]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
325ea10c IM |
2 | /* |
3 | * Simple CPU accounting cgroup controller | |
4 | */ | |
73fbec60 | 5 | #include "sched.h" |
73fbec60 FW |
6 | |
7 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
8 | ||
9 | /* | |
10 | * There are no locks covering percpu hardirq/softirq time. | |
bf9fae9f | 11 | * They are only modified in vtime_account, on corresponding CPU |
73fbec60 FW |
12 | * with interrupts disabled. So, writes are safe. |
13 | * They are read and saved off onto struct rq in update_rq_clock(). | |
14 | * This may result in other CPU reading this CPU's irq time and can | |
bf9fae9f | 15 | * race with irq/vtime_account on this CPU. We would either get old |
73fbec60 FW |
16 | * or new value with a side effect of accounting a slice of irq time to wrong |
17 | * task when irq is in progress while we read rq->clock. That is a worthy | |
18 | * compromise in place of having locks on each irq in account_system_time. | |
19 | */ | |
19d23dbf | 20 | DEFINE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 | 21 | |
73fbec60 FW |
22 | static int sched_clock_irqtime; |
23 | ||
24 | void enable_sched_clock_irqtime(void) | |
25 | { | |
26 | sched_clock_irqtime = 1; | |
27 | } | |
28 | ||
29 | void disable_sched_clock_irqtime(void) | |
30 | { | |
31 | sched_clock_irqtime = 0; | |
32 | } | |
33 | ||
25e2d8c1 FW |
34 | static void irqtime_account_delta(struct irqtime *irqtime, u64 delta, |
35 | enum cpu_usage_stat idx) | |
36 | { | |
37 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
38 | ||
39 | u64_stats_update_begin(&irqtime->sync); | |
40 | cpustat[idx] += delta; | |
41 | irqtime->total += delta; | |
42 | irqtime->tick_delta += delta; | |
43 | u64_stats_update_end(&irqtime->sync); | |
44 | } | |
45 | ||
73fbec60 FW |
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 | { |
19d23dbf | 52 | struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime); |
73fbec60 FW |
53 | s64 delta; |
54 | int cpu; | |
55 | ||
56 | if (!sched_clock_irqtime) | |
57 | return; | |
58 | ||
73fbec60 | 59 | cpu = smp_processor_id(); |
19d23dbf FW |
60 | delta = sched_clock_cpu(cpu) - irqtime->irq_start_time; |
61 | irqtime->irq_start_time += delta; | |
73fbec60 | 62 | |
73fbec60 FW |
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 | */ | |
25e2d8c1 FW |
69 | if (hardirq_count()) |
70 | irqtime_account_delta(irqtime, delta, CPUTIME_IRQ); | |
71 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) | |
72 | irqtime_account_delta(irqtime, delta, CPUTIME_SOFTIRQ); | |
73fbec60 | 73 | } |
3e1df4f5 | 74 | EXPORT_SYMBOL_GPL(irqtime_account_irq); |
73fbec60 | 75 | |
2b1f967d | 76 | static u64 irqtime_tick_accounted(u64 maxtime) |
73fbec60 | 77 | { |
a499a5a1 | 78 | struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime); |
2b1f967d | 79 | u64 delta; |
73fbec60 | 80 | |
2b1f967d FW |
81 | delta = min(irqtime->tick_delta, maxtime); |
82 | irqtime->tick_delta -= delta; | |
2810f611 | 83 | |
a499a5a1 | 84 | return delta; |
73fbec60 FW |
85 | } |
86 | ||
87 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
88 | ||
89 | #define sched_clock_irqtime (0) | |
90 | ||
2b1f967d | 91 | static u64 irqtime_tick_accounted(u64 dummy) |
57430218 RR |
92 | { |
93 | return 0; | |
94 | } | |
95 | ||
73fbec60 FW |
96 | #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */ |
97 | ||
98 | static inline void task_group_account_field(struct task_struct *p, int index, | |
99 | u64 tmp) | |
100 | { | |
73fbec60 FW |
101 | /* |
102 | * Since all updates are sure to touch the root cgroup, we | |
103 | * get ourselves ahead and touch it first. If the root cgroup | |
104 | * is the only cgroup, then nothing else should be necessary. | |
105 | * | |
106 | */ | |
a4f61cc0 | 107 | __this_cpu_add(kernel_cpustat.cpustat[index], tmp); |
73fbec60 | 108 | |
d2cc5ed6 | 109 | cgroup_account_cputime_field(p, index, tmp); |
73fbec60 FW |
110 | } |
111 | ||
112 | /* | |
97fb7a0a IM |
113 | * Account user CPU time to a process. |
114 | * @p: the process that the CPU time gets accounted to | |
115 | * @cputime: the CPU time spent in user space since the last update | |
73fbec60 | 116 | */ |
23244a5c | 117 | void account_user_time(struct task_struct *p, u64 cputime) |
73fbec60 FW |
118 | { |
119 | int index; | |
120 | ||
121 | /* Add user time to process. */ | |
23244a5c FW |
122 | p->utime += cputime; |
123 | account_group_user_time(p, cputime); | |
73fbec60 | 124 | |
d0ea0268 | 125 | index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
73fbec60 FW |
126 | |
127 | /* Add user time to cpustat. */ | |
23244a5c | 128 | task_group_account_field(p, index, cputime); |
73fbec60 FW |
129 | |
130 | /* Account for user time used */ | |
6fac4829 | 131 | acct_account_cputime(p); |
73fbec60 FW |
132 | } |
133 | ||
134 | /* | |
97fb7a0a IM |
135 | * Account guest CPU time to a process. |
136 | * @p: the process that the CPU time gets accounted to | |
137 | * @cputime: the CPU time spent in virtual machine since the last update | |
73fbec60 | 138 | */ |
fb8b049c | 139 | void account_guest_time(struct task_struct *p, u64 cputime) |
73fbec60 FW |
140 | { |
141 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
142 | ||
143 | /* Add guest time to process. */ | |
fb8b049c FW |
144 | p->utime += cputime; |
145 | account_group_user_time(p, cputime); | |
146 | p->gtime += cputime; | |
73fbec60 FW |
147 | |
148 | /* Add guest time to cpustat. */ | |
d0ea0268 | 149 | if (task_nice(p) > 0) { |
fb8b049c FW |
150 | cpustat[CPUTIME_NICE] += cputime; |
151 | cpustat[CPUTIME_GUEST_NICE] += cputime; | |
73fbec60 | 152 | } else { |
fb8b049c FW |
153 | cpustat[CPUTIME_USER] += cputime; |
154 | cpustat[CPUTIME_GUEST] += cputime; | |
73fbec60 FW |
155 | } |
156 | } | |
157 | ||
158 | /* | |
97fb7a0a IM |
159 | * Account system CPU time to a process and desired cpustat field |
160 | * @p: the process that the CPU time gets accounted to | |
161 | * @cputime: the CPU time spent in kernel space since the last update | |
40565b5a | 162 | * @index: pointer to cpustat field that has to be updated |
73fbec60 | 163 | */ |
c31cc6a5 | 164 | void account_system_index_time(struct task_struct *p, |
fb8b049c | 165 | u64 cputime, enum cpu_usage_stat index) |
73fbec60 FW |
166 | { |
167 | /* Add system time to process. */ | |
fb8b049c FW |
168 | p->stime += cputime; |
169 | account_group_system_time(p, cputime); | |
73fbec60 FW |
170 | |
171 | /* Add system time to cpustat. */ | |
fb8b049c | 172 | task_group_account_field(p, index, cputime); |
73fbec60 FW |
173 | |
174 | /* Account for system time used */ | |
6fac4829 | 175 | acct_account_cputime(p); |
73fbec60 FW |
176 | } |
177 | ||
178 | /* | |
97fb7a0a IM |
179 | * Account system CPU time to a process. |
180 | * @p: the process that the CPU time gets accounted to | |
73fbec60 | 181 | * @hardirq_offset: the offset to subtract from hardirq_count() |
97fb7a0a | 182 | * @cputime: the CPU time spent in kernel space since the last update |
73fbec60 | 183 | */ |
fb8b049c | 184 | void account_system_time(struct task_struct *p, int hardirq_offset, u64 cputime) |
73fbec60 FW |
185 | { |
186 | int index; | |
187 | ||
188 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { | |
40565b5a | 189 | account_guest_time(p, cputime); |
73fbec60 FW |
190 | return; |
191 | } | |
192 | ||
193 | if (hardirq_count() - hardirq_offset) | |
194 | index = CPUTIME_IRQ; | |
195 | else if (in_serving_softirq()) | |
196 | index = CPUTIME_SOFTIRQ; | |
197 | else | |
198 | index = CPUTIME_SYSTEM; | |
199 | ||
c31cc6a5 | 200 | account_system_index_time(p, cputime, index); |
73fbec60 FW |
201 | } |
202 | ||
203 | /* | |
204 | * Account for involuntary wait time. | |
97fb7a0a | 205 | * @cputime: the CPU time spent in involuntary wait |
73fbec60 | 206 | */ |
be9095ed | 207 | void account_steal_time(u64 cputime) |
73fbec60 FW |
208 | { |
209 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
210 | ||
be9095ed | 211 | cpustat[CPUTIME_STEAL] += cputime; |
73fbec60 FW |
212 | } |
213 | ||
214 | /* | |
215 | * Account for idle time. | |
97fb7a0a | 216 | * @cputime: the CPU time spent in idle wait |
73fbec60 | 217 | */ |
18b43a9b | 218 | void account_idle_time(u64 cputime) |
73fbec60 FW |
219 | { |
220 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
221 | struct rq *rq = this_rq(); | |
222 | ||
223 | if (atomic_read(&rq->nr_iowait) > 0) | |
18b43a9b | 224 | cpustat[CPUTIME_IOWAIT] += cputime; |
73fbec60 | 225 | else |
18b43a9b | 226 | cpustat[CPUTIME_IDLE] += cputime; |
73fbec60 FW |
227 | } |
228 | ||
03cbc732 WL |
229 | /* |
230 | * When a guest is interrupted for a longer amount of time, missed clock | |
231 | * ticks are not redelivered later. Due to that, this function may on | |
232 | * occasion account more time than the calling functions think elapsed. | |
233 | */ | |
2b1f967d | 234 | static __always_inline u64 steal_account_process_time(u64 maxtime) |
73fbec60 FW |
235 | { |
236 | #ifdef CONFIG_PARAVIRT | |
237 | if (static_key_false(¶virt_steal_enabled)) { | |
2b1f967d | 238 | u64 steal; |
73fbec60 FW |
239 | |
240 | steal = paravirt_steal_clock(smp_processor_id()); | |
241 | steal -= this_rq()->prev_steal_time; | |
2b1f967d FW |
242 | steal = min(steal, maxtime); |
243 | account_steal_time(steal); | |
244 | this_rq()->prev_steal_time += steal; | |
73fbec60 | 245 | |
2b1f967d | 246 | return steal; |
73fbec60 FW |
247 | } |
248 | #endif | |
807e5b80 | 249 | return 0; |
73fbec60 FW |
250 | } |
251 | ||
57430218 RR |
252 | /* |
253 | * Account how much elapsed time was spent in steal, irq, or softirq time. | |
254 | */ | |
2b1f967d | 255 | static inline u64 account_other_time(u64 max) |
57430218 | 256 | { |
2b1f967d | 257 | u64 accounted; |
57430218 | 258 | |
2c11dba0 | 259 | lockdep_assert_irqs_disabled(); |
2810f611 | 260 | |
57430218 RR |
261 | accounted = steal_account_process_time(max); |
262 | ||
263 | if (accounted < max) | |
a499a5a1 | 264 | accounted += irqtime_tick_accounted(max - accounted); |
57430218 RR |
265 | |
266 | return accounted; | |
267 | } | |
268 | ||
a1eb1411 SG |
269 | #ifdef CONFIG_64BIT |
270 | static inline u64 read_sum_exec_runtime(struct task_struct *t) | |
271 | { | |
272 | return t->se.sum_exec_runtime; | |
273 | } | |
274 | #else | |
275 | static u64 read_sum_exec_runtime(struct task_struct *t) | |
276 | { | |
277 | u64 ns; | |
278 | struct rq_flags rf; | |
279 | struct rq *rq; | |
280 | ||
281 | rq = task_rq_lock(t, &rf); | |
282 | ns = t->se.sum_exec_runtime; | |
283 | task_rq_unlock(rq, t, &rf); | |
284 | ||
285 | return ns; | |
286 | } | |
287 | #endif | |
288 | ||
a634f933 FW |
289 | /* |
290 | * Accumulate raw cputime values of dead tasks (sig->[us]time) and live | |
291 | * tasks (sum on group iteration) belonging to @tsk's group. | |
292 | */ | |
293 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) | |
294 | { | |
295 | struct signal_struct *sig = tsk->signal; | |
5613fda9 | 296 | u64 utime, stime; |
a634f933 | 297 | struct task_struct *t; |
e78c3496 | 298 | unsigned int seq, nextseq; |
9c368b5b | 299 | unsigned long flags; |
a634f933 | 300 | |
a1eb1411 SG |
301 | /* |
302 | * Update current task runtime to account pending time since last | |
303 | * scheduler action or thread_group_cputime() call. This thread group | |
304 | * might have other running tasks on different CPUs, but updating | |
305 | * their runtime can affect syscall performance, so we skip account | |
306 | * those pending times and rely only on values updated on tick or | |
307 | * other scheduler action. | |
308 | */ | |
309 | if (same_thread_group(current, tsk)) | |
310 | (void) task_sched_runtime(current); | |
311 | ||
a634f933 | 312 | rcu_read_lock(); |
e78c3496 RR |
313 | /* Attempt a lockless read on the first round. */ |
314 | nextseq = 0; | |
315 | do { | |
316 | seq = nextseq; | |
9c368b5b | 317 | flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq); |
e78c3496 RR |
318 | times->utime = sig->utime; |
319 | times->stime = sig->stime; | |
320 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
321 | ||
322 | for_each_thread(tsk, t) { | |
323 | task_cputime(t, &utime, &stime); | |
324 | times->utime += utime; | |
325 | times->stime += stime; | |
a1eb1411 | 326 | times->sum_exec_runtime += read_sum_exec_runtime(t); |
e78c3496 RR |
327 | } |
328 | /* If lockless access failed, take the lock. */ | |
329 | nextseq = 1; | |
330 | } while (need_seqretry(&sig->stats_lock, seq)); | |
9c368b5b | 331 | done_seqretry_irqrestore(&sig->stats_lock, seq, flags); |
a634f933 FW |
332 | rcu_read_unlock(); |
333 | } | |
334 | ||
73fbec60 FW |
335 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
336 | /* | |
337 | * Account a tick to a process and cpustat | |
97fb7a0a | 338 | * @p: the process that the CPU time gets accounted to |
73fbec60 FW |
339 | * @user_tick: is the tick from userspace |
340 | * @rq: the pointer to rq | |
341 | * | |
342 | * Tick demultiplexing follows the order | |
343 | * - pending hardirq update | |
344 | * - pending softirq update | |
345 | * - user_time | |
346 | * - idle_time | |
347 | * - system time | |
348 | * - check for guest_time | |
349 | * - else account as system_time | |
350 | * | |
351 | * Check for hardirq is done both for system and user time as there is | |
352 | * no timer going off while we are on hardirq and hence we may never get an | |
353 | * opportunity to update it solely in system time. | |
354 | * p->stime and friends are only updated on system time and not on irq | |
355 | * softirq as those do not count in task exec_runtime any more. | |
356 | */ | |
357 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 358 | struct rq *rq, int ticks) |
73fbec60 | 359 | { |
2b1f967d | 360 | u64 other, cputime = TICK_NSEC * ticks; |
73fbec60 | 361 | |
57430218 RR |
362 | /* |
363 | * When returning from idle, many ticks can get accounted at | |
364 | * once, including some ticks of steal, irq, and softirq time. | |
365 | * Subtract those ticks from the amount of time accounted to | |
366 | * idle, or potentially user or system time. Due to rounding, | |
367 | * other time can exceed ticks occasionally. | |
368 | */ | |
03cbc732 | 369 | other = account_other_time(ULONG_MAX); |
2b1f967d | 370 | if (other >= cputime) |
73fbec60 | 371 | return; |
23244a5c | 372 | |
2b1f967d | 373 | cputime -= other; |
73fbec60 | 374 | |
57430218 | 375 | if (this_cpu_ksoftirqd() == p) { |
73fbec60 FW |
376 | /* |
377 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
378 | * So, we have to handle it separately here. | |
379 | * Also, p->stime needs to be updated for ksoftirqd. | |
380 | */ | |
fb8b049c | 381 | account_system_index_time(p, cputime, CPUTIME_SOFTIRQ); |
73fbec60 | 382 | } else if (user_tick) { |
40565b5a | 383 | account_user_time(p, cputime); |
73fbec60 | 384 | } else if (p == rq->idle) { |
18b43a9b | 385 | account_idle_time(cputime); |
73fbec60 | 386 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ |
fb8b049c | 387 | account_guest_time(p, cputime); |
73fbec60 | 388 | } else { |
fb8b049c | 389 | account_system_index_time(p, cputime, CPUTIME_SYSTEM); |
73fbec60 FW |
390 | } |
391 | } | |
392 | ||
393 | static void irqtime_account_idle_ticks(int ticks) | |
394 | { | |
73fbec60 FW |
395 | struct rq *rq = this_rq(); |
396 | ||
2d513868 | 397 | irqtime_account_process_tick(current, 0, rq, ticks); |
73fbec60 FW |
398 | } |
399 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
97fb7a0a | 400 | static inline void irqtime_account_idle_ticks(int ticks) { } |
3f4724ea | 401 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, |
97fb7a0a | 402 | struct rq *rq, int nr_ticks) { } |
73fbec60 FW |
403 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
404 | ||
73fbec60 FW |
405 | /* |
406 | * Use precise platform statistics if available: | |
407 | */ | |
408 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
97fb7a0a | 409 | # ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
b0493406 | 410 | void vtime_common_task_switch(struct task_struct *prev) |
e3942ba0 FW |
411 | { |
412 | if (is_idle_task(prev)) | |
413 | vtime_account_idle(prev); | |
414 | else | |
415 | vtime_account_system(prev); | |
416 | ||
c8d7dabf | 417 | vtime_flush(prev); |
e3942ba0 FW |
418 | arch_vtime_task_switch(prev); |
419 | } | |
97fb7a0a | 420 | # endif |
0cfdf9a1 FW |
421 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ |
422 | ||
423 | ||
424 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE | |
a7e1a9e3 FW |
425 | /* |
426 | * Archs that account the whole time spent in the idle task | |
427 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 428 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
429 | * have other meaning of the idle time (s390 only includes the |
430 | * time spent by the CPU when it's in low power mode) must override | |
431 | * vtime_account(). | |
432 | */ | |
433 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
0cfdf9a1 | 434 | void vtime_account_irq_enter(struct task_struct *tsk) |
a7e1a9e3 | 435 | { |
0cfdf9a1 FW |
436 | if (!in_interrupt() && is_idle_task(tsk)) |
437 | vtime_account_idle(tsk); | |
438 | else | |
439 | vtime_account_system(tsk); | |
a7e1a9e3 | 440 | } |
0cfdf9a1 | 441 | EXPORT_SYMBOL_GPL(vtime_account_irq_enter); |
a7e1a9e3 | 442 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ |
9fbc42ea | 443 | |
8157a7fa TH |
444 | void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev, |
445 | u64 *ut, u64 *st) | |
446 | { | |
447 | *ut = curr->utime; | |
448 | *st = curr->stime; | |
449 | } | |
450 | ||
5613fda9 | 451 | void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
9fbc42ea FW |
452 | { |
453 | *ut = p->utime; | |
454 | *st = p->stime; | |
455 | } | |
9eec50b8 | 456 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
a7e1a9e3 | 457 | |
5613fda9 | 458 | void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
9fbc42ea FW |
459 | { |
460 | struct task_cputime cputime; | |
73fbec60 | 461 | |
9fbc42ea FW |
462 | thread_group_cputime(p, &cputime); |
463 | ||
464 | *ut = cputime.utime; | |
465 | *st = cputime.stime; | |
466 | } | |
97fb7a0a IM |
467 | |
468 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE: */ | |
469 | ||
9fbc42ea | 470 | /* |
97fb7a0a IM |
471 | * Account a single tick of CPU time. |
472 | * @p: the process that the CPU time gets accounted to | |
9fbc42ea FW |
473 | * @user_tick: indicates if the tick is a user or a system tick |
474 | */ | |
475 | void account_process_tick(struct task_struct *p, int user_tick) | |
73fbec60 | 476 | { |
2b1f967d | 477 | u64 cputime, steal; |
9fbc42ea | 478 | struct rq *rq = this_rq(); |
73fbec60 | 479 | |
55dbdcfa | 480 | if (vtime_accounting_cpu_enabled()) |
9fbc42ea FW |
481 | return; |
482 | ||
483 | if (sched_clock_irqtime) { | |
2d513868 | 484 | irqtime_account_process_tick(p, user_tick, rq, 1); |
9fbc42ea FW |
485 | return; |
486 | } | |
487 | ||
2b1f967d | 488 | cputime = TICK_NSEC; |
03cbc732 | 489 | steal = steal_account_process_time(ULONG_MAX); |
57430218 | 490 | |
2b1f967d | 491 | if (steal >= cputime) |
9fbc42ea | 492 | return; |
73fbec60 | 493 | |
2b1f967d | 494 | cputime -= steal; |
57430218 | 495 | |
9fbc42ea | 496 | if (user_tick) |
40565b5a | 497 | account_user_time(p, cputime); |
9fbc42ea | 498 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
fb8b049c | 499 | account_system_time(p, HARDIRQ_OFFSET, cputime); |
73fbec60 | 500 | else |
18b43a9b | 501 | account_idle_time(cputime); |
9fbc42ea | 502 | } |
73fbec60 | 503 | |
9fbc42ea FW |
504 | /* |
505 | * Account multiple ticks of idle time. | |
506 | * @ticks: number of stolen ticks | |
507 | */ | |
508 | void account_idle_ticks(unsigned long ticks) | |
509 | { | |
18b43a9b | 510 | u64 cputime, steal; |
26f2c75c | 511 | |
9fbc42ea FW |
512 | if (sched_clock_irqtime) { |
513 | irqtime_account_idle_ticks(ticks); | |
514 | return; | |
515 | } | |
516 | ||
18b43a9b | 517 | cputime = ticks * TICK_NSEC; |
2b1f967d | 518 | steal = steal_account_process_time(ULONG_MAX); |
f9bcf1e0 WL |
519 | |
520 | if (steal >= cputime) | |
521 | return; | |
522 | ||
523 | cputime -= steal; | |
524 | account_idle_time(cputime); | |
9fbc42ea | 525 | } |
73fbec60 | 526 | |
d9a3c982 | 527 | /* |
55eaa7c1 | 528 | * Perform (stime * rtime) / total, but avoid multiplication overflow by |
dfcb245e | 529 | * losing precision when the numbers are big. |
d9a3c982 | 530 | */ |
5613fda9 | 531 | static u64 scale_stime(u64 stime, u64 rtime, u64 total) |
73fbec60 | 532 | { |
55eaa7c1 | 533 | u64 scaled; |
73fbec60 | 534 | |
55eaa7c1 SG |
535 | for (;;) { |
536 | /* Make sure "rtime" is the bigger of stime/rtime */ | |
84f9f3a1 SG |
537 | if (stime > rtime) |
538 | swap(rtime, stime); | |
55eaa7c1 SG |
539 | |
540 | /* Make sure 'total' fits in 32 bits */ | |
541 | if (total >> 32) | |
542 | goto drop_precision; | |
543 | ||
544 | /* Does rtime (and thus stime) fit in 32 bits? */ | |
545 | if (!(rtime >> 32)) | |
546 | break; | |
547 | ||
548 | /* Can we just balance rtime/stime rather than dropping bits? */ | |
549 | if (stime >> 31) | |
550 | goto drop_precision; | |
551 | ||
552 | /* We can grow stime and shrink rtime and try to make them both fit */ | |
553 | stime <<= 1; | |
554 | rtime >>= 1; | |
555 | continue; | |
556 | ||
557 | drop_precision: | |
558 | /* We drop from rtime, it has more bits than stime */ | |
559 | rtime >>= 1; | |
560 | total >>= 1; | |
d9a3c982 | 561 | } |
73fbec60 | 562 | |
55eaa7c1 SG |
563 | /* |
564 | * Make sure gcc understands that this is a 32x32->64 multiply, | |
565 | * followed by a 64/32->64 divide. | |
566 | */ | |
567 | scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total); | |
5613fda9 | 568 | return scaled; |
73fbec60 FW |
569 | } |
570 | ||
347abad9 | 571 | /* |
9d7fb042 PZ |
572 | * Adjust tick based cputime random precision against scheduler runtime |
573 | * accounting. | |
347abad9 | 574 | * |
9d7fb042 PZ |
575 | * Tick based cputime accounting depend on random scheduling timeslices of a |
576 | * task to be interrupted or not by the timer. Depending on these | |
577 | * circumstances, the number of these interrupts may be over or | |
578 | * under-optimistic, matching the real user and system cputime with a variable | |
579 | * precision. | |
580 | * | |
581 | * Fix this by scaling these tick based values against the total runtime | |
582 | * accounted by the CFS scheduler. | |
583 | * | |
584 | * This code provides the following guarantees: | |
585 | * | |
586 | * stime + utime == rtime | |
587 | * stime_i+1 >= stime_i, utime_i+1 >= utime_i | |
588 | * | |
589 | * Assuming that rtime_i+1 >= rtime_i. | |
fa092057 | 590 | */ |
cfb766da TH |
591 | void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev, |
592 | u64 *ut, u64 *st) | |
73fbec60 | 593 | { |
5613fda9 | 594 | u64 rtime, stime, utime; |
9d7fb042 | 595 | unsigned long flags; |
fa092057 | 596 | |
9d7fb042 PZ |
597 | /* Serialize concurrent callers such that we can honour our guarantees */ |
598 | raw_spin_lock_irqsave(&prev->lock, flags); | |
5613fda9 | 599 | rtime = curr->sum_exec_runtime; |
73fbec60 | 600 | |
772c808a | 601 | /* |
9d7fb042 PZ |
602 | * This is possible under two circumstances: |
603 | * - rtime isn't monotonic after all (a bug); | |
604 | * - we got reordered by the lock. | |
605 | * | |
606 | * In both cases this acts as a filter such that the rest of the code | |
607 | * can assume it is monotonic regardless of anything else. | |
772c808a SG |
608 | */ |
609 | if (prev->stime + prev->utime >= rtime) | |
610 | goto out; | |
611 | ||
5a8e01f8 SG |
612 | stime = curr->stime; |
613 | utime = curr->utime; | |
614 | ||
173be9a1 | 615 | /* |
3b9c08ae IM |
616 | * If either stime or utime are 0, assume all runtime is userspace. |
617 | * Once a task gets some ticks, the monotonicy code at 'update:' | |
618 | * will ensure things converge to the observed ratio. | |
173be9a1 | 619 | */ |
3b9c08ae IM |
620 | if (stime == 0) { |
621 | utime = rtime; | |
622 | goto update; | |
9d7fb042 | 623 | } |
5a8e01f8 | 624 | |
3b9c08ae IM |
625 | if (utime == 0) { |
626 | stime = rtime; | |
627 | goto update; | |
628 | } | |
629 | ||
630 | stime = scale_stime(stime, rtime, stime + utime); | |
631 | ||
632 | update: | |
9d7fb042 PZ |
633 | /* |
634 | * Make sure stime doesn't go backwards; this preserves monotonicity | |
635 | * for utime because rtime is monotonic. | |
636 | * | |
637 | * utime_i+1 = rtime_i+1 - stime_i | |
638 | * = rtime_i+1 - (rtime_i - utime_i) | |
639 | * = (rtime_i+1 - rtime_i) + utime_i | |
640 | * >= utime_i | |
641 | */ | |
642 | if (stime < prev->stime) | |
643 | stime = prev->stime; | |
644 | utime = rtime - stime; | |
645 | ||
646 | /* | |
647 | * Make sure utime doesn't go backwards; this still preserves | |
648 | * monotonicity for stime, analogous argument to above. | |
649 | */ | |
650 | if (utime < prev->utime) { | |
651 | utime = prev->utime; | |
652 | stime = rtime - utime; | |
653 | } | |
d37f761d | 654 | |
9d7fb042 PZ |
655 | prev->stime = stime; |
656 | prev->utime = utime; | |
772c808a | 657 | out: |
d37f761d FW |
658 | *ut = prev->utime; |
659 | *st = prev->stime; | |
9d7fb042 | 660 | raw_spin_unlock_irqrestore(&prev->lock, flags); |
d37f761d | 661 | } |
73fbec60 | 662 | |
5613fda9 | 663 | void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
d37f761d FW |
664 | { |
665 | struct task_cputime cputime = { | |
d37f761d FW |
666 | .sum_exec_runtime = p->se.sum_exec_runtime, |
667 | }; | |
668 | ||
6fac4829 | 669 | task_cputime(p, &cputime.utime, &cputime.stime); |
d37f761d | 670 | cputime_adjust(&cputime, &p->prev_cputime, ut, st); |
73fbec60 | 671 | } |
9eec50b8 | 672 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
73fbec60 | 673 | |
5613fda9 | 674 | void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
73fbec60 | 675 | { |
73fbec60 | 676 | struct task_cputime cputime; |
73fbec60 FW |
677 | |
678 | thread_group_cputime(p, &cputime); | |
d37f761d | 679 | cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); |
73fbec60 | 680 | } |
9fbc42ea | 681 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ |
abf917cd FW |
682 | |
683 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN | |
bac5b6b6 | 684 | static u64 vtime_delta(struct vtime *vtime) |
6a61671b | 685 | { |
2a42eb95 | 686 | unsigned long long clock; |
6a61671b | 687 | |
0e4097c3 | 688 | clock = sched_clock(); |
2a42eb95 | 689 | if (clock < vtime->starttime) |
6a61671b | 690 | return 0; |
abf917cd | 691 | |
2a42eb95 | 692 | return clock - vtime->starttime; |
6a61671b FW |
693 | } |
694 | ||
bac5b6b6 | 695 | static u64 get_vtime_delta(struct vtime *vtime) |
abf917cd | 696 | { |
2a42eb95 WL |
697 | u64 delta = vtime_delta(vtime); |
698 | u64 other; | |
abf917cd | 699 | |
03cbc732 WL |
700 | /* |
701 | * Unlike tick based timing, vtime based timing never has lost | |
702 | * ticks, and no need for steal time accounting to make up for | |
703 | * lost ticks. Vtime accounts a rounded version of actual | |
704 | * elapsed time. Limit account_other_time to prevent rounding | |
705 | * errors from causing elapsed vtime to go negative. | |
706 | */ | |
b58c3584 | 707 | other = account_other_time(delta); |
bac5b6b6 | 708 | WARN_ON_ONCE(vtime->state == VTIME_INACTIVE); |
2a42eb95 | 709 | vtime->starttime += delta; |
abf917cd | 710 | |
b58c3584 | 711 | return delta - other; |
abf917cd FW |
712 | } |
713 | ||
2a42eb95 WL |
714 | static void __vtime_account_system(struct task_struct *tsk, |
715 | struct vtime *vtime) | |
6a61671b | 716 | { |
2a42eb95 WL |
717 | vtime->stime += get_vtime_delta(vtime); |
718 | if (vtime->stime >= TICK_NSEC) { | |
719 | account_system_time(tsk, irq_count(), vtime->stime); | |
720 | vtime->stime = 0; | |
721 | } | |
722 | } | |
723 | ||
724 | static void vtime_account_guest(struct task_struct *tsk, | |
725 | struct vtime *vtime) | |
726 | { | |
727 | vtime->gtime += get_vtime_delta(vtime); | |
728 | if (vtime->gtime >= TICK_NSEC) { | |
729 | account_guest_time(tsk, vtime->gtime); | |
730 | vtime->gtime = 0; | |
731 | } | |
6a61671b FW |
732 | } |
733 | ||
abf917cd FW |
734 | void vtime_account_system(struct task_struct *tsk) |
735 | { | |
bac5b6b6 FW |
736 | struct vtime *vtime = &tsk->vtime; |
737 | ||
738 | if (!vtime_delta(vtime)) | |
ff9a9b4c RR |
739 | return; |
740 | ||
bac5b6b6 | 741 | write_seqcount_begin(&vtime->seqcount); |
2a42eb95 | 742 | /* We might have scheduled out from guest path */ |
14c5126e | 743 | if (tsk->flags & PF_VCPU) |
2a42eb95 WL |
744 | vtime_account_guest(tsk, vtime); |
745 | else | |
746 | __vtime_account_system(tsk, vtime); | |
bac5b6b6 | 747 | write_seqcount_end(&vtime->seqcount); |
6a61671b | 748 | } |
3f4724ea | 749 | |
1c3eda01 | 750 | void vtime_user_enter(struct task_struct *tsk) |
abf917cd | 751 | { |
bac5b6b6 FW |
752 | struct vtime *vtime = &tsk->vtime; |
753 | ||
754 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 | 755 | __vtime_account_system(tsk, vtime); |
bac5b6b6 FW |
756 | vtime->state = VTIME_USER; |
757 | write_seqcount_end(&vtime->seqcount); | |
6a61671b FW |
758 | } |
759 | ||
1c3eda01 | 760 | void vtime_user_exit(struct task_struct *tsk) |
6a61671b | 761 | { |
bac5b6b6 FW |
762 | struct vtime *vtime = &tsk->vtime; |
763 | ||
764 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 WL |
765 | vtime->utime += get_vtime_delta(vtime); |
766 | if (vtime->utime >= TICK_NSEC) { | |
767 | account_user_time(tsk, vtime->utime); | |
768 | vtime->utime = 0; | |
769 | } | |
bac5b6b6 FW |
770 | vtime->state = VTIME_SYS; |
771 | write_seqcount_end(&vtime->seqcount); | |
6a61671b FW |
772 | } |
773 | ||
774 | void vtime_guest_enter(struct task_struct *tsk) | |
775 | { | |
bac5b6b6 | 776 | struct vtime *vtime = &tsk->vtime; |
5b206d48 FW |
777 | /* |
778 | * The flags must be updated under the lock with | |
60a9ce57 | 779 | * the vtime_starttime flush and update. |
5b206d48 FW |
780 | * That enforces a right ordering and update sequence |
781 | * synchronization against the reader (task_gtime()) | |
782 | * that can thus safely catch up with a tickless delta. | |
783 | */ | |
bac5b6b6 | 784 | write_seqcount_begin(&vtime->seqcount); |
2a42eb95 | 785 | __vtime_account_system(tsk, vtime); |
14c5126e | 786 | tsk->flags |= PF_VCPU; |
bac5b6b6 | 787 | write_seqcount_end(&vtime->seqcount); |
6a61671b | 788 | } |
48d6a816 | 789 | EXPORT_SYMBOL_GPL(vtime_guest_enter); |
6a61671b FW |
790 | |
791 | void vtime_guest_exit(struct task_struct *tsk) | |
792 | { | |
bac5b6b6 FW |
793 | struct vtime *vtime = &tsk->vtime; |
794 | ||
795 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 | 796 | vtime_account_guest(tsk, vtime); |
14c5126e | 797 | tsk->flags &= ~PF_VCPU; |
bac5b6b6 | 798 | write_seqcount_end(&vtime->seqcount); |
abf917cd | 799 | } |
48d6a816 | 800 | EXPORT_SYMBOL_GPL(vtime_guest_exit); |
abf917cd FW |
801 | |
802 | void vtime_account_idle(struct task_struct *tsk) | |
803 | { | |
bac5b6b6 | 804 | account_idle_time(get_vtime_delta(&tsk->vtime)); |
abf917cd | 805 | } |
3f4724ea | 806 | |
6a61671b FW |
807 | void arch_vtime_task_switch(struct task_struct *prev) |
808 | { | |
bac5b6b6 | 809 | struct vtime *vtime = &prev->vtime; |
6a61671b | 810 | |
bac5b6b6 FW |
811 | write_seqcount_begin(&vtime->seqcount); |
812 | vtime->state = VTIME_INACTIVE; | |
813 | write_seqcount_end(&vtime->seqcount); | |
814 | ||
815 | vtime = ¤t->vtime; | |
816 | ||
817 | write_seqcount_begin(&vtime->seqcount); | |
818 | vtime->state = VTIME_SYS; | |
0e4097c3 | 819 | vtime->starttime = sched_clock(); |
bac5b6b6 | 820 | write_seqcount_end(&vtime->seqcount); |
6a61671b FW |
821 | } |
822 | ||
45eacc69 | 823 | void vtime_init_idle(struct task_struct *t, int cpu) |
6a61671b | 824 | { |
bac5b6b6 | 825 | struct vtime *vtime = &t->vtime; |
6a61671b FW |
826 | unsigned long flags; |
827 | ||
b7ce2277 | 828 | local_irq_save(flags); |
bac5b6b6 FW |
829 | write_seqcount_begin(&vtime->seqcount); |
830 | vtime->state = VTIME_SYS; | |
0e4097c3 | 831 | vtime->starttime = sched_clock(); |
bac5b6b6 | 832 | write_seqcount_end(&vtime->seqcount); |
b7ce2277 | 833 | local_irq_restore(flags); |
6a61671b FW |
834 | } |
835 | ||
16a6d9be | 836 | u64 task_gtime(struct task_struct *t) |
6a61671b | 837 | { |
bac5b6b6 | 838 | struct vtime *vtime = &t->vtime; |
6a61671b | 839 | unsigned int seq; |
16a6d9be | 840 | u64 gtime; |
6a61671b | 841 | |
e5925394 | 842 | if (!vtime_accounting_enabled()) |
2541117b HS |
843 | return t->gtime; |
844 | ||
6a61671b | 845 | do { |
bac5b6b6 | 846 | seq = read_seqcount_begin(&vtime->seqcount); |
6a61671b FW |
847 | |
848 | gtime = t->gtime; | |
bac5b6b6 | 849 | if (vtime->state == VTIME_SYS && t->flags & PF_VCPU) |
2a42eb95 | 850 | gtime += vtime->gtime + vtime_delta(vtime); |
6a61671b | 851 | |
bac5b6b6 | 852 | } while (read_seqcount_retry(&vtime->seqcount, seq)); |
6a61671b FW |
853 | |
854 | return gtime; | |
855 | } | |
856 | ||
857 | /* | |
858 | * Fetch cputime raw values from fields of task_struct and | |
859 | * add up the pending nohz execution time since the last | |
860 | * cputime snapshot. | |
861 | */ | |
5613fda9 | 862 | void task_cputime(struct task_struct *t, u64 *utime, u64 *stime) |
6a61671b | 863 | { |
bac5b6b6 | 864 | struct vtime *vtime = &t->vtime; |
6a61671b | 865 | unsigned int seq; |
bac5b6b6 | 866 | u64 delta; |
6a61671b | 867 | |
353c50eb SG |
868 | if (!vtime_accounting_enabled()) { |
869 | *utime = t->utime; | |
870 | *stime = t->stime; | |
871 | return; | |
872 | } | |
6a61671b | 873 | |
353c50eb | 874 | do { |
bac5b6b6 | 875 | seq = read_seqcount_begin(&vtime->seqcount); |
6a61671b | 876 | |
353c50eb SG |
877 | *utime = t->utime; |
878 | *stime = t->stime; | |
6a61671b FW |
879 | |
880 | /* Task is sleeping, nothing to add */ | |
bac5b6b6 | 881 | if (vtime->state == VTIME_INACTIVE || is_idle_task(t)) |
6a61671b FW |
882 | continue; |
883 | ||
bac5b6b6 | 884 | delta = vtime_delta(vtime); |
6a61671b FW |
885 | |
886 | /* | |
887 | * Task runs either in user or kernel space, add pending nohz time to | |
888 | * the right place. | |
889 | */ | |
bac5b6b6 | 890 | if (vtime->state == VTIME_USER || t->flags & PF_VCPU) |
2a42eb95 | 891 | *utime += vtime->utime + delta; |
bac5b6b6 | 892 | else if (vtime->state == VTIME_SYS) |
2a42eb95 | 893 | *stime += vtime->stime + delta; |
bac5b6b6 | 894 | } while (read_seqcount_retry(&vtime->seqcount, seq)); |
6a61671b | 895 | } |
abf917cd | 896 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |