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 | 115 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE) |
c5bfece2 | 116 | && !tick_nohz_full_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 | ||
c5bfece2 FW |
146 | #ifdef CONFIG_NO_HZ_FULL |
147 | static cpumask_var_t nohz_full_mask; | |
148 | bool have_nohz_full_mask; | |
a831881b | 149 | |
c5bfece2 | 150 | int tick_nohz_full_cpu(int cpu) |
a831881b | 151 | { |
c5bfece2 | 152 | if (!have_nohz_full_mask) |
a831881b FW |
153 | return 0; |
154 | ||
c5bfece2 | 155 | return cpumask_test_cpu(cpu, nohz_full_mask); |
a831881b FW |
156 | } |
157 | ||
158 | /* Parse the boot-time nohz CPU list from the kernel parameters. */ | |
c5bfece2 | 159 | static int __init tick_nohz_full_setup(char *str) |
a831881b | 160 | { |
0453b435 FW |
161 | int cpu; |
162 | ||
c5bfece2 | 163 | alloc_bootmem_cpumask_var(&nohz_full_mask); |
0453b435 | 164 | if (cpulist_parse(str, nohz_full_mask) < 0) { |
c5bfece2 | 165 | pr_warning("NOHZ: Incorrect nohz_full cpumask\n"); |
0453b435 FW |
166 | return 1; |
167 | } | |
168 | ||
169 | cpu = smp_processor_id(); | |
170 | if (cpumask_test_cpu(cpu, nohz_full_mask)) { | |
171 | pr_warning("NO_HZ: Clearing %d from nohz_full range for timekeeping\n", cpu); | |
172 | cpumask_clear_cpu(cpu, nohz_full_mask); | |
173 | } | |
174 | have_nohz_full_mask = true; | |
175 | ||
a831881b FW |
176 | return 1; |
177 | } | |
c5bfece2 | 178 | __setup("nohz_full=", tick_nohz_full_setup); |
a831881b | 179 | |
a382bf93 FW |
180 | static int __cpuinit tick_nohz_cpu_down_callback(struct notifier_block *nfb, |
181 | unsigned long action, | |
182 | void *hcpu) | |
183 | { | |
184 | unsigned int cpu = (unsigned long)hcpu; | |
185 | ||
186 | switch (action & ~CPU_TASKS_FROZEN) { | |
187 | case CPU_DOWN_PREPARE: | |
188 | /* | |
189 | * If we handle the timekeeping duty for full dynticks CPUs, | |
190 | * we can't safely shutdown that CPU. | |
191 | */ | |
c5bfece2 | 192 | if (have_nohz_full_mask && tick_do_timer_cpu == cpu) |
a382bf93 FW |
193 | return -EINVAL; |
194 | break; | |
195 | } | |
196 | return NOTIFY_OK; | |
197 | } | |
198 | ||
1034fc2f FW |
199 | /* |
200 | * Worst case string length in chunks of CPU range seems 2 steps | |
201 | * separations: 0,2,4,6,... | |
202 | * This is NR_CPUS + sizeof('\0') | |
203 | */ | |
c5bfece2 | 204 | static char __initdata nohz_full_buf[NR_CPUS + 1]; |
1034fc2f | 205 | |
d1e43fa5 | 206 | void __init tick_nohz_init(void) |
a831881b | 207 | { |
d1e43fa5 FW |
208 | int cpu; |
209 | ||
210 | if (!have_nohz_full_mask) | |
211 | return; | |
212 | ||
213 | cpu_notifier(tick_nohz_cpu_down_callback, 0); | |
214 | ||
215 | /* Make sure full dynticks CPU are also RCU nocbs */ | |
216 | for_each_cpu(cpu, nohz_full_mask) { | |
217 | if (!rcu_is_nocb_cpu(cpu)) { | |
218 | pr_warning("NO_HZ: CPU %d is not RCU nocb: " | |
219 | "cleared from nohz_full range", cpu); | |
220 | cpumask_clear_cpu(cpu, nohz_full_mask); | |
221 | } | |
222 | } | |
a831881b | 223 | |
c5bfece2 FW |
224 | cpulist_scnprintf(nohz_full_buf, sizeof(nohz_full_buf), nohz_full_mask); |
225 | pr_info("NO_HZ: Full dynticks CPUs: %s.\n", nohz_full_buf); | |
a831881b | 226 | } |
a831881b | 227 | #else |
c5bfece2 | 228 | #define have_nohz_full_mask (0) |
a831881b FW |
229 | #endif |
230 | ||
79bf2bb3 TG |
231 | /* |
232 | * NOHZ - aka dynamic tick functionality | |
233 | */ | |
3451d024 | 234 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
235 | /* |
236 | * NO HZ enabled ? | |
237 | */ | |
9d2ad243 | 238 | int tick_nohz_enabled __read_mostly = 1; |
79bf2bb3 TG |
239 | |
240 | /* | |
241 | * Enable / Disable tickless mode | |
242 | */ | |
243 | static int __init setup_tick_nohz(char *str) | |
244 | { | |
245 | if (!strcmp(str, "off")) | |
246 | tick_nohz_enabled = 0; | |
247 | else if (!strcmp(str, "on")) | |
248 | tick_nohz_enabled = 1; | |
249 | else | |
250 | return 0; | |
251 | return 1; | |
252 | } | |
253 | ||
254 | __setup("nohz=", setup_tick_nohz); | |
255 | ||
256 | /** | |
257 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
258 | * | |
259 | * Called from interrupt entry when the CPU was idle | |
260 | * | |
261 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
262 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
263 | * value. We do this unconditionally on any cpu, as we don't know whether the | |
264 | * cpu, which has the update task assigned is in a long sleep. | |
265 | */ | |
eed3b9cf | 266 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 TG |
267 | { |
268 | int cpu = smp_processor_id(); | |
269 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
270 | unsigned long flags; | |
79bf2bb3 | 271 | |
5df7fa1c | 272 | ts->idle_waketime = now; |
79bf2bb3 TG |
273 | |
274 | local_irq_save(flags); | |
275 | tick_do_update_jiffies64(now); | |
276 | local_irq_restore(flags); | |
02ff3755 IM |
277 | |
278 | touch_softlockup_watchdog(); | |
79bf2bb3 TG |
279 | } |
280 | ||
595aac48 AV |
281 | /* |
282 | * Updates the per cpu time idle statistics counters | |
283 | */ | |
8d63bf94 | 284 | static void |
8c215bd3 | 285 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 286 | { |
eed3b9cf | 287 | ktime_t delta; |
6378ddb5 | 288 | |
595aac48 AV |
289 | if (ts->idle_active) { |
290 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 291 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 292 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
293 | else |
294 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 295 | ts->idle_entrytime = now; |
595aac48 | 296 | } |
8d63bf94 | 297 | |
e0e37c20 | 298 | if (last_update_time) |
8d63bf94 AV |
299 | *last_update_time = ktime_to_us(now); |
300 | ||
595aac48 AV |
301 | } |
302 | ||
303 | static void tick_nohz_stop_idle(int cpu, ktime_t now) | |
304 | { | |
305 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
306 | ||
8c215bd3 | 307 | update_ts_time_stats(cpu, ts, now, NULL); |
eed3b9cf | 308 | ts->idle_active = 0; |
56c7426b | 309 | |
eed3b9cf | 310 | sched_clock_idle_wakeup_event(0); |
6378ddb5 VP |
311 | } |
312 | ||
8c215bd3 | 313 | static ktime_t tick_nohz_start_idle(int cpu, struct tick_sched *ts) |
6378ddb5 | 314 | { |
430ee881 | 315 | ktime_t now = ktime_get(); |
595aac48 | 316 | |
6378ddb5 VP |
317 | ts->idle_entrytime = now; |
318 | ts->idle_active = 1; | |
56c7426b | 319 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
320 | return now; |
321 | } | |
322 | ||
b1f724c3 AV |
323 | /** |
324 | * get_cpu_idle_time_us - get the total idle time of a cpu | |
325 | * @cpu: CPU number to query | |
09a1d34f MH |
326 | * @last_update_time: variable to store update time in. Do not update |
327 | * counters if NULL. | |
b1f724c3 AV |
328 | * |
329 | * Return the cummulative idle time (since boot) for a given | |
6beea0cd | 330 | * CPU, in microseconds. |
b1f724c3 AV |
331 | * |
332 | * This time is measured via accounting rather than sampling, | |
333 | * and is as accurate as ktime_get() is. | |
334 | * | |
335 | * This function returns -1 if NOHZ is not enabled. | |
336 | */ | |
6378ddb5 VP |
337 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
338 | { | |
339 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 340 | ktime_t now, idle; |
6378ddb5 | 341 | |
8083e4ad | 342 | if (!tick_nohz_enabled) |
343 | return -1; | |
344 | ||
09a1d34f MH |
345 | now = ktime_get(); |
346 | if (last_update_time) { | |
347 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
348 | idle = ts->idle_sleeptime; | |
349 | } else { | |
350 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
351 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
352 | ||
353 | idle = ktime_add(ts->idle_sleeptime, delta); | |
354 | } else { | |
355 | idle = ts->idle_sleeptime; | |
356 | } | |
357 | } | |
358 | ||
359 | return ktime_to_us(idle); | |
8083e4ad | 360 | |
6378ddb5 | 361 | } |
8083e4ad | 362 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 363 | |
6beea0cd | 364 | /** |
0224cf4c AV |
365 | * get_cpu_iowait_time_us - get the total iowait time of a cpu |
366 | * @cpu: CPU number to query | |
09a1d34f MH |
367 | * @last_update_time: variable to store update time in. Do not update |
368 | * counters if NULL. | |
0224cf4c AV |
369 | * |
370 | * Return the cummulative iowait time (since boot) for a given | |
371 | * CPU, in microseconds. | |
372 | * | |
373 | * This time is measured via accounting rather than sampling, | |
374 | * and is as accurate as ktime_get() is. | |
375 | * | |
376 | * This function returns -1 if NOHZ is not enabled. | |
377 | */ | |
378 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
379 | { | |
380 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 381 | ktime_t now, iowait; |
0224cf4c AV |
382 | |
383 | if (!tick_nohz_enabled) | |
384 | return -1; | |
385 | ||
09a1d34f MH |
386 | now = ktime_get(); |
387 | if (last_update_time) { | |
388 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
389 | iowait = ts->iowait_sleeptime; | |
390 | } else { | |
391 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
392 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 393 | |
09a1d34f MH |
394 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
395 | } else { | |
396 | iowait = ts->iowait_sleeptime; | |
397 | } | |
398 | } | |
0224cf4c | 399 | |
09a1d34f | 400 | return ktime_to_us(iowait); |
0224cf4c AV |
401 | } |
402 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
403 | ||
84bf1bcc FW |
404 | static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts, |
405 | ktime_t now, int cpu) | |
79bf2bb3 | 406 | { |
280f0677 | 407 | unsigned long seq, last_jiffies, next_jiffies, delta_jiffies; |
84bf1bcc | 408 | ktime_t last_update, expires, ret = { .tv64 = 0 }; |
aa9b1630 | 409 | unsigned long rcu_delta_jiffies; |
4f86d3a8 | 410 | struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev; |
98962465 | 411 | u64 time_delta; |
79bf2bb3 | 412 | |
79bf2bb3 TG |
413 | /* Read jiffies and the time when jiffies were updated last */ |
414 | do { | |
d6ad4187 | 415 | seq = read_seqbegin(&jiffies_lock); |
79bf2bb3 TG |
416 | last_update = last_jiffies_update; |
417 | last_jiffies = jiffies; | |
27185016 | 418 | time_delta = timekeeping_max_deferment(); |
d6ad4187 | 419 | } while (read_seqretry(&jiffies_lock, seq)); |
79bf2bb3 | 420 | |
74876a98 | 421 | if (rcu_needs_cpu(cpu, &rcu_delta_jiffies) || |
00b42959 | 422 | arch_needs_cpu(cpu) || irq_work_needs_cpu()) { |
3c5d92a0 | 423 | next_jiffies = last_jiffies + 1; |
6ba9b346 | 424 | delta_jiffies = 1; |
3c5d92a0 MS |
425 | } else { |
426 | /* Get the next timer wheel timer */ | |
427 | next_jiffies = get_next_timer_interrupt(last_jiffies); | |
428 | delta_jiffies = next_jiffies - last_jiffies; | |
aa9b1630 PM |
429 | if (rcu_delta_jiffies < delta_jiffies) { |
430 | next_jiffies = last_jiffies + rcu_delta_jiffies; | |
431 | delta_jiffies = rcu_delta_jiffies; | |
432 | } | |
3c5d92a0 | 433 | } |
79bf2bb3 TG |
434 | /* |
435 | * Do not stop the tick, if we are only one off | |
436 | * or if the cpu is required for rcu | |
437 | */ | |
6ba9b346 | 438 | if (!ts->tick_stopped && delta_jiffies == 1) |
79bf2bb3 TG |
439 | goto out; |
440 | ||
441 | /* Schedule the tick, if we are at least one jiffie off */ | |
442 | if ((long)delta_jiffies >= 1) { | |
443 | ||
00147449 WR |
444 | /* |
445 | * If this cpu is the one which updates jiffies, then | |
446 | * give up the assignment and let it be taken by the | |
447 | * cpu which runs the tick timer next, which might be | |
448 | * this cpu as well. If we don't drop this here the | |
449 | * jiffies might be stale and do_timer() never | |
27185016 TG |
450 | * invoked. Keep track of the fact that it was the one |
451 | * which had the do_timer() duty last. If this cpu is | |
452 | * the one which had the do_timer() duty last, we | |
453 | * limit the sleep time to the timekeeping | |
454 | * max_deferement value which we retrieved | |
455 | * above. Otherwise we can sleep as long as we want. | |
00147449 | 456 | */ |
27185016 | 457 | if (cpu == tick_do_timer_cpu) { |
00147449 | 458 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; |
27185016 TG |
459 | ts->do_timer_last = 1; |
460 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
461 | time_delta = KTIME_MAX; | |
462 | ts->do_timer_last = 0; | |
463 | } else if (!ts->do_timer_last) { | |
464 | time_delta = KTIME_MAX; | |
465 | } | |
466 | ||
00147449 | 467 | /* |
98962465 JH |
468 | * calculate the expiry time for the next timer wheel |
469 | * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals | |
470 | * that there is no timer pending or at least extremely | |
471 | * far into the future (12 days for HZ=1000). In this | |
472 | * case we set the expiry to the end of time. | |
473 | */ | |
474 | if (likely(delta_jiffies < NEXT_TIMER_MAX_DELTA)) { | |
475 | /* | |
476 | * Calculate the time delta for the next timer event. | |
477 | * If the time delta exceeds the maximum time delta | |
478 | * permitted by the current clocksource then adjust | |
479 | * the time delta accordingly to ensure the | |
480 | * clocksource does not wrap. | |
481 | */ | |
482 | time_delta = min_t(u64, time_delta, | |
483 | tick_period.tv64 * delta_jiffies); | |
98962465 | 484 | } |
00147449 | 485 | |
27185016 TG |
486 | if (time_delta < KTIME_MAX) |
487 | expires = ktime_add_ns(last_update, time_delta); | |
488 | else | |
489 | expires.tv64 = KTIME_MAX; | |
00147449 | 490 | |
00147449 WR |
491 | /* Skip reprogram of event if its not changed */ |
492 | if (ts->tick_stopped && ktime_equal(expires, dev->next_event)) | |
493 | goto out; | |
494 | ||
84bf1bcc FW |
495 | ret = expires; |
496 | ||
79bf2bb3 TG |
497 | /* |
498 | * nohz_stop_sched_tick can be called several times before | |
499 | * the nohz_restart_sched_tick is called. This happens when | |
500 | * interrupts arrive which do not cause a reschedule. In the | |
501 | * first call we save the current tick time, so we can restart | |
502 | * the scheduler tick in nohz_restart_sched_tick. | |
503 | */ | |
504 | if (!ts->tick_stopped) { | |
c1cc017c | 505 | nohz_balance_enter_idle(cpu); |
5167e8d5 | 506 | calc_load_enter_idle(); |
46cb4b7c | 507 | |
f5d411c9 | 508 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
79bf2bb3 | 509 | ts->tick_stopped = 1; |
79bf2bb3 | 510 | } |
d3ed7824 | 511 | |
eaad084b | 512 | /* |
98962465 JH |
513 | * If the expiration time == KTIME_MAX, then |
514 | * in this case we simply stop the tick timer. | |
eaad084b | 515 | */ |
98962465 | 516 | if (unlikely(expires.tv64 == KTIME_MAX)) { |
eaad084b TG |
517 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
518 | hrtimer_cancel(&ts->sched_timer); | |
519 | goto out; | |
520 | } | |
521 | ||
79bf2bb3 TG |
522 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
523 | hrtimer_start(&ts->sched_timer, expires, | |
5c333864 | 524 | HRTIMER_MODE_ABS_PINNED); |
79bf2bb3 TG |
525 | /* Check, if the timer was already in the past */ |
526 | if (hrtimer_active(&ts->sched_timer)) | |
527 | goto out; | |
4c9dc641 | 528 | } else if (!tick_program_event(expires, 0)) |
79bf2bb3 TG |
529 | goto out; |
530 | /* | |
531 | * We are past the event already. So we crossed a | |
532 | * jiffie boundary. Update jiffies and raise the | |
533 | * softirq. | |
534 | */ | |
535 | tick_do_update_jiffies64(ktime_get()); | |
79bf2bb3 TG |
536 | } |
537 | raise_softirq_irqoff(TIMER_SOFTIRQ); | |
538 | out: | |
539 | ts->next_jiffies = next_jiffies; | |
540 | ts->last_jiffies = last_jiffies; | |
4f86d3a8 | 541 | ts->sleep_length = ktime_sub(dev->next_event, now); |
84bf1bcc FW |
542 | |
543 | return ret; | |
280f0677 FW |
544 | } |
545 | ||
5b39939a FW |
546 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
547 | { | |
548 | /* | |
549 | * If this cpu is offline and it is the one which updates | |
550 | * jiffies, then give up the assignment and let it be taken by | |
551 | * the cpu which runs the tick timer next. If we don't drop | |
552 | * this here the jiffies might be stale and do_timer() never | |
553 | * invoked. | |
554 | */ | |
555 | if (unlikely(!cpu_online(cpu))) { | |
556 | if (cpu == tick_do_timer_cpu) | |
557 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
558 | } | |
559 | ||
560 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) | |
561 | return false; | |
562 | ||
563 | if (need_resched()) | |
564 | return false; | |
565 | ||
566 | if (unlikely(local_softirq_pending() && cpu_online(cpu))) { | |
567 | static int ratelimit; | |
568 | ||
803b0eba PM |
569 | if (ratelimit < 10 && |
570 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
5b39939a FW |
571 | printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n", |
572 | (unsigned int) local_softirq_pending()); | |
573 | ratelimit++; | |
574 | } | |
575 | return false; | |
576 | } | |
577 | ||
c5bfece2 | 578 | if (have_nohz_full_mask) { |
a382bf93 FW |
579 | /* |
580 | * Keep the tick alive to guarantee timekeeping progression | |
581 | * if there are full dynticks CPUs around | |
582 | */ | |
583 | if (tick_do_timer_cpu == cpu) | |
584 | return false; | |
585 | /* | |
586 | * Boot safety: make sure the timekeeping duty has been | |
587 | * assigned before entering dyntick-idle mode, | |
588 | */ | |
589 | if (tick_do_timer_cpu == TICK_DO_TIMER_NONE) | |
590 | return false; | |
591 | } | |
592 | ||
5b39939a FW |
593 | return true; |
594 | } | |
595 | ||
19f5f736 FW |
596 | static void __tick_nohz_idle_enter(struct tick_sched *ts) |
597 | { | |
84bf1bcc | 598 | ktime_t now, expires; |
5b39939a | 599 | int cpu = smp_processor_id(); |
19f5f736 | 600 | |
5b39939a | 601 | now = tick_nohz_start_idle(cpu, ts); |
2ac0d98f | 602 | |
5b39939a FW |
603 | if (can_stop_idle_tick(cpu, ts)) { |
604 | int was_stopped = ts->tick_stopped; | |
605 | ||
606 | ts->idle_calls++; | |
84bf1bcc FW |
607 | |
608 | expires = tick_nohz_stop_sched_tick(ts, now, cpu); | |
609 | if (expires.tv64 > 0LL) { | |
610 | ts->idle_sleeps++; | |
611 | ts->idle_expires = expires; | |
612 | } | |
5b39939a FW |
613 | |
614 | if (!was_stopped && ts->tick_stopped) | |
615 | ts->idle_jiffies = ts->last_jiffies; | |
616 | } | |
280f0677 FW |
617 | } |
618 | ||
619 | /** | |
620 | * tick_nohz_idle_enter - stop the idle tick from the idle task | |
621 | * | |
622 | * When the next event is more than a tick into the future, stop the idle tick | |
623 | * Called when we start the idle loop. | |
2bbb6817 | 624 | * |
1268fbc7 | 625 | * The arch is responsible of calling: |
2bbb6817 FW |
626 | * |
627 | * - rcu_idle_enter() after its last use of RCU before the CPU is put | |
628 | * to sleep. | |
629 | * - rcu_idle_exit() before the first use of RCU after the CPU is woken up. | |
280f0677 | 630 | */ |
1268fbc7 | 631 | void tick_nohz_idle_enter(void) |
280f0677 FW |
632 | { |
633 | struct tick_sched *ts; | |
634 | ||
1268fbc7 FW |
635 | WARN_ON_ONCE(irqs_disabled()); |
636 | ||
0db49b72 LT |
637 | /* |
638 | * Update the idle state in the scheduler domain hierarchy | |
639 | * when tick_nohz_stop_sched_tick() is called from the idle loop. | |
640 | * State will be updated to busy during the first busy tick after | |
641 | * exiting idle. | |
642 | */ | |
643 | set_cpu_sd_state_idle(); | |
644 | ||
1268fbc7 FW |
645 | local_irq_disable(); |
646 | ||
280f0677 FW |
647 | ts = &__get_cpu_var(tick_cpu_sched); |
648 | /* | |
649 | * set ts->inidle unconditionally. even if the system did not | |
650 | * switch to nohz mode the cpu frequency governers rely on the | |
651 | * update of the idle time accounting in tick_nohz_start_idle(). | |
652 | */ | |
653 | ts->inidle = 1; | |
19f5f736 | 654 | __tick_nohz_idle_enter(ts); |
1268fbc7 FW |
655 | |
656 | local_irq_enable(); | |
280f0677 | 657 | } |
4dbd2771 | 658 | EXPORT_SYMBOL_GPL(tick_nohz_idle_enter); |
280f0677 FW |
659 | |
660 | /** | |
661 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
662 | * | |
663 | * When an interrupt fires while we are idle and it doesn't cause | |
664 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
665 | * an RCU callback, etc... | |
666 | * So we need to re-calculate and reprogram the next tick event. | |
667 | */ | |
668 | void tick_nohz_irq_exit(void) | |
669 | { | |
670 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
671 | ||
672 | if (!ts->inidle) | |
673 | return; | |
674 | ||
69a37bea YS |
675 | /* Cancel the timer because CPU already waken up from the C-states*/ |
676 | menu_hrtimer_cancel(); | |
19f5f736 | 677 | __tick_nohz_idle_enter(ts); |
79bf2bb3 TG |
678 | } |
679 | ||
4f86d3a8 LB |
680 | /** |
681 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
682 | * | |
683 | * Called from power state control code with interrupts disabled | |
684 | */ | |
685 | ktime_t tick_nohz_get_sleep_length(void) | |
686 | { | |
687 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
688 | ||
689 | return ts->sleep_length; | |
690 | } | |
691 | ||
c34bec5a TG |
692 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
693 | { | |
694 | hrtimer_cancel(&ts->sched_timer); | |
f5d411c9 | 695 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); |
c34bec5a TG |
696 | |
697 | while (1) { | |
698 | /* Forward the time to expire in the future */ | |
699 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
700 | ||
701 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { | |
268a3dcf | 702 | hrtimer_start_expires(&ts->sched_timer, |
5c333864 | 703 | HRTIMER_MODE_ABS_PINNED); |
c34bec5a TG |
704 | /* Check, if the timer was already in the past */ |
705 | if (hrtimer_active(&ts->sched_timer)) | |
706 | break; | |
707 | } else { | |
268a3dcf TG |
708 | if (!tick_program_event( |
709 | hrtimer_get_expires(&ts->sched_timer), 0)) | |
c34bec5a TG |
710 | break; |
711 | } | |
6f103929 | 712 | /* Reread time and update jiffies */ |
c34bec5a | 713 | now = ktime_get(); |
6f103929 | 714 | tick_do_update_jiffies64(now); |
c34bec5a TG |
715 | } |
716 | } | |
717 | ||
19f5f736 | 718 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
79bf2bb3 | 719 | { |
79bf2bb3 | 720 | /* Update jiffies first */ |
79bf2bb3 | 721 | tick_do_update_jiffies64(now); |
5aaa0b7a | 722 | update_cpu_load_nohz(); |
79bf2bb3 | 723 | |
749c8814 | 724 | calc_load_exit_idle(); |
2ac0d98f FW |
725 | touch_softlockup_watchdog(); |
726 | /* | |
727 | * Cancel the scheduled timer and restore the tick | |
728 | */ | |
729 | ts->tick_stopped = 0; | |
730 | ts->idle_exittime = now; | |
731 | ||
732 | tick_nohz_restart(ts, now); | |
733 | } | |
734 | ||
735 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) | |
736 | { | |
3f4724ea | 737 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 738 | unsigned long ticks; |
3f4724ea FW |
739 | |
740 | if (vtime_accounting_enabled()) | |
741 | return; | |
79bf2bb3 TG |
742 | /* |
743 | * We stopped the tick in idle. Update process times would miss the | |
744 | * time we slept as update_process_times does only a 1 tick | |
745 | * accounting. Enforce that this is accounted to idle ! | |
746 | */ | |
747 | ticks = jiffies - ts->idle_jiffies; | |
748 | /* | |
749 | * We might be one off. Do not randomly account a huge number of ticks! | |
750 | */ | |
79741dd3 MS |
751 | if (ticks && ticks < LONG_MAX) |
752 | account_idle_ticks(ticks); | |
753 | #endif | |
19f5f736 FW |
754 | } |
755 | ||
79bf2bb3 | 756 | /** |
280f0677 | 757 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
758 | * |
759 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
760 | * This also exit the RCU extended quiescent state. The CPU |
761 | * can use RCU again after this function is called. | |
79bf2bb3 | 762 | */ |
280f0677 | 763 | void tick_nohz_idle_exit(void) |
79bf2bb3 TG |
764 | { |
765 | int cpu = smp_processor_id(); | |
766 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
6378ddb5 | 767 | ktime_t now; |
79bf2bb3 | 768 | |
6378ddb5 | 769 | local_irq_disable(); |
2bbb6817 | 770 | |
15f827be FW |
771 | WARN_ON_ONCE(!ts->inidle); |
772 | ||
773 | ts->inidle = 0; | |
774 | ||
69a37bea YS |
775 | /* Cancel the timer because CPU already waken up from the C-states*/ |
776 | menu_hrtimer_cancel(); | |
15f827be | 777 | if (ts->idle_active || ts->tick_stopped) |
eed3b9cf MS |
778 | now = ktime_get(); |
779 | ||
780 | if (ts->idle_active) | |
781 | tick_nohz_stop_idle(cpu, now); | |
6378ddb5 | 782 | |
2ac0d98f | 783 | if (ts->tick_stopped) { |
19f5f736 | 784 | tick_nohz_restart_sched_tick(ts, now); |
2ac0d98f | 785 | tick_nohz_account_idle_ticks(ts); |
6378ddb5 | 786 | } |
79bf2bb3 | 787 | |
79bf2bb3 TG |
788 | local_irq_enable(); |
789 | } | |
4dbd2771 | 790 | EXPORT_SYMBOL_GPL(tick_nohz_idle_exit); |
79bf2bb3 TG |
791 | |
792 | static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now) | |
793 | { | |
794 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
cc584b21 | 795 | return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0); |
79bf2bb3 TG |
796 | } |
797 | ||
798 | /* | |
799 | * The nohz low res interrupt handler | |
800 | */ | |
801 | static void tick_nohz_handler(struct clock_event_device *dev) | |
802 | { | |
803 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
804 | struct pt_regs *regs = get_irq_regs(); | |
805 | ktime_t now = ktime_get(); | |
806 | ||
807 | dev->next_event.tv64 = KTIME_MAX; | |
808 | ||
5bb96226 | 809 | tick_sched_do_timer(now); |
9e8f559b | 810 | tick_sched_handle(ts, regs); |
79bf2bb3 | 811 | |
79bf2bb3 TG |
812 | while (tick_nohz_reprogram(ts, now)) { |
813 | now = ktime_get(); | |
814 | tick_do_update_jiffies64(now); | |
815 | } | |
816 | } | |
817 | ||
818 | /** | |
819 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
820 | */ | |
821 | static void tick_nohz_switch_to_nohz(void) | |
822 | { | |
823 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
824 | ktime_t next; | |
825 | ||
826 | if (!tick_nohz_enabled) | |
827 | return; | |
828 | ||
829 | local_irq_disable(); | |
830 | if (tick_switch_to_oneshot(tick_nohz_handler)) { | |
831 | local_irq_enable(); | |
832 | return; | |
833 | } | |
834 | ||
835 | ts->nohz_mode = NOHZ_MODE_LOWRES; | |
836 | ||
837 | /* | |
838 | * Recycle the hrtimer in ts, so we can share the | |
839 | * hrtimer_forward with the highres code. | |
840 | */ | |
841 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
842 | /* Get the next period */ | |
843 | next = tick_init_jiffy_update(); | |
844 | ||
845 | for (;;) { | |
cc584b21 | 846 | hrtimer_set_expires(&ts->sched_timer, next); |
79bf2bb3 TG |
847 | if (!tick_program_event(next, 0)) |
848 | break; | |
849 | next = ktime_add(next, tick_period); | |
850 | } | |
851 | local_irq_enable(); | |
79bf2bb3 TG |
852 | } |
853 | ||
fb02fbc1 TG |
854 | /* |
855 | * When NOHZ is enabled and the tick is stopped, we need to kick the | |
856 | * tick timer from irq_enter() so that the jiffies update is kept | |
857 | * alive during long running softirqs. That's ugly as hell, but | |
858 | * correctness is key even if we need to fix the offending softirq in | |
859 | * the first place. | |
860 | * | |
861 | * Note, this is different to tick_nohz_restart. We just kick the | |
862 | * timer and do not touch the other magic bits which need to be done | |
863 | * when idle is left. | |
864 | */ | |
eed3b9cf | 865 | static void tick_nohz_kick_tick(int cpu, ktime_t now) |
fb02fbc1 | 866 | { |
ae99286b TG |
867 | #if 0 |
868 | /* Switch back to 2.6.27 behaviour */ | |
869 | ||
fb02fbc1 | 870 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); |
eed3b9cf | 871 | ktime_t delta; |
fb02fbc1 | 872 | |
c4bd822e TG |
873 | /* |
874 | * Do not touch the tick device, when the next expiry is either | |
875 | * already reached or less/equal than the tick period. | |
876 | */ | |
268a3dcf | 877 | delta = ktime_sub(hrtimer_get_expires(&ts->sched_timer), now); |
c4bd822e TG |
878 | if (delta.tv64 <= tick_period.tv64) |
879 | return; | |
880 | ||
881 | tick_nohz_restart(ts, now); | |
ae99286b | 882 | #endif |
fb02fbc1 TG |
883 | } |
884 | ||
eed3b9cf MS |
885 | static inline void tick_check_nohz(int cpu) |
886 | { | |
887 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
888 | ktime_t now; | |
889 | ||
890 | if (!ts->idle_active && !ts->tick_stopped) | |
891 | return; | |
892 | now = ktime_get(); | |
893 | if (ts->idle_active) | |
894 | tick_nohz_stop_idle(cpu, now); | |
895 | if (ts->tick_stopped) { | |
896 | tick_nohz_update_jiffies(now); | |
897 | tick_nohz_kick_tick(cpu, now); | |
898 | } | |
899 | } | |
900 | ||
79bf2bb3 TG |
901 | #else |
902 | ||
903 | static inline void tick_nohz_switch_to_nohz(void) { } | |
eed3b9cf | 904 | static inline void tick_check_nohz(int cpu) { } |
79bf2bb3 | 905 | |
3451d024 | 906 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 907 | |
719254fa TG |
908 | /* |
909 | * Called from irq_enter to notify about the possible interruption of idle() | |
910 | */ | |
911 | void tick_check_idle(int cpu) | |
912 | { | |
fb02fbc1 | 913 | tick_check_oneshot_broadcast(cpu); |
eed3b9cf | 914 | tick_check_nohz(cpu); |
719254fa TG |
915 | } |
916 | ||
79bf2bb3 TG |
917 | /* |
918 | * High resolution timer specific code | |
919 | */ | |
920 | #ifdef CONFIG_HIGH_RES_TIMERS | |
921 | /* | |
4c9dc641 | 922 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 923 | * Called with interrupts disabled. |
79bf2bb3 TG |
924 | */ |
925 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
926 | { | |
927 | struct tick_sched *ts = | |
928 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
929 | struct pt_regs *regs = get_irq_regs(); |
930 | ktime_t now = ktime_get(); | |
d3ed7824 | 931 | |
5bb96226 | 932 | tick_sched_do_timer(now); |
79bf2bb3 TG |
933 | |
934 | /* | |
935 | * Do not call, when we are not in irq context and have | |
936 | * no valid regs pointer | |
937 | */ | |
9e8f559b FW |
938 | if (regs) |
939 | tick_sched_handle(ts, regs); | |
79bf2bb3 | 940 | |
79bf2bb3 TG |
941 | hrtimer_forward(timer, now, tick_period); |
942 | ||
943 | return HRTIMER_RESTART; | |
944 | } | |
945 | ||
5307c955 MG |
946 | static int sched_skew_tick; |
947 | ||
62cf20b3 TG |
948 | static int __init skew_tick(char *str) |
949 | { | |
950 | get_option(&str, &sched_skew_tick); | |
951 | ||
952 | return 0; | |
953 | } | |
954 | early_param("skew_tick", skew_tick); | |
955 | ||
79bf2bb3 TG |
956 | /** |
957 | * tick_setup_sched_timer - setup the tick emulation timer | |
958 | */ | |
959 | void tick_setup_sched_timer(void) | |
960 | { | |
961 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
962 | ktime_t now = ktime_get(); | |
963 | ||
964 | /* | |
965 | * Emulate tick processing via per-CPU hrtimers: | |
966 | */ | |
967 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
968 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 969 | |
3704540b | 970 | /* Get the next period (per cpu) */ |
cc584b21 | 971 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 972 | |
9c3f9e28 | 973 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
974 | if (sched_skew_tick) { |
975 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
976 | do_div(offset, num_possible_cpus()); | |
977 | offset *= smp_processor_id(); | |
978 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
979 | } | |
980 | ||
79bf2bb3 TG |
981 | for (;;) { |
982 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
5c333864 AB |
983 | hrtimer_start_expires(&ts->sched_timer, |
984 | HRTIMER_MODE_ABS_PINNED); | |
79bf2bb3 TG |
985 | /* Check, if the timer was already in the past */ |
986 | if (hrtimer_active(&ts->sched_timer)) | |
987 | break; | |
988 | now = ktime_get(); | |
989 | } | |
990 | ||
3451d024 | 991 | #ifdef CONFIG_NO_HZ_COMMON |
29c158e8 | 992 | if (tick_nohz_enabled) |
79bf2bb3 TG |
993 | ts->nohz_mode = NOHZ_MODE_HIGHRES; |
994 | #endif | |
995 | } | |
3c4fbe5e | 996 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 997 | |
3451d024 | 998 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
999 | void tick_cancel_sched_timer(int cpu) |
1000 | { | |
1001 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1002 | ||
3c4fbe5e | 1003 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1004 | if (ts->sched_timer.base) |
1005 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1006 | # endif |
a7901766 | 1007 | |
79bf2bb3 TG |
1008 | ts->nohz_mode = NOHZ_MODE_INACTIVE; |
1009 | } | |
3c4fbe5e | 1010 | #endif |
79bf2bb3 TG |
1011 | |
1012 | /** | |
1013 | * Async notification about clocksource changes | |
1014 | */ | |
1015 | void tick_clock_notify(void) | |
1016 | { | |
1017 | int cpu; | |
1018 | ||
1019 | for_each_possible_cpu(cpu) | |
1020 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * Async notification about clock event changes | |
1025 | */ | |
1026 | void tick_oneshot_notify(void) | |
1027 | { | |
1028 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1029 | ||
1030 | set_bit(0, &ts->check_clocks); | |
1031 | } | |
1032 | ||
1033 | /** | |
1034 | * Check, if a change happened, which makes oneshot possible. | |
1035 | * | |
1036 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1037 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1038 | * mode, because high resolution timers are disabled (either compile | |
1039 | * or runtime). | |
1040 | */ | |
1041 | int tick_check_oneshot_change(int allow_nohz) | |
1042 | { | |
1043 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1044 | ||
1045 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1046 | return 0; | |
1047 | ||
1048 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1049 | return 0; | |
1050 | ||
cf4fc6cb | 1051 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1052 | return 0; |
1053 | ||
1054 | if (!allow_nohz) | |
1055 | return 1; | |
1056 | ||
1057 | tick_nohz_switch_to_nohz(); | |
1058 | return 0; | |
1059 | } |