]>
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 | |
f26af895 | 189 | WARN_ON_ONCE(tick_nohz_full_running); |
08ae95f4 | 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; |
f268c373 | 233 | EXPORT_SYMBOL_GPL(tick_nohz_full_mask); |
73867dcd | 234 | bool tick_nohz_full_running; |
ae9e557b | 235 | EXPORT_SYMBOL_GPL(tick_nohz_full_running); |
f009a7a7 | 236 | static atomic_t tick_dep_mask; |
a831881b | 237 | |
f009a7a7 | 238 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 239 | { |
f009a7a7 FW |
240 | int val = atomic_read(dep); |
241 | ||
242 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 243 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 244 | return true; |
d027d45d FW |
245 | } |
246 | ||
f009a7a7 | 247 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 248 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 249 | return true; |
d027d45d FW |
250 | } |
251 | ||
f009a7a7 | 252 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 253 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 254 | return true; |
d027d45d FW |
255 | } |
256 | ||
f009a7a7 | 257 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 258 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
259 | return true; |
260 | } | |
261 | ||
01b4c399 FW |
262 | if (val & TICK_DEP_MASK_RCU) { |
263 | trace_tick_stop(0, TICK_DEP_MASK_RCU); | |
264 | return true; | |
265 | } | |
266 | ||
f009a7a7 | 267 | return false; |
d027d45d FW |
268 | } |
269 | ||
57ccdf44 | 270 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d | 271 | { |
ebf3adba | 272 | lockdep_assert_irqs_disabled(); |
9014c45d | 273 | |
57ccdf44 WL |
274 | if (unlikely(!cpu_online(cpu))) |
275 | return false; | |
276 | ||
f009a7a7 | 277 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 278 | return false; |
d027d45d | 279 | |
f009a7a7 | 280 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 281 | return false; |
d027d45d | 282 | |
f009a7a7 | 283 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 284 | return false; |
d027d45d | 285 | |
f009a7a7 | 286 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 287 | return false; |
d027d45d | 288 | |
9014c45d FW |
289 | return true; |
290 | } | |
291 | ||
d027d45d | 292 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 293 | { |
73738a95 | 294 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
295 | } |
296 | ||
7a9f50a0 PZ |
297 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = |
298 | IRQ_WORK_INIT_HARD(nohz_full_kick_func); | |
76c24fb0 | 299 | |
40bea039 FW |
300 | /* |
301 | * Kick this CPU if it's full dynticks in order to force it to | |
302 | * re-evaluate its dependency on the tick and restart it if necessary. | |
303 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
304 | * is NMI safe. | |
305 | */ | |
555e0c1e | 306 | static void tick_nohz_full_kick(void) |
40bea039 FW |
307 | { |
308 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
309 | return; | |
310 | ||
56e4dea8 | 311 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
312 | } |
313 | ||
76c24fb0 | 314 | /* |
3d36aebc | 315 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
316 | * re-evaluate its dependency on the tick and restart it if necessary. |
317 | */ | |
3d36aebc | 318 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 319 | { |
3d36aebc FW |
320 | if (!tick_nohz_full_cpu(cpu)) |
321 | return; | |
322 | ||
323 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
324 | } |
325 | ||
29721b85 FW |
326 | static void tick_nohz_kick_task(struct task_struct *tsk) |
327 | { | |
a1dfb631 MT |
328 | int cpu; |
329 | ||
330 | /* | |
331 | * If the task is not running, run_posix_cpu_timers() | |
332 | * has nothing to elapse, IPI can then be spared. | |
333 | * | |
334 | * activate_task() STORE p->tick_dep_mask | |
335 | * STORE p->on_rq | |
336 | * __schedule() (switch to task 'p') smp_mb() (atomic_fetch_or()) | |
337 | * LOCK rq->lock LOAD p->on_rq | |
338 | * smp_mb__after_spin_lock() | |
339 | * tick_nohz_task_switch() | |
340 | * LOAD p->tick_dep_mask | |
341 | */ | |
342 | if (!sched_task_on_rq(tsk)) | |
343 | return; | |
29721b85 FW |
344 | |
345 | /* | |
346 | * If the task concurrently migrates to another CPU, | |
347 | * we guarantee it sees the new tick dependency upon | |
348 | * schedule. | |
349 | * | |
29721b85 FW |
350 | * set_task_cpu(p, cpu); |
351 | * STORE p->cpu = @cpu | |
352 | * __schedule() (switch to task 'p') | |
353 | * LOCK rq->lock | |
354 | * smp_mb__after_spin_lock() STORE p->tick_dep_mask | |
355 | * tick_nohz_task_switch() smp_mb() (atomic_fetch_or()) | |
356 | * LOAD p->tick_dep_mask LOAD p->cpu | |
357 | */ | |
a1dfb631 | 358 | cpu = task_cpu(tsk); |
29721b85 FW |
359 | |
360 | preempt_disable(); | |
361 | if (cpu_online(cpu)) | |
362 | tick_nohz_full_kick_cpu(cpu); | |
363 | preempt_enable(); | |
364 | } | |
365 | ||
76c24fb0 FW |
366 | /* |
367 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
368 | * their dependency on the tick and restart it if necessary. | |
369 | */ | |
b7878300 | 370 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 371 | { |
8537bb95 FW |
372 | int cpu; |
373 | ||
73867dcd | 374 | if (!tick_nohz_full_running) |
76c24fb0 FW |
375 | return; |
376 | ||
377 | preempt_disable(); | |
8537bb95 FW |
378 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
379 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
380 | preempt_enable(); |
381 | } | |
382 | ||
f009a7a7 | 383 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
384 | enum tick_dep_bits bit) |
385 | { | |
f009a7a7 | 386 | int prev; |
d027d45d | 387 | |
a1cc5bcf | 388 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
389 | if (!prev) |
390 | tick_nohz_full_kick_all(); | |
391 | } | |
392 | ||
393 | /* | |
394 | * Set a global tick dependency. Used by perf events that rely on freq and | |
395 | * by unstable clock. | |
396 | */ | |
397 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
398 | { | |
399 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
400 | } | |
401 | ||
402 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
403 | { | |
f009a7a7 | 404 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
405 | } |
406 | ||
407 | /* | |
408 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
409 | * manage events throttling. | |
410 | */ | |
411 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
412 | { | |
f009a7a7 | 413 | int prev; |
d027d45d FW |
414 | struct tick_sched *ts; |
415 | ||
416 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
417 | ||
a1cc5bcf | 418 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
419 | if (!prev) { |
420 | preempt_disable(); | |
421 | /* Perf needs local kick that is NMI safe */ | |
422 | if (cpu == smp_processor_id()) { | |
423 | tick_nohz_full_kick(); | |
424 | } else { | |
425 | /* Remote irq work not NMI-safe */ | |
426 | if (!WARN_ON_ONCE(in_nmi())) | |
427 | tick_nohz_full_kick_cpu(cpu); | |
428 | } | |
429 | preempt_enable(); | |
430 | } | |
431 | } | |
01b4c399 | 432 | EXPORT_SYMBOL_GPL(tick_nohz_dep_set_cpu); |
d027d45d FW |
433 | |
434 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
435 | { | |
436 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
437 | ||
f009a7a7 | 438 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d | 439 | } |
01b4c399 | 440 | EXPORT_SYMBOL_GPL(tick_nohz_dep_clear_cpu); |
d027d45d FW |
441 | |
442 | /* | |
3c8920e2 FW |
443 | * Set a per-task tick dependency. RCU need this. Also posix CPU timers |
444 | * in order to elapse per task timers. | |
d027d45d FW |
445 | */ |
446 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
447 | { | |
29721b85 FW |
448 | if (!atomic_fetch_or(BIT(bit), &tsk->tick_dep_mask)) |
449 | tick_nohz_kick_task(tsk); | |
d027d45d | 450 | } |
ae9e557b | 451 | EXPORT_SYMBOL_GPL(tick_nohz_dep_set_task); |
d027d45d FW |
452 | |
453 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
454 | { | |
f009a7a7 | 455 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d | 456 | } |
ae9e557b | 457 | EXPORT_SYMBOL_GPL(tick_nohz_dep_clear_task); |
d027d45d FW |
458 | |
459 | /* | |
460 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
461 | * per process timers. | |
462 | */ | |
1e4ca26d MT |
463 | void tick_nohz_dep_set_signal(struct task_struct *tsk, |
464 | enum tick_dep_bits bit) | |
d027d45d | 465 | { |
1e4ca26d MT |
466 | int prev; |
467 | struct signal_struct *sig = tsk->signal; | |
468 | ||
469 | prev = atomic_fetch_or(BIT(bit), &sig->tick_dep_mask); | |
470 | if (!prev) { | |
471 | struct task_struct *t; | |
472 | ||
473 | lockdep_assert_held(&tsk->sighand->siglock); | |
474 | __for_each_thread(sig, t) | |
475 | tick_nohz_kick_task(t); | |
476 | } | |
d027d45d FW |
477 | } |
478 | ||
479 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
480 | { | |
f009a7a7 | 481 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
482 | } |
483 | ||
99e5ada9 FW |
484 | /* |
485 | * Re-evaluate the need for the tick as we switch the current task. | |
486 | * It might need the tick due to per task/process properties: | |
0de7611a | 487 | * perf events, posix CPU timers, ... |
99e5ada9 | 488 | */ |
de734f89 | 489 | void __tick_nohz_task_switch(void) |
99e5ada9 | 490 | { |
d027d45d | 491 | struct tick_sched *ts; |
99e5ada9 | 492 | |
6296ace4 | 493 | if (!tick_nohz_full_cpu(smp_processor_id())) |
0fdcccfa | 494 | return; |
6296ace4 | 495 | |
d027d45d | 496 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 497 | |
d027d45d | 498 | if (ts->tick_stopped) { |
f009a7a7 FW |
499 | if (atomic_read(¤t->tick_dep_mask) || |
500 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
501 | tick_nohz_full_kick(); |
502 | } | |
99e5ada9 FW |
503 | } |
504 | ||
6f1982fe FW |
505 | /* Get the boot-time nohz CPU list from the kernel parameters. */ |
506 | void __init tick_nohz_full_setup(cpumask_var_t cpumask) | |
a831881b | 507 | { |
73867dcd | 508 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
6f1982fe | 509 | cpumask_copy(tick_nohz_full_mask, cpumask); |
73867dcd | 510 | tick_nohz_full_running = true; |
a831881b | 511 | } |
ae9e557b | 512 | EXPORT_SYMBOL_GPL(tick_nohz_full_setup); |
a831881b | 513 | |
31eff243 | 514 | static int tick_nohz_cpu_down(unsigned int cpu) |
a382bf93 | 515 | { |
31eff243 | 516 | /* |
08ae95f4 NP |
517 | * The tick_do_timer_cpu CPU handles housekeeping duty (unbound |
518 | * timers, workqueues, timekeeping, ...) on behalf of full dynticks | |
31eff243 SAS |
519 | * CPUs. It must remain online when nohz full is enabled. |
520 | */ | |
521 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) | |
522 | return -EBUSY; | |
523 | return 0; | |
a382bf93 FW |
524 | } |
525 | ||
d1e43fa5 | 526 | void __init tick_nohz_init(void) |
a831881b | 527 | { |
31eff243 | 528 | int cpu, ret; |
d1e43fa5 | 529 | |
a7c8655b PM |
530 | if (!tick_nohz_full_running) |
531 | return; | |
d1e43fa5 | 532 | |
9b01f5bf FW |
533 | /* |
534 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
535 | * locking contexts. But then we need irq work to raise its own | |
536 | * interrupts to avoid circular dependency on the tick | |
537 | */ | |
538 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 539 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf | 540 | cpumask_clear(tick_nohz_full_mask); |
9b01f5bf FW |
541 | tick_nohz_full_running = false; |
542 | return; | |
543 | } | |
544 | ||
08ae95f4 NP |
545 | if (IS_ENABLED(CONFIG_PM_SLEEP_SMP) && |
546 | !IS_ENABLED(CONFIG_PM_SLEEP_SMP_NONZERO_CPU)) { | |
547 | cpu = smp_processor_id(); | |
4327b15f | 548 | |
08ae95f4 NP |
549 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { |
550 | pr_warn("NO_HZ: Clearing %d from nohz_full range " | |
551 | "for timekeeping\n", cpu); | |
552 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); | |
553 | } | |
4327b15f FW |
554 | } |
555 | ||
73867dcd | 556 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
557 | context_tracking_cpu_set(cpu); |
558 | ||
31eff243 SAS |
559 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
560 | "kernel/nohz:predown", NULL, | |
561 | tick_nohz_cpu_down); | |
562 | WARN_ON(ret < 0); | |
ffda22c1 TH |
563 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
564 | cpumask_pr_args(tick_nohz_full_mask)); | |
a831881b | 565 | } |
a831881b FW |
566 | #endif |
567 | ||
79bf2bb3 TG |
568 | /* |
569 | * NOHZ - aka dynamic tick functionality | |
570 | */ | |
3451d024 | 571 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
572 | /* |
573 | * NO HZ enabled ? | |
574 | */ | |
4cc7ecb7 | 575 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 576 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
577 | /* |
578 | * Enable / Disable tickless mode | |
579 | */ | |
580 | static int __init setup_tick_nohz(char *str) | |
581 | { | |
4cc7ecb7 | 582 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
583 | } |
584 | ||
585 | __setup("nohz=", setup_tick_nohz); | |
586 | ||
a3642983 | 587 | bool tick_nohz_tick_stopped(void) |
c1797baf | 588 | { |
2bc629a6 FW |
589 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
590 | ||
591 | return ts->tick_stopped; | |
c1797baf TG |
592 | } |
593 | ||
22ab8bc0 FW |
594 | bool tick_nohz_tick_stopped_cpu(int cpu) |
595 | { | |
596 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
597 | ||
598 | return ts->tick_stopped; | |
599 | } | |
600 | ||
79bf2bb3 TG |
601 | /** |
602 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
603 | * | |
604 | * Called from interrupt entry when the CPU was idle | |
605 | * | |
606 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
607 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
608 | * value. We do this unconditionally on any CPU, as we don't know whether the |
609 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 610 | */ |
eed3b9cf | 611 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 612 | { |
79bf2bb3 | 613 | unsigned long flags; |
79bf2bb3 | 614 | |
e8fcaa5c | 615 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
616 | |
617 | local_irq_save(flags); | |
618 | tick_do_update_jiffies64(now); | |
619 | local_irq_restore(flags); | |
02ff3755 | 620 | |
03e0d461 | 621 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
622 | } |
623 | ||
595aac48 | 624 | /* |
0de7611a | 625 | * Updates the per-CPU time idle statistics counters |
595aac48 | 626 | */ |
8d63bf94 | 627 | static void |
8c215bd3 | 628 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 629 | { |
eed3b9cf | 630 | ktime_t delta; |
6378ddb5 | 631 | |
595aac48 AV |
632 | if (ts->idle_active) { |
633 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 634 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 635 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
636 | else |
637 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 638 | ts->idle_entrytime = now; |
595aac48 | 639 | } |
8d63bf94 | 640 | |
e0e37c20 | 641 | if (last_update_time) |
8d63bf94 AV |
642 | *last_update_time = ktime_to_us(now); |
643 | ||
595aac48 AV |
644 | } |
645 | ||
e8fcaa5c | 646 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 647 | { |
e8fcaa5c | 648 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 649 | ts->idle_active = 0; |
56c7426b | 650 | |
ac1e843f | 651 | sched_clock_idle_wakeup_event(); |
6378ddb5 VP |
652 | } |
653 | ||
0e776768 | 654 | static void tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 655 | { |
0e776768 | 656 | ts->idle_entrytime = ktime_get(); |
6378ddb5 | 657 | ts->idle_active = 1; |
56c7426b | 658 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
659 | } |
660 | ||
b1f724c3 | 661 | /** |
0de7611a | 662 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 663 | * @cpu: CPU number to query |
09a1d34f MH |
664 | * @last_update_time: variable to store update time in. Do not update |
665 | * counters if NULL. | |
b1f724c3 | 666 | * |
6168f8ed | 667 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 668 | * CPU, in microseconds. |
b1f724c3 AV |
669 | * |
670 | * This time is measured via accounting rather than sampling, | |
671 | * and is as accurate as ktime_get() is. | |
672 | * | |
673 | * This function returns -1 if NOHZ is not enabled. | |
674 | */ | |
6378ddb5 VP |
675 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
676 | { | |
677 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 678 | ktime_t now, idle; |
6378ddb5 | 679 | |
d689fe22 | 680 | if (!tick_nohz_active) |
8083e4ad | 681 | return -1; |
682 | ||
09a1d34f MH |
683 | now = ktime_get(); |
684 | if (last_update_time) { | |
685 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
686 | idle = ts->idle_sleeptime; | |
687 | } else { | |
688 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
689 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
690 | ||
691 | idle = ktime_add(ts->idle_sleeptime, delta); | |
692 | } else { | |
693 | idle = ts->idle_sleeptime; | |
694 | } | |
695 | } | |
696 | ||
697 | return ktime_to_us(idle); | |
8083e4ad | 698 | |
6378ddb5 | 699 | } |
8083e4ad | 700 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 701 | |
6beea0cd | 702 | /** |
0de7611a | 703 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 704 | * @cpu: CPU number to query |
09a1d34f MH |
705 | * @last_update_time: variable to store update time in. Do not update |
706 | * counters if NULL. | |
0224cf4c | 707 | * |
6168f8ed | 708 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
709 | * CPU, in microseconds. |
710 | * | |
711 | * This time is measured via accounting rather than sampling, | |
712 | * and is as accurate as ktime_get() is. | |
713 | * | |
714 | * This function returns -1 if NOHZ is not enabled. | |
715 | */ | |
716 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
717 | { | |
718 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 719 | ktime_t now, iowait; |
0224cf4c | 720 | |
d689fe22 | 721 | if (!tick_nohz_active) |
0224cf4c AV |
722 | return -1; |
723 | ||
09a1d34f MH |
724 | now = ktime_get(); |
725 | if (last_update_time) { | |
726 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
727 | iowait = ts->iowait_sleeptime; | |
728 | } else { | |
729 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
730 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 731 | |
09a1d34f MH |
732 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
733 | } else { | |
734 | iowait = ts->iowait_sleeptime; | |
735 | } | |
736 | } | |
0224cf4c | 737 | |
09a1d34f | 738 | return ktime_to_us(iowait); |
0224cf4c AV |
739 | } |
740 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
741 | ||
0ff53d09 TG |
742 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
743 | { | |
744 | hrtimer_cancel(&ts->sched_timer); | |
745 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
746 | ||
747 | /* Forward the time to expire in the future */ | |
b9965449 | 748 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
0ff53d09 | 749 | |
902a9f9c SAS |
750 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
751 | hrtimer_start_expires(&ts->sched_timer, | |
752 | HRTIMER_MODE_ABS_PINNED_HARD); | |
753 | } else { | |
0ff53d09 | 754 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
902a9f9c | 755 | } |
411fe24e FW |
756 | |
757 | /* | |
758 | * Reset to make sure next tick stop doesn't get fooled by past | |
759 | * cached clock deadline. | |
760 | */ | |
761 | ts->next_tick = 0; | |
0ff53d09 TG |
762 | } |
763 | ||
5d62c183 TG |
764 | static inline bool local_timer_softirq_pending(void) |
765 | { | |
80d20d35 | 766 | return local_softirq_pending() & BIT(TIMER_SOFTIRQ); |
5d62c183 TG |
767 | } |
768 | ||
23a8d888 | 769 | static ktime_t tick_nohz_next_event(struct tick_sched *ts, int cpu) |
79bf2bb3 | 770 | { |
c1ad348b | 771 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
e1e41b6c RV |
772 | unsigned long basejiff; |
773 | unsigned int seq; | |
855a0fc3 | 774 | |
79bf2bb3 TG |
775 | /* Read jiffies and the time when jiffies were updated last */ |
776 | do { | |
e5d4d175 | 777 | seq = read_seqcount_begin(&jiffies_seq); |
2456e855 | 778 | basemono = last_jiffies_update; |
c1ad348b | 779 | basejiff = jiffies; |
e5d4d175 | 780 | } while (read_seqcount_retry(&jiffies_seq, seq)); |
c1ad348b | 781 | ts->last_jiffies = basejiff; |
23a8d888 | 782 | ts->timer_expires_base = basemono; |
79bf2bb3 | 783 | |
5d62c183 TG |
784 | /* |
785 | * Keep the periodic tick, when RCU, architecture or irq_work | |
786 | * requests it. | |
787 | * Aside of that check whether the local timer softirq is | |
788 | * pending. If so its a bad idea to call get_next_timer_interrupt() | |
789 | * because there is an already expired timer, so it will request | |
4bf07f65 | 790 | * immediate expiry, which rearms the hardware timer with a |
5d62c183 TG |
791 | * minimal delta which brings us back to this place |
792 | * immediately. Lather, rinse and repeat... | |
793 | */ | |
794 | if (rcu_needs_cpu(basemono, &next_rcu) || arch_needs_cpu() || | |
795 | irq_work_needs_cpu() || local_timer_softirq_pending()) { | |
c1ad348b | 796 | next_tick = basemono + TICK_NSEC; |
3c5d92a0 | 797 | } else { |
c1ad348b TG |
798 | /* |
799 | * Get the next pending timer. If high resolution | |
800 | * timers are enabled this only takes the timer wheel | |
801 | * timers into account. If high resolution timers are | |
802 | * disabled this also looks at the next expiring | |
803 | * hrtimer. | |
804 | */ | |
805 | next_tmr = get_next_timer_interrupt(basejiff, basemono); | |
806 | ts->next_timer = next_tmr; | |
807 | /* Take the next rcu event into account */ | |
808 | next_tick = next_rcu < next_tmr ? next_rcu : next_tmr; | |
3c5d92a0 | 809 | } |
47aa8b6c | 810 | |
c1ad348b TG |
811 | /* |
812 | * If the tick is due in the next period, keep it ticking or | |
82bbe34b | 813 | * force prod the timer. |
c1ad348b TG |
814 | */ |
815 | delta = next_tick - basemono; | |
816 | if (delta <= (u64)TICK_NSEC) { | |
a683f390 TG |
817 | /* |
818 | * Tell the timer code that the base is not idle, i.e. undo | |
819 | * the effect of get_next_timer_interrupt(): | |
820 | */ | |
821 | timer_clear_idle(); | |
82bbe34b PZ |
822 | /* |
823 | * We've not stopped the tick yet, and there's a timer in the | |
824 | * next period, so no point in stopping it either, bail. | |
825 | */ | |
f99973e1 | 826 | if (!ts->tick_stopped) { |
23a8d888 | 827 | ts->timer_expires = 0; |
157d29e1 TG |
828 | goto out; |
829 | } | |
830 | } | |
831 | ||
23a8d888 RW |
832 | /* |
833 | * If this CPU is the one which had the do_timer() duty last, we limit | |
834 | * the sleep time to the timekeeping max_deferment value. | |
835 | * Otherwise we can sleep as long as we want. | |
836 | */ | |
837 | delta = timekeeping_max_deferment(); | |
838 | if (cpu != tick_do_timer_cpu && | |
839 | (tick_do_timer_cpu != TICK_DO_TIMER_NONE || !ts->do_timer_last)) | |
840 | delta = KTIME_MAX; | |
841 | ||
842 | /* Calculate the next expiry time */ | |
843 | if (delta < (KTIME_MAX - basemono)) | |
844 | expires = basemono + delta; | |
845 | else | |
846 | expires = KTIME_MAX; | |
847 | ||
848 | ts->timer_expires = min_t(u64, expires, next_tick); | |
849 | ||
850 | out: | |
851 | return ts->timer_expires; | |
852 | } | |
853 | ||
854 | static void tick_nohz_stop_tick(struct tick_sched *ts, int cpu) | |
855 | { | |
856 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); | |
857 | u64 basemono = ts->timer_expires_base; | |
858 | u64 expires = ts->timer_expires; | |
859 | ktime_t tick = expires; | |
860 | ||
861 | /* Make sure we won't be trying to stop it twice in a row. */ | |
862 | ts->timer_expires_base = 0; | |
863 | ||
79bf2bb3 | 864 | /* |
0de7611a IM |
865 | * If this CPU is the one which updates jiffies, then give up |
866 | * the assignment and let it be taken by the CPU which runs | |
867 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
868 | * don't drop this here the jiffies might be stale and |
869 | * do_timer() never invoked. Keep track of the fact that it | |
23a8d888 | 870 | * was the one which had the do_timer() duty last. |
79bf2bb3 | 871 | */ |
157d29e1 TG |
872 | if (cpu == tick_do_timer_cpu) { |
873 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
874 | ts->do_timer_last = 1; | |
875 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
157d29e1 | 876 | ts->do_timer_last = 0; |
157d29e1 | 877 | } |
27185016 | 878 | |
157d29e1 | 879 | /* Skip reprogram of event if its not changed */ |
411fe24e FW |
880 | if (ts->tick_stopped && (expires == ts->next_tick)) { |
881 | /* Sanity check: make sure clockevent is actually programmed */ | |
d4af6d93 | 882 | if (tick == KTIME_MAX || ts->next_tick == hrtimer_get_expires(&ts->sched_timer)) |
23a8d888 | 883 | return; |
411fe24e FW |
884 | |
885 | WARN_ON_ONCE(1); | |
886 | printk_once("basemono: %llu ts->next_tick: %llu dev->next_event: %llu timer->active: %d timer->expires: %llu\n", | |
887 | basemono, ts->next_tick, dev->next_event, | |
888 | hrtimer_active(&ts->sched_timer), hrtimer_get_expires(&ts->sched_timer)); | |
ce6cf9a1 | 889 | } |
84bf1bcc | 890 | |
157d29e1 TG |
891 | /* |
892 | * nohz_stop_sched_tick can be called several times before | |
893 | * the nohz_restart_sched_tick is called. This happens when | |
894 | * interrupts arrive which do not cause a reschedule. In the | |
895 | * first call we save the current tick time, so we can restart | |
896 | * the scheduler tick in nohz_restart_sched_tick. | |
897 | */ | |
898 | if (!ts->tick_stopped) { | |
3c85d6db | 899 | calc_load_nohz_start(); |
62cb1188 | 900 | quiet_vmstat(); |
d3ed7824 | 901 | |
157d29e1 TG |
902 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
903 | ts->tick_stopped = 1; | |
e6e6cc22 | 904 | trace_tick_stop(1, TICK_DEP_MASK_NONE); |
157d29e1 | 905 | } |
eaad084b | 906 | |
411fe24e FW |
907 | ts->next_tick = tick; |
908 | ||
157d29e1 | 909 | /* |
c1ad348b TG |
910 | * If the expiration time == KTIME_MAX, then we simply stop |
911 | * the tick timer. | |
157d29e1 | 912 | */ |
c1ad348b | 913 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
914 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
915 | hrtimer_cancel(&ts->sched_timer); | |
23a8d888 | 916 | return; |
79bf2bb3 | 917 | } |
0ff53d09 | 918 | |
1f71addd | 919 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
902a9f9c SAS |
920 | hrtimer_start(&ts->sched_timer, tick, |
921 | HRTIMER_MODE_ABS_PINNED_HARD); | |
1f71addd TG |
922 | } else { |
923 | hrtimer_set_expires(&ts->sched_timer, tick); | |
c1ad348b | 924 | tick_program_event(tick, 1); |
1f71addd | 925 | } |
280f0677 FW |
926 | } |
927 | ||
23a8d888 RW |
928 | static void tick_nohz_retain_tick(struct tick_sched *ts) |
929 | { | |
930 | ts->timer_expires_base = 0; | |
931 | } | |
932 | ||
933 | #ifdef CONFIG_NO_HZ_FULL | |
934 | static void tick_nohz_stop_sched_tick(struct tick_sched *ts, int cpu) | |
935 | { | |
936 | if (tick_nohz_next_event(ts, cpu)) | |
937 | tick_nohz_stop_tick(ts, cpu); | |
938 | else | |
939 | tick_nohz_retain_tick(ts); | |
940 | } | |
941 | #endif /* CONFIG_NO_HZ_FULL */ | |
942 | ||
1f41906a | 943 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
944 | { |
945 | /* Update jiffies first */ | |
946 | tick_do_update_jiffies64(now); | |
59d2c7ca | 947 | |
a683f390 TG |
948 | /* |
949 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
950 | * the clock forward checks in the enqueue path: | |
951 | */ | |
952 | timer_clear_idle(); | |
953 | ||
3c85d6db | 954 | calc_load_nohz_stop(); |
03e0d461 | 955 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
956 | /* |
957 | * Cancel the scheduled timer and restore the tick | |
958 | */ | |
959 | ts->tick_stopped = 0; | |
59d2c7ca FW |
960 | tick_nohz_restart(ts, now); |
961 | } | |
73738a95 | 962 | |
a5183862 YY |
963 | static void __tick_nohz_full_update_tick(struct tick_sched *ts, |
964 | ktime_t now) | |
5811d996 FW |
965 | { |
966 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 967 | int cpu = smp_processor_id(); |
5811d996 | 968 | |
a5183862 YY |
969 | if (can_stop_full_tick(cpu, ts)) |
970 | tick_nohz_stop_sched_tick(ts, cpu); | |
971 | else if (ts->tick_stopped) | |
972 | tick_nohz_restart_sched_tick(ts, now); | |
973 | #endif | |
974 | } | |
975 | ||
976 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
977 | { | |
978 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
e9a2eb40 | 979 | return; |
5811d996 | 980 | |
e9a2eb40 AS |
981 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
982 | return; | |
5811d996 | 983 | |
a5183862 | 984 | __tick_nohz_full_update_tick(ts, ktime_get()); |
5811d996 FW |
985 | } |
986 | ||
5b39939a FW |
987 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
988 | { | |
989 | /* | |
0de7611a | 990 | * If this CPU is offline and it is the one which updates |
5b39939a | 991 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 992 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
993 | * this here the jiffies might be stale and do_timer() never |
994 | * invoked. | |
995 | */ | |
996 | if (unlikely(!cpu_online(cpu))) { | |
997 | if (cpu == tick_do_timer_cpu) | |
998 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
411fe24e FW |
999 | /* |
1000 | * Make sure the CPU doesn't get fooled by obsolete tick | |
1001 | * deadline if it comes back online later. | |
1002 | */ | |
1003 | ts->next_tick = 0; | |
f7ea0fd6 | 1004 | return false; |
5b39939a FW |
1005 | } |
1006 | ||
23a8d888 | 1007 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) |
5b39939a FW |
1008 | return false; |
1009 | ||
1010 | if (need_resched()) | |
1011 | return false; | |
1012 | ||
d59e0ba1 | 1013 | if (unlikely(local_softirq_pending())) { |
5b39939a FW |
1014 | static int ratelimit; |
1015 | ||
47c218dc | 1016 | if (ratelimit < 10 && !local_bh_blocked() && |
803b0eba | 1017 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { |
bca37119 | 1018 | pr_warn("NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #%02x!!!\n", |
cfea7d7e | 1019 | (unsigned int) local_softirq_pending()); |
5b39939a FW |
1020 | ratelimit++; |
1021 | } | |
1022 | return false; | |
1023 | } | |
1024 | ||
460775df | 1025 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
1026 | /* |
1027 | * Keep the tick alive to guarantee timekeeping progression | |
1028 | * if there are full dynticks CPUs around | |
1029 | */ | |
1030 | if (tick_do_timer_cpu == cpu) | |
1031 | return false; | |
08ae95f4 NP |
1032 | |
1033 | /* Should not happen for nohz-full */ | |
1034 | if (WARN_ON_ONCE(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) | |
a382bf93 FW |
1035 | return false; |
1036 | } | |
1037 | ||
5b39939a FW |
1038 | return true; |
1039 | } | |
1040 | ||
0e776768 | 1041 | static void __tick_nohz_idle_stop_tick(struct tick_sched *ts) |
19f5f736 | 1042 | { |
0e776768 | 1043 | ktime_t expires; |
5b39939a | 1044 | int cpu = smp_processor_id(); |
19f5f736 | 1045 | |
554c8aa8 RW |
1046 | /* |
1047 | * If tick_nohz_get_sleep_length() ran tick_nohz_next_event(), the | |
1048 | * tick timer expiration time is known already. | |
1049 | */ | |
1050 | if (ts->timer_expires_base) | |
1051 | expires = ts->timer_expires; | |
1052 | else if (can_stop_idle_tick(cpu, ts)) | |
1053 | expires = tick_nohz_next_event(ts, cpu); | |
1054 | else | |
1055 | return; | |
23a8d888 RW |
1056 | |
1057 | ts->idle_calls++; | |
08d07259 | 1058 | |
23a8d888 | 1059 | if (expires > 0LL) { |
5b39939a FW |
1060 | int was_stopped = ts->tick_stopped; |
1061 | ||
23a8d888 | 1062 | tick_nohz_stop_tick(ts, cpu); |
84bf1bcc | 1063 | |
23a8d888 RW |
1064 | ts->idle_sleeps++; |
1065 | ts->idle_expires = expires; | |
5b39939a | 1066 | |
a0db971e | 1067 | if (!was_stopped && ts->tick_stopped) { |
5b39939a | 1068 | ts->idle_jiffies = ts->last_jiffies; |
a0db971e FW |
1069 | nohz_balance_enter_idle(cpu); |
1070 | } | |
23a8d888 RW |
1071 | } else { |
1072 | tick_nohz_retain_tick(ts); | |
5b39939a | 1073 | } |
280f0677 FW |
1074 | } |
1075 | ||
1076 | /** | |
0e776768 | 1077 | * tick_nohz_idle_stop_tick - stop the idle tick from the idle task |
280f0677 FW |
1078 | * |
1079 | * When the next event is more than a tick into the future, stop the idle tick | |
0e776768 RW |
1080 | */ |
1081 | void tick_nohz_idle_stop_tick(void) | |
1082 | { | |
1083 | __tick_nohz_idle_stop_tick(this_cpu_ptr(&tick_cpu_sched)); | |
1084 | } | |
1085 | ||
554c8aa8 RW |
1086 | void tick_nohz_idle_retain_tick(void) |
1087 | { | |
1088 | tick_nohz_retain_tick(this_cpu_ptr(&tick_cpu_sched)); | |
1089 | /* | |
1090 | * Undo the effect of get_next_timer_interrupt() called from | |
1091 | * tick_nohz_next_event(). | |
1092 | */ | |
1093 | timer_clear_idle(); | |
1094 | } | |
1095 | ||
0e776768 RW |
1096 | /** |
1097 | * tick_nohz_idle_enter - prepare for entering idle on the current CPU | |
2bbb6817 | 1098 | * |
0e776768 | 1099 | * Called when we start the idle loop. |
280f0677 | 1100 | */ |
1268fbc7 | 1101 | void tick_nohz_idle_enter(void) |
280f0677 FW |
1102 | { |
1103 | struct tick_sched *ts; | |
1104 | ||
ebf3adba | 1105 | lockdep_assert_irqs_enabled(); |
0db49b72 | 1106 | |
1268fbc7 FW |
1107 | local_irq_disable(); |
1108 | ||
22127e93 | 1109 | ts = this_cpu_ptr(&tick_cpu_sched); |
23a8d888 RW |
1110 | |
1111 | WARN_ON_ONCE(ts->timer_expires_base); | |
1112 | ||
280f0677 | 1113 | ts->inidle = 1; |
0e776768 | 1114 | tick_nohz_start_idle(ts); |
1268fbc7 FW |
1115 | |
1116 | local_irq_enable(); | |
280f0677 FW |
1117 | } |
1118 | ||
1119 | /** | |
1120 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
1121 | * | |
1122 | * When an interrupt fires while we are idle and it doesn't cause | |
1123 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
1124 | * an RCU callback, etc... | |
1125 | * So we need to re-calculate and reprogram the next tick event. | |
1126 | */ | |
1127 | void tick_nohz_irq_exit(void) | |
1128 | { | |
22127e93 | 1129 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 1130 | |
14851912 | 1131 | if (ts->inidle) |
0e776768 | 1132 | tick_nohz_start_idle(ts); |
14851912 | 1133 | else |
73738a95 | 1134 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
1135 | } |
1136 | ||
4f86d3a8 | 1137 | /** |
45f1ff59 RW |
1138 | * tick_nohz_idle_got_tick - Check whether or not the tick handler has run |
1139 | */ | |
1140 | bool tick_nohz_idle_got_tick(void) | |
1141 | { | |
1142 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1143 | ||
2bc629a6 FW |
1144 | if (ts->got_idle_tick) { |
1145 | ts->got_idle_tick = 0; | |
45f1ff59 RW |
1146 | return true; |
1147 | } | |
1148 | return false; | |
1149 | } | |
1150 | ||
6f9b83ac UH |
1151 | /** |
1152 | * tick_nohz_get_next_hrtimer - return the next expiration time for the hrtimer | |
1153 | * or the tick, whatever that expires first. Note that, if the tick has been | |
1154 | * stopped, it returns the next hrtimer. | |
1155 | * | |
1156 | * Called from power state control code with interrupts disabled | |
1157 | */ | |
1158 | ktime_t tick_nohz_get_next_hrtimer(void) | |
1159 | { | |
1160 | return __this_cpu_read(tick_cpu_device.evtdev)->next_event; | |
1161 | } | |
1162 | ||
4f86d3a8 | 1163 | /** |
554c8aa8 | 1164 | * tick_nohz_get_sleep_length - return the expected length of the current sleep |
296bb1e5 | 1165 | * @delta_next: duration until the next event if the tick cannot be stopped |
4f86d3a8 | 1166 | * |
4c81cb7e RW |
1167 | * Called from power state control code with interrupts disabled. |
1168 | * | |
1169 | * The return value of this function and/or the value returned by it through the | |
1170 | * @delta_next pointer can be negative which must be taken into account by its | |
1171 | * callers. | |
4f86d3a8 | 1172 | */ |
296bb1e5 | 1173 | ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next) |
4f86d3a8 | 1174 | { |
554c8aa8 | 1175 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
22127e93 | 1176 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
554c8aa8 RW |
1177 | int cpu = smp_processor_id(); |
1178 | /* | |
1179 | * The idle entry time is expected to be a sufficient approximation of | |
1180 | * the current time at this point. | |
1181 | */ | |
1182 | ktime_t now = ts->idle_entrytime; | |
1183 | ktime_t next_event; | |
1184 | ||
1185 | WARN_ON_ONCE(!ts->inidle); | |
1186 | ||
296bb1e5 RW |
1187 | *delta_next = ktime_sub(dev->next_event, now); |
1188 | ||
554c8aa8 | 1189 | if (!can_stop_idle_tick(cpu, ts)) |
296bb1e5 | 1190 | return *delta_next; |
554c8aa8 RW |
1191 | |
1192 | next_event = tick_nohz_next_event(ts, cpu); | |
1193 | if (!next_event) | |
296bb1e5 | 1194 | return *delta_next; |
554c8aa8 RW |
1195 | |
1196 | /* | |
1197 | * If the next highres timer to expire is earlier than next_event, the | |
1198 | * idle governor needs to know that. | |
1199 | */ | |
1200 | next_event = min_t(u64, next_event, | |
1201 | hrtimer_next_event_without(&ts->sched_timer)); | |
4f86d3a8 | 1202 | |
554c8aa8 | 1203 | return ktime_sub(next_event, now); |
4f86d3a8 LB |
1204 | } |
1205 | ||
466a2b42 JF |
1206 | /** |
1207 | * tick_nohz_get_idle_calls_cpu - return the current idle calls counter value | |
1208 | * for a particular CPU. | |
1209 | * | |
1210 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1211 | */ | |
1212 | unsigned long tick_nohz_get_idle_calls_cpu(int cpu) | |
1213 | { | |
1214 | struct tick_sched *ts = tick_get_tick_sched(cpu); | |
1215 | ||
1216 | return ts->idle_calls; | |
1217 | } | |
1218 | ||
b7eaf1aa RW |
1219 | /** |
1220 | * tick_nohz_get_idle_calls - return the current idle calls counter value | |
1221 | * | |
1222 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1223 | */ | |
1224 | unsigned long tick_nohz_get_idle_calls(void) | |
1225 | { | |
1226 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1227 | ||
1228 | return ts->idle_calls; | |
1229 | } | |
1230 | ||
96c9b903 YY |
1231 | static void tick_nohz_account_idle_time(struct tick_sched *ts, |
1232 | ktime_t now) | |
2ac0d98f | 1233 | { |
2ac0d98f | 1234 | unsigned long ticks; |
3f4724ea | 1235 | |
96c9b903 YY |
1236 | ts->idle_exittime = now; |
1237 | ||
e44fcb4b | 1238 | if (vtime_accounting_enabled_this_cpu()) |
3f4724ea | 1239 | return; |
79bf2bb3 TG |
1240 | /* |
1241 | * We stopped the tick in idle. Update process times would miss the | |
1242 | * time we slept as update_process_times does only a 1 tick | |
1243 | * accounting. Enforce that this is accounted to idle ! | |
1244 | */ | |
1245 | ticks = jiffies - ts->idle_jiffies; | |
1246 | /* | |
1247 | * We might be one off. Do not randomly account a huge number of ticks! | |
1248 | */ | |
79741dd3 MS |
1249 | if (ticks && ticks < LONG_MAX) |
1250 | account_idle_ticks(ticks); | |
19f5f736 FW |
1251 | } |
1252 | ||
a5183862 | 1253 | void tick_nohz_idle_restart_tick(void) |
2aaf709a | 1254 | { |
a5183862 YY |
1255 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
1256 | ||
1257 | if (ts->tick_stopped) { | |
96c9b903 YY |
1258 | ktime_t now = ktime_get(); |
1259 | tick_nohz_restart_sched_tick(ts, now); | |
1260 | tick_nohz_account_idle_time(ts, now); | |
a5183862 | 1261 | } |
2aaf709a RW |
1262 | } |
1263 | ||
a5183862 | 1264 | static void tick_nohz_idle_update_tick(struct tick_sched *ts, ktime_t now) |
2aaf709a | 1265 | { |
a5183862 YY |
1266 | if (tick_nohz_full_cpu(smp_processor_id())) |
1267 | __tick_nohz_full_update_tick(ts, now); | |
1268 | else | |
1269 | tick_nohz_restart_sched_tick(ts, now); | |
2aaf709a | 1270 | |
96c9b903 | 1271 | tick_nohz_account_idle_time(ts, now); |
2aaf709a RW |
1272 | } |
1273 | ||
79bf2bb3 | 1274 | /** |
280f0677 | 1275 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1276 | * |
1277 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1278 | * This also exit the RCU extended quiescent state. The CPU |
1279 | * can use RCU again after this function is called. | |
79bf2bb3 | 1280 | */ |
280f0677 | 1281 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1282 | { |
4a32fea9 | 1283 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
bbe9a70a | 1284 | bool idle_active, tick_stopped; |
6378ddb5 | 1285 | ktime_t now; |
79bf2bb3 | 1286 | |
6378ddb5 | 1287 | local_irq_disable(); |
2bbb6817 | 1288 | |
15f827be | 1289 | WARN_ON_ONCE(!ts->inidle); |
23a8d888 | 1290 | WARN_ON_ONCE(ts->timer_expires_base); |
15f827be FW |
1291 | |
1292 | ts->inidle = 0; | |
bbe9a70a AB |
1293 | idle_active = ts->idle_active; |
1294 | tick_stopped = ts->tick_stopped; | |
15f827be | 1295 | |
bbe9a70a | 1296 | if (idle_active || tick_stopped) |
eed3b9cf MS |
1297 | now = ktime_get(); |
1298 | ||
bbe9a70a | 1299 | if (idle_active) |
e8fcaa5c | 1300 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1301 | |
bbe9a70a | 1302 | if (tick_stopped) |
a5183862 | 1303 | tick_nohz_idle_update_tick(ts, now); |
79bf2bb3 | 1304 | |
79bf2bb3 TG |
1305 | local_irq_enable(); |
1306 | } | |
1307 | ||
79bf2bb3 TG |
1308 | /* |
1309 | * The nohz low res interrupt handler | |
1310 | */ | |
1311 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1312 | { | |
22127e93 | 1313 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1314 | struct pt_regs *regs = get_irq_regs(); |
1315 | ktime_t now = ktime_get(); | |
1316 | ||
2456e855 | 1317 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1318 | |
ff7de620 | 1319 | tick_sched_do_timer(ts, now); |
9e8f559b | 1320 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1321 | |
b5e995e6 VK |
1322 | /* No need to reprogram if we are running tickless */ |
1323 | if (unlikely(ts->tick_stopped)) | |
1324 | return; | |
1325 | ||
b9965449 | 1326 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
0ff53d09 | 1327 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
79bf2bb3 TG |
1328 | } |
1329 | ||
bc7a34b8 TG |
1330 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1331 | { | |
1332 | if (!tick_nohz_enabled) | |
1333 | return; | |
1334 | ts->nohz_mode = mode; | |
1335 | /* One update is enough */ | |
1336 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
ae67bada | 1337 | timers_update_nohz(); |
bc7a34b8 TG |
1338 | } |
1339 | ||
79bf2bb3 TG |
1340 | /** |
1341 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1342 | */ | |
1343 | static void tick_nohz_switch_to_nohz(void) | |
1344 | { | |
22127e93 | 1345 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1346 | ktime_t next; |
1347 | ||
27630532 | 1348 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1349 | return; |
1350 | ||
6b442bc8 | 1351 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1352 | return; |
6b442bc8 | 1353 | |
79bf2bb3 TG |
1354 | /* |
1355 | * Recycle the hrtimer in ts, so we can share the | |
1356 | * hrtimer_forward with the highres code. | |
1357 | */ | |
71fed982 | 1358 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 TG |
1359 | /* Get the next period */ |
1360 | next = tick_init_jiffy_update(); | |
1361 | ||
0ff53d09 | 1362 | hrtimer_set_expires(&ts->sched_timer, next); |
b9965449 | 1363 | hrtimer_forward_now(&ts->sched_timer, TICK_NSEC); |
1ca8ec53 | 1364 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
bc7a34b8 | 1365 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1366 | } |
1367 | ||
5acac1be | 1368 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1369 | { |
4a32fea9 | 1370 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1371 | ktime_t now; |
1372 | ||
1373 | if (!ts->idle_active && !ts->tick_stopped) | |
1374 | return; | |
1375 | now = ktime_get(); | |
1376 | if (ts->idle_active) | |
e8fcaa5c | 1377 | tick_nohz_stop_idle(ts, now); |
ff006732 | 1378 | if (ts->tick_stopped) |
eed3b9cf | 1379 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1380 | } |
1381 | ||
79bf2bb3 TG |
1382 | #else |
1383 | ||
1384 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1385 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1386 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1387 | |
3451d024 | 1388 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1389 | |
719254fa TG |
1390 | /* |
1391 | * Called from irq_enter to notify about the possible interruption of idle() | |
1392 | */ | |
5acac1be | 1393 | void tick_irq_enter(void) |
719254fa | 1394 | { |
e8fcaa5c | 1395 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1396 | tick_nohz_irq_enter(); |
719254fa TG |
1397 | } |
1398 | ||
79bf2bb3 TG |
1399 | /* |
1400 | * High resolution timer specific code | |
1401 | */ | |
1402 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1403 | /* | |
4c9dc641 | 1404 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1405 | * Called with interrupts disabled. |
79bf2bb3 TG |
1406 | */ |
1407 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1408 | { | |
1409 | struct tick_sched *ts = | |
1410 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1411 | struct pt_regs *regs = get_irq_regs(); |
1412 | ktime_t now = ktime_get(); | |
d3ed7824 | 1413 | |
ff7de620 | 1414 | tick_sched_do_timer(ts, now); |
79bf2bb3 TG |
1415 | |
1416 | /* | |
1417 | * Do not call, when we are not in irq context and have | |
1418 | * no valid regs pointer | |
1419 | */ | |
9e8f559b FW |
1420 | if (regs) |
1421 | tick_sched_handle(ts, regs); | |
7c259045 FW |
1422 | else |
1423 | ts->next_tick = 0; | |
79bf2bb3 | 1424 | |
2a16fc93 VK |
1425 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1426 | if (unlikely(ts->tick_stopped)) | |
1427 | return HRTIMER_NORESTART; | |
1428 | ||
b9965449 | 1429 | hrtimer_forward(timer, now, TICK_NSEC); |
79bf2bb3 TG |
1430 | |
1431 | return HRTIMER_RESTART; | |
1432 | } | |
1433 | ||
5307c955 MG |
1434 | static int sched_skew_tick; |
1435 | ||
62cf20b3 TG |
1436 | static int __init skew_tick(char *str) |
1437 | { | |
1438 | get_option(&str, &sched_skew_tick); | |
1439 | ||
1440 | return 0; | |
1441 | } | |
1442 | early_param("skew_tick", skew_tick); | |
1443 | ||
79bf2bb3 TG |
1444 | /** |
1445 | * tick_setup_sched_timer - setup the tick emulation timer | |
1446 | */ | |
1447 | void tick_setup_sched_timer(void) | |
1448 | { | |
22127e93 | 1449 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1450 | ktime_t now = ktime_get(); |
1451 | ||
1452 | /* | |
1453 | * Emulate tick processing via per-CPU hrtimers: | |
1454 | */ | |
902a9f9c | 1455 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 | 1456 | ts->sched_timer.function = tick_sched_timer; |
79bf2bb3 | 1457 | |
0de7611a | 1458 | /* Get the next period (per-CPU) */ |
cc584b21 | 1459 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1460 | |
9c3f9e28 | 1461 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 | 1462 | if (sched_skew_tick) { |
b9965449 | 1463 | u64 offset = TICK_NSEC >> 1; |
5307c955 MG |
1464 | do_div(offset, num_possible_cpus()); |
1465 | offset *= smp_processor_id(); | |
1466 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1467 | } | |
1468 | ||
b9965449 | 1469 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
902a9f9c | 1470 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED_HARD); |
bc7a34b8 | 1471 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1472 | } |
3c4fbe5e | 1473 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1474 | |
3451d024 | 1475 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1476 | void tick_cancel_sched_timer(int cpu) |
1477 | { | |
1478 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1479 | ||
3c4fbe5e | 1480 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1481 | if (ts->sched_timer.base) |
1482 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1483 | # endif |
a7901766 | 1484 | |
4b0c0f29 | 1485 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1486 | } |
3c4fbe5e | 1487 | #endif |
79bf2bb3 TG |
1488 | |
1489 | /** | |
1490 | * Async notification about clocksource changes | |
1491 | */ | |
1492 | void tick_clock_notify(void) | |
1493 | { | |
1494 | int cpu; | |
1495 | ||
1496 | for_each_possible_cpu(cpu) | |
1497 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * Async notification about clock event changes | |
1502 | */ | |
1503 | void tick_oneshot_notify(void) | |
1504 | { | |
22127e93 | 1505 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1506 | |
1507 | set_bit(0, &ts->check_clocks); | |
1508 | } | |
1509 | ||
1510 | /** | |
1511 | * Check, if a change happened, which makes oneshot possible. | |
1512 | * | |
1513 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1514 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1515 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1516 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1517 | */ |
1518 | int tick_check_oneshot_change(int allow_nohz) | |
1519 | { | |
22127e93 | 1520 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1521 | |
1522 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1523 | return 0; | |
1524 | ||
1525 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1526 | return 0; | |
1527 | ||
cf4fc6cb | 1528 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1529 | return 0; |
1530 | ||
1531 | if (!allow_nohz) | |
1532 | return 1; | |
1533 | ||
1534 | tick_nohz_switch_to_nohz(); | |
1535 | return 0; | |
1536 | } |