]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - kernel/time/tick-sched.c
timers: Framework for identifying pinned timers
[mirror_ubuntu-artful-kernel.git] / kernel / time / tick-sched.c
CommitLineData
79bf2bb3
TG
1/*
2 * linux/kernel/time/tick-sched.c
3 *
4 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner
7 *
8 * No idle tick implementation for low and high resolution timers
9 *
10 * Started by: Thomas Gleixner and Ingo Molnar
11 *
b10db7f0 12 * Distribute under GPLv2.
79bf2bb3
TG
13 */
14#include <linux/cpu.h>
15#include <linux/err.h>
16#include <linux/hrtimer.h>
17#include <linux/interrupt.h>
18#include <linux/kernel_stat.h>
19#include <linux/percpu.h>
20#include <linux/profile.h>
21#include <linux/sched.h>
22#include <linux/tick.h>
8083e4ad 23#include <linux/module.h>
79bf2bb3 24
9e203bcc
DM
25#include <asm/irq_regs.h>
26
79bf2bb3
TG
27#include "tick-internal.h"
28
29/*
30 * Per cpu nohz control structure
31 */
32static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched);
33
34/*
35 * The time, when the last jiffy update happened. Protected by xtime_lock.
36 */
37static ktime_t last_jiffies_update;
38
289f480a
IM
39struct tick_sched *tick_get_tick_sched(int cpu)
40{
41 return &per_cpu(tick_cpu_sched, cpu);
42}
43
79bf2bb3
TG
44/*
45 * Must be called with interrupts disabled !
46 */
47static void tick_do_update_jiffies64(ktime_t now)
48{
49 unsigned long ticks = 0;
50 ktime_t delta;
51
7a14ce1d
IM
52 /*
53 * Do a quick check without holding xtime_lock:
54 */
55 delta = ktime_sub(now, last_jiffies_update);
56 if (delta.tv64 < tick_period.tv64)
57 return;
58
79bf2bb3
TG
59 /* Reevalute with xtime_lock held */
60 write_seqlock(&xtime_lock);
61
62 delta = ktime_sub(now, last_jiffies_update);
63 if (delta.tv64 >= tick_period.tv64) {
64
65 delta = ktime_sub(delta, tick_period);
66 last_jiffies_update = ktime_add(last_jiffies_update,
67 tick_period);
68
69 /* Slow path for long timeouts */
70 if (unlikely(delta.tv64 >= tick_period.tv64)) {
71 s64 incr = ktime_to_ns(tick_period);
72
73 ticks = ktime_divns(delta, incr);
74
75 last_jiffies_update = ktime_add_ns(last_jiffies_update,
76 incr * ticks);
77 }
78 do_timer(++ticks);
49d670fb
TG
79
80 /* Keep the tick_next_period variable up to date */
81 tick_next_period = ktime_add(last_jiffies_update, tick_period);
79bf2bb3
TG
82 }
83 write_sequnlock(&xtime_lock);
84}
85
86/*
87 * Initialize and return retrieve the jiffies update.
88 */
89static ktime_t tick_init_jiffy_update(void)
90{
91 ktime_t period;
92
93 write_seqlock(&xtime_lock);
94 /* Did we start the jiffies update yet ? */
95 if (last_jiffies_update.tv64 == 0)
96 last_jiffies_update = tick_next_period;
97 period = last_jiffies_update;
98 write_sequnlock(&xtime_lock);
99 return period;
100}
101
102/*
103 * NOHZ - aka dynamic tick functionality
104 */
105#ifdef CONFIG_NO_HZ
106/*
107 * NO HZ enabled ?
108 */
109static int tick_nohz_enabled __read_mostly = 1;
110
111/*
112 * Enable / Disable tickless mode
113 */
114static int __init setup_tick_nohz(char *str)
115{
116 if (!strcmp(str, "off"))
117 tick_nohz_enabled = 0;
118 else if (!strcmp(str, "on"))
119 tick_nohz_enabled = 1;
120 else
121 return 0;
122 return 1;
123}
124
125__setup("nohz=", setup_tick_nohz);
126
127/**
128 * tick_nohz_update_jiffies - update jiffies when idle was interrupted
129 *
130 * Called from interrupt entry when the CPU was idle
131 *
132 * In case the sched_tick was stopped on this CPU, we have to check if jiffies
133 * must be updated. Otherwise an interrupt handler could use a stale jiffy
134 * value. We do this unconditionally on any cpu, as we don't know whether the
135 * cpu, which has the update task assigned is in a long sleep.
136 */
934d96ea 137static void tick_nohz_update_jiffies(void)
79bf2bb3
TG
138{
139 int cpu = smp_processor_id();
140 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
141 unsigned long flags;
142 ktime_t now;
143
144 if (!ts->tick_stopped)
145 return;
146
6a7b3dc3 147 cpumask_clear_cpu(cpu, nohz_cpu_mask);
79bf2bb3 148 now = ktime_get();
5df7fa1c 149 ts->idle_waketime = now;
79bf2bb3
TG
150
151 local_irq_save(flags);
152 tick_do_update_jiffies64(now);
153 local_irq_restore(flags);
02ff3755
IM
154
155 touch_softlockup_watchdog();
79bf2bb3
TG
156}
157
719254fa 158static void tick_nohz_stop_idle(int cpu)
6378ddb5
VP
159{
160 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
161
162 if (ts->idle_active) {
163 ktime_t now, delta;
164 now = ktime_get();
165 delta = ktime_sub(now, ts->idle_entrytime);
166 ts->idle_lastupdate = now;
167 ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
168 ts->idle_active = 0;
56c7426b
PZ
169
170 sched_clock_idle_wakeup_event(0);
6378ddb5
VP
171 }
172}
173
903b8a8d 174static ktime_t tick_nohz_start_idle(struct tick_sched *ts)
6378ddb5 175{
6378ddb5
VP
176 ktime_t now, delta;
177
178 now = ktime_get();
179 if (ts->idle_active) {
180 delta = ktime_sub(now, ts->idle_entrytime);
181 ts->idle_lastupdate = now;
182 ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
183 }
184 ts->idle_entrytime = now;
185 ts->idle_active = 1;
56c7426b 186 sched_clock_idle_sleep_event();
6378ddb5
VP
187 return now;
188}
189
190u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
191{
192 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
193
8083e4ad 194 if (!tick_nohz_enabled)
195 return -1;
196
197 if (ts->idle_active)
198 *last_update_time = ktime_to_us(ts->idle_lastupdate);
199 else
200 *last_update_time = ktime_to_us(ktime_get());
201
6378ddb5
VP
202 return ktime_to_us(ts->idle_sleeptime);
203}
8083e4ad 204EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
6378ddb5 205
79bf2bb3
TG
206/**
207 * tick_nohz_stop_sched_tick - stop the idle tick from the idle task
208 *
209 * When the next event is more than a tick into the future, stop the idle tick
210 * Called either from the idle loop or from irq_exit() when an idle period was
211 * just interrupted by an interrupt which did not cause a reschedule.
212 */
b8f8c3cf 213void tick_nohz_stop_sched_tick(int inidle)
79bf2bb3
TG
214{
215 unsigned long seq, last_jiffies, next_jiffies, delta_jiffies, flags;
216 struct tick_sched *ts;
6378ddb5 217 ktime_t last_update, expires, now;
4f86d3a8 218 struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev;
79bf2bb3
TG
219 int cpu;
220
221 local_irq_save(flags);
222
223 cpu = smp_processor_id();
224 ts = &per_cpu(tick_cpu_sched, cpu);
903b8a8d 225 now = tick_nohz_start_idle(ts);
79bf2bb3 226
5e41d0d6
TG
227 /*
228 * If this cpu is offline and it is the one which updates
229 * jiffies, then give up the assignment and let it be taken by
230 * the cpu which runs the tick timer next. If we don't drop
231 * this here the jiffies might be stale and do_timer() never
232 * invoked.
233 */
234 if (unlikely(!cpu_online(cpu))) {
235 if (cpu == tick_do_timer_cpu)
6441402b 236 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
5e41d0d6
TG
237 }
238
79bf2bb3
TG
239 if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE))
240 goto end;
241
b8f8c3cf
TG
242 if (!inidle && !ts->inidle)
243 goto end;
244
245 ts->inidle = 1;
246
79bf2bb3
TG
247 if (need_resched())
248 goto end;
249
fa116ea3 250 if (unlikely(local_softirq_pending() && cpu_online(cpu))) {
35282316
TG
251 static int ratelimit;
252
253 if (ratelimit < 10) {
254 printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n",
255 local_softirq_pending());
256 ratelimit++;
257 }
857f3fd7 258 goto end;
35282316 259 }
79bf2bb3 260
79bf2bb3 261 ts->idle_calls++;
79bf2bb3
TG
262 /* Read jiffies and the time when jiffies were updated last */
263 do {
264 seq = read_seqbegin(&xtime_lock);
265 last_update = last_jiffies_update;
266 last_jiffies = jiffies;
267 } while (read_seqretry(&xtime_lock, seq));
268
269 /* Get the next timer wheel timer */
270 next_jiffies = get_next_timer_interrupt(last_jiffies);
271 delta_jiffies = next_jiffies - last_jiffies;
272
b845b517 273 if (rcu_needs_cpu(cpu) || printk_needs_cpu(cpu))
6ba9b346 274 delta_jiffies = 1;
79bf2bb3
TG
275 /*
276 * Do not stop the tick, if we are only one off
277 * or if the cpu is required for rcu
278 */
6ba9b346 279 if (!ts->tick_stopped && delta_jiffies == 1)
79bf2bb3
TG
280 goto out;
281
282 /* Schedule the tick, if we are at least one jiffie off */
283 if ((long)delta_jiffies >= 1) {
284
00147449
WR
285 /*
286 * calculate the expiry time for the next timer wheel
287 * timer
288 */
289 expires = ktime_add_ns(last_update, tick_period.tv64 *
290 delta_jiffies);
291
292 /*
293 * If this cpu is the one which updates jiffies, then
294 * give up the assignment and let it be taken by the
295 * cpu which runs the tick timer next, which might be
296 * this cpu as well. If we don't drop this here the
297 * jiffies might be stale and do_timer() never
298 * invoked.
299 */
300 if (cpu == tick_do_timer_cpu)
301 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
302
6ba9b346 303 if (delta_jiffies > 1)
6a7b3dc3 304 cpumask_set_cpu(cpu, nohz_cpu_mask);
00147449
WR
305
306 /* Skip reprogram of event if its not changed */
307 if (ts->tick_stopped && ktime_equal(expires, dev->next_event))
308 goto out;
309
79bf2bb3
TG
310 /*
311 * nohz_stop_sched_tick can be called several times before
312 * the nohz_restart_sched_tick is called. This happens when
313 * interrupts arrive which do not cause a reschedule. In the
314 * first call we save the current tick time, so we can restart
315 * the scheduler tick in nohz_restart_sched_tick.
316 */
317 if (!ts->tick_stopped) {
46cb4b7c
SS
318 if (select_nohz_load_balancer(1)) {
319 /*
320 * sched tick not stopped!
321 */
6a7b3dc3 322 cpumask_clear_cpu(cpu, nohz_cpu_mask);
46cb4b7c
SS
323 goto out;
324 }
325
cc584b21 326 ts->idle_tick = hrtimer_get_expires(&ts->sched_timer);
79bf2bb3
TG
327 ts->tick_stopped = 1;
328 ts->idle_jiffies = last_jiffies;
2232c2d8 329 rcu_enter_nohz();
79bf2bb3 330 }
d3ed7824 331
eaad084b
TG
332 ts->idle_sleeps++;
333
334 /*
335 * delta_jiffies >= NEXT_TIMER_MAX_DELTA signals that
336 * there is no timer pending or at least extremly far
337 * into the future (12 days for HZ=1000). In this case
338 * we simply stop the tick timer:
339 */
340 if (unlikely(delta_jiffies >= NEXT_TIMER_MAX_DELTA)) {
341 ts->idle_expires.tv64 = KTIME_MAX;
342 if (ts->nohz_mode == NOHZ_MODE_HIGHRES)
343 hrtimer_cancel(&ts->sched_timer);
344 goto out;
345 }
346
00147449 347 /* Mark expiries */
79bf2bb3 348 ts->idle_expires = expires;
79bf2bb3
TG
349
350 if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
351 hrtimer_start(&ts->sched_timer, expires,
352 HRTIMER_MODE_ABS);
353 /* Check, if the timer was already in the past */
354 if (hrtimer_active(&ts->sched_timer))
355 goto out;
4c9dc641 356 } else if (!tick_program_event(expires, 0))
79bf2bb3
TG
357 goto out;
358 /*
359 * We are past the event already. So we crossed a
360 * jiffie boundary. Update jiffies and raise the
361 * softirq.
362 */
363 tick_do_update_jiffies64(ktime_get());
6a7b3dc3 364 cpumask_clear_cpu(cpu, nohz_cpu_mask);
79bf2bb3
TG
365 }
366 raise_softirq_irqoff(TIMER_SOFTIRQ);
367out:
368 ts->next_jiffies = next_jiffies;
369 ts->last_jiffies = last_jiffies;
4f86d3a8 370 ts->sleep_length = ktime_sub(dev->next_event, now);
79bf2bb3
TG
371end:
372 local_irq_restore(flags);
373}
374
4f86d3a8
LB
375/**
376 * tick_nohz_get_sleep_length - return the length of the current sleep
377 *
378 * Called from power state control code with interrupts disabled
379 */
380ktime_t tick_nohz_get_sleep_length(void)
381{
382 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
383
384 return ts->sleep_length;
385}
386
c34bec5a
TG
387static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
388{
389 hrtimer_cancel(&ts->sched_timer);
268a3dcf 390 hrtimer_set_expires(&ts->sched_timer, ts->idle_tick);
c34bec5a
TG
391
392 while (1) {
393 /* Forward the time to expire in the future */
394 hrtimer_forward(&ts->sched_timer, now, tick_period);
395
396 if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
268a3dcf 397 hrtimer_start_expires(&ts->sched_timer,
c34bec5a
TG
398 HRTIMER_MODE_ABS);
399 /* Check, if the timer was already in the past */
400 if (hrtimer_active(&ts->sched_timer))
401 break;
402 } else {
268a3dcf
TG
403 if (!tick_program_event(
404 hrtimer_get_expires(&ts->sched_timer), 0))
c34bec5a
TG
405 break;
406 }
407 /* Update jiffies and reread time */
408 tick_do_update_jiffies64(now);
409 now = ktime_get();
410 }
411}
412
79bf2bb3 413/**
8dce39c2 414 * tick_nohz_restart_sched_tick - restart the idle tick from the idle task
79bf2bb3
TG
415 *
416 * Restart the idle tick when the CPU is woken up from idle
417 */
418void tick_nohz_restart_sched_tick(void)
419{
420 int cpu = smp_processor_id();
421 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
79741dd3 422#ifndef CONFIG_VIRT_CPU_ACCOUNTING
79bf2bb3 423 unsigned long ticks;
79741dd3 424#endif
6378ddb5 425 ktime_t now;
79bf2bb3 426
6378ddb5
VP
427 local_irq_disable();
428 tick_nohz_stop_idle(cpu);
429
b8f8c3cf
TG
430 if (!ts->inidle || !ts->tick_stopped) {
431 ts->inidle = 0;
6378ddb5 432 local_irq_enable();
79bf2bb3 433 return;
6378ddb5 434 }
79bf2bb3 435
b8f8c3cf
TG
436 ts->inidle = 0;
437
2232c2d8
SR
438 rcu_exit_nohz();
439
79bf2bb3 440 /* Update jiffies first */
46cb4b7c 441 select_nohz_load_balancer(0);
6378ddb5 442 now = ktime_get();
79bf2bb3 443 tick_do_update_jiffies64(now);
6a7b3dc3 444 cpumask_clear_cpu(cpu, nohz_cpu_mask);
79bf2bb3 445
79741dd3 446#ifndef CONFIG_VIRT_CPU_ACCOUNTING
79bf2bb3
TG
447 /*
448 * We stopped the tick in idle. Update process times would miss the
449 * time we slept as update_process_times does only a 1 tick
450 * accounting. Enforce that this is accounted to idle !
451 */
452 ticks = jiffies - ts->idle_jiffies;
453 /*
454 * We might be one off. Do not randomly account a huge number of ticks!
455 */
79741dd3
MS
456 if (ticks && ticks < LONG_MAX)
457 account_idle_ticks(ticks);
458#endif
79bf2bb3 459
126e01bf 460 touch_softlockup_watchdog();
79bf2bb3
TG
461 /*
462 * Cancel the scheduled timer and restore the tick
463 */
464 ts->tick_stopped = 0;
5df7fa1c 465 ts->idle_exittime = now;
79bf2bb3 466
c34bec5a 467 tick_nohz_restart(ts, now);
79bf2bb3 468
79bf2bb3
TG
469 local_irq_enable();
470}
471
472static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now)
473{
474 hrtimer_forward(&ts->sched_timer, now, tick_period);
cc584b21 475 return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0);
79bf2bb3
TG
476}
477
478/*
479 * The nohz low res interrupt handler
480 */
481static void tick_nohz_handler(struct clock_event_device *dev)
482{
483 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
484 struct pt_regs *regs = get_irq_regs();
d3ed7824 485 int cpu = smp_processor_id();
79bf2bb3
TG
486 ktime_t now = ktime_get();
487
488 dev->next_event.tv64 = KTIME_MAX;
489
d3ed7824
TG
490 /*
491 * Check if the do_timer duty was dropped. We don't care about
492 * concurrency: This happens only when the cpu in charge went
493 * into a long sleep. If two cpus happen to assign themself to
494 * this duty, then the jiffies update is still serialized by
495 * xtime_lock.
496 */
6441402b 497 if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
d3ed7824
TG
498 tick_do_timer_cpu = cpu;
499
79bf2bb3 500 /* Check, if the jiffies need an update */
d3ed7824
TG
501 if (tick_do_timer_cpu == cpu)
502 tick_do_update_jiffies64(now);
79bf2bb3
TG
503
504 /*
505 * When we are idle and the tick is stopped, we have to touch
506 * the watchdog as we might not schedule for a really long
507 * time. This happens on complete idle SMP systems while
508 * waiting on the login prompt. We also increment the "start
509 * of idle" jiffy stamp so the idle accounting adjustment we
510 * do when we go busy again does not account too much ticks.
511 */
512 if (ts->tick_stopped) {
513 touch_softlockup_watchdog();
514 ts->idle_jiffies++;
515 }
516
517 update_process_times(user_mode(regs));
518 profile_tick(CPU_PROFILING);
519
79bf2bb3
TG
520 while (tick_nohz_reprogram(ts, now)) {
521 now = ktime_get();
522 tick_do_update_jiffies64(now);
523 }
524}
525
526/**
527 * tick_nohz_switch_to_nohz - switch to nohz mode
528 */
529static void tick_nohz_switch_to_nohz(void)
530{
531 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
532 ktime_t next;
533
534 if (!tick_nohz_enabled)
535 return;
536
537 local_irq_disable();
538 if (tick_switch_to_oneshot(tick_nohz_handler)) {
539 local_irq_enable();
540 return;
541 }
542
543 ts->nohz_mode = NOHZ_MODE_LOWRES;
544
545 /*
546 * Recycle the hrtimer in ts, so we can share the
547 * hrtimer_forward with the highres code.
548 */
549 hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
550 /* Get the next period */
551 next = tick_init_jiffy_update();
552
553 for (;;) {
cc584b21 554 hrtimer_set_expires(&ts->sched_timer, next);
79bf2bb3
TG
555 if (!tick_program_event(next, 0))
556 break;
557 next = ktime_add(next, tick_period);
558 }
559 local_irq_enable();
560
561 printk(KERN_INFO "Switched to NOHz mode on CPU #%d\n",
562 smp_processor_id());
563}
564
fb02fbc1
TG
565/*
566 * When NOHZ is enabled and the tick is stopped, we need to kick the
567 * tick timer from irq_enter() so that the jiffies update is kept
568 * alive during long running softirqs. That's ugly as hell, but
569 * correctness is key even if we need to fix the offending softirq in
570 * the first place.
571 *
572 * Note, this is different to tick_nohz_restart. We just kick the
573 * timer and do not touch the other magic bits which need to be done
574 * when idle is left.
575 */
576static void tick_nohz_kick_tick(int cpu)
577{
ae99286b
TG
578#if 0
579 /* Switch back to 2.6.27 behaviour */
580
fb02fbc1 581 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
c4bd822e 582 ktime_t delta, now;
fb02fbc1
TG
583
584 if (!ts->tick_stopped)
585 return;
586
c4bd822e
TG
587 /*
588 * Do not touch the tick device, when the next expiry is either
589 * already reached or less/equal than the tick period.
590 */
591 now = ktime_get();
268a3dcf 592 delta = ktime_sub(hrtimer_get_expires(&ts->sched_timer), now);
c4bd822e
TG
593 if (delta.tv64 <= tick_period.tv64)
594 return;
595
596 tick_nohz_restart(ts, now);
ae99286b 597#endif
fb02fbc1
TG
598}
599
79bf2bb3
TG
600#else
601
602static inline void tick_nohz_switch_to_nohz(void) { }
603
604#endif /* NO_HZ */
605
719254fa
TG
606/*
607 * Called from irq_enter to notify about the possible interruption of idle()
608 */
609void tick_check_idle(int cpu)
610{
fb02fbc1 611 tick_check_oneshot_broadcast(cpu);
719254fa
TG
612#ifdef CONFIG_NO_HZ
613 tick_nohz_stop_idle(cpu);
614 tick_nohz_update_jiffies();
fb02fbc1 615 tick_nohz_kick_tick(cpu);
719254fa
TG
616#endif
617}
618
79bf2bb3
TG
619/*
620 * High resolution timer specific code
621 */
622#ifdef CONFIG_HIGH_RES_TIMERS
623/*
4c9dc641 624 * We rearm the timer until we get disabled by the idle code.
79bf2bb3
TG
625 * Called with interrupts disabled and timer->base->cpu_base->lock held.
626 */
627static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer)
628{
629 struct tick_sched *ts =
630 container_of(timer, struct tick_sched, sched_timer);
79bf2bb3
TG
631 struct pt_regs *regs = get_irq_regs();
632 ktime_t now = ktime_get();
d3ed7824
TG
633 int cpu = smp_processor_id();
634
635#ifdef CONFIG_NO_HZ
636 /*
637 * Check if the do_timer duty was dropped. We don't care about
638 * concurrency: This happens only when the cpu in charge went
639 * into a long sleep. If two cpus happen to assign themself to
640 * this duty, then the jiffies update is still serialized by
641 * xtime_lock.
642 */
6441402b 643 if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
d3ed7824
TG
644 tick_do_timer_cpu = cpu;
645#endif
79bf2bb3
TG
646
647 /* Check, if the jiffies need an update */
d3ed7824
TG
648 if (tick_do_timer_cpu == cpu)
649 tick_do_update_jiffies64(now);
79bf2bb3
TG
650
651 /*
652 * Do not call, when we are not in irq context and have
653 * no valid regs pointer
654 */
655 if (regs) {
656 /*
657 * When we are idle and the tick is stopped, we have to touch
658 * the watchdog as we might not schedule for a really long
659 * time. This happens on complete idle SMP systems while
660 * waiting on the login prompt. We also increment the "start of
661 * idle" jiffy stamp so the idle accounting adjustment we do
662 * when we go busy again does not account too much ticks.
663 */
664 if (ts->tick_stopped) {
665 touch_softlockup_watchdog();
666 ts->idle_jiffies++;
667 }
79bf2bb3
TG
668 update_process_times(user_mode(regs));
669 profile_tick(CPU_PROFILING);
79bf2bb3
TG
670 }
671
79bf2bb3
TG
672 hrtimer_forward(timer, now, tick_period);
673
674 return HRTIMER_RESTART;
675}
676
677/**
678 * tick_setup_sched_timer - setup the tick emulation timer
679 */
680void tick_setup_sched_timer(void)
681{
682 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
683 ktime_t now = ktime_get();
3704540b 684 u64 offset;
79bf2bb3
TG
685
686 /*
687 * Emulate tick processing via per-CPU hrtimers:
688 */
689 hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
690 ts->sched_timer.function = tick_sched_timer;
79bf2bb3 691
3704540b 692 /* Get the next period (per cpu) */
cc584b21 693 hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update());
3704540b 694 offset = ktime_to_ns(tick_period) >> 1;
b2d9323d 695 do_div(offset, num_possible_cpus());
3704540b 696 offset *= smp_processor_id();
cc584b21 697 hrtimer_add_expires_ns(&ts->sched_timer, offset);
79bf2bb3
TG
698
699 for (;;) {
700 hrtimer_forward(&ts->sched_timer, now, tick_period);
cc584b21 701 hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS);
79bf2bb3
TG
702 /* Check, if the timer was already in the past */
703 if (hrtimer_active(&ts->sched_timer))
704 break;
705 now = ktime_get();
706 }
707
708#ifdef CONFIG_NO_HZ
709 if (tick_nohz_enabled)
710 ts->nohz_mode = NOHZ_MODE_HIGHRES;
711#endif
712}
3c4fbe5e 713#endif /* HIGH_RES_TIMERS */
79bf2bb3 714
3c4fbe5e 715#if defined CONFIG_NO_HZ || defined CONFIG_HIGH_RES_TIMERS
79bf2bb3
TG
716void tick_cancel_sched_timer(int cpu)
717{
718 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
719
3c4fbe5e 720# ifdef CONFIG_HIGH_RES_TIMERS
79bf2bb3
TG
721 if (ts->sched_timer.base)
722 hrtimer_cancel(&ts->sched_timer);
3c4fbe5e 723# endif
a7901766 724
79bf2bb3
TG
725 ts->nohz_mode = NOHZ_MODE_INACTIVE;
726}
3c4fbe5e 727#endif
79bf2bb3
TG
728
729/**
730 * Async notification about clocksource changes
731 */
732void tick_clock_notify(void)
733{
734 int cpu;
735
736 for_each_possible_cpu(cpu)
737 set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks);
738}
739
740/*
741 * Async notification about clock event changes
742 */
743void tick_oneshot_notify(void)
744{
745 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
746
747 set_bit(0, &ts->check_clocks);
748}
749
750/**
751 * Check, if a change happened, which makes oneshot possible.
752 *
753 * Called cyclic from the hrtimer softirq (driven by the timer
754 * softirq) allow_nohz signals, that we can switch into low-res nohz
755 * mode, because high resolution timers are disabled (either compile
756 * or runtime).
757 */
758int tick_check_oneshot_change(int allow_nohz)
759{
760 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
761
762 if (!test_and_clear_bit(0, &ts->check_clocks))
763 return 0;
764
765 if (ts->nohz_mode != NOHZ_MODE_INACTIVE)
766 return 0;
767
cf4fc6cb 768 if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available())
79bf2bb3
TG
769 return 0;
770
771 if (!allow_nohz)
772 return 1;
773
774 tick_nohz_switch_to_nohz();
775 return 0;
776}