]>
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 | |
1966aaf7 | 112 | cpuacct_account_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 | |
2810f611 FW |
262 | /* Shall be converted to a lockdep-enabled lightweight check */ |
263 | WARN_ON_ONCE(!irqs_disabled()); | |
264 | ||
57430218 RR |
265 | accounted = steal_account_process_time(max); |
266 | ||
267 | if (accounted < max) | |
a499a5a1 | 268 | accounted += irqtime_tick_accounted(max - accounted); |
57430218 RR |
269 | |
270 | return accounted; | |
271 | } | |
272 | ||
a1eb1411 SG |
273 | #ifdef CONFIG_64BIT |
274 | static inline u64 read_sum_exec_runtime(struct task_struct *t) | |
275 | { | |
276 | return t->se.sum_exec_runtime; | |
277 | } | |
278 | #else | |
279 | static u64 read_sum_exec_runtime(struct task_struct *t) | |
280 | { | |
281 | u64 ns; | |
282 | struct rq_flags rf; | |
283 | struct rq *rq; | |
284 | ||
285 | rq = task_rq_lock(t, &rf); | |
286 | ns = t->se.sum_exec_runtime; | |
287 | task_rq_unlock(rq, t, &rf); | |
288 | ||
289 | return ns; | |
290 | } | |
291 | #endif | |
292 | ||
a634f933 FW |
293 | /* |
294 | * Accumulate raw cputime values of dead tasks (sig->[us]time) and live | |
295 | * tasks (sum on group iteration) belonging to @tsk's group. | |
296 | */ | |
297 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) | |
298 | { | |
299 | struct signal_struct *sig = tsk->signal; | |
5613fda9 | 300 | u64 utime, stime; |
a634f933 | 301 | struct task_struct *t; |
e78c3496 | 302 | unsigned int seq, nextseq; |
9c368b5b | 303 | unsigned long flags; |
a634f933 | 304 | |
a1eb1411 SG |
305 | /* |
306 | * Update current task runtime to account pending time since last | |
307 | * scheduler action or thread_group_cputime() call. This thread group | |
308 | * might have other running tasks on different CPUs, but updating | |
309 | * their runtime can affect syscall performance, so we skip account | |
310 | * those pending times and rely only on values updated on tick or | |
311 | * other scheduler action. | |
312 | */ | |
313 | if (same_thread_group(current, tsk)) | |
314 | (void) task_sched_runtime(current); | |
315 | ||
a634f933 | 316 | rcu_read_lock(); |
e78c3496 RR |
317 | /* Attempt a lockless read on the first round. */ |
318 | nextseq = 0; | |
319 | do { | |
320 | seq = nextseq; | |
9c368b5b | 321 | flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq); |
e78c3496 RR |
322 | times->utime = sig->utime; |
323 | times->stime = sig->stime; | |
324 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
325 | ||
326 | for_each_thread(tsk, t) { | |
327 | task_cputime(t, &utime, &stime); | |
328 | times->utime += utime; | |
329 | times->stime += stime; | |
a1eb1411 | 330 | times->sum_exec_runtime += read_sum_exec_runtime(t); |
e78c3496 RR |
331 | } |
332 | /* If lockless access failed, take the lock. */ | |
333 | nextseq = 1; | |
334 | } while (need_seqretry(&sig->stats_lock, seq)); | |
9c368b5b | 335 | done_seqretry_irqrestore(&sig->stats_lock, seq, flags); |
a634f933 FW |
336 | rcu_read_unlock(); |
337 | } | |
338 | ||
73fbec60 FW |
339 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
340 | /* | |
341 | * Account a tick to a process and cpustat | |
342 | * @p: the process that the cpu time gets accounted to | |
343 | * @user_tick: is the tick from userspace | |
344 | * @rq: the pointer to rq | |
345 | * | |
346 | * Tick demultiplexing follows the order | |
347 | * - pending hardirq update | |
348 | * - pending softirq update | |
349 | * - user_time | |
350 | * - idle_time | |
351 | * - system time | |
352 | * - check for guest_time | |
353 | * - else account as system_time | |
354 | * | |
355 | * Check for hardirq is done both for system and user time as there is | |
356 | * no timer going off while we are on hardirq and hence we may never get an | |
357 | * opportunity to update it solely in system time. | |
358 | * p->stime and friends are only updated on system time and not on irq | |
359 | * softirq as those do not count in task exec_runtime any more. | |
360 | */ | |
361 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 362 | struct rq *rq, int ticks) |
73fbec60 | 363 | { |
2b1f967d | 364 | u64 other, cputime = TICK_NSEC * ticks; |
73fbec60 | 365 | |
57430218 RR |
366 | /* |
367 | * When returning from idle, many ticks can get accounted at | |
368 | * once, including some ticks of steal, irq, and softirq time. | |
369 | * Subtract those ticks from the amount of time accounted to | |
370 | * idle, or potentially user or system time. Due to rounding, | |
371 | * other time can exceed ticks occasionally. | |
372 | */ | |
03cbc732 | 373 | other = account_other_time(ULONG_MAX); |
2b1f967d | 374 | if (other >= cputime) |
73fbec60 | 375 | return; |
23244a5c | 376 | |
2b1f967d | 377 | cputime -= other; |
73fbec60 | 378 | |
57430218 | 379 | if (this_cpu_ksoftirqd() == p) { |
73fbec60 FW |
380 | /* |
381 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
382 | * So, we have to handle it separately here. | |
383 | * Also, p->stime needs to be updated for ksoftirqd. | |
384 | */ | |
fb8b049c | 385 | account_system_index_time(p, cputime, CPUTIME_SOFTIRQ); |
73fbec60 | 386 | } else if (user_tick) { |
40565b5a | 387 | account_user_time(p, cputime); |
73fbec60 | 388 | } else if (p == rq->idle) { |
18b43a9b | 389 | account_idle_time(cputime); |
73fbec60 | 390 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ |
fb8b049c | 391 | account_guest_time(p, cputime); |
73fbec60 | 392 | } else { |
fb8b049c | 393 | account_system_index_time(p, cputime, CPUTIME_SYSTEM); |
73fbec60 FW |
394 | } |
395 | } | |
396 | ||
397 | static void irqtime_account_idle_ticks(int ticks) | |
398 | { | |
73fbec60 FW |
399 | struct rq *rq = this_rq(); |
400 | ||
2d513868 | 401 | irqtime_account_process_tick(current, 0, rq, ticks); |
73fbec60 FW |
402 | } |
403 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
3f4724ea FW |
404 | static inline void irqtime_account_idle_ticks(int ticks) {} |
405 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 406 | struct rq *rq, int nr_ticks) {} |
73fbec60 FW |
407 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
408 | ||
73fbec60 FW |
409 | /* |
410 | * Use precise platform statistics if available: | |
411 | */ | |
412 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
a7e1a9e3 | 413 | |
e3942ba0 | 414 | #ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
b0493406 | 415 | void vtime_common_task_switch(struct task_struct *prev) |
e3942ba0 FW |
416 | { |
417 | if (is_idle_task(prev)) | |
418 | vtime_account_idle(prev); | |
419 | else | |
420 | vtime_account_system(prev); | |
421 | ||
c8d7dabf | 422 | vtime_flush(prev); |
e3942ba0 FW |
423 | arch_vtime_task_switch(prev); |
424 | } | |
425 | #endif | |
11113334 | 426 | |
0cfdf9a1 FW |
427 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ |
428 | ||
429 | ||
430 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE | |
a7e1a9e3 FW |
431 | /* |
432 | * Archs that account the whole time spent in the idle task | |
