cputime.c 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838
  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>
  6. #include <linux/context_tracking.h>
  7. #include "sched.h"
  8. #ifdef CONFIG_IRQ_TIME_ACCOUNTING
  9. /*
  10. * There are no locks covering percpu hardirq/softirq time.
  11. * They are only modified in vtime_account, on corresponding CPU
  12. * with interrupts disabled. So, writes are safe.
  13. * They are read and saved off onto struct rq in update_rq_clock().
  14. * This may result in other CPU reading this CPU's irq time and can
  15. * race with irq/vtime_account on this CPU. We would either get old
  16. * or new value with a side effect of accounting a slice of irq time to wrong
  17. * task when irq is in progress while we read rq->clock. That is a worthy
  18. * compromise in place of having locks on each irq in account_system_time.
  19. */
  20. DEFINE_PER_CPU(u64, cpu_hardirq_time);
  21. DEFINE_PER_CPU(u64, cpu_softirq_time);
  22. static DEFINE_PER_CPU(u64, irq_start_time);
  23. static int sched_clock_irqtime;
  24. void enable_sched_clock_irqtime(void)
  25. {
  26. sched_clock_irqtime = 1;
  27. }
  28. void disable_sched_clock_irqtime(void)
  29. {
  30. sched_clock_irqtime = 0;
  31. }
  32. #ifndef CONFIG_64BIT
  33. DEFINE_PER_CPU(seqcount_t, irq_time_seq);
  34. #endif /* CONFIG_64BIT */
  35. /*
  36. * Called before incrementing preempt_count on {soft,}irq_enter
  37. * and before decrementing preempt_count on {soft,}irq_exit.
  38. */
  39. void irqtime_account_irq(struct task_struct *curr)
  40. {
  41. unsigned long flags;
  42. s64 delta;
  43. int cpu;
  44. if (!sched_clock_irqtime)
  45. return;
  46. local_irq_save(flags);
  47. cpu = smp_processor_id();
  48. delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time);
  49. __this_cpu_add(irq_start_time, delta);
  50. irq_time_write_begin();
  51. /*
  52. * We do not account for softirq time from ksoftirqd here.
  53. * We want to continue accounting softirq time to ksoftirqd thread
  54. * in that case, so as not to confuse scheduler with a special task
  55. * that do not consume any time, but still wants to run.
  56. */
  57. if (hardirq_count())
  58. __this_cpu_add(cpu_hardirq_time, delta);
  59. else if (in_serving_softirq() && curr != this_cpu_ksoftirqd())
  60. __this_cpu_add(cpu_softirq_time, delta);
  61. irq_time_write_end();
  62. local_irq_restore(flags);
  63. }
  64. EXPORT_SYMBOL_GPL(irqtime_account_irq);
  65. static int irqtime_account_hi_update(void)
  66. {
  67. u64 *cpustat = kcpustat_this_cpu->cpustat;
  68. unsigned long flags;
  69. u64 latest_ns;
  70. int ret = 0;
  71. local_irq_save(flags);
  72. latest_ns = this_cpu_read(cpu_hardirq_time);
  73. if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ])
  74. ret = 1;
  75. local_irq_restore(flags);
  76. return ret;
  77. }
  78. static int irqtime_account_si_update(void)
  79. {
  80. u64 *cpustat = kcpustat_this_cpu->cpustat;
  81. unsigned long flags;
  82. u64 latest_ns;
  83. int ret = 0;
  84. local_irq_save(flags);
  85. latest_ns = this_cpu_read(cpu_softirq_time);
  86. if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ])
  87. ret = 1;
  88. local_irq_restore(flags);
  89. return ret;
  90. }
  91. #else /* CONFIG_IRQ_TIME_ACCOUNTING */
  92. #define sched_clock_irqtime (0)
  93. #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
  94. static inline void task_group_account_field(struct task_struct *p, int index,
  95. u64 tmp)
  96. {
  97. /*
  98. * Since all updates are sure to touch the root cgroup, we
  99. * get ourselves ahead and touch it first. If the root cgroup
  100. * is the only cgroup, then nothing else should be necessary.
  101. *
  102. */
  103. __this_cpu_add(kernel_cpustat.cpustat[index], tmp);
  104. cpuacct_account_field(p, index, tmp);
  105. }
  106. /*
  107. * Account user cpu time to a process.
  108. * @p: the process that the cpu time gets accounted to
  109. * @cputime: the cpu time spent in user space since the last update
  110. * @cputime_scaled: cputime scaled by cpu frequency
  111. */
  112. void account_user_time(struct task_struct *p, cputime_t cputime,
  113. cputime_t cputime_scaled)
  114. {
  115. int index;
  116. /* Add user time to process. */
  117. p->utime += cputime;
  118. p->utimescaled += cputime_scaled;
  119. account_group_user_time(p, cputime);
  120. index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER;
  121. /* Add user time to cpustat. */
  122. task_group_account_field(p, index, (__force u64) cputime);
  123. /* Account for user time used */
  124. acct_account_cputime(p);
  125. }
  126. /*
  127. * Account guest cpu time to a process.
  128. * @p: the process that the cpu time gets accounted to
  129. * @cputime: the cpu time spent in virtual machine since the last update
  130. * @cputime_scaled: cputime scaled by cpu frequency
  131. */
  132. static void account_guest_time(struct task_struct *p, cputime_t cputime,
  133. cputime_t cputime_scaled)
  134. {
  135. u64 *cpustat = kcpustat_this_cpu->cpustat;
  136. /* Add guest time to process. */
  137. p->utime += cputime;
  138. p->utimescaled += cputime_scaled;
  139. account_group_user_time(p, cputime);
  140. p->gtime += cputime;
  141. /* Add guest time to cpustat. */
  142. if (task_nice(p) > 0) {
  143. cpustat[CPUTIME_NICE] += (__force u64) cputime;
  144. cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime;
  145. } else {
  146. cpustat[CPUTIME_USER] += (__force u64) cputime;
  147. cpustat[CPUTIME_GUEST] += (__force u64) cputime;
  148. }
  149. }
  150. /*
  151. * Account system cpu time to a process and desired cpustat field
  152. * @p: the process that the cpu time gets accounted to
  153. * @cputime: the cpu time spent in kernel space since the last update
  154. * @cputime_scaled: cputime scaled by cpu frequency
  155. * @target_cputime64: pointer to cpustat field that has to be updated
  156. */
  157. static inline
  158. void __account_system_time(struct task_struct *p, cputime_t cputime,
  159. cputime_t cputime_scaled, int index)
  160. {
  161. /* Add system time to process. */
  162. p->stime += cputime;
  163. p->stimescaled += cputime_scaled;
  164. account_group_system_time(p, cputime);
  165. /* Add system time to cpustat. */
  166. task_group_account_field(p, index, (__force u64) cputime);
  167. /* Account for system time used */
  168. acct_account_cputime(p);
  169. }
  170. /*
  171. * Account system cpu time to a process.
  172. * @p: the process that the cpu time gets accounted to
  173. * @hardirq_offset: the offset to subtract from hardirq_count()
  174. * @cputime: the cpu time spent in kernel space since the last update
  175. * @cputime_scaled: cputime scaled by cpu frequency
  176. */
  177. void account_system_time(struct task_struct *p, int hardirq_offset,
  178. cputime_t cputime, cputime_t cputime_scaled)
  179. {
  180. int index;
  181. if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
  182. account_guest_time(p, cputime, cputime_scaled);
  183. return;
  184. }
  185. if (hardirq_count() - hardirq_offset)
  186. index = CPUTIME_IRQ;
  187. else if (in_serving_softirq())
  188. index = CPUTIME_SOFTIRQ;
  189. else
  190. index = CPUTIME_SYSTEM;
  191. __account_system_time(p, cputime, cputime_scaled, index);
  192. }
  193. /*
  194. * Account for involuntary wait time.
  195. * @cputime: the cpu time spent in involuntary wait
  196. */
  197. void account_steal_time(cputime_t cputime)
  198. {
  199. u64 *cpustat = kcpustat_this_cpu->cpustat;
  200. cpustat[CPUTIME_STEAL] += (__force u64) cputime;
  201. }
  202. /*
  203. * Account for idle time.
  204. * @cputime: the cpu time spent in idle wait
  205. */
  206. void account_idle_time(cputime_t cputime)
  207. {
  208. u64 *cpustat = kcpustat_this_cpu->cpustat;
  209. struct rq *rq = this_rq();
  210. if (atomic_read(&rq->nr_iowait) > 0)
  211. cpustat[CPUTIME_IOWAIT] += (__force u64) cputime;
  212. else
  213. cpustat[CPUTIME_IDLE] += (__force u64) cputime;
  214. }
  215. static __always_inline bool steal_account_process_tick(void)
  216. {
  217. #ifdef CONFIG_PARAVIRT
  218. if (static_key_false(&paravirt_steal_enabled)) {
  219. u64 steal;
  220. cputime_t steal_ct;
  221. steal = paravirt_steal_clock(smp_processor_id());
  222. steal -= this_rq()->prev_steal_time;
  223. /*
  224. * cputime_t may be less precise than nsecs (eg: if it's
  225. * based on jiffies). Lets cast the result to cputime
  226. * granularity and account the rest on the next rounds.
  227. */
  228. steal_ct = nsecs_to_cputime(steal);
  229. this_rq()->prev_steal_time += cputime_to_nsecs(steal_ct);
  230. account_steal_time(steal_ct);
  231. return steal_ct;
  232. }
  233. #endif
  234. return false;
  235. }
  236. /*
  237. * Accumulate raw cputime values of dead tasks (sig->[us]time) and live
  238. * tasks (sum on group iteration) belonging to @tsk's group.
  239. */
  240. void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
  241. {
  242. struct signal_struct *sig = tsk->signal;
  243. cputime_t utime, stime;
  244. struct task_struct *t;
  245. times->utime = sig->utime;
  246. times->stime = sig->stime;
  247. times->sum_exec_runtime = sig->sum_sched_runtime;
  248. rcu_read_lock();
  249. /* make sure we can trust tsk->thread_group list */
  250. if (!likely(pid_alive(tsk)))
  251. goto out;
  252. t = tsk;
  253. do {
  254. task_cputime(t, &utime, &stime);
  255. times->utime += utime;
  256. times->stime += stime;
  257. times->sum_exec_runtime += task_sched_runtime(t);
  258. } while_each_thread(tsk, t);
  259. out:
  260. rcu_read_unlock();
  261. }
  262. #ifdef CONFIG_IRQ_TIME_ACCOUNTING
  263. /*
  264. * Account a tick to a process and cpustat
  265. * @p: the process that the cpu time gets accounted to
  266. * @user_tick: is the tick from userspace
  267. * @rq: the pointer to rq
  268. *
  269. * Tick demultiplexing follows the order
  270. * - pending hardirq update
  271. * - pending softirq update
  272. * - user_time
  273. * - idle_time
  274. * - system time
  275. * - check for guest_time
  276. * - else account as system_time
  277. *
  278. * Check for hardirq is done both for system and user time as there is
  279. * no timer going off while we are on hardirq and hence we may never get an
  280. * opportunity to update it solely in system time.
  281. * p->stime and friends are only updated on system time and not on irq
  282. * softirq as those do not count in task exec_runtime any more.
  283. */
  284. static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
  285. struct rq *rq)
  286. {
  287. cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
  288. u64 *cpustat = kcpustat_this_cpu->cpustat;
  289. if (steal_account_process_tick())
  290. return;
  291. if (irqtime_account_hi_update()) {
  292. cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy;
  293. } else if (irqtime_account_si_update()) {
  294. cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy;
  295. } else if (this_cpu_ksoftirqd() == p) {
  296. /*
  297. * ksoftirqd time do not get accounted in cpu_softirq_time.
  298. * So, we have to handle it separately here.
  299. * Also, p->stime needs to be updated for ksoftirqd.
  300. */
  301. __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
  302. CPUTIME_SOFTIRQ);
  303. } else if (user_tick) {
  304. account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
  305. } else if (p == rq->idle) {
  306. account_idle_time(cputime_one_jiffy);
  307. } else if (p->flags & PF_VCPU) { /* System time or guest time */
  308. account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled);
  309. } else {
  310. __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
  311. CPUTIME_SYSTEM);
  312. }
  313. }
  314. static void irqtime_account_idle_ticks(int ticks)
  315. {
  316. int i;
  317. struct rq *rq = this_rq();
  318. for (i = 0; i < ticks; i++)
  319. irqtime_account_process_tick(current, 0, rq);
  320. }
  321. #else /* CONFIG_IRQ_TIME_ACCOUNTING */
  322. static inline void irqtime_account_idle_ticks(int ticks) {}
  323. static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick,
  324. struct rq *rq) {}
  325. #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
  326. /*
  327. * Use precise platform statistics if available:
  328. */
  329. #ifdef CONFIG_VIRT_CPU_ACCOUNTING
  330. #ifndef __ARCH_HAS_VTIME_TASK_SWITCH
  331. void vtime_common_task_switch(struct task_struct *prev)
  332. {
  333. if (is_idle_task(prev))
  334. vtime_account_idle(prev);
  335. else
  336. vtime_account_system(prev);
  337. #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
  338. vtime_account_user(prev);
  339. #endif
  340. arch_vtime_task_switch(prev);
  341. }
  342. #endif
  343. /*
  344. * Archs that account the whole time spent in the idle task
  345. * (outside irq) as idle time can rely on this and just implement
  346. * vtime_account_system() and vtime_account_idle(). Archs that
  347. * have other meaning of the idle time (s390 only includes the
  348. * time spent by the CPU when it's in low power mode) must override
  349. * vtime_account().
  350. */
  351. #ifndef __ARCH_HAS_VTIME_ACCOUNT
  352. void vtime_common_account_irq_enter(struct task_struct *tsk)
  353. {
  354. if (!in_interrupt()) {
  355. /*
  356. * If we interrupted user, context_tracking_in_user()
  357. * is 1 because the context tracking don't hook
  358. * on irq entry/exit. This way we know if
  359. * we need to flush user time on kernel entry.
  360. */
  361. if (context_tracking_in_user()) {
  362. vtime_account_user(tsk);
  363. return;
  364. }
  365. if (is_idle_task(tsk)) {
  366. vtime_account_idle(tsk);
  367. return;
  368. }
  369. }
  370. vtime_account_system(tsk);
  371. }
  372. EXPORT_SYMBOL_GPL(vtime_common_account_irq_enter);
  373. #endif /* __ARCH_HAS_VTIME_ACCOUNT */
  374. #endif /* CONFIG_VIRT_CPU_ACCOUNTING */
  375. #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
  376. void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
  377. {
  378. *ut = p->utime;
  379. *st = p->stime;
  380. }
  381. void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
  382. {
  383. struct task_cputime cputime;
  384. thread_group_cputime(p, &cputime);
  385. *ut = cputime.utime;
  386. *st = cputime.stime;
  387. }
  388. #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
  389. /*
  390. * Account a single tick of cpu time.
  391. * @p: the process that the cpu time gets accounted to
  392. * @user_tick: indicates if the tick is a user or a system tick
  393. */
  394. void account_process_tick(struct task_struct *p, int user_tick)
  395. {
  396. cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
  397. struct rq *rq = this_rq();
  398. if (vtime_accounting_enabled())
  399. return;
  400. if (sched_clock_irqtime) {
  401. irqtime_account_process_tick(p, user_tick, rq);
  402. return;
  403. }
  404. if (steal_account_process_tick())
  405. return;
  406. if (user_tick)
  407. account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
  408. else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
  409. account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
  410. one_jiffy_scaled);
  411. else
  412. account_idle_time(cputime_one_jiffy);
  413. }
  414. /*
  415. * Account multiple ticks of steal time.
  416. * @p: the process from which the cpu time has been stolen
  417. * @ticks: number of stolen ticks
  418. */
  419. void account_steal_ticks(unsigned long ticks)
  420. {
  421. account_steal_time(jiffies_to_cputime(ticks));
  422. }
  423. /*
  424. * Account multiple ticks of idle time.
  425. * @ticks: number of stolen ticks
  426. */
  427. void account_idle_ticks(unsigned long ticks)
  428. {
  429. if (sched_clock_irqtime) {
  430. irqtime_account_idle_ticks(ticks);
  431. return;
  432. }
  433. account_idle_time(jiffies_to_cputime(ticks));
  434. }
  435. /*
  436. * Perform (stime * rtime) / total, but avoid multiplication overflow by
  437. * loosing precision when the numbers are big.
  438. */
  439. static cputime_t scale_stime(u64 stime, u64 rtime, u64 total)
  440. {
  441. u64 scaled;
  442. for (;;) {
  443. /* Make sure "rtime" is the bigger of stime/rtime */
  444. if (stime > rtime)
  445. swap(rtime, stime);
  446. /* Make sure 'total' fits in 32 bits */
  447. if (total >> 32)
  448. goto drop_precision;
  449. /* Does rtime (and thus stime) fit in 32 bits? */
  450. if (!(rtime >> 32))
  451. break;
  452. /* Can we just balance rtime/stime rather than dropping bits? */
  453. if (stime >> 31)
  454. goto drop_precision;
  455. /* We can grow stime and shrink rtime and try to make them both fit */
  456. stime <<= 1;
  457. rtime >>= 1;
  458. continue;
  459. drop_precision:
  460. /* We drop from rtime, it has more bits than stime */
  461. rtime >>= 1;
  462. total >>= 1;
  463. }
  464. /*
  465. * Make sure gcc understands that this is a 32x32->64 multiply,
  466. * followed by a 64/32->64 divide.
  467. */
  468. scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total);
  469. return (__force cputime_t) scaled;
  470. }
  471. /*
  472. * Adjust tick based cputime random precision against scheduler
  473. * runtime accounting.
  474. */
  475. static void cputime_adjust(struct task_cputime *curr,
  476. struct cputime *prev,
  477. cputime_t *ut, cputime_t *st)
  478. {
  479. cputime_t rtime, stime, utime;
  480. /*
  481. * Tick based cputime accounting depend on random scheduling
  482. * timeslices of a task to be interrupted or not by the timer.
  483. * Depending on these circumstances, the number of these interrupts
  484. * may be over or under-optimistic, matching the real user and system
  485. * cputime with a variable precision.
  486. *
  487. * Fix this by scaling these tick based values against the total
  488. * runtime accounted by the CFS scheduler.
  489. */
  490. rtime = nsecs_to_cputime(curr->sum_exec_runtime);
  491. /*
  492. * Update userspace visible utime/stime values only if actual execution
  493. * time is bigger than already exported. Note that can happen, that we
  494. * provided bigger values due to scaling inaccuracy on big numbers.
  495. */
  496. if (prev->stime + prev->utime >= rtime)
  497. goto out;
  498. stime = curr->stime;
  499. utime = curr->utime;
  500. if (utime == 0) {
  501. stime = rtime;
  502. } else if (stime == 0) {
  503. utime = rtime;
  504. } else {
  505. cputime_t total = stime + utime;
  506. stime = scale_stime((__force u64)stime,
  507. (__force u64)rtime, (__force u64)total);
  508. utime = rtime - stime;
  509. }
  510. /*
  511. * If the tick based count grows faster than the scheduler one,
  512. * the result of the scaling may go backward.
  513. * Let's enforce monotonicity.
  514. */
  515. prev->stime = max(prev->stime, stime);
  516. prev->utime = max(prev->utime, utime);
  517. out:
  518. *ut = prev->utime;
  519. *st = prev->stime;
  520. }
  521. void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
  522. {
  523. struct task_cputime cputime = {
  524. .sum_exec_runtime = p->se.sum_exec_runtime,
  525. };
  526. task_cputime(p, &cputime.utime, &cputime.stime);
  527. cputime_adjust(&cputime, &p->prev_cputime, ut, st);
  528. }
  529. /*
  530. * Must be called with siglock held.
  531. */
  532. void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
  533. {
  534. struct task_cputime cputime;
  535. thread_group_cputime(p, &cputime);
  536. cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st);
  537. }
  538. #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
  539. #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
  540. static unsigned long long vtime_delta(struct task_struct *tsk)
  541. {
  542. unsigned long long clock;
  543. clock = local_clock();
  544. if (clock < tsk->vtime_snap)
  545. return 0;
  546. return clock - tsk->vtime_snap;
  547. }
  548. static cputime_t get_vtime_delta(struct task_struct *tsk)
  549. {
  550. unsigned long long delta = vtime_delta(tsk);
  551. WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_SLEEPING);
  552. tsk->vtime_snap += delta;
  553. /* CHECKME: always safe to convert nsecs to cputime? */
  554. return nsecs_to_cputime(delta);
  555. }
  556. static void __vtime_account_system(struct task_struct *tsk)
  557. {
  558. cputime_t delta_cpu = get_vtime_delta(tsk);
  559. account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu));
  560. }
  561. void vtime_account_system(struct task_struct *tsk)
  562. {
  563. write_seqlock(&tsk->vtime_seqlock);
  564. __vtime_account_system(tsk);
  565. write_sequnlock(&tsk->vtime_seqlock);
  566. }
  567. void vtime_gen_account_irq_exit(struct task_struct *tsk)
  568. {
  569. write_seqlock(&tsk->vtime_seqlock);
  570. __vtime_account_system(tsk);
  571. if (context_tracking_in_user())
  572. tsk->vtime_snap_whence = VTIME_USER;
  573. write_sequnlock(&tsk->vtime_seqlock);
  574. }
  575. void vtime_account_user(struct task_struct *tsk)
  576. {
  577. cputime_t delta_cpu;
  578. write_seqlock(&tsk->vtime_seqlock);
  579. delta_cpu = get_vtime_delta(tsk);
  580. tsk->vtime_snap_whence = VTIME_SYS;
  581. account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu));
  582. write_sequnlock(&tsk->vtime_seqlock);
  583. }
  584. void vtime_user_enter(struct task_struct *tsk)
  585. {
  586. write_seqlock(&tsk->vtime_seqlock);
  587. __vtime_account_system(tsk);
  588. tsk->vtime_snap_whence = VTIME_USER;
  589. write_sequnlock(&tsk->vtime_seqlock);
  590. }
  591. void vtime_guest_enter(struct task_struct *tsk)
  592. {
  593. /*
  594. * The flags must be updated under the lock with
  595. * the vtime_snap flush and update.
  596. * That enforces a right ordering and update sequence
  597. * synchronization against the reader (task_gtime())
  598. * that can thus safely catch up with a tickless delta.
  599. */
  600. write_seqlock(&tsk->vtime_seqlock);
  601. __vtime_account_system(tsk);
  602. current->flags |= PF_VCPU;
  603. write_sequnlock(&tsk->vtime_seqlock);
  604. }
  605. EXPORT_SYMBOL_GPL(vtime_guest_enter);
  606. void vtime_guest_exit(struct task_struct *tsk)
  607. {
  608. write_seqlock(&tsk->vtime_seqlock);
  609. __vtime_account_system(tsk);
  610. current->flags &= ~PF_VCPU;
  611. write_sequnlock(&tsk->vtime_seqlock);
  612. }
  613. EXPORT_SYMBOL_GPL(vtime_guest_exit);
  614. void vtime_account_idle(struct task_struct *tsk)
  615. {
  616. cputime_t delta_cpu = get_vtime_delta(tsk);
  617. account_idle_time(delta_cpu);
  618. }
  619. void arch_vtime_task_switch(struct task_struct *prev)
  620. {
  621. write_seqlock(&prev->vtime_seqlock);
  622. prev->vtime_snap_whence = VTIME_SLEEPING;
  623. write_sequnlock(&prev->vtime_seqlock);
  624. write_seqlock(&current->vtime_seqlock);
  625. current->vtime_snap_whence = VTIME_SYS;
  626. current->vtime_snap = sched_clock_cpu(smp_processor_id());
  627. write_sequnlock(&current->vtime_seqlock);
  628. }
  629. void vtime_init_idle(struct task_struct *t, int cpu)
  630. {
  631. unsigned long flags;
  632. write_seqlock_irqsave(&t->vtime_seqlock, flags);
  633. t->vtime_snap_whence = VTIME_SYS;
  634. t->vtime_snap = sched_clock_cpu(cpu);
  635. write_sequnlock_irqrestore(&t->vtime_seqlock, flags);
  636. }
  637. cputime_t task_gtime(struct task_struct *t)
  638. {
  639. unsigned int seq;
  640. cputime_t gtime;
  641. do {
  642. seq = read_seqbegin(&t->vtime_seqlock);
  643. gtime = t->gtime;
  644. if (t->flags & PF_VCPU)
  645. gtime += vtime_delta(t);
  646. } while (read_seqretry(&t->vtime_seqlock, seq));
  647. return gtime;
  648. }
  649. /*
  650. * Fetch cputime raw values from fields of task_struct and
  651. * add up the pending nohz execution time since the last
  652. * cputime snapshot.
  653. */
  654. static void
  655. fetch_task_cputime(struct task_struct *t,
  656. cputime_t *u_dst, cputime_t *s_dst,
  657. cputime_t *u_src, cputime_t *s_src,
  658. cputime_t *udelta, cputime_t *sdelta)
  659. {
  660. unsigned int seq;
  661. unsigned long long delta;
  662. do {
  663. *udelta = 0;
  664. *sdelta = 0;
  665. seq = read_seqbegin(&t->vtime_seqlock);
  666. if (u_dst)
  667. *u_dst = *u_src;
  668. if (s_dst)
  669. *s_dst = *s_src;
  670. /* Task is sleeping, nothing to add */
  671. if (t->vtime_snap_whence == VTIME_SLEEPING ||
  672. is_idle_task(t))
  673. continue;
  674. delta = vtime_delta(t);
  675. /*
  676. * Task runs either in user or kernel space, add pending nohz time to
  677. * the right place.
  678. */
  679. if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) {
  680. *udelta = delta;
  681. } else {
  682. if (t->vtime_snap_whence == VTIME_SYS)
  683. *sdelta = delta;
  684. }
  685. } while (read_seqretry(&t->vtime_seqlock, seq));
  686. }
  687. void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime)
  688. {
  689. cputime_t udelta, sdelta;
  690. fetch_task_cputime(t, utime, stime, &t->utime,
  691. &t->stime, &udelta, &sdelta);
  692. if (utime)
  693. *utime += udelta;
  694. if (stime)
  695. *stime += sdelta;
  696. }
  697. void task_cputime_scaled(struct task_struct *t,
  698. cputime_t *utimescaled, cputime_t *stimescaled)
  699. {
  700. cputime_t udelta, sdelta;
  701. fetch_task_cputime(t, utimescaled, stimescaled,
  702. &t->utimescaled, &t->stimescaled, &udelta, &sdelta);
  703. if (utimescaled)
  704. *utimescaled += cputime_to_scaled(udelta);
  705. if (stimescaled)
  706. *stimescaled += cputime_to_scaled(sdelta);
  707. }
  708. #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */