]>
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); |
73fbec60 FW |
47 | s64 delta; |
48 | int cpu; | |
49 | ||
50 | if (!sched_clock_irqtime) | |
51 | return; | |
52 | ||
73fbec60 | 53 | cpu = smp_processor_id(); |
19d23dbf FW |
54 | delta = sched_clock_cpu(cpu) - irqtime->irq_start_time; |
55 | irqtime->irq_start_time += delta; | |
73fbec60 | 56 | |
19d23dbf | 57 | u64_stats_update_begin(&irqtime->sync); |
73fbec60 FW |
58 | /* |
59 | * We do not account for softirq time from ksoftirqd here. | |
60 | * We want to continue accounting softirq time to ksoftirqd thread | |
61 | * in that case, so as not to confuse scheduler with a special task | |
62 | * that do not consume any time, but still wants to run. | |
63 | */ | |
64 | if (hardirq_count()) | |
19d23dbf | 65 | irqtime->hardirq_time += delta; |
73fbec60 | 66 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
19d23dbf | 67 | irqtime->softirq_time += delta; |
73fbec60 | 68 | |
19d23dbf | 69 | u64_stats_update_end(&irqtime->sync); |
73fbec60 | 70 | } |
3e1df4f5 | 71 | EXPORT_SYMBOL_GPL(irqtime_account_irq); |
73fbec60 | 72 | |
447976ef | 73 | static cputime_t irqtime_account_update(u64 irqtime, int idx, cputime_t maxtime) |
73fbec60 FW |
74 | { |
75 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
57430218 | 76 | cputime_t irq_cputime; |
73fbec60 | 77 | |
447976ef | 78 | irq_cputime = nsecs_to_cputime64(irqtime) - cpustat[idx]; |
57430218 | 79 | irq_cputime = min(irq_cputime, maxtime); |
447976ef | 80 | cpustat[idx] += irq_cputime; |
2810f611 | 81 | |
57430218 | 82 | return irq_cputime; |
73fbec60 FW |
83 | } |
84 | ||
447976ef | 85 | static cputime_t irqtime_account_hi_update(cputime_t maxtime) |
73fbec60 | 86 | { |
447976ef FW |
87 | return irqtime_account_update(__this_cpu_read(cpu_irqtime.hardirq_time), |
88 | CPUTIME_IRQ, maxtime); | |
89 | } | |
2810f611 | 90 | |
447976ef FW |
91 | static cputime_t irqtime_account_si_update(cputime_t maxtime) |
92 | { | |
93 | return irqtime_account_update(__this_cpu_read(cpu_irqtime.softirq_time), | |
94 | CPUTIME_SOFTIRQ, maxtime); | |
73fbec60 FW |
95 | } |
96 | ||
97 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
98 | ||
99 | #define sched_clock_irqtime (0) | |
100 | ||
57430218 RR |
101 | static cputime_t irqtime_account_hi_update(cputime_t dummy) |
102 | { | |
103 | return 0; | |
104 | } | |
105 | ||
106 | static cputime_t irqtime_account_si_update(cputime_t dummy) | |
107 | { | |
108 | return 0; | |
109 | } | |
110 | ||
73fbec60 FW |
111 | #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */ |
112 | ||
113 | static inline void task_group_account_field(struct task_struct *p, int index, | |
114 | u64 tmp) | |
115 | { | |
73fbec60 FW |
116 | /* |
117 | * Since all updates are sure to touch the root cgroup, we | |
118 | * get ourselves ahead and touch it first. If the root cgroup | |
119 | * is the only cgroup, then nothing else should be necessary. | |
120 | * | |
121 | */ | |
a4f61cc0 | 122 | __this_cpu_add(kernel_cpustat.cpustat[index], tmp); |
73fbec60 | 123 | |
1966aaf7 | 124 | cpuacct_account_field(p, index, tmp); |
73fbec60 FW |
125 | } |
126 | ||
127 | /* | |
128 | * Account user cpu time to a process. | |
129 | * @p: the process that the cpu time gets accounted to | |
130 | * @cputime: the cpu time spent in user space since the last update | |
131 | * @cputime_scaled: cputime scaled by cpu frequency | |
132 | */ | |
133 | void account_user_time(struct task_struct *p, cputime_t cputime, | |
134 | cputime_t cputime_scaled) | |
135 | { | |
136 | int index; | |
137 | ||
138 | /* Add user time to process. */ | |
139 | p->utime += cputime; | |
140 | p->utimescaled += cputime_scaled; | |
141 | account_group_user_time(p, cputime); | |
142 | ||
d0ea0268 | 143 | index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; |
73fbec60 FW |
144 | |
145 | /* Add user time to cpustat. */ | |
146 | task_group_account_field(p, index, (__force u64) cputime); | |
147 | ||
148 | /* Account for user time used */ | |
6fac4829 | 149 | acct_account_cputime(p); |
73fbec60 FW |
150 | } |
151 | ||
152 | /* | |
153 | * Account guest cpu time to a process. | |
154 | * @p: the process that the cpu time gets accounted to | |
155 | * @cputime: the cpu time spent in virtual machine since the last update | |
156 | * @cputime_scaled: cputime scaled by cpu frequency | |
157 | */ | |
158 | static void account_guest_time(struct task_struct *p, cputime_t cputime, | |
159 | cputime_t cputime_scaled) | |
160 | { | |
161 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
162 | ||
163 | /* Add guest time to process. */ | |
164 | p->utime += cputime; | |
165 | p->utimescaled += cputime_scaled; | |
166 | account_group_user_time(p, cputime); | |
167 | p->gtime += cputime; | |
168 | ||
169 | /* Add guest time to cpustat. */ | |
d0ea0268 | 170 | if (task_nice(p) > 0) { |
73fbec60 FW |
171 | cpustat[CPUTIME_NICE] += (__force u64) cputime; |
172 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
173 | } else { | |
174 | cpustat[CPUTIME_USER] += (__force u64) cputime; | |
175 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
176 | } | |
177 | } | |
178 | ||
179 | /* | |
180 | * Account system cpu time to a process and desired cpustat field | |
181 | * @p: the process that the cpu time gets accounted to | |
182 | * @cputime: the cpu time spent in kernel space since the last update | |
183 | * @cputime_scaled: cputime scaled by cpu frequency | |
184 | * @target_cputime64: pointer to cpustat field that has to be updated | |
185 | */ | |
186 | static inline | |
187 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
188 | cputime_t cputime_scaled, int index) | |
189 | { | |
190 | /* Add system time to process. */ | |
191 | p->stime += cputime; | |
192 | p->stimescaled += cputime_scaled; | |
193 | account_group_system_time(p, cputime); | |
194 | ||
195 | /* Add system time to cpustat. */ | |
196 | task_group_account_field(p, index, (__force u64) cputime); | |
197 | ||
198 | /* Account for system time used */ | |
6fac4829 | 199 | acct_account_cputime(p); |
73fbec60 FW |
200 | } |
201 | ||
202 | /* | |
203 | * Account system cpu time to a process. | |
204 | * @p: the process that the cpu time gets accounted to | |
205 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
206 | * @cputime: the cpu time spent in kernel space since the last update | |
207 | * @cputime_scaled: cputime scaled by cpu frequency | |
208 | */ | |
209 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
210 | cputime_t cputime, cputime_t cputime_scaled) | |
211 | { | |
212 | int index; | |
213 | ||
214 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { | |
215 | account_guest_time(p, cputime, cputime_scaled); | |
216 | return; | |
217 | } | |
218 | ||
219 | if (hardirq_count() - hardirq_offset) | |
220 | index = CPUTIME_IRQ; | |
221 | else if (in_serving_softirq()) | |
222 | index = CPUTIME_SOFTIRQ; | |
223 | else | |
224 | index = CPUTIME_SYSTEM; | |
225 | ||
226 | __account_system_time(p, cputime, cputime_scaled, index); | |
227 | } | |
228 | ||
229 | /* | |
230 | * Account for involuntary wait time. | |
231 | * @cputime: the cpu time spent in involuntary wait | |
232 | */ | |
233 | void account_steal_time(cputime_t cputime) | |
234 | { | |
235 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
236 | ||
237 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; | |
238 | } | |
239 | ||
240 | /* | |
241 | * Account for idle time. | |
242 | * @cputime: the cpu time spent in idle wait | |
243 | */ | |
244 | void account_idle_time(cputime_t cputime) | |
245 | { | |
246 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
247 | struct rq *rq = this_rq(); | |
248 | ||
249 | if (atomic_read(&rq->nr_iowait) > 0) | |
250 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; | |
251 | else | |
252 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; | |
253 | } | |
254 | ||
03cbc732 WL |
255 | /* |
256 | * When a guest is interrupted for a longer amount of time, missed clock | |
257 | * ticks are not redelivered later. Due to that, this function may on | |
258 | * occasion account more time than the calling functions think elapsed. | |
259 | */ | |
57430218 | 260 | static __always_inline cputime_t steal_account_process_time(cputime_t maxtime) |
73fbec60 FW |
261 | { |
262 | #ifdef CONFIG_PARAVIRT | |
263 | if (static_key_false(¶virt_steal_enabled)) { | |
57430218 | 264 | cputime_t steal_cputime; |
dee08a72 | 265 | u64 steal; |
73fbec60 FW |
266 | |
267 | steal = paravirt_steal_clock(smp_processor_id()); | |
268 | steal -= this_rq()->prev_steal_time; | |
269 | ||
57430218 RR |
270 | steal_cputime = min(nsecs_to_cputime(steal), maxtime); |
271 | account_steal_time(steal_cputime); | |
272 | this_rq()->prev_steal_time += cputime_to_nsecs(steal_cputime); | |
73fbec60 | 273 | |
57430218 | 274 | return steal_cputime; |
73fbec60 FW |
275 | } |
276 | #endif | |
807e5b80 | 277 | return 0; |
73fbec60 FW |
278 | } |
279 | ||
57430218 RR |
280 | /* |
281 | * Account how much elapsed time was spent in steal, irq, or softirq time. | |
282 | */ | |
283 | static inline cputime_t account_other_time(cputime_t max) | |
284 | { | |
285 | cputime_t accounted; | |
286 | ||
2810f611 FW |
287 | /* Shall be converted to a lockdep-enabled lightweight check */ |
288 | WARN_ON_ONCE(!irqs_disabled()); | |
289 | ||
57430218 RR |
290 | accounted = steal_account_process_time(max); |
291 | ||
292 | if (accounted < max) | |
293 | accounted += irqtime_account_hi_update(max - accounted); | |
294 | ||
295 | if (accounted < max) | |
296 | accounted += irqtime_account_si_update(max - accounted); | |
297 | ||
298 | return accounted; | |
299 | } | |
300 | ||
a1eb1411 SG |
301 | #ifdef CONFIG_64BIT |
302 | static inline u64 read_sum_exec_runtime(struct task_struct *t) | |
303 | { | |
304 | return t->se.sum_exec_runtime; | |
305 | } | |
306 | #else | |
307 | static u64 read_sum_exec_runtime(struct task_struct *t) | |
308 | { | |
309 | u64 ns; | |
310 | struct rq_flags rf; | |
311 | struct rq *rq; | |
312 | ||
313 | rq = task_rq_lock(t, &rf); | |
314 | ns = t->se.sum_exec_runtime; | |
315 | task_rq_unlock(rq, t, &rf); | |
316 | ||
317 | return ns; | |
318 | } | |
319 | #endif | |
320 | ||
a634f933 FW |
321 | /* |
322 | * Accumulate raw cputime values of dead tasks (sig->[us]time) and live | |
323 | * tasks (sum on group iteration) belonging to @tsk's group. | |
324 | */ | |
325 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) | |
326 | { | |
327 | struct signal_struct *sig = tsk->signal; | |
6fac4829 | 328 | cputime_t utime, stime; |
a634f933 | 329 | struct task_struct *t; |
e78c3496 | 330 | unsigned int seq, nextseq; |
9c368b5b | 331 | unsigned long flags; |
a634f933 | 332 | |
a1eb1411 SG |
333 | /* |
334 | * Update current task runtime to account pending time since last | |
335 | * scheduler action or thread_group_cputime() call. This thread group | |
336 | * might have other running tasks on different CPUs, but updating | |
337 | * their runtime can affect syscall performance, so we skip account | |
338 | * those pending times and rely only on values updated on tick or | |
339 | * other scheduler action. | |
340 | */ | |
341 | if (same_thread_group(current, tsk)) | |
342 | (void) task_sched_runtime(current); | |
343 | ||
a634f933 | 344 | rcu_read_lock(); |
e78c3496 RR |
345 | /* Attempt a lockless read on the first round. */ |
346 | nextseq = 0; | |
347 | do { | |
348 | seq = nextseq; | |
9c368b5b | 349 | flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq); |
e78c3496 RR |
350 | times->utime = sig->utime; |
351 | times->stime = sig->stime; | |
352 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
353 | ||
354 | for_each_thread(tsk, t) { | |
355 | task_cputime(t, &utime, &stime); | |
356 | times->utime += utime; | |
357 | times->stime += stime; | |
a1eb1411 | 358 | times->sum_exec_runtime += read_sum_exec_runtime(t); |
e78c3496 RR |
359 | } |
360 | /* If lockless access failed, take the lock. */ | |
361 | nextseq = 1; | |
362 | } while (need_seqretry(&sig->stats_lock, seq)); | |
9c368b5b | 363 | done_seqretry_irqrestore(&sig->stats_lock, seq, flags); |
a634f933 FW |
364 | rcu_read_unlock(); |
365 | } | |
366 | ||
73fbec60 FW |
367 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
368 | /* | |
369 | * Account a tick to a process and cpustat | |
370 | * @p: the process that the cpu time gets accounted to | |
371 | * @user_tick: is the tick from userspace | |
372 | * @rq: the pointer to rq | |
373 | * | |
374 | * Tick demultiplexing follows the order | |
375 | * - pending hardirq update | |
376 | * - pending softirq update | |
377 | * - user_time | |
378 | * - idle_time | |
379 | * - system time | |
380 | * - check for guest_time | |
381 | * - else account as system_time | |
382 | * | |
383 | * Check for hardirq is done both for system and user time as there is | |
384 | * no timer going off while we are on hardirq and hence we may never get an | |
385 | * opportunity to update it solely in system time. | |
386 | * p->stime and friends are only updated on system time and not on irq | |
387 | * softirq as those do not count in task exec_runtime any more. | |
388 | */ | |
389 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 390 | struct rq *rq, int ticks) |
73fbec60 | 391 | { |
57430218 RR |
392 | u64 cputime = (__force u64) cputime_one_jiffy * ticks; |
393 | cputime_t scaled, other; | |
73fbec60 | 394 | |
57430218 RR |
395 | /* |
396 | * When returning from idle, many ticks can get accounted at | |
397 | * once, including some ticks of steal, irq, and softirq time. | |
398 | * Subtract those ticks from the amount of time accounted to | |
399 | * idle, or potentially user or system time. Due to rounding, | |
400 | * other time can exceed ticks occasionally. | |
401 | */ | |
03cbc732 | 402 | other = account_other_time(ULONG_MAX); |
57430218 | 403 | if (other >= cputime) |
73fbec60 | 404 | return; |
57430218 RR |
405 | cputime -= other; |
406 | scaled = cputime_to_scaled(cputime); | |
73fbec60 | 407 | |
57430218 | 408 | if (this_cpu_ksoftirqd() == p) { |
73fbec60 FW |
409 | /* |
410 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
411 | * So, we have to handle it separately here. | |
412 | * Also, p->stime needs to be updated for ksoftirqd. | |
413 | */ | |
2d513868 | 414 | __account_system_time(p, cputime, scaled, CPUTIME_SOFTIRQ); |
73fbec60 | 415 | } else if (user_tick) { |
2d513868 | 416 | account_user_time(p, cputime, scaled); |
73fbec60 | 417 | } else if (p == rq->idle) { |
2d513868 | 418 | account_idle_time(cputime); |
73fbec60 | 419 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ |
2d513868 | 420 | account_guest_time(p, cputime, scaled); |
73fbec60 | 421 | } else { |
2d513868 | 422 | __account_system_time(p, cputime, scaled, CPUTIME_SYSTEM); |
73fbec60 FW |
423 | } |
424 | } | |
425 | ||
426 | static void irqtime_account_idle_ticks(int ticks) | |
427 | { | |
73fbec60 FW |
428 | struct rq *rq = this_rq(); |
429 | ||
2d513868 | 430 | irqtime_account_process_tick(current, 0, rq, ticks); |
73fbec60 FW |
431 | } |
432 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
3f4724ea FW |
433 | static inline void irqtime_account_idle_ticks(int ticks) {} |
434 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
2d513868 | 435 | struct rq *rq, int nr_ticks) {} |
73fbec60 FW |
436 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
437 | ||
73fbec60 FW |
438 | /* |
439 | * Use precise platform statistics if available: | |
440 | */ | |
441 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
a7e1a9e3 | 442 | |
e3942ba0 | 443 | #ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
b0493406 | 444 | void vtime_common_task_switch(struct task_struct *prev) |
e3942ba0 FW |
445 | { |
446 | if (is_idle_task(prev)) | |
447 | vtime_account_idle(prev); | |
448 | else | |
449 | vtime_account_system(prev); | |
450 | ||
abf917cd | 451 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
e3942ba0 | 452 | vtime_account_user(prev); |
abf917cd | 453 | #endif |
e3942ba0 FW |
454 | arch_vtime_task_switch(prev); |
455 | } | |
456 | #endif | |
11113334 | 457 | |
0cfdf9a1 FW |
458 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ |
459 | ||
460 | ||
461 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE | |
a7e1a9e3 FW |
462 | /* |
463 | * Archs that account the whole time spent in the idle task | |
464 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 465 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
466 | * have other meaning of the idle time (s390 only includes the |
467 | * time spent by the CPU when it's in low power mode) must override | |
468 | * vtime_account(). | |
469 | */ | |
470 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
0cfdf9a1 | 471 | void vtime_account_irq_enter(struct task_struct *tsk) |
a7e1a9e3 | 472 | { |
0cfdf9a1 FW |
473 | if (!in_interrupt() && is_idle_task(tsk)) |
474 | vtime_account_idle(tsk); | |
475 | else | |
476 | vtime_account_system(tsk); | |
a7e1a9e3 | 477 | } |
0cfdf9a1 | 478 | EXPORT_SYMBOL_GPL(vtime_account_irq_enter); |
a7e1a9e3 | 479 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ |
9fbc42ea | 480 | |
9fbc42ea FW |
481 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
482 | { | |
483 | *ut = p->utime; | |
484 | *st = p->stime; | |
485 | } | |
9eec50b8 | 486 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
a7e1a9e3 | 487 | |
9fbc42ea FW |
488 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
489 | { | |
490 | struct task_cputime cputime; | |
73fbec60 | 491 | |
9fbc42ea FW |
492 | thread_group_cputime(p, &cputime); |
493 | ||
494 | *ut = cputime.utime; | |
495 | *st = cputime.stime; | |
496 | } | |
497 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ | |
498 | /* | |
499 | * Account a single tick of cpu time. | |
500 | * @p: the process that the cpu time gets accounted to | |
501 | * @user_tick: indicates if the tick is a user or a system tick | |
502 | */ | |
503 | void account_process_tick(struct task_struct *p, int user_tick) | |
73fbec60 | 504 | { |
57430218 | 505 | cputime_t cputime, scaled, steal; |
9fbc42ea | 506 | struct rq *rq = this_rq(); |
73fbec60 | 507 | |
55dbdcfa | 508 | if (vtime_accounting_cpu_enabled()) |
9fbc42ea FW |
509 | return; |
510 | ||
511 | if (sched_clock_irqtime) { | |
2d513868 | 512 | irqtime_account_process_tick(p, user_tick, rq, 1); |
9fbc42ea FW |
513 | return; |
514 | } | |
515 | ||
57430218 | 516 | cputime = cputime_one_jiffy; |
03cbc732 | 517 | steal = steal_account_process_time(ULONG_MAX); |
57430218 RR |
518 | |
519 | if (steal >= cputime) | |
9fbc42ea | 520 | return; |
73fbec60 | 521 | |
57430218 RR |
522 | cputime -= steal; |
523 | scaled = cputime_to_scaled(cputime); | |
524 | ||
9fbc42ea | 525 | if (user_tick) |
57430218 | 526 | account_user_time(p, cputime, scaled); |
9fbc42ea | 527 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
57430218 | 528 | account_system_time(p, HARDIRQ_OFFSET, cputime, scaled); |
73fbec60 | 529 | else |
57430218 | 530 | account_idle_time(cputime); |
9fbc42ea | 531 | } |
73fbec60 | 532 | |
9fbc42ea FW |
533 | /* |
534 | * Account multiple ticks of idle time. | |
535 | * @ticks: number of stolen ticks | |
536 | */ | |
537 | void account_idle_ticks(unsigned long ticks) | |
538 | { | |
f9bcf1e0 | 539 | cputime_t cputime, steal; |
26f2c75c | 540 | |
9fbc42ea FW |
541 | if (sched_clock_irqtime) { |
542 | irqtime_account_idle_ticks(ticks); | |
543 | return; | |
544 | } | |
545 | ||
26f2c75c | 546 | cputime = jiffies_to_cputime(ticks); |
03cbc732 | 547 | steal = steal_account_process_time(ULONG_MAX); |
f9bcf1e0 WL |
548 | |
549 | if (steal >= cputime) | |
550 | return; | |
551 | ||
552 | cputime -= steal; | |
553 | account_idle_time(cputime); | |
9fbc42ea | 554 | } |
73fbec60 | 555 | |
d9a3c982 | 556 | /* |
55eaa7c1 SG |
557 | * Perform (stime * rtime) / total, but avoid multiplication overflow by |
558 | * loosing precision when the numbers are big. | |
d9a3c982 FW |
559 | */ |
560 | static cputime_t scale_stime(u64 stime, u64 rtime, u64 total) | |
73fbec60 | 561 | { |
55eaa7c1 | 562 | u64 scaled; |
73fbec60 | 563 | |
55eaa7c1 SG |
564 | for (;;) { |
565 | /* Make sure "rtime" is the bigger of stime/rtime */ | |
84f9f3a1 SG |
566 | if (stime > rtime) |
567 | swap(rtime, stime); | |
55eaa7c1 SG |
568 | |
569 | /* Make sure 'total' fits in 32 bits */ | |
570 | if (total >> 32) | |
571 | goto drop_precision; | |
572 | ||
573 | /* Does rtime (and thus stime) fit in 32 bits? */ | |
574 | if (!(rtime >> 32)) | |
575 | break; | |
576 | ||
577 | /* Can we just balance rtime/stime rather than dropping bits? */ | |
578 | if (stime >> 31) | |
579 | goto drop_precision; | |
580 | ||
581 | /* We can grow stime and shrink rtime and try to make them both fit */ | |
582 | stime <<= 1; | |
583 | rtime >>= 1; | |
584 | continue; | |
585 | ||
586 | drop_precision: | |
587 | /* We drop from rtime, it has more bits than stime */ | |
588 | rtime >>= 1; | |
589 | total >>= 1; | |
d9a3c982 | 590 | } |
73fbec60 | 591 | |
55eaa7c1 SG |
592 | /* |
593 | * Make sure gcc understands that this is a 32x32->64 multiply, | |
594 | * followed by a 64/32->64 divide. | |
595 | */ | |
596 | scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total); | |
d9a3c982 | 597 | return (__force cputime_t) scaled; |
73fbec60 FW |
598 | } |
599 | ||
347abad9 | 600 | /* |
9d7fb042 PZ |
601 | * Adjust tick based cputime random precision against scheduler runtime |
602 | * accounting. | |
347abad9 | 603 | * |
9d7fb042 PZ |
604 | * Tick based cputime accounting depend on random scheduling timeslices of a |
605 | * task to be interrupted or not by the timer. Depending on these | |
606 | * circumstances, the number of these interrupts may be over or | |
607 | * under-optimistic, matching the real user and system cputime with a variable | |
608 | * precision. | |
609 | * | |
610 | * Fix this by scaling these tick based values against the total runtime | |
611 | * accounted by the CFS scheduler. | |
612 | * | |
613 | * This code provides the following guarantees: | |
614 | * | |
615 | * stime + utime == rtime | |
616 | * stime_i+1 >= stime_i, utime_i+1 >= utime_i | |
617 | * | |
618 | * Assuming that rtime_i+1 >= rtime_i. | |
fa092057 | 619 | */ |
d37f761d | 620 | static void cputime_adjust(struct task_cputime *curr, |
9d7fb042 | 621 | struct prev_cputime *prev, |
d37f761d | 622 | cputime_t *ut, cputime_t *st) |
73fbec60 | 623 | { |
5a8e01f8 | 624 | cputime_t rtime, stime, utime; |
9d7fb042 | 625 | unsigned long flags; |
fa092057 | 626 | |
9d7fb042 PZ |
627 | /* Serialize concurrent callers such that we can honour our guarantees */ |
628 | raw_spin_lock_irqsave(&prev->lock, flags); | |
d37f761d | 629 | rtime = nsecs_to_cputime(curr->sum_exec_runtime); |
73fbec60 | 630 | |
772c808a | 631 | /* |
9d7fb042 PZ |
632 | * This is possible under two circumstances: |
633 | * - rtime isn't monotonic after all (a bug); | |
634 | * - we got reordered by the lock. | |
635 | * | |
636 | * In both cases this acts as a filter such that the rest of the code | |
637 | * can assume it is monotonic regardless of anything else. | |
772c808a SG |
638 | */ |
639 | if (prev->stime + prev->utime >= rtime) | |
640 | goto out; | |
641 | ||
5a8e01f8 SG |
642 | stime = curr->stime; |
643 | utime = curr->utime; | |
644 | ||
173be9a1 PZ |
645 | /* |
646 | * If either stime or both stime and utime are 0, assume all runtime is | |
647 | * userspace. Once a task gets some ticks, the monotonicy code at | |
648 | * 'update' will ensure things converge to the observed ratio. | |
649 | */ | |
650 | if (stime == 0) { | |
651 | utime = rtime; | |
9d7fb042 PZ |
652 | goto update; |
653 | } | |
5a8e01f8 | 654 | |
173be9a1 PZ |
655 | if (utime == 0) { |
656 | stime = rtime; | |
9d7fb042 | 657 | goto update; |
d9a3c982 | 658 | } |
73fbec60 | 659 | |
9d7fb042 PZ |
660 | stime = scale_stime((__force u64)stime, (__force u64)rtime, |
661 | (__force u64)(stime + utime)); | |
662 | ||
173be9a1 | 663 | update: |
9d7fb042 PZ |
664 | /* |
665 | * Make sure stime doesn't go backwards; this preserves monotonicity | |
666 | * for utime because rtime is monotonic. | |
667 | * | |
668 | * utime_i+1 = rtime_i+1 - stime_i | |
669 | * = rtime_i+1 - (rtime_i - utime_i) | |
670 | * = (rtime_i+1 - rtime_i) + utime_i | |
671 | * >= utime_i | |
672 | */ | |
673 | if (stime < prev->stime) | |
674 | stime = prev->stime; | |
675 | utime = rtime - stime; | |
676 | ||
677 | /* | |
678 | * Make sure utime doesn't go backwards; this still preserves | |
679 | * monotonicity for stime, analogous argument to above. | |
680 | */ | |
681 | if (utime < prev->utime) { | |
682 | utime = prev->utime; | |
683 | stime = rtime - utime; | |
684 | } | |
d37f761d | 685 | |
9d7fb042 PZ |
686 | prev->stime = stime; |
687 | prev->utime = utime; | |
772c808a | 688 | out: |
d37f761d FW |
689 | *ut = prev->utime; |
690 | *st = prev->stime; | |
9d7fb042 | 691 | raw_spin_unlock_irqrestore(&prev->lock, flags); |
d37f761d | 692 | } |
73fbec60 | 693 | |
d37f761d FW |
694 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
695 | { | |
696 | struct task_cputime cputime = { | |
d37f761d FW |
697 | .sum_exec_runtime = p->se.sum_exec_runtime, |
698 | }; | |
699 | ||
6fac4829 | 700 | task_cputime(p, &cputime.utime, &cputime.stime); |
d37f761d | 701 | cputime_adjust(&cputime, &p->prev_cputime, ut, st); |
73fbec60 | 702 | } |
9eec50b8 | 703 | EXPORT_SYMBOL_GPL(task_cputime_adjusted); |
73fbec60 | 704 | |
e80d0a1a | 705 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 | 706 | { |
73fbec60 | 707 | struct task_cputime cputime; |
73fbec60 FW |
708 | |
709 | thread_group_cputime(p, &cputime); | |
d37f761d | 710 | cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); |
73fbec60 | 711 | } |
9fbc42ea | 712 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ |
abf917cd FW |
713 | |
714 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN | |
ff9a9b4c | 715 | static cputime_t vtime_delta(struct task_struct *tsk) |
6a61671b | 716 | { |
ff9a9b4c | 717 | unsigned long now = READ_ONCE(jiffies); |
6a61671b | 718 | |
ff9a9b4c | 719 | if (time_before(now, (unsigned long)tsk->vtime_snap)) |
6a61671b | 720 | return 0; |
abf917cd | 721 | |
ff9a9b4c | 722 | return jiffies_to_cputime(now - tsk->vtime_snap); |
6a61671b FW |
723 | } |
724 | ||
725 | static cputime_t get_vtime_delta(struct task_struct *tsk) | |
abf917cd | 726 | { |
ff9a9b4c | 727 | unsigned long now = READ_ONCE(jiffies); |
b58c3584 | 728 | cputime_t delta, other; |
abf917cd | 729 | |
03cbc732 WL |
730 | /* |
731 | * Unlike tick based timing, vtime based timing never has lost | |
732 | * ticks, and no need for steal time accounting to make up for | |
733 | * lost ticks. Vtime accounts a rounded version of actual | |
734 | * elapsed time. Limit account_other_time to prevent rounding | |
735 | * errors from causing elapsed vtime to go negative. | |
736 | */ | |
57430218 | 737 | delta = jiffies_to_cputime(now - tsk->vtime_snap); |
b58c3584 | 738 | other = account_other_time(delta); |
7098c1ea | 739 | WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_INACTIVE); |
ff9a9b4c | 740 | tsk->vtime_snap = now; |
abf917cd | 741 | |
b58c3584 | 742 | return delta - other; |
abf917cd FW |
743 | } |
744 | ||
6a61671b FW |
745 | static void __vtime_account_system(struct task_struct *tsk) |
746 | { | |
747 | cputime_t delta_cpu = get_vtime_delta(tsk); | |
748 | ||
749 | account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu)); | |
750 | } | |
751 | ||
abf917cd FW |
752 | void vtime_account_system(struct task_struct *tsk) |
753 | { | |
ff9a9b4c RR |
754 | if (!vtime_delta(tsk)) |
755 | return; | |
756 | ||
b7ce2277 | 757 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b | 758 | __vtime_account_system(tsk); |
b7ce2277 | 759 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b | 760 | } |
3f4724ea | 761 | |
abf917cd FW |
762 | void vtime_account_user(struct task_struct *tsk) |
763 | { | |
3f4724ea FW |
764 | cputime_t delta_cpu; |
765 | ||
b7ce2277 | 766 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b | 767 | tsk->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c RR |
768 | if (vtime_delta(tsk)) { |
769 | delta_cpu = get_vtime_delta(tsk); | |
770 | account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu)); | |
771 | } | |
b7ce2277 | 772 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b FW |
773 | } |
774 | ||
775 | void vtime_user_enter(struct task_struct *tsk) | |
776 | { | |
b7ce2277 | 777 | write_seqcount_begin(&tsk->vtime_seqcount); |
ff9a9b4c RR |
778 | if (vtime_delta(tsk)) |
779 | __vtime_account_system(tsk); | |
af2350bd | 780 | tsk->vtime_snap_whence = VTIME_USER; |
b7ce2277 | 781 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b FW |
782 | } |
783 | ||
784 | void vtime_guest_enter(struct task_struct *tsk) | |
785 | { | |
5b206d48 FW |
786 | /* |
787 | * The flags must be updated under the lock with | |
788 | * the vtime_snap flush and update. | |
789 | * That enforces a right ordering and update sequence | |
790 | * synchronization against the reader (task_gtime()) | |
791 | * that can thus safely catch up with a tickless delta. | |
792 | */ | |
b7ce2277 | 793 | write_seqcount_begin(&tsk->vtime_seqcount); |
ff9a9b4c RR |
794 | if (vtime_delta(tsk)) |
795 | __vtime_account_system(tsk); | |
6a61671b | 796 | current->flags |= PF_VCPU; |
b7ce2277 | 797 | write_seqcount_end(&tsk->vtime_seqcount); |
6a61671b | 798 | } |
48d6a816 | 799 | EXPORT_SYMBOL_GPL(vtime_guest_enter); |
6a61671b FW |
800 | |
801 | void vtime_guest_exit(struct task_struct *tsk) | |
802 | { | |
b7ce2277 | 803 | write_seqcount_begin(&tsk->vtime_seqcount); |
6a61671b FW |
804 | __vtime_account_system(tsk); |
805 | current->flags &= ~PF_VCPU; | |
b7ce2277 | 806 | write_seqcount_end(&tsk->vtime_seqcount); |
abf917cd | 807 | } |
48d6a816 | 808 | EXPORT_SYMBOL_GPL(vtime_guest_exit); |
abf917cd FW |
809 | |
810 | void vtime_account_idle(struct task_struct *tsk) | |
811 | { | |
6a61671b | 812 | cputime_t delta_cpu = get_vtime_delta(tsk); |
abf917cd FW |
813 | |
814 | account_idle_time(delta_cpu); | |
815 | } | |
3f4724ea | 816 | |
6a61671b FW |
817 | void arch_vtime_task_switch(struct task_struct *prev) |
818 | { | |
b7ce2277 | 819 | write_seqcount_begin(&prev->vtime_seqcount); |
7098c1ea | 820 | prev->vtime_snap_whence = VTIME_INACTIVE; |
b7ce2277 | 821 | write_seqcount_end(&prev->vtime_seqcount); |
6a61671b | 822 | |
b7ce2277 | 823 | write_seqcount_begin(¤t->vtime_seqcount); |
6a61671b | 824 | current->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c | 825 | current->vtime_snap = jiffies; |
b7ce2277 | 826 | write_seqcount_end(¤t->vtime_seqcount); |
6a61671b FW |
827 | } |
828 | ||
45eacc69 | 829 | void vtime_init_idle(struct task_struct *t, int cpu) |
6a61671b FW |
830 | { |
831 | unsigned long flags; | |
832 | ||
b7ce2277 FW |
833 | local_irq_save(flags); |
834 | write_seqcount_begin(&t->vtime_seqcount); | |
6a61671b | 835 | t->vtime_snap_whence = VTIME_SYS; |
ff9a9b4c | 836 | t->vtime_snap = jiffies; |
b7ce2277 FW |
837 | write_seqcount_end(&t->vtime_seqcount); |
838 | local_irq_restore(flags); | |
6a61671b FW |
839 | } |
840 | ||
841 | cputime_t task_gtime(struct task_struct *t) | |
842 | { | |
6a61671b FW |
843 | unsigned int seq; |
844 | cputime_t gtime; | |
845 | ||
e5925394 | 846 | if (!vtime_accounting_enabled()) |
2541117b HS |
847 | return t->gtime; |
848 | ||
6a61671b | 849 | do { |
b7ce2277 | 850 | seq = read_seqcount_begin(&t->vtime_seqcount); |
6a61671b FW |
851 | |
852 | gtime = t->gtime; | |
cab245d6 | 853 | if (t->vtime_snap_whence == VTIME_SYS && t->flags & PF_VCPU) |
6a61671b FW |
854 | gtime += vtime_delta(t); |
855 | ||
b7ce2277 | 856 | } while (read_seqcount_retry(&t->vtime_seqcount, seq)); |
6a61671b FW |
857 | |
858 | return gtime; | |
859 | } | |
860 | ||
861 | /* | |
862 | * Fetch cputime raw values from fields of task_struct and | |
863 | * add up the pending nohz execution time since the last | |
864 | * cputime snapshot. | |
865 | */ | |
866 | static void | |
867 | fetch_task_cputime(struct task_struct *t, | |
868 | cputime_t *u_dst, cputime_t *s_dst, | |
869 | cputime_t *u_src, cputime_t *s_src, | |
870 | cputime_t *udelta, cputime_t *sdelta) | |
871 | { | |
6a61671b FW |
872 | unsigned int seq; |
873 | unsigned long long delta; | |
874 | ||
875 | do { | |
876 | *udelta = 0; | |
877 | *sdelta = 0; | |
878 | ||
b7ce2277 | 879 | seq = read_seqcount_begin(&t->vtime_seqcount); |
6a61671b FW |
880 | |
881 | if (u_dst) | |
882 | *u_dst = *u_src; | |
883 | if (s_dst) | |
884 | *s_dst = *s_src; | |
885 | ||
886 | /* Task is sleeping, nothing to add */ | |
7098c1ea | 887 | if (t->vtime_snap_whence == VTIME_INACTIVE || |
6a61671b FW |
888 | is_idle_task(t)) |
889 | continue; | |
890 | ||
891 | delta = vtime_delta(t); | |
892 | ||
893 | /* | |
894 | * Task runs either in user or kernel space, add pending nohz time to | |
895 | * the right place. | |
896 | */ | |
897 | if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) { | |
898 | *udelta = delta; | |
899 | } else { | |
900 | if (t->vtime_snap_whence == VTIME_SYS) | |
901 | *sdelta = delta; | |
902 | } | |
b7ce2277 | 903 | } while (read_seqcount_retry(&t->vtime_seqcount, seq)); |
6a61671b FW |
904 | } |
905 | ||
906 | ||
907 | void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime) | |
908 | { | |
909 | cputime_t udelta, sdelta; | |
910 | ||
e5925394 | 911 | if (!vtime_accounting_enabled()) { |
7877a0ba HS |
912 | if (utime) |
913 | *utime = t->utime; | |
914 | if (stime) | |
915 | *stime = t->stime; | |
916 | return; | |
917 | } | |
918 | ||
6a61671b FW |
919 | fetch_task_cputime(t, utime, stime, &t->utime, |
920 | &t->stime, &udelta, &sdelta); | |
921 | if (utime) | |
922 | *utime += udelta; | |
923 | if (stime) | |
924 | *stime += sdelta; | |
925 | } | |
926 | ||
927 | void task_cputime_scaled(struct task_struct *t, | |
928 | cputime_t *utimescaled, cputime_t *stimescaled) | |
929 | { | |
930 | cputime_t udelta, sdelta; | |
931 | ||
e5925394 | 932 | if (!vtime_accounting_enabled()) { |
7877a0ba HS |
933 | if (utimescaled) |
934 | *utimescaled = t->utimescaled; | |
935 | if (stimescaled) | |
936 | *stimescaled = t->stimescaled; | |
937 | return; | |
938 | } | |
939 | ||
6a61671b FW |
940 | fetch_task_cputime(t, utimescaled, stimescaled, |
941 | &t->utimescaled, &t->stimescaled, &udelta, &sdelta); | |
942 | if (utimescaled) | |
943 | *utimescaled += cputime_to_scaled(udelta); | |
944 | if (stimescaled) | |
945 | *stimescaled += cputime_to_scaled(sdelta); | |
946 | } | |
abf917cd | 947 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |