]>
Commit | Line | Data |
---|---|---|
35728b82 | 1 | // SPDX-License-Identifier: GPL-2.0 |
79bf2bb3 | 2 | /* |
79bf2bb3 TG |
3 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> |
4 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar | |
5 | * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner | |
6 | * | |
7 | * No idle tick implementation for low and high resolution timers | |
8 | * | |
9 | * Started by: Thomas Gleixner and Ingo Molnar | |
79bf2bb3 TG |
10 | */ |
11 | #include <linux/cpu.h> | |
12 | #include <linux/err.h> | |
13 | #include <linux/hrtimer.h> | |
14 | #include <linux/interrupt.h> | |
15 | #include <linux/kernel_stat.h> | |
16 | #include <linux/percpu.h> | |
38b8d208 | 17 | #include <linux/nmi.h> |
79bf2bb3 | 18 | #include <linux/profile.h> |
3f07c014 | 19 | #include <linux/sched/signal.h> |
e6017571 | 20 | #include <linux/sched/clock.h> |
03441a34 | 21 | #include <linux/sched/stat.h> |
370c9135 | 22 | #include <linux/sched/nohz.h> |
896b969e | 23 | #include <linux/sched/loadavg.h> |
8083e4ad | 24 | #include <linux/module.h> |
00b42959 | 25 | #include <linux/irq_work.h> |
9014c45d | 26 | #include <linux/posix-timers.h> |
2e709338 | 27 | #include <linux/context_tracking.h> |
62cb1188 | 28 | #include <linux/mm.h> |
79bf2bb3 | 29 | |
9e203bcc DM |
30 | #include <asm/irq_regs.h> |
31 | ||
79bf2bb3 TG |
32 | #include "tick-internal.h" |
33 | ||
cb41a290 FW |
34 | #include <trace/events/timer.h> |
35 | ||
79bf2bb3 | 36 | /* |
0de7611a | 37 | * Per-CPU nohz control structure |
79bf2bb3 | 38 | */ |
c1797baf | 39 | static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); |
79bf2bb3 | 40 | |
289f480a IM |
41 | struct tick_sched *tick_get_tick_sched(int cpu) |
42 | { | |
43 | return &per_cpu(tick_cpu_sched, cpu); | |
44 | } | |
45 | ||
7809998a AB |
46 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
47 | /* | |
c398960c TG |
48 | * The time, when the last jiffy update happened. Write access must hold |
49 | * jiffies_lock and jiffies_seq. tick_nohz_next_event() needs to get a | |
50 | * consistent view of jiffies and last_jiffies_update. | |
7809998a AB |
51 | */ |
52 | static ktime_t last_jiffies_update; | |
53 | ||
79bf2bb3 TG |
54 | /* |
55 | * Must be called with interrupts disabled ! | |
56 | */ | |
57 | static void tick_do_update_jiffies64(ktime_t now) | |
58 | { | |
7a35bf2a | 59 | unsigned long ticks = 1; |
aa3b66f4 | 60 | ktime_t delta, nextp; |
79bf2bb3 | 61 | |
7a14ce1d | 62 | /* |
aa3b66f4 TG |
63 | * 64bit can do a quick check without holding jiffies lock and |
64 | * without looking at the sequence count. The smp_load_acquire() | |
372acbba TG |
65 | * pairs with the update done later in this function. |
66 | * | |
aa3b66f4 TG |
67 | * 32bit cannot do that because the store of tick_next_period |
68 | * consists of two 32bit stores and the first store could move it | |
69 | * to a random point in the future. | |
7a14ce1d | 70 | */ |
aa3b66f4 TG |
71 | if (IS_ENABLED(CONFIG_64BIT)) { |
72 | if (ktime_before(now, smp_load_acquire(&tick_next_period))) | |
73 | return; | |
74 | } else { | |
75 | unsigned int seq; | |
76 | ||
77 | /* | |
78 | * Avoid contention on jiffies_lock and protect the quick | |
79 | * check with the sequence count. | |
80 | */ | |
81 | do { | |
82 | seq = read_seqcount_begin(&jiffies_seq); | |
83 | nextp = tick_next_period; | |
84 | } while (read_seqcount_retry(&jiffies_seq, seq)); | |
85 | ||
86 | if (ktime_before(now, nextp)) | |
87 | return; | |
88 | } | |
7a14ce1d | 89 | |
aa3b66f4 | 90 | /* Quick check failed, i.e. update is required. */ |
e5d4d175 | 91 | raw_spin_lock(&jiffies_lock); |
aa3b66f4 TG |
92 | /* |
93 | * Reevaluate with the lock held. Another CPU might have done the | |
94 | * update already. | |
95 | */ | |
94ad2e3c | 96 | if (ktime_before(now, tick_next_period)) { |
e5d4d175 | 97 | raw_spin_unlock(&jiffies_lock); |
03e6bdc5 | 98 | return; |
79bf2bb3 | 99 | } |
94ad2e3c YY |
100 | |
101 | write_seqcount_begin(&jiffies_seq); | |
102 | ||
94ad2e3c | 103 | delta = ktime_sub(now, tick_next_period); |
b9965449 | 104 | if (unlikely(delta >= TICK_NSEC)) { |
94ad2e3c | 105 | /* Slow path for long idle sleep times */ |
b9965449 | 106 | s64 incr = TICK_NSEC; |
94ad2e3c | 107 | |
7a35bf2a | 108 | ticks += ktime_divns(delta, incr); |
94ad2e3c YY |
109 | |
110 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
111 | incr * ticks); | |
7a35bf2a | 112 | } else { |
b9965449 TG |
113 | last_jiffies_update = ktime_add_ns(last_jiffies_update, |
114 | TICK_NSEC); | |
94ad2e3c YY |
115 | } |
116 | ||
896b969e YY |
117 | /* Advance jiffies to complete the jiffies_seq protected job */ |
118 | jiffies_64 += ticks; | |
94ad2e3c YY |
119 | |
120 | /* | |
aa3b66f4 | 121 | * Keep the tick_next_period variable up to date. |
94ad2e3c | 122 | */ |
aa3b66f4 TG |
123 | nextp = ktime_add_ns(last_jiffies_update, TICK_NSEC); |
124 | ||
125 | if (IS_ENABLED(CONFIG_64BIT)) { | |
126 | /* | |
127 | * Pairs with smp_load_acquire() in the lockless quick | |
128 | * check above and ensures that the update to jiffies_64 is | |
129 | * not reordered vs. the store to tick_next_period, neither | |
130 | * by the compiler nor by the CPU. | |
131 | */ | |
132 | smp_store_release(&tick_next_period, nextp); | |
133 | } else { | |
134 | /* | |
135 | * A plain store is good enough on 32bit as the quick check | |
136 | * above is protected by the sequence count. | |
137 | */ | |
138 | tick_next_period = nextp; | |
139 | } | |
94ad2e3c | 140 | |
896b969e YY |
141 | /* |
142 | * Release the sequence count. calc_global_load() below is not | |
143 | * protected by it, but jiffies_lock needs to be held to prevent | |
144 | * concurrent invocations. | |
145 | */ | |
e5d4d175 | 146 | write_seqcount_end(&jiffies_seq); |
896b969e YY |
147 | |
148 | calc_global_load(); | |
149 | ||
e5d4d175 | 150 | raw_spin_unlock(&jiffies_lock); |
47a1b796 | 151 | update_wall_time(); |
79bf2bb3 TG |
152 | } |
153 | ||
154 | /* | |
155 | * Initialize and return retrieve the jiffies update. | |
156 | */ | |
157 | static ktime_t tick_init_jiffy_update(void) | |
158 | { | |
159 | ktime_t period; | |
160 | ||
e5d4d175 TG |
161 | raw_spin_lock(&jiffies_lock); |
162 | write_seqcount_begin(&jiffies_seq); | |
79bf2bb3 | 163 | /* Did we start the jiffies update yet ? */ |
2456e855 | 164 | if (last_jiffies_update == 0) |
79bf2bb3 TG |
165 | last_jiffies_update = tick_next_period; |
166 | period = last_jiffies_update; | |
e5d4d175 TG |
167 | write_seqcount_end(&jiffies_seq); |
168 | raw_spin_unlock(&jiffies_lock); | |
79bf2bb3 TG |
169 | return period; |
170 | } | |
171 | ||
ff7de620 | 172 | static void tick_sched_do_timer(struct tick_sched *ts, ktime_t now) |
5bb96226 FW |
173 | { |
174 | int cpu = smp_processor_id(); | |
175 | ||
3451d024 | 176 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
177 | /* |
178 | * Check if the do_timer duty was dropped. We don't care about | |
0de7611a IM |
