Commit | Line | Data |
---|---|---|
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> | |
23 | ||
9e203bcc DM |
24 | #include <asm/irq_regs.h> |
25 | ||
79bf2bb3 TG |
26 | #include "tick-internal.h" |
27 | ||
28 | /* | |
29 | * Per cpu nohz control structure | |
30 | */ | |
31 | static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); | |
32 | ||
33 | /* | |
34 | * The time, when the last jiffy update happened. Protected by xtime_lock. | |
35 | */ | |
36 | static ktime_t last_jiffies_update; | |
37 | ||
289f480a IM |
38 | struct tick_sched *tick_get_tick_sched(int cpu) |
39 | { | |
40 | return &per_cpu(tick_cpu_sched, cpu); | |
41 | } | |
42 | ||
79bf2bb3 TG |
43 | /* |
44 | * Must be called with interrupts disabled ! | |
45 | */ | |
46 | static void tick_do_update_jiffies64(ktime_t now) | |
47 | { | |
48 | unsigned long ticks = 0; | |
49 | ktime_t delta; | |
50 | ||
7a14ce1d IM |
51 | /* |
52 | * Do a quick check without holding xtime_lock: | |
53 | */ | |
54 | delta = ktime_sub(now, last_jiffies_update); | |
55 | if (delta.tv64 < tick_period.tv64) | |
56 | return; | |
57 | ||
79bf2bb3 TG |
58 | /* Reevalute with xtime_lock held */ |
59 | write_seqlock(&xtime_lock); | |
60 | ||
61 | delta = ktime_sub(now, last_jiffies_update); | |
62 | if (delta.tv64 >= tick_period.tv64) { | |
63 | ||
64 | delta = ktime_sub(delta, tick_period); | |
65 | last_jiffies_update = ktime_add(last_jiffies_update, | |
66 | tick_period); | |
67 | ||
68 | /* Slow path for long timeouts */ | |
69 | if (unlikely(delta.tv64 >= tick_period.tv64)) { | |
70 | s64 incr = ktime_to_ns(tick_period); | |
71 | ||
72 | ticks = ktime_divns(delta, incr); | |
73 | ||
74 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
75 | incr * ticks); | |
76 | } | |
77 | do_timer(++ticks); | |
49d670fb TG |
78 | |
79 | /* Keep the tick_next_period variable up to date */ | |
80 | tick_next_period = ktime_add(last_jiffies_update, tick_period); | |
79bf2bb3 TG |
81 | } |
82 | write_sequnlock(&xtime_lock); | |
83 | } | |
84 | ||
85 | /* | |
86 | * Initialize and return retrieve the jiffies update. | |
87 | */ | |
88 | static ktime_t tick_init_jiffy_update(void) | |
89 | { | |
90 | ktime_t period; | |
91 | ||
92 | write_seqlock(&xtime_lock); | |
93 | /* Did we start the jiffies update yet ? */ | |
94 | if (last_jiffies_update.tv64 == 0) | |
95 | last_jiffies_update = tick_next_period; | |
96 | period = last_jiffies_update; | |
97 | write_sequnlock(&xtime_lock); | |
98 | return period; | |
99 | } | |
100 | ||
101 | /* | |
102 | * NOHZ - aka dynamic tick functionality | |
103 | */ | |
104 | #ifdef CONFIG_NO_HZ | |
105 | /* | |
106 | * NO HZ enabled ? | |
107 | */ | |
108 | static int tick_nohz_enabled __read_mostly = 1; | |
109 | ||
110 | /* | |
111 | * Enable / Disable tickless mode | |
112 | */ | |
113 | static int __init setup_tick_nohz(char *str) | |
114 | { | |
115 | if (!strcmp(str, "off")) | |
116 | tick_nohz_enabled = 0; | |
117 | else if (!strcmp(str, "on")) | |
118 | tick_nohz_enabled = 1; | |
119 | else | |
120 | return 0; | |
121 | return 1; | |
122 | } | |
123 | ||
124 | __setup("nohz=", setup_tick_nohz); | |
125 | ||
126 | /** | |
127 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
128 | * | |
129 | * Called from interrupt entry when the CPU was idle | |
130 | * | |
131 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
132 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
133 | * value. We do this unconditionally on any cpu, as we don't know whether the | |
134 | * cpu, which has the update task assigned is in a long sleep. | |
135 | */ | |
136 | void tick_nohz_update_jiffies(void) | |
137 | { | |
138 | int cpu = smp_processor_id(); | |
139 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
140 | unsigned long flags; | |
141 | ktime_t now; | |
142 | ||
143 | if (!ts->tick_stopped) | |
144 | return; | |
145 | ||
146 | cpu_clear(cpu, nohz_cpu_mask); | |
147 | now = ktime_get(); | |
5df7fa1c | 148 | ts->idle_waketime = now; |
79bf2bb3 TG |
149 | |
150 | local_irq_save(flags); | |
151 | tick_do_update_jiffies64(now); | |
152 | local_irq_restore(flags); | |
02ff3755 IM |
153 | |
154 | touch_softlockup_watchdog(); | |
79bf2bb3 TG |
155 | } |
156 | ||
6378ddb5 VP |
157 | void tick_nohz_stop_idle(int cpu) |
158 | { | |
159 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
160 | ||
161 | if (ts->idle_active) { | |
162 | ktime_t now, delta; | |
163 | now = ktime_get(); | |
164 | delta = ktime_sub(now, ts->idle_entrytime); | |
165 | ts->idle_lastupdate = now; | |
166 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
167 | ts->idle_active = 0; | |
56c7426b PZ |
168 | |
169 | sched_clock_idle_wakeup_event(0); | |
6378ddb5 VP |
170 | } |
171 | } | |
172 | ||
903b8a8d | 173 | static ktime_t tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 174 | { |
6378ddb5 VP |
175 | ktime_t now, delta; |
176 | ||
177 | now = ktime_get(); | |
178 | if (ts->idle_active) { | |
179 | delta = ktime_sub(now, ts->idle_entrytime); | |
180 | ts->idle_lastupdate = now; | |
181 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
182 | } | |
183 | ts->idle_entrytime = now; | |
184 | ts->idle_active = 1; | |
56c7426b | 185 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
186 | return now; |
187 | } | |
188 | ||
189 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) | |
190 | { | |
191 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
192 | ||
193 | *last_update_time = ktime_to_us(ts->idle_lastupdate); | |
194 | return ktime_to_us(ts->idle_sleeptime); | |
195 | } | |
196 | ||
79bf2bb3 TG |
197 | /** |
198 | * tick_nohz_stop_sched_tick - stop the idle tick from the idle task | |
199 | * | |
200 | * When the next event is more than a tick into the future, stop the idle tick | |
201 | * Called either from the idle loop or from irq_exit() when an idle period was | |
202 | * just interrupted by an interrupt which did not cause a reschedule. | |
203 | */ | |
b8f8c3cf | 204 | void tick_nohz_stop_sched_tick(int inidle) |
79bf2bb3 TG |
205 | { |
206 | unsigned long seq, last_jiffies, next_jiffies, delta_jiffies, flags; | |
207 | struct tick_sched *ts; | |
6378ddb5 | 208 | ktime_t last_update, expires, now; |
4f86d3a8 | 209 | struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev; |
79bf2bb3 TG |
210 | int cpu; |
211 | ||
212 | local_irq_save(flags); | |
213 | ||
214 | cpu = smp_processor_id(); | |
215 | ts = &per_cpu(tick_cpu_sched, cpu); | |
903b8a8d | 216 | now = tick_nohz_start_idle(ts); |
79bf2bb3 | 217 | |
5e41d0d6 TG |
218 | /* |
219 | * If this cpu is offline and it is the one which updates | |
220 | * jiffies, then give up the assignment and let it be taken by | |
221 | * the cpu which runs the tick timer next. If we don't drop | |
222 | * this here the jiffies might be stale and do_timer() never | |
223 | * invoked. | |
224 | */ | |
225 | if (unlikely(!cpu_online(cpu))) { | |
226 | if (cpu == tick_do_timer_cpu) | |
6441402b | 227 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; |
5e41d0d6 TG |
228 | } |
229 | ||
79bf2bb3 TG |
230 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) |
231 | goto end; | |
232 | ||
b8f8c3cf TG |
233 | if (!inidle && !ts->inidle) |
234 | goto end; | |
235 | ||
236 | ts->inidle = 1; | |
237 | ||
79bf2bb3 TG |
238 | if (need_resched()) |
239 | goto end; | |
240 | ||
35282316 TG |
241 | if (unlikely(local_softirq_pending())) { |
242 | static int ratelimit; | |
243 | ||
244 | if (ratelimit < 10) { | |
245 | printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n", | |
246 | local_softirq_pending()); | |
247 | ratelimit++; | |
248 | } | |
857f3fd7 | 249 | goto end; |
35282316 | 250 | } |
79bf2bb3 | 251 | |
79bf2bb3 | 252 | ts->idle_calls++; |
79bf2bb3 TG |
253 | /* Read jiffies and the time when jiffies were updated last */ |
254 | do { | |
255 | seq = read_seqbegin(&xtime_lock); | |
256 | last_update = last_jiffies_update; | |
257 | last_jiffies = jiffies; | |
258 | } while (read_seqretry(&xtime_lock, seq)); | |
259 | ||
260 | /* Get the next timer wheel timer */ | |
261 | next_jiffies = get_next_timer_interrupt(last_jiffies); | |
262 | delta_jiffies = next_jiffies - last_jiffies; | |
263 | ||
6ba9b346 IM |
264 | if (rcu_needs_cpu(cpu)) |
265 | delta_jiffies = 1; | |
79bf2bb3 TG |
266 | /* |
267 | * Do not stop the tick, if we are only one off | |
268 | * or if the cpu is required for rcu | |
269 | */ | |
6ba9b346 | 270 | if (!ts->tick_stopped && delta_jiffies == 1) |
79bf2bb3 TG |
271 | goto out; |
272 | ||
273 | /* Schedule the tick, if we are at least one jiffie off */ | |
274 | if ((long)delta_jiffies >= 1) { | |
275 | ||
6ba9b346 | 276 | if (delta_jiffies > 1) |
79bf2bb3 TG |
277 | cpu_set(cpu, nohz_cpu_mask); |
278 | /* | |
279 | * nohz_stop_sched_tick can be called several times before | |
280 | * the nohz_restart_sched_tick is called. This happens when | |
281 | * interrupts arrive which do not cause a reschedule. In the | |
282 | * first call we save the current tick time, so we can restart | |
283 | * the scheduler tick in nohz_restart_sched_tick. | |
284 | */ | |
285 | if (!ts->tick_stopped) { | |
46cb4b7c SS |
286 | if (select_nohz_load_balancer(1)) { |
287 | /* | |
288 | * sched tick not stopped! | |
289 | */ | |
290 | cpu_clear(cpu, nohz_cpu_mask); | |
291 | goto out; | |
292 | } | |
293 | ||
79bf2bb3 TG |
294 | ts->idle_tick = ts->sched_timer.expires; |
295 | ts->tick_stopped = 1; | |
296 | ts->idle_jiffies = last_jiffies; | |
2232c2d8 | 297 | rcu_enter_nohz(); |
79bf2bb3 | 298 | } |
d3ed7824 TG |
299 | |
300 | /* | |
301 | * If this cpu is the one which updates jiffies, then | |
302 | * give up the assignment and let it be taken by the | |
303 | * cpu which runs the tick timer next, which might be | |
304 | * this cpu as well. If we don't drop this here the | |
305 | * jiffies might be stale and do_timer() never | |
306 | * invoked. | |
307 | */ | |
308 | if (cpu == tick_do_timer_cpu) | |
6441402b | 309 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; |
d3ed7824 | 310 | |
eaad084b TG |
311 | ts->idle_sleeps++; |
312 | ||
313 | /* | |
314 | * delta_jiffies >= NEXT_TIMER_MAX_DELTA signals that | |
315 | * there is no timer pending or at least extremly far | |
316 | * into the future (12 days for HZ=1000). In this case | |
317 | * we simply stop the tick timer: | |
318 | */ | |
319 | if (unlikely(delta_jiffies >= NEXT_TIMER_MAX_DELTA)) { | |
320 | ts->idle_expires.tv64 = KTIME_MAX; | |
321 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) | |
322 | hrtimer_cancel(&ts->sched_timer); | |
323 | goto out; | |
324 | } | |
325 | ||
79bf2bb3 TG |
326 | /* |
327 | * calculate the expiry time for the next timer wheel | |
328 | * timer | |
329 | */ | |
330 | expires = ktime_add_ns(last_update, tick_period.tv64 * | |
331 | delta_jiffies); | |
332 | ts->idle_expires = expires; | |
79bf2bb3 TG |
333 | |
334 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { | |
335 | hrtimer_start(&ts->sched_timer, expires, | |
336 | HRTIMER_MODE_ABS); | |
337 | /* Check, if the timer was already in the past */ | |
338 | if (hrtimer_active(&ts->sched_timer)) | |
339 | goto out; | |
4c9dc641 | 340 | } else if (!tick_program_event(expires, 0)) |
79bf2bb3 TG |
341 | goto out; |
342 | /* | |
343 | * We are past the event already. So we crossed a | |
344 | * jiffie boundary. Update jiffies and raise the | |
345 | * softirq. | |
346 | */ | |
347 | tick_do_update_jiffies64(ktime_get()); | |
348 | cpu_clear(cpu, nohz_cpu_mask); | |
349 | } | |
350 | raise_softirq_irqoff(TIMER_SOFTIRQ); | |
351 | out: | |
352 | ts->next_jiffies = next_jiffies; | |
353 | ts->last_jiffies = last_jiffies; | |
4f86d3a8 | 354 | ts->sleep_length = ktime_sub(dev->next_event, now); |
79bf2bb3 TG |
355 | end: |
356 | local_irq_restore(flags); | |
357 | } | |
358 | ||
4f86d3a8 LB |
359 | /** |
360 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
361 | * | |
362 | * Called from power state control code with interrupts disabled | |
363 | */ | |
364 | ktime_t tick_nohz_get_sleep_length(void) | |
365 | { | |
366 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
367 | ||
368 | return ts->sleep_length; | |
369 | } | |
370 | ||
79bf2bb3 | 371 | /** |
8dce39c2 | 372 | * tick_nohz_restart_sched_tick - restart the idle tick from the idle task |
79bf2bb3 TG |
373 | * |
374 | * Restart the idle tick when the CPU is woken up from idle | |
375 | */ | |
376 | void tick_nohz_restart_sched_tick(void) | |
377 | { | |
378 | int cpu = smp_processor_id(); | |
379 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
380 | unsigned long ticks; | |
6378ddb5 | 381 | ktime_t now; |
79bf2bb3 | 382 | |
6378ddb5 VP |
383 | local_irq_disable(); |
384 | tick_nohz_stop_idle(cpu); | |
385 | ||
b8f8c3cf TG |
386 | if (!ts->inidle || !ts->tick_stopped) { |
387 | ts->inidle = 0; | |
6378ddb5 | 388 | local_irq_enable(); |
79bf2bb3 | 389 | return; |
6378ddb5 | 390 | } |
79bf2bb3 | 391 | |
b8f8c3cf TG |
392 | ts->inidle = 0; |
393 | ||
2232c2d8 SR |
394 | rcu_exit_nohz(); |
395 | ||
79bf2bb3 | 396 | /* Update jiffies first */ |
46cb4b7c | 397 | select_nohz_load_balancer(0); |
6378ddb5 | 398 | now = ktime_get(); |
79bf2bb3 TG |
399 | tick_do_update_jiffies64(now); |
400 | cpu_clear(cpu, nohz_cpu_mask); | |
401 | ||
79bf2bb3 TG |
402 | /* |
403 | * We stopped the tick in idle. Update process times would miss the | |
404 | * time we slept as update_process_times does only a 1 tick | |
405 | * accounting. Enforce that this is accounted to idle ! | |
406 | */ | |
407 | ticks = jiffies - ts->idle_jiffies; | |
408 | /* | |
409 | * We might be one off. Do not randomly account a huge number of ticks! | |
410 | */ | |
411 | if (ticks && ticks < LONG_MAX) { | |
412 | add_preempt_count(HARDIRQ_OFFSET); | |
413 | account_system_time(current, HARDIRQ_OFFSET, | |
414 | jiffies_to_cputime(ticks)); | |
415 | sub_preempt_count(HARDIRQ_OFFSET); | |
416 | } | |
417 | ||
126e01bf | 418 | touch_softlockup_watchdog(); |
79bf2bb3 TG |
419 | /* |
420 | * Cancel the scheduled timer and restore the tick | |
421 | */ | |
422 | ts->tick_stopped = 0; | |
5df7fa1c | 423 | ts->idle_exittime = now; |
79bf2bb3 TG |
424 | hrtimer_cancel(&ts->sched_timer); |
425 | ts->sched_timer.expires = ts->idle_tick; | |
426 | ||
427 | while (1) { | |
428 | /* Forward the time to expire in the future */ | |
429 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
430 | ||
431 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { | |
432 | hrtimer_start(&ts->sched_timer, | |
433 | ts->sched_timer.expires, | |
434 | HRTIMER_MODE_ABS); | |
435 | /* Check, if the timer was already in the past */ | |
436 | if (hrtimer_active(&ts->sched_timer)) | |
437 | break; | |
438 | } else { | |
439 | if (!tick_program_event(ts->sched_timer.expires, 0)) | |
440 | break; | |
441 | } | |
442 | /* Update jiffies and reread time */ | |
443 | tick_do_update_jiffies64(now); | |
444 | now = ktime_get(); | |
445 | } | |
446 | local_irq_enable(); | |
447 | } | |
448 | ||
449 | static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now) | |
450 | { | |
451 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
452 | return tick_program_event(ts->sched_timer.expires, 0); | |
453 | } | |
454 | ||
455 | /* | |
456 | * The nohz low res interrupt handler | |
457 | */ | |
458 | static void tick_nohz_handler(struct clock_event_device *dev) | |
459 | { | |
460 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
461 | struct pt_regs *regs = get_irq_regs(); | |
d3ed7824 | 462 | int cpu = smp_processor_id(); |
79bf2bb3 TG |
463 | ktime_t now = ktime_get(); |
464 | ||
465 | dev->next_event.tv64 = KTIME_MAX; | |
466 | ||
d3ed7824 TG |
467 | /* |
468 | * Check if the do_timer duty was dropped. We don't care about | |
469 | * concurrency: This happens only when the cpu in charge went | |
470 | * into a long sleep. If two cpus happen to assign themself to | |
471 | * this duty, then the jiffies update is still serialized by | |
472 | * xtime_lock. | |
473 | */ | |
6441402b | 474 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) |
d3ed7824 TG |
475 | tick_do_timer_cpu = cpu; |
476 | ||
79bf2bb3 | 477 | /* Check, if the jiffies need an update */ |
d3ed7824 TG |
478 | if (tick_do_timer_cpu == cpu) |
479 | tick_do_update_jiffies64(now); | |
79bf2bb3 TG |
480 | |
481 | /* | |
482 | * When we are idle and the tick is stopped, we have to touch | |
483 | * the watchdog as we might not schedule for a really long | |
484 | * time. This happens on complete idle SMP systems while | |
485 | * waiting on the login prompt. We also increment the "start | |
486 | * of idle" jiffy stamp so the idle accounting adjustment we | |
487 | * do when we go busy again does not account too much ticks. | |
488 | */ | |
489 | if (ts->tick_stopped) { | |
490 | touch_softlockup_watchdog(); | |
491 | ts->idle_jiffies++; | |
492 | } | |
493 | ||
494 | update_process_times(user_mode(regs)); | |
495 | profile_tick(CPU_PROFILING); | |
496 | ||
497 | /* Do not restart, when we are in the idle loop */ | |
498 | if (ts->tick_stopped) | |
499 | return; | |
500 | ||
501 | while (tick_nohz_reprogram(ts, now)) { | |
502 | now = ktime_get(); | |
503 | tick_do_update_jiffies64(now); | |
504 | } | |
505 | } | |
506 | ||
507 | /** | |
508 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
509 | */ | |
510 | static void tick_nohz_switch_to_nohz(void) | |
511 | { | |
512 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
513 | ktime_t next; | |
514 | ||
515 | if (!tick_nohz_enabled) | |
516 | return; | |
517 | ||
518 | local_irq_disable(); | |
519 | if (tick_switch_to_oneshot(tick_nohz_handler)) { | |
520 | local_irq_enable(); | |
521 | return; | |
522 | } | |
523 | ||
524 | ts->nohz_mode = NOHZ_MODE_LOWRES; | |
525 | ||
526 | /* | |
527 | * Recycle the hrtimer in ts, so we can share the | |
528 | * hrtimer_forward with the highres code. | |
529 | */ | |
530 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
531 | /* Get the next period */ | |
532 | next = tick_init_jiffy_update(); | |
533 | ||
534 | for (;;) { | |
535 | ts->sched_timer.expires = next; | |
536 | if (!tick_program_event(next, 0)) | |
537 | break; | |
538 | next = ktime_add(next, tick_period); | |
539 | } | |
540 | local_irq_enable(); | |
541 | ||
542 | printk(KERN_INFO "Switched to NOHz mode on CPU #%d\n", | |
543 | smp_processor_id()); | |
544 | } | |
545 | ||
546 | #else | |
547 | ||
548 | static inline void tick_nohz_switch_to_nohz(void) { } | |
549 | ||
550 | #endif /* NO_HZ */ | |
551 | ||
552 | /* | |
553 | * High resolution timer specific code | |
554 | */ | |
555 | #ifdef CONFIG_HIGH_RES_TIMERS | |
556 | /* | |
4c9dc641 | 557 | * We rearm the timer until we get disabled by the idle code. |
79bf2bb3 TG |
558 | * Called with interrupts disabled and timer->base->cpu_base->lock held. |
559 | */ | |
560 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
561 | { | |
562 | struct tick_sched *ts = | |
563 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
564 | struct pt_regs *regs = get_irq_regs(); |
565 | ktime_t now = ktime_get(); | |
d3ed7824 TG |
566 | int cpu = smp_processor_id(); |
567 | ||
568 | #ifdef CONFIG_NO_HZ | |
569 | /* | |
570 | * Check if the do_timer duty was dropped. We don't care about | |
571 | * concurrency: This happens only when the cpu in charge went | |
572 | * into a long sleep. If two cpus happen to assign themself to | |
573 | * this duty, then the jiffies update is still serialized by | |
574 | * xtime_lock. | |
575 | */ | |
6441402b | 576 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) |
d3ed7824 TG |
577 | tick_do_timer_cpu = cpu; |
578 | #endif | |
79bf2bb3 TG |
579 | |
580 | /* Check, if the jiffies need an update */ | |
d3ed7824 TG |
581 | if (tick_do_timer_cpu == cpu) |
582 | tick_do_update_jiffies64(now); | |
79bf2bb3 TG |
583 | |
584 | /* | |
585 | * Do not call, when we are not in irq context and have | |
586 | * no valid regs pointer | |
587 | */ | |
588 | if (regs) { | |
589 | /* | |
590 | * When we are idle and the tick is stopped, we have to touch | |
591 | * the watchdog as we might not schedule for a really long | |
592 | * time. This happens on complete idle SMP systems while | |
593 | * waiting on the login prompt. We also increment the "start of | |
594 | * idle" jiffy stamp so the idle accounting adjustment we do | |
595 | * when we go busy again does not account too much ticks. | |
596 | */ | |
597 | if (ts->tick_stopped) { | |
598 | touch_softlockup_watchdog(); | |
599 | ts->idle_jiffies++; | |
600 | } | |
79bf2bb3 TG |
601 | update_process_times(user_mode(regs)); |
602 | profile_tick(CPU_PROFILING); | |
79bf2bb3 TG |
603 | } |
604 | ||
605 | /* Do not restart, when we are in the idle loop */ | |
606 | if (ts->tick_stopped) | |
607 | return HRTIMER_NORESTART; | |
608 | ||
609 | hrtimer_forward(timer, now, tick_period); | |
610 | ||
611 | return HRTIMER_RESTART; | |
612 | } | |
613 | ||
614 | /** | |
615 | * tick_setup_sched_timer - setup the tick emulation timer | |
616 | */ | |
617 | void tick_setup_sched_timer(void) | |
618 | { | |
619 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
620 | ktime_t now = ktime_get(); | |
3704540b | 621 | u64 offset; |
79bf2bb3 TG |
622 | |
623 | /* | |
624 | * Emulate tick processing via per-CPU hrtimers: | |
625 | */ | |
626 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
627 | ts->sched_timer.function = tick_sched_timer; | |
ccc7dadf | 628 | ts->sched_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU; |
79bf2bb3 | 629 | |
3704540b | 630 | /* Get the next period (per cpu) */ |
79bf2bb3 | 631 | ts->sched_timer.expires = tick_init_jiffy_update(); |
3704540b | 632 | offset = ktime_to_ns(tick_period) >> 1; |
b2d9323d | 633 | do_div(offset, num_possible_cpus()); |
3704540b | 634 | offset *= smp_processor_id(); |
635 | ts->sched_timer.expires = ktime_add_ns(ts->sched_timer.expires, offset); | |
79bf2bb3 TG |
636 | |
637 | for (;;) { | |
638 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
639 | hrtimer_start(&ts->sched_timer, ts->sched_timer.expires, | |
640 | HRTIMER_MODE_ABS); | |
641 | /* Check, if the timer was already in the past */ | |
642 | if (hrtimer_active(&ts->sched_timer)) | |
643 | break; | |
644 | now = ktime_get(); | |
645 | } | |
646 | ||
647 | #ifdef CONFIG_NO_HZ | |
648 | if (tick_nohz_enabled) | |
649 | ts->nohz_mode = NOHZ_MODE_HIGHRES; | |
650 | #endif | |
651 | } | |
3c4fbe5e | 652 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 653 | |
3c4fbe5e | 654 | #if defined CONFIG_NO_HZ || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
655 | void tick_cancel_sched_timer(int cpu) |
656 | { | |
657 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
658 | ||
3c4fbe5e | 659 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
660 | if (ts->sched_timer.base) |
661 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 662 | # endif |
a7901766 | 663 | |
79bf2bb3 TG |
664 | ts->nohz_mode = NOHZ_MODE_INACTIVE; |
665 | } | |
3c4fbe5e | 666 | #endif |
79bf2bb3 TG |
667 | |
668 | /** | |
669 | * Async notification about clocksource changes | |
670 | */ | |
671 | void tick_clock_notify(void) | |
672 | { | |
673 | int cpu; | |
674 | ||
675 | for_each_possible_cpu(cpu) | |
676 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
677 | } | |
678 | ||
679 | /* | |
680 | * Async notification about clock event changes | |
681 | */ | |
682 | void tick_oneshot_notify(void) | |
683 | { | |
684 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
685 | ||
686 | set_bit(0, &ts->check_clocks); | |
687 | } | |
688 | ||
689 | /** | |
690 | * Check, if a change happened, which makes oneshot possible. | |
691 | * | |
692 | * Called cyclic from the hrtimer softirq (driven by the timer | |
693 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
694 | * mode, because high resolution timers are disabled (either compile | |
695 | * or runtime). | |
696 | */ | |
697 | int tick_check_oneshot_change(int allow_nohz) | |
698 | { | |
699 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
700 | ||
701 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
702 | return 0; | |
703 | ||
704 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
705 | return 0; | |
706 | ||
cf4fc6cb | 707 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
708 | return 0; |
709 | ||
710 | if (!allow_nohz) | |
711 | return 1; | |
712 | ||
713 | tick_nohz_switch_to_nohz(); | |
714 | return 0; | |
715 | } |