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> |
8083e4ad | 23 | #include <linux/module.h> |
00b42959 | 24 | #include <linux/irq_work.h> |
9014c45d | 25 | #include <linux/posix-timers.h> |
2e709338 | 26 | #include <linux/context_tracking.h> |
62cb1188 | 27 | #include <linux/mm.h> |
79bf2bb3 | 28 | |
9e203bcc DM |
29 | #include <asm/irq_regs.h> |
30 | ||
79bf2bb3 TG |
31 | #include "tick-internal.h" |
32 | ||
cb41a290 FW |
33 | #include <trace/events/timer.h> |
34 | ||
79bf2bb3 | 35 | /* |
0de7611a | 36 | * Per-CPU nohz control structure |
79bf2bb3 | 37 | */ |
c1797baf | 38 | static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); |
79bf2bb3 | 39 | |
289f480a IM |
40 | struct tick_sched *tick_get_tick_sched(int cpu) |
41 | { | |
42 | return &per_cpu(tick_cpu_sched, cpu); | |
43 | } | |
44 | ||
7809998a AB |
45 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
46 | /* | |
c398960c TG |
47 | * The time, when the last jiffy update happened. Write access must hold |
48 | * jiffies_lock and jiffies_seq. tick_nohz_next_event() needs to get a | |
49 | * consistent view of jiffies and last_jiffies_update. | |
7809998a AB |
50 | */ |
51 | static ktime_t last_jiffies_update; | |
52 | ||
79bf2bb3 TG |
53 | /* |
54 | * Must be called with interrupts disabled ! | |
55 | */ | |
56 | static void tick_do_update_jiffies64(ktime_t now) | |
57 | { | |
7a35bf2a | 58 | unsigned long ticks = 1; |
79bf2bb3 TG |
59 | ktime_t delta; |
60 | ||
7a14ce1d | 61 | /* |
372acbba TG |
62 | * Do a quick check without holding jiffies_lock. The READ_ONCE() |
63 | * pairs with the update done later in this function. | |
64 | * | |
65 | * This is also an intentional data race which is even safe on | |
66 | * 32bit in theory. If there is a concurrent update then the check | |
67 | * might give a random answer. It does not matter because if it | |
68 | * returns then the concurrent update is already taking care, if it | |
69 | * falls through then it will pointlessly contend on jiffies_lock. | |
70 | * | |
71 | * Though there is one nasty case on 32bit due to store tearing of | |
72 | * the 64bit value. If the first 32bit store makes the quick check | |
73 | * return on all other CPUs and the writing CPU context gets | |
74 | * delayed to complete the second store (scheduled out on virt) | |
75 | * then jiffies can become stale for up to ~2^32 nanoseconds | |
76 | * without noticing. After that point all CPUs will wait for | |
77 | * jiffies lock. | |
78 | * | |
79 | * OTOH, this is not any different than the situation with NOHZ=off | |
80 | * where one CPU is responsible for updating jiffies and | |
81 | * timekeeping. If that CPU goes out for lunch then all other CPUs | |
82 | * will operate on stale jiffies until it decides to come back. | |
7a14ce1d | 83 | */ |
372acbba | 84 | if (ktime_before(now, READ_ONCE(tick_next_period))) |
7a14ce1d IM |
85 | return; |
86 | ||
6168f8ed | 87 | /* Reevaluate with jiffies_lock held */ |
e5d4d175 | 88 | raw_spin_lock(&jiffies_lock); |
94ad2e3c | 89 | if (ktime_before(now, tick_next_period)) { |
e5d4d175 | 90 | raw_spin_unlock(&jiffies_lock); |
03e6bdc5 | 91 | return; |
79bf2bb3 | 92 | } |
94ad2e3c YY |
93 | |
94 | write_seqcount_begin(&jiffies_seq); | |
95 | ||
94ad2e3c YY |
96 | delta = ktime_sub(now, tick_next_period); |
97 | if (unlikely(delta >= tick_period)) { | |
98 | /* Slow path for long idle sleep times */ | |
99 | s64 incr = ktime_to_ns(tick_period); | |
100 | ||
7a35bf2a | 101 | ticks += ktime_divns(delta, incr); |
94ad2e3c YY |
102 | |
103 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
104 | incr * ticks); | |
7a35bf2a TG |
105 | } else { |
106 | last_jiffies_update = ktime_add(last_jiffies_update, | |
107 | tick_period); | |
94ad2e3c YY |
108 | } |
109 | ||
7a35bf2a | 110 | do_timer(ticks); |
94ad2e3c YY |
111 | |
112 | /* | |
113 | * Keep the tick_next_period variable up to date. WRITE_ONCE() | |
114 | * pairs with the READ_ONCE() in the lockless quick check above. | |
115 | */ | |
116 | WRITE_ONCE(tick_next_period, | |
117 | ktime_add(last_jiffies_update, tick_period)); | |
118 | ||
e5d4d175 TG |
119 | write_seqcount_end(&jiffies_seq); |
120 | raw_spin_unlock(&jiffies_lock); | |
47a1b796 | 121 | update_wall_time(); |
79bf2bb3 TG |
122 | } |
123 | ||
124 | /* | |
125 | * Initialize and return retrieve the jiffies update. | |
126 | */ | |
127 | static ktime_t tick_init_jiffy_update(void) | |
128 | { | |
129 | ktime_t period; | |
130 | ||
e5d4d175 TG |
131 | raw_spin_lock(&jiffies_lock); |
132 | write_seqcount_begin(&jiffies_seq); | |
79bf2bb3 | 133 | /* Did we start the jiffies update yet ? */ |
2456e855 | 134 | if (last_jiffies_update == 0) |
79bf2bb3 TG |
135 | last_jiffies_update = tick_next_period; |
136 | period = last_jiffies_update; | |
e5d4d175 TG |
137 | write_seqcount_end(&jiffies_seq); |
138 | raw_spin_unlock(&jiffies_lock); | |
79bf2bb3 TG |
139 | return period; |
140 | } | |
141 | ||
ff7de620 | 142 | static void tick_sched_do_timer(struct tick_sched *ts, ktime_t now) |
5bb96226 FW |
143 | { |
144 | int cpu = smp_processor_id(); | |
145 | ||
3451d024 | 146 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
147 | /* |
148 | * Check if the do_timer duty was dropped. We don't care about | |
0de7611a IM |
149 | * concurrency: This happens only when the CPU in charge went |
150 | * into a long sleep. If two CPUs happen to assign themselves to | |
5bb96226 | 151 | * this duty, then the jiffies update is still serialized by |
9c3f9e28 | 152 | * jiffies_lock. |
08ae95f4 NP |
153 | * |
154 | * If nohz_full is enabled, this should not happen because the | |
155 | * tick_do_timer_cpu never relinquishes. | |
5bb96226 | 156 | */ |
08ae95f4 NP |
157 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) { |
158 | #ifdef CONFIG_NO_HZ_FULL | |
159 | WARN_ON(tick_nohz_full_running); | |
160 | #endif | |
5bb96226 | 161 | tick_do_timer_cpu = cpu; |
08ae95f4 | 162 | } |
5bb96226 FW |
163 | #endif |
164 | ||
165 | /* Check, if the jiffies need an update */ | |
166 | if (tick_do_timer_cpu == cpu) | |
167 | tick_do_update_jiffies64(now); | |
ff7de620 RW |
168 | |
169 | if (ts->inidle) | |
170 | ts->got_idle_tick = 1; | |
5bb96226 FW |
171 | } |
172 | ||
9e8f559b FW |
173 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
174 | { | |
3451d024 | 175 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
176 | /* |
177 | * When we are idle and the tick is stopped, we have to touch | |
178 | * the watchdog as we might not schedule for a really long | |
179 | * time. This happens on complete idle SMP systems while | |
180 | * waiting on the