179 | * concurrency: This happens only when the CPU in charge went |
180 | * into a long sleep. If two CPUs happen to assign themselves to | |
5bb96226 | 181 | * this duty, then the jiffies update is still serialized by |
9c3f9e28 | 182 | * jiffies_lock. |
08ae95f4 NP |
183 | * |
184 | * If nohz_full is enabled, this should not happen because the | |
185 | * tick_do_timer_cpu never relinquishes. | |
5bb96226 | 186 | */ |
08ae95f4 NP |
187 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) { |
188 | #ifdef CONFIG_NO_HZ_FULL | |
189 | WARN_ON(tick_nohz_full_running); | |
190 | #endif | |
5bb96226 | 191 | tick_do_timer_cpu = cpu; |
08ae95f4 | 192 | } |
5bb96226 FW |
193 | #endif |
194 | ||
195 | /* Check, if the jiffies need an update */ | |
196 | if (tick_do_timer_cpu == cpu) | |
197 | tick_do_update_jiffies64(now); | |
ff7de620 RW |
198 | |
199 | if (ts->inidle) | |
200 | ts->got_idle_tick = 1; | |
5bb96226 FW |
201 | } |
202 | ||
9e8f559b FW |
203 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
204 | { | |
3451d024 | 205 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
206 | /* |
207 | * When we are idle and the tick is stopped, we have to touch | |
208 | * the watchdog as we might not schedule for a really long | |
209 | * time. This happens on complete idle SMP systems while | |
210 | * waiting on the login prompt. We also increment the "start of | |
211 | * idle" jiffy stamp so the idle accounting adjustment we do | |
212 | * when we go busy again does not account too much ticks. | |
213 | */ | |
214 | if (ts->tick_stopped) { | |
03e0d461 | 215 | touch_softlockup_watchdog_sched(); |
9e8f559b FW |
216 | if (is_idle_task(current)) |
217 | ts->idle_jiffies++; | |
411fe24e FW |
218 | /* |
219 | * In case the current tick fired too early past its expected | |
220 | * expiration, make sure we don't bypass the next clock reprogramming | |
221 | * to the same deadline. | |
222 | */ | |
223 | ts->next_tick = 0; | |
9e8f559b | 224 | } |
94a57140 | 225 | #endif |
9e8f559b FW |
226 | update_process_times(user_mode(regs)); |
227 | profile_tick(CPU_PROFILING); | |
228 | } | |
7809998a | 229 | #endif |
9e8f559b | 230 | |
c5bfece2 | 231 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 232 | cpumask_var_t tick_nohz_full_mask; |
73867dcd | 233 | bool tick_nohz_full_running; |
ae9e557b | 234 | EXPORT_SYMBOL_GPL(tick_nohz_full_running); |
f009a7a7 | 235 | static atomic_t tick_dep_mask; |
a831881b | 236 | |
f009a7a7 | 237 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 238 | { |
f009a7a7 FW |
239 | int val = atomic_read(dep); |
240 | ||
241 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 242 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 243 | return true; |
d027d45d FW |
244 | } |
245 | ||
f009a7a7 | 246 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 247 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 248 | return true; |
d027d45d FW |
249 | } |
250 | ||
f009a7a7 | 251 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 252 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 253 | return true; |
d027d45d FW |
254 | } |
255 | ||
f009a7a7 | 256 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 257 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
258 | return true; |
259 | } | |
260 | ||
01b4c399 FW |
261 | if (val & TICK_DEP_MASK_RCU) { |
262 | trace_tick_stop(0, TICK_DEP_MASK_RCU); | |
263 | return true; | |
264 | } | |
265 | ||
f009a7a7 | 266 | return false; |
d027d45d FW |
267 | } |
268 | ||
57ccdf44 | 269 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d | 270 | { |
ebf3adba | 271 | lockdep_assert_irqs_disabled(); |
9014c45d | 272 | |
57ccdf44 WL |
273 | if (unlikely(!cpu_online(cpu))) |
274 | return false; | |
275 | ||
f009a7a7 | 276 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 277 | return false; |
d027d45d | 278 | |
f009a7a7 | 279 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 280 | return false; |
d027d45d | 281 | |
f009a7a7 | 282 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 283 | return false; |
d027d45d | 284 | |
f009a7a7 | 285 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 286 | return false; |
d027d45d | 287 | |
9014c45d FW |
288 | return true; |
289 | } | |
290 | ||
d027d45d | 291 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 292 | { |
73738a95 | 293 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
294 | } |
295 | ||
7a9f50a0 PZ |
296 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = |
297 | IRQ_WORK_INIT_HARD(nohz_full_kick_func); | |
76c24fb0 | 298 | |
40bea039 FW |
299 | /* |
300 | * Kick this CPU if it's full dynticks in order to force it to | |
301 | * re-evaluate its dependency on the tick and restart it if necessary. | |
302 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
303 | * is NMI safe. | |
304 | */ | |
555e0c1e | 305 | static void tick_nohz_full_kick(void) |
40bea039 FW |
306 | { |
307 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
308 | return; | |
309 | ||
56e4dea8 | 310 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
311 | } |
312 | ||
76c24fb0 | 313 | /* |
3d36aebc | 314 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
315 | * re-evaluate its dependency on the tick and restart it if necessary. |
316 | */ | |
3d36aebc | 317 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 318 | { |
3d36aebc FW |
319 | if (!tick_nohz_full_cpu(cpu)) |
320 | return; | |
321 | ||
322 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
323 | } |
324 | ||
76c24fb0 FW |
325 | /* |
326 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
327 | * their dependency on the tick and restart it if necessary. | |
328 | */ | |
b7878300 | 329 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 330 | { |
8537bb95 FW |
331 | int cpu; |
332 | ||
73867dcd | 333 | if (!tick_nohz_full_running) |
76c24fb0 FW |
334 | return; |
335 | ||
336 | preempt_disable(); | |
8537bb95 FW |
337 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
338 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
339 | preempt_enable(); |
340 | } | |
341 | ||
f009a7a7 | 342 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
343 | enum tick_dep_bits bit) |
344 | { | |
f009a7a7 | 345 | int prev; |
d027d45d | 346 | |
a1cc5bcf | 347 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
348 | if (!prev) |
349 | tick_nohz_full_kick_all(); | |
350 | } | |
351 | ||
352 | /* | |
353 | * Set a global tick dependency. Used by perf events that rely on freq and | |
354 | * by unstable clock. | |
355 | */ | |
356 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
357 | { | |
358 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
359 | } | |
360 | ||
361 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
362 | { | |
f009a7a7 | 363 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
364 | } |
365 | ||
366 | /* | |
367 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
368 | * manage events throttling. | |
369 | */ | |
370 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
371 | { | |
f009a7a7 | 372 | int prev; |
d027d45d FW |
373 | struct tick_sched *ts; |
374 | ||
375 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
376 | ||
a1cc5bcf | 377 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
378 | if (!prev) { |
379 | preempt_disable(); | |
380 | /* Perf needs local kick that is NMI safe */ | |
381 | if (cpu == smp_processor_id()) { | |
382 | tick_nohz_full_kick(); | |
383 | } else { | |
384 | /* Remote irq work not NMI-safe */ | |
385 | if (!WARN_ON_ONCE(in_nmi())) | |
386 | tick_nohz_full_kick_cpu(cpu); | |
387 | } | |
388 | preempt_enable(); | |
389 | } | |
390 | } | |
01b4c399 | 391 | EXPORT_SYMBOL_GPL(tick_nohz_dep_set_cpu); |
d027d45d FW |
392 | |
393 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
394 | { | |
395 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
396 | ||
f009a7a7 | 397 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d | 398 | } |
01b4c399 | 399 | EXPORT_SYMBOL_GPL(tick_nohz_dep_clear_cpu); |
d027d45d FW |
400 | |
401 | /* | |
3c8920e2 FW |
402 | * Set a per-task tick dependency. RCU need this. Also posix CPU timers |
403 | * in order to elapse per task timers. | |
d027d45d FW |
404 | */ |
405 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
406 | { | |
3c8920e2 FW |
407 | if (!atomic_fetch_or(BIT(bit), &tsk->tick_dep_mask)) { |
408 | if (tsk == current) { | |
409 | preempt_disable(); | |
410 | tick_nohz_full_kick(); | |
411 | preempt_enable(); | |
412 | } else { | |
413 | /* | |
414 | * Some future tick_nohz_full_kick_task() | |
415 | * should optimize this. | |
416 | */ | |
417 | tick_nohz_full_kick_all(); | |
418 | } | |
419 | } | |
d027d45d | 420 | } |
ae9e557b | 421 | EXPORT_SYMBOL_GPL(tick_nohz_dep_set_task); |
d027d45d FW |
422 | |
423 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
424 | { | |
f009a7a7 | 425 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d | 426 | } |
ae9e557b | 427 | EXPORT_SYMBOL_GPL(tick_nohz_dep_clear_task); |
d027d45d FW |
428 | |
429 | /* | |
430 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
431 | * per process timers. | |
432 | */ | |
433 | void tick_nohz_dep_set_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
434 | { | |
435 | tick_nohz_dep_set_all(&sig->tick_dep_mask, bit); | |
436 | } | |
437 | ||
438 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
439 | { | |
f009a7a7 | 440 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
441 | } |
442 | ||
99e5ada9 FW |
443 | /* |
444 | * Re-evaluate the need for the tick as we switch the current task. | |
445 | * It might need the tick due to per task/process properties: | |
0de7611a | 446 | * perf events, posix CPU timers, ... |
99e5ada9 | 447 | */ |
de734f89 | 448 | void __tick_nohz_task_switch(void) |
99e5ada9 FW |
449 | { |
450 | unsigned long flags; | |
d027d45d | 451 | struct tick_sched *ts; |
99e5ada9 | 452 | |
99e5ada9 FW |
453 | local_irq_save(flags); |
454 | ||
6296ace4 LZ |
455 | if (!tick_nohz_full_cpu(smp_processor_id())) |
456 | goto out; | |
457 | ||
d027d45d | 458 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 459 | |
d027d45d | 460 | if (ts->tick_stopped) { |
f009a7a7 FW |
461 | if (atomic_read(¤t->tick_dep_mask) || |
462 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
463 | tick_nohz_full_kick(); |
464 | } | |
6296ace4 | 465 | out: |
99e5ada9 FW |
466 | local_irq_restore(flags); |
467 | } | |
468 | ||
6f1982fe FW |
469 | /* Get the boot-time nohz CPU list from the kernel parameters. */ |
470 | void __init tick_nohz_full_setup(cpumask_var_t cpumask) | |
a831881b | 471 | { |
73867dcd | 472 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
6f1982fe | 473 | cpumask_copy(tick_nohz_full_mask, cpumask); |
73867dcd | 474 | tick_nohz_full_running = true; |
a831881b | 475 | } |
ae9e557b | 476 | EXPORT_SYMBOL_GPL(tick_nohz_full_setup); |
a831881b | 477 | |
31eff243 | 478 | static int tick_nohz_cpu_down(unsigned int cpu) |
a382bf93 | 479 | { |
31eff243 | 480 | /* |
08ae95f4 NP |
481 | * The tick_do_timer_cpu CPU handles housekeeping duty (unbound |
482 | * timers, workqueues, timekeeping, ...) on behalf of full dynticks | |
31eff243 SAS |
483 | * CPUs. It must remain online when nohz full is enabled. |
484 | */ | |
485 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) | |
486 | return -EBUSY; | |
487 | return 0; | |
a382bf93 FW |
488 | } |
489 | ||
d1e43fa5 | 490 | void __init tick_nohz_init(void) |
a831881b | 491 | { |
31eff243 | 492 | int cpu, ret; |
d1e43fa5 | 493 | |
a7c8655b PM |
494 | if (!tick_nohz_full_running) |
495 | return; | |
d1e43fa5 | 496 | |
9b01f5bf FW |
497 | /* |
498 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
499 | * locking contexts. But then we need irq work to raise its own | |
500 | * interrupts to avoid circular dependency on the tick | |
501 | */ | |
502 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 503 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf | 504 | cpumask_clear(tick_nohz_full_mask); |
9b01f5bf FW |
505 | tick_nohz_full_running = false; |
506 | return; | |
507 | } | |
508 | ||
08ae95f4 NP |
509 | if (IS_ENABLED(CONFIG_PM_SLEEP_SMP) && |
510 | !IS_ENABLED(CONFIG_PM_SLEEP_SMP_NONZERO_CPU)) { | |
511 | cpu = smp_processor_id(); | |
4327b15f | 512 | |
08ae95f4 NP |
513 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { |
514 | pr_warn("NO_HZ: Clearing %d from nohz_full range " | |
515 | "for timekeeping\n", cpu); | |
516 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); | |
517 | } | |
4327b15f FW |
518 | } |
519 | ||
73867dcd | 520 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
521 | context_tracking_cpu_set(cpu); |
522 | ||
31eff243 SAS |
523 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
524 | "kernel/nohz:predown", NULL, | |
525 | tick_nohz_cpu_down); | |
526 | WARN_ON(ret < 0); | |
ffda22c1 TH |
527 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
528 | cpumask_pr_args(tick_nohz_full_mask)); | |
a831881b | 529 | } |
a831881b FW |
530 | #endif |
531 | ||
79bf2bb3 TG |
532 | /* |
533 | * NOHZ - aka dynamic tick functionality | |
534 | */ | |
3451d024 | 535 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
536 | /* |
537 | * NO HZ enabled ? | |
538 | */ | |
4cc7ecb7 | 539 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 540 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
541 | /* |
542 | * Enable / Disable tickless mode | |
543 | */ | |
544 | static int __init setup_tick_nohz(char *str) | |
545 | { | |
4cc7ecb7 | 546 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
547 | } |
548 | ||
549 | __setup("nohz=", setup_tick_nohz); | |
550 | ||
a3642983 | 551 | bool tick_nohz_tick_stopped(void) |
c1797baf | 552 | { |
2bc629a6 FW |
553 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
554 | ||
555 | return ts->tick_stopped; | |
c1797baf TG |
556 | } |
557 | ||
22ab8bc0 FW |
558 | bool tick_nohz_tick_stopped_cpu(int cpu) |
559 | { | |
560 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
561 | ||
562 | return ts->tick_stopped; | |
563 | } | |
564 | ||
79bf2bb3 TG |
565 | /** |
566 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
567 | * | |
568 | * Called from interrupt entry when the CPU was idle | |
569 | * | |
570 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
571 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
572 | * value. We do this unconditionally on any CPU, as we don't know whether the |
573 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 574 | */ |
eed3b9cf | 575 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 576 | { |
79bf2bb3 | 577 | unsigned long flags; |
79bf2bb3 | 578 | |
e8fcaa5c | 579 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
580 | |
581 | local_irq_save(flags); | |
582 | tick_do_update_jiffies64(now); | |
583 | local_irq_restore(flags); | |
02ff3755 | 584 | |
03e0d461 | 585 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
586 | } |
587 | ||
595aac48 | 588 | /* |
0de7611a | 589 | * Updates the per-CPU time idle statistics counters |
595aac48 | 590 | */ |
8d63bf94 | 591 | static void |
8c215bd3 | 592 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 593 | { |
eed3b9cf | 594 | ktime_t delta; |
6378ddb5 | 595 | |
595aac48 AV |
596 | if (ts->idle_active) { |
597 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 598 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 599 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
600 | else |
601 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 602 | ts->idle_entrytime = now; |
595aac48 | 603 | } |
8d63bf94 | 604 | |
e0e37c20 | 605 | if (last_update_time) |
8d63bf94 AV |
606 | *last_update_time = ktime_to_us(now); |
607 | ||
595aac48 AV |
608 | } |
609 | ||
e8fcaa5c | 610 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 611 | { |
e8fcaa5c | 612 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 613 | ts->idle_active = 0; |
56c7426b | 614 | |
ac1e843f | 615 | sched_clock_idle_wakeup_event(); |
6378ddb5 VP |
616 | } |
617 | ||
0e776768 | 618 | static void tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 619 | { |
0e776768 | 620 | ts->idle_entrytime = ktime_get(); |
6378ddb5 | 621 | ts->idle_active = 1; |
56c7426b | 622 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
623 | } |
624 | ||
b1f724c3 | 625 | /** |
0de7611a | 626 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 627 | * @cpu: CPU number to query |
09a1d34f MH |
628 | * @last_update_time: variable to store update time in. Do not update |
629 | * counters if NULL. | |
b1f724c3 | 630 | * |
6168f8ed | 631 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 632 | * CPU, in microseconds. |
b1f724c3 AV |
633 | * |
634 | * This time is measured via accounting rather than sampling, | |
635 | * and is as accurate as ktime_get() is. | |
636 | * | |
637 | * This function returns -1 if NOHZ is not enabled. | |
638 | */ | |
6378ddb5 VP |
639 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
640 | { | |
641 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 642 | ktime_t now, idle; |
6378ddb5 | 643 | |
d689fe22 | 644 | if (!tick_nohz_active) |
8083e4ad | 645 | return -1; |
646 | ||
09a1d34f MH |
647 | now = ktime_get(); |
648 | if (last_update_time) { | |
649 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
650 | idle = ts->idle_sleeptime; | |
651 | } else { | |
652 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
653 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
654 | ||
655 | idle = ktime_add(ts->idle_sleeptime, delta); | |
656 | } else { | |
657 | idle = ts->idle_sleeptime; | |
658 | } | |
659 | } | |
660 | ||
661 | return ktime_to_us(idle); | |
8083e4ad | 662 | |
6378ddb5 | 663 | } |
8083e4ad | 664 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 665 | |
6beea0cd | 666 | /** |
0de7611a | 667 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 668 | * @cpu: CPU number to query |
09a1d34f MH |
669 | * @last_update_time: variable to store update time in. Do not update |
670 | * counters if NULL. | |
0224cf4c | 671 | * |
6168f8ed | 672 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
673 | * CPU, in microseconds. |
674 | * | |
675 | * This time is measured via accounting rather than sampling, | |
676 | * and is as accurate as ktime_get() is. | |
677 | * | |
678 | * This function returns -1 if NOHZ is not enabled. | |
679 | */ | |
680 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
681 | { | |
682 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 683 | ktime_t now, iowait; |
0224cf4c | 684 | |
d689fe22 | 685 | if (!tick_nohz_active) |
0224cf4c AV |
686 | return -1; |
687 | ||
09a1d34f MH |
688 | now = ktime_get(); |
689 | if (last_update_time) { | |
690 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
691 | iowait = ts->iowait_sleeptime; | |
692 | } else { | |
693 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
694 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 695 | |
09a1d34f MH |
696 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
697 | } else { | |
698 | iowait = ts->iowait_sleeptime; | |
699 | } | |
700 | } | |
0224cf4c | 701 | |
09a1d34f | 702 | return ktime_to_us(iowait); |
0224cf4c AV |
703 | } |
704 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
705 | ||
0ff53d09 TG |
706 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
707 | { | |
708 | hrtimer_cancel(&ts->sched_timer); | |
709 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
710 | ||
711 | /* Forward the time to expire in the future */ | |
b9965449 | 712 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
0ff53d09 | 713 | |
902a9f9c SAS |
714 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
715 | hrtimer_start_expires(&ts->sched_timer, | |
716 | HRTIMER_MODE_ABS_PINNED_HARD); | |
717 | } else { | |
0ff53d09 | 718 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
902a9f9c | 719 | } |
411fe24e FW |
720 | |
721 | /* | |
722 | * Reset to make sure next tick stop doesn't get fooled by past | |
723 | * cached clock deadline. | |
724 | */ | |
725 | ts->next_tick = 0; | |
0ff53d09 TG |
726 | } |
727 | ||
5d62c183 TG |
728 | static inline bool local_timer_softirq_pending(void) |
729 | { | |
80d20d35 | 730 | return local_softirq_pending() & BIT(TIMER_SOFTIRQ); |
5d62c183 TG |
731 | } |
732 | ||
23a8d888 | 733 | static ktime_t tick_nohz_next_event(struct tick_sched *ts, int cpu) |
79bf2bb3 | 734 | { |
c1ad348b | 735 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
e1e41b6c RV |
736 | unsigned long basejiff; |
737 | unsigned int seq; | |
855a0fc3 | 738 | |
79bf2bb3 TG |
739 | /* Read jiffies and the time when jiffies were updated last */ |
740 | do { | |
e5d4d175 | 741 | seq = read_seqcount_begin(&jiffies_seq); |
2456e855 | 742 | basemono = last_jiffies_update; |
c1ad348b | 743 | basejiff = jiffies; |
e5d4d175 | 744 | } while (read_seqcount_retry(&jiffies_seq, seq)); |
c1ad348b | 745 | ts->last_jiffies = basejiff; |
