]>
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> |
32ef5517 | 7 | #include <linux/sched/cputime.h> |
73fbec60 | 8 | #include "sched.h" |
73fbec60 FW |
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 | */ | |
19d23dbf | 23 | DEFINE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 | 24 | |
73fbec60 FW |
25 | static int sched_clock_irqtime; |
26 | ||
27 | void enable_sched_clock_irqtime(void) | |
28 | { | |
29 | sched_clock_irqtime = 1; | |
30 | } | |
31 | ||
32 | void disable_sched_clock_irqtime(void) | |
33 | { | |
34 | sched_clock_irqtime = 0; | |
35 | } | |
36 | ||
25e2d8c1 FW |
37 | static void irqtime_account_delta(struct irqtime *irqtime, u64 delta, |
38 | enum cpu_usage_stat idx) | |
39 | { | |
40 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
41 | ||
42 | u64_stats_update_begin(&irqtime->sync); | |
43 | cpustat[idx] += delta; | |
44 | irqtime->total += delta; | |
45 | irqtime->tick_delta += delta; | |
46 | u64_stats_update_end(&irqtime->sync); | |
47 | } | |
48 | ||
73fbec60 FW |
49 | /* |
50 | * Called before incrementing preempt_count on {soft,}irq_enter | |
51 | * and before decrementing preempt_count on {soft,}irq_exit. | |
52 | */ | |
3e1df4f5 | 53 | void irqtime_account_irq(struct task_struct *curr) |
73fbec60 | 54 | { |
19d23dbf | 55 | struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime); |
73fbec60 FW |
56 | s64 delta; |
57 | int cpu; | |
58 | ||
59 | if (!sched_clock_irqtime) | |
60 | return; | |
61 | ||
73fbec60 | 62 | cpu = smp_processor_id(); |
19d23dbf FW |
63 | delta = sched_clock_cpu(cpu) - irqtime->irq_start_time; |
64 | irqtime->irq_start_time += delta; | |
73fbec60 | 65 | |
73fbec60 FW |
66 | /* |
67 | * We do not account for softirq time from ksoftirqd here. | |
68 | * We want to continue accounting softirq time to ksoftirqd thread | |
69 | * in that case, so as not to confuse scheduler with a special task | |
70 | * that do not consume any time, but still wants to run. | |
71 | */ | |
25e2d8c1 FW |
72 | if (hardirq_count()) |
73 | irqtime_account_delta(irqtime, delta, CPUTIME_IRQ); | |
74 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) | |
75 | irqtime_account_delta(irqtime, delta, CPUTIME_SOFTIRQ); | |
73fbec60 | 76 | } |
3e1df4f5 | 77 | EXPORT_SYMBOL_GPL(irqtime_account_irq); |
73fbec60 | 78 | |
2b1f967d | 79 | static u64 irqtime_tick_accounted(u64 maxtime) |
73fbec60 | 80 | { |
a499a5a1 | 81 | struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime); |
2b1f967d | 82 | u64 delta; |
73fbec60 | 83 | |
2b1f967d FW |
84 | delta = min(irqtime->tick_delta, maxtime); |
85 | irqtime->tick_delta -= delta; | |
2810f611 | 86 | |
a499a5a1 | 87 | return delta; |
73fbec60 FW |
88 | } |
89 | ||
90 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
91 | ||
92 | #define sched_clock_irqtime (0) | |
93 | ||
2b1f967d | 94 | static u64 irqtime_tick_accounted(u64 dummy) |
57430218 RR |
95 | { |
96 | return 0; | |
97 | } | |
98 | ||
73fbec60 FW |
99 | #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */ |
100 | ||
101 | static inline void task_group_account_field(struct task_struct *p, int index, | |
102 | u64 tmp) | |
103 | { | |
73fbec60 FW |
104 | /* |
105 | * Since all updates are sure to touch the root cgroup, we | |
106 | * get ourselves ahead and touch it first. If the root cgroup | |
107 | * is the only cgroup, then nothing else should be necessary. | |
108 | * | |
109 | */ | |
a4f61cc0 | 110 | __this_cpu_add(kernel_cpustat.cpustat[index], tmp); |
73fbec60 | 111 | |
d2cc5ed6 | 112 | cgroup_account_cputime_field(p, index, tmp); |
73fbec60 FW |
113 | } |
114 | ||
115 | /* | |
116 | * Account user cpu time to a process. | |
117 | * @p: the process that the cpu time gets accounted to | |
118 | * @cputime: the cpu time spent in user space since the last update | |
73fbec60 | 119 | */ |
23244a5c | 120 | void account_user_time(struct task_struct *p, u64 cputime) |
73fbec60 FW |
121 | { |
122 | int index; | |
123 | ||
124 | /* Add user time to process. */ | |
23244a5c FW |
125 | p->utime += cputime; |
126 | account_group_user_time(p, cputime); | |
73fbec60 | 127 | |
d0ea0268 | 128 | index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
73fbec60 FW |
129 | |
130 | /* Add user time to cpustat. */ | |
23244a5c | 131 | task_group_account_field(p, index, cputime); |
73fbec60 FW |
132 | |
133 | /* Account for user time used */ | |
6fac4829 | 134 | acct_account_cputime(p); |
73fbec60 FW |
135 | } |
136 | ||
137 | /* | |
138 | * Account guest cpu time to a process. | |
139 | * @p: the process that the cpu time gets accounted to | |
140 | * @cputime: the cpu time spent in virtual machine since the last update | |
73fbec60 | 141 | */ |
fb8b049c | 142 | void account_guest_time(struct task_struct *p, u64 cputime) |
73fbec60 FW |
143 | { |
144 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
145 | ||
146 | /* Add guest time to process. */ | |
fb8b049c FW |
147 | p->utime += cputime; |
148 | account_group_user_time(p, cputime); | |
149 | p->gtime += cputime; | |
73fbec60 FW |
150 | |
151 | /* Add guest time to cpustat. */ | |
d0ea0268 | 152 | if (task_nice(p) > 0) { |
fb8b049c FW |
153 | cpustat[CPUTIME_NICE] += cputime; |
154 | cpustat[CPUTIME_GUEST_NICE] += cputime; | |
73fbec60 | 155 | } else { |
fb8b049c FW |
156 | cpustat[CPUTIME_USER] += cputime; |
157 | cpustat[CPUTIME_GUEST] += cputime; | |
73fbec60 FW |
158 | } |
159 | } | |
160 | ||
161 | /* | |
162 | * Account system cpu time to a process and desired cpustat field | |
163 | * @p: the process that the cpu time gets accounted to | |
164 | * @cputime: the cpu time spent in kernel space since the last update | |
40565b5a | 165 | * @index: pointer to cpustat field that has to be updated |
73fbec60 | 166 | */ |
c31cc6a5 | 167 | void account_system_index_time(struct task_struct *p, |
fb8b049c | 168 | u64 cputime, enum cpu_usage_stat index) |
73fbec60 FW |
169 | { |
170 | /* Add system time to process. */ | |
fb8b049c FW |
171 | p->stime += cputime; |
172 | account_group_system_time(p, cputime); | |
73fbec60 FW |
173 | |
174 | /* Add system time to cpustat. */ | |
fb8b049c | 175 | task_group_account_field(p, index, cputime); |
73fbec60 FW |
176 | |
177 | /* Account for system time used */ | |
6fac4829 | 178 | acct_account_cputime(p); |
73fbec60 FW |
179 | } |
180 | ||
181 | /* | |
182 | * Account system cpu time to a process. | |
183 | * @p: the process that the cpu time gets accounted to | |
184 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
185 | * @cputime: the cpu time spent in kernel space since the last update | |
73fbec60 | 186 | */ |
fb8b049c | 187 | void account_system_time(struct task_struct *p, int hardirq_offset, u64 cputime) |
73fbec60 FW |
188 | { |
189 | int index; | |
190 | ||
191 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { | |
40565b5a | 192 | account_guest_time(p, cputime); |
73fbec60 FW |
193 | return; |
194 | } | |
195 | ||
196 | if (hardirq_count() - hardirq_offset) | |
197 | index = CPUTIME_IRQ; | |
198 | else if (in_serving_softirq()) | |
199 | index = CPUTIME_SOFTIRQ; | |
200 | else | |
201 | index = CPUTIME_SYSTEM; | |
202 | ||
c31cc6a5 | 203 | account_system_index_time(p, cputime, index); |
73fbec60 FW |
204 | } |
205 | ||
206 | /* | |
207 | * Account for involuntary wait time. | |
208 | * @cputime: the cpu time spent in involuntary wait | |
209 | */ | |
be9095ed | 210 | void account_steal_time(u64 cputime) |
73fbec60 FW |
211 | { |
212 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
213 | ||
be9095ed | 214 | cpustat[CPUTIME_STEAL] += cputime; |
73fbec60 FW |
215 | } |
216 | ||
217 | /* | |
218 | * Account for idle time. | |
219 | * @cputime: the cpu time spent in idle wait | |
220 | */ | |
18b43a9b | 221 | void account_idle_time(u64 cputime) |
73fbec60 FW |
222 | { |
223 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
224 | struct rq *rq = this_rq(); | |
225 | ||
226 | if (atomic_read(&rq->nr_iowait) > 0) | |
18b43a9b | 227 | cpustat[CPUTIME_IOWAIT] += cputime; |
73fbec60 | 228 | else |
18b43a9b | 229 | cpustat[CPUTIME_IDLE] += cputime; |
73fbec60 FW |
230 | } |
231 | ||
03cbc732 WL |
232 | /* |
233 | * When a guest is interrupted for a longer amount of time, missed clock | |
234 | * ticks are not redelivered later. Due to that, this function may on | |
235 | * occasion account more time than the calling functions think elapsed. | |
236 | */ | |
2b1f967d | 237 | static __always_inline u64 steal_account_process_time(u64 maxtime) |
73fbec60 FW |
238 | { |
239 | #ifdef CONFIG_PARAVIRT | |
240 | if (static_key_false(¶virt_steal_enabled)) { | |
2b1f967d | 241 | u64 steal; |
73fbec60 FW |
242 | |
243 | steal = paravirt_steal_clock(smp_processor_id()); | |
244 | steal -= this_rq()->prev_steal_time; | |
2b1f967d FW |
245 | steal = min(steal, maxtime); |
246 | account_steal_time(steal); | |
247 | this_rq()->prev_steal_time += steal; | |
73fbec60 | 248 | |
2b1f967d | 249 | return steal; |
73fbec60 FW |
250 | } |
251 | #endif | |
807e5b80 | 252 | return 0; |
73fbec60 FW |
253 | } |
254 | ||
57430218 RR |
255 | /* |
256 | * Account how much elapsed time was spent in steal, irq, or softirq time. | |
257 | */ | |
2b1f967d | 258 | static inline u64 account_other_time(u64 max) |
57430218 | 259 | { |
2b1f967d | 260 | u64 accounted; |
57430218 | 261 | |
2c11dba0 | 262 | lockdep_assert_irqs_disabled(); |
2810f611 | 263 | |
57430218 RR |
264 | accounted = steal_account_process_time(max); |
265 | ||
266 | if (accounted < max) | |
a499a5a1 | 267 | accounted += irqtime_tick_accounted(max - accounted); |
57430218 RR |
268 | |
269 | return accounted; | |
270 | } | |
271 | ||
a1eb1411 SG |
272 | #ifdef CONFIG_64BIT |
273 | static inline u64 read_sum_exec_runtime(struct task_struct *t) | |
274 | { | |
275 | return t->se.sum_exec_runtime; | |
276 | } | |
277 | #else | |
278 | static u64 read_sum_exec_runtime(struct task_struct *t) | |
279 | { | |
280 | u64 ns; | |
281 | struct rq_flags rf; | |
282 | struct rq *rq; | |
283 | ||
284 | rq = task_rq_lock(t, &rf); | |
285 | ns = t->se.sum_exec_runtime; | |
286 | task_rq_unlock(rq, t, &rf); | |
287 | ||
288 | return ns; | |
289 | } | |
290 | #endif | |
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; | |
5613fda9 | 299 | u64 utime, stime; |
a634f933 | 300 | struct task_struct *t; |
e78c3496 | 301 | unsigned int seq, nextseq; |
9c368b5b | 302 | unsigned long flags; |
a634f933 | 303 | |
a1eb1411 SG |
304 | /* |
305 | * Update current task runtime to account pending time since last | |
306 | * scheduler action or thread_group_cputime() call. This thread group | |
307 | * might have other running tasks on different CPUs, but updating | |
308 | * their runtime can affect syscall performance, so we skip account | |
309 | * those pending times and rely only on values updated on tick or | |
310 | * other scheduler action. | |
311 | */ | |
312 | if (same_thread_group(current, tsk)) | |
313 | (void) task_sched_runtime(current); | |
314 | ||
a634f933 | 315 | rcu_read_lock(); |
e78c3496 RR |
316 | /* Attempt a lockless read on the first round. */ |
317 | nextseq = 0; | |
318 | do { | |
319 | seq = nextseq; | |
9c368b5b | 320 | flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq); |
e78c3496 RR |
321 | times->utime = sig->utime; |
322 | times->stime = sig->stime; | |
323 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
324 | ||
325 | for_each_thread(tsk, t) { | |
326 | task_cputime(t, &utime, &stime); | |
327 | times->utime += utime; | |
328 | times->stime += stime; | |
a1eb1411 | 329 | times->sum_exec_runtime += read_sum_exec_runtime(t); |
e78c3496 RR |
330 | } |
331 | /* If lockless access failed, take the lock. */ | |
332 | nextseq = 1; | |
333 | } while (need_seqretry(&sig->stats_lock, seq)); | |
9c368b5b | 334 | done_seqretry_irqrestore(&sig->stats_lock, seq, flags); |
a634f933 FW |
335 | rcu_read_unlock(); |
336 | } | |
337 | ||
73fbec60 FW |
338 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
339 | /* | |
340 | * Account a tick to a process and cpustat | |
341 | * @p: the process that the cpu time gets accounted to | |
342 | * @user_tick: is the tick from userspace | |
343 | * @rq: the pointer to rq | |
344 | * | |
345 | * Tick demultiplexing follows the order | |
346 | * - pending hardirq update | |
347 | * - pending softirq update | |
348 | * - user_time | |
349 | * - idle_time | |
350 | * - system time | |
351 | * - check for guest_time | |
352 | * - else account as system_time | |
353 | * | |
354 | * Check for hardirq is done both for system and user time as there is | |
355 | * no timer going off while we are on hardirq and hence we may never get an | |
356 | * opportunity to update it solely in system time. | |
357 | * p->stime and friends are only updated on system time and not on irq | |
358 | * softirq as those do not count in task exec_runtime any more. | |
359 | */ | |
360 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 361 | struct rq *rq, int ticks) |
73fbec60 | 362 | { |
2b1f967d | 363 | u64 other, cputime = TICK_NSEC * ticks; |
73fbec60 | 364 | |
57430218 RR |
365 | /* |
366 | * When returning from idle, many ticks can get accounted at | |
367 | * once, including some ticks of steal, irq, and softirq time. | |
368 | * Subtract those ticks from the amount of time accounted to | |
369 | * idle, or potentially user or system time. Due to rounding, | |
370 | * other time can exceed ticks occasionally. | |
371 | */ | |
03cbc732 | 372 | other = account_other_time(ULONG_MAX); |
2b1f967d | 373 | if (other >= cputime) |
73fbec60 | 374 | return; |
23244a5c | 375 | |
2b1f967d | 376 | cputime -= other; |
73fbec60 | 377 | |
57430218 | 378 | if (this_cpu_ksoftirqd() == p) { |
73fbec60 FW |
379 | /* |
380 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
381 | * So, we have to handle it separately here. | |
382 | * Also, p->stime needs to be updated for ksoftirqd. | |
383 | */ | |
fb8b049c | 384 | account_system_index_time(p, cputime, CPUTIME_SOFTIRQ); |
73fbec60 | 385 | } else if (user_tick) { |
40565b5a | 386 | account_user_time(p, cputime); |
73fbec60 | 387 | } else if (p == rq->idle) { |
18b43a9b | 388 | account_idle_time(cputime); |
73fbec60 | 389 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ |
fb8b049c | 390 | account_guest_time(p, cputime); |
73fbec60 | 391 | } else { |
fb8b049c | 392 | account_system_index_time(p, cputime, CPUTIME_SYSTEM); |
73fbec60 FW |
393 | } |
394 | } | |
395 | ||
396 | static void irqtime_account_idle_ticks(int ticks) | |
397 | { | |
73fbec60 FW |
398 | struct rq *rq = this_rq(); |
399 | ||
2d513868 | 400 | irqtime_account_process_tick(current, 0, rq, ticks); |
73fbec60 FW |
401 | } |
402 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
3f4724ea FW |
403 | static inline void irqtime_account_idle_ticks(int ticks) {} |
404 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 405 | struct rq *rq, int nr_ticks) {} |
73fbec60 FW |
406 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
407 | ||
73fbec60 FW |
408 | /* |
409 | * Use precise platform statistics if available: | |
410 | */ | |
411 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
a7e1a9e3 | 412 | |
e3942ba0 | 413 | #ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
b0493406 | 414 | void vtime_common_task_switch(struct task_struct *prev) |
e3942ba0 FW |
415 | { |
416 | if (is_idle_task(prev)) | |
417 | vtime_account_idle(prev); | |
418 | else | |
419 | vtime_account_system(prev); | |
420 | ||
c8d7dabf | 421 | vtime_flush(prev); |
e3942ba0 FW |
422 | arch_vtime_task_switch(prev); |
423 | } | |
424 | #endif | |
11113334 | 425 | |
0cfdf9a1 FW |
426 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ |
427 | ||
428 | ||
429 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE | |
a7e1a9e3 FW |
430 | /* |
431 | * Archs that account the whole time spent in the idle task | |
432 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 433 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
434 | * have other meaning of the idle time (s390 only includes the |
435 | * time spent by the CPU when it's in low power mode) must override | |
436 | * vtime_account(). | |
437 | */ | |
438 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
0cfdf9a1 | 439 | void vtime_account_irq_enter(struct task_struct *tsk) |
a7e1a9e3 | 440 | { |
0cfdf9a1 FW |
441 | if (!in_interrupt() && is_idle_task(tsk)) |
442 | vtime_account_idle(tsk); | |
443 | else | |
444 | vtime_account_system(tsk); | |
a7e1a9e3 | 445 | } |
0cfdf9a1 | 446 | EXPORT_SYMBOL_GPL(vtime_account_irq_enter); |
a7e1a9e3 | 447 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ |
9fbc42ea | 448 | |
8157a7fa TH |
449 | void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev, |
450 | u64 *ut, u64 *st) | |
451 | { | |
452 | *ut = curr->utime; | |
453 | *st = curr->stime; | |
454 | } | |
455 | ||
5613fda9 | 456 | void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
9fbc42ea FW |
457 | { |
458 | *ut = p->utime; | |
459 | *st = p->stime; | |
460 | } | |
9eec50b8 | 461 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
a7e1a9e3 | 462 | |
5613fda9 | 463 | void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
9fbc42ea FW |
464 | { |
465 | struct task_cputime cputime; | |
73fbec60 | 466 | |
9fbc42ea FW |
467 | thread_group_cputime(p, &cputime); |
468 | ||
469 | *ut = cputime.utime; | |
470 | *st = cputime.stime; | |
471 | } | |
472 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ | |
473 | /* | |
474 | * Account a single tick of cpu time. | |
475 | * @p: the process that the cpu time gets accounted to | |
476 | * @user_tick: indicates if the tick is a user or a system tick | |
477 | */ | |
478 | void account_process_tick(struct task_struct *p, int user_tick) | |
73fbec60 | 479 | { |
2b1f967d | 480 | u64 cputime, steal; |
9fbc42ea | 481 | struct rq *rq = this_rq(); |
73fbec60 | 482 | |
55dbdcfa | 483 | if (vtime_accounting_cpu_enabled()) |
9fbc42ea FW |
484 | return; |
485 | ||
486 | if (sched_clock_irqtime) { | |
2d513868 | 487 | irqtime_account_process_tick(p, user_tick, rq, 1); |
9fbc42ea FW |
488 | return; |
489 | } | |
490 | ||
2b1f967d | 491 | cputime = TICK_NSEC; |
03cbc732 | 492 | steal = steal_account_process_time(ULONG_MAX); |
57430218 | 493 | |
2b1f967d | 494 | if (steal >= cputime) |
9fbc42ea | 495 | return; |
73fbec60 | 496 | |
2b1f967d | 497 | cputime -= steal; |
57430218 | 498 | |
9fbc42ea | 499 | if (user_tick) |
40565b5a | 500 | account_user_time(p, cputime); |
9fbc42ea | 501 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
fb8b049c | 502 | account_system_time(p, HARDIRQ_OFFSET, cputime); |
73fbec60 | 503 | else |
18b43a9b | 504 | account_idle_time(cputime); |
9fbc42ea | 505 | } |
73fbec60 | 506 | |
9fbc42ea FW |
507 | /* |
508 | * Account multiple ticks of idle time. | |
509 | * @ticks: number of stolen ticks | |
510 | */ | |
511 | void account_idle_ticks(unsigned long ticks) | |
512 | { | |
18b43a9b | 513 | u64 cputime, steal; |
26f2c75c | 514 | |
9fbc42ea FW |
515 | if (sched_clock_irqtime) { |
516 | irqtime_account_idle_ticks(ticks); | |
517 | return; | |
518 | } | |
519 | ||
18b43a9b | 520 | cputime = ticks * TICK_NSEC; |
2b1f967d | 521 | steal = steal_account_process_time(ULONG_MAX); |
f9bcf1e0 WL |
522 | |
523 | if (steal >= cputime) | |
524 | return; | |
525 | ||
526 | cputime -= steal; | |
527 | account_idle_time(cputime); | |
9fbc42ea | 528 | } |
73fbec60 | 529 | |
d9a3c982 | 530 | /* |
55eaa7c1 SG |
531 | * Perform (stime * rtime) / total, but avoid multiplication overflow by |
532 | * loosing precision when the numbers are big. | |
d9a3c982 | 533 | */ |
5613fda9 | 534 | static u64 scale_stime(u64 stime, u64 rtime, u64 total) |
73fbec60 | 535 | { |
55eaa7c1 | 536 | u64 scaled; |
73fbec60 | 537 | |
55eaa7c1 SG |
538 | for (;;) { |
539 | /* Make sure "rtime" is the bigger of stime/rtime */ | |
84f9f3a1 SG |
540 | if (stime > rtime) |
541 | swap(rtime, stime); | |
55eaa7c1 SG |
542 | |
543 | /* Make sure 'total' fits in 32 bits */ | |
544 | if (total >> 32) | |
545 | goto drop_precision; | |
546 | ||
547 | /* Does rtime (and thus stime) fit in 32 bits? */ | |
548 | if (!(rtime >> 32)) | |
549 | break; | |
550 | ||
551 | /* Can we just balance rtime/stime rather than dropping bits? */ | |
552 | if (stime >> 31) | |
553 | goto drop_precision; | |
554 | ||
555 | /* We can grow stime and shrink rtime and try to make them both fit */ | |
556 | stime <<= 1; | |
557 | rtime >>= 1; | |
558 | continue; | |
559 | ||
560 | drop_precision: | |
561 | /* We drop from rtime, it has more bits than stime */ | |
562 | rtime >>= 1; | |
563 | total >>= 1; | |
d9a3c982 | 564 | } |
73fbec60 | 565 | |
55eaa7c1 SG |
566 | /* |
567 | * Make sure gcc understands that this is a 32x32->64 multiply, | |
568 | * followed by a 64/32->64 divide. | |
569 | */ | |
570 | scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total); | |
5613fda9 | 571 | return scaled; |
73fbec60 FW |
572 | } |
573 | ||
347abad9 | 574 | /* |
9d7fb042 PZ |
575 | * Adjust tick based cputime random precision against scheduler runtime |
576 | * accounting. | |
347abad9 | 577 | * |
9d7fb042 PZ |
578 | * Tick based cputime accounting depend on random scheduling timeslices of a |
579 | * task to be interrupted or not by the timer. Depending on these | |
580 | * circumstances, the number of these interrupts may be over or | |
581 | * under-optimistic, matching the real user and system cputime with a variable | |
582 | * precision. | |
583 | * | |
584 | * Fix this by scaling these tick based values against the total runtime | |
585 | * accounted by the CFS scheduler. | |
586 | * | |
587 | * This code provides the following guarantees: | |
588 | * | |
589 | * stime + utime == rtime | |
590 | * stime_i+1 >= stime_i, utime_i+1 >= utime_i | |
591 | * | |
592 | * Assuming that rtime_i+1 >= rtime_i. | |
fa092057 | 593 | */ |
cfb766da TH |
594 | void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev, |
595 | u64 *ut, u64 *st) | |
73fbec60 | 596 | { |
5613fda9 | 597 | u64 rtime, stime, utime; |
9d7fb042 | 598 | unsigned long flags; |
fa092057 | 599 | |
9d7fb042 PZ |
600 | /* Serialize concurrent callers such that we can honour our guarantees */ |
601 | raw_spin_lock_irqsave(&prev->lock, flags); | |
5613fda9 | 602 | rtime = curr->sum_exec_runtime; |
73fbec60 | 603 | |
772c808a | 604 | /* |
9d7fb042 PZ |
605 | * This is possible under two circumstances: |
606 | * - rtime isn't monotonic after all (a bug); | |
607 | * - we got reordered by the lock. | |
608 | * | |
609 | * In both cases this acts as a filter such that the rest of the code | |
610 | * can assume it is monotonic regardless of anything else. | |
772c808a SG |
611 | */ |
612 | if (prev->stime + prev->utime >= rtime) | |
613 | goto out; | |
614 | ||
5a8e01f8 SG |
615 | stime = curr->stime; |
616 | utime = curr->utime; | |
617 | ||
173be9a1 | 618 | /* |
3b9c08ae IM |
619 | * If either stime or utime are 0, assume all runtime is userspace. |
620 | * Once a task gets some ticks, the monotonicy code at 'update:' | |
621 | * will ensure things converge to the observed ratio. | |
173be9a1 | 622 | */ |
3b9c08ae IM |
623 | if (stime == 0) { |
624 | utime = rtime; | |
625 | goto update; | |
9d7fb042 | 626 | } |
5a8e01f8 | 627 | |
3b9c08ae IM |
628 | if (utime == 0) { |
629 | stime = rtime; | |
630 | goto update; | |
631 | } | |
632 | ||
633 | stime = scale_stime(stime, rtime, stime + utime); | |
634 | ||
635 | update: | |
9d7fb042 PZ |
636 | /* |
637 | * Make sure stime doesn't go backwards; this preserves monotonicity | |
638 | * for utime because rtime is monotonic. | |
639 | * | |
640 | * utime_i+1 = rtime_i+1 - stime_i | |
641 | * = rtime_i+1 - (rtime_i - utime_i) | |
642 | * = (rtime_i+1 - rtime_i) + utime_i | |
643 | * >= utime_i | |
644 | */ | |
645 | if (stime < prev->stime) | |
646 | stime = prev->stime; | |
647 | utime = rtime - stime; | |
648 | ||
649 | /* | |
650 | * Make sure utime doesn't go backwards; this still preserves | |
651 | * monotonicity for stime, analogous argument to above. | |
652 | */ | |
653 | if (utime < prev->utime) { | |
654 | utime = prev->utime; | |
655 | stime = rtime - utime; | |
656 | } | |
d37f761d | 657 | |
9d7fb042 PZ |
658 | prev->stime = stime; |
659 | prev->utime = utime; | |
772c808a | 660 | out: |
d37f761d FW |
661 | *ut = prev->utime; |
662 | *st = prev->stime; | |
9d7fb042 | 663 | raw_spin_unlock_irqrestore(&prev->lock, flags); |
d37f761d | 664 | } |
73fbec60 | 665 | |
5613fda9 | 666 | void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
d37f761d FW |
667 | { |
668 | struct task_cputime cputime = { | |
d37f761d FW |
669 | .sum_exec_runtime = p->se.sum_exec_runtime, |
670 | }; | |
671 | ||
6fac4829 | 672 | task_cputime(p, &cputime.utime, &cputime.stime); |
d37f761d | 673 | cputime_adjust(&cputime, &p->prev_cputime, ut, st); |
73fbec60 | 674 | } |
9eec50b8 | 675 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
73fbec60 | 676 | |
5613fda9 | 677 | void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
73fbec60 | 678 | { |
73fbec60 | 679 | struct task_cputime cputime; |
73fbec60 FW |
680 | |
681 | thread_group_cputime(p, &cputime); | |
d37f761d | 682 | cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); |
73fbec60 | 683 | } |
9fbc42ea | 684 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ |
abf917cd FW |
685 | |
686 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN | |
bac5b6b6 | 687 | static u64 vtime_delta(struct vtime *vtime) |
6a61671b | 688 | { |
2a42eb95 | 689 | unsigned long long clock; |
6a61671b | 690 | |
0e4097c3 | 691 | clock = sched_clock(); |
2a42eb95 | 692 | if (clock < vtime->starttime) |
6a61671b | 693 | return 0; |
abf917cd | 694 | |
2a42eb95 | 695 | return clock - vtime->starttime; |
6a61671b FW |
696 | } |
697 | ||
bac5b6b6 | 698 | static u64 get_vtime_delta(struct vtime *vtime) |
abf917cd | 699 | { |
2a42eb95 WL |
700 | u64 delta = vtime_delta(vtime); |
701 | u64 other; | |
abf917cd | 702 | |
03cbc732 WL |
703 | /* |
704 | * Unlike tick based timing, vtime based timing never has lost | |
705 | * ticks, and no need for steal time accounting to make up for | |
706 | * lost ticks. Vtime accounts a rounded version of actual | |
707 | * elapsed time. Limit account_other_time to prevent rounding | |
708 | * errors from causing elapsed vtime to go negative. | |
709 | */ | |
b58c3584 | 710 | other = account_other_time(delta); |
bac5b6b6 | 711 | WARN_ON_ONCE(vtime->state == VTIME_INACTIVE); |
2a42eb95 | 712 | vtime->starttime += delta; |
abf917cd | 713 | |
b58c3584 | 714 | return delta - other; |
abf917cd FW |
715 | } |
716 | ||
2a42eb95 WL |
717 | static void __vtime_account_system(struct task_struct *tsk, |
718 | struct vtime *vtime) | |
6a61671b | 719 | { |
2a42eb95 WL |
720 | vtime->stime += get_vtime_delta(vtime); |
721 | if (vtime->stime >= TICK_NSEC) { | |
722 | account_system_time(tsk, irq_count(), vtime->stime); | |
723 | vtime->stime = 0; | |
724 | } | |
725 | } | |
726 | ||
727 | static void vtime_account_guest(struct task_struct *tsk, | |
728 | struct vtime *vtime) | |
729 | { | |
730 | vtime->gtime += get_vtime_delta(vtime); | |
731 | if (vtime->gtime >= TICK_NSEC) { | |
732 | account_guest_time(tsk, vtime->gtime); | |
733 | vtime->gtime = 0; | |
734 | } | |
6a61671b FW |
735 | } |
736 | ||
abf917cd FW |
737 | void vtime_account_system(struct task_struct *tsk) |
738 | { | |
bac5b6b6 FW |
739 | struct vtime *vtime = &tsk->vtime; |
740 | ||
741 | if (!vtime_delta(vtime)) | |
ff9a9b4c RR |
742 | return; |
743 | ||
bac5b6b6 | 744 | write_seqcount_begin(&vtime->seqcount); |
2a42eb95 WL |
745 | /* We might have scheduled out from guest path */ |
746 | if (current->flags & PF_VCPU) | |
747 | vtime_account_guest(tsk, vtime); | |
748 | else | |
749 | __vtime_account_system(tsk, vtime); | |
bac5b6b6 | 750 | write_seqcount_end(&vtime->seqcount); |
6a61671b | 751 | } |
3f4724ea | 752 | |
1c3eda01 | 753 | void vtime_user_enter(struct task_struct *tsk) |
abf917cd | 754 | { |
bac5b6b6 FW |
755 | struct vtime *vtime = &tsk->vtime; |
756 | ||
757 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 | 758 | __vtime_account_system(tsk, vtime); |
bac5b6b6 FW |
759 | vtime->state = VTIME_USER; |
760 | write_seqcount_end(&vtime->seqcount); | |
6a61671b FW |
761 | } |
762 | ||
1c3eda01 | 763 | void vtime_user_exit(struct task_struct *tsk) |
6a61671b | 764 | { |
bac5b6b6 FW |
765 | struct vtime *vtime = &tsk->vtime; |
766 | ||
767 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 WL |
768 | vtime->utime += get_vtime_delta(vtime); |
769 | if (vtime->utime >= TICK_NSEC) { | |
770 | account_user_time(tsk, vtime->utime); | |
771 | vtime->utime = 0; | |
772 | } | |
bac5b6b6 FW |
773 | vtime->state = VTIME_SYS; |
774 | write_seqcount_end(&vtime->seqcount); | |
6a61671b FW |
775 | } |
776 | ||
777 | void vtime_guest_enter(struct task_struct *tsk) | |
778 | { | |
bac5b6b6 | 779 | struct vtime *vtime = &tsk->vtime; |
5b206d48 FW |
780 | /* |
781 | * The flags must be updated under the lock with | |
60a9ce57 | 782 | * the vtime_starttime flush and update. |
5b206d48 FW |
783 | * That enforces a right ordering and update sequence |
784 | * synchronization against the reader (task_gtime()) | |
785 | * that can thus safely catch up with a tickless delta. | |
786 | */ | |
bac5b6b6 | 787 | write_seqcount_begin(&vtime->seqcount); |
2a42eb95 | 788 | __vtime_account_system(tsk, vtime); |
6a61671b | 789 | current->flags |= PF_VCPU; |
bac5b6b6 | 790 | write_seqcount_end(&vtime->seqcount); |
6a61671b | 791 | } |
48d6a816 | 792 | EXPORT_SYMBOL_GPL(vtime_guest_enter); |
6a61671b FW |
793 | |
794 | void vtime_guest_exit(struct task_struct *tsk) | |
795 | { | |
bac5b6b6 FW |
796 | struct vtime *vtime = &tsk->vtime; |
797 | ||
798 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 | 799 | vtime_account_guest(tsk, vtime); |
6a61671b | 800 | current->flags &= ~PF_VCPU; |
bac5b6b6 | 801 | write_seqcount_end(&vtime->seqcount); |
abf917cd | 802 | } |
48d6a816 | 803 | EXPORT_SYMBOL_GPL(vtime_guest_exit); |
abf917cd FW |
804 | |
805 | void vtime_account_idle(struct task_struct *tsk) | |
806 | { | |
bac5b6b6 | 807 | account_idle_time(get_vtime_delta(&tsk->vtime)); |
abf917cd | 808 | } |
3f4724ea | 809 | |
6a61671b FW |
810 | void arch_vtime_task_switch(struct task_struct *prev) |
811 | { | |
bac5b6b6 | 812 | struct vtime *vtime = &prev->vtime; |
6a61671b | 813 | |
bac5b6b6 FW |
814 | write_seqcount_begin(&vtime->seqcount); |
815 | vtime->state = VTIME_INACTIVE; | |
816 | write_seqcount_end(&vtime->seqcount); | |
817 | ||
818 | vtime = ¤t->vtime; | |
819 | ||
820 | write_seqcount_begin(&vtime->seqcount); | |
821 | vtime->state = VTIME_SYS; | |
0e4097c3 | 822 | vtime->starttime = sched_clock(); |
bac5b6b6 | 823 | write_seqcount_end(&vtime->seqcount); |
6a61671b FW |
824 | } |
825 | ||
45eacc69 | 826 | void vtime_init_idle(struct task_struct *t, int cpu) |
6a61671b | 827 | { |
bac5b6b6 | 828 | struct vtime *vtime = &t->vtime; |
6a61671b FW |
829 | unsigned long flags; |
830 | ||
b7ce2277 | 831 | local_irq_save(flags); |
bac5b6b6 FW |
832 | write_seqcount_begin(&vtime->seqcount); |
833 | vtime->state = VTIME_SYS; | |
0e4097c3 | 834 | vtime->starttime = sched_clock(); |
bac5b6b6 | 835 | write_seqcount_end(&vtime->seqcount); |
b7ce2277 | 836 | local_irq_restore(flags); |
6a61671b FW |
837 | } |
838 | ||
16a6d9be | 839 | u64 task_gtime(struct task_struct *t) |
6a61671b | 840 | { |
bac5b6b6 | 841 | struct vtime *vtime = &t->vtime; |
6a61671b | 842 | unsigned int seq; |
16a6d9be | 843 | u64 gtime; |
6a61671b | 844 | |
e5925394 | 845 | if (!vtime_accounting_enabled()) |
2541117b HS |
846 | return t->gtime; |
847 | ||
6a61671b | 848 | do { |
bac5b6b6 | 849 | seq = read_seqcount_begin(&vtime->seqcount); |
6a61671b FW |
850 | |
851 | gtime = t->gtime; | |
bac5b6b6 | 852 | if (vtime->state == VTIME_SYS && t->flags & PF_VCPU) |
2a42eb95 | 853 | gtime += vtime->gtime + vtime_delta(vtime); |
6a61671b | 854 | |
bac5b6b6 | 855 | } while (read_seqcount_retry(&vtime->seqcount, seq)); |
6a61671b FW |
856 | |
857 | return gtime; | |
858 | } | |
859 | ||
860 | /* | |
861 | * Fetch cputime raw values from fields of task_struct and | |
862 | * add up the pending nohz execution time since the last | |
863 | * cputime snapshot. | |
864 | */ | |
5613fda9 | 865 | void task_cputime(struct task_struct *t, u64 *utime, u64 *stime) |
6a61671b | 866 | { |
bac5b6b6 | 867 | struct vtime *vtime = &t->vtime; |
6a61671b | 868 | unsigned int seq; |
bac5b6b6 | 869 | u64 delta; |
6a61671b | 870 | |
353c50eb SG |
871 | if (!vtime_accounting_enabled()) { |
872 | *utime = t->utime; | |
873 | *stime = t->stime; | |
874 | return; | |
875 | } | |
6a61671b | 876 | |
353c50eb | 877 | do { |
bac5b6b6 | 878 | seq = read_seqcount_begin(&vtime->seqcount); |
6a61671b | 879 | |
353c50eb SG |
880 | *utime = t->utime; |
881 | *stime = t->stime; | |
6a61671b FW |
882 | |
883 | /* Task is sleeping, nothing to add */ | |
bac5b6b6 | 884 | if (vtime->state == VTIME_INACTIVE || is_idle_task(t)) |
6a61671b FW |
885 | continue; |
886 | ||
bac5b6b6 | 887 | delta = vtime_delta(vtime); |
6a61671b FW |
888 | |
889 | /* | |
890 | * Task runs either in user or kernel space, add pending nohz time to | |
891 | * the right place. | |
892 | */ | |
bac5b6b6 | 893 | if (vtime->state == VTIME_USER || t->flags & PF_VCPU) |
2a42eb95 | 894 | *utime += vtime->utime + delta; |
bac5b6b6 | 895 | else if (vtime->state == VTIME_SYS) |
2a42eb95 | 896 | *stime += vtime->stime + delta; |
bac5b6b6 | 897 | } while (read_seqcount_retry(&vtime->seqcount, seq)); |
6a61671b | 898 | } |
abf917cd | 899 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |