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> | |
8083e4ad | 22 | #include <linux/module.h> |
79bf2bb3 | 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 | */ | |
eed3b9cf | 136 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 TG |
137 | { |
138 | int cpu = smp_processor_id(); | |
139 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
140 | unsigned long flags; | |
79bf2bb3 | 141 | |
6a7b3dc3 | 142 | cpumask_clear_cpu(cpu, nohz_cpu_mask); |
5df7fa1c | 143 | ts->idle_waketime = now; |
79bf2bb3 TG |
144 | |
145 | local_irq_save(flags); | |
146 | tick_do_update_jiffies64(now); | |
147 | local_irq_restore(flags); | |
02ff3755 IM |
148 | |
149 | touch_softlockup_watchdog(); | |
79bf2bb3 TG |
150 | } |
151 | ||
595aac48 AV |
152 | /* |
153 | * Updates the per cpu time idle statistics counters | |
154 | */ | |
8d63bf94 | 155 | static void |
8c215bd3 | 156 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 157 | { |
eed3b9cf | 158 | ktime_t delta; |
6378ddb5 | 159 | |
595aac48 AV |
160 | if (ts->idle_active) { |
161 | delta = ktime_sub(now, ts->idle_entrytime); | |
162 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c215bd3 | 163 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 164 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
8c7b09f4 | 165 | ts->idle_entrytime = now; |
595aac48 | 166 | } |
8d63bf94 | 167 | |
e0e37c20 | 168 | if (last_update_time) |
8d63bf94 AV |
169 | *last_update_time = ktime_to_us(now); |
170 | ||
595aac48 AV |
171 | } |
172 | ||
173 | static void tick_nohz_stop_idle(int cpu, ktime_t now) | |
174 | { | |
175 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
176 | ||
8c215bd3 | 177 | update_ts_time_stats(cpu, ts, now, NULL); |
eed3b9cf | 178 | ts->idle_active = 0; |
56c7426b | 179 | |
eed3b9cf | 180 | sched_clock_idle_wakeup_event(0); |
6378ddb5 VP |
181 | } |
182 | ||
8c215bd3 | 183 | static ktime_t tick_nohz_start_idle(int cpu, struct tick_sched *ts) |
6378ddb5 | 184 | { |
595aac48 | 185 | ktime_t now; |
6378ddb5 VP |
186 | |
187 | now = ktime_get(); | |
595aac48 | 188 | |
8c215bd3 | 189 | update_ts_time_stats(cpu, ts, now, NULL); |
595aac48 | 190 | |
6378ddb5 VP |
191 | ts->idle_entrytime = now; |
192 | ts->idle_active = 1; | |
56c7426b | 193 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
194 | return now; |
195 | } | |
196 | ||
b1f724c3 AV |
197 | /** |
198 | * get_cpu_idle_time_us - get the total idle time of a cpu | |
199 | * @cpu: CPU number to query | |
200 | * @last_update_time: variable to store update time in | |
201 | * | |
202 | * Return the cummulative idle time (since boot) for a given | |
203 | * CPU, in microseconds. The idle time returned includes | |
204 | * the iowait time (unlike what "top" and co report). | |
205 | * | |
206 | * This time is measured via accounting rather than sampling, | |
207 | * and is as accurate as ktime_get() is. | |
208 | * | |
209 | * This function returns -1 if NOHZ is not enabled. | |
210 | */ | |
6378ddb5 VP |
211 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
212 | { | |
213 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
214 | ||
8083e4ad | 215 | if (!tick_nohz_enabled) |
216 | return -1; | |
217 | ||
8c215bd3 | 218 | update_ts_time_stats(cpu, ts, ktime_get(), last_update_time); |
8083e4ad | 219 | |
6378ddb5 VP |
220 | return ktime_to_us(ts->idle_sleeptime); |
221 | } | |
8083e4ad | 222 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 223 | |
0224cf4c AV |
224 | /* |
225 | * get_cpu_iowait_time_us - get the total iowait time of a cpu | |
226 | * @cpu: CPU number to query | |
227 | * @last_update_time: variable to store update time in | |
228 | * | |
229 | * Return the cummulative iowait time (since boot) for a given | |
230 | * CPU, in microseconds. | |
231 | * | |
232 | * This time is measured via accounting rather than sampling, | |
233 | * and is as accurate as ktime_get() is. | |
234 | * | |
235 | * This function returns -1 if NOHZ is not enabled. | |
236 | */ | |
237 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
238 | { | |
239 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
240 | ||
241 | if (!tick_nohz_enabled) | |
242 | return -1; | |
243 | ||
8c215bd3 | 244 | update_ts_time_stats(cpu, ts, ktime_get(), last_update_time); |
0224cf4c AV |
245 | |
246 | return ktime_to_us(ts->iowait_sleeptime); | |
247 | } | |
248 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
249 | ||
79bf2bb3 TG |
250 | /** |
251 | * tick_nohz_stop_sched_tick - stop the idle tick from the idle task | |
252 | * | |
253 | * When the next event is more than a tick into the future, stop the idle tick | |
254 | * Called either from the idle loop or from irq_exit() when an idle period was | |
255 | * just interrupted by an interrupt which did not cause a reschedule. | |
256 | */ | |
b8f8c3cf | 257 | void tick_nohz_stop_sched_tick(int inidle) |
79bf2bb3 TG |
258 | { |
259 | unsigned long seq, last_jiffies, next_jiffies, delta_jiffies, flags; | |
260 | struct tick_sched *ts; | |
6378ddb5 | 261 | ktime_t last_update, expires, now; |
4f86d3a8 | 262 | struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev; |
98962465 | 263 | u64 time_delta; |
79bf2bb3 TG |
264 | int cpu; |
265 | ||
266 | local_irq_save(flags); | |
267 | ||
268 | cpu = smp_processor_id(); | |
269 | ts = &per_cpu(tick_cpu_sched, cpu); | |
f2e21c96 EN |
270 | |
271 | /* | |
272 | * Call to tick_nohz_start_idle stops the last_update_time from being | |
273 | * updated. Thus, it must not be called in the event we are called from | |
274 | * irq_exit() with the prior state different than idle. | |
275 | */ | |
276 | if (!inidle && !ts->inidle) | |
277 | goto end; | |
278 | ||
fdc6f192 EN |
279 | /* |
280 | * Set ts->inidle unconditionally. Even if the system did not | |
281 | * switch to NOHZ mode the cpu frequency governers rely on the | |
282 | * update of the idle time accounting in tick_nohz_start_idle(). | |
283 | */ | |
284 | ts->inidle = 1; | |
285 | ||
8c215bd3 | 286 | now = tick_nohz_start_idle(cpu, ts); |
79bf2bb3 | 287 | |
5e41d0d6 TG |
288 | /* |
289 | * If this cpu is offline and it is the one which updates | |
290 | * jiffies, then give up the assignment and let it be taken by | |
291 | * the cpu which runs the tick timer next. If we don't drop | |
292 | * this here the jiffies might be stale and do_timer() never | |
293 | * invoked. | |
294 | */ | |
295 | if (unlikely(!cpu_online(cpu))) { | |
296 | if (cpu == tick_do_timer_cpu) | |
6441402b | 297 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; |
5e41d0d6 TG |
298 | } |
299 | ||
79bf2bb3 TG |
300 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) |
301 | goto end; | |
302 | ||
303 | if (need_resched()) | |
304 | goto end; | |
305 | ||
fa116ea3 | 306 | if (unlikely(local_softirq_pending() && cpu_online(cpu))) { |
35282316 TG |
307 | static int ratelimit; |
308 | ||
309 | if (ratelimit < 10) { | |
310 | printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n", | |
529eaccd | 311 | (unsigned int) local_softirq_pending()); |
35282316 TG |
312 | ratelimit++; |
313 | } | |
857f3fd7 | 314 | goto end; |
35282316 | 315 | } |
79bf2bb3 | 316 | |
79bf2bb3 | 317 | ts->idle_calls++; |
79bf2bb3 TG |
318 | /* Read jiffies and the time when jiffies were updated last */ |
319 | do { | |
320 | seq = read_seqbegin(&xtime_lock); | |
321 | last_update = last_jiffies_update; | |
322 | last_jiffies = jiffies; | |
27185016 | 323 | time_delta = timekeeping_max_deferment(); |
79bf2bb3 TG |
324 | } while (read_seqretry(&xtime_lock, seq)); |
325 | ||
3c5d92a0 | 326 | if (rcu_needs_cpu(cpu) || printk_needs_cpu(cpu) || |
396e894d | 327 | arch_needs_cpu(cpu)) { |
3c5d92a0 | 328 | next_jiffies = last_jiffies + 1; |
6ba9b346 | 329 | delta_jiffies = 1; |
3c5d92a0 MS |
330 | } else { |
331 | /* Get the next timer wheel timer */ | |
332 | next_jiffies = get_next_timer_interrupt(last_jiffies); | |
333 | delta_jiffies = next_jiffies - last_jiffies; | |
334 | } | |
79bf2bb3 TG |
335 | /* |
336 | * Do not stop the tick, if we are only one off | |
337 | * or if the cpu is required for rcu | |
338 | */ | |
6ba9b346 | 339 | if (!ts->tick_stopped && delta_jiffies == 1) |
79bf2bb3 TG |
340 | goto out; |
341 | ||
342 | /* Schedule the tick, if we are at least one jiffie off */ | |
343 | if ((long)delta_jiffies >= 1) { | |
344 | ||
00147449 WR |
345 | /* |
346 | * If this cpu is the one which updates jiffies, then | |
347 | * give up the assignment and let it be taken by the | |
348 | * cpu which runs the tick timer next, which might be | |
349 | * this cpu as well. If we don't drop this here the | |
350 | * jiffies might be stale and do_timer() never | |
27185016 TG |
351 | * invoked. Keep track of the fact that it was the one |
352 | * which had the do_timer() duty last. If this cpu is | |
353 | * the one which had the do_timer() duty last, we | |
354 | * limit the sleep time to the timekeeping | |
355 | * max_deferement value which we retrieved | |
356 | * above. Otherwise we can sleep as long as we want. | |
00147449 | 357 | */ |
27185016 | 358 | if (cpu == tick_do_timer_cpu) { |
00147449 | 359 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; |
27185016 TG |
360 | ts->do_timer_last = 1; |
361 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
362 | time_delta = KTIME_MAX; | |
363 | ts->do_timer_last = 0; | |
364 | } else if (!ts->do_timer_last) { | |
365 | time_delta = KTIME_MAX; | |
366 | } | |
367 | ||
00147449 | 368 | /* |
98962465 JH |
369 | * calculate the expiry time for the next timer wheel |
370 | * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals | |
371 | * that there is no timer pending or at least extremely | |
372 | * far into the future (12 days for HZ=1000). In this | |
373 | * case we set the expiry to the end of time. | |
374 | */ | |
375 | if (likely(delta_jiffies < NEXT_TIMER_MAX_DELTA)) { | |
376 | /* | |
377 | * Calculate the time delta for the next timer event. | |
378 | * If the time delta exceeds the maximum time delta | |
379 | * permitted by the current clocksource then adjust | |
380 | * the time delta accordingly to ensure the | |
381 | * clocksource does not wrap. | |
382 | */ | |
383 | time_delta = min_t(u64, time_delta, | |
384 | tick_period.tv64 * delta_jiffies); | |
98962465 | 385 | } |
00147449 | 386 | |
27185016 TG |
387 | if (time_delta < KTIME_MAX) |
388 | expires = ktime_add_ns(last_update, time_delta); | |
389 | else | |
390 | expires.tv64 = KTIME_MAX; | |
00147449 | 391 | |
6ba9b346 | 392 | if (delta_jiffies > 1) |
6a7b3dc3 | 393 | cpumask_set_cpu(cpu, nohz_cpu_mask); |
00147449 WR |
394 | |
395 | /* Skip reprogram of event if its not changed */ | |
396 | if (ts->tick_stopped && ktime_equal(expires, dev->next_event)) | |
397 | goto out; | |
398 | ||
79bf2bb3 TG |
399 | /* |
400 | * nohz_stop_sched_tick can be called several times before | |
401 | * the nohz_restart_sched_tick is called. This happens when | |
402 | * interrupts arrive which do not cause a reschedule. In the | |
403 | * first call we save the current tick time, so we can restart | |
404 | * the scheduler tick in nohz_restart_sched_tick. | |
405 | */ | |
406 | if (!ts->tick_stopped) { | |
83cd4fe2 | 407 | select_nohz_load_balancer(1); |
46cb4b7c | 408 | |
cc584b21 | 409 | ts->idle_tick = hrtimer_get_expires(&ts->sched_timer); |
79bf2bb3 TG |
410 | ts->tick_stopped = 1; |
411 | ts->idle_jiffies = last_jiffies; | |
2232c2d8 | 412 | rcu_enter_nohz(); |
79bf2bb3 | 413 | } |
d3ed7824 | 414 | |
eaad084b TG |
415 | ts->idle_sleeps++; |
416 | ||
98962465 JH |
417 | /* Mark expires */ |
418 | ts->idle_expires = expires; | |
419 | ||
eaad084b | 420 | /* |
98962465 JH |
421 | * If the expiration time == KTIME_MAX, then |
422 | * in this case we simply stop the tick timer. | |
eaad084b | 423 | */ |
98962465 | 424 | if (unlikely(expires.tv64 == KTIME_MAX)) { |
eaad084b TG |
425 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
426 | hrtimer_cancel(&ts->sched_timer); | |
427 | goto out; | |
428 | } | |
429 | ||
79bf2bb3 TG |
430 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
431 | hrtimer_start(&ts->sched_timer, expires, | |
5c333864 | 432 | HRTIMER_MODE_ABS_PINNED); |
79bf2bb3 TG |
433 | /* Check, if the timer was already in the past */ |
434 | if (hrtimer_active(&ts->sched_timer)) | |
435 | goto out; | |
4c9dc641 | 436 | } else if (!tick_program_event(expires, 0)) |
79bf2bb3 TG |
437 | goto out; |
438 | /* | |
439 | * We are past the event already. So we crossed a | |
440 | * jiffie boundary. Update jiffies and raise the | |
441 | * softirq. | |
442 | */ | |
443 | tick_do_update_jiffies64(ktime_get()); | |
6a7b3dc3 | 444 | cpumask_clear_cpu(cpu, nohz_cpu_mask); |
79bf2bb3 TG |
445 | } |
446 | raise_softirq_irqoff(TIMER_SOFTIRQ); | |
447 | out: | |
448 | ts->next_jiffies = next_jiffies; | |
449 | ts->last_jiffies = last_jiffies; | |
4f86d3a8 | 450 | ts->sleep_length = ktime_sub(dev->next_event, now); |
79bf2bb3 TG |
451 | end: |
452 | local_irq_restore(flags); | |
453 | } | |
454 | ||
4f86d3a8 LB |
455 | /** |
456 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
457 | * | |
458 | * Called from power state control code with interrupts disabled | |
459 | */ | |
460 | ktime_t tick_nohz_get_sleep_length(void) | |
461 | { | |
462 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
463 | ||
464 | return ts->sleep_length; | |
465 | } | |
466 | ||
c34bec5a TG |
467 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
468 | { | |
469 | hrtimer_cancel(&ts->sched_timer); | |
268a3dcf | 470 | hrtimer_set_expires(&ts->sched_timer, ts->idle_tick); |
c34bec5a TG |
471 | |
472 | while (1) { | |
473 | /* Forward the time to expire in the future */ | |
474 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
475 | ||
476 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { | |
268a3dcf | 477 | hrtimer_start_expires(&ts->sched_timer, |
5c333864 | 478 | HRTIMER_MODE_ABS_PINNED); |
c34bec5a TG |
479 | /* Check, if the timer was already in the past */ |
480 | if (hrtimer_active(&ts->sched_timer)) | |
481 | break; | |
482 | } else { | |
268a3dcf TG |
483 | if (!tick_program_event( |
484 | hrtimer_get_expires(&ts->sched_timer), 0)) | |
c34bec5a TG |
485 | break; |
486 | } | |
487 | /* Update jiffies and reread time */ | |
488 | tick_do_update_jiffies64(now); | |
489 | now = ktime_get(); | |
490 | } | |
491 | } | |
492 | ||
79bf2bb3 | 493 | /** |
8dce39c2 | 494 | * tick_nohz_restart_sched_tick - restart the idle tick from the idle task |
79bf2bb3 TG |
495 | * |
496 | * Restart the idle tick when the CPU is woken up from idle | |
497 | */ | |
498 | void tick_nohz_restart_sched_tick(void) | |
499 | { | |
500 | int cpu = smp_processor_id(); | |
501 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
79741dd3 | 502 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
79bf2bb3 | 503 | unsigned long ticks; |
79741dd3 | 504 | #endif |
6378ddb5 | 505 | ktime_t now; |
79bf2bb3 | 506 | |
6378ddb5 | 507 | local_irq_disable(); |
eed3b9cf MS |
508 | if (ts->idle_active || (ts->inidle && ts->tick_stopped)) |
509 | now = ktime_get(); | |
510 | ||
511 | if (ts->idle_active) | |
512 | tick_nohz_stop_idle(cpu, now); | |
6378ddb5 | 513 | |
b8f8c3cf TG |
514 | if (!ts->inidle || !ts->tick_stopped) { |
515 | ts->inidle = 0; | |
6378ddb5 | 516 | local_irq_enable(); |
79bf2bb3 | 517 | return; |
6378ddb5 | 518 | } |
79bf2bb3 | 519 | |
b8f8c3cf TG |
520 | ts->inidle = 0; |
521 | ||
2232c2d8 SR |
522 | rcu_exit_nohz(); |
523 | ||
79bf2bb3 | 524 | /* Update jiffies first */ |
46cb4b7c | 525 | select_nohz_load_balancer(0); |
79bf2bb3 | 526 | tick_do_update_jiffies64(now); |
6a7b3dc3 | 527 | cpumask_clear_cpu(cpu, nohz_cpu_mask); |
79bf2bb3 | 528 | |
79741dd3 | 529 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
79bf2bb3 TG |
530 | /* |
531 | * We stopped the tick in idle. Update process times would miss the | |
532 | * time we slept as update_process_times does only a 1 tick | |
533 | * accounting. Enforce that this is accounted to idle ! | |
534 | */ | |
535 | ticks = jiffies - ts->idle_jiffies; | |
536 | /* | |
537 | * We might be one off. Do not randomly account a huge number of ticks! | |
538 | */ | |
79741dd3 MS |
539 | if (ticks && ticks < LONG_MAX) |
540 | account_idle_ticks(ticks); | |
541 | #endif | |
79bf2bb3 | 542 | |
126e01bf | 543 | touch_softlockup_watchdog(); |
79bf2bb3 TG |
544 | /* |
545 | * Cancel the scheduled timer and restore the tick | |
546 | */ | |
547 | ts->tick_stopped = 0; | |
5df7fa1c | 548 | ts->idle_exittime = now; |
79bf2bb3 | 549 | |
c34bec5a | 550 | tick_nohz_restart(ts, now); |
79bf2bb3 | 551 | |
79bf2bb3 TG |
552 | local_irq_enable(); |
553 | } | |
554 | ||
555 | static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now) | |
556 | { | |
557 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
cc584b21 | 558 | return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0); |
79bf2bb3 TG |
559 | } |
560 | ||
561 | /* | |
562 | * The nohz low res interrupt handler | |
563 | */ | |
564 | static void tick_nohz_handler(struct clock_event_device *dev) | |
565 | { | |
566 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
567 | struct pt_regs *regs = get_irq_regs(); | |
d3ed7824 | 568 | int cpu = smp_processor_id(); |
79bf2bb3 TG |
569 | ktime_t now = ktime_get(); |
570 | ||
571 | dev->next_event.tv64 = KTIME_MAX; | |
572 | ||
d3ed7824 TG |
573 | /* |
574 | * Check if the do_timer duty was dropped. We don't care about | |
575 | * concurrency: This happens only when the cpu in charge went | |
576 | * into a long sleep. If two cpus happen to assign themself to | |
577 | * this duty, then the jiffies update is still serialized by | |
578 | * xtime_lock. | |
579 | */ | |
6441402b | 580 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) |
d3ed7824 TG |
581 | tick_do_timer_cpu = cpu; |
582 | ||
79bf2bb3 | 583 | /* Check, if the jiffies need an update */ |
d3ed7824 TG |
584 | if (tick_do_timer_cpu == cpu) |
585 | tick_do_update_jiffies64(now); | |
79bf2bb3 TG |
586 | |
587 | /* | |
588 | * When we are idle and the tick is stopped, we have to touch | |
589 | * the watchdog as we might not schedule for a really long | |
590 | * time. This happens on complete idle SMP systems while | |
591 | * waiting on the login prompt. We also increment the "start | |
592 | * of idle" jiffy stamp so the idle accounting adjustment we | |
593 | * do when we go busy again does not account too much ticks. | |
594 | */ | |
595 | if (ts->tick_stopped) { | |
596 | touch_softlockup_watchdog(); | |
597 | ts->idle_jiffies++; | |
598 | } | |
599 | ||
600 | update_process_times(user_mode(regs)); | |
601 | profile_tick(CPU_PROFILING); | |
602 | ||
79bf2bb3 TG |
603 | while (tick_nohz_reprogram(ts, now)) { |
604 | now = ktime_get(); | |
605 | tick_do_update_jiffies64(now); | |
606 | } | |
607 | } | |
608 | ||
609 | /** | |
610 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
611 | */ | |
612 | static void tick_nohz_switch_to_nohz(void) | |
613 | { | |
614 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
615 | ktime_t next; | |
616 | ||
617 | if (!tick_nohz_enabled) | |
618 | return; | |
619 | ||
620 | local_irq_disable(); | |
621 | if (tick_switch_to_oneshot(tick_nohz_handler)) { | |
622 | local_irq_enable(); | |
623 | return; | |
624 | } | |
625 | ||
626 | ts->nohz_mode = NOHZ_MODE_LOWRES; | |
627 | ||
628 | /* | |
629 | * Recycle the hrtimer in ts, so we can share the | |
630 | * hrtimer_forward with the highres code. | |
631 | */ | |
632 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
633 | /* Get the next period */ | |
634 | next = tick_init_jiffy_update(); | |
635 | ||
636 | for (;;) { | |
cc584b21 | 637 | hrtimer_set_expires(&ts->sched_timer, next); |
79bf2bb3 TG |
638 | if (!tick_program_event(next, 0)) |
639 | break; | |
640 | next = ktime_add(next, tick_period); | |
641 | } | |
642 | local_irq_enable(); | |
643 | ||
2d0640b4 | 644 | printk(KERN_INFO "Switched to NOHz mode on CPU #%d\n", smp_processor_id()); |
79bf2bb3 TG |
645 | } |
646 | ||
fb02fbc1 TG |
647 | /* |
648 | * When NOHZ is enabled and the tick is stopped, we need to kick the | |
649 | * tick timer from irq_enter() so that the jiffies update is kept | |
650 | * alive during long running softirqs. That's ugly as hell, but | |
651 | * correctness is key even if we need to fix the offending softirq in | |
652 | * the first place. | |
653 | * | |
654 | * Note, this is different to tick_nohz_restart. We just kick the | |
655 | * timer and do not touch the other magic bits which need to be done | |
656 | * when idle is left. | |
657 | */ | |
eed3b9cf | 658 | static void tick_nohz_kick_tick(int cpu, ktime_t now) |
fb02fbc1 | 659 | { |
ae99286b TG |
660 | #if 0 |
661 | /* Switch back to 2.6.27 behaviour */ | |
662 | ||
fb02fbc1 | 663 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); |
eed3b9cf | 664 | ktime_t delta; |
fb02fbc1 | 665 | |
c4bd822e TG |
666 | /* |
667 | * Do not touch the tick device, when the next expiry is either | |
668 | * already reached or less/equal than the tick period. | |
669 | */ | |
268a3dcf | 670 | delta = ktime_sub(hrtimer_get_expires(&ts->sched_timer), now); |
c4bd822e TG |
671 | if (delta.tv64 <= tick_period.tv64) |
672 | return; | |
673 | ||
674 | tick_nohz_restart(ts, now); | |
ae99286b | 675 | #endif |
fb02fbc1 TG |
676 | } |
677 | ||
eed3b9cf MS |
678 | static inline void tick_check_nohz(int cpu) |
679 | { | |
680 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
681 | ktime_t now; | |
682 | ||
683 | if (!ts->idle_active && !ts->tick_stopped) | |
684 | return; | |
685 | now = ktime_get(); | |
686 | if (ts->idle_active) | |
687 | tick_nohz_stop_idle(cpu, now); | |
688 | if (ts->tick_stopped) { | |
689 | tick_nohz_update_jiffies(now); | |
690 | tick_nohz_kick_tick(cpu, now); | |
691 | } | |
692 | } | |
693 | ||
79bf2bb3 TG |
694 | #else |
695 | ||
696 | static inline void tick_nohz_switch_to_nohz(void) { } | |
eed3b9cf | 697 | static inline void tick_check_nohz(int cpu) { } |
79bf2bb3 TG |
698 | |
699 | #endif /* NO_HZ */ | |
700 | ||
719254fa TG |
701 | /* |
702 | * Called from irq_enter to notify about the possible interruption of idle() | |
703 | */ | |
704 | void tick_check_idle(int cpu) | |
705 | { | |
fb02fbc1 | 706 | tick_check_oneshot_broadcast(cpu); |
eed3b9cf | 707 | tick_check_nohz(cpu); |
719254fa TG |
708 | } |
709 | ||
79bf2bb3 TG |
710 | /* |
711 | * High resolution timer specific code | |
712 | */ | |
713 | #ifdef CONFIG_HIGH_RES_TIMERS | |
714 | /* | |
4c9dc641 | 715 | * We rearm the timer until we get disabled by the idle code. |
79bf2bb3 TG |
716 | * Called with interrupts disabled and timer->base->cpu_base->lock held. |
717 | */ | |
718 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
719 | { | |
720 | struct tick_sched *ts = | |
721 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
722 | struct pt_regs *regs = get_irq_regs(); |
723 | ktime_t now = ktime_get(); | |
d3ed7824 TG |
724 | int cpu = smp_processor_id(); |
725 | ||
726 | #ifdef CONFIG_NO_HZ | |
727 | /* | |
728 | * Check if the do_timer duty was dropped. We don't care about | |
729 | * concurrency: This happens only when the cpu in charge went | |
730 | * into a long sleep. If two cpus happen to assign themself to | |
731 | * this duty, then the jiffies update is still serialized by | |
732 | * xtime_lock. | |
733 | */ | |
6441402b | 734 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) |
d3ed7824 TG |
735 | tick_do_timer_cpu = cpu; |
736 | #endif | |
79bf2bb3 TG |
737 | |
738 | /* Check, if the jiffies need an update */ | |
d3ed7824 TG |
739 | if (tick_do_timer_cpu == cpu) |
740 | tick_do_update_jiffies64(now); | |
79bf2bb3 TG |
741 | |
742 | /* | |
743 | * Do not call, when we are not in irq context and have | |
744 | * no valid regs pointer | |
745 | */ | |
746 | if (regs) { | |
747 | /* | |
748 | * When we are idle and the tick is stopped, we have to touch | |
749 | * the watchdog as we might not schedule for a really long | |
750 | * time. This happens on complete idle SMP systems while | |
751 | * waiting on the login prompt. We also increment the "start of | |
752 | * idle" jiffy stamp so the idle accounting adjustment we do | |
753 | * when we go busy again does not account too much ticks. | |
754 | */ | |
755 | if (ts->tick_stopped) { | |
756 | touch_softlockup_watchdog(); | |
757 | ts->idle_jiffies++; | |
758 | } | |
79bf2bb3 TG |
759 | update_process_times(user_mode(regs)); |
760 | profile_tick(CPU_PROFILING); | |
79bf2bb3 TG |
761 | } |
762 | ||
79bf2bb3 TG |
763 | hrtimer_forward(timer, now, tick_period); |
764 | ||
765 | return HRTIMER_RESTART; | |
766 | } | |
767 | ||
768 | /** | |
769 | * tick_setup_sched_timer - setup the tick emulation timer | |
770 | */ | |
771 | void tick_setup_sched_timer(void) | |
772 | { | |
773 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
774 | ktime_t now = ktime_get(); | |
775 | ||
776 | /* | |
777 | * Emulate tick processing via per-CPU hrtimers: | |
778 | */ | |
779 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
780 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 781 | |
3704540b | 782 | /* Get the next period (per cpu) */ |
cc584b21 | 783 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 TG |
784 | |
785 | for (;;) { | |
786 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
5c333864 AB |
787 | hrtimer_start_expires(&ts->sched_timer, |
788 | HRTIMER_MODE_ABS_PINNED); | |
79bf2bb3 TG |
789 | /* Check, if the timer was already in the past */ |
790 | if (hrtimer_active(&ts->sched_timer)) | |
791 | break; | |
792 | now = ktime_get(); | |
793 | } | |
794 | ||
795 | #ifdef CONFIG_NO_HZ | |
2d0640b4 | 796 | if (tick_nohz_enabled) { |
79bf2bb3 | 797 | ts->nohz_mode = NOHZ_MODE_HIGHRES; |
2d0640b4 SB |
798 | printk(KERN_INFO "Switched to NOHz mode on CPU #%d\n", smp_processor_id()); |
799 | } | |
79bf2bb3 TG |
800 | #endif |
801 | } | |
3c4fbe5e | 802 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 803 | |
3c4fbe5e | 804 | #if defined CONFIG_NO_HZ || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
805 | void tick_cancel_sched_timer(int cpu) |
806 | { | |
807 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
808 | ||
3c4fbe5e | 809 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
810 | if (ts->sched_timer.base) |
811 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 812 | # endif |
a7901766 | 813 | |
79bf2bb3 TG |
814 | ts->nohz_mode = NOHZ_MODE_INACTIVE; |
815 | } | |
3c4fbe5e | 816 | #endif |
79bf2bb3 TG |
817 | |
818 | /** | |
819 | * Async notification about clocksource changes | |
820 | */ | |
821 | void tick_clock_notify(void) | |
822 | { | |
823 | int cpu; | |
824 | ||
825 | for_each_possible_cpu(cpu) | |
826 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
827 | } | |
828 | ||
829 | /* | |
830 | * Async notification about clock event changes | |
831 | */ | |
832 | void tick_oneshot_notify(void) | |
833 | { | |
834 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
835 | ||
836 | set_bit(0, &ts->check_clocks); | |
837 | } | |
838 | ||
839 | /** | |
840 | * Check, if a change happened, which makes oneshot possible. | |
841 | * | |
842 | * Called cyclic from the hrtimer softirq (driven by the timer | |
843 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
844 | * mode, because high resolution timers are disabled (either compile | |
845 | * or runtime). | |
846 | */ | |
847 | int tick_check_oneshot_change(int allow_nohz) | |
848 | { | |
849 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
850 | ||
851 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
852 | return 0; | |
853 | ||
854 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
855 | return 0; | |
856 | ||
cf4fc6cb | 857 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
858 | return 0; |
859 | ||
860 | if (!allow_nohz) | |
861 | return 1; | |
862 | ||
863 | tick_nohz_switch_to_nohz(); | |
864 | return 0; | |
865 | } |