]>
Commit | Line | Data |
---|---|---|
73fbec60 FW |
1 | #include <linux/export.h> |
2 | #include <linux/sched.h> | |
3 | #include <linux/tsacct_kern.h> | |
4 | #include <linux/kernel_stat.h> | |
5 | #include <linux/static_key.h> | |
abf917cd | 6 | #include <linux/context_tracking.h> |
73fbec60 | 7 | #include "sched.h" |
1fe7c4ef SS |
8 | #ifdef CONFIG_PARAVIRT |
9 | #include <asm/paravirt.h> | |
10 | #endif | |
73fbec60 FW |
11 | |
12 | ||
13 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
14 | ||
15 | /* | |
16 | * There are no locks covering percpu hardirq/softirq time. | |
bf9fae9f | 17 | * They are only modified in vtime_account, on corresponding CPU |
73fbec60 FW |
18 | * with interrupts disabled. So, writes are safe. |
19 | * They are read and saved off onto struct rq in update_rq_clock(). | |
20 | * This may result in other CPU reading this CPU's irq time and can | |
bf9fae9f | 21 | * race with irq/vtime_account on this CPU. We would either get old |
73fbec60 FW |
22 | * or new value with a side effect of accounting a slice of irq time to wrong |
23 | * task when irq is in progress while we read rq->clock. That is a worthy | |
24 | * compromise in place of having locks on each irq in account_system_time. | |
25 | */ | |
19d23dbf | 26 | DEFINE_PER_CPU(struct irqtime, cpu_irqtime); |
73fbec60 | 27 | |
73fbec60 FW |
28 | static int sched_clock_irqtime; |
29 | ||
30 | void enable_sched_clock_irqtime(void) | |
31 | { | |
32 | sched_clock_irqtime = 1; | |
33 | } | |
34 | ||
35 | void disable_sched_clock_irqtime(void) | |
36 | { | |
37 | sched_clock_irqtime = 0; | |
38 | } | |
39 | ||
73fbec60 FW |
40 | /* |
41 | * Called before incrementing preempt_count on {soft,}irq_enter | |
42 | * and before decrementing preempt_count on {soft,}irq_exit. | |
43 | */ | |
3e1df4f5 | 44 | void irqtime_account_irq(struct task_struct *curr) |
73fbec60 | 45 | { |
19d23dbf | 46 | struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime); |
a499a5a1 | 47 | u64 *cpustat = kcpustat_this_cpu->cpustat; |
73fbec60 FW |
48 | s64 delta; |
49 | int cpu; | |
50 | ||
51 | if (!sched_clock_irqtime) | |
52 | return; | |
53 | ||
73fbec60 | 54 | cpu = smp_processor_id(); |
19d23dbf FW |
55 | delta = sched_clock_cpu(cpu) - irqtime->irq_start_time; |
56 | irqtime->irq_start_time += delta; | |
73fbec60 | 57 | |
19d23dbf | 58 | u64_stats_update_begin(&irqtime->sync); |
73fbec60 FW |
59 | /* |
60 | * We do not account for softirq time from ksoftirqd here. | |
61 | * We want to continue accounting softirq time to ksoftirqd thread | |
62 | * in that case, so as not to confuse scheduler with a special task | |
63 | * that do not consume any time, but still wants to run. | |
64 | */ | |
a499a5a1 FW |
65 | if (hardirq_count()) { |
66 | cpustat[CPUTIME_IRQ] += delta; | |
67 | irqtime->tick_delta += delta; | |
68 | } else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) { | |
69 | cpustat[CPUTIME_SOFTIRQ] += delta; | |
70 | irqtime->tick_delta += delta; | |
71 | } | |
73fbec60 | 72 | |
19d23dbf | 73 | u64_stats_update_end(&irqtime->sync); |
73fbec60 | 74 | } |
3e1df4f5 | 75 | EXPORT_SYMBOL_GPL(irqtime_account_irq); |
73fbec60 | 76 | |
a499a5a1 | 77 | static cputime_t irqtime_tick_accounted(cputime_t maxtime) |
73fbec60 | 78 | { |
a499a5a1 FW |
79 | struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime); |
80 | cputime_t delta; | |
73fbec60 | 81 | |
a499a5a1 FW |
82 | delta = nsecs_to_cputime(irqtime->tick_delta); |
83 | delta = min(delta, maxtime); | |
84 | irqtime->tick_delta -= cputime_to_nsecs(delta); | |
2810f611 | 85 | |
a499a5a1 | 86 | return delta; |
73fbec60 FW |
87 | } |
88 | ||
89 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
90 | ||
91 | #define sched_clock_irqtime (0) | |
92 | ||
a499a5a1 | 93 | static cputime_t irqtime_tick_accounted(cputime_t dummy) |
57430218 RR |
94 | { |
95 | return 0; | |
96 | } | |
97 | ||
73fbec60 FW |
98 | #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */ |
99 | ||
100 | static inline void task_group_account_field(struct task_struct *p, int index, | |
101 | u64 tmp) | |
102 | { | |
73fbec60 FW |
103 | /* |
104 | * Since all updates are sure to touch the root cgroup, we | |
105 | * get ourselves ahead and touch it first. If the root cgroup | |
106 | * is the only cgroup, then nothing else should be necessary. | |
107 | * | |
108 | */ | |
a4f61cc0 | 109 | __this_cpu_add(kernel_cpustat.cpustat[index], tmp); |
73fbec60 | 110 | |
1966aaf7 | 111 | cpuacct_account_field(p, index, tmp); |
73fbec60 FW |
112 | } |
113 | ||
114 | /* | |
115 | * Account user cpu time to a process. | |
116 | * @p: the process that the cpu time gets accounted to | |
117 | * @cputime: the cpu time spent in user space since the last update | |
73fbec60 | 118 | */ |
40565b5a | 119 | void account_user_time(struct task_struct *p, cputime_t cputime) |
73fbec60 FW |
120 | { |
121 | int index; | |
122 | ||
123 | /* Add user time to process. */ | |
5613fda9 | 124 | p->utime += cputime_to_nsecs(cputime); |
ebd7e7fc | 125 | account_group_user_time(p, cputime_to_nsecs(cputime)); |
73fbec60 | 126 | |
d0ea0268 | 127 | index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
73fbec60 FW |
128 | |
129 | /* Add user time to cpustat. */ | |
7fb1327e | 130 | task_group_account_field(p, index, cputime_to_nsecs(cputime)); |
73fbec60 FW |
131 | |
132 | /* Account for user time used */ | |
6fac4829 | 133 | acct_account_cputime(p); |
73fbec60 FW |
134 | } |
135 | ||
136 | /* | |
137 | * Account guest cpu time to a process. | |
138 | * @p: the process that the cpu time gets accounted to | |
139 | * @cputime: the cpu time spent in virtual machine since the last update | |
73fbec60 | 140 | */ |
1213699a | 141 | void account_guest_time(struct task_struct *p, cputime_t cputime) |
73fbec60 FW |
142 | { |
143 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
144 | ||
145 | /* Add guest time to process. */ | |
5613fda9 | 146 | p->utime += cputime_to_nsecs(cputime); |
ebd7e7fc | 147 | account_group_user_time(p, cputime_to_nsecs(cputime)); |
16a6d9be | 148 | p->gtime += cputime_to_nsecs(cputime); |
73fbec60 FW |
149 | |
150 | /* Add guest time to cpustat. */ | |
d0ea0268 | 151 | if (task_nice(p) > 0) { |
7fb1327e FW |
152 | cpustat[CPUTIME_NICE] += cputime_to_nsecs(cputime); |
153 | cpustat[CPUTIME_GUEST_NICE] += cputime_to_nsecs(cputime); | |
73fbec60 | 154 | } else { |
7fb1327e FW |
155 | cpustat[CPUTIME_USER] += cputime_to_nsecs(cputime); |
156 | cpustat[CPUTIME_GUEST] += cputime_to_nsecs(cputime); | |
73fbec60 FW |
157 | } |
158 | } | |
159 | ||
160 | /* | |
161 | * Account system cpu time to a process and desired cpustat field | |
162 | * @p: the process that the cpu time gets accounted to | |
163 | * @cputime: the cpu time spent in kernel space since the last update | |
40565b5a | 164 | * @index: pointer to cpustat field that has to be updated |
73fbec60 | 165 | */ |
c31cc6a5 FW |
166 | void account_system_index_time(struct task_struct *p, |
167 | cputime_t cputime, enum cpu_usage_stat index) | |
73fbec60 FW |
168 | { |
169 | /* Add system time to process. */ | |
5613fda9 | 170 | p->stime += cputime_to_nsecs(cputime); |
ebd7e7fc | 171 | account_group_system_time(p, cputime_to_nsecs(cputime)); |
73fbec60 FW |
172 | |
173 | /* Add system time to cpustat. */ | |
7fb1327e | 174 | task_group_account_field(p, index, cputime_to_nsecs(cputime)); |
73fbec60 FW |
175 | |
176 | /* Account for system time used */ | |
6fac4829 | 177 | acct_account_cputime(p); |
73fbec60 FW |
178 | } |
179 | ||
180 | /* | |
181 | * Account system cpu time to a process. | |
182 | * @p: the process that the cpu time gets accounted to | |
183 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
184 | * @cputime: the cpu time spent in kernel space since the last update | |
73fbec60 FW |
185 | */ |
186 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
40565b5a | 187 | cputime_t 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 | */ | |
210 | void account_steal_time(cputime_t cputime) | |
211 | { | |
212 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
213 | ||
7fb1327e | 214 | cpustat[CPUTIME_STEAL] += cputime_to_nsecs(cputime); |
73fbec60 FW |
215 | } |
216 | ||
217 | /* | |
218 | * Account for idle time. | |
219 | * @cputime: the cpu time spent in idle wait | |
220 | */ | |
221 | void account_idle_time(cputime_t cputime) | |
222 | { | |
223 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
224 | struct rq *rq = this_rq(); | |
225 | ||
226 | if (atomic_read(&rq->nr_iowait) > 0) | |
7fb1327e | 227 | cpustat[CPUTIME_IOWAIT] += cputime_to_nsecs(cputime); |
73fbec60 | 228 | else |
7fb1327e | 229 | cpustat[CPUTIME_IDLE] += cputime_to_nsecs(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 | */ | |
57430218 | 237 | static __always_inline cputime_t steal_account_process_time(cputime_t maxtime) |
73fbec60 FW |
238 | { |
239 | #ifdef CONFIG_PARAVIRT | |
240 | if (static_key_false(¶virt_steal_enabled)) { | |
57430218 | 241 | cputime_t steal_cputime; |
dee08a72 | 242 | u64 steal; |
73fbec60 FW |
243 | |
244 | steal = paravirt_steal_clock(smp_processor_id()); | |
245 | steal -= this_rq()->prev_steal_time; | |
246 | ||
57430218 RR |
247 | steal_cputime = min(nsecs_to_cputime(steal), maxtime); |
248 | account_steal_time(steal_cputime); | |
249 | this_rq()->prev_steal_time += cputime_to_nsecs(steal_cputime); | |
73fbec60 | 250 | |
57430218 | 251 | return steal_cputime; |
73fbec60 FW |
252 | } |
253 | #endif | |
807e5b80 | 254 | return 0; |
73fbec60 FW |
255 | } |
256 | ||
57430218 RR |
257 | /* |
258 | * Account how much elapsed time was spent in steal, irq, or softirq time. | |
259 | */ | |
260 | static inline cputime_t account_other_time(cputime_t max) | |
261 | { | |
262 | cputime_t accounted; | |
263 | ||
2810f611 FW |
264 | /* Shall be converted to a lockdep-enabled lightweight check */ |
265 | WARN_ON_ONCE(!irqs_disabled()); | |
266 | ||
57430218 RR |
267 | accounted = steal_account_process_time(max); |
268 | ||
269 | if (accounted < max) | |
a499a5a1 | 270 | accounted += irqtime_tick_accounted(max - accounted); |
57430218 RR |
271 | |
272 | return accounted; | |
273 | } | |
274 | ||
a1eb1411 SG |
275 | #ifdef CONFIG_64BIT |
276 | static inline u64 read_sum_exec_runtime(struct task_struct *t) | |
277 | { | |
278 | return t->se.sum_exec_runtime; | |
279 | } | |
280 | #else | |
281 | static u64 read_sum_exec_runtime(struct task_struct *t) | |
282 | { | |
283 | u64 ns; | |
284 | struct rq_flags rf; | |
285 | struct rq *rq; | |
286 | ||
287 | rq = task_rq_lock(t, &rf); | |
288 | ns = t->se.sum_exec_runtime; | |
289 | task_rq_unlock(rq, t, &rf); | |
290 | ||
291 | return ns; | |
292 | } | |
293 | #endif | |
294 | ||
a634f933 FW |
295 | /* |
296 | * Accumulate raw cputime values of dead tasks (sig->[us]time) and live | |
297 | * tasks (sum on group iteration) belonging to @tsk's group. | |
298 | */ | |
299 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) | |
300 | { | |
301 | struct signal_struct *sig = tsk->signal; | |
5613fda9 | 302 | u64 utime, stime; |
a634f933 | 303 | struct task_struct *t; |
e78c3496 | 304 | unsigned int seq, nextseq; |
9c368b5b | 305 | unsigned long flags; |
a634f933 | 306 | |
a1eb1411 SG |
307 | /* |
308 | * Update current task runtime to account pending time since last | |
309 | * scheduler action or thread_group_cputime() call. This thread group | |
310 | * might have other running tasks on different CPUs, but updating | |
311 | * their runtime can affect syscall performance, so we skip account | |
312 | * those pending times and rely only on values updated on tick or | |
313 | * other scheduler action. | |
314 | */ | |
315 | if (same_thread_group(current, tsk)) | |
316 | (void) task_sched_runtime(current); | |
317 | ||
a634f933 | 318 | rcu_read_lock(); |
e78c3496 RR |
319 | /* Attempt a lockless read on the first round. */ |
320 | nextseq = 0; | |
321 | do { | |
322 | seq = nextseq; | |
9c368b5b | 323 | flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq); |
e78c3496 RR |
324 | times->utime = sig->utime; |
325 | times->stime = sig->stime; | |
326 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
327 | ||
328 | for_each_thread(tsk, t) { | |
329 | task_cputime(t, &utime, &stime); | |
330 | times->utime += utime; | |
331 | times->stime += stime; | |
a1eb1411 | 332 | times->sum_exec_runtime += read_sum_exec_runtime(t); |
e78c3496 RR |
333 | } |
334 | /* If lockless access failed, take the lock. */ | |
335 | nextseq = 1; | |
336 | } while (need_seqretry(&sig->stats_lock, seq)); | |
9c368b5b | 337 | done_seqretry_irqrestore(&sig->stats_lock, seq, flags); |
a634f933 FW |
338 | rcu_read_unlock(); |
339 | } | |
340 | ||
73fbec60 FW |
341 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
342 | /* | |
343 | * Account a tick to a process and cpustat | |
344 | * @p: the process that the cpu time gets accounted to | |
345 | * @user_tick: is the tick from userspace | |
346 | * @rq: the pointer to rq | |
347 | * | |
348 | * Tick demultiplexing follows the order | |
349 | * - pending hardirq update | |
350 | * - pending softirq update | |
351 | * - user_time | |
352 | * - idle_time | |
353 | * - system time | |
354 | * - check for guest_time | |
355 | * - else account as system_time | |
356 | * | |
357 | * Check for hardirq is done both for system and user time as there is | |
358 | * no timer going off while we are on hardirq and hence we may never get an | |
359 | * opportunity to update it solely in system time. | |
360 | * p->stime and friends are only updated on system time and not on irq | |
361 | * softirq as those do not count in task exec_runtime any more. | |
362 | */ | |
363 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 364 | struct rq *rq, int ticks) |
73fbec60 | 365 | { |
57430218 | 366 | u64 cputime = (__force u64) cputime_one_jiffy * ticks; |
981ee2d4 | 367 | cputime_t other; |
73fbec60 | 368 | |
57430218 RR |
369 | /* |
370 | * When returning from idle, many ticks can get accounted at | |
371 | * once, including some ticks of steal, irq, and softirq time. | |
372 | * Subtract those ticks from the amount of time accounted to | |
373 | * idle, or potentially user or system time. Due to rounding, | |
374 | * other time can exceed ticks occasionally. | |
375 | */ | |
03cbc732 | 376 | other = account_other_time(ULONG_MAX); |
57430218 | 377 | if (other >= cputime) |
73fbec60 | 378 | return; |
57430218 | 379 | cputime -= other; |
73fbec60 | 380 | |
57430218 | 381 | if (this_cpu_ksoftirqd() == p) { |
73fbec60 FW |
382 | /* |
383 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
384 | * So, we have to handle it separately here. | |
385 | * Also, p->stime needs to be updated for ksoftirqd. | |
386 | */ | |
c31cc6a5 | 387 | account_system_index_time(p, cputime, CPUTIME_SOFTIRQ); |
73fbec60 | 388 | } else if (user_tick) { |
40565b5a | 389 | account_user_time(p, cputime); |
73fbec60 | 390 | } else if (p == rq->idle) { |
2d513868 | 391 | account_idle_time(cputime); |
73fbec60 | 392 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ |
40565b5a | 393 | account_guest_time(p, cputime); |
73fbec60 | 394 | } else { |
c31cc6a5 | 395 | account_system_index_time(p, cputime, CPUTIME_SYSTEM); |
73fbec60 FW |
396 | } |
397 | } | |
398 | ||
399 | static void irqtime_account_idle_ticks(int ticks) | |
400 | { | |
73fbec60 FW |
401 | struct rq *rq = this_rq(); |
402 | ||
2d513868 | 403 | irqtime_account_process_tick(current, 0, rq, ticks); |
73fbec60 FW |
404 | } |
405 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
3f4724ea FW |
406 | static inline void irqtime_account_idle_ticks(int ticks) {} |
407 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 408 | struct rq *rq, int nr_ticks) {} |
73fbec60 FW |
409 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
410 | ||
73fbec60 FW |
411 | /* |
412 | * Use precise platform statistics if available: | |
413 | */ | |
414 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
a7e1a9e3 | 415 | |
e3942ba0 | 416 | #ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
b0493406 | 417 | void vtime_common_task_switch(struct task_struct *prev) |
e3942ba0 FW |
418 | { |
419 | if (is_idle_task(prev)) | |
420 | vtime_account_idle(prev); | |
421 | else | |
422 | vtime_account_system(prev); | |
423 | ||
c8d7dabf | 424 | vtime_flush(prev); |
e3942ba0 FW |
425 | arch_vtime_task_switch(prev); |
426 | } | |
427 | #endif | |
11113334 | 428 | |
0cfdf9a1 FW |
429 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ |
430 | ||
431 | ||
432 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE | |
a7e1a9e3 FW |
433 | /* |
434 | * Archs that account the whole time spent in the idle task | |
435 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 436 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
437 | * have other meaning of the idle time (s390 only includes the |
438 | * time spent by the CPU when it's in low power mode) must override | |
439 | * vtime_account(). | |
440 | */ | |
441 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
0cfdf9a1 | 442 | void vtime_account_irq_enter(struct task_struct *tsk) |
a7e1a9e3 | 443 | { |
0cfdf9a1 FW |
444 | if (!in_interrupt() && is_idle_task(tsk)) |
445 | vtime_account_idle(tsk); | |
446 | else | |
447 | vtime_account_system(tsk); | |
a7e1a9e3 | 448 | } |
0cfdf9a1 | 449 | EXPORT_SYMBOL_GPL(vtime_account_irq_enter); |
a7e1a9e3 | 450 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ |
9fbc42ea | 451 | |
5613fda9 | 452 | void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
9fbc42ea FW |
453 | { |
454 | *ut = p->utime; | |
455 | *st = p->stime; | |
456 | } | |
9eec50b8 | 457 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
a7e1a9e3 | 458 | |
5613fda9 | 459 | void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st) |
9fbc42ea FW |
460 | { |
461 | struct task_cputime cputime; | |
73fbec60 | 462 | |
9fbc42ea FW |
463 | thread_group_cputime(p, &cputime); |
464 | ||
465 | *ut = cputime.utime; | |
466 | *st = cputime.stime; | |
467 | } | |
468 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ | |
469 | /* | |
470 | * Account a single tick of cpu time. | |
471 | * @p: the process that the cpu time gets accounted to | |
472 | * @user_tick: indicates if the tick is a user or a system tick | |
473 | */ | |
474 | void account_process_tick(struct task_struct *p, int user_tick) | |
73fbec60 | 475 | { |
981ee2d4 | 476 | cputime_t cputime, steal; |
9fbc42ea | 477 | struct rq *rq = this_rq(); |
73fbec60 | 478 | |
55dbdcfa | 479 | if (vtime_accounting_cpu_enabled()) |
9fbc42ea FW |
480 | return; |
481 | ||
482 | if (sched_clock_irqtime) { | |
2d513868 | 483 | irqtime_account_process_tick(p, user_tick, rq, 1); |
9fbc42ea FW |
484 | return; |
485 | } | |
486 | ||
57430218 | 487 | cputime = cputime_one_jiffy; |
03cbc732 | 488 | steal = steal_account_process_time(ULONG_MAX); |
57430218 RR |
489 | |
490 | if (steal >= cputime) | |
9fbc42ea | 491 | return; |
73fbec60 | 492 | |
57430218 | 493 | cputime -= steal; |
57430218 | 494 | |
9fbc42ea | 495 | if (user_tick) |
40565b5a | 496 | account_user_time(p, cputime); |
9fbc42ea | 497 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
40565b5a | 498 | account_system_time(p, HARDIRQ_OFFSET, cputime); |
73fbec60 | 499 | else |
57430218 | 500 | account_idle_time(cputime); |
9fbc42ea | 501 | } |
73fbec60 | 502 | |
9fbc42ea FW |
503 | /* |
504 | * Account multiple ticks of idle time. | |
505 | * @ticks: number of stolen ticks | |
506 | */ | |
507 | void account_idle_ticks(unsigned long ticks) | |
508 | { | |
f9bcf1e0 | 509 | cputime_t cputime, steal; |
26f2c75c | 510 | |
9fbc42ea FW |
511 | if (sched_clock_irqtime) { |
512 | irqtime_account_idle_ticks(ticks); | |
513 | return; | |
514 | } | |
515 | ||
26f2c75c | 516 | cputime = jiffies_to_cputime(ticks); |
03cbc732 | 517 | steal = steal_account_process_time(ULONG_MAX); |
f9bcf1e0 WL |
518 | |
519 | if (steal >= cputime) | |
520 | return; | |
521 | ||
522 | cputime -= steal; | |
523 | account_idle_time(cputime); | |
9fbc42ea | 524 | } |
73fbec60 | 525 | |
d9a3c982 | 526 | /* |
55eaa7c1 SG |
527 | * Perform (stime * rtime) / total, but avoid multiplication overflow by |
528 | * loosing precision when the numbers are big. | |
d9a3c982 | 529 | */ |
5613fda9 | 530 | static u64 scale_stime(u64 stime, u64 rtime, u64 total) |
73fbec60 | 531 | { |
55eaa7c1 | 532 | u64 scaled; |
73fbec60 | 533 | |
55eaa7c1 SG |
534 | for (;;) { |
535 | /* Make sure "rtime" is the bigger of stime/rtime */ | |
84f9f3a1 SG |
536 | if (stime > rtime) |
537 | swap(rtime, stime); | |
55eaa7c1 SG |
538 | |
539 | /* Make sure 'total' fits in 32 bits */ | |
540 | if (total >> 32) | |
541 | goto drop_precision; | |
542 | ||
543 | /* Does rtime (and thus stime) fit in 32 bits? */ | |
544 | if (!(rtime >> 32)) | |
545 | break; | |
546 | ||
547 | /* Can we just balance rtime/stime rather than dropping bits? */ | |
548 | if (stime >> 31) | |
549 | goto drop_precision; | |
550 | ||
551 | /* We can grow stime and shrink rtime and try to make them both fit */ | |
552 | stime <<= 1; | |
553 | rtime >>= 1; | |
554 | continue; | |
555 | ||
556 | drop_precision: | |
557 | /* We drop from rtime, it has more bits than stime */ | |
558 | rtime >>= 1; | |
559 | total >>= 1; | |
d9a3c982 | 560 | } |
73fbec60 | 561 | |
55eaa7c1 SG |
562 | /* |
563 | * Make sure gcc understands that this is a 32x32->64 multiply, | |
564 | * followed by a 64/32->64 divide. | |
565 | */ | |
566 | scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total); | |
5613fda9 | 567 | return scaled; |
73fbec60 FW |
568 | } |
569 | ||
347abad9 | 570 | /* |
9d7fb042 PZ |
571 | * Adjust tick based cputime random precision against scheduler runtime |
572 | * accounting. | |
347abad9 | 573 | * |
9d7fb042 PZ |
574 | * Tick based cputime accounting depend on random scheduling timeslices of a |
575 | * task to be interrupted or not by the timer. Depending on these | |
576 | * circumstances, the number of these interrupts may be over or | |
577 | * under-optimistic, matching the real user and system cputime with a variable | |
578 | * precision. | |
579 | * | |
580 | * Fix this by scaling these tick based values against the total runtime | |
581 | * accounted by the CFS scheduler. | |
582 | * | |
583 | * This code provides the following guarantees: | |
584 | * | |
585 | * stime + utime == rtime | |
586 | * stime_i+1 >= stime_i, utime_i+1 >= utime_i | |
587 | * | |
588 | * Assuming that rtime_i+1 >= rtime_i. | |
fa092057 | 589 | */ |
d37f761d | 590 | static void cputime_adjust(struct task_cputime *curr, |
9d7fb042 | 591 | struct prev_cputime *prev, |
5613fda9 | 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 PZ |
615 | /* |
616 | * If either stime or both stime and utime are 0, assume all runtime is | |
617 | * userspace. Once a task gets some ticks, the monotonicy code at | |
618 | * 'update' will ensure things converge to the observed ratio. | |
619 | */ | |
620 | if (stime == 0) { | |
621 | utime = rtime; | |
9d7fb042 PZ |
622 | goto update; |
623 | } | |
5a8e01f8 | 624 | |
173be9a1 PZ |
625 | if (utime == 0) { |
626 | stime = rtime; | |
9d7fb042 | 627 | goto update; |
d9a3c982 | 628 | } |
73fbec60 | 629 | |
5613fda9 | 630 | stime = scale_stime(stime, rtime, stime + utime); |
9d7fb042 | 631 | |
173be9a1 | 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 | |
ff9a9b4c | 684 | static cputime_t vtime_delta(struct task_struct *tsk) |
6a61671b | 685 | { |
ff9a9b4c | 686 | unsigned long now = READ_ONCE(jiffies); |
6a61671b | 687 | |
ff9a9b4c | 688 | if (time_before(now, (unsigned long)tsk->vtime_snap)) |
6a61671b | 689 | return 0; |
abf917cd | 690 | |
ff9a9b4c | 691 | return jiffies_to_cputime(now - tsk->vtime_snap); |
6a61671b FW |
692 | } |
693 | ||
694 | static cputime_t get_vtime_delta(struct task_struct *tsk) | |
abf917cd | 695 | { |
ff9a9b4c | 696 | unsigned long now = READ_ONCE(jiffies); |
b58c3584 | 697 | cputime_t delta, other; |
abf917cd | 698 | |
03cbc732 WL |
699 | /* |
700 | * Unlike tick based timing, vtime based timing never has lost | |
701 | * ticks, and no need for steal time accounting to make up for | |
702 | * lost ticks. Vtime accounts a rounded version of actual | |
703 | * elapsed time. Limit account_other_time to prevent rounding | |
704 | * errors from causing elapsed vtime to go negative. | |
705 | */ | |
57430218 | 706 | delta = jiffies_to_cputime(now - tsk->vtime_snap); |
b58c3584 | 707 | other = account_other_time(delta); |
7098c1ea | 708 | WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_INACTIVE); |
ff9a9b4c | 709 | tsk->vtime_snap = now; |
abf917cd | 710 | |
b58c3584 | 711 | return delta - other; |
abf917cd FW |
712 | } |
713 | ||
6a61671b FW |
714 | static void __vtime_account_system(struct task_struct *tsk) |
715 | { | |
716 | cputime_t delta_cpu = get_vtime_delta(tsk); | |
717 | ||
40565b5a | 718 | account_system_time(tsk, irq_count(), delta_cpu); |
6a61671b FW |
719 | } |
720 | ||
abf917cd FW |
721 | void vtime_account_system(struct task_struct *tsk) |
722 | { | |
ff9a9b4c RR |
723 | if (!vtime_delta(tsk)) |
724 | return; | |
725 | ||
b7ce2277 | 726 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b | 727 | __vtime_account_system(tsk); |
b7ce2277 | 728 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b | 729 | } |
3f4724ea | 730 | |
abf917cd FW |
731 | void vtime_account_user(struct task_struct *tsk) |
732 | { | |
3f4724ea FW |
733 | cputime_t delta_cpu; |
734 | ||
b7ce2277 | 735 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b | 736 | tsk->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c RR |
737 | if (vtime_delta(tsk)) { |
738 | delta_cpu = get_vtime_delta(tsk); | |
40565b5a | 739 | account_user_time(tsk, delta_cpu); |
ff9a9b4c | 740 | } |
b7ce2277 | 741 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b FW |
742 | } |
743 | ||
744 | void vtime_user_enter(struct task_struct *tsk) | |
745 | { | |
b7ce2277 | 746 | write_seqcount_begin(&tsk->vtime_seqcount); |
ff9a9b4c RR |
747 | if (vtime_delta(tsk)) |
748 | __vtime_account_system(tsk); | |
af2350bd | 749 | tsk->vtime_snap_whence = VTIME_USER; |
b7ce2277 | 750 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b FW |
751 | } |
752 | ||
753 | void vtime_guest_enter(struct task_struct *tsk) | |
754 | { | |
5b206d48 FW |
755 | /* |
756 | * The flags must be updated under the lock with | |
757 | * the vtime_snap flush and update. | |
758 | * That enforces a right ordering and update sequence | |
759 | * synchronization against the reader (task_gtime()) | |
760 | * that can thus safely catch up with a tickless delta. | |
761 | */ | |
b7ce2277 | 762 | write_seqcount_begin(&tsk->vtime_seqcount); |
ff9a9b4c RR |
763 | if (vtime_delta(tsk)) |
764 | __vtime_account_system(tsk); | |
6a61671b | 765 | current->flags |= PF_VCPU; |
b7ce2277 | 766 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b | 767 | } |
48d6a816 | 768 | EXPORT_SYMBOL_GPL(vtime_guest_enter); |
6a61671b FW |
769 | |
770 | void vtime_guest_exit(struct task_struct *tsk) | |
771 | { | |
b7ce2277 | 772 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b FW |
773 | __vtime_account_system(tsk); |
774 | current->flags &= ~PF_VCPU; | |
b7ce2277 | 775 | write_seqcount_end(&tsk->vtime_seqcount); |
abf917cd | 776 | } |
48d6a816 | 777 | EXPORT_SYMBOL_GPL(vtime_guest_exit); |
abf917cd FW |
778 | |
779 | void vtime_account_idle(struct task_struct *tsk) | |
780 | { | |
6a61671b | 781 | cputime_t delta_cpu = get_vtime_delta(tsk); |
abf917cd FW |
782 | |
783 | account_idle_time(delta_cpu); | |
784 | } | |
3f4724ea | 785 | |
6a61671b FW |
786 | void arch_vtime_task_switch(struct task_struct *prev) |
787 | { | |
b7ce2277 | 788 | write_seqcount_begin(&prev->vtime_seqcount); |
7098c1ea | 789 | prev->vtime_snap_whence = VTIME_INACTIVE; |
b7ce2277 | 790 | write_seqcount_end(&prev->vtime_seqcount); |
6a61671b | 791 | |
b7ce2277 | 792 | write_seqcount_begin(¤t->vtime_seqcount); |
6a61671b | 793 | current->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c | 794 | current->vtime_snap = jiffies; |
b7ce2277 | 795 | write_seqcount_end(¤t->vtime_seqcount); |
6a61671b FW |
796 | } |
797 | ||
45eacc69 | 798 | void vtime_init_idle(struct task_struct *t, int cpu) |
6a61671b FW |
799 | { |
800 | unsigned long flags; | |
801 | ||
b7ce2277 FW |
802 | local_irq_save(flags); |
803 | write_seqcount_begin(&t->vtime_seqcount); | |
6a61671b | 804 | t->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c | 805 | t->vtime_snap = jiffies; |
b7ce2277 FW |
806 | write_seqcount_end(&t->vtime_seqcount); |
807 | local_irq_restore(flags); | |
6a61671b FW |
808 | } |
809 | ||
16a6d9be | 810 | u64 task_gtime(struct task_struct *t) |
6a61671b | 811 | { |
6a61671b | 812 | unsigned int seq; |
16a6d9be | 813 | u64 gtime; |
6a61671b | 814 | |
e5925394 | 815 | if (!vtime_accounting_enabled()) |
2541117b HS |
816 | return t->gtime; |
817 | ||
6a61671b | 818 | do { |
b7ce2277 | 819 | seq = read_seqcount_begin(&t->vtime_seqcount); |
6a61671b FW |
820 | |
821 | gtime = t->gtime; | |
cab245d6 | 822 | if (t->vtime_snap_whence == VTIME_SYS && t->flags & PF_VCPU) |
16a6d9be | 823 | gtime += cputime_to_nsecs(vtime_delta(t)); |
6a61671b | 824 | |
b7ce2277 | 825 | } while (read_seqcount_retry(&t->vtime_seqcount, seq)); |
6a61671b FW |
826 | |
827 | return gtime; | |
828 | } | |
829 | ||
830 | /* | |
831 | * Fetch cputime raw values from fields of task_struct and | |
832 | * add up the pending nohz execution time since the last | |
833 | * cputime snapshot. | |
834 | */ | |
5613fda9 | 835 | void task_cputime(struct task_struct *t, u64 *utime, u64 *stime) |
6a61671b | 836 | { |
5613fda9 | 837 | u64 delta; |
6a61671b | 838 | unsigned int seq; |
6a61671b | 839 | |
353c50eb SG |
840 | if (!vtime_accounting_enabled()) { |
841 | *utime = t->utime; | |
842 | *stime = t->stime; | |
843 | return; | |
844 | } | |
6a61671b | 845 | |
353c50eb | 846 | do { |
b7ce2277 | 847 | seq = read_seqcount_begin(&t->vtime_seqcount); |
6a61671b | 848 | |
353c50eb SG |
849 | *utime = t->utime; |
850 | *stime = t->stime; | |
6a61671b FW |
851 | |
852 | /* Task is sleeping, nothing to add */ | |
353c50eb | 853 | if (t->vtime_snap_whence == VTIME_INACTIVE || is_idle_task(t)) |
6a61671b FW |
854 | continue; |
855 | ||
5613fda9 | 856 | delta = cputime_to_nsecs(vtime_delta(t)); |
6a61671b FW |
857 | |
858 | /* | |
859 | * Task runs either in user or kernel space, add pending nohz time to | |
860 | * the right place. | |
861 | */ | |
353c50eb SG |
862 | if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) |
863 | *utime += delta; | |
864 | else if (t->vtime_snap_whence == VTIME_SYS) | |
865 | *stime += delta; | |
b7ce2277 | 866 | } while (read_seqcount_retry(&t->vtime_seqcount, seq)); |
6a61671b | 867 | } |
abf917cd | 868 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |