]>
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> |
9014c45d FW |
24 | #include <linux/posix-timers.h> |
25 | #include <linux/perf_event.h> | |
79bf2bb3 | 26 | |
9e203bcc DM |
27 | #include <asm/irq_regs.h> |
28 | ||
79bf2bb3 TG |
29 | #include "tick-internal.h" |
30 | ||
cb41a290 FW |
31 | #include <trace/events/timer.h> |
32 | ||
79bf2bb3 TG |
33 | /* |
34 | * Per cpu nohz control structure | |
35 | */ | |
33a5f626 | 36 | DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); |
79bf2bb3 TG |
37 | |
38 | /* | |
d6ad4187 | 39 | * The time, when the last jiffy update happened. Protected by jiffies_lock. |
79bf2bb3 TG |
40 | */ |
41 | static ktime_t last_jiffies_update; | |
42 | ||
289f480a IM |
43 | struct tick_sched *tick_get_tick_sched(int cpu) |
44 | { | |
45 | return &per_cpu(tick_cpu_sched, cpu); | |
46 | } | |
47 | ||
79bf2bb3 TG |
48 | /* |
49 | * Must be called with interrupts disabled ! | |
50 | */ | |
51 | static void tick_do_update_jiffies64(ktime_t now) | |
52 | { | |
53 | unsigned long ticks = 0; | |
54 | ktime_t delta; | |
55 | ||
7a14ce1d | 56 | /* |
d6ad4187 | 57 | * Do a quick check without holding jiffies_lock: |
7a14ce1d IM |
58 | */ |
59 | delta = ktime_sub(now, last_jiffies_update); | |
60 | if (delta.tv64 < tick_period.tv64) | |
61 | return; | |
62 | ||
d6ad4187 JS |
63 | /* Reevalute with jiffies_lock held */ |
64 | write_seqlock(&jiffies_lock); | |
79bf2bb3 TG |
65 | |
66 | delta = ktime_sub(now, last_jiffies_update); | |
67 | if (delta.tv64 >= tick_period.tv64) { | |
68 | ||
69 | delta = ktime_sub(delta, tick_period); | |
70 | last_jiffies_update = ktime_add(last_jiffies_update, | |
71 | tick_period); | |
72 | ||
73 | /* Slow path for long timeouts */ | |
74 | if (unlikely(delta.tv64 >= tick_period.tv64)) { | |
75 | s64 incr = ktime_to_ns(tick_period); | |
76 | ||
77 | ticks = ktime_divns(delta, incr); | |
78 | ||
79 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
80 | incr * ticks); | |
81 | } | |
82 | do_timer(++ticks); | |
49d670fb TG |
83 | |
84 | /* Keep the tick_next_period variable up to date */ | |
85 | tick_next_period = ktime_add(last_jiffies_update, tick_period); | |
79bf2bb3 | 86 | } |
d6ad4187 | 87 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
88 | } |
89 | ||
90 | /* | |
91 | * Initialize and return retrieve the jiffies update. | |
92 | */ | |
93 | static ktime_t tick_init_jiffy_update(void) | |
94 | { | |
95 | ktime_t period; | |
96 | ||
d6ad4187 | 97 | write_seqlock(&jiffies_lock); |
79bf2bb3 TG |
98 | /* Did we start the jiffies update yet ? */ |
99 | if (last_jiffies_update.tv64 == 0) | |
100 | last_jiffies_update = tick_next_period; | |
101 | period = last_jiffies_update; | |
d6ad4187 | 102 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
103 | return period; |
104 | } | |
105 | ||
5bb96226 FW |
106 | |
107 | static void tick_sched_do_timer(ktime_t now) | |
108 | { | |
109 | int cpu = smp_processor_id(); | |
110 | ||
3451d024 | 111 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
112 | /* |
113 | * Check if the do_timer duty was dropped. We don't care about | |
114 | * concurrency: This happens only when the cpu in charge went | |
115 | * into a long sleep. If two cpus happen to assign themself to | |
116 | * this duty, then the jiffies update is still serialized by | |
9c3f9e28 | 117 | * jiffies_lock. |
5bb96226 | 118 | */ |
a382bf93 | 119 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE) |
c5bfece2 | 120 | && !tick_nohz_full_cpu(cpu)) |
5bb96226 FW |
121 | tick_do_timer_cpu = cpu; |
122 | #endif | |
123 | ||
124 | /* Check, if the jiffies need an update */ | |
125 | if (tick_do_timer_cpu == cpu) | |
126 | tick_do_update_jiffies64(now); | |
127 | } | |
128 | ||
9e8f559b FW |
129 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
130 | { | |
3451d024 | 131 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
132 | /* |
133 | * When we are idle and the tick is stopped, we have to touch | |
134 | * the watchdog as we might not schedule for a really long | |
135 | * time. This happens on complete idle SMP systems while | |
136 | * waiting on the login prompt. We also increment the "start of | |
137 | * idle" jiffy stamp so the idle accounting adjustment we do | |
138 | * when we go busy again does not account too much ticks. | |
139 | */ | |
140 | if (ts->tick_stopped) { | |
141 | touch_softlockup_watchdog(); | |
142 | if (is_idle_task(current)) | |
143 | ts->idle_jiffies++; | |
144 | } | |
94a57140 | 145 | #endif |
9e8f559b FW |
146 | update_process_times(user_mode(regs)); |
147 | profile_tick(CPU_PROFILING); | |
148 | } | |
149 | ||
c5bfece2 FW |
150 | #ifdef CONFIG_NO_HZ_FULL |
151 | static cpumask_var_t nohz_full_mask; | |
152 | bool have_nohz_full_mask; | |
a831881b | 153 | |
9014c45d FW |
154 | static bool can_stop_full_tick(void) |
155 | { | |
156 | WARN_ON_ONCE(!irqs_disabled()); | |
157 | ||
cb41a290 FW |
158 | if (!sched_can_stop_tick()) { |
159 | trace_tick_stop(0, "more than 1 task in runqueue\n"); | |
9014c45d | 160 | return false; |
cb41a290 | 161 | } |
9014c45d | 162 | |
cb41a290 FW |
163 | if (!posix_cpu_timers_can_stop_tick(current)) { |
164 | trace_tick_stop(0, "posix timers running\n"); | |
9014c45d | 165 | return false; |
cb41a290 | 166 | } |
9014c45d | 167 | |
cb41a290 FW |
168 | if (!perf_event_can_stop_tick()) { |
169 | trace_tick_stop(0, "perf events running\n"); | |
9014c45d | 170 | return false; |
cb41a290 | 171 | } |
9014c45d FW |
172 | |
173 | /* sched_clock_tick() needs us? */ | |
174 | #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK | |
175 | /* | |
176 | * TODO: kick full dynticks CPUs when | |
177 | * sched_clock_stable is set. | |
178 | */ | |
cb41a290 FW |
179 | if (!sched_clock_stable) { |
180 | trace_tick_stop(0, "unstable sched clock\n"); | |
e12d0271 SR |
181 | /* |
182 | * Don't allow the user to think they can get | |
183 | * full NO_HZ with this machine. | |
184 | */ | |
543487c7 SR |
185 | WARN_ONCE(have_nohz_full_mask, |
186 | "NO_HZ FULL will not work with unstable sched clock"); | |
9014c45d | 187 | return false; |
cb41a290 | 188 | } |
9014c45d FW |
189 | #endif |
190 | ||
191 | return true; | |
192 | } | |
193 | ||
194 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now); | |
195 | ||
76c24fb0 FW |
196 | /* |
197 | * Re-evaluate the need for the tick on the current CPU | |
198 | * and restart it if necessary. | |
199 | */ | |
ff442c51 | 200 | void tick_nohz_full_check(void) |
76c24fb0 | 201 | { |
9014c45d FW |
202 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); |
203 | ||
204 | if (tick_nohz_full_cpu(smp_processor_id())) { | |
205 | if (ts->tick_stopped && !is_idle_task(current)) { | |
206 | if (!can_stop_full_tick()) | |
207 | tick_nohz_restart_sched_tick(ts, ktime_get()); | |
208 | } | |
209 | } | |
76c24fb0 FW |
210 | } |
211 | ||
212 | static void nohz_full_kick_work_func(struct irq_work *work) | |
213 | { | |
214 | tick_nohz_full_check(); | |
215 | } | |
216 | ||
217 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = { | |
218 | .func = nohz_full_kick_work_func, | |
219 | }; | |
220 | ||
221 | /* | |
222 | * Kick the current CPU if it's full dynticks in order to force it to | |
223 | * re-evaluate its dependency on the tick and restart it if necessary. | |
224 | */ | |
225 | void tick_nohz_full_kick(void) | |
226 | { | |
227 | if (tick_nohz_full_cpu(smp_processor_id())) | |
228 | irq_work_queue(&__get_cpu_var(nohz_full_kick_work)); | |
229 | } | |
230 | ||
231 | static void nohz_full_kick_ipi(void *info) | |
232 | { | |
233 | tick_nohz_full_check(); | |
234 | } | |
235 | ||
236 | /* | |
237 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
238 | * their dependency on the tick and restart it if necessary. | |
239 | */ | |
240 | void tick_nohz_full_kick_all(void) | |
241 | { | |
242 | if (!have_nohz_full_mask) | |
243 | return; | |
244 | ||
245 | preempt_disable(); | |
246 | smp_call_function_many(nohz_full_mask, | |
247 | nohz_full_kick_ipi, NULL, false); | |
248 | preempt_enable(); | |
249 | } | |
250 | ||
99e5ada9 FW |
251 | /* |
252 | * Re-evaluate the need for the tick as we switch the current task. | |
253 | * It might need the tick due to per task/process properties: | |
254 | * perf events, posix cpu timers, ... | |
255 | */ | |
256 | void tick_nohz_task_switch(struct task_struct *tsk) | |
257 | { | |
258 | unsigned long flags; | |
259 | ||
99e5ada9 FW |
260 | local_irq_save(flags); |
261 | ||
6296ace4 LZ |
262 | if (!tick_nohz_full_cpu(smp_processor_id())) |
263 | goto out; | |
264 | ||
99e5ada9 FW |
265 | if (tick_nohz_tick_stopped() && !can_stop_full_tick()) |
266 | tick_nohz_full_kick(); | |
267 | ||
6296ace4 | 268 | out: |
99e5ada9 FW |
269 | local_irq_restore(flags); |
270 | } | |
271 | ||
c5bfece2 | 272 | int tick_nohz_full_cpu(int cpu) |
a831881b | 273 | { |
c5bfece2 | 274 | if (!have_nohz_full_mask) |
a831881b FW |
275 | return 0; |
276 | ||
c5bfece2 | 277 | return cpumask_test_cpu(cpu, nohz_full_mask); |
a831881b FW |
278 | } |
279 | ||
280 | /* Parse the boot-time nohz CPU list from the kernel parameters. */ | |
c5bfece2 | 281 | static int __init tick_nohz_full_setup(char *str) |
a831881b | 282 | { |
0453b435 FW |
283 | int cpu; |
284 | ||
c5bfece2 | 285 | alloc_bootmem_cpumask_var(&nohz_full_mask); |
0453b435 | 286 | if (cpulist_parse(str, nohz_full_mask) < 0) { |
c5bfece2 | 287 | pr_warning("NOHZ: Incorrect nohz_full cpumask\n"); |
0453b435 FW |
288 | return 1; |
289 | } | |
290 | ||
291 | cpu = smp_processor_id(); | |
292 | if (cpumask_test_cpu(cpu, nohz_full_mask)) { | |
293 | pr_warning("NO_HZ: Clearing %d from nohz_full range for timekeeping\n", cpu); | |
294 | cpumask_clear_cpu(cpu, nohz_full_mask); | |
295 | } | |
296 | have_nohz_full_mask = true; | |
297 | ||
a831881b FW |
298 | return 1; |
299 | } | |
c5bfece2 | 300 | __setup("nohz_full=", tick_nohz_full_setup); |
a831881b | 301 | |
0db0628d | 302 | static int tick_nohz_cpu_down_callback(struct notifier_block *nfb, |
a382bf93 FW |
303 | unsigned long action, |
304 | void *hcpu) | |
305 | { | |
306 | unsigned int cpu = (unsigned long)hcpu; | |
307 | ||
308 | switch (action & ~CPU_TASKS_FROZEN) { | |
309 | case CPU_DOWN_PREPARE: | |
310 | /* | |
311 | * If we handle the timekeeping duty for full dynticks CPUs, | |
312 | * we can't safely shutdown that CPU. | |
313 | */ | |
c5bfece2 | 314 | if (have_nohz_full_mask && tick_do_timer_cpu == cpu) |
1a7f829f | 315 | return NOTIFY_BAD; |
a382bf93 FW |
316 | break; |
317 | } | |
318 | return NOTIFY_OK; | |
319 | } | |
320 | ||
1034fc2f FW |
321 | /* |
322 | * Worst case string length in chunks of CPU range seems 2 steps | |
323 | * separations: 0,2,4,6,... | |
324 | * This is NR_CPUS + sizeof('\0') | |
325 | */ | |
c5bfece2 | 326 | static char __initdata nohz_full_buf[NR_CPUS + 1]; |
1034fc2f | 327 | |
f98823ac FW |
328 | static int tick_nohz_init_all(void) |
329 | { | |
330 | int err = -1; | |
331 | ||
332 | #ifdef CONFIG_NO_HZ_FULL_ALL | |
333 | if (!alloc_cpumask_var(&nohz_full_mask, GFP_KERNEL)) { | |
334 | pr_err("NO_HZ: Can't allocate full dynticks cpumask\n"); | |
335 | return err; | |
336 | } | |
337 | err = 0; | |
338 | cpumask_setall(nohz_full_mask); | |
339 | cpumask_clear_cpu(smp_processor_id(), nohz_full_mask); | |
340 | have_nohz_full_mask = true; | |
341 | #endif | |
342 | return err; | |
343 | } | |
344 | ||
d1e43fa5 | 345 | void __init tick_nohz_init(void) |
a831881b | 346 | { |
f98823ac FW |
347 | if (!have_nohz_full_mask) { |
348 | if (tick_nohz_init_all() < 0) | |
349 | return; | |
350 | } | |
d1e43fa5 FW |
351 | |
352 | cpu_notifier(tick_nohz_cpu_down_callback, 0); | |
c5bfece2 FW |
353 | cpulist_scnprintf(nohz_full_buf, sizeof(nohz_full_buf), nohz_full_mask); |
354 | pr_info("NO_HZ: Full dynticks CPUs: %s.\n", nohz_full_buf); | |
a831881b | 355 | } |
a831881b | 356 | #else |
c5bfece2 | 357 | #define have_nohz_full_mask (0) |
a831881b FW |
358 | #endif |
359 | ||
79bf2bb3 TG |
360 | /* |
361 | * NOHZ - aka dynamic tick functionality | |
362 | */ | |
3451d024 | 363 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
364 | /* |
365 | * NO HZ enabled ? | |
366 | */ | |
9d2ad243 | 367 | int tick_nohz_enabled __read_mostly = 1; |
79bf2bb3 TG |
368 | |
369 | /* | |
370 | * Enable / Disable tickless mode | |
371 | */ | |
372 | static int __init setup_tick_nohz(char *str) | |
373 | { | |
374 | if (!strcmp(str, "off")) | |
375 | tick_nohz_enabled = 0; | |
376 | else if (!strcmp(str, "on")) | |
377 | tick_nohz_enabled = 1; | |
378 | else | |
379 | return 0; | |
380 | return 1; | |
381 | } | |
382 | ||
383 | __setup("nohz=", setup_tick_nohz); | |
384 | ||
385 | /** | |
386 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
387 | * | |
388 | * Called from interrupt entry when the CPU was idle | |
389 | * | |
390 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
391 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
392 | * value. We do this unconditionally on any cpu, as we don't know whether the | |
393 | * cpu, which has the update task assigned is in a long sleep. | |
394 | */ | |
eed3b9cf | 395 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 TG |
396 | { |
397 | int cpu = smp_processor_id(); | |
398 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
399 | unsigned long flags; | |
79bf2bb3 | 400 | |
5df7fa1c | 401 | ts->idle_waketime = now; |
79bf2bb3 TG |
402 | |
403 | local_irq_save(flags); | |
404 | tick_do_update_jiffies64(now); | |
405 | local_irq_restore(flags); | |
02ff3755 IM |
406 | |
407 | touch_softlockup_watchdog(); | |
79bf2bb3 TG |
408 | } |
409 | ||
595aac48 AV |
410 | /* |
411 | * Updates the per cpu time idle statistics counters | |
412 | */ | |
8d63bf94 | 413 | static void |
8c215bd3 | 414 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 415 | { |
eed3b9cf | 416 | ktime_t delta; |
6378ddb5 | 417 | |
595aac48 AV |
418 | if (ts->idle_active) { |
419 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 420 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 421 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
422 | else |
423 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 424 | ts->idle_entrytime = now; |
595aac48 | 425 | } |
8d63bf94 | 426 | |
e0e37c20 | 427 | if (last_update_time) |
8d63bf94 AV |
428 | *last_update_time = ktime_to_us(now); |
429 | ||
595aac48 AV |
430 | } |
431 | ||
432 | static void tick_nohz_stop_idle(int cpu, ktime_t now) | |
433 | { | |
434 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
435 | ||
8c215bd3 | 436 | update_ts_time_stats(cpu, ts, now, NULL); |
eed3b9cf | 437 | ts->idle_active = 0; |
56c7426b | 438 | |
eed3b9cf | 439 | sched_clock_idle_wakeup_event(0); |
6378ddb5 VP |
440 | } |
441 | ||
8c215bd3 | 442 | static ktime_t tick_nohz_start_idle(int cpu, struct tick_sched *ts) |
6378ddb5 | 443 | { |
430ee881 | 444 | ktime_t now = ktime_get(); |
595aac48 | 445 | |
6378ddb5 VP |
446 | ts->idle_entrytime = now; |
447 | ts->idle_active = 1; | |
56c7426b | 448 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
449 | return now; |
450 | } | |
451 | ||
b1f724c3 AV |
452 | /** |
453 | * get_cpu_idle_time_us - get the total idle time of a cpu | |
454 | * @cpu: CPU number to query | |
09a1d34f MH |
455 | * @last_update_time: variable to store update time in. Do not update |
456 | * counters if NULL. | |
b1f724c3 AV |
457 | * |
458 | * Return the cummulative idle time (since boot) for a given | |
6beea0cd | 459 | * CPU, in microseconds. |
b1f724c3 AV |
460 | * |
461 | * This time is measured via accounting rather than sampling, | |
462 | * and is as accurate as ktime_get() is. | |
463 | * | |
464 | * This function returns -1 if NOHZ is not enabled. | |
465 | */ | |
6378ddb5 VP |
466 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
467 | { | |
468 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 469 | ktime_t now, idle; |
6378ddb5 | 470 | |
8083e4ad | 471 | if (!tick_nohz_enabled) |
472 | return -1; | |
473 | ||
09a1d34f MH |
474 | now = ktime_get(); |
475 | if (last_update_time) { | |
476 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
477 | idle = ts->idle_sleeptime; | |
478 | } else { | |
479 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
480 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
481 | ||
482 | idle = ktime_add(ts->idle_sleeptime, delta); | |
483 | } else { | |
484 | idle = ts->idle_sleeptime; | |
485 | } | |
486 | } | |
487 | ||
488 | return ktime_to_us(idle); | |
8083e4ad | 489 | |
6378ddb5 | 490 | } |
8083e4ad | 491 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 492 | |
6beea0cd | 493 | /** |
0224cf4c AV |
494 | * get_cpu_iowait_time_us - get the total iowait time of a cpu |
495 | * @cpu: CPU number to query | |
09a1d34f MH |
496 | * @last_update_time: variable to store update time in. Do not update |
497 | * counters if NULL. | |
0224cf4c AV |
498 | * |
499 | * Return the cummulative iowait time (since boot) for a given | |
500 | * CPU, in microseconds. | |
501 | * | |
502 | * This time is measured via accounting rather than sampling, | |
503 | * and is as accurate as ktime_get() is. | |
504 | * | |
505 | * This function returns -1 if NOHZ is not enabled. | |
506 | */ | |
507 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
508 | { | |
509 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 510 | ktime_t now, iowait; |
0224cf4c AV |
511 | |
512 | if (!tick_nohz_enabled) | |
513 | return -1; | |
514 | ||
09a1d34f MH |
515 | now = ktime_get(); |
516 | if (last_update_time) { | |
517 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
518 | iowait = ts->iowait_sleeptime; | |
519 | } else { | |
520 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
521 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 522 | |
09a1d34f MH |
523 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
524 | } else { | |
525 | iowait = ts->iowait_sleeptime; | |
526 | } | |
527 | } | |
0224cf4c | 528 | |
09a1d34f | 529 | return ktime_to_us(iowait); |
0224cf4c AV |
530 | } |
531 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
532 | ||
84bf1bcc FW |
533 | static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts, |
534 | ktime_t now, int cpu) | |
79bf2bb3 | 535 | { |
280f0677 | 536 | unsigned long seq, last_jiffies, next_jiffies, delta_jiffies; |
84bf1bcc | 537 | ktime_t last_update, expires, ret = { .tv64 = 0 }; |
aa9b1630 | 538 | unsigned long rcu_delta_jiffies; |
4f86d3a8 | 539 | struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev; |
98962465 | 540 | u64 time_delta; |
79bf2bb3 | 541 | |
79bf2bb3 TG |
542 | /* Read jiffies and the time when jiffies were updated last */ |
543 | do { | |
d6ad4187 | 544 | seq = read_seqbegin(&jiffies_lock); |
79bf2bb3 TG |
545 | last_update = last_jiffies_update; |
546 | last_jiffies = jiffies; | |
27185016 | 547 | time_delta = timekeeping_max_deferment(); |
d6ad4187 | 548 | } while (read_seqretry(&jiffies_lock, seq)); |
79bf2bb3 | 549 | |
74876a98 | 550 | if (rcu_needs_cpu(cpu, &rcu_delta_jiffies) || |
00b42959 | 551 | arch_needs_cpu(cpu) || irq_work_needs_cpu()) { |
3c5d92a0 | 552 | next_jiffies = last_jiffies + 1; |
6ba9b346 | 553 | delta_jiffies = 1; |
3c5d92a0 MS |
554 | } else { |
555 | /* Get the next timer wheel timer */ | |
556 | next_jiffies = get_next_timer_interrupt(last_jiffies); | |
557 | delta_jiffies = next_jiffies - last_jiffies; | |
aa9b1630 PM |
558 | if (rcu_delta_jiffies < delta_jiffies) { |
559 | next_jiffies = last_jiffies + rcu_delta_jiffies; | |
560 | delta_jiffies = rcu_delta_jiffies; | |
561 | } | |
3c5d92a0 | 562 | } |
47aa8b6c | 563 | |
79bf2bb3 | 564 | /* |
47aa8b6c IM |
565 | * Do not stop the tick, if we are only one off (or less) |
566 | * or if the cpu is required for RCU: | |
79bf2bb3 | 567 | */ |
47aa8b6c | 568 | if (!ts->tick_stopped && delta_jiffies <= 1) |
79bf2bb3 TG |
569 | goto out; |
570 | ||
571 | /* Schedule the tick, if we are at least one jiffie off */ | |
572 | if ((long)delta_jiffies >= 1) { | |
573 | ||
00147449 WR |
574 | /* |
575 | * If this cpu is the one which updates jiffies, then | |
576 | * give up the assignment and let it be taken by the | |
577 | * cpu which runs the tick timer next, which might be | |
578 | * this cpu as well. If we don't drop this here the | |
579 | * jiffies might be stale and do_timer() never | |
27185016 TG |
580 | * invoked. Keep track of the fact that it was the one |
581 | * which had the do_timer() duty last. If this cpu is | |
582 | * the one which had the do_timer() duty last, we | |
583 | * limit the sleep time to the timekeeping | |
584 | * max_deferement value which we retrieved | |
585 | * above. Otherwise we can sleep as long as we want. | |
00147449 | 586 | */ |
27185016 | 587 | if (cpu == tick_do_timer_cpu) { |
00147449 | 588 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; |
27185016 TG |
589 | ts->do_timer_last = 1; |
590 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
591 | time_delta = KTIME_MAX; | |
592 | ts->do_timer_last = 0; | |
593 | } else if (!ts->do_timer_last) { | |
594 | time_delta = KTIME_MAX; | |
595 | } | |
596 | ||
265f22a9 FW |
597 | #ifdef CONFIG_NO_HZ_FULL |
598 | if (!ts->inidle) { | |
599 | time_delta = min(time_delta, | |
600 | scheduler_tick_max_deferment()); | |
601 | } | |
602 | #endif | |
603 | ||
00147449 | 604 | /* |
98962465 JH |
605 | * calculate the expiry time for the next timer wheel |
606 | * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals | |
607 | * that there is no timer pending or at least extremely | |
608 | * far into the future (12 days for HZ=1000). In this | |
609 | * case we set the expiry to the end of time. | |
610 | */ | |
611 | if (likely(delta_jiffies < NEXT_TIMER_MAX_DELTA)) { | |
612 | /* | |
613 | * Calculate the time delta for the next timer event. | |
614 | * If the time delta exceeds the maximum time delta | |
615 | * permitted by the current clocksource then adjust | |
616 | * the time delta accordingly to ensure the | |
617 | * clocksource does not wrap. | |
618 | */ | |
619 | time_delta = min_t(u64, time_delta, | |
620 | tick_period.tv64 * delta_jiffies); | |
98962465 | 621 | } |
00147449 | 622 | |
27185016 TG |
623 | if (time_delta < KTIME_MAX) |
624 | expires = ktime_add_ns(last_update, time_delta); | |
625 | else | |
626 | expires.tv64 = KTIME_MAX; | |
00147449 | 627 | |
00147449 WR |
628 | /* Skip reprogram of event if its not changed */ |
629 | if (ts->tick_stopped && ktime_equal(expires, dev->next_event)) | |
630 | goto out; | |
631 | ||
84bf1bcc FW |
632 | ret = expires; |
633 | ||
79bf2bb3 TG |
634 | /* |
635 | * nohz_stop_sched_tick can be called several times before | |
636 | * the nohz_restart_sched_tick is called. This happens when | |
637 | * interrupts arrive which do not cause a reschedule. In the | |
638 | * first call we save the current tick time, so we can restart | |
639 | * the scheduler tick in nohz_restart_sched_tick. | |
640 | */ | |
641 | if (!ts->tick_stopped) { | |
c1cc017c | 642 | nohz_balance_enter_idle(cpu); |
5167e8d5 | 643 | calc_load_enter_idle(); |
46cb4b7c | 644 | |
f5d411c9 | 645 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
79bf2bb3 | 646 | ts->tick_stopped = 1; |
cb41a290 | 647 | trace_tick_stop(1, " "); |
79bf2bb3 | 648 | } |
d3ed7824 | 649 | |
eaad084b | 650 | /* |
98962465 JH |
651 | * If the expiration time == KTIME_MAX, then |
652 | * in this case we simply stop the tick timer. | |
eaad084b | 653 | */ |
98962465 | 654 | if (unlikely(expires.tv64 == KTIME_MAX)) { |
eaad084b TG |
655 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
656 | hrtimer_cancel(&ts->sched_timer); | |
657 | goto out; | |
658 | } | |
659 | ||
79bf2bb3 TG |
660 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
661 | hrtimer_start(&ts->sched_timer, expires, | |
5c333864 | 662 | HRTIMER_MODE_ABS_PINNED); |
79bf2bb3 TG |
663 | /* Check, if the timer was already in the past */ |
664 | if (hrtimer_active(&ts->sched_timer)) | |
665 | goto out; | |
4c9dc641 | 666 | } else if (!tick_program_event(expires, 0)) |
79bf2bb3 TG |
667 | goto out; |
668 | /* | |
669 | * We are past the event already. So we crossed a | |
670 | * jiffie boundary. Update jiffies and raise the | |
671 | * softirq. | |
672 | */ | |
673 | tick_do_update_jiffies64(ktime_get()); | |
79bf2bb3 TG |
674 | } |
675 | raise_softirq_irqoff(TIMER_SOFTIRQ); | |
676 | out: | |
677 | ts->next_jiffies = next_jiffies; | |
678 | ts->last_jiffies = last_jiffies; | |
4f86d3a8 | 679 | ts->sleep_length = ktime_sub(dev->next_event, now); |
84bf1bcc FW |
680 | |
681 | return ret; | |
280f0677 FW |
682 | } |
683 | ||
5811d996 FW |
684 | static void tick_nohz_full_stop_tick(struct tick_sched *ts) |
685 | { | |
686 | #ifdef CONFIG_NO_HZ_FULL | |
687 | int cpu = smp_processor_id(); | |
688 | ||
689 | if (!tick_nohz_full_cpu(cpu) || is_idle_task(current)) | |
690 | return; | |
691 | ||
692 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) | |
693 | return; | |
694 | ||
695 | if (!can_stop_full_tick()) | |
696 | return; | |
697 | ||
698 | tick_nohz_stop_sched_tick(ts, ktime_get(), cpu); | |
699 | #endif | |
700 | } | |
701 | ||
5b39939a FW |
702 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
703 | { | |
704 | /* | |
705 | * If this cpu is offline and it is the one which updates | |
706 | * jiffies, then give up the assignment and let it be taken by | |
707 | * the cpu which runs the tick timer next. If we don't drop | |
708 | * this here the jiffies might be stale and do_timer() never | |
709 | * invoked. | |
710 | */ | |
711 | if (unlikely(!cpu_online(cpu))) { | |
712 | if (cpu == tick_do_timer_cpu) | |
713 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
f7ea0fd6 | 714 | return false; |
5b39939a FW |
715 | } |
716 | ||
717 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) | |
718 | return false; | |
719 | ||
720 | if (need_resched()) | |
721 | return false; | |
722 | ||
723 | if (unlikely(local_softirq_pending() && cpu_online(cpu))) { | |
724 | static int ratelimit; | |
725 | ||
803b0eba PM |
726 | if (ratelimit < 10 && |
727 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
cfea7d7e RV |
728 | pr_warn("NOHZ: local_softirq_pending %02x\n", |
729 | (unsigned int) local_softirq_pending()); | |
5b39939a FW |
730 | ratelimit++; |
731 | } | |
732 | return false; | |
733 | } | |
734 | ||
c5bfece2 | 735 | if (have_nohz_full_mask) { |
a382bf93 FW |
736 | /* |
737 | * Keep the tick alive to guarantee timekeeping progression | |
738 | * if there are full dynticks CPUs around | |
739 | */ | |
740 | if (tick_do_timer_cpu == cpu) | |
741 | return false; | |
742 | /* | |
743 | * Boot safety: make sure the timekeeping duty has been | |
744 | * assigned before entering dyntick-idle mode, | |
745 | */ | |
746 | if (tick_do_timer_cpu == TICK_DO_TIMER_NONE) | |
747 | return false; | |
748 | } | |
749 | ||
5b39939a FW |
750 | return true; |
751 | } | |
752 | ||
19f5f736 FW |
753 | static void __tick_nohz_idle_enter(struct tick_sched *ts) |
754 | { | |
84bf1bcc | 755 | ktime_t now, expires; |
5b39939a | 756 | int cpu = smp_processor_id(); |
19f5f736 | 757 | |
5b39939a | 758 | now = tick_nohz_start_idle(cpu, ts); |
2ac0d98f | 759 | |
5b39939a FW |
760 | if (can_stop_idle_tick(cpu, ts)) { |
761 | int was_stopped = ts->tick_stopped; | |
762 | ||
763 | ts->idle_calls++; | |
84bf1bcc FW |
764 | |
765 | expires = tick_nohz_stop_sched_tick(ts, now, cpu); | |
766 | if (expires.tv64 > 0LL) { | |
767 | ts->idle_sleeps++; | |
768 | ts->idle_expires = expires; | |
769 | } | |
5b39939a FW |
770 | |
771 | if (!was_stopped && ts->tick_stopped) | |
772 | ts->idle_jiffies = ts->last_jiffies; | |
773 | } | |
280f0677 FW |
774 | } |
775 | ||
776 | /** | |
777 | * tick_nohz_idle_enter - stop the idle tick from the idle task | |
778 | * | |
779 | * When the next event is more than a tick into the future, stop the idle tick | |
780 | * Called when we start the idle loop. | |
2bbb6817 | 781 | * |
1268fbc7 | 782 | * The arch is responsible of calling: |
2bbb6817 FW |
783 | * |
784 | * - rcu_idle_enter() after its last use of RCU before the CPU is put | |
785 | * to sleep. | |
786 | * - rcu_idle_exit() before the first use of RCU after the CPU is woken up. | |
280f0677 | 787 | */ |
1268fbc7 | 788 | void tick_nohz_idle_enter(void) |
280f0677 FW |
789 | { |
790 | struct tick_sched *ts; | |
791 | ||
1268fbc7 FW |
792 | WARN_ON_ONCE(irqs_disabled()); |
793 | ||
0db49b72 LT |
794 | /* |
795 | * Update the idle state in the scheduler domain hierarchy | |
796 | * when tick_nohz_stop_sched_tick() is called from the idle loop. | |
797 | * State will be updated to busy during the first busy tick after | |
798 | * exiting idle. | |
799 | */ | |
800 | set_cpu_sd_state_idle(); | |
801 | ||
1268fbc7 FW |
802 | local_irq_disable(); |
803 | ||
280f0677 FW |
804 | ts = &__get_cpu_var(tick_cpu_sched); |
805 | /* | |
806 | * set ts->inidle unconditionally. even if the system did not | |
807 | * switch to nohz mode the cpu frequency governers rely on the | |
808 | * update of the idle time accounting in tick_nohz_start_idle(). | |
809 | */ | |
810 | ts->inidle = 1; | |
19f5f736 | 811 | __tick_nohz_idle_enter(ts); |
1268fbc7 FW |
812 | |
813 | local_irq_enable(); | |
280f0677 | 814 | } |
4dbd2771 | 815 | EXPORT_SYMBOL_GPL(tick_nohz_idle_enter); |
280f0677 FW |
816 | |
817 | /** | |
818 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
819 | * | |
820 | * When an interrupt fires while we are idle and it doesn't cause | |
821 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
822 | * an RCU callback, etc... | |
823 | * So we need to re-calculate and reprogram the next tick event. | |
824 | */ | |
825 | void tick_nohz_irq_exit(void) | |
826 | { | |
827 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
828 | ||
5811d996 FW |
829 | if (ts->inidle) { |
830 | /* Cancel the timer because CPU already waken up from the C-states*/ | |
831 | menu_hrtimer_cancel(); | |
832 | __tick_nohz_idle_enter(ts); | |
833 | } else { | |
834 | tick_nohz_full_stop_tick(ts); | |
835 | } | |
79bf2bb3 TG |
836 | } |
837 | ||
4f86d3a8 LB |
838 | /** |
839 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
840 | * | |
841 | * Called from power state control code with interrupts disabled | |
842 | */ | |
843 | ktime_t tick_nohz_get_sleep_length(void) | |
844 | { | |
845 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
846 | ||
847 | return ts->sleep_length; | |
848 | } | |
849 | ||
c34bec5a TG |
850 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
851 | { | |
852 | hrtimer_cancel(&ts->sched_timer); | |
f5d411c9 | 853 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); |
c34bec5a TG |
854 | |
855 | while (1) { | |
856 | /* Forward the time to expire in the future */ | |
857 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
858 | ||
859 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { | |
268a3dcf | 860 | hrtimer_start_expires(&ts->sched_timer, |
5c333864 | 861 | HRTIMER_MODE_ABS_PINNED); |
c34bec5a TG |
862 | /* Check, if the timer was already in the past */ |
863 | if (hrtimer_active(&ts->sched_timer)) | |
864 | break; | |
865 | } else { | |
268a3dcf TG |
866 | if (!tick_program_event( |
867 | hrtimer_get_expires(&ts->sched_timer), 0)) | |
c34bec5a TG |
868 | break; |
869 | } | |
6f103929 | 870 | /* Reread time and update jiffies */ |
c34bec5a | 871 | now = ktime_get(); |
6f103929 | 872 | tick_do_update_jiffies64(now); |
c34bec5a TG |
873 | } |
874 | } | |
875 | ||
19f5f736 | 876 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
79bf2bb3 | 877 | { |
79bf2bb3 | 878 | /* Update jiffies first */ |
79bf2bb3 | 879 | tick_do_update_jiffies64(now); |
5aaa0b7a | 880 | update_cpu_load_nohz(); |
79bf2bb3 | 881 | |
749c8814 | 882 | calc_load_exit_idle(); |
2ac0d98f FW |
883 | touch_softlockup_watchdog(); |
884 | /* | |
885 | * Cancel the scheduled timer and restore the tick | |
886 | */ | |
887 | ts->tick_stopped = 0; | |
888 | ts->idle_exittime = now; | |
889 | ||
890 | tick_nohz_restart(ts, now); | |
891 | } | |
892 | ||
893 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) | |
894 | { | |
3f4724ea | 895 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 896 | unsigned long ticks; |
3f4724ea FW |
897 | |
898 | if (vtime_accounting_enabled()) | |
899 | return; | |
79bf2bb3 TG |
900 | /* |
901 | * We stopped the tick in idle. Update process times would miss the | |
902 | * time we slept as update_process_times does only a 1 tick | |
903 | * accounting. Enforce that this is accounted to idle ! | |
904 | */ | |
905 | ticks = jiffies - ts->idle_jiffies; | |
906 | /* | |
907 | * We might be one off. Do not randomly account a huge number of ticks! | |
908 | */ | |
79741dd3 MS |
909 | if (ticks && ticks < LONG_MAX) |
910 | account_idle_ticks(ticks); | |
911 | #endif | |
19f5f736 FW |
912 | } |
913 | ||
79bf2bb3 | 914 | /** |
280f0677 | 915 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
916 | * |
917 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
918 | * This also exit the RCU extended quiescent state. The CPU |
919 | * can use RCU again after this function is called. | |
79bf2bb3 | 920 | */ |
280f0677 | 921 | void tick_nohz_idle_exit(void) |
79bf2bb3 TG |
922 | { |
923 | int cpu = smp_processor_id(); | |
924 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
6378ddb5 | 925 | ktime_t now; |
79bf2bb3 | 926 | |
6378ddb5 | 927 | local_irq_disable(); |
2bbb6817 | 928 | |
15f827be FW |
929 | WARN_ON_ONCE(!ts->inidle); |
930 | ||
931 | ts->inidle = 0; | |
932 | ||
69a37bea YS |
933 | /* Cancel the timer because CPU already waken up from the C-states*/ |
934 | menu_hrtimer_cancel(); | |
15f827be | 935 | if (ts->idle_active || ts->tick_stopped) |
eed3b9cf MS |
936 | now = ktime_get(); |
937 | ||
938 | if (ts->idle_active) | |
939 | tick_nohz_stop_idle(cpu, now); | |
6378ddb5 | 940 | |
2ac0d98f | 941 | if (ts->tick_stopped) { |
19f5f736 | 942 | tick_nohz_restart_sched_tick(ts, now); |
2ac0d98f | 943 | tick_nohz_account_idle_ticks(ts); |
6378ddb5 | 944 | } |
79bf2bb3 | 945 | |
79bf2bb3 TG |
946 | local_irq_enable(); |
947 | } | |
4dbd2771 | 948 | EXPORT_SYMBOL_GPL(tick_nohz_idle_exit); |
79bf2bb3 TG |
949 | |
950 | static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now) | |
951 | { | |
952 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
cc584b21 | 953 | return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0); |
79bf2bb3 TG |
954 | } |
955 | ||
956 | /* | |
957 | * The nohz low res interrupt handler | |
958 | */ | |
959 | static void tick_nohz_handler(struct clock_event_device *dev) | |
960 | { | |
961 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
962 | struct pt_regs *regs = get_irq_regs(); | |
963 | ktime_t now = ktime_get(); | |
964 | ||
965 | dev->next_event.tv64 = KTIME_MAX; | |
966 | ||
5bb96226 | 967 | tick_sched_do_timer(now); |
9e8f559b | 968 | tick_sched_handle(ts, regs); |
79bf2bb3 | 969 | |
79bf2bb3 TG |
970 | while (tick_nohz_reprogram(ts, now)) { |
971 | now = ktime_get(); | |
972 | tick_do_update_jiffies64(now); | |
973 | } | |
974 | } | |
975 | ||
976 | /** | |
977 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
978 | */ | |
979 | static void tick_nohz_switch_to_nohz(void) | |
980 | { | |
981 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
982 | ktime_t next; | |
983 | ||
984 | if (!tick_nohz_enabled) | |
985 | return; | |
986 | ||
987 | local_irq_disable(); | |
988 | if (tick_switch_to_oneshot(tick_nohz_handler)) { | |
989 | local_irq_enable(); | |
990 | return; | |
991 | } | |
992 | ||
993 | ts->nohz_mode = NOHZ_MODE_LOWRES; | |
994 | ||
995 | /* | |
996 | * Recycle the hrtimer in ts, so we can share the | |
997 | * hrtimer_forward with the highres code. | |
998 | */ | |
999 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1000 | /* Get the next period */ | |
1001 | next = tick_init_jiffy_update(); | |
1002 | ||
1003 | for (;;) { | |
cc584b21 | 1004 | hrtimer_set_expires(&ts->sched_timer, next); |
79bf2bb3 TG |
1005 | if (!tick_program_event(next, 0)) |
1006 | break; | |
1007 | next = ktime_add(next, tick_period); | |
1008 | } | |
1009 | local_irq_enable(); | |
79bf2bb3 TG |
1010 | } |
1011 | ||
fb02fbc1 TG |
1012 | /* |
1013 | * When NOHZ is enabled and the tick is stopped, we need to kick the | |
1014 | * tick timer from irq_enter() so that the jiffies update is kept | |
1015 | * alive during long running softirqs. That's ugly as hell, but | |
1016 | * correctness is key even if we need to fix the offending softirq in | |
1017 | * the first place. | |
1018 | * | |
1019 | * Note, this is different to tick_nohz_restart. We just kick the | |
1020 | * timer and do not touch the other magic bits which need to be done | |
1021 | * when idle is left. | |
1022 | */ | |
eed3b9cf | 1023 | static void tick_nohz_kick_tick(int cpu, ktime_t now) |
fb02fbc1 | 1024 | { |
ae99286b TG |
1025 | #if 0 |
1026 | /* Switch back to 2.6.27 behaviour */ | |
1027 | ||
fb02fbc1 | 1028 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); |
eed3b9cf | 1029 | ktime_t delta; |
fb02fbc1 | 1030 | |
c4bd822e TG |
1031 | /* |
1032 | * Do not touch the tick device, when the next expiry is either | |
1033 | * already reached or less/equal than the tick period. | |
1034 | */ | |
268a3dcf | 1035 | delta = ktime_sub(hrtimer_get_expires(&ts->sched_timer), now); |
c4bd822e TG |
1036 | if (delta.tv64 <= tick_period.tv64) |
1037 | return; | |
1038 | ||
1039 | tick_nohz_restart(ts, now); | |
ae99286b | 1040 | #endif |
fb02fbc1 TG |
1041 | } |
1042 | ||
eed3b9cf MS |
1043 | static inline void tick_check_nohz(int cpu) |
1044 | { | |
1045 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1046 | ktime_t now; | |
1047 | ||
1048 | if (!ts->idle_active && !ts->tick_stopped) | |
1049 | return; | |
1050 | now = ktime_get(); | |
1051 | if (ts->idle_active) | |
1052 | tick_nohz_stop_idle(cpu, now); | |
1053 | if (ts->tick_stopped) { | |
1054 | tick_nohz_update_jiffies(now); | |
1055 | tick_nohz_kick_tick(cpu, now); | |
1056 | } | |
1057 | } | |
1058 | ||
79bf2bb3 TG |
1059 | #else |
1060 | ||
1061 | static inline void tick_nohz_switch_to_nohz(void) { } | |
eed3b9cf | 1062 | static inline void tick_check_nohz(int cpu) { } |
79bf2bb3 | 1063 | |
3451d024 | 1064 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1065 | |
719254fa TG |
1066 | /* |
1067 | * Called from irq_enter to notify about the possible interruption of idle() | |
1068 | */ | |
1069 | void tick_check_idle(int cpu) | |
1070 | { | |
fb02fbc1 | 1071 | tick_check_oneshot_broadcast(cpu); |
eed3b9cf | 1072 | tick_check_nohz(cpu); |
719254fa TG |
1073 | } |
1074 | ||
79bf2bb3 TG |
1075 | /* |
1076 | * High resolution timer specific code | |
1077 | */ | |
1078 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1079 | /* | |
4c9dc641 | 1080 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1081 | * Called with interrupts disabled. |
79bf2bb3 TG |
1082 | */ |
1083 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1084 | { | |
1085 | struct tick_sched *ts = | |
1086 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1087 | struct pt_regs *regs = get_irq_regs(); |
1088 | ktime_t now = ktime_get(); | |
d3ed7824 | 1089 | |
5bb96226 | 1090 | tick_sched_do_timer(now); |
79bf2bb3 TG |
1091 | |
1092 | /* | |
1093 | * Do not call, when we are not in irq context and have | |
1094 | * no valid regs pointer | |
1095 | */ | |
9e8f559b FW |
1096 | if (regs) |
1097 | tick_sched_handle(ts, regs); | |
79bf2bb3 | 1098 | |
79bf2bb3 TG |
1099 | hrtimer_forward(timer, now, tick_period); |
1100 | ||
1101 | return HRTIMER_RESTART; | |
1102 | } | |
1103 | ||
5307c955 MG |
1104 | static int sched_skew_tick; |
1105 | ||
62cf20b3 TG |
1106 | static int __init skew_tick(char *str) |
1107 | { | |
1108 | get_option(&str, &sched_skew_tick); | |
1109 | ||
1110 | return 0; | |
1111 | } | |
1112 | early_param("skew_tick", skew_tick); | |
1113 | ||
79bf2bb3 TG |
1114 | /** |
1115 | * tick_setup_sched_timer - setup the tick emulation timer | |
1116 | */ | |
1117 | void tick_setup_sched_timer(void) | |
1118 | { | |
1119 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1120 | ktime_t now = ktime_get(); | |
1121 | ||
1122 | /* | |
1123 | * Emulate tick processing via per-CPU hrtimers: | |
1124 | */ | |
1125 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1126 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 1127 | |
3704540b | 1128 | /* Get the next period (per cpu) */ |
cc584b21 | 1129 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1130 | |
9c3f9e28 | 1131 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
1132 | if (sched_skew_tick) { |
1133 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
1134 | do_div(offset, num_possible_cpus()); | |
1135 | offset *= smp_processor_id(); | |
1136 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1137 | } | |
1138 | ||
79bf2bb3 TG |
1139 | for (;;) { |
1140 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
5c333864 AB |
1141 | hrtimer_start_expires(&ts->sched_timer, |
1142 | HRTIMER_MODE_ABS_PINNED); | |
79bf2bb3 TG |
1143 | /* Check, if the timer was already in the past */ |
1144 | if (hrtimer_active(&ts->sched_timer)) | |
1145 | break; | |
1146 | now = ktime_get(); | |
1147 | } | |
1148 | ||
3451d024 | 1149 | #ifdef CONFIG_NO_HZ_COMMON |
29c158e8 | 1150 | if (tick_nohz_enabled) |
79bf2bb3 TG |
1151 | ts->nohz_mode = NOHZ_MODE_HIGHRES; |
1152 | #endif | |
1153 | } | |
3c4fbe5e | 1154 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1155 | |
3451d024 | 1156 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1157 | void tick_cancel_sched_timer(int cpu) |
1158 | { | |
1159 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1160 | ||
3c4fbe5e | 1161 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1162 | if (ts->sched_timer.base) |
1163 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1164 | # endif |
a7901766 | 1165 | |
4b0c0f29 | 1166 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1167 | } |
3c4fbe5e | 1168 | #endif |
79bf2bb3 TG |
1169 | |
1170 | /** | |
1171 | * Async notification about clocksource changes | |
1172 | */ | |
1173 | void tick_clock_notify(void) | |
1174 | { | |
1175 | int cpu; | |
1176 | ||
1177 | for_each_possible_cpu(cpu) | |
1178 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * Async notification about clock event changes | |
1183 | */ | |
1184 | void tick_oneshot_notify(void) | |
1185 | { | |
1186 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1187 | ||
1188 | set_bit(0, &ts->check_clocks); | |
1189 | } | |
1190 | ||
1191 | /** | |
1192 | * Check, if a change happened, which makes oneshot possible. | |
1193 | * | |
1194 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1195 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1196 | * mode, because high resolution timers are disabled (either compile | |
1197 | * or runtime). | |
1198 | */ | |
1199 | int tick_check_oneshot_change(int allow_nohz) | |
1200 | { | |
1201 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1202 | ||
1203 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1204 | return 0; | |
1205 | ||
1206 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1207 | return 0; | |
1208 | ||
cf4fc6cb | 1209 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1210 | return 0; |
1211 | ||
1212 | if (!allow_nohz) | |
1213 | return 1; | |
1214 | ||
1215 | tick_nohz_switch_to_nohz(); | |
1216 | return 0; | |
1217 | } |