23a8d888 | 746 | ts->timer_expires_base = basemono; |
79bf2bb3 | 747 | |
5d62c183 TG |
748 | /* |
749 | * Keep the periodic tick, when RCU, architecture or irq_work | |
750 | * requests it. | |
751 | * Aside of that check whether the local timer softirq is | |
752 | * pending. If so its a bad idea to call get_next_timer_interrupt() | |
753 | * because there is an already expired timer, so it will request | |
754 | * immeditate expiry, which rearms the hardware timer with a | |
755 | * minimal delta which brings us back to this place | |
756 | * immediately. Lather, rinse and repeat... | |
757 | */ | |
758 | if (rcu_needs_cpu(basemono, &next_rcu) || arch_needs_cpu() || | |
759 | irq_work_needs_cpu() || local_timer_softirq_pending()) { | |
c1ad348b | 760 | next_tick = basemono + TICK_NSEC; |
3c5d92a0 | 761 | } else { |
c1ad348b TG |
762 | /* |
763 | * Get the next pending timer. If high resolution | |
764 | * timers are enabled this only takes the timer wheel | |
765 | * timers into account. If high resolution timers are | |
766 | * disabled this also looks at the next expiring | |
767 | * hrtimer. | |
768 | */ | |
769 | next_tmr = get_next_timer_interrupt(basejiff, basemono); | |
770 | ts->next_timer = next_tmr; | |
771 | /* Take the next rcu event into account */ | |
772 | next_tick = next_rcu < next_tmr ? next_rcu : next_tmr; | |
3c5d92a0 | 773 | } |
47aa8b6c | 774 | |
c1ad348b TG |
775 | /* |
776 | * If the tick is due in the next period, keep it ticking or | |
82bbe34b | 777 | * force prod the timer. |
c1ad348b TG |
778 | */ |
779 | delta = next_tick - basemono; | |
780 | if (delta <= (u64)TICK_NSEC) { | |
a683f390 TG |
781 | /* |
782 | * Tell the timer code that the base is not idle, i.e. undo | |
783 | * the effect of get_next_timer_interrupt(): | |
784 | */ | |
785 | timer_clear_idle(); | |
82bbe34b PZ |
786 | /* |
787 | * We've not stopped the tick yet, and there's a timer in the | |
788 | * next period, so no point in stopping it either, bail. | |
789 | */ | |
f99973e1 | 790 | if (!ts->tick_stopped) { |
23a8d888 | 791 | ts->timer_expires = 0; |
157d29e1 TG |
792 | goto out; |
793 | } | |
794 | } | |
795 | ||
23a8d888 RW |
796 | /* |
797 | * If this CPU is the one which had the do_timer() duty last, we limit | |
798 | * the sleep time to the timekeeping max_deferment value. | |
799 | * Otherwise we can sleep as long as we want. | |
800 | */ | |
801 | delta = timekeeping_max_deferment(); | |
802 | if (cpu != tick_do_timer_cpu && | |
803 | (tick_do_timer_cpu != TICK_DO_TIMER_NONE || !ts->do_timer_last)) | |
804 | delta = KTIME_MAX; | |
805 | ||
806 | /* Calculate the next expiry time */ | |
807 | if (delta < (KTIME_MAX - basemono)) | |
808 | expires = basemono + delta; | |
809 | else | |
810 | expires = KTIME_MAX; | |
811 | ||
812 | ts->timer_expires = min_t(u64, expires, next_tick); | |
813 | ||
814 | out: | |
815 | return ts->timer_expires; | |
816 | } | |
817 | ||
818 | static void tick_nohz_stop_tick(struct tick_sched *ts, int cpu) | |
819 | { | |
820 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); | |
821 | u64 basemono = ts->timer_expires_base; | |
822 | u64 expires = ts->timer_expires; | |
823 | ktime_t tick = expires; | |
824 | ||
825 | /* Make sure we won't be trying to stop it twice in a row. */ | |
826 | ts->timer_expires_base = 0; | |
827 | ||
79bf2bb3 | 828 | /* |
0de7611a IM |
829 | * If this CPU is the one which updates jiffies, then give up |
830 | * the assignment and let it be taken by the CPU which runs | |
831 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
832 | * don't drop this here the jiffies might be stale and |
833 | * do_timer() never invoked. Keep track of the fact that it | |
23a8d888 | 834 | * was the one which had the do_timer() duty last. |
79bf2bb3 | 835 | */ |
157d29e1 TG |
836 | if (cpu == tick_do_timer_cpu) { |
837 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
838 | ts->do_timer_last = 1; | |
839 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
157d29e1 | 840 | ts->do_timer_last = 0; |
157d29e1 | 841 | } |
27185016 | 842 | |
157d29e1 | 843 | /* Skip reprogram of event if its not changed */ |
411fe24e FW |
844 | if (ts->tick_stopped && (expires == ts->next_tick)) { |
845 | /* Sanity check: make sure clockevent is actually programmed */ | |
d4af6d93 | 846 | if (tick == KTIME_MAX || ts->next_tick == hrtimer_get_expires(&ts->sched_timer)) |
23a8d888 | 847 | return; |
411fe24e FW |
848 | |
849 | WARN_ON_ONCE(1); | |
850 | printk_once("basemono: %llu ts->next_tick: %llu dev->next_event: %llu timer->active: %d timer->expires: %llu\n", | |
851 | basemono, ts->next_tick, dev->next_event, | |
852 | hrtimer_active(&ts->sched_timer), hrtimer_get_expires(&ts->sched_timer)); | |
ce6cf9a1 | 853 | } |
84bf1bcc | 854 | |
157d29e1 TG |
855 | /* |
856 | * nohz_stop_sched_tick can be called several times before | |
857 | * the nohz_restart_sched_tick is called. This happens when | |
858 | * interrupts arrive which do not cause a reschedule. In the | |
859 | * first call we save the current tick time, so we can restart | |
860 | * the scheduler tick in nohz_restart_sched_tick. | |
861 | */ | |
862 | if (!ts->tick_stopped) { | |
3c85d6db | 863 | calc_load_nohz_start(); |
62cb1188 | 864 | quiet_vmstat(); |
d3ed7824 | 865 | |
157d29e1 TG |
866 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
867 | ts->tick_stopped = 1; | |
e6e6cc22 | 868 | trace_tick_stop(1, TICK_DEP_MASK_NONE); |
157d29e1 | 869 | } |
eaad084b | 870 | |
411fe24e FW |
871 | ts->next_tick = tick; |
872 | ||
157d29e1 | 873 | /* |
c1ad348b TG |
874 | * If the expiration time == KTIME_MAX, then we simply stop |
875 | * the tick timer. | |
157d29e1 | 876 | */ |
c1ad348b | 877 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
878 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
879 | hrtimer_cancel(&ts->sched_timer); | |
23a8d888 | 880 | return; |
79bf2bb3 | 881 | } |
0ff53d09 | 882 | |
1f71addd | 883 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
902a9f9c SAS |
884 | hrtimer_start(&ts->sched_timer, tick, |
885 | HRTIMER_MODE_ABS_PINNED_HARD); | |
1f71addd TG |
886 | } else { |
887 | hrtimer_set_expires(&ts->sched_timer, tick); | |
c1ad348b | 888 | tick_program_event(tick, 1); |
1f71addd | 889 | } |
280f0677 FW |
890 | } |
891 | ||
23a8d888 RW |
892 | static void tick_nohz_retain_tick(struct tick_sched *ts) |
893 | { | |
894 | ts->timer_expires_base = 0; | |
895 | } | |
896 | ||
897 | #ifdef CONFIG_NO_HZ_FULL | |
898 | static void tick_nohz_stop_sched_tick(struct tick_sched *ts, int cpu) | |
899 | { | |
900 | if (tick_nohz_next_event(ts, cpu)) | |
901 | tick_nohz_stop_tick(ts, cpu); | |
902 | else | |
903 | tick_nohz_retain_tick(ts); | |
904 | } | |
905 | #endif /* CONFIG_NO_HZ_FULL */ | |
906 | ||
1f41906a | 907 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
908 | { |
909 | /* Update jiffies first */ | |
910 | tick_do_update_jiffies64(now); | |
59d2c7ca | 911 | |
a683f390 TG |
912 | /* |
913 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
914 | * the clock forward checks in the enqueue path: | |
915 | */ | |
916 | timer_clear_idle(); | |
917 | ||
3c85d6db | 918 | calc_load_nohz_stop(); |
03e0d461 | 919 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
920 | /* |
921 | * Cancel the scheduled timer and restore the tick | |
922 | */ | |
923 | ts->tick_stopped = 0; | |
924 | ts->idle_exittime = now; | |
925 | ||
926 | tick_nohz_restart(ts, now); | |
927 | } | |
73738a95 FW |
928 | |
929 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
5811d996 FW |
930 | { |
931 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 932 | int cpu = smp_processor_id(); |
5811d996 | 933 | |
59449359 | 934 | if (!tick_nohz_full_cpu(cpu)) |
e9a2eb40 | 935 | return; |
5811d996 | 936 | |
e9a2eb40 AS |
937 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
938 | return; | |
5811d996 | 939 | |
57ccdf44 | 940 | if (can_stop_full_tick(cpu, ts)) |
23a8d888 | 941 | tick_nohz_stop_sched_tick(ts, cpu); |
73738a95 | 942 | else if (ts->tick_stopped) |
1f41906a | 943 | tick_nohz_restart_sched_tick(ts, ktime_get()); |
5811d996 FW |
944 | #endif |
945 | } | |
946 | ||
5b39939a FW |
947 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
948 | { | |
949 | /* | |
0de7611a | 950 | * If this CPU is offline and it is the one which updates |
5b39939a | 951 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 952 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
953 | * this here the jiffies might be stale and do_timer() never |
954 | * invoked. | |
955 | */ | |
956 | if (unlikely(!cpu_online(cpu))) { | |
957 | if (cpu == tick_do_timer_cpu) | |
958 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
411fe24e FW |
959 | /* |
960 | * Make sure the CPU doesn't get fooled by obsolete tick | |
961 | * deadline if it comes back online later. | |
962 | */ | |
963 | ts->next_tick = 0; | |
f7ea0fd6 | 964 | return false; |
5b39939a FW |
965 | } |
966 | ||
23a8d888 | 967 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) |
5b39939a FW |
968 | return false; |
969 | ||
970 | if (need_resched()) | |
971 | return false; | |
972 | ||
d59e0ba1 | 973 | if (unlikely(local_softirq_pending())) { |
5b39939a FW |
974 | static int ratelimit; |
975 | ||
803b0eba PM |
976 | if (ratelimit < 10 && |
977 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
bca37119 | 978 | pr_warn("NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #%02x!!!\n", |
cfea7d7e | 979 | (unsigned int) local_softirq_pending()); |
5b39939a FW |
980 | ratelimit++; |
981 | } | |
982 | return false; | |
983 | } | |
984 | ||
460775df | 985 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
986 | /* |
987 | * Keep the tick alive to guarantee timekeeping progression | |
988 | * if there are full dynticks CPUs around | |
989 | */ | |
990 | if (tick_do_timer_cpu == cpu) | |
991 | return false; | |
08ae95f4 NP |
992 | |
993 | /* Should not happen for nohz-full */ | |
994 | if (WARN_ON_ONCE(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) | |
a382bf93 FW |
995 | return false; |
996 | } | |
997 | ||
5b39939a FW |
998 | return true; |
999 | } | |
1000 | ||
0e776768 | 1001 | static void __tick_nohz_idle_stop_tick(struct tick_sched *ts) |
19f5f736 | 1002 | { |
0e776768 | 1003 | ktime_t expires; |
5b39939a | 1004 | int cpu = smp_processor_id(); |
19f5f736 | 1005 | |
554c8aa8 RW |
1006 | /* |
1007 | * If tick_nohz_get_sleep_length() ran tick_nohz_next_event(), the | |
1008 | * tick timer expiration time is known already. | |
1009 | */ | |
1010 | if (ts->timer_expires_base) | |
1011 | expires = ts->timer_expires; | |
1012 | else if (can_stop_idle_tick(cpu, ts)) | |
1013 | expires = tick_nohz_next_event(ts, cpu); | |
1014 | else | |
1015 | return; | |
23a8d888 RW |
1016 | |
1017 | ts->idle_calls++; | |
08d07259 | 1018 | |
23a8d888 | 1019 | if (expires > 0LL) { |
5b39939a FW |
1020 | int was_stopped = ts->tick_stopped; |
1021 | ||
23a8d888 | 1022 | tick_nohz_stop_tick(ts, cpu); |
84bf1bcc | 1023 | |
23a8d888 RW |
1024 | ts->idle_sleeps++; |
1025 | ts->idle_expires = expires; | |
5b39939a | 1026 | |
a0db971e | 1027 | if (!was_stopped && ts->tick_stopped) { |
5b39939a | 1028 | ts->idle_jiffies = ts->last_jiffies; |
a0db971e FW |
1029 | nohz_balance_enter_idle(cpu); |
1030 | } | |
23a8d888 RW |
1031 | } else { |
1032 | tick_nohz_retain_tick(ts); | |
5b39939a | 1033 | } |
280f0677 FW |
1034 | } |
1035 | ||
1036 | /** | |
0e776768 | 1037 | * tick_nohz_idle_stop_tick - stop the idle tick from the idle task |
280f0677 FW |
1038 | * |
1039 | * When the next event is more than a tick into the future, stop the idle tick | |
0e776768 RW |
1040 | */ |
1041 | void tick_nohz_idle_stop_tick(void) | |
1042 | { | |
1043 | __tick_nohz_idle_stop_tick(this_cpu_ptr(&tick_cpu_sched)); | |
1044 | } | |
1045 | ||
554c8aa8 RW |
1046 | void tick_nohz_idle_retain_tick(void) |
1047 | { | |
1048 | tick_nohz_retain_tick(this_cpu_ptr(&tick_cpu_sched)); | |
1049 | /* | |
1050 | * Undo the effect of get_next_timer_interrupt() called from | |
1051 | * tick_nohz_next_event(). | |
1052 | */ | |
1053 | timer_clear_idle(); | |
1054 | } | |
1055 | ||
0e776768 RW |
1056 | /** |
1057 | * tick_nohz_idle_enter - prepare for entering idle on the current CPU | |
2bbb6817 | 1058 | * |
0e776768 | 1059 | * Called when we start the idle loop. |
280f0677 | 1060 | */ |
1268fbc7 | 1061 | void tick_nohz_idle_enter(void) |
280f0677 FW |
1062 | { |
1063 | struct tick_sched *ts; | |
1064 | ||
ebf3adba | 1065 | lockdep_assert_irqs_enabled(); |
0db49b72 | 1066 | |
1268fbc7 FW |
1067 | local_irq_disable(); |
1068 | ||
22127e93 | 1069 | ts = this_cpu_ptr(&tick_cpu_sched); |
23a8d888 RW |
1070 | |
1071 | WARN_ON_ONCE(ts->timer_expires_base); | |
1072 | ||
280f0677 | 1073 | ts->inidle = 1; |
0e776768 | 1074 | tick_nohz_start_idle(ts); |
1268fbc7 FW |
1075 | |
1076 | local_irq_enable(); | |
280f0677 FW |
1077 | } |
1078 | ||
1079 | /** | |
1080 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
1081 | * | |
1082 | * When an interrupt fires while we are idle and it doesn't cause | |
1083 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
1084 | * an RCU callback, etc... | |
1085 | * So we need to re-calculate and reprogram the next tick event. | |
1086 | */ | |
1087 | void tick_nohz_irq_exit(void) | |
1088 | { | |
22127e93 | 1089 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 1090 | |
14851912 | 1091 | if (ts->inidle) |
0e776768 | 1092 | tick_nohz_start_idle(ts); |
14851912 | 1093 | else |
73738a95 | 1094 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
1095 | } |
1096 | ||
4f86d3a8 | 1097 | /** |
45f1ff59 RW |
1098 | * tick_nohz_idle_got_tick - Check whether or not the tick handler has run |
1099 | */ | |
1100 | bool tick_nohz_idle_got_tick(void) | |
1101 | { | |
1102 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1103 | ||
2bc629a6 FW |
1104 | if (ts->got_idle_tick) { |
1105 | ts->got_idle_tick = 0; | |
45f1ff59 RW |
1106 | return true; |
1107 | } | |
1108 | return false; | |
1109 | } | |
1110 | ||
6f9b83ac UH |
1111 | /** |
1112 | * tick_nohz_get_next_hrtimer - return the next expiration time for the hrtimer | |
1113 | * or the tick, whatever that expires first. Note that, if the tick has been | |
1114 | * stopped, it returns the next hrtimer. | |
1115 | * | |
1116 | * Called from power state control code with interrupts disabled | |
1117 | */ | |
1118 | ktime_t tick_nohz_get_next_hrtimer(void) | |
1119 | { | |
1120 | return __this_cpu_read(tick_cpu_device.evtdev)->next_event; | |
1121 | } | |
1122 | ||
4f86d3a8 | 1123 | /** |
554c8aa8 | 1124 | * tick_nohz_get_sleep_length - return the expected length of the current sleep |
296bb1e5 | 1125 | * @delta_next: duration until the next event if the tick cannot be stopped |
4f86d3a8 LB |
1126 | * |
1127 | * Called from power state control code with interrupts disabled | |
1128 | */ | |
296bb1e5 | 1129 | ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next) |
4f86d3a8 | 1130 | { |
554c8aa8 | 1131 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
22127e93 | 1132 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
554c8aa8 RW |
1133 | int cpu = smp_processor_id(); |
1134 | /* | |
1135 | * The idle entry time is expected to be a sufficient approximation of | |
1136 | * the current time at this point. | |
1137 | */ | |
1138 | ktime_t now = ts->idle_entrytime; | |
1139 | ktime_t next_event; | |
1140 | ||
1141 | WARN_ON_ONCE(!ts->inidle); | |
1142 | ||
296bb1e5 RW |
1143 | *delta_next = ktime_sub(dev->next_event, now); |
1144 | ||
554c8aa8 | 1145 | if (!can_stop_idle_tick(cpu, ts)) |
296bb1e5 | 1146 | return *delta_next; |
554c8aa8 RW |
1147 | |
1148 | next_event = tick_nohz_next_event(ts, cpu); | |
1149 | if (!next_event) | |
296bb1e5 | 1150 | return *delta_next; |
554c8aa8 RW |
1151 | |
1152 | /* | |
1153 | * If the next highres timer to expire is earlier than next_event, the | |
1154 | * idle governor needs to know that. | |
1155 | */ | |
1156 | next_event = min_t(u64, next_event, | |
1157 | hrtimer_next_event_without(&ts->sched_timer)); | |
4f86d3a8 | 1158 | |
554c8aa8 | 1159 | return ktime_sub(next_event, now); |
4f86d3a8 LB |
1160 | } |
1161 | ||
466a2b42 JF |
1162 | /** |
1163 | * tick_nohz_get_idle_calls_cpu - return the current idle calls counter value | |
1164 | * for a particular CPU. | |
1165 | * | |
1166 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1167 | */ | |
1168 | unsigned long tick_nohz_get_idle_calls_cpu(int cpu) | |
1169 | { | |
1170 | struct tick_sched *ts = tick_get_tick_sched(cpu); | |
1171 | ||
1172 | return ts->idle_calls; | |
1173 | } | |
1174 | ||
b7eaf1aa RW |
1175 | /** |
1176 | * tick_nohz_get_idle_calls - return the current idle calls counter value | |
1177 | * | |
1178 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1179 | */ | |
1180 | unsigned long tick_nohz_get_idle_calls(void) | |
1181 | { | |
1182 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1183 | ||
1184 | return ts->idle_calls; | |
1185 | } | |
1186 | ||
2ac0d98f FW |
1187 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) |
1188 | { | |
3f4724ea | 1189 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 1190 | unsigned long ticks; |
3f4724ea | 1191 | |
e44fcb4b | 1192 | if (vtime_accounting_enabled_this_cpu()) |
3f4724ea | 1193 | return; |
79bf2bb3 TG |
1194 | /* |
1195 | * We stopped the tick in idle. Update process times would miss the | |
1196 | * time we slept as update_process_times does only a 1 tick | |
1197 | * accounting. Enforce that this is accounted to idle ! | |
1198 | */ | |
1199 | ticks = jiffies - ts->idle_jiffies; | |
1200 | /* | |
1201 | * We might be one off. Do not randomly account a huge number of ticks! | |
1202 | */ | |
79741dd3 MS |
1203 | if (ticks && ticks < LONG_MAX) |
1204 | account_idle_ticks(ticks); | |
1205 | #endif | |
19f5f736 FW |
1206 | } |
1207 | ||
2aaf709a RW |
1208 | static void __tick_nohz_idle_restart_tick(struct tick_sched *ts, ktime_t now) |
1209 | { | |
1210 | tick_nohz_restart_sched_tick(ts, now); | |
1211 | tick_nohz_account_idle_ticks(ts); | |
1212 | } | |
1213 | ||
1214 | void tick_nohz_idle_restart_tick(void) | |
1215 | { | |
1216 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1217 | ||
1218 | if (ts->tick_stopped) | |
1219 | __tick_nohz_idle_restart_tick(ts, ktime_get()); | |
1220 | } | |
1221 | ||
79bf2bb3 | 1222 | /** |
280f0677 | 1223 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1224 | * |
1225 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1226 | * This also exit the RCU extended quiescent state. The CPU |
1227 | * can use RCU again after this function is called. | |
79bf2bb3 | 1228 | */ |
280f0677 | 1229 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1230 | { |
4a32fea9 | 1231 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
bbe9a70a | 1232 | bool idle_active, tick_stopped; |
6378ddb5 | 1233 | ktime_t now; |
79bf2bb3 | 1234 | |
6378ddb5 | 1235 | local_irq_disable(); |
2bbb6817 | 1236 | |
15f827be | 1237 | WARN_ON_ONCE(!ts->inidle); |
23a8d888 | 1238 | WARN_ON_ONCE(ts->timer_expires_base); |
15f827be FW |
1239 | |
1240 | ts->inidle = 0; | |
bbe9a70a AB |
1241 | idle_active = ts->idle_active; |
1242 | tick_stopped = ts->tick_stopped; | |
15f827be | 1243 | |
bbe9a70a | 1244 | if (idle_active || tick_stopped) |
eed3b9cf MS |
1245 | now = ktime_get(); |
1246 | ||
bbe9a70a | 1247 | if (idle_active) |
e8fcaa5c | 1248 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1249 | |
bbe9a70a | 1250 | if (tick_stopped) |
2aaf709a | 1251 | __tick_nohz_idle_restart_tick(ts, now); |
79bf2bb3 | 1252 | |
79bf2bb3 TG |
1253 | local_irq_enable(); |
1254 | } | |
1255 | ||
79bf2bb3 TG |
1256 | /* |
1257 | * The nohz low res interrupt handler | |
1258 | */ | |
1259 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1260 | { | |
22127e93 | 1261 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1262 | struct pt_regs *regs = get_irq_regs(); |
1263 | ktime_t now = ktime_get(); | |
1264 | ||
2456e855 | 1265 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1266 | |
ff7de620 | 1267 | tick_sched_do_timer(ts, now); |
9e8f559b | 1268 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1269 | |
b5e995e6 VK |
1270 | /* No need to reprogram if we are running tickless */ |
1271 | if (unlikely(ts->tick_stopped)) | |
1272 | return; | |
1273 | ||
b9965449 | 1274 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
0ff53d09 | 1275 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
79bf2bb3 TG |
1276 | } |
1277 | ||
bc7a34b8 TG |
1278 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1279 | { | |
1280 | if (!tick_nohz_enabled) | |
1281 | return; | |
1282 | ts->nohz_mode = mode; | |
1283 | /* One update is enough */ | |
1284 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
ae67bada | 1285 | timers_update_nohz(); |
bc7a34b8 TG |
1286 | } |
1287 | ||
79bf2bb3 TG |
1288 | /** |
1289 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1290 | */ | |
1291 | static void tick_nohz_switch_to_nohz(void) | |
1292 | { | |
22127e93 | 1293 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1294 | ktime_t next; |
1295 | ||
27630532 | 1296 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1297 | return; |
1298 | ||
6b442bc8 | 1299 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1300 | return; |
6b442bc8 | 1301 | |
79bf2bb3 TG |
1302 | /* |
1303 | * Recycle the hrtimer in ts, so we can share the | |
1304 | * hrtimer_forward with the highres code. | |
1305 | */ | |
71fed982 | 1306 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 TG |
1307 | /* Get the next period */ |
1308 | next = tick_init_jiffy_update(); | |
1309 | ||
0ff53d09 | 1310 | hrtimer_set_expires(&ts->sched_timer, next); |
b9965449 | 1311 | hrtimer_forward_now(&ts->sched_timer, TICK_NSEC); |
1ca8ec53 | 1312 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
bc7a34b8 | 1313 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1314 | } |
1315 | ||
5acac1be | 1316 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1317 | { |
4a32fea9 | 1318 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1319 | ktime_t now; |
1320 | ||
1321 | if (!ts->idle_active && !ts->tick_stopped) | |
1322 | return; | |
1323 | now = ktime_get(); | |
1324 | if (ts->idle_active) | |
e8fcaa5c | 1325 | tick_nohz_stop_idle(ts, now); |
ff006732 | 1326 | if (ts->tick_stopped) |
eed3b9cf | 1327 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1328 | } |
1329 | ||
79bf2bb3 TG |
1330 | #else |
1331 | ||
1332 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1333 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1334 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1335 | |
3451d024 | 1336 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1337 | |
719254fa TG |
1338 | /* |
1339 | * Called from irq_enter to notify about the possible interruption of idle() | |
1340 | */ | |
5acac1be | 1341 | void tick_irq_enter(void) |
719254fa | 1342 | { |
e8fcaa5c | 1343 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1344 | tick_nohz_irq_enter(); |
719254fa TG |
1345 | } |
1346 | ||
79bf2bb3 TG |
1347 | /* |
1348 | * High resolution timer specific code | |
1349 | */ | |
1350 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1351 | /* | |
4c9dc641 | 1352 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1353 | * Called with interrupts disabled. |
79bf2bb3 TG |
1354 | */ |
1355 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1356 | { | |
1357 | struct tick_sched *ts = | |
1358 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1359 | struct pt_regs *regs = get_irq_regs(); |
1360 | ktime_t now = ktime_get(); | |
d3ed7824 | 1361 | |
ff7de620 | 1362 | tick_sched_do_timer(ts, now); |
79bf2bb3 TG |
1363 | |
1364 | /* | |
1365 | * Do not call, when we are not in irq context and have | |
1366 | * no valid regs pointer | |
1367 | */ | |
9e8f559b FW |
1368 | if (regs) |
1369 | tick_sched_handle(ts, regs); | |
7c259045 FW |
1370 | else |
1371 | ts->next_tick = 0; | |
79bf2bb3 | 1372 | |
2a16fc93 VK |
1373 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1374 | if (unlikely(ts->tick_stopped)) | |
1375 | return HRTIMER_NORESTART; | |
1376 | ||
b9965449 | 1377 | hrtimer_forward(timer, now, TICK_NSEC); |
79bf2bb3 TG |
1378 | |
1379 | return HRTIMER_RESTART; | |
1380 | } | |
1381 | ||
5307c955 MG |
1382 | static int sched_skew_tick; |
1383 | ||
62cf20b3 TG |
1384 | static int __init skew_tick(char *str) |
1385 | { | |
1386 | get_option(&str, &sched_skew_tick); | |
1387 | ||
1388 | return 0; | |
1389 | } | |
1390 | early_param("skew_tick", skew_tick); | |
1391 | ||
79bf2bb3 TG |
1392 | /** |
1393 | * tick_setup_sched_timer - setup the tick emulation timer | |
1394 | */ | |
1395 | void tick_setup_sched_timer(void) | |
1396 | { | |
22127e93 | 1397 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1398 | ktime_t now = ktime_get(); |
1399 | ||
1400 | /* | |
1401 | * Emulate tick processing via per-CPU hrtimers: | |
1402 | */ | |
902a9f9c | 1403 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 | 1404 | ts->sched_timer.function = tick_sched_timer; |
79bf2bb3 | 1405 | |
0de7611a | 1406 | /* Get the next period (per-CPU) */ |
cc584b21 | 1407 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1408 | |
9c3f9e28 | 1409 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 | 1410 | if (sched_skew_tick) { |
b9965449 | 1411 | u64 offset = TICK_NSEC >> 1; |
5307c955 MG |
1412 | do_div(offset, num_possible_cpus()); |
1413 | offset *= smp_processor_id(); | |
1414 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1415 | } | |
1416 | ||
b9965449 | 1417 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
902a9f9c | 1418 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED_HARD); |
bc7a34b8 | 1419 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1420 | } |
3c4fbe5e | 1421 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1422 | |
3451d024 | 1423 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1424 | void tick_cancel_sched_timer(int cpu) |
1425 | { | |
1426 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1427 | ||
3c4fbe5e | 1428 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1429 | if (ts->sched_timer.base) |
1430 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1431 | # endif |
a7901766 | 1432 | |
4b0c0f29 | 1433 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1434 | } |
3c4fbe5e | 1435 | #endif |
79bf2bb3 TG |
1436 | |
1437 | /** | |
1438 | * Async notification about clocksource changes | |
1439 | */ | |
1440 | void tick_clock_notify(void) | |
1441 | { | |
1442 | int cpu; | |
1443 | ||
1444 | for_each_possible_cpu(cpu) | |
1445 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1446 | } | |
1447 | ||
1448 | /* | |
1449 | * Async notification about clock event changes | |
1450 | */ | |
1451 | void tick_oneshot_notify(void) | |
1452 | { | |
22127e93 | 1453 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1454 | |
1455 | set_bit(0, &ts->check_clocks); | |
1456 | } | |
1457 | ||
1458 | /** | |
1459 | * Check, if a change happened, which makes oneshot possible. | |
1460 | * | |
1461 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1462 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1463 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1464 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1465 | */ |
1466 | int tick_check_oneshot_change(int allow_nohz) | |
1467 | { | |
22127e93 | 1468 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1469 | |
1470 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1471 | return 0; | |
1472 | ||
1473 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1474 | return 0; | |
1475 | ||
cf4fc6cb | 1476 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1477 | return 0; |
1478 | ||
1479 | if (!allow_nohz) | |
1480 | return 1; | |
1481 | ||
1482 | tick_nohz_switch_to_nohz(); | |
1483 | return 0; | |
1484 | } |