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 | ||
a1eb1411 SG |
309 | #ifdef CONFIG_64BIT |
310 | static inline u64 read_sum_exec_runtime(struct task_struct *t) | |
311 | { | |
312 | return t->se.sum_exec_runtime; | |
313 | } | |
314 | #else | |
315 | static u64 read_sum_exec_runtime(struct task_struct *t) | |
316 | { | |
317 | u64 ns; | |
318 | struct rq_flags rf; | |
319 | struct rq *rq; | |
320 | ||
321 | rq = task_rq_lock(t, &rf); | |
322 | ns = t->se.sum_exec_runtime; | |
323 | task_rq_unlock(rq, t, &rf); | |
324 | ||
325 | return ns; | |
326 | } | |
327 | #endif | |
328 | ||
a634f933 FW |
329 | /* |
330 | * Accumulate raw cputime values of dead tasks (sig->[us]time) and live | |
331 | * tasks (sum on group iteration) belonging to @tsk's group. | |
332 | */ | |
333 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) | |
334 | { | |
335 | struct signal_struct *sig = tsk->signal; | |
6fac4829 | 336 | cputime_t utime, stime; |
a634f933 | 337 | struct task_struct *t; |
e78c3496 | 338 | unsigned int seq, nextseq; |
9c368b5b | 339 | unsigned long flags; |
a634f933 | 340 | |
a1eb1411 SG |
341 | /* |
342 | * Update current task runtime to account pending time since last | |
343 | * scheduler action or thread_group_cputime() call. This thread group | |
344 | * might have other running tasks on different CPUs, but updating | |
345 | * their runtime can affect syscall performance, so we skip account | |
346 | * those pending times and rely only on values updated on tick or | |
347 | * other scheduler action. | |
348 | */ | |
349 | if (same_thread_group(current, tsk)) | |
350 | (void) task_sched_runtime(current); | |
351 | ||
a634f933 | 352 | rcu_read_lock(); |
e78c3496 RR |
353 | /* Attempt a lockless read on the first round. */ |
354 | nextseq = 0; | |
355 | do { | |
356 | seq = nextseq; | |
9c368b5b | 357 | flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq); |
e78c3496 RR |
358 | times->utime = sig->utime; |
359 | times->stime = sig->stime; | |
360 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
361 | ||
362 | for_each_thread(tsk, t) { | |
363 | task_cputime(t, &utime, &stime); | |
364 | times->utime += utime; | |
365 | times->stime += stime; | |
a1eb1411 | 366 | times->sum_exec_runtime += read_sum_exec_runtime(t); |
e78c3496 RR |
367 | } |
368 | /* If lockless access failed, take the lock. */ | |
369 | nextseq = 1; | |
370 | } while (need_seqretry(&sig->stats_lock, seq)); | |
9c368b5b | 371 | done_seqretry_irqrestore(&sig->stats_lock, seq, flags); |
a634f933 FW |
372 | rcu_read_unlock(); |
373 | } | |
374 | ||
73fbec60 FW |
375 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
376 | /* | |
377 | * Account a tick to a process and cpustat | |
378 | * @p: the process that the cpu time gets accounted to | |
379 | * @user_tick: is the tick from userspace | |
380 | * @rq: the pointer to rq | |
381 | * | |
382 | * Tick demultiplexing follows the order | |
383 | * - pending hardirq update | |
384 | * - pending softirq update | |
385 | * - user_time | |
386 | * - idle_time | |
387 | * - system time | |
388 | * - check for guest_time | |
389 | * - else account as system_time | |
390 | * | |
391 | * Check for hardirq is done both for system and user time as there is | |
392 | * no timer going off while we are on hardirq and hence we may never get an | |
393 | * opportunity to update it solely in system time. | |
394 | * p->stime and friends are only updated on system time and not on irq | |
395 | * softirq as those do not count in task exec_runtime any more. | |
396 | */ | |
397 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 398 | struct rq *rq, int ticks) |
73fbec60 | 399 | { |
57430218 RR |
400 | u64 cputime = (__force u64) cputime_one_jiffy * ticks; |
401 | cputime_t scaled, other; | |
73fbec60 | 402 | |
57430218 RR |
403 | /* |
404 | * When returning from idle, many ticks can get accounted at | |
405 | * once, including some ticks of steal, irq, and softirq time. | |
406 | * Subtract those ticks from the amount of time accounted to | |
407 | * idle, or potentially user or system time. Due to rounding, | |
408 | * other time can exceed ticks occasionally. | |
409 | */ | |
03cbc732 | 410 | other = account_other_time(ULONG_MAX); |
57430218 | 411 | if (other >= cputime) |
73fbec60 | 412 | return; |
57430218 RR |
413 | cputime -= other; |
414 | scaled = cputime_to_scaled(cputime); | |
73fbec60 | 415 | |
57430218 | 416 | if (this_cpu_ksoftirqd() == p) { |
73fbec60 FW |
417 | /* |
418 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
419 | * So, we have to handle it separately here. | |
420 | * Also, p->stime needs to be updated for ksoftirqd. | |
421 | */ | |
2d513868 | 422 | __account_system_time(p, cputime, scaled, CPUTIME_SOFTIRQ); |
73fbec60 | 423 | } else if (user_tick) { |
2d513868 | 424 | account_user_time(p, cputime, scaled); |
73fbec60 | 425 | } else if (p == rq->idle) { |
2d513868 | 426 | account_idle_time(cputime); |
73fbec60 | 427 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ |
2d513868 | 428 | account_guest_time(p, cputime, scaled); |
73fbec60 | 429 | } else { |
2d513868 | 430 | __account_system_time(p, cputime, scaled, CPUTIME_SYSTEM); |
73fbec60 FW |
431 | } |
432 | } | |
433 | ||
434 | static void irqtime_account_idle_ticks(int ticks) | |
435 | { | |
73fbec60 FW |
436 | struct rq *rq = this_rq(); |
437 | ||
2d513868 | 438 | irqtime_account_process_tick(current, 0, rq, ticks); |
73fbec60 FW |
439 | } |
440 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
3f4724ea FW |
441 | static inline void irqtime_account_idle_ticks(int ticks) {} |
442 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 443 | struct rq *rq, int nr_ticks) {} |
73fbec60 FW |
444 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
445 | ||
73fbec60 FW |
446 | /* |
447 | * Use precise platform statistics if available: | |
448 | */ | |
449 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
a7e1a9e3 | 450 | |
e3942ba0 | 451 | #ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
b0493406 | 452 | void vtime_common_task_switch(struct task_struct *prev) |
e3942ba0 FW |
453 | { |
454 | if (is_idle_task(prev)) | |
455 | vtime_account_idle(prev); | |
456 | else | |
457 | vtime_account_system(prev); | |
458 | ||
abf917cd | 459 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
e3942ba0 | 460 | vtime_account_user(prev); |
abf917cd | 461 | #endif |
e3942ba0 FW |
462 | arch_vtime_task_switch(prev); |
463 | } | |
464 | #endif | |
11113334 | 465 | |
0cfdf9a1 FW |
466 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ |
467 | ||
468 | ||
469 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE | |
a7e1a9e3 FW |
470 | /* |
471 | * Archs that account the whole time spent in the idle task | |
472 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 473 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
474 | * have other meaning of the idle time (s390 only includes the |
475 | * time spent by the CPU when it's in low power mode) must override | |
476 | * vtime_account(). | |
477 | */ | |
478 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
0cfdf9a1 | 479 | void vtime_account_irq_enter(struct task_struct *tsk) |
a7e1a9e3 | 480 | { |
0cfdf9a1 FW |
481 | if (!in_interrupt() && is_idle_task(tsk)) |
482 | vtime_account_idle(tsk); | |
483 | else | |
484 | vtime_account_system(tsk); | |
a7e1a9e3 | 485 | } |
0cfdf9a1 | 486 | EXPORT_SYMBOL_GPL(vtime_account_irq_enter); |
a7e1a9e3 | 487 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ |
9fbc42ea | 488 | |
9fbc42ea FW |
489 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
490 | { | |
491 | *ut = p->utime; | |
492 | *st = p->stime; | |
493 | } | |
9eec50b8 | 494 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
a7e1a9e3 | 495 | |
9fbc42ea FW |
496 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
497 | { | |
498 | struct task_cputime cputime; | |
73fbec60 | 499 | |
9fbc42ea FW |
500 | thread_group_cputime(p, &cputime); |
501 | ||
502 | *ut = cputime.utime; | |
503 | *st = cputime.stime; | |
504 | } | |
505 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ | |
506 | /* | |
507 | * Account a single tick of cpu time. | |
508 | * @p: the process that the cpu time gets accounted to | |
509 | * @user_tick: indicates if the tick is a user or a system tick | |
510 | */ | |
511 | void account_process_tick(struct task_struct *p, int user_tick) | |
73fbec60 | 512 | { |
57430218 | 513 | cputime_t cputime, scaled, steal; |
9fbc42ea | 514 | struct rq *rq = this_rq(); |
73fbec60 | 515 | |
55dbdcfa | 516 | if (vtime_accounting_cpu_enabled()) |
9fbc42ea FW |
517 | return; |
518 | ||
519 | if (sched_clock_irqtime) { | |
2d513868 | 520 | irqtime_account_process_tick(p, user_tick, rq, 1); |
9fbc42ea FW |
521 | return; |
522 | } | |
523 | ||
57430218 | 524 | cputime = cputime_one_jiffy; |
03cbc732 | 525 | steal = steal_account_process_time(ULONG_MAX); |
57430218 RR |
526 | |
527 | if (steal >= cputime) | |
9fbc42ea | 528 | return; |
73fbec60 | 529 | |
57430218 RR |
530 | cputime -= steal; |
531 | scaled = cputime_to_scaled(cputime); | |
532 | ||
9fbc42ea | 533 | if (user_tick) |
57430218 | 534 | account_user_time(p, cputime, scaled); |
9fbc42ea | 535 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
57430218 | 536 | account_system_time(p, HARDIRQ_OFFSET, cputime, scaled); |
73fbec60 | 537 | else |
57430218 | 538 | account_idle_time(cputime); |
9fbc42ea | 539 | } |
73fbec60 | 540 | |
9fbc42ea FW |
541 | /* |
542 | * Account multiple ticks of idle time. | |
543 | * @ticks: number of stolen ticks | |
544 | */ | |
545 | void account_idle_ticks(unsigned long ticks) | |
546 | { | |
f9bcf1e0 | 547 | cputime_t cputime, steal; |
26f2c75c | 548 | |
9fbc42ea FW |
549 | if (sched_clock_irqtime) { |
550 | irqtime_account_idle_ticks(ticks); | |
551 | return; | |
552 | } | |
553 | ||
26f2c75c | 554 | cputime = jiffies_to_cputime(ticks); |
03cbc732 | 555 | steal = steal_account_process_time(ULONG_MAX); |
f9bcf1e0 WL |
556 | |
557 | if (steal >= cputime) | |
558 | return; | |
559 | ||
560 | cputime -= steal; | |
561 | account_idle_time(cputime); | |
9fbc42ea | 562 | } |
73fbec60 | 563 | |
d9a3c982 | 564 | /* |
55eaa7c1 SG |
565 | * Perform (stime * rtime) / total, but avoid multiplication overflow by |
566 | * loosing precision when the numbers are big. | |
d9a3c982 FW |
567 | */ |
568 | static cputime_t scale_stime(u64 stime, u64 rtime, u64 total) | |
73fbec60 | 569 | { |
55eaa7c1 | 570 | u64 scaled; |
73fbec60 | 571 | |
55eaa7c1 SG |
572 | for (;;) { |
573 | /* Make sure "rtime" is the bigger of stime/rtime */ | |
84f9f3a1 SG |
574 | if (stime > rtime) |
575 | swap(rtime, stime); | |
55eaa7c1 SG |
576 | |
577 | /* Make sure 'total' fits in 32 bits */ | |
578 | if (total >> 32) | |
579 | goto drop_precision; | |
580 | ||
581 | /* Does rtime (and thus stime) fit in 32 bits? */ | |
582 | if (!(rtime >> 32)) | |
583 | break; | |
584 | ||
585 | /* Can we just balance rtime/stime rather than dropping bits? */ | |
586 | if (stime >> 31) | |
587 | goto drop_precision; | |
588 | ||
589 | /* We can grow stime and shrink rtime and try to make them both fit */ | |
590 | stime <<= 1; | |
591 | rtime >>= 1; | |
592 | continue; | |
593 | ||
594 | drop_precision: | |
595 | /* We drop from rtime, it has more bits than stime */ | |
596 | rtime >>= 1; | |
597 | total >>= 1; | |
d9a3c982 | 598 | } |
73fbec60 | 599 | |
55eaa7c1 SG |
600 | /* |
601 | * Make sure gcc understands that this is a 32x32->64 multiply, | |
602 | * followed by a 64/32->64 divide. | |
603 | */ | |
604 | scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total); | |
d9a3c982 | 605 | return (__force cputime_t) scaled; |
73fbec60 FW |
606 | } |
607 | ||
347abad9 | 608 | /* |
9d7fb042 PZ |
609 | * Adjust tick based cputime random precision against scheduler runtime |
610 | * accounting. | |
347abad9 | 611 | * |
9d7fb042 PZ |
612 | * Tick based cputime accounting depend on random scheduling timeslices of a |
613 | * task to be interrupted or not by the timer. Depending on these | |
614 | * circumstances, the number of these interrupts may be over or | |
615 | * under-optimistic, matching the real user and system cputime with a variable | |
616 | * precision. | |
617 | * | |
618 | * Fix this by scaling these tick based values against the total runtime | |
619 | * accounted by the CFS scheduler. | |
620 | * | |
621 | * This code provides the following guarantees: | |
622 | * | |
623 | * stime + utime == rtime | |
624 | * stime_i+1 >= stime_i, utime_i+1 >= utime_i | |
625 | * | |
626 | * Assuming that rtime_i+1 >= rtime_i. | |
fa092057 | 627 | */ |
d37f761d | 628 | static void cputime_adjust(struct task_cputime *curr, |
9d7fb042 | 629 | struct prev_cputime *prev, |
d37f761d | 630 | cputime_t *ut, cputime_t *st) |
73fbec60 | 631 | { |
5a8e01f8 | 632 | cputime_t rtime, stime, utime; |
9d7fb042 | 633 | unsigned long flags; |
fa092057 | 634 | |
9d7fb042 PZ |
635 | /* Serialize concurrent callers such that we can honour our guarantees */ |
636 | raw_spin_lock_irqsave(&prev->lock, flags); | |
d37f761d | 637 | rtime = nsecs_to_cputime(curr->sum_exec_runtime); |
73fbec60 | 638 | |
772c808a | 639 | /* |
9d7fb042 PZ |
640 | * This is possible under two circumstances: |
641 | * - rtime isn't monotonic after all (a bug); | |
642 | * - we got reordered by the lock. | |
643 | * | |
644 | * In both cases this acts as a filter such that the rest of the code | |
645 | * can assume it is monotonic regardless of anything else. | |
772c808a SG |
646 | */ |
647 | if (prev->stime + prev->utime >= rtime) | |
648 | goto out; | |
649 | ||
5a8e01f8 SG |
650 | stime = curr->stime; |
651 | utime = curr->utime; | |
652 | ||
173be9a1 PZ |
653 | /* |
654 | * If either stime or both stime and utime are 0, assume all runtime is | |
655 | * userspace. Once a task gets some ticks, the monotonicy code at | |
656 | * 'update' will ensure things converge to the observed ratio. | |
657 | */ | |
658 | if (stime == 0) { | |
659 | utime = rtime; | |
9d7fb042 PZ |
660 | goto update; |
661 | } | |
5a8e01f8 | 662 | |
173be9a1 PZ |
663 | if (utime == 0) { |
664 | stime = rtime; | |
9d7fb042 | 665 | goto update; |
d9a3c982 | 666 | } |
73fbec60 | 667 | |
9d7fb042 PZ |
668 | stime = scale_stime((__force u64)stime, (__force u64)rtime, |
669 | (__force u64)(stime + utime)); | |
670 | ||
173be9a1 | 671 | update: |
9d7fb042 PZ |
672 | /* |
673 | * Make sure stime doesn't go backwards; this preserves monotonicity | |
674 | * for utime because rtime is monotonic. | |
675 | * | |
676 | * utime_i+1 = rtime_i+1 - stime_i | |
677 | * = rtime_i+1 - (rtime_i - utime_i) | |
678 | * = (rtime_i+1 - rtime_i) + utime_i | |
679 | * >= utime_i | |
680 | */ | |
681 | if (stime < prev->stime) | |
682 | stime = prev->stime; | |
683 | utime = rtime - stime; | |
684 | ||
685 | /* | |
686 | * Make sure utime doesn't go backwards; this still preserves | |
687 | * monotonicity for stime, analogous argument to above. | |
688 | */ | |
689 | if (utime < prev->utime) { | |
690 | utime = prev->utime; | |
691 | stime = rtime - utime; | |
692 | } | |
d37f761d | 693 | |
9d7fb042 PZ |
694 | prev->stime = stime; |
695 | prev->utime = utime; | |
772c808a | 696 | out: |
d37f761d FW |
697 | *ut = prev->utime; |
698 | *st = prev->stime; | |
9d7fb042 | 699 | raw_spin_unlock_irqrestore(&prev->lock, flags); |
d37f761d | 700 | } |
73fbec60 | 701 | |
d37f761d FW |
702 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
703 | { | |
704 | struct task_cputime cputime = { | |
d37f761d FW |
705 | .sum_exec_runtime = p->se.sum_exec_runtime, |
706 | }; | |
707 | ||
6fac4829 | 708 | task_cputime(p, &cputime.utime, &cputime.stime); |
d37f761d | 709 | cputime_adjust(&cputime, &p->prev_cputime, ut, st); |
73fbec60 | 710 | } |
9eec50b8 | 711 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
73fbec60 | 712 | |
e80d0a1a | 713 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 | 714 | { |
73fbec60 | 715 | struct task_cputime cputime; |
73fbec60 FW |
716 | |
717 | thread_group_cputime(p, &cputime); | |
d37f761d | 718 | cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); |
73fbec60 | 719 | } |
9fbc42ea | 720 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ |
abf917cd FW |
721 | |
722 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN | |
ff9a9b4c | 723 | static cputime_t vtime_delta(struct task_struct *tsk) |
6a61671b | 724 | { |
ff9a9b4c | 725 | unsigned long now = READ_ONCE(jiffies); |
6a61671b | 726 | |
ff9a9b4c | 727 | if (time_before(now, (unsigned long)tsk->vtime_snap)) |
6a61671b | 728 | return 0; |
abf917cd | 729 | |
ff9a9b4c | 730 | return jiffies_to_cputime(now - tsk->vtime_snap); |
6a61671b FW |
731 | } |
732 | ||
733 | static cputime_t get_vtime_delta(struct task_struct *tsk) | |
abf917cd | 734 | { |
ff9a9b4c | 735 | unsigned long now = READ_ONCE(jiffies); |
b58c3584 | 736 | cputime_t delta, other; |
abf917cd | 737 | |
03cbc732 WL |
738 | /* |
739 | * Unlike tick based timing, vtime based timing never has lost | |
740 | * ticks, and no need for steal time accounting to make up for | |
741 | * lost ticks. Vtime accounts a rounded version of actual | |
742 | * elapsed time. Limit account_other_time to prevent rounding | |
743 | * errors from causing elapsed vtime to go negative. | |
744 | */ | |
57430218 | 745 | delta = jiffies_to_cputime(now - tsk->vtime_snap); |
b58c3584 | 746 | other = account_other_time(delta); |
7098c1ea | 747 | WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_INACTIVE); |
ff9a9b4c | 748 | tsk->vtime_snap = now; |
abf917cd | 749 | |
b58c3584 | 750 | return delta - other; |
abf917cd FW |
751 | } |
752 | ||
6a61671b FW |
753 | static void __vtime_account_system(struct task_struct *tsk) |
754 | { | |
755 | cputime_t delta_cpu = get_vtime_delta(tsk); | |
756 | ||
757 | account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu)); | |
758 | } | |
759 | ||
abf917cd FW |
760 | void vtime_account_system(struct task_struct *tsk) |
761 | { | |
ff9a9b4c RR |
762 | if (!vtime_delta(tsk)) |
763 | return; | |
764 | ||
b7ce2277 | 765 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b | 766 | __vtime_account_system(tsk); |
b7ce2277 | 767 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b | 768 | } |
3f4724ea | 769 | |
abf917cd FW |
770 | void vtime_account_user(struct task_struct *tsk) |
771 | { | |
3f4724ea FW |
772 | cputime_t delta_cpu; |
773 | ||
b7ce2277 | 774 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b | 775 | tsk->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c RR |
776 | if (vtime_delta(tsk)) { |
777 | delta_cpu = get_vtime_delta(tsk); | |
778 | account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu)); | |
779 | } | |
b7ce2277 | 780 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b FW |
781 | } |
782 | ||
783 | void vtime_user_enter(struct task_struct *tsk) | |
784 | { | |
b7ce2277 | 785 | write_seqcount_begin(&tsk->vtime_seqcount); |
ff9a9b4c RR |
786 | if (vtime_delta(tsk)) |
787 | __vtime_account_system(tsk); | |
af2350bd | 788 | tsk->vtime_snap_whence = VTIME_USER; |
b7ce2277 | 789 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b FW |
790 | } |
791 | ||
792 | void vtime_guest_enter(struct task_struct *tsk) | |
793 | { | |
5b206d48 FW |
794 | /* |
795 | * The flags must be updated under the lock with | |
796 | * the vtime_snap flush and update. | |
797 | * That enforces a right ordering and update sequence | |
798 | * synchronization against the reader (task_gtime()) | |
799 | * that can thus safely catch up with a tickless delta. | |
800 | */ | |
b7ce2277 | 801 | write_seqcount_begin(&tsk->vtime_seqcount); |
ff9a9b4c RR |
802 | if (vtime_delta(tsk)) |
803 | __vtime_account_system(tsk); | |
6a61671b | 804 | current->flags |= PF_VCPU; |
b7ce2277 | 805 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b | 806 | } |
48d6a816 | 807 | EXPORT_SYMBOL_GPL(vtime_guest_enter); |
6a61671b FW |
808 | |
809 | void vtime_guest_exit(struct task_struct *tsk) | |
810 | { | |
b7ce2277 | 811 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b FW |
812 | __vtime_account_system(tsk); |
813 | current->flags &= ~PF_VCPU; | |
b7ce2277 | 814 | write_seqcount_end(&tsk->vtime_seqcount); |
abf917cd | 815 | } |
48d6a816 | 816 | EXPORT_SYMBOL_GPL(vtime_guest_exit); |
abf917cd FW |
817 | |
818 | void vtime_account_idle(struct task_struct *tsk) | |
819 | { | |
6a61671b | 820 | cputime_t delta_cpu = get_vtime_delta(tsk); |
abf917cd FW |
821 | |
822 | account_idle_time(delta_cpu); | |
823 | } | |
3f4724ea | 824 | |
6a61671b FW |
825 | void arch_vtime_task_switch(struct task_struct *prev) |
826 | { | |
b7ce2277 | 827 | write_seqcount_begin(&prev->vtime_seqcount); |
7098c1ea | 828 | prev->vtime_snap_whence = VTIME_INACTIVE; |
b7ce2277 | 829 | write_seqcount_end(&prev->vtime_seqcount); |
6a61671b | 830 | |
b7ce2277 | 831 | write_seqcount_begin(¤t->vtime_seqcount); |
6a61671b | 832 | current->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c | 833 | current->vtime_snap = jiffies; |
b7ce2277 | 834 | write_seqcount_end(¤t->vtime_seqcount); |
6a61671b FW |
835 | } |
836 | ||
45eacc69 | 837 | void vtime_init_idle(struct task_struct *t, int cpu) |
6a61671b FW |
838 | { |
839 | unsigned long flags; | |
840 | ||
b7ce2277 FW |
841 | local_irq_save(flags); |
842 | write_seqcount_begin(&t->vtime_seqcount); | |
6a61671b | 843 | t->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c | 844 | t->vtime_snap = jiffies; |
b7ce2277 FW |
845 | write_seqcount_end(&t->vtime_seqcount); |
846 | local_irq_restore(flags); | |
6a61671b FW |
847 | } |
848 | ||
849 | cputime_t task_gtime(struct task_struct *t) | |
850 | { | |
6a61671b FW |
851 | unsigned int seq; |
852 | cputime_t gtime; | |
853 | ||
e5925394 | 854 | if (!vtime_accounting_enabled()) |
2541117b HS |
855 | return t->gtime; |
856 | ||
6a61671b | 857 | do { |
b7ce2277 | 858 | seq = read_seqcount_begin(&t->vtime_seqcount); |
6a61671b FW |
859 | |
860 | gtime = t->gtime; | |
cab245d6 | 861 | if (t->vtime_snap_whence == VTIME_SYS && t->flags & PF_VCPU) |
6a61671b FW |
862 | gtime += vtime_delta(t); |
863 | ||
b7ce2277 | 864 | } while (read_seqcount_retry(&t->vtime_seqcount, seq)); |
6a61671b FW |
865 | |
866 | return gtime; | |
867 | } | |
868 | ||
869 | /* | |
870 | * Fetch cputime raw values from fields of task_struct and | |
871 | * add up the pending nohz execution time since the last | |
872 | * cputime snapshot. | |
873 | */ | |
874 | static void | |
875 | fetch_task_cputime(struct task_struct *t, | |
876 | cputime_t *u_dst, cputime_t *s_dst, | |
877 | cputime_t *u_src, cputime_t *s_src, | |
878 | cputime_t *udelta, cputime_t *sdelta) | |
879 | { | |
6a61671b FW |
880 | unsigned int seq; |
881 | unsigned long long delta; | |
882 | ||
883 | do { | |
884 | *udelta = 0; | |
885 | *sdelta = 0; | |
886 | ||
b7ce2277 | 887 | seq = read_seqcount_begin(&t->vtime_seqcount); |
6a61671b FW |
888 | |
889 | if (u_dst) | |
890 | *u_dst = *u_src; | |
891 | if (s_dst) | |
892 | *s_dst = *s_src; | |
893 | ||
894 | /* Task is sleeping, nothing to add */ | |
7098c1ea | 895 | if (t->vtime_snap_whence == VTIME_INACTIVE || |
6a61671b FW |
896 | is_idle_task(t)) |
897 | continue; | |
898 | ||
899 | delta = vtime_delta(t); | |
900 | ||
901 | /* | |
902 | * Task runs either in user or kernel space, add pending nohz time to | |
903 | * the right place. | |
904 | */ | |
905 | if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) { | |
906 | *udelta = delta; | |
907 | } else { | |
908 | if (t->vtime_snap_whence == VTIME_SYS) | |
909 | *sdelta = delta; | |
910 | } | |
b7ce2277 | 911 | } while (read_seqcount_retry(&t->vtime_seqcount, seq)); |
6a61671b FW |
912 | } |
913 | ||
914 | ||
915 | void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime) | |
916 | { | |
917 | cputime_t udelta, sdelta; | |
918 | ||
e5925394 | 919 | if (!vtime_accounting_enabled()) { |
7877a0ba HS |
920 | if (utime) |
921 | *utime = t->utime; | |
922 | if (stime) | |
923 | *stime = t->stime; | |
924 | return; | |
925 | } | |
926 | ||
6a61671b FW |
927 | fetch_task_cputime(t, utime, stime, &t->utime, |
928 | &t->stime, &udelta, &sdelta); | |
929 | if (utime) | |
930 | *utime += udelta; | |
931 | if (stime) | |
932 | *stime += sdelta; | |
933 | } | |
934 | ||
935 | void task_cputime_scaled(struct task_struct *t, | |
936 | cputime_t *utimescaled, cputime_t *stimescaled) | |
937 | { | |
938 | cputime_t udelta, sdelta; | |
939 | ||
e5925394 | 940 | if (!vtime_accounting_enabled()) { |
7877a0ba HS |
941 | if (utimescaled) |
942 | *utimescaled = t->utimescaled; | |
943 | if (stimescaled) | |
944 | *stimescaled = t->stimescaled; | |
945 | return; | |
946 | } | |
947 | ||
6a61671b FW |
948 | fetch_task_cputime(t, utimescaled, stimescaled, |
949 | &t->utimescaled, &t->stimescaled, &udelta, &sdelta); | |
950 | if (utimescaled) | |
951 | *utimescaled += cputime_to_scaled(udelta); | |
952 | if (stimescaled) | |
953 | *stimescaled += cputime_to_scaled(sdelta); | |
954 | } | |
abf917cd | 955 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |