime_i | |
646 | * = rtime_i+1 - (rtime_i - utime_i) | |
647 | * = (rtime_i+1 - rtime_i) + utime_i | |
648 | * >= utime_i | |
649 | */ | |
650 | if (stime < prev->stime) | |
651 | stime = prev->stime; | |
652 | utime = rtime - stime; | |
653 | ||
654 | /* | |
655 | * Make sure utime doesn't go backwards; this still preserves | |
656 | * monotonicity for stime, analogous argument to above. | |
657 | */ | |
658 | if (utime < prev->utime) { | |
659 | utime = prev->utime; | |
660 | stime = rtime - utime; | |
661 | } | |
d37f761d | 662 | |
9d7fb042 PZ |
663 | prev->stime = stime; |
664 | prev->utime = utime; | |
772c808a | 665 | out: |
d37f761d FW |
666 | *ut = prev->utime; |
667 | *st = prev->stime; | |
9d7fb042 | 668 | raw_spin_unlock_irqrestore(&prev->lock, flags); |
d37f761d | 669 | } |
73fbec60 | 670 | |
d37f761d FW |
671 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
672 | { | |
673 | struct task_cputime cputime = { | |
d37f761d FW |
674 | .sum_exec_runtime = p->se.sum_exec_runtime, |
675 | }; | |
676 | ||
6fac4829 | 677 | task_cputime(p, &cputime.utime, &cputime.stime); |
d37f761d | 678 | cputime_adjust(&cputime, &p->prev_cputime, ut, st); |
73fbec60 | 679 | } |
9eec50b8 | 680 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
73fbec60 | 681 | |
e80d0a1a | 682 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 | 683 | { |
73fbec60 | 684 | struct task_cputime cputime; |
73fbec60 FW |
685 | |
686 | thread_group_cputime(p, &cputime); | |
d37f761d | 687 | cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); |
73fbec60 | 688 | } |
9fbc42ea | 689 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ |
abf917cd FW |
690 | |
691 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN | |
ff9a9b4c | 692 | static cputime_t vtime_delta(struct task_struct *tsk) |
6a61671b | 693 | { |
ff9a9b4c | 694 | unsigned long now = READ_ONCE(jiffies); |
6a61671b | 695 | |
ff9a9b4c | 696 | if (time_before(now, (unsigned long)tsk->vtime_snap)) |
6a61671b | 697 | return 0; |
abf917cd | 698 | |
ff9a9b4c | 699 | return jiffies_to_cputime(now - tsk->vtime_snap); |
6a61671b FW |
700 | } |
701 | ||
702 | static cputime_t get_vtime_delta(struct task_struct *tsk) | |
abf917cd | 703 | { |
ff9a9b4c | 704 | unsigned long now = READ_ONCE(jiffies); |
b58c3584 | 705 | cputime_t delta, other; |
abf917cd | 706 | |
03cbc732 WL |
707 | /* |
708 | * Unlike tick based timing, vtime based timing never has lost | |
709 | * ticks, and no need for steal time accounting to make up for | |
710 | * lost ticks. Vtime accounts a rounded version of actual | |
711 | * elapsed time. Limit account_other_time to prevent rounding | |
712 | * errors from causing elapsed vtime to go negative. | |
713 | */ | |
57430218 | 714 | delta = jiffies_to_cputime(now - tsk->vtime_snap); |
b58c3584 | 715 | other = account_other_time(delta); |
7098c1ea | 716 | WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_INACTIVE); |
ff9a9b4c | 717 | tsk->vtime_snap = now; |
abf917cd | 718 | |
b58c3584 | 719 | return delta - other; |
abf917cd FW |
720 | } |
721 | ||
6a61671b FW |
722 | static void __vtime_account_system(struct task_struct *tsk) |
723 | { | |
724 | cputime_t delta_cpu = get_vtime_delta(tsk); | |
725 | ||
726 | account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu)); | |
727 | } | |
728 | ||
abf917cd FW |
729 | void vtime_account_system(struct task_struct *tsk) |
730 | { | |
ff9a9b4c RR |
731 | if (!vtime_delta(tsk)) |
732 | return; | |
733 | ||
b7ce2277 | 734 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b | 735 | __vtime_account_system(tsk); |
b7ce2277 | 736 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b | 737 | } |
3f4724ea | 738 | |
abf917cd FW |
739 | void vtime_account_user(struct task_struct *tsk) |
740 | { | |
3f4724ea FW |
741 | cputime_t delta_cpu; |
742 | ||
b7ce2277 | 743 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b | 744 | tsk->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c RR |
745 | if (vtime_delta(tsk)) { |
746 | delta_cpu = get_vtime_delta(tsk); | |
747 | account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu)); | |
748 | } | |
b7ce2277 | 749 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b FW |
750 | } |
751 | ||
752 | void vtime_user_enter(struct task_struct *tsk) | |
753 | { | |
b7ce2277 | 754 | write_seqcount_begin(&tsk->vtime_seqcount); |
ff9a9b4c RR |
755 | if (vtime_delta(tsk)) |
756 | __vtime_account_system(tsk); | |
af2350bd | 757 | tsk->vtime_snap_whence = VTIME_USER; |
b7ce2277 | 758 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b FW |
759 | } |
760 | ||
761 | void vtime_guest_enter(struct task_struct *tsk) | |
762 | { | |
5b206d48 FW |
763 | /* |
764 | * The flags must be updated under the lock with | |
765 | * the vtime_snap flush and update. | |
766 | * That enforces a right ordering and update sequence | |
767 | * synchronization against the reader (task_gtime()) | |
768 | * that can thus safely catch up with a tickless delta. | |
769 | */ | |
b7ce2277 | 770 | write_seqcount_begin(&tsk->vtime_seqcount); |
ff9a9b4c RR |
771 | if (vtime_delta(tsk)) |
772 | __vtime_account_system(tsk); | |
6a61671b | 773 | current->flags |= PF_VCPU; |
b7ce2277 | 774 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b | 775 | } |
48d6a816 | 776 | EXPORT_SYMBOL_GPL(vtime_guest_enter); |
6a61671b FW |
777 | |
778 | void vtime_guest_exit(struct task_struct *tsk) | |
779 | { | |
b7ce2277 | 780 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b FW |
781 | __vtime_account_system(tsk); |
782 | current->flags &= ~PF_VCPU; | |
b7ce2277 | 783 | write_seqcount_end(&tsk->vtime_seqcount); |
abf917cd | 784 | } |
48d6a816 | 785 | EXPORT_SYMBOL_GPL(vtime_guest_exit); |
abf917cd FW |
786 | |
787 | void vtime_account_idle(struct task_struct *tsk) | |
788 | { | |
6a61671b | 789 | cputime_t delta_cpu = get_vtime_delta(tsk); |
abf917cd FW |
790 | |
791 | account_idle_time(delta_cpu); | |
792 | } | |
3f4724ea | 793 | |
6a61671b FW |
794 | void arch_vtime_task_switch(struct task_struct *prev) |
795 | { | |
b7ce2277 | 796 | write_seqcount_begin(&prev->vtime_seqcount); |
7098c1ea | 797 | prev->vtime_snap_whence = VTIME_INACTIVE; |
b7ce2277 | 798 | write_seqcount_end(&prev->vtime_seqcount); |
6a61671b | 799 | |
b7ce2277 | 800 | write_seqcount_begin(¤t->vtime_seqcount); |
6a61671b | 801 | current->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c | 802 | current->vtime_snap = jiffies; |
b7ce2277 | 803 | write_seqcount_end(¤t->vtime_seqcount); |
6a61671b FW |
804 | } |
805 | ||
45eacc69 | 806 | void vtime_init_idle(struct task_struct *t, int cpu) |
6a61671b FW |
807 | { |
808 | unsigned long flags; | |
809 | ||
b7ce2277 FW |
810 | local_irq_save(flags); |
811 | write_seqcount_begin(&t->vtime_seqcount); | |
6a61671b | 812 | t->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c | 813 | t->vtime_snap = jiffies; |
b7ce2277 FW |
814 | write_seqcount_end(&t->vtime_seqcount); |
815 | local_irq_restore(flags); | |
6a61671b FW |
816 | } |
817 | ||
818 | cputime_t task_gtime(struct task_struct *t) | |
819 | { | |
6a61671b FW |
820 | unsigned int seq; |
821 | cputime_t gtime; | |
822 | ||
e5925394 | 823 | if (!vtime_accounting_enabled()) |
2541117b HS |
824 | return t->gtime; |
825 | ||
6a61671b | 826 | do { |
b7ce2277 | 827 | seq = read_seqcount_begin(&t->vtime_seqcount); |
6a61671b FW |
828 | |
829 | gtime = t->gtime; | |
cab245d6 | 830 | if (t->vtime_snap_whence == VTIME_SYS && t->flags & PF_VCPU) |
6a61671b FW |
831 | gtime += vtime_delta(t); |
832 | ||
b7ce2277 | 833 | } while (read_seqcount_retry(&t->vtime_seqcount, seq)); |
6a61671b FW |
834 | |
835 | return gtime; | |
836 | } | |
837 | ||
838 | /* | |
839 | * Fetch cputime raw values from fields of task_struct and | |
840 | * add up the pending nohz execution time since the last | |
841 | * cputime snapshot. | |
842 | */ | |
843 | static void | |
844 | fetch_task_cputime(struct task_struct *t, | |
845 | cputime_t *u_dst, cputime_t *s_dst, | |
846 | cputime_t *u_src, cputime_t *s_src, | |
847 | cputime_t *udelta, cputime_t *sdelta) | |
848 | { | |
6a61671b FW |
849 | unsigned int seq; |
850 | unsigned long long delta; | |
851 | ||
852 | do { | |
853 | *udelta = 0; | |
854 | *sdelta = 0; | |
855 | ||
b7ce2277 | 856 | seq = read_seqcount_begin(&t->vtime_seqcount); |
6a61671b FW |
857 | |
858 | if (u_dst) | |
859 | *u_dst = *u_src; | |
860 | if (s_dst) | |
861 | *s_dst = *s_src; | |
862 | ||
863 | /* Task is sleeping, nothing to add */ | |
7098c1ea | 864 | if (t->vtime_snap_whence == VTIME_INACTIVE || |
6a61671b FW |
865 | is_idle_task(t)) |
866 | continue; | |
867 | ||
868 | delta = vtime_delta(t); | |
869 | ||
870 | /* | |
871 | * Task runs either in user or kernel space, add pending nohz time to | |
872 | * the right place. | |
873 | */ | |
874 | if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) { | |
875 | *udelta = delta; | |
876 | } else { | |
877 | if (t->vtime_snap_whence == VTIME_SYS) | |
878 | *sdelta = delta; | |
879 | } | |
b7ce2277 | 880 | } while (read_seqcount_retry(&t->vtime_seqcount, seq)); |
6a61671b FW |
881 | } |
882 | ||
883 | ||
884 | void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime) | |
885 | { | |
886 | cputime_t udelta, sdelta; | |
887 | ||
e5925394 | 888 | if (!vtime_accounting_enabled()) { |
7877a0ba HS |
889 | if (utime) |
890 | *utime = t->utime; | |
891 | if (stime) | |
892 | *stime = t->stime; | |
893 | return; | |
894 | } | |
895 | ||
6a61671b FW |
896 | fetch_task_cputime(t, utime, stime, &t->utime, |
897 | &t->stime, &udelta, &sdelta); | |
898 | if (utime) | |
899 | *utime += udelta; | |
900 | if (stime) | |
901 | *stime += sdelta; | |
902 | } | |
903 | ||
904 | void task_cputime_scaled(struct task_struct *t, | |
905 | cputime_t *utimescaled, cputime_t *stimescaled) | |
906 | { | |
907 | cputime_t udelta, sdelta; | |
908 | ||
e5925394 | 909 | if (!vtime_accounting_enabled()) { |
7877a0ba HS |
910 | if (utimescaled) |
911 | *utimescaled = t->utimescaled; | |
912 | if (stimescaled) | |
913 | *stimescaled = t->stimescaled; | |
914 | return; | |
915 | } | |
916 | ||
6a61671b FW |
917 | fetch_task_cputime(t, utimescaled, stimescaled, |
918 | &t->utimescaled, &t->stimescaled, &udelta, &sdelta); | |
919 | if (utimescaled) | |
920 | *utimescaled += cputime_to_scaled(udelta); | |
921 | if (stimescaled) | |
922 | *stimescaled += cputime_to_scaled(sdelta); | |
923 | } | |
abf917cd | 924 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |