1#include <linux/cpufreq.h> 2#include <linux/export.h> 3#include <linux/sched.h> 4#include <linux/tsacct_kern.h> 5#include <linux/kernel_stat.h> 6#include <linux/static_key.h> 7#include <linux/context_tracking.h> 8#include "sched.h" 9 10 11#ifdef CONFIG_IRQ_TIME_ACCOUNTING 12 13/* 14 * There are no locks covering percpu hardirq/softirq time. 15 * They are only modified in vtime_account, on corresponding CPU 16 * with interrupts disabled. So, writes are safe. 17 * They are read and saved off onto struct rq in update_rq_clock(). 18 * This may result in other CPU reading this CPU's irq time and can 19 * race with irq/vtime_account on this CPU. We would either get old 20 * or new value with a side effect of accounting a slice of irq time to wrong 21 * task when irq is in progress while we read rq->clock. That is a worthy 22 * compromise in place of having locks on each irq in account_system_time. 23 */ 24DEFINE_PER_CPU(u64, cpu_hardirq_time); 25DEFINE_PER_CPU(u64, cpu_softirq_time); 26 27static DEFINE_PER_CPU(u64, irq_start_time); 28static int sched_clock_irqtime; 29 30void enable_sched_clock_irqtime(void) 31{ 32 sched_clock_irqtime = 1; 33} 34 35void disable_sched_clock_irqtime(void) 36{ 37 sched_clock_irqtime = 0; 38} 39 40#ifndef CONFIG_64BIT 41DEFINE_PER_CPU(seqcount_t, irq_time_seq); 42#endif /* CONFIG_64BIT */ 43 44/* 45 * Called before incrementing preempt_count on {soft,}irq_enter 46 * and before decrementing preempt_count on {soft,}irq_exit. 47 */ 48void irqtime_account_irq(struct task_struct *curr) 49{ 50 unsigned long flags; 51 s64 delta; 52 int cpu; 53 54 if (!sched_clock_irqtime) 55 return; 56 57 local_irq_save(flags); 58 59 cpu = smp_processor_id(); 60 delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); 61 __this_cpu_add(irq_start_time, delta); 62 63 irq_time_write_begin(); 64 /* 65 * We do not account for softirq time from ksoftirqd here. 66 * We want to continue accounting softirq time to ksoftirqd thread 67 * in that case, so as not to confuse scheduler with a special task 68 * that do not consume any time, but still wants to run. 69 */ 70 if (hardirq_count()) 71 __this_cpu_add(cpu_hardirq_time, delta); 72 else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) 73 __this_cpu_add(cpu_softirq_time, delta); 74 75 irq_time_write_end(); 76 local_irq_restore(flags); 77} 78EXPORT_SYMBOL_GPL(irqtime_account_irq); 79 80static int irqtime_account_hi_update(void) 81{ 82 u64 *cpustat = kcpustat_this_cpu->cpustat; 83 unsigned long flags; 84 u64 latest_ns; 85 int ret = 0; 86 87 local_irq_save(flags); 88 latest_ns = this_cpu_read(cpu_hardirq_time); 89 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ]) 90 ret = 1; 91 local_irq_restore(flags); 92 return ret; 93} 94 95static int irqtime_account_si_update(void) 96{ 97 u64 *cpustat = kcpustat_this_cpu->cpustat; 98 unsigned long flags; 99 u64 latest_ns; 100 int ret = 0; 101 102 local_irq_save(flags); 103 latest_ns = this_cpu_read(cpu_softirq_time); 104 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ]) 105 ret = 1; 106 local_irq_restore(flags); 107 return ret; 108} 109 110#else /* CONFIG_IRQ_TIME_ACCOUNTING */ 111 112#define sched_clock_irqtime (0) 113 114#endif /* !CONFIG_IRQ_TIME_ACCOUNTING */ 115 116static inline void task_group_account_field(struct task_struct *p, int index, 117 u64 tmp) 118{ 119 /* 120 * Since all updates are sure to touch the root cgroup, we 121 * get ourselves ahead and touch it first. If the root cgroup 122 * is the only cgroup, then nothing else should be necessary. 123 * 124 */ 125 __this_cpu_add(kernel_cpustat.cpustat[index], tmp); 126 127 cpuacct_account_field(p, index, tmp); 128} 129 130/* 131 * Account user cpu time to a process. 132 * @p: the process that the cpu time gets accounted to 133 * @cputime: the cpu time spent in user space since the last update 134 * @cputime_scaled: cputime scaled by cpu frequency 135 */ 136void account_user_time(struct task_struct *p, cputime_t cputime, 137 cputime_t cputime_scaled) 138{ 139 int index; 140 141 /* Add user time to process. */ 142 p->utime += cputime; 143 p->utimescaled += cputime_scaled; 144 account_group_user_time(p, cputime); 145 146 index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; 147 148 /* Add user time to cpustat. */ 149 task_group_account_field(p, index, (__force u64) cputime); 150 151 /* Account for user time used */ 152 acct_account_cputime(p); 153 154#ifdef CONFIG_CPU_FREQ_STAT 155 /* Account power usage for user time */ 156 acct_update_power(p, cputime); 157#endif 158} 159 160/* 161 * Account guest cpu time to a process. 162 * @p: the process that the cpu time gets accounted to 163 * @cputime: the cpu time spent in virtual machine since the last update 164 * @cputime_scaled: cputime scaled by cpu frequency 165 */ 166static void account_guest_time(struct task_struct *p, cputime_t cputime, 167 cputime_t cputime_scaled) 168{ 169 u64 *cpustat = kcpustat_this_cpu->cpustat; 170 171 /* Add guest time to process. */ 172 p->utime += cputime; 173 p->utimescaled += cputime_scaled; 174 account_group_user_time(p, cputime); 175 p->gtime += cputime; 176 177 /* Add guest time to cpustat. */ 178 if (task_nice(p) > 0) { 179 cpustat[CPUTIME_NICE] += (__force u64) cputime; 180 cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; 181 } else { 182 cpustat[CPUTIME_USER] += (__force u64) cputime; 183 cpustat[CPUTIME_GUEST] += (__force u64) cputime; 184 } 185} 186 187/* 188 * Account system cpu time to a process and desired cpustat field 189 * @p: the process that the cpu time gets accounted to 190 * @cputime: the cpu time spent in kernel space since the last update 191 * @cputime_scaled: cputime scaled by cpu frequency 192 * @target_cputime64: pointer to cpustat field that has to be updated 193 */ 194static inline 195void __account_system_time(struct task_struct *p, cputime_t cputime, 196 cputime_t cputime_scaled, int index) 197{ 198 /* Add system time to process. */ 199 p->stime += cputime; 200 p->stimescaled += cputime_scaled; 201 account_group_system_time(p, cputime); 202 203 /* Add system time to cpustat. */ 204 task_group_account_field(p, index, (__force u64) cputime); 205 206 /* Account for system time used */ 207 acct_account_cputime(p); 208 209#ifdef CONFIG_CPU_FREQ_STAT 210 /* Account power usage for system time */ 211 acct_update_power(p, cputime); 212#endif 213} 214 215/* 216 * Account system cpu time to a process. 217 * @p: the process that the cpu time gets accounted to 218 * @hardirq_offset: the offset to subtract from hardirq_count() 219 * @cputime: the cpu time spent in kernel space since the last update 220 * @cputime_scaled: cputime scaled by cpu frequency 221 */ 222void account_system_time(struct task_struct *p, int hardirq_offset, 223 cputime_t cputime, cputime_t cputime_scaled) 224{ 225 int index; 226 227 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { 228 account_guest_time(p, cputime, cputime_scaled); 229 return; 230 } 231 232 if (hardirq_count() - hardirq_offset) 233 index = CPUTIME_IRQ; 234 else if (in_serving_softirq()) 235 index = CPUTIME_SOFTIRQ; 236 else 237 index = CPUTIME_SYSTEM; 238 239 __account_system_time(p, cputime, cputime_scaled, index); 240} 241 242/* 243 * Account for involuntary wait time. 244 * @cputime: the cpu time spent in involuntary wait 245 */ 246void account_steal_time(cputime_t cputime) 247{ 248 u64 *cpustat = kcpustat_this_cpu->cpustat; 249 250 cpustat[CPUTIME_STEAL] += (__force u64) cputime; 251} 252 253/* 254 * Account for idle time. 255 * @cputime: the cpu time spent in idle wait 256 */ 257void account_idle_time(cputime_t cputime) 258{ 259 u64 *cpustat = kcpustat_this_cpu->cpustat; 260 struct rq *rq = this_rq(); 261 262 if (atomic_read(&rq->nr_iowait) > 0) 263 cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; 264 else 265 cpustat[CPUTIME_IDLE] += (__force u64) cputime; 266} 267 268static __always_inline bool steal_account_process_tick(void) 269{ 270#ifdef CONFIG_PARAVIRT 271 if (static_key_false(¶virt_steal_enabled)) { 272 u64 steal; 273 cputime_t steal_ct; 274 275 steal = paravirt_steal_clock(smp_processor_id()); 276 steal -= this_rq()->prev_steal_time; 277 278 /* 279 * cputime_t may be less precise than nsecs (eg: if it's 280 * based on jiffies). Lets cast the result to cputime 281 * granularity and account the rest on the next rounds. 282 */ 283 steal_ct = nsecs_to_cputime(steal); 284 this_rq()->prev_steal_time += cputime_to_nsecs(steal_ct); 285 286 account_steal_time(steal_ct); 287 return steal_ct; 288 } 289#endif 290 return false; 291} 292 293/* 294 * Accumulate raw cputime values of dead tasks (sig->[us]time) and live 295 * tasks (sum on group iteration) belonging to @tsk's group. 296 */ 297void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) 298{ 299 struct signal_struct *sig = tsk->signal; 300 cputime_t utime, stime; 301 struct task_struct *t; 302 unsigned int seq, nextseq; 303 unsigned long flags; 304 305 rcu_read_lock(); 306 /* Attempt a lockless read on the first round. */ 307 nextseq = 0; 308 do { 309 seq = nextseq; 310 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq); 311 times->utime = sig->utime; 312 times->stime = sig->stime; 313 times->sum_exec_runtime = sig->sum_sched_runtime; 314 315 for_each_thread(tsk, t) { 316 task_cputime(t, &utime, &stime); 317 times->utime += utime; 318 times->stime += stime; 319 times->sum_exec_runtime += task_sched_runtime(t); 320 } 321 /* If lockless access failed, take the lock. */ 322 nextseq = 1; 323 } while (need_seqretry(&sig->stats_lock, seq)); 324 done_seqretry_irqrestore(&sig->stats_lock, seq, flags); 325 rcu_read_unlock(); 326} 327 328#ifdef CONFIG_IRQ_TIME_ACCOUNTING 329/* 330 * Account a tick to a process and cpustat 331 * @p: the process that the cpu time gets accounted to 332 * @user_tick: is the tick from userspace 333 * @rq: the pointer to rq 334 * 335 * Tick demultiplexing follows the order 336 * - pending hardirq update 337 * - pending softirq update 338 * - user_time 339 * - idle_time 340 * - system time 341 * - check for guest_time 342 * - else account as system_time 343 * 344 * Check for hardirq is done both for system and user time as there is 345 * no timer going off while we are on hardirq and hence we may never get an 346 * opportunity to update it solely in system time. 347 * p->stime and friends are only updated on system time and not on irq 348 * softirq as those do not count in task exec_runtime any more. 349 */ 350static void irqtime_account_process_tick(struct task_struct *p, int user_tick, 351 struct rq *rq, int ticks) 352{ 353 cputime_t scaled = cputime_to_scaled(cputime_one_jiffy); 354 u64 cputime = (__force u64) cputime_one_jiffy; 355 u64 *cpustat = kcpustat_this_cpu->cpustat; 356 357 if (steal_account_process_tick()) 358 return; 359 360 cputime *= ticks; 361 scaled *= ticks; 362 363 if (irqtime_account_hi_update()) { 364 cpustat[CPUTIME_IRQ] += cputime; 365 } else if (irqtime_account_si_update()) { 366 cpustat[CPUTIME_SOFTIRQ] += cputime; 367 } else if (this_cpu_ksoftirqd() == p) { 368 /* 369 * ksoftirqd time do not get accounted in cpu_softirq_time. 370 * So, we have to handle it separately here. 371 * Also, p->stime needs to be updated for ksoftirqd. 372 */ 373 __account_system_time(p, cputime, scaled, CPUTIME_SOFTIRQ); 374 } else if (user_tick) { 375 account_user_time(p, cputime, scaled); 376 } else if (p == rq->idle) { 377 account_idle_time(cputime); 378 } else if (p->flags & PF_VCPU) { /* System time or guest time */ 379 account_guest_time(p, cputime, scaled); 380 } else { 381 __account_system_time(p, cputime, scaled, CPUTIME_SYSTEM); 382 } 383} 384 385static void irqtime_account_idle_ticks(int ticks) 386{ 387 struct rq *rq = this_rq(); 388 389 irqtime_account_process_tick(current, 0, rq, ticks); 390} 391#else /* CONFIG_IRQ_TIME_ACCOUNTING */ 392static inline void irqtime_account_idle_ticks(int ticks) {} 393static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, 394 struct rq *rq, int nr_ticks) {} 395#endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 396 397/* 398 * Use precise platform statistics if available: 399 */ 400#ifdef CONFIG_VIRT_CPU_ACCOUNTING 401 402#ifndef __ARCH_HAS_VTIME_TASK_SWITCH 403void vtime_common_task_switch(struct task_struct *prev) 404{ 405 if (is_idle_task(prev)) 406 vtime_account_idle(prev); 407 else 408 vtime_account_system(prev); 409 410#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE 411 vtime_account_user(prev); 412#endif 413 arch_vtime_task_switch(prev); 414} 415#endif 416 417/* 418 * Archs that account the whole time spent in the idle task 419 * (outside irq) as idle time can rely on this and just implement 420 * vtime_account_system() and vtime_account_idle(). Archs that 421 * have other meaning of the idle time (s390 only includes the 422 * time spent by the CPU when it's in low power mode) must override 423 * vtime_account(). 424 */ 425#ifndef __ARCH_HAS_VTIME_ACCOUNT 426void vtime_common_account_irq_enter(struct task_struct *tsk) 427{ 428 if (!in_interrupt()) { 429 /* 430 * If we interrupted user, context_tracking_in_user() 431 * is 1 because the context tracking don't hook 432 * on irq entry/exit. This way we know if 433 * we need to flush user time on kernel entry. 434 */ 435 if (context_tracking_in_user()) { 436 vtime_account_user(tsk); 437 return; 438 } 439 440 if (is_idle_task(tsk)) { 441 vtime_account_idle(tsk); 442 return; 443 } 444 } 445 vtime_account_system(tsk); 446} 447EXPORT_SYMBOL_GPL(vtime_common_account_irq_enter); 448#endif /* __ARCH_HAS_VTIME_ACCOUNT */ 449#endif /* CONFIG_VIRT_CPU_ACCOUNTING */ 450 451 452#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE 453void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) 454{ 455 *ut = p->utime; 456 *st = p->stime; 457} 458 459void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) 460{ 461 struct task_cputime cputime; 462 463 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 */ 474void account_process_tick(struct task_struct *p, int user_tick) 475{ 476 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); 477 struct rq *rq = this_rq(); 478 479 if (vtime_accounting_enabled()) 480 return; 481 482 if (sched_clock_irqtime) { 483 irqtime_account_process_tick(p, user_tick, rq, 1); 484 return; 485 } 486 487 if (steal_account_process_tick()) 488 return; 489 490 if (user_tick) 491 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); 492 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) 493 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, 494 one_jiffy_scaled); 495 else 496 account_idle_time(cputime_one_jiffy); 497} 498 499/* 500 * Account multiple ticks of steal time. 501 * @p: the process from which the cpu time has been stolen 502 * @ticks: number of stolen ticks 503 */ 504void account_steal_ticks(unsigned long ticks) 505{ 506 account_steal_time(jiffies_to_cputime(ticks)); 507} 508 509/* 510 * Account multiple ticks of idle time. 511 * @ticks: number of stolen ticks 512 */ 513void account_idle_ticks(unsigned long ticks) 514{ 515 516 if (sched_clock_irqtime) { 517 irqtime_account_idle_ticks(ticks); 518 return; 519 } 520 521 account_idle_time(jiffies_to_cputime(ticks)); 522} 523 524/* 525 * Perform (stime * rtime) / total, but avoid multiplication overflow by 526 * loosing precision when the numbers are big. 527 */ 528static cputime_t scale_stime(u64 stime, u64 rtime, u64 total) 529{ 530 u64 scaled; 531 532 for (;;) { 533 /* Make sure "rtime" is the bigger of stime/rtime */ 534 if (stime > rtime) 535 swap(rtime, stime); 536 537 /* Make sure 'total' fits in 32 bits */ 538 if (total >> 32) 539 goto drop_precision; 540 541 /* Does rtime (and thus stime) fit in 32 bits? */ 542 if (!(rtime >> 32)) 543 break; 544 545 /* Can we just balance rtime/stime rather than dropping bits? */ 546 if (stime >> 31) 547 goto drop_precision; 548 549 /* We can grow stime and shrink rtime and try to make them both fit */ 550 stime <<= 1; 551 rtime >>= 1; 552 continue; 553 554drop_precision: 555 /* We drop from rtime, it has more bits than stime */ 556 rtime >>= 1; 557 total >>= 1; 558 } 559 560 /* 561 * Make sure gcc understands that this is a 32x32->64 multiply, 562 * followed by a 64/32->64 divide. 563 */ 564 scaled = div_u64((u64) (u32) stime * (u64) (u32) rtime, (u32)total); 565 return (__force cputime_t) scaled; 566} 567 568/* 569 * Atomically advance counter to the new value. Interrupts, vcpu 570 * scheduling, and scaling inaccuracies can cause cputime_advance 571 * to be occasionally called with a new value smaller than counter. 572 * Let's enforce atomicity. 573 * 574 * Normally a caller will only go through this loop once, or not 575 * at all in case a previous caller updated counter the same jiffy. 576 */ 577static void cputime_advance(cputime_t *counter, cputime_t new) 578{ 579 cputime_t old; 580 581 while (new > (old = ACCESS_ONCE(*counter))) 582 cmpxchg_cputime(counter, old, new); 583} 584 585/* 586 * Adjust tick based cputime random precision against scheduler 587 * runtime accounting. 588 */ 589static void cputime_adjust(struct task_cputime *curr, 590 struct cputime *prev, 591 cputime_t *ut, cputime_t *st) 592{ 593 cputime_t rtime, stime, utime; 594 595 /* 596 * Tick based cputime accounting depend on random scheduling 597 * timeslices of a task to be interrupted or not by the timer. 598 * Depending on these circumstances, the number of these interrupts 599 * may be over or under-optimistic, matching the real user and system 600 * cputime with a variable precision. 601 * 602 * Fix this by scaling these tick based values against the total 603 * runtime accounted by the CFS scheduler. 604 */ 605 rtime = nsecs_to_cputime(curr->sum_exec_runtime); 606 607 /* 608 * Update userspace visible utime/stime values only if actual execution 609 * time is bigger than already exported. Note that can happen, that we 610 * provided bigger values due to scaling inaccuracy on big numbers. 611 */ 612 if (prev->stime + prev->utime >= rtime) 613 goto out; 614 615 stime = curr->stime; 616 utime = curr->utime; 617 618 if (utime == 0) { 619 stime = rtime; 620 } else if (stime == 0) { 621 utime = rtime; 622 } else { 623 cputime_t total = stime + utime; 624 625 stime = scale_stime((__force u64)stime, 626 (__force u64)rtime, (__force u64)total); 627 utime = rtime - stime; 628 } 629 630 cputime_advance(&prev->stime, stime); 631 cputime_advance(&prev->utime, utime); 632 633out: 634 *ut = prev->utime; 635 *st = prev->stime; 636} 637 638void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) 639{ 640 struct task_cputime cputime = { 641 .sum_exec_runtime = p->se.sum_exec_runtime, 642 }; 643 644 task_cputime(p, &cputime.utime, &cputime.stime); 645 cputime_adjust(&cputime, &p->prev_cputime, ut, st); 646} 647 648void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) 649{ 650 struct task_cputime cputime; 651 652 thread_group_cputime(p, &cputime); 653 cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); 654} 655#endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ 656 657#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN 658static unsigned long long vtime_delta(struct task_struct *tsk) 659{ 660 unsigned long long clock; 661 662 clock = local_clock(); 663 if (clock < tsk->vtime_snap) 664 return 0; 665 666 return clock - tsk->vtime_snap; 667} 668 669static cputime_t get_vtime_delta(struct task_struct *tsk) 670{ 671 unsigned long long delta = vtime_delta(tsk); 672 673 WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_SLEEPING); 674 tsk->vtime_snap += delta; 675 676 /* CHECKME: always safe to convert nsecs to cputime? */ 677 return nsecs_to_cputime(delta); 678} 679 680static void __vtime_account_system(struct task_struct *tsk) 681{ 682 cputime_t delta_cpu = get_vtime_delta(tsk); 683 684 account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu)); 685} 686 687void vtime_account_system(struct task_struct *tsk) 688{ 689 write_seqlock(&tsk->vtime_seqlock); 690 __vtime_account_system(tsk); 691 write_sequnlock(&tsk->vtime_seqlock); 692} 693 694void vtime_gen_account_irq_exit(struct task_struct *tsk) 695{ 696 write_seqlock(&tsk->vtime_seqlock); 697 __vtime_account_system(tsk); 698 if (context_tracking_in_user()) 699 tsk->vtime_snap_whence = VTIME_USER; 700 write_sequnlock(&tsk->vtime_seqlock); 701} 702 703void vtime_account_user(struct task_struct *tsk) 704{ 705 cputime_t delta_cpu; 706 707 write_seqlock(&tsk->vtime_seqlock); 708 delta_cpu = get_vtime_delta(tsk); 709 tsk->vtime_snap_whence = VTIME_SYS; 710 account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu)); 711 write_sequnlock(&tsk->vtime_seqlock); 712} 713 714void vtime_user_enter(struct task_struct *tsk) 715{ 716 write_seqlock(&tsk->vtime_seqlock); 717 __vtime_account_system(tsk); 718 tsk->vtime_snap_whence = VTIME_USER; 719 write_sequnlock(&tsk->vtime_seqlock); 720} 721 722void vtime_guest_enter(struct task_struct *tsk) 723{ 724 /* 725 * The flags must be updated under the lock with 726 * the vtime_snap flush and update. 727 * That enforces a right ordering and update sequence 728 * synchronization against the reader (task_gtime()) 729 * that can thus safely catch up with a tickless delta. 730 */ 731 write_seqlock(&tsk->vtime_seqlock); 732 __vtime_account_system(tsk); 733 current->flags |= PF_VCPU; 734 write_sequnlock(&tsk->vtime_seqlock); 735} 736EXPORT_SYMBOL_GPL(vtime_guest_enter); 737 738void vtime_guest_exit(struct task_struct *tsk) 739{ 740 write_seqlock(&tsk->vtime_seqlock); 741 __vtime_account_system(tsk); 742 current->flags &= ~PF_VCPU; 743 write_sequnlock(&tsk->vtime_seqlock); 744} 745EXPORT_SYMBOL_GPL(vtime_guest_exit); 746 747void vtime_account_idle(struct task_struct *tsk) 748{ 749 cputime_t delta_cpu = get_vtime_delta(tsk); 750 751 account_idle_time(delta_cpu); 752} 753 754void arch_vtime_task_switch(struct task_struct *prev) 755{ 756 write_seqlock(&prev->vtime_seqlock); 757 prev->vtime_snap_whence = VTIME_SLEEPING; 758 write_sequnlock(&prev->vtime_seqlock); 759 760 write_seqlock(¤t->vtime_seqlock); 761 current->vtime_snap_whence = VTIME_SYS; 762 current->vtime_snap = sched_clock_cpu(smp_processor_id()); 763 write_sequnlock(¤t->vtime_seqlock); 764} 765 766void vtime_init_idle(struct task_struct *t, int cpu) 767{ 768 unsigned long flags; 769 770 write_seqlock_irqsave(&t->vtime_seqlock, flags); 771 t->vtime_snap_whence = VTIME_SYS; 772 t->vtime_snap = sched_clock_cpu(cpu); 773 write_sequnlock_irqrestore(&t->vtime_seqlock, flags); 774} 775 776cputime_t task_gtime(struct task_struct *t) 777{ 778 unsigned int seq; 779 cputime_t gtime; 780 781 do { 782 seq = read_seqbegin(&t->vtime_seqlock); 783 784 gtime = t->gtime; 785 if (t->flags & PF_VCPU) 786 gtime += vtime_delta(t); 787 788 } while (read_seqretry(&t->vtime_seqlock, seq)); 789 790 return gtime; 791} 792 793/* 794 * Fetch cputime raw values from fields of task_struct and 795 * add up the pending nohz execution time since the last 796 * cputime snapshot. 797 */ 798static void 799fetch_task_cputime(struct task_struct *t, 800 cputime_t *u_dst, cputime_t *s_dst, 801 cputime_t *u_src, cputime_t *s_src, 802 cputime_t *udelta, cputime_t *sdelta) 803{ 804 unsigned int seq; 805 unsigned long long delta; 806 807 do { 808 *udelta = 0; 809 *sdelta = 0; 810 811 seq = read_seqbegin(&t->vtime_seqlock); 812 813 if (u_dst) 814 *u_dst = *u_src; 815 if (s_dst) 816 *s_dst = *s_src; 817 818 /* Task is sleeping, nothing to add */ 819 if (t->vtime_snap_whence == VTIME_SLEEPING || 820 is_idle_task(t)) 821 continue; 822 823 delta = vtime_delta(t); 824 825 /* 826 * Task runs either in user or kernel space, add pending nohz time to 827 * the right place. 828 */ 829 if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) { 830 *udelta = delta; 831 } else { 832 if (t->vtime_snap_whence == VTIME_SYS) 833 *sdelta = delta; 834 } 835 } while (read_seqretry(&t->vtime_seqlock, seq)); 836} 837 838 839void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime) 840{ 841 cputime_t udelta, sdelta; 842 843 fetch_task_cputime(t, utime, stime, &t->utime, 844 &t->stime, &udelta, &sdelta); 845 if (utime) 846 *utime += udelta; 847 if (stime) 848 *stime += sdelta; 849} 850 851void task_cputime_scaled(struct task_struct *t, 852 cputime_t *utimescaled, cputime_t *stimescaled) 853{ 854 cputime_t udelta, sdelta; 855 856 fetch_task_cputime(t, utimescaled, stimescaled, 857 &t->utimescaled, &t->stimescaled, &udelta, &sdelta); 858 if (utimescaled) 859 *utimescaled += cputime_to_scaled(udelta); 860 if (stimescaled) 861 *stimescaled += cputime_to_scaled(sdelta); 862} 863#endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ 864