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> |
00b42959 | 23 | #include <linux/irq_work.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 | */ | |
33a5f626 | 32 | DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); |
79bf2bb3 TG |
33 | |
34 | /* | |
d6ad4187 | 35 | * The time, when the last jiffy update happened. Protected by jiffies_lock. |
79bf2bb3 TG |
36 | */ |
37 | static ktime_t last_jiffies_update; | |
38 | ||
289f480a IM |
39 | struct 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 | */ | |
47 | static void tick_do_update_jiffies64(ktime_t now) | |
48 | { | |
49 | unsigned long ticks = 0; | |
50 | ktime_t delta; | |
51 | ||
7a14ce1d | 52 | /* |
d6ad4187 | 53 | * Do a quick check without holding jiffies_lock: |
7a14ce1d IM |
54 | */ |
55 | delta = ktime_sub(now, last_jiffies_update); | |
56 | if (delta.tv64 < tick_period.tv64) | |
57 | return; | |
58 | ||
d6ad4187 JS |
59 | /* Reevalute with jiffies_lock held */ |
60 | write_seqlock(&jiffies_lock); | |
79bf2bb3 TG |
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 | 82 | } |
d6ad4187 | 83 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
84 | } |
85 | ||
86 | /* | |
87 | * Initialize and return retrieve the jiffies update. | |
88 | */ | |
89 | static ktime_t tick_init_jiffy_update(void) | |
90 | { | |
91 | ktime_t period; | |
92 | ||
d6ad4187 | 93 | write_seqlock(&jiffies_lock); |
79bf2bb3 TG |
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; | |
d6ad4187 | 98 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
99 | return period; |
100 | } | |
101 | ||
5bb96226 FW |
102 | |
103 | static void tick_sched_do_timer(ktime_t now) | |
104 | { | |
105 | int cpu = smp_processor_id(); | |
106 | ||
3451d024 | 107 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
108 | /* |
109 | * Check if the do_timer duty was dropped. We don't care about | |
110 | * concurrency: This happens only when the cpu in charge went | |
111 | * into a long sleep. If two cpus happen to assign themself to | |
112 | * this duty, then the jiffies update is still serialized by | |
9c3f9e28 | 113 | * jiffies_lock. |
5bb96226 | 114 | */ |
a382bf93 FW |
115 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE) |
116 | && !tick_nohz_extended_cpu(cpu)) | |
5bb96226 FW |
117 | tick_do_timer_cpu = cpu; |
118 | #endif | |
119 | ||
120 | /* Check, if the jiffies need an update */ | |
121 | if (tick_do_timer_cpu == cpu) | |
122 | tick_do_update_jiffies64(now); | |
123 | } | |
124 | ||
9e8f559b FW |
125 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
126 | { | |
3451d024 | 127 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
128 | /* |
129 | * When we are idle and the tick is stopped, we have to touch | |
130 | * the watchdog as we might not schedule for a really long | |
131 | * time. This happens on complete idle SMP systems while | |
132 | * waiting on the login prompt. We also increment the "start of | |
133 | * idle" jiffy stamp so the idle accounting adjustment we do | |
134 | * when we go busy again does not account too much ticks. | |
135 | */ | |
136 | if (ts->tick_stopped) { | |
137 | touch_softlockup_watchdog(); | |
138 | if (is_idle_task(current)) | |
139 | ts->idle_jiffies++; | |
140 | } | |
94a57140 | 141 | #endif |
9e8f559b FW |
142 | update_process_times(user_mode(regs)); |
143 | profile_tick(CPU_PROFILING); | |
144 | } | |
145 | ||
a831881b FW |
146 | #ifdef CONFIG_NO_HZ_EXTENDED |
147 | static cpumask_var_t nohz_extended_mask; | |
148 | bool have_nohz_extended_mask; | |
149 | ||
150 | int tick_nohz_extended_cpu(int cpu) | |
151 | { | |
152 | if (!have_nohz_extended_mask) | |
153 | return 0; | |
154 | ||
155 | return cpumask_test_cpu(cpu, nohz_extended_mask); | |
156 | } | |
157 | ||
158 | /* Parse the boot-time nohz CPU list from the kernel parameters. */ | |
159 | static int __init tick_nohz_extended_setup(char *str) | |
160 | { | |
161 | alloc_bootmem_cpumask_var(&nohz_extended_mask); | |
162 | if (cpulist_parse(str, nohz_extended_mask) < 0) | |
163 | pr_warning("NOHZ: Incorrect nohz_extended cpumask\n"); | |
164 | else | |
165 | have_nohz_extended_mask = true; | |
166 | return 1; | |
167 | } | |
168 | __setup("nohz_extended=", tick_nohz_extended_setup); | |
169 | ||
a382bf93 FW |
170 | static int __cpuinit tick_nohz_cpu_down_callback(struct notifier_block *nfb, |
171 | unsigned long action, | |
172 | void *hcpu) | |
173 | { | |
174 | unsigned int cpu = (unsigned long)hcpu; | |
175 | ||
176 | switch (action & ~CPU_TASKS_FROZEN) { | |
177 | case CPU_DOWN_PREPARE: | |
178 | /* | |
179 | * If we handle the timekeeping duty for full dynticks CPUs, | |
180 | * we can't safely shutdown that CPU. | |
181 | */ | |
182 | if (have_nohz_extended_mask && tick_do_timer_cpu == cpu) | |
183 | return -EINVAL; | |
184 | break; | |
185 | } | |
186 | return NOTIFY_OK; | |
187 | } | |
188 | ||
1034fc2f FW |
189 | /* |
190 | * Worst case string length in chunks of CPU range seems 2 steps | |
191 | * separations: 0,2,4,6,... | |
192 | * This is NR_CPUS + sizeof('\0') | |
193 | */ | |
194 | static char __initdata nohz_ext_buf[NR_CPUS + 1]; | |
195 | ||
a831881b FW |
196 | static int __init init_tick_nohz_extended(void) |
197 | { | |
198 | cpumask_var_t online_nohz; | |
199 | int cpu; | |
200 | ||
201 | if (!have_nohz_extended_mask) | |
202 | return 0; | |
203 | ||
a382bf93 FW |
204 | cpu_notifier(tick_nohz_cpu_down_callback, 0); |
205 | ||
a831881b FW |
206 | if (!zalloc_cpumask_var(&online_nohz, GFP_KERNEL)) { |
207 | pr_warning("NO_HZ: Not enough memory to check extended nohz mask\n"); | |
208 | return -ENOMEM; | |
209 | } | |
210 | ||
211 | /* | |
212 | * CPUs can probably not be concurrently offlined on initcall time. | |
213 | * But we are paranoid, aren't we? | |
214 | */ | |
215 | get_online_cpus(); | |
216 | ||
217 | /* Ensure we keep a CPU outside the dynticks range for timekeeping */ | |
218 | cpumask_and(online_nohz, cpu_online_mask, nohz_extended_mask); | |
219 | if (cpumask_equal(online_nohz, cpu_online_mask)) { | |
a831881b FW |
220 | pr_warning("NO_HZ: Must keep at least one online CPU " |
221 | "out of nohz_extended range\n"); | |
a382bf93 FW |
222 | /* |
223 | * We know the current CPU doesn't have its tick stopped. | |
224 | * Let's use it for the timekeeping duty. | |
225 | */ | |
226 | preempt_disable(); | |
227 | cpu = smp_processor_id(); | |
a831881b FW |
228 | pr_warning("NO_HZ: Clearing %d from nohz_extended range\n", cpu); |
229 | cpumask_clear_cpu(cpu, nohz_extended_mask); | |
a382bf93 | 230 | preempt_enable(); |
a831881b FW |
231 | } |
232 | put_online_cpus(); | |
233 | free_cpumask_var(online_nohz); | |
234 | ||
1034fc2f FW |
235 | cpulist_scnprintf(nohz_ext_buf, sizeof(nohz_ext_buf), nohz_extended_mask); |
236 | pr_info("NO_HZ: Full dynticks CPUs: %s.\n", nohz_ext_buf); | |
237 | ||
a831881b FW |
238 | return 0; |
239 | } | |
240 | core_initcall(init_tick_nohz_extended); | |
241 | #else | |
242 | #define have_nohz_extended_mask (0) | |
243 | #endif | |
244 | ||
79bf2bb3 TG |
245 | /* |
246 | * NOHZ - aka dynamic tick functionality | |
247 | */ | |
3451d024 | 248 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
249 | /* |
250 | * NO HZ enabled ? | |
251 | */ | |
9d2ad243 | 252 | int tick_nohz_enabled __read_mostly = 1; |
79bf2bb3 TG |
253 | |
254 | /* | |
255 | * Enable / Disable tickless mode | |
256 | */ | |
257 | static int __init setup_tick_nohz(char *str) | |
258 | { | |
259 | if (!strcmp(str, "off")) | |
260 | tick_nohz_enabled = 0; | |
261 | else if (!strcmp(str, "on")) | |
262 | tick_nohz_enabled = 1; | |
263 | else | |
264 | return 0; | |
265 | return 1; | |
266 | } | |
267 | ||
268 | __setup("nohz=", setup_tick_nohz); | |
269 | ||
270 | /** | |
271 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
272 | * | |
273 | * Called from interrupt entry when the CPU was idle | |
274 | * | |
275 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
276 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
277 | * value. We do this unconditionally on any cpu, as we don't know whether the | |
278 | * cpu, which has the update task assigned is in a long sleep. | |
279 | */ | |
eed3b9cf | 280 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 TG |
281 | { |
282 | int cpu = smp_processor_id(); | |
283 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
284 | unsigned long flags; | |
79bf2bb3 | 285 | |
5df7fa1c | 286 | ts->idle_waketime = now; |
79bf2bb3 TG |
287 | |
288 | local_irq_save(flags); | |
289 | tick_do_update_jiffies64(now); | |
290 | local_irq_restore(flags); | |
02ff3755 IM |
291 | |
292 | touch_softlockup_watchdog(); | |
79bf2bb3 TG |
293 | } |
294 | ||
595aac48 AV |
295 | /* |
296 | * Updates the per cpu time idle statistics counters | |
297 | */ | |
8d63bf94 | 298 | static void |
8c215bd3 | 299 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 300 | { |
eed3b9cf | 301 | ktime_t delta; |
6378ddb5 | 302 | |
595aac48 AV |
303 | if (ts->idle_active) { |
304 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 305 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 306 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
307 | else |
308 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 309 | ts->idle_entrytime = now; |
595aac48 | 310 | } |
8d63bf94 | 311 | |
e0e37c20 | 312 | if (last_update_time) |
8d63bf94 AV |
313 | *last_update_time = ktime_to_us(now); |
314 | ||
595aac48 AV |
315 | } |
316 | ||
317 | static void tick_nohz_stop_idle(int cpu, ktime_t now) | |
318 | { | |
319 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
320 | ||
8c215bd3 | 321 | update_ts_time_stats(cpu, ts, now, NULL); |
eed3b9cf | 322 | ts->idle_active = 0; |
56c7426b | 323 | |
eed3b9cf | 324 | sched_clock_idle_wakeup_event(0); |
6378ddb5 VP |
325 | } |
326 | ||
8c215bd3 | 327 | static ktime_t tick_nohz_start_idle(int cpu, struct tick_sched *ts) |
6378ddb5 | 328 | { |
430ee881 | 329 | ktime_t now = ktime_get(); |
595aac48 | 330 | |
6378ddb5 VP |
331 | ts->idle_entrytime = now; |
332 | ts->idle_active = 1; | |
56c7426b | 333 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
334 | return now; |
335 | } | |
336 | ||
b1f724c3 AV |
337 | /** |
338 | * get_cpu_idle_time_us - get the total idle time of a cpu | |
339 | * @cpu: CPU number to query | |
09a1d34f MH |
340 | * @last_update_time: variable to store update time in. Do not update |
341 | * counters if NULL. | |
b1f724c3 AV |
342 | * |
343 | * Return the cummulative idle time (since boot) for a given | |
6beea0cd | 344 | * CPU, in microseconds. |
b1f724c3 AV |
345 | * |
346 | * This time is measured via accounting rather than sampling, | |
347 | * and is as accurate as ktime_get() is. | |
348 | * | |
349 | * This function returns -1 if NOHZ is not enabled. | |
350 | */ | |
6378ddb5 VP |
351 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
352 | { | |
353 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 354 | ktime_t now, idle; |
6378ddb5 | 355 | |
8083e4ad | 356 | if (!tick_nohz_enabled) |
357 | return -1; | |
358 | ||
09a1d34f MH |
359 | now = ktime_get(); |
360 | if (last_update_time) { | |
361 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
362 | idle = ts->idle_sleeptime; | |
363 | } else { | |
364 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
365 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
366 | ||
367 | idle = ktime_add(ts->idle_sleeptime, delta); | |
368 | } else { | |
369 | idle = ts->idle_sleeptime; | |
370 | } | |
371 | } | |
372 | ||
373 | return ktime_to_us(idle); | |
8083e4ad | 374 | |
6378ddb5 | 375 | } |
8083e4ad | 376 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 377 | |
6beea0cd | 378 | /** |
0224cf4c AV |
379 | * get_cpu_iowait_time_us - get the total iowait time of a cpu |
380 | * @cpu: CPU number to query | |
09a1d34f MH |
381 | * @last_update_time: variable to store update time in. Do not update |
382 | * counters if NULL. | |
0224cf4c AV |
383 | * |
384 | * Return the cummulative iowait time (since boot) for a given | |
385 | * CPU, in microseconds. | |
386 | * | |
387 | * This time is measured via accounting rather than sampling, | |
388 | * and is as accurate as ktime_get() is. | |
389 | * | |
390 | * This function returns -1 if NOHZ is not enabled. | |
391 | */ | |
392 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
393 | { | |
394 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 395 | ktime_t now, iowait; |
0224cf4c AV |
396 | |
397 | if (!tick_nohz_enabled) | |
398 | return -1; | |
399 | ||
09a1d34f MH |
400 | now = ktime_get(); |
401 | if (last_update_time) { | |
402 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
403 | iowait = ts->iowait_sleeptime; | |
404 | } else { | |
405 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
406 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 407 | |
09a1d34f MH |
408 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
409 | } else { | |
410 | iowait = ts->iowait_sleeptime; | |
411 | } | |
412 | } | |
0224cf4c | 413 | |
09a1d34f | 414 | return ktime_to_us(iowait); |
0224cf4c AV |
415 | } |
416 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
417 | ||
84bf1bcc FW |
418 | static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts, |
419 | ktime_t now, int cpu) | |
79bf2bb3 | 420 | { |
280f0677 | 421 | unsigned long seq, last_jiffies, next_jiffies, delta_jiffies; |
84bf1bcc | 422 | ktime_t last_update, expires, ret = { .tv64 = 0 }; |
aa9b1630 | 423 | unsigned long rcu_delta_jiffies; |
4f86d3a8 | 424 | struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev; |
98962465 | 425 | u64 time_delta; |
79bf2bb3 | 426 | |
79bf2bb3 TG |
427 | /* Read jiffies and the time when jiffies were updated last */ |
428 | do { | |
d6ad4187 | 429 | seq = read_seqbegin(&jiffies_lock); |
79bf2bb3 TG |
430 | last_update = last_jiffies_update; |
431 | last_jiffies = jiffies; | |
27185016 | 432 | time_delta = timekeeping_max_deferment(); |
d6ad4187 | 433 | } while (read_seqretry(&jiffies_lock, seq)); |
79bf2bb3 | 434 | |
74876a98 | 435 | if (rcu_needs_cpu(cpu, &rcu_delta_jiffies) || |
00b42959 | 436 | arch_needs_cpu(cpu) || irq_work_needs_cpu()) { |
3c5d92a0 | 437 | next_jiffies = last_jiffies + 1; |
6ba9b346 | 438 | delta_jiffies = 1; |
3c5d92a0 MS |
439 | } else { |
440 | /* Get the next timer wheel timer */ | |
441 | next_jiffies = get_next_timer_interrupt(last_jiffies); | |
442 | delta_jiffies = next_jiffies - last_jiffies; | |
aa9b1630 PM |
443 | if (rcu_delta_jiffies < delta_jiffies) { |
444 | next_jiffies = last_jiffies + rcu_delta_jiffies; | |
445 | delta_jiffies = rcu_delta_jiffies; | |
446 | } | |
3c5d92a0 | 447 | } |
79bf2bb3 TG |
448 | /* |
449 | * Do not stop the tick, if we are only one off | |
450 | * or if the cpu is required for rcu | |
451 | */ | |
6ba9b346 | 452 | if (!ts->tick_stopped && delta_jiffies == 1) |
79bf2bb3 TG |
453 | goto out; |
454 | ||
455 | /* Schedule the tick, if we are at least one jiffie off */ | |
456 | if ((long)delta_jiffies >= 1) { | |
457 | ||
00147449 WR |
458 | /* |
459 | * If this cpu is the one which updates jiffies, then | |
460 | * give up the assignment and let it be taken by the | |
461 | * cpu which runs the tick timer next, which might be | |
462 | * this cpu as well. If we don't drop this here the | |
463 | * jiffies might be stale and do_timer() never | |
27185016 TG |
464 | * invoked. Keep track of the fact that it was the one |
465 | * which had the do_timer() duty last. If this cpu is | |
466 | * the one which had the do_timer() duty last, we | |
467 | * limit the sleep time to the timekeeping | |
468 | * max_deferement value which we retrieved | |
469 | * above. Otherwise we can sleep as long as we want. | |
00147449 | 470 | */ |
27185016 | 471 | if (cpu == tick_do_timer_cpu) { |
00147449 | 472 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; |
27185016 TG |
473 | ts->do_timer_last = 1; |
474 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
475 | time_delta = KTIME_MAX; | |
476 | ts->do_timer_last = 0; | |
477 | } else if (!ts->do_timer_last) { | |
478 | time_delta = KTIME_MAX; | |
479 | } | |
480 | ||
00147449 | 481 | /* |
98962465 JH |
482 | * calculate the expiry time for the next timer wheel |
483 | * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals | |
484 | * that there is no timer pending or at least extremely | |
485 | * far into the future (12 days for HZ=1000). In this | |
486 | * case we set the expiry to the end of time. | |
487 | */ | |
488 | if (likely(delta_jiffies < NEXT_TIMER_MAX_DELTA)) { | |
489 | /* | |
490 | * Calculate the time delta for the next timer event. | |
491 | * If the time delta exceeds the maximum time delta | |
492 | * permitted by the current clocksource then adjust | |
493 | * the time delta accordingly to ensure the | |
494 | * clocksource does not wrap. | |
495 | */ | |
496 | time_delta = min_t(u64, time_delta, | |
497 | tick_period.tv64 * delta_jiffies); | |
98962465 | 498 | } |
00147449 | 499 | |
27185016 TG |
500 | if (time_delta < KTIME_MAX) |
501 | expires = ktime_add_ns(last_update, time_delta); | |
502 | else | |
503 | expires.tv64 = KTIME_MAX; | |
00147449 | 504 | |
00147449 WR |
505 | /* Skip reprogram of event if its not changed */ |
506 | if (ts->tick_stopped && ktime_equal(expires, dev->next_event)) | |
507 | goto out; | |
508 | ||
84bf1bcc FW |
509 | ret = expires; |
510 | ||
79bf2bb3 TG |
511 | /* |
512 | * nohz_stop_sched_tick can be called several times before | |
513 | * the nohz_restart_sched_tick is called. This happens when | |
514 | * interrupts arrive which do not cause a reschedule. In the | |
515 | * first call we save the current tick time, so we can restart | |
516 | * the scheduler tick in nohz_restart_sched_tick. | |
517 | */ | |
518 | if (!ts->tick_stopped) { | |
c1cc017c | 519 | nohz_balance_enter_idle(cpu); |
5167e8d5 | 520 | calc_load_enter_idle(); |
46cb4b7c | 521 | |
f5d411c9 | 522 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
79bf2bb3 | 523 | ts->tick_stopped = 1; |
79bf2bb3 | 524 | } |
d3ed7824 | 525 | |
eaad084b | 526 | /* |
98962465 JH |
527 | * If the expiration time == KTIME_MAX, then |
528 | * in this case we simply stop the tick timer. | |
eaad084b | 529 | */ |
98962465 | 530 | if (unlikely(expires.tv64 == KTIME_MAX)) { |
eaad084b TG |
531 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
532 | hrtimer_cancel(&ts->sched_timer); | |
533 | goto out; | |
534 | } | |
535 | ||
79bf2bb3 TG |
536 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
537 | hrtimer_start(&ts->sched_timer, expires, | |
5c333864 | 538 | HRTIMER_MODE_ABS_PINNED); |
79bf2bb3 TG |
539 | /* Check, if the timer was already in the past */ |
540 | if (hrtimer_active(&ts->sched_timer)) | |
541 | goto out; | |
4c9dc641 | 542 | } else if (!tick_program_event(expires, 0)) |
79bf2bb3 TG |
543 | goto out; |
544 | /* | |
545 | * We are past the event already. So we crossed a | |
546 | * jiffie boundary. Update jiffies and raise the | |
547 | * softirq. | |
548 | */ | |
549 | tick_do_update_jiffies64(ktime_get()); | |
79bf2bb3 TG |
550 | } |
551 | raise_softirq_irqoff(TIMER_SOFTIRQ); | |
552 | out: | |
553 | ts->next_jiffies = next_jiffies; | |
554 | ts->last_jiffies = last_jiffies; | |
4f86d3a8 | 555 | ts->sleep_length = ktime_sub(dev->next_event, now); |
84bf1bcc FW |
556 | |
557 | return ret; | |
280f0677 FW |
558 | } |
559 | ||
5b39939a FW |
560 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
561 | { | |
562 | /* | |
563 | * If this cpu is offline and it is the one which updates | |
564 | * jiffies, then give up the assignment and let it be taken by | |
565 | * the cpu which runs the tick timer next. If we don't drop | |
566 | * this here the jiffies might be stale and do_timer() never | |
567 | * invoked. | |
568 | */ | |
569 | if (unlikely(!cpu_online(cpu))) { | |
570 | if (cpu == tick_do_timer_cpu) | |
571 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
572 | } | |
573 | ||
574 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) | |
575 | return false; | |
576 | ||
577 | if (need_resched()) | |
578 | return false; | |
579 | ||
580 | if (unlikely(local_softirq_pending() && cpu_online(cpu))) { | |
581 | static int ratelimit; | |
582 | ||
803b0eba PM |
583 | if (ratelimit < 10 && |
584 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
5b39939a FW |
585 | printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n", |
586 | (unsigned int) local_softirq_pending()); | |
587 | ratelimit++; | |
588 | } | |
589 | return false; | |
590 | } | |
591 | ||
a382bf93 FW |
592 | if (have_nohz_extended_mask) { |
593 | /* | |
594 | * Keep the tick alive to guarantee timekeeping progression | |
595 | * if there are full dynticks CPUs around | |
596 | */ | |
597 | if (tick_do_timer_cpu == cpu) | |
598 | return false; | |
599 | /* | |
600 | * Boot safety: make sure the timekeeping duty has been | |
601 | * assigned before entering dyntick-idle mode, | |
602 | */ | |
603 | if (tick_do_timer_cpu == TICK_DO_TIMER_NONE) | |
604 | return false; | |
605 | } | |
606 | ||
5b39939a FW |
607 | return true; |
608 | } | |
609 | ||
19f5f736 FW |
610 | static void __tick_nohz_idle_enter(struct tick_sched *ts) |
611 | { | |
84bf1bcc | 612 | ktime_t now, expires; |
5b39939a | 613 | int cpu = smp_processor_id(); |
19f5f736 | 614 | |
5b39939a | 615 | now = tick_nohz_start_idle(cpu, ts); |
2ac0d98f | 616 | |
5b39939a FW |
617 | if (can_stop_idle_tick(cpu, ts)) { |
618 | int was_stopped = ts->tick_stopped; | |
619 | ||
620 | ts->idle_calls++; | |
84bf1bcc FW |
621 | |
622 | expires = tick_nohz_stop_sched_tick(ts, now, cpu); | |
623 | if (expires.tv64 > 0LL) { | |
624 | ts->idle_sleeps++; | |
625 | ts->idle_expires = expires; | |
626 | } | |
5b39939a FW |
627 | |
628 | if (!was_stopped && ts->tick_stopped) | |
629 | ts->idle_jiffies = ts->last_jiffies; | |
630 | } | |
280f0677 FW |
631 | } |
632 | ||
633 | /** | |
634 | * tick_nohz_idle_enter - stop the idle tick from the idle task | |
635 | * | |
636 | * When the next event is more than a tick into the future, stop the idle tick | |
637 | * Called when we start the idle loop. | |
2bbb6817 | 638 | * |
1268fbc7 | 639 | * The arch is responsible of calling: |
2bbb6817 FW |
640 | * |
641 | * - rcu_idle_enter() after its last use of RCU before the CPU is put | |
642 | * to sleep. | |
643 | * - rcu_idle_exit() before the first use of RCU after the CPU is woken up. | |
280f0677 | 644 | */ |
1268fbc7 | 645 | void tick_nohz_idle_enter(void) |
280f0677 FW |
646 | { |
647 | struct tick_sched *ts; | |
648 | ||
1268fbc7 FW |
649 | WARN_ON_ONCE(irqs_disabled()); |
650 | ||
0db49b72 LT |
651 | /* |
652 | * Update the idle state in the scheduler domain hierarchy | |
653 | * when tick_nohz_stop_sched_tick() is called from the idle loop. | |
654 | * State will be updated to busy during the first busy tick after | |
655 | * exiting idle. | |
656 | */ | |
657 | set_cpu_sd_state_idle(); | |
658 | ||
1268fbc7 FW |
659 | local_irq_disable(); |
660 | ||
280f0677 FW |
661 | ts = &__get_cpu_var(tick_cpu_sched); |
662 | /* | |
663 | * set ts->inidle unconditionally. even if the system did not | |
664 | * switch to nohz mode the cpu frequency governers rely on the | |
665 | * update of the idle time accounting in tick_nohz_start_idle(). | |
666 | */ | |
667 | ts->inidle = 1; | |
19f5f736 | 668 | __tick_nohz_idle_enter(ts); |
1268fbc7 FW |
669 | |
670 | local_irq_enable(); | |
280f0677 | 671 | } |
4dbd2771 | 672 | EXPORT_SYMBOL_GPL(tick_nohz_idle_enter); |
280f0677 FW |
673 | |
674 | /** | |
675 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
676 | * | |
677 | * When an interrupt fires while we are idle and it doesn't cause | |
678 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
679 | * an RCU callback, etc... | |
680 | * So we need to re-calculate and reprogram the next tick event. | |
681 | */ | |
682 | void tick_nohz_irq_exit(void) | |
683 | { | |
684 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
685 | ||
686 | if (!ts->inidle) | |
687 | return; | |
688 | ||
69a37bea YS |
689 | /* Cancel the timer because CPU already waken up from the C-states*/ |
690 | menu_hrtimer_cancel(); | |
19f5f736 | 691 | __tick_nohz_idle_enter(ts); |
79bf2bb3 TG |
692 | } |
693 | ||
4f86d3a8 LB |
694 | /** |
695 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
696 | * | |
697 | * Called from power state control code with interrupts disabled | |
698 | */ | |
699 | ktime_t tick_nohz_get_sleep_length(void) | |
700 | { | |
701 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
702 | ||
703 | return ts->sleep_length; | |
704 | } | |
705 | ||
c34bec5a TG |
706 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
707 | { | |
708 | hrtimer_cancel(&ts->sched_timer); | |
f5d411c9 | 709 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); |
c34bec5a TG |
710 | |
711 | while (1) { | |
712 | /* Forward the time to expire in the future */ | |
713 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
714 | ||
715 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { | |
268a3dcf | 716 | hrtimer_start_expires(&ts->sched_timer, |
5c333864 | 717 | HRTIMER_MODE_ABS_PINNED); |
c34bec5a TG |
718 | /* Check, if the timer was already in the past */ |
719 | if (hrtimer_active(&ts->sched_timer)) | |
720 | break; | |
721 | } else { | |
268a3dcf TG |
722 | if (!tick_program_event( |
723 | hrtimer_get_expires(&ts->sched_timer), 0)) | |
c34bec5a TG |
724 | break; |
725 | } | |
6f103929 | 726 | /* Reread time and update jiffies */ |
c34bec5a | 727 | now = ktime_get(); |
6f103929 | 728 | tick_do_update_jiffies64(now); |
c34bec5a TG |
729 | } |
730 | } | |
731 | ||
19f5f736 | 732 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
79bf2bb3 | 733 | { |
79bf2bb3 | 734 | /* Update jiffies first */ |
79bf2bb3 | 735 | tick_do_update_jiffies64(now); |
5aaa0b7a | 736 | update_cpu_load_nohz(); |
79bf2bb3 | 737 | |
749c8814 | 738 | calc_load_exit_idle(); |
2ac0d98f FW |
739 | touch_softlockup_watchdog(); |
740 | /* | |
741 | * Cancel the scheduled timer and restore the tick | |
742 | */ | |
743 | ts->tick_stopped = 0; | |
744 | ts->idle_exittime = now; | |
745 | ||
746 | tick_nohz_restart(ts, now); | |
747 | } | |
748 | ||
749 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) | |
750 | { | |
3f4724ea | 751 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 752 | unsigned long ticks; |
3f4724ea FW |
753 | |
754 | if (vtime_accounting_enabled()) | |
755 | return; | |
79bf2bb3 TG |
756 | /* |
757 | * We stopped the tick in idle. Update process times would miss the | |
758 | * time we slept as update_process_times does only a 1 tick | |
759 | * accounting. Enforce that this is accounted to idle ! | |
760 | */ | |
761 | ticks = jiffies - ts->idle_jiffies; | |
762 | /* | |
763 | * We might be one off. Do not randomly account a huge number of ticks! | |
764 | */ | |
79741dd3 MS |
765 | if (ticks && ticks < LONG_MAX) |
766 | account_idle_ticks(ticks); | |
767 | #endif | |
19f5f736 FW |
768 | } |
769 | ||
79bf2bb3 | 770 | /** |
280f0677 | 771 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
772 | * |
773 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
774 | * This also exit the RCU extended quiescent state. The CPU |
775 | * can use RCU again after this function is called. | |
79bf2bb3 | 776 | */ |
280f0677 | 777 | void tick_nohz_idle_exit(void) |
79bf2bb3 TG |
778 | { |
779 | int cpu = smp_processor_id(); | |
780 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
6378ddb5 | 781 | ktime_t now; |
79bf2bb3 | 782 | |
6378ddb5 | 783 | local_irq_disable(); |
2bbb6817 | 784 | |
15f827be FW |
785 | WARN_ON_ONCE(!ts->inidle); |
786 | ||
787 | ts->inidle = 0; | |
788 | ||
69a37bea YS |
789 | /* Cancel the timer because CPU already waken up from the C-states*/ |
790 | menu_hrtimer_cancel(); | |
15f827be | 791 | if (ts->idle_active || ts->tick_stopped) |
eed3b9cf MS |
792 | now = ktime_get(); |
793 | ||
794 | if (ts->idle_active) | |
795 | tick_nohz_stop_idle(cpu, now); | |
6378ddb5 | 796 | |
2ac0d98f | 797 | if (ts->tick_stopped) { |
19f5f736 | 798 | tick_nohz_restart_sched_tick(ts, now); |
2ac0d98f | 799 | tick_nohz_account_idle_ticks(ts); |
6378ddb5 | 800 | } |
79bf2bb3 | 801 | |
79bf2bb3 TG |
802 | local_irq_enable(); |
803 | } | |
4dbd2771 | 804 | EXPORT_SYMBOL_GPL(tick_nohz_idle_exit); |
79bf2bb3 TG |
805 | |
806 | static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now) | |
807 | { | |
808 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
cc584b21 | 809 | return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0); |
79bf2bb3 TG |
810 | } |
811 | ||
812 | /* | |
813 | * The nohz low res interrupt handler | |
814 | */ | |
815 | static void tick_nohz_handler(struct clock_event_device *dev) | |
816 | { | |
817 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
818 | struct pt_regs *regs = get_irq_regs(); | |
819 | ktime_t now = ktime_get(); | |
820 | ||
821 | dev->next_event.tv64 = KTIME_MAX; | |
822 | ||
5bb96226 | 823 | tick_sched_do_timer(now); |
9e8f559b | 824 | tick_sched_handle(ts, regs); |
79bf2bb3 | 825 | |
79bf2bb3 TG |
826 | while (tick_nohz_reprogram(ts, now)) { |
827 | now = ktime_get(); | |
828 | tick_do_update_jiffies64(now); | |
829 | } | |
830 | } | |
831 | ||
832 | /** | |
833 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
834 | */ | |
835 | static void tick_nohz_switch_to_nohz(void) | |
836 | { | |
837 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
838 | ktime_t next; | |
839 | ||
840 | if (!tick_nohz_enabled) | |
841 | return; | |
842 | ||
843 | local_irq_disable(); | |
844 | if (tick_switch_to_oneshot(tick_nohz_handler)) { | |
845 | local_irq_enable(); | |
846 | return; | |
847 | } | |
848 | ||
849 | ts->nohz_mode = NOHZ_MODE_LOWRES; | |
850 | ||
851 | /* | |
852 | * Recycle the hrtimer in ts, so we can share the | |
853 | * hrtimer_forward with the highres code. | |
854 | */ | |
855 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
856 | /* Get the next period */ | |
857 | next = tick_init_jiffy_update(); | |
858 | ||
859 | for (;;) { | |
cc584b21 | 860 | hrtimer_set_expires(&ts->sched_timer, next); |
79bf2bb3 TG |
861 | if (!tick_program_event(next, 0)) |
862 | break; | |
863 | next = ktime_add(next, tick_period); | |
864 | } | |
865 | local_irq_enable(); | |
79bf2bb3 TG |
866 | } |
867 | ||
fb02fbc1 TG |
868 | /* |
869 | * When NOHZ is enabled and the tick is stopped, we need to kick the | |
870 | * tick timer from irq_enter() so that the jiffies update is kept | |
871 | * alive during long running softirqs. That's ugly as hell, but | |
872 | * correctness is key even if we need to fix the offending softirq in | |
873 | * the first place. | |
874 | * | |
875 | * Note, this is different to tick_nohz_restart. We just kick the | |
876 | * timer and do not touch the other magic bits which need to be done | |
877 | * when idle is left. | |
878 | */ | |
eed3b9cf | 879 | static void tick_nohz_kick_tick(int cpu, ktime_t now) |
fb02fbc1 | 880 | { |
ae99286b TG |
881 | #if 0 |
882 | /* Switch back to 2.6.27 behaviour */ | |
883 | ||
fb02fbc1 | 884 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); |
eed3b9cf | 885 | ktime_t delta; |
fb02fbc1 | 886 | |
c4bd822e TG |
887 | /* |
888 | * Do not touch the tick device, when the next expiry is either | |
889 | * already reached or less/equal than the tick period. | |
890 | */ | |
268a3dcf | 891 | delta = ktime_sub(hrtimer_get_expires(&ts->sched_timer), now); |
c4bd822e TG |
892 | if (delta.tv64 <= tick_period.tv64) |
893 | return; | |
894 | ||
895 | tick_nohz_restart(ts, now); | |
ae99286b | 896 | #endif |
fb02fbc1 TG |
897 | } |
898 | ||
eed3b9cf MS |
899 | static inline void tick_check_nohz(int cpu) |
900 | { | |
901 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
902 | ktime_t now; | |
903 | ||
904 | if (!ts->idle_active && !ts->tick_stopped) | |
905 | return; | |
906 | now = ktime_get(); | |
907 | if (ts->idle_active) | |
908 | tick_nohz_stop_idle(cpu, now); | |
909 | if (ts->tick_stopped) { | |
910 | tick_nohz_update_jiffies(now); | |
911 | tick_nohz_kick_tick(cpu, now); | |
912 | } | |
913 | } | |
914 | ||
79bf2bb3 TG |
915 | #else |
916 | ||
917 | static inline void tick_nohz_switch_to_nohz(void) { } | |
eed3b9cf | 918 | static inline void tick_check_nohz(int cpu) { } |
79bf2bb3 | 919 | |
3451d024 | 920 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 921 | |
719254fa TG |
922 | /* |
923 | * Called from irq_enter to notify about the possible interruption of idle() | |
924 | */ | |
925 | void tick_check_idle(int cpu) | |
926 | { | |
fb02fbc1 | 927 | tick_check_oneshot_broadcast(cpu); |
eed3b9cf | 928 | tick_check_nohz(cpu); |
719254fa TG |
929 | } |
930 | ||
79bf2bb3 TG |
931 | /* |
932 | * High resolution timer specific code | |
933 | */ | |
934 | #ifdef CONFIG_HIGH_RES_TIMERS | |
935 | /* | |
4c9dc641 | 936 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 937 | * Called with interrupts disabled. |
79bf2bb3 TG |
938 | */ |
939 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
940 | { | |
941 | struct tick_sched *ts = | |
942 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
943 | struct pt_regs *regs = get_irq_regs(); |
944 | ktime_t now = ktime_get(); | |
d3ed7824 | 945 | |
5bb96226 | 946 | tick_sched_do_timer(now); |
79bf2bb3 TG |
947 | |
948 | /* | |
949 | * Do not call, when we are not in irq context and have | |
950 | * no valid regs pointer | |
951 | */ | |
9e8f559b FW |
952 | if (regs) |
953 | tick_sched_handle(ts, regs); | |
79bf2bb3 | 954 | |
79bf2bb3 TG |
955 | hrtimer_forward(timer, now, tick_period); |
956 | ||
957 | return HRTIMER_RESTART; | |
958 | } | |
959 | ||
5307c955 MG |
960 | static int sched_skew_tick; |
961 | ||
62cf20b3 TG |
962 | static int __init skew_tick(char *str) |
963 | { | |
964 | get_option(&str, &sched_skew_tick); | |
965 | ||
966 | return 0; | |
967 | } | |
968 | early_param("skew_tick", skew_tick); | |
969 | ||
79bf2bb3 TG |
970 | /** |
971 | * tick_setup_sched_timer - setup the tick emulation timer | |
972 | */ | |
973 | void tick_setup_sched_timer(void) | |
974 | { | |
975 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
976 | ktime_t now = ktime_get(); | |
977 | ||
978 | /* | |
979 | * Emulate tick processing via per-CPU hrtimers: | |
980 | */ | |
981 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
982 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 983 | |
3704540b | 984 | /* Get the next period (per cpu) */ |
cc584b21 | 985 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 986 | |
9c3f9e28 | 987 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
988 | if (sched_skew_tick) { |
989 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
990 | do_div(offset, num_possible_cpus()); | |
991 | offset *= smp_processor_id(); | |
992 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
993 | } | |
994 | ||
79bf2bb3 TG |
995 | for (;;) { |
996 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
5c333864 AB |
997 | hrtimer_start_expires(&ts->sched_timer, |
998 | HRTIMER_MODE_ABS_PINNED); | |
79bf2bb3 TG |
999 | /* Check, if the timer was already in the past */ |
1000 | if (hrtimer_active(&ts->sched_timer)) | |
1001 | break; | |
1002 | now = ktime_get(); | |
1003 | } | |
1004 | ||
3451d024 | 1005 | #ifdef CONFIG_NO_HZ_COMMON |
29c158e8 | 1006 | if (tick_nohz_enabled) |
79bf2bb3 TG |
1007 | ts->nohz_mode = NOHZ_MODE_HIGHRES; |
1008 | #endif | |
1009 | } | |
3c4fbe5e | 1010 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1011 | |
3451d024 | 1012 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1013 | void tick_cancel_sched_timer(int cpu) |
1014 | { | |
1015 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1016 | ||
3c4fbe5e | 1017 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1018 | if (ts->sched_timer.base) |
1019 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1020 | # endif |
a7901766 | 1021 | |
79bf2bb3 TG |
1022 | ts->nohz_mode = NOHZ_MODE_INACTIVE; |
1023 | } | |
3c4fbe5e | 1024 | #endif |
79bf2bb3 TG |
1025 | |
1026 | /** | |
1027 | * Async notification about clocksource changes | |
1028 | */ | |
1029 | void tick_clock_notify(void) | |
1030 | { | |
1031 | int cpu; | |
1032 | ||
1033 | for_each_possible_cpu(cpu) | |
1034 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * Async notification about clock event changes | |
1039 | */ | |
1040 | void tick_oneshot_notify(void) | |
1041 | { | |
1042 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1043 | ||
1044 | set_bit(0, &ts->check_clocks); | |
1045 | } | |
1046 | ||
1047 | /** | |
1048 | * Check, if a change happened, which makes oneshot possible. | |
1049 | * | |
1050 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1051 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1052 | * mode, because high resolution timers are disabled (either compile | |
1053 | * or runtime). | |
1054 | */ | |
1055 | int tick_check_oneshot_change(int allow_nohz) | |
1056 | { | |
1057 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1058 | ||
1059 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1060 | return 0; | |
1061 | ||
1062 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1063 | return 0; | |
1064 | ||
cf4fc6cb | 1065 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1066 | return 0; |
1067 | ||
1068 | if (!allow_nohz) | |
1069 | return 1; | |
1070 | ||
1071 | tick_nohz_switch_to_nohz(); | |
1072 | return 0; | |
1073 | } |