433 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 434 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
435 | * have other meaning of the idle time (s390 only includes the |
436 | * time spent by the CPU when it's in low power mode) must override | |
437 | * vtime_account(). | |
438 | */ | |
439 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
0cfdf9a1 | 440 | void vtime_account_irq_enter(struct task_struct *tsk) |
a7e1a9e3 | 441 | { |
0cfdf9a1 FW |
442 | if (!in_interrupt() && is_idle_task(tsk)) |
443 | vtime_account_idle(tsk); | |
444 | else | |
445 | vtime_account_system(tsk); | |
a7e1a9e3 | 446 | } |
0cfdf9a1 | 447 | EXPORT_SYMBOL_GPL(vtime_account_irq_enter); |
a7e1a9e3 | 448 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ |
9fbc42ea | 449 | |
5613fda9 | 450 | void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
9fbc42ea FW |
451 | { |
452 | *ut = p->utime; | |
453 | *st = p->stime; | |
454 | } | |
9eec50b8 | 455 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
a7e1a9e3 | 456 | |
5613fda9 | 457 | void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
9fbc42ea FW |
458 | { |
459 | struct task_cputime cputime; | |
73fbec60 | 460 | |
9fbc42ea FW |
461 | thread_group_cputime(p, &cputime); |
462 | ||
463 | *ut = cputime.utime; | |
464 | *st = cputime.stime; | |
465 | } | |
466 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ | |
467 | /* | |
468 | * Account a single tick of cpu time. | |
469 | * @p: the process that the cpu time gets accounted to | |
470 | * @user_tick: indicates if the tick is a user or a system tick | |
471 | */ | |
472 | void account_process_tick(struct task_struct *p, int user_tick) | |
73fbec60 | 473 | { |
2b1f967d | 474 | u64 cputime, steal; |
9fbc42ea | 475 | struct rq *rq = this_rq(); |
73fbec60 | 476 | |
55dbdcfa | 477 | if (vtime_accounting_cpu_enabled()) |
9fbc42ea FW |
478 | return; |
479 | ||
480 | if (sched_clock_irqtime) { | |
2d513868 | 481 | irqtime_account_process_tick(p, user_tick, rq, 1); |
9fbc42ea FW |
482 | return; |
483 | } | |
484 | ||
2b1f967d | 485 | cputime = TICK_NSEC; |
03cbc732 | 486 | steal = steal_account_process_time(ULONG_MAX); |
57430218 | 487 | |
2b1f967d | 488 | if (steal >= cputime) |
9fbc42ea | 489 | return; |
73fbec60 | 490 | |
2b1f967d | 491 | cputime -= steal; |
57430218 | 492 | |
9fbc42ea | 493 | if (user_tick) |
40565b5a | 494 | account_user_time(p, cputime); |
9fbc42ea | 495 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
fb8b049c | 496 | account_system_time(p, HARDIRQ_OFFSET, cputime); |
73fbec60 | 497 | else |
18b43a9b | 498 | account_idle_time(cputime); |
9fbc42ea | 499 | } |
73fbec60 | 500 | |
9fbc42ea FW |
501 | /* |
502 | * Account multiple ticks of idle time. | |
503 | * @ticks: number of stolen ticks | |
504 | */ | |
505 | void account_idle_ticks(unsigned long ticks) | |
506 | { | |
18b43a9b | 507 | u64 cputime, steal; |
26f2c75c | 508 | |
9fbc42ea FW |
509 | if (sched_clock_irqtime) { |
510 | irqtime_account_idle_ticks(ticks); | |
511 | return; | |
512 | } | |
513 | ||
18b43a9b | 514 | cputime = ticks * TICK_NSEC; |
2b1f967d | 515 | steal = steal_account_process_time(ULONG_MAX); |
f9bcf1e0 WL |
516 | |
517 | if (steal >= cputime) | |
518 | return; | |
519 | ||
520 | cputime -= steal; | |
521 | account_idle_time(cputime); | |
9fbc42ea | 522 | } |
73fbec60 | 523 | |
d9a3c982 | 524 | /* |
55eaa7c1 SG |
525 | * Perform (stime * rtime) / total, but avoid multiplication overflow by |
526 | * loosing precision when the numbers are big. | |
d9a3c982 | 527 | */ |
5613fda9 | 528 | static u64 scale_stime(u64 stime, u64 rtime, u64 total) |
73fbec60 | 529 | { |
55eaa7c1 | 530 | u64 scaled; |
73fbec60 | 531 | |
55eaa7c1 SG |
532 | for (;;) { |
533 | /* Make sure "rtime" is the bigger of stime/rtime */ | |
84f9f3a1 SG |
534 | if (stime > rtime) |
535 | swap(rtime, stime); | |
55eaa7c1 SG |
536 | |
537 | /* Make sure 'total' fits in 32 bits */ | |
538 | if (total >> 32) | |
539 | goto drop_precision; | |
540 | ||
541 | /* Does rtime (and thus stime) fit in 32 bits? */ | |
542 | if (!(rtime >> 32)) | |
543 | break; | |
544 | ||
545 | /* Can we just balance rtime/stime rather than dropping bits? */ | |
546 | if (stime >> 31) | |
547 | goto drop_precision; | |
548 | ||
549 | /* We can grow stime and shrink rtime and try to make them both fit */ | |
550 | stime <<= 1; | |
551 | rtime >>= 1; | |
552 | continue; | |
553 | ||
554 | drop_precision: | |
555 | /* We drop from rtime, it has more bits than stime */ | |
556 | rtime >>= 1; | |
557 | total >>= 1; | |
d9a3c982 | 558 | } |
73fbec60 | 559 | |
55eaa7c1 SG |
560 | /* |
561 | * Make sure gcc understands that this is a 32x32->64 multiply, | |
562 | * followed by a 64/32->64 divide. | |
563 | */ | |
564 | scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total); | |
5613fda9 | 565 | return scaled; |
73fbec60 FW |
566 | } |
567 | ||
347abad9 | 568 | /* |
9d7fb042 PZ |
569 | * Adjust tick based cputime random precision against scheduler runtime |
570 | * accounting. | |
347abad9 | 571 | * |
9d7fb042 PZ |
572 | * Tick based cputime accounting depend on random scheduling timeslices of a |
573 | * task to be interrupted or not by the timer. Depending on these | |
574 | * circumstances, the number of these interrupts may be over or | |
575 | * under-optimistic, matching the real user and system cputime with a variable | |
576 | * precision. | |
577 | * | |
578 | * Fix this by scaling these tick based values against the total runtime | |
579 | * accounted by the CFS scheduler. | |
580 | * | |
581 | * This code provides the following guarantees: | |
582 | * | |
583 | * stime + utime == rtime | |
584 | * stime_i+1 >= stime_i, utime_i+1 >= utime_i | |
585 | * | |
586 | * Assuming that rtime_i+1 >= rtime_i. | |
fa092057 | 587 | */ |
d37f761d | 588 | static void cputime_adjust(struct task_cputime *curr, |
9d7fb042 | 589 | struct prev_cputime *prev, |
5613fda9 | 590 | u64 *ut, u64 *st) |
73fbec60 | 591 | { |
5613fda9 | 592 | u64 rtime, stime, utime; |
9d7fb042 | 593 | unsigned long flags; |
fa092057 | 594 | |
9d7fb042 PZ |
595 | /* Serialize concurrent callers such that we can honour our guarantees */ |
596 | raw_spin_lock_irqsave(&prev->lock, flags); | |
5613fda9 | 597 | rtime = curr->sum_exec_runtime; |
73fbec60 | 598 | |
772c808a | 599 | /* |
9d7fb042 PZ |
600 | * This is possible under two circumstances: |
601 | * - rtime isn't monotonic after all (a bug); | |
602 | * - we got reordered by the lock. | |
603 | * | |
604 | * In both cases this acts as a filter such that the rest of the code | |
605 | * can assume it is monotonic regardless of anything else. | |
772c808a SG |
606 | */ |
607 | if (prev->stime + prev->utime >= rtime) | |
608 | goto out; | |
609 | ||
5a8e01f8 SG |
610 | stime = curr->stime; |
611 | utime = curr->utime; | |
612 | ||
173be9a1 | 613 | /* |
3b9c08ae IM |
614 | * If either stime or utime are 0, assume all runtime is userspace. |
615 | * Once a task gets some ticks, the monotonicy code at 'update:' | |
616 | * will ensure things converge to the observed ratio. | |
173be9a1 | 617 | */ |
3b9c08ae IM |
618 | if (stime == 0) { |
619 | utime = rtime; | |
620 | goto update; | |
9d7fb042 | 621 | } |
5a8e01f8 | 622 | |
3b9c08ae IM |
623 | if (utime == 0) { |
624 | stime = rtime; | |
625 | goto update; | |
626 | } | |
627 | ||
628 | stime = scale_stime(stime, rtime, stime + utime); | |
629 | ||
630 | update: | |
9d7fb042 PZ |
631 | /* |
632 | * Make sure stime doesn't go backwards; this preserves monotonicity | |
633 | * for utime because rtime is monotonic. | |
634 | * | |
635 | * utime_i+1 = rtime_i+1 - stime_i | |
636 | * = rtime_i+1 - (rtime_i - utime_i) | |
637 | * = (rtime_i+1 - rtime_i) + utime_i | |
638 | * >= utime_i | |
639 | */ | |
640 | if (stime < prev->stime) | |
641 | stime = prev->stime; | |
642 | utime = rtime - stime; | |
643 | ||
644 | /* | |
645 | * Make sure utime doesn't go backwards; this still preserves | |
646 | * monotonicity for stime, analogous argument to above. | |
647 | */ | |
648 | if (utime < prev->utime) { | |
649 | utime = prev->utime; | |
650 | stime = rtime - utime; | |
651 | } | |
d37f761d | 652 | |
9d7fb042 PZ |
653 | prev->stime = stime; |
654 | prev->utime = utime; | |
772c808a | 655 | out: |
d37f761d FW |
656 | *ut = prev->utime; |
657 | *st = prev->stime; | |
9d7fb042 | 658 | raw_spin_unlock_irqrestore(&prev->lock, flags); |
d37f761d | 659 | } |
73fbec60 | 660 | |
5613fda9 | 661 | void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
d37f761d FW |
662 | { |
663 | struct task_cputime cputime = { | |
d37f761d FW |
664 | .sum_exec_runtime = p->se.sum_exec_runtime, |
665 | }; | |
666 | ||
6fac4829 | 667 | task_cputime(p, &cputime.utime, &cputime.stime); |
d37f761d | 668 | cputime_adjust(&cputime, &p->prev_cputime, ut, st); |
73fbec60 | 669 | } |
9eec50b8 | 670 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
73fbec60 | 671 | |
5613fda9 | 672 | void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
73fbec60 | 673 | { |
73fbec60 | 674 | struct task_cputime cputime; |
73fbec60 FW |
675 | |
676 | thread_group_cputime(p, &cputime); | |
d37f761d | 677 | cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); |
73fbec60 | 678 | } |
9fbc42ea | 679 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ |
abf917cd FW |
680 | |
681 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN | |
bac5b6b6 | 682 | static u64 vtime_delta(struct vtime *vtime) |
6a61671b | 683 | { |
2a42eb95 | 684 | unsigned long long clock; |
6a61671b | 685 | |
0e4097c3 | 686 | clock = sched_clock(); |
2a42eb95 | 687 | if (clock < vtime->starttime) |
6a61671b | 688 | return 0; |
abf917cd | 689 | |
2a42eb95 | 690 | return clock - vtime->starttime; |
6a61671b FW |
691 | } |
692 | ||
bac5b6b6 | 693 | static u64 get_vtime_delta(struct vtime *vtime) |
abf917cd | 694 | { |
2a42eb95 WL |
695 | u64 delta = vtime_delta(vtime); |
696 | u64 other; | |
abf917cd | 697 | |
03cbc732 WL |
698 | /* |
699 | * Unlike tick based timing, vtime based timing never has lost | |
700 | * ticks, and no need for steal time accounting to make up for | |
701 | * lost ticks. Vtime accounts a rounded version of actual | |
702 | * elapsed time. Limit account_other_time to prevent rounding | |
703 | * errors from causing elapsed vtime to go negative. | |
704 | */ | |
b58c3584 | 705 | other = account_other_time(delta); |
bac5b6b6 | 706 | WARN_ON_ONCE(vtime->state == VTIME_INACTIVE); |
2a42eb95 | 707 | vtime->starttime += delta; |
abf917cd | 708 | |
b58c3584 | 709 | return delta - other; |
abf917cd FW |
710 | } |
711 | ||
2a42eb95 WL |
712 | static void __vtime_account_system(struct task_struct *tsk, |
713 | struct vtime *vtime) | |
6a61671b | 714 | { |
2a42eb95 WL |
715 | vtime->stime += get_vtime_delta(vtime); |
716 | if (vtime->stime >= TICK_NSEC) { | |
717 | account_system_time(tsk, irq_count(), vtime->stime); | |
718 | vtime->stime = 0; | |
719 | } | |
720 | } | |
721 | ||
722 | static void vtime_account_guest(struct task_struct *tsk, | |
723 | struct vtime *vtime) | |
724 | { | |
725 | vtime->gtime += get_vtime_delta(vtime); | |
726 | if (vtime->gtime >= TICK_NSEC) { | |
727 | account_guest_time(tsk, vtime->gtime); | |
728 | vtime->gtime = 0; | |
729 | } | |
6a61671b FW |
730 | } |
731 | ||
abf917cd FW |
732 | void vtime_account_system(struct task_struct *tsk) |
733 | { | |
bac5b6b6 FW |
734 | struct vtime *vtime = &tsk->vtime; |
735 | ||
736 | if (!vtime_delta(vtime)) | |
ff9a9b4c RR |
737 | return; |
738 | ||
bac5b6b6 | 739 | write_seqcount_begin(&vtime->seqcount); |
2a42eb95 WL |
740 | /* We might have scheduled out from guest path */ |
741 | if (current->flags & PF_VCPU) | |
742 | vtime_account_guest(tsk, vtime); | |
743 | else | |
744 | __vtime_account_system(tsk, vtime); | |
bac5b6b6 | 745 | write_seqcount_end(&vtime->seqcount); |
6a61671b | 746 | } |
3f4724ea | 747 | |
1c3eda01 | 748 | void vtime_user_enter(struct task_struct *tsk) |
abf917cd | 749 | { |
bac5b6b6 FW |
750 | struct vtime *vtime = &tsk->vtime; |
751 | ||
752 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 | 753 | __vtime_account_system(tsk, vtime); |
bac5b6b6 FW |
754 | vtime->state = VTIME_USER; |
755 | write_seqcount_end(&vtime->seqcount); | |
6a61671b FW |
756 | } |
757 | ||
1c3eda01 | 758 | void vtime_user_exit(struct task_struct *tsk) |
6a61671b | 759 | { |
bac5b6b6 FW |
760 | struct vtime *vtime = &tsk->vtime; |
761 | ||
762 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 WL |
763 | vtime->utime += get_vtime_delta(vtime); |
764 | if (vtime->utime >= TICK_NSEC) { | |
765 | account_user_time(tsk, vtime->utime); | |
766 | vtime->utime = 0; | |
767 | } | |
bac5b6b6 FW |
768 | vtime->state = VTIME_SYS; |
769 | write_seqcount_end(&vtime->seqcount); | |
6a61671b FW |
770 | } |
771 | ||
772 | void vtime_guest_enter(struct task_struct *tsk) | |
773 | { | |
bac5b6b6 | 774 | struct vtime *vtime = &tsk->vtime; |
5b206d48 FW |
775 | /* |
776 | * The flags must be updated under the lock with | |
60a9ce57 | 777 | * the vtime_starttime flush and update. |
5b206d48 FW |
778 | * That enforces a right ordering and update sequence |
779 | * synchronization against the reader (task_gtime()) | |
780 | * that can thus safely catch up with a tickless delta. | |
781 | */ | |
bac5b6b6 | 782 | write_seqcount_begin(&vtime->seqcount); |
2a42eb95 | 783 | __vtime_account_system(tsk, vtime); |
6a61671b | 784 | current->flags |= PF_VCPU; |
bac5b6b6 | 785 | write_seqcount_end(&vtime->seqcount); |
6a61671b | 786 | } |
48d6a816 | 787 | EXPORT_SYMBOL_GPL(vtime_guest_enter); |
6a61671b FW |
788 | |
789 | void vtime_guest_exit(struct task_struct *tsk) | |
790 | { | |
bac5b6b6 FW |
791 | struct vtime *vtime = &tsk->vtime; |
792 | ||
793 | write_seqcount_begin(&vtime->seqcount); | |
2a42eb95 | 794 | vtime_account_guest(tsk, vtime); |
6a61671b | 795 | current->flags &= ~PF_VCPU; |
bac5b6b6 | 796 | write_seqcount_end(&vtime->seqcount); |
abf917cd | 797 | } |
48d6a816 | 798 | EXPORT_SYMBOL_GPL(vtime_guest_exit); |
abf917cd FW |
799 | |
800 | void vtime_account_idle(struct task_struct *tsk) | |
801 | { | |
bac5b6b6 | 802 | account_idle_time(get_vtime_delta(&tsk->vtime)); |
abf917cd | 803 | } |
3f4724ea | 804 | |
6a61671b FW |
805 | void arch_vtime_task_switch(struct task_struct *prev) |
806 | { | |
bac5b6b6 | 807 | struct vtime *vtime = &prev->vtime; |
6a61671b | 808 | |
bac5b6b6 FW |
809 | write_seqcount_begin(&vtime->seqcount); |
810 | vtime->state = VTIME_INACTIVE; | |
811 | write_seqcount_end(&vtime->seqcount); | |
812 | ||
813 | vtime = ¤t->vtime; | |
814 | ||
815 | write_seqcount_begin(&vtime->seqcount); | |
816 | vtime->state = VTIME_SYS; | |
0e4097c3 | 817 | vtime->starttime = sched_clock(); |
bac5b6b6 | 818 | write_seqcount_end(&vtime->seqcount); |
6a61671b FW |
819 | } |
820 | ||
45eacc69 | 821 | void vtime_init_idle(struct task_struct *t, int cpu) |
6a61671b | 822 | { |
bac5b6b6 | 823 | struct vtime *vtime = &t->vtime; |
6a61671b FW |
824 | unsigned long flags; |
825 | ||
b7ce2277 | 826 | local_irq_save(flags); |
bac5b6b6 FW |
827 | write_seqcount_begin(&vtime->seqcount); |
828 | vtime->state = VTIME_SYS; | |
0e4097c3 | 829 | vtime->starttime = sched_clock(); |
bac5b6b6 | 830 | write_seqcount_end(&vtime->seqcount); |
b7ce2277 | 831 | local_irq_restore(flags); |
6a61671b FW |
832 | } |
833 | ||
16a6d9be | 834 | u64 task_gtime(struct task_struct *t) |
6a61671b | 835 | { |
bac5b6b6 | 836 | struct vtime *vtime = &t->vtime; |
6a61671b | 837 | unsigned int seq; |
16a6d9be | 838 | u64 gtime; |
6a61671b | 839 | |
e5925394 | 840 | if (!vtime_accounting_enabled()) |
2541117b HS |
841 | return t->gtime; |
842 | ||
6a61671b | 843 | do { |
bac5b6b6 | 844 | seq = read_seqcount_begin(&vtime->seqcount); |
6a61671b FW |
845 | |
846 | gtime = t->gtime; | |
bac5b6b6 | 847 | if (vtime->state == VTIME_SYS && t->flags & PF_VCPU) |
2a42eb95 | 848 | gtime += vtime->gtime + vtime_delta(vtime); |
6a61671b | 849 | |
bac5b6b6 | 850 | } while (read_seqcount_retry(&vtime->seqcount, seq)); |
6a61671b FW |
851 | |
852 | return gtime; | |
853 | } | |
854 | ||
855 | /* | |
856 | * Fetch cputime raw values from fields of task_struct and | |
857 | * add up the pending nohz execution time since the last | |
858 | * cputime snapshot. | |
859 | */ | |
5613fda9 | 860 | void task_cputime(struct task_struct *t, u64 *utime, u64 *stime) |
6a61671b | 861 | { |
bac5b6b6 | 862 | struct vtime *vtime = &t->vtime; |
6a61671b | 863 | unsigned int seq; |
bac5b6b6 | 864 | u64 delta; |
6a61671b | 865 | |
353c50eb SG |
866 | if (!vtime_accounting_enabled()) { |
867 | *utime = t->utime; | |
868 | *stime = t->stime; | |
869 | return; | |
870 | } | |
6a61671b | 871 | |
353c50eb | 872 | do { |
bac5b6b6 | 873 | seq = read_seqcount_begin(&vtime->seqcount); |
6a61671b | 874 | |
353c50eb SG |
875 | *utime = t->utime; |
876 | *stime = t->stime; | |
6a61671b FW |
877 | |
878 | /* Task is sleeping, nothing to add */ | |
bac5b6b6 | 879 | if (vtime->state == VTIME_INACTIVE || is_idle_task(t)) |
6a61671b FW |
880 | continue; |
881 | ||
bac5b6b6 | 882 | delta = vtime_delta(vtime); |
6a61671b FW |
883 | |
884 | /* | |
885 | * Task runs either in user or kernel space, add pending nohz time to | |
886 | * the right place. | |
887 | */ | |
bac5b6b6 | 888 | if (vtime->state == VTIME_USER || t->flags & PF_VCPU) |
2a42eb95 | 889 | *utime += vtime->utime + delta; |
bac5b6b6 | 890 | else if (vtime->state == VTIME_SYS) |
2a42eb95 | 891 | *stime += vtime->stime + delta; |
bac5b6b6 | 892 | } while (read_seqcount_retry(&vtime->seqcount, seq)); |
6a61671b | 893 | } |
abf917cd | 894 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |