Merge branch 'fix/hda' into for-linus
[linux-2.6-block.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 */
eed3b9cf 137static void tick_nohz_update_jiffies(ktime_t now)
79bf2bb3
TG
138{
139 int cpu = smp_processor_id();
140 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
141 unsigned long flags;
79bf2bb3 142
6a7b3dc3 143 cpumask_clear_cpu(cpu, nohz_cpu_mask);
5df7fa1c 144 ts->idle_waketime = now;
79bf2bb3
TG
145
146 local_irq_save(flags);
147 tick_do_update_jiffies64(now);
148 local_irq_restore(flags);
02ff3755
IM
149
150 touch_softlockup_watchdog();
79bf2bb3
TG
151}
152
eed3b9cf 153static void tick_nohz_stop_idle(int cpu, ktime_t now)
6378ddb5
VP
154{
155 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
eed3b9cf 156 ktime_t delta;
6378ddb5 157
eed3b9cf
MS
158 delta = ktime_sub(now, ts->idle_entrytime);
159 ts->idle_lastupdate = now;
160 ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
161 ts->idle_active = 0;
56c7426b 162
eed3b9cf 163 sched_clock_idle_wakeup_event(0);
6378ddb5
VP
164}
165
903b8a8d 166static ktime_t tick_nohz_start_idle(struct tick_sched *ts)
6378ddb5 167{
6378ddb5
VP
168 ktime_t now, delta;
169
170 now = ktime_get();
171 if (ts->idle_active) {
172 delta = ktime_sub(now, ts->idle_entrytime);
173 ts->idle_lastupdate = now;
174 ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
175 }
176 ts->idle_entrytime = now;
177 ts->idle_active = 1;
56c7426b 178 sched_clock_idle_sleep_event();
6378ddb5
VP
179 return now;
180}
181
182u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
183{
184 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
185
8083e4ad 186 if (!tick_nohz_enabled)
187 return -1;
188
189 if (ts->idle_active)
190 *last_update_time = ktime_to_us(ts->idle_lastupdate);
191 else
192 *last_update_time = ktime_to_us(ktime_get());
193
6378ddb5
VP
194 return ktime_to_us(ts->idle_sleeptime);
195}
8083e4ad 196EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
6378ddb5 197
79bf2bb3
TG
198/**
199 * tick_nohz_stop_sched_tick - stop the idle tick from the idle task
200 *
201 * When the next event is more than a tick into the future, stop the idle tick
202 * Called either from the idle loop or from irq_exit() when an idle period was
203 * just interrupted by an interrupt which did not cause a reschedule.
204 */
b8f8c3cf 205void tick_nohz_stop_sched_tick(int inidle)
79bf2bb3
TG
206{
207 unsigned long seq, last_jiffies, next_jiffies, delta_jiffies, flags;
208 struct tick_sched *ts;
6378ddb5 209 ktime_t last_update, expires, now;
4f86d3a8 210 struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev;
98962465 211 u64 time_delta;
79bf2bb3
TG
212 int cpu;
213
214 local_irq_save(flags);
215
216 cpu = smp_processor_id();
217 ts = &per_cpu(tick_cpu_sched, cpu);
f2e21c96
EN
218
219 /*
220 * Call to tick_nohz_start_idle stops the last_update_time from being
221 * updated. Thus, it must not be called in the event we are called from
222 * irq_exit() with the prior state different than idle.
223 */
224 if (!inidle && !ts->inidle)
225 goto end;
226
fdc6f192
EN
227 /*
228 * Set ts->inidle unconditionally. Even if the system did not
229 * switch to NOHZ mode the cpu frequency governers rely on the
230 * update of the idle time accounting in tick_nohz_start_idle().
231 */
232 ts->inidle = 1;
233
903b8a8d 234 now = tick_nohz_start_idle(ts);
79bf2bb3 235
5e41d0d6
TG
236 /*
237 * If this cpu is offline and it is the one which updates
238 * jiffies, then give up the assignment and let it be taken by
239 * the cpu which runs the tick timer next. If we don't drop
240 * this here the jiffies might be stale and do_timer() never
241 * invoked.
242 */
243 if (unlikely(!cpu_online(cpu))) {
244 if (cpu == tick_do_timer_cpu)
6441402b 245 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
5e41d0d6
TG
246 }
247
79bf2bb3
TG
248 if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE))
249 goto end;
250
251 if (need_resched())
252 goto end;
253
fa116ea3 254 if (unlikely(local_softirq_pending() && cpu_online(cpu))) {
35282316
TG
255 static int ratelimit;
256
257 if (ratelimit < 10) {
258 printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n",
529eaccd 259 (unsigned int) local_softirq_pending());
35282316
TG
260 ratelimit++;
261 }
857f3fd7 262 goto end;
35282316 263 }
79bf2bb3 264
79bf2bb3 265 ts->idle_calls++;
79bf2bb3
TG
266 /* Read jiffies and the time when jiffies were updated last */
267 do {
268 seq = read_seqbegin(&xtime_lock);
269 last_update = last_jiffies_update;
270 last_jiffies = jiffies;
27185016 271 time_delta = timekeeping_max_deferment();
79bf2bb3
TG
272 } while (read_seqretry(&xtime_lock, seq));
273
3c5d92a0
MS
274 if (rcu_needs_cpu(cpu) || printk_needs_cpu(cpu) ||
275 arch_needs_cpu(cpu)) {
276 next_jiffies = last_jiffies + 1;
6ba9b346 277 delta_jiffies = 1;
3c5d92a0
MS
278 } else {
279 /* Get the next timer wheel timer */
280 next_jiffies = get_next_timer_interrupt(last_jiffies);
281 delta_jiffies = next_jiffies - last_jiffies;
282 }
79bf2bb3
TG
283 /*
284 * Do not stop the tick, if we are only one off
285 * or if the cpu is required for rcu
286 */
6ba9b346 287 if (!ts->tick_stopped && delta_jiffies == 1)
79bf2bb3
TG
288 goto out;
289
290 /* Schedule the tick, if we are at least one jiffie off */
291 if ((long)delta_jiffies >= 1) {
292
00147449
WR
293 /*
294 * If this cpu is the one which updates jiffies, then
295 * give up the assignment and let it be taken by the
296 * cpu which runs the tick timer next, which might be
297 * this cpu as well. If we don't drop this here the
298 * jiffies might be stale and do_timer() never
27185016
TG
299 * invoked. Keep track of the fact that it was the one
300 * which had the do_timer() duty last. If this cpu is
301 * the one which had the do_timer() duty last, we
302 * limit the sleep time to the timekeeping
303 * max_deferement value which we retrieved
304 * above. Otherwise we can sleep as long as we want.
00147449 305 */
27185016 306 if (cpu == tick_do_timer_cpu) {
00147449 307 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
27185016
TG
308 ts->do_timer_last = 1;
309 } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) {
310 time_delta = KTIME_MAX;
311 ts->do_timer_last = 0;
312 } else if (!ts->do_timer_last) {
313 time_delta = KTIME_MAX;
314 }
315
00147449 316 /*
98962465
JH
317 * calculate the expiry time for the next timer wheel
318 * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals
319 * that there is no timer pending or at least extremely
320 * far into the future (12 days for HZ=1000). In this
321 * case we set the expiry to the end of time.
322 */
323 if (likely(delta_jiffies < NEXT_TIMER_MAX_DELTA)) {
324 /*
325 * Calculate the time delta for the next timer event.
326 * If the time delta exceeds the maximum time delta
327 * permitted by the current clocksource then adjust
328 * the time delta accordingly to ensure the
329 * clocksource does not wrap.
330 */
331 time_delta = min_t(u64, time_delta,
332 tick_period.tv64 * delta_jiffies);
98962465 333 }
00147449 334
27185016
TG
335 if (time_delta < KTIME_MAX)
336 expires = ktime_add_ns(last_update, time_delta);
337 else
338 expires.tv64 = KTIME_MAX;
00147449 339
6ba9b346 340 if (delta_jiffies > 1)
6a7b3dc3 341 cpumask_set_cpu(cpu, nohz_cpu_mask);
00147449
WR
342
343 /* Skip reprogram of event if its not changed */
344 if (ts->tick_stopped && ktime_equal(expires, dev->next_event))
345 goto out;
346
79bf2bb3
TG
347 /*
348 * nohz_stop_sched_tick can be called several times before
349 * the nohz_restart_sched_tick is called. This happens when
350 * interrupts arrive which do not cause a reschedule. In the
351 * first call we save the current tick time, so we can restart
352 * the scheduler tick in nohz_restart_sched_tick.
353 */
354 if (!ts->tick_stopped) {
46cb4b7c
SS
355 if (select_nohz_load_balancer(1)) {
356 /*
357 * sched tick not stopped!
358 */
6a7b3dc3 359 cpumask_clear_cpu(cpu, nohz_cpu_mask);
46cb4b7c
SS
360 goto out;
361 }
362
cc584b21 363 ts->idle_tick = hrtimer_get_expires(&ts->sched_timer);
79bf2bb3
TG
364 ts->tick_stopped = 1;
365 ts->idle_jiffies = last_jiffies;
2232c2d8 366 rcu_enter_nohz();
79bf2bb3 367 }
d3ed7824 368
eaad084b
TG
369 ts->idle_sleeps++;
370
98962465
JH
371 /* Mark expires */
372 ts->idle_expires = expires;
373
eaad084b 374 /*
98962465
JH
375 * If the expiration time == KTIME_MAX, then
376 * in this case we simply stop the tick timer.
eaad084b 377 */
98962465 378 if (unlikely(expires.tv64 == KTIME_MAX)) {
eaad084b
TG
379 if (ts->nohz_mode == NOHZ_MODE_HIGHRES)
380 hrtimer_cancel(&ts->sched_timer);
381 goto out;
382 }
383
79bf2bb3
TG
384 if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
385 hrtimer_start(&ts->sched_timer, expires,
5c333864 386 HRTIMER_MODE_ABS_PINNED);
79bf2bb3
TG
387 /* Check, if the timer was already in the past */
388 if (hrtimer_active(&ts->sched_timer))
389 goto out;
4c9dc641 390 } else if (!tick_program_event(expires, 0))
79bf2bb3
TG
391 goto out;
392 /*
393 * We are past the event already. So we crossed a
394 * jiffie boundary. Update jiffies and raise the
395 * softirq.
396 */
397 tick_do_update_jiffies64(ktime_get());
6a7b3dc3 398 cpumask_clear_cpu(cpu, nohz_cpu_mask);
79bf2bb3
TG
399 }
400 raise_softirq_irqoff(TIMER_SOFTIRQ);
401out:
402 ts->next_jiffies = next_jiffies;
403 ts->last_jiffies = last_jiffies;
4f86d3a8 404 ts->sleep_length = ktime_sub(dev->next_event, now);
79bf2bb3
TG
405end:
406 local_irq_restore(flags);
407}
408
4f86d3a8
LB
409/**
410 * tick_nohz_get_sleep_length - return the length of the current sleep
411 *
412 * Called from power state control code with interrupts disabled
413 */
414ktime_t tick_nohz_get_sleep_length(void)
415{
416 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
417
418 return ts->sleep_length;
419}
420
c34bec5a
TG
421static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
422{
423 hrtimer_cancel(&ts->sched_timer);
268a3dcf 424 hrtimer_set_expires(&ts->sched_timer, ts->idle_tick);
c34bec5a
TG
425
426 while (1) {
427 /* Forward the time to expire in the future */
428 hrtimer_forward(&ts->sched_timer, now, tick_period);
429
430 if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
268a3dcf 431 hrtimer_start_expires(&ts->sched_timer,
5c333864 432 HRTIMER_MODE_ABS_PINNED);
c34bec5a
TG
433 /* Check, if the timer was already in the past */
434 if (hrtimer_active(&ts->sched_timer))
435 break;
436 } else {
268a3dcf
TG
437 if (!tick_program_event(
438 hrtimer_get_expires(&ts->sched_timer), 0))
c34bec5a
TG
439 break;
440 }
441 /* Update jiffies and reread time */
442 tick_do_update_jiffies64(now);
443 now = ktime_get();
444 }
445}
446
79bf2bb3 447/**
8dce39c2 448 * tick_nohz_restart_sched_tick - restart the idle tick from the idle task
79bf2bb3
TG
449 *
450 * Restart the idle tick when the CPU is woken up from idle
451 */
452void tick_nohz_restart_sched_tick(void)
453{
454 int cpu = smp_processor_id();
455 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
79741dd3 456#ifndef CONFIG_VIRT_CPU_ACCOUNTING
79bf2bb3 457 unsigned long ticks;
79741dd3 458#endif
6378ddb5 459 ktime_t now;
79bf2bb3 460
6378ddb5 461 local_irq_disable();
eed3b9cf
MS
462 if (ts->idle_active || (ts->inidle && ts->tick_stopped))
463 now = ktime_get();
464
465 if (ts->idle_active)
466 tick_nohz_stop_idle(cpu, now);
6378ddb5 467
b8f8c3cf
TG
468 if (!ts->inidle || !ts->tick_stopped) {
469 ts->inidle = 0;
6378ddb5 470 local_irq_enable();
79bf2bb3 471 return;
6378ddb5 472 }
79bf2bb3 473
b8f8c3cf
TG
474 ts->inidle = 0;
475
2232c2d8
SR
476 rcu_exit_nohz();
477
79bf2bb3 478 /* Update jiffies first */
46cb4b7c 479 select_nohz_load_balancer(0);
79bf2bb3 480 tick_do_update_jiffies64(now);
6a7b3dc3 481 cpumask_clear_cpu(cpu, nohz_cpu_mask);
79bf2bb3 482
79741dd3 483#ifndef CONFIG_VIRT_CPU_ACCOUNTING
79bf2bb3
TG
484 /*
485 * We stopped the tick in idle. Update process times would miss the
486 * time we slept as update_process_times does only a 1 tick
487 * accounting. Enforce that this is accounted to idle !
488 */
489 ticks = jiffies - ts->idle_jiffies;
490 /*
491 * We might be one off. Do not randomly account a huge number of ticks!
492 */
79741dd3
MS
493 if (ticks && ticks < LONG_MAX)
494 account_idle_ticks(ticks);
495#endif
79bf2bb3 496
126e01bf 497 touch_softlockup_watchdog();
79bf2bb3
TG
498 /*
499 * Cancel the scheduled timer and restore the tick
500 */
501 ts->tick_stopped = 0;
5df7fa1c 502 ts->idle_exittime = now;
79bf2bb3 503
c34bec5a 504 tick_nohz_restart(ts, now);
79bf2bb3 505
79bf2bb3
TG
506 local_irq_enable();
507}
508
509static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now)
510{
511 hrtimer_forward(&ts->sched_timer, now, tick_period);
cc584b21 512 return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0);
79bf2bb3
TG
513}
514
515/*
516 * The nohz low res interrupt handler
517 */
518static void tick_nohz_handler(struct clock_event_device *dev)
519{
520 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
521 struct pt_regs *regs = get_irq_regs();
d3ed7824 522 int cpu = smp_processor_id();
79bf2bb3
TG
523 ktime_t now = ktime_get();
524
525 dev->next_event.tv64 = KTIME_MAX;
526
d3ed7824
TG
527 /*
528 * Check if the do_timer duty was dropped. We don't care about
529 * concurrency: This happens only when the cpu in charge went
530 * into a long sleep. If two cpus happen to assign themself to
531 * this duty, then the jiffies update is still serialized by
532 * xtime_lock.
533 */
6441402b 534 if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
d3ed7824
TG
535 tick_do_timer_cpu = cpu;
536
79bf2bb3 537 /* Check, if the jiffies need an update */
d3ed7824
TG
538 if (tick_do_timer_cpu == cpu)
539 tick_do_update_jiffies64(now);
79bf2bb3
TG
540
541 /*
542 * When we are idle and the tick is stopped, we have to touch
543 * the watchdog as we might not schedule for a really long
544 * time. This happens on complete idle SMP systems while
545 * waiting on the login prompt. We also increment the "start
546 * of idle" jiffy stamp so the idle accounting adjustment we
547 * do when we go busy again does not account too much ticks.
548 */
549 if (ts->tick_stopped) {
550 touch_softlockup_watchdog();
551 ts->idle_jiffies++;
552 }
553
554 update_process_times(user_mode(regs));
555 profile_tick(CPU_PROFILING);
556
79bf2bb3
TG
557 while (tick_nohz_reprogram(ts, now)) {
558 now = ktime_get();
559 tick_do_update_jiffies64(now);
560 }
561}
562
563/**
564 * tick_nohz_switch_to_nohz - switch to nohz mode
565 */
566static void tick_nohz_switch_to_nohz(void)
567{
568 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
569 ktime_t next;
570
571 if (!tick_nohz_enabled)
572 return;
573
574 local_irq_disable();
575 if (tick_switch_to_oneshot(tick_nohz_handler)) {
576 local_irq_enable();
577 return;
578 }
579
580 ts->nohz_mode = NOHZ_MODE_LOWRES;
581
582 /*
583 * Recycle the hrtimer in ts, so we can share the
584 * hrtimer_forward with the highres code.
585 */
586 hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
587 /* Get the next period */
588 next = tick_init_jiffy_update();
589
590 for (;;) {
cc584b21 591 hrtimer_set_expires(&ts->sched_timer, next);
79bf2bb3
TG
592 if (!tick_program_event(next, 0))
593 break;
594 next = ktime_add(next, tick_period);
595 }
596 local_irq_enable();
597
598 printk(KERN_INFO "Switched to NOHz mode on CPU #%d\n",
599 smp_processor_id());
600}
601
fb02fbc1
TG
602/*
603 * When NOHZ is enabled and the tick is stopped, we need to kick the
604 * tick timer from irq_enter() so that the jiffies update is kept
605 * alive during long running softirqs. That's ugly as hell, but
606 * correctness is key even if we need to fix the offending softirq in
607 * the first place.
608 *
609 * Note, this is different to tick_nohz_restart. We just kick the
610 * timer and do not touch the other magic bits which need to be done
611 * when idle is left.
612 */
eed3b9cf 613static void tick_nohz_kick_tick(int cpu, ktime_t now)
fb02fbc1 614{
ae99286b
TG
615#if 0
616 /* Switch back to 2.6.27 behaviour */
617
fb02fbc1 618 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
eed3b9cf 619 ktime_t delta;
fb02fbc1 620
c4bd822e
TG
621 /*
622 * Do not touch the tick device, when the next expiry is either
623 * already reached or less/equal than the tick period.
624 */
268a3dcf 625 delta = ktime_sub(hrtimer_get_expires(&ts->sched_timer), now);
c4bd822e
TG
626 if (delta.tv64 <= tick_period.tv64)
627 return;
628
629 tick_nohz_restart(ts, now);
ae99286b 630#endif
fb02fbc1
TG
631}
632
eed3b9cf
MS
633static inline void tick_check_nohz(int cpu)
634{
635 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
636 ktime_t now;
637
638 if (!ts->idle_active && !ts->tick_stopped)
639 return;
640 now = ktime_get();
641 if (ts->idle_active)
642 tick_nohz_stop_idle(cpu, now);
643 if (ts->tick_stopped) {
644 tick_nohz_update_jiffies(now);
645 tick_nohz_kick_tick(cpu, now);
646 }
647}
648
79bf2bb3
TG
649#else
650
651static inline void tick_nohz_switch_to_nohz(void) { }
eed3b9cf 652static inline void tick_check_nohz(int cpu) { }
79bf2bb3
TG
653
654#endif /* NO_HZ */
655
719254fa
TG
656/*
657 * Called from irq_enter to notify about the possible interruption of idle()
658 */
659void tick_check_idle(int cpu)
660{
fb02fbc1 661 tick_check_oneshot_broadcast(cpu);
eed3b9cf 662 tick_check_nohz(cpu);
719254fa
TG
663}
664
79bf2bb3
TG
665/*
666 * High resolution timer specific code
667 */
668#ifdef CONFIG_HIGH_RES_TIMERS
669/*
4c9dc641 670 * We rearm the timer until we get disabled by the idle code.
79bf2bb3
TG
671 * Called with interrupts disabled and timer->base->cpu_base->lock held.
672 */
673static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer)
674{
675 struct tick_sched *ts =
676 container_of(timer, struct tick_sched, sched_timer);
79bf2bb3
TG
677 struct pt_regs *regs = get_irq_regs();
678 ktime_t now = ktime_get();
d3ed7824
TG
679 int cpu = smp_processor_id();
680
681#ifdef CONFIG_NO_HZ
682 /*
683 * Check if the do_timer duty was dropped. We don't care about
684 * concurrency: This happens only when the cpu in charge went
685 * into a long sleep. If two cpus happen to assign themself to
686 * this duty, then the jiffies update is still serialized by
687 * xtime_lock.
688 */
6441402b 689 if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
d3ed7824
TG
690 tick_do_timer_cpu = cpu;
691#endif
79bf2bb3
TG
692
693 /* Check, if the jiffies need an update */
d3ed7824
TG
694 if (tick_do_timer_cpu == cpu)
695 tick_do_update_jiffies64(now);
79bf2bb3
TG
696
697 /*
698 * Do not call, when we are not in irq context and have
699 * no valid regs pointer
700 */
701 if (regs) {
702 /*
703 * When we are idle and the tick is stopped, we have to touch
704 * the watchdog as we might not schedule for a really long
705 * time. This happens on complete idle SMP systems while
706 * waiting on the login prompt. We also increment the "start of
707 * idle" jiffy stamp so the idle accounting adjustment we do
708 * when we go busy again does not account too much ticks.
709 */
710 if (ts->tick_stopped) {
711 touch_softlockup_watchdog();
712 ts->idle_jiffies++;
713 }
79bf2bb3
TG
714 update_process_times(user_mode(regs));
715 profile_tick(CPU_PROFILING);
79bf2bb3
TG
716 }
717
79bf2bb3
TG
718 hrtimer_forward(timer, now, tick_period);
719
720 return HRTIMER_RESTART;
721}
722
723/**
724 * tick_setup_sched_timer - setup the tick emulation timer
725 */
726void tick_setup_sched_timer(void)
727{
728 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
729 ktime_t now = ktime_get();
3704540b 730 u64 offset;
79bf2bb3
TG
731
732 /*
733 * Emulate tick processing via per-CPU hrtimers:
734 */
735 hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
736 ts->sched_timer.function = tick_sched_timer;
79bf2bb3 737
3704540b 738 /* Get the next period (per cpu) */
cc584b21 739 hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update());
3704540b 740 offset = ktime_to_ns(tick_period) >> 1;
b2d9323d 741 do_div(offset, num_possible_cpus());
3704540b 742 offset *= smp_processor_id();
cc584b21 743 hrtimer_add_expires_ns(&ts->sched_timer, offset);
79bf2bb3
TG
744
745 for (;;) {
746 hrtimer_forward(&ts->sched_timer, now, tick_period);
5c333864
AB
747 hrtimer_start_expires(&ts->sched_timer,
748 HRTIMER_MODE_ABS_PINNED);
79bf2bb3
TG
749 /* Check, if the timer was already in the past */
750 if (hrtimer_active(&ts->sched_timer))
751 break;
752 now = ktime_get();
753 }
754
755#ifdef CONFIG_NO_HZ
756 if (tick_nohz_enabled)
757 ts->nohz_mode = NOHZ_MODE_HIGHRES;
758#endif
759}
3c4fbe5e 760#endif /* HIGH_RES_TIMERS */
79bf2bb3 761
3c4fbe5e 762#if defined CONFIG_NO_HZ || defined CONFIG_HIGH_RES_TIMERS
79bf2bb3
TG
763void tick_cancel_sched_timer(int cpu)
764{
765 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
766
3c4fbe5e 767# ifdef CONFIG_HIGH_RES_TIMERS
79bf2bb3
TG
768 if (ts->sched_timer.base)
769 hrtimer_cancel(&ts->sched_timer);
3c4fbe5e 770# endif
a7901766 771
79bf2bb3
TG
772 ts->nohz_mode = NOHZ_MODE_INACTIVE;
773}
3c4fbe5e 774#endif
79bf2bb3
TG
775
776/**
777 * Async notification about clocksource changes
778 */
779void tick_clock_notify(void)
780{
781 int cpu;
782
783 for_each_possible_cpu(cpu)
784 set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks);
785}
786
787/*
788 * Async notification about clock event changes
789 */
790void tick_oneshot_notify(void)
791{
792 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
793
794 set_bit(0, &ts->check_clocks);
795}
796
797/**
798 * Check, if a change happened, which makes oneshot possible.
799 *
800 * Called cyclic from the hrtimer softirq (driven by the timer
801 * softirq) allow_nohz signals, that we can switch into low-res nohz
802 * mode, because high resolution timers are disabled (either compile
803 * or runtime).
804 */
805int tick_check_oneshot_change(int allow_nohz)
806{
807 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
808
809 if (!test_and_clear_bit(0, &ts->check_clocks))
810 return 0;
811
812 if (ts->nohz_mode != NOHZ_MODE_INACTIVE)
813 return 0;
814
cf4fc6cb 815 if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available())
79bf2bb3
TG
816 return 0;
817
818 if (!allow_nohz)
819 return 1;
820
821 tick_nohz_switch_to_nohz();
822 return 0;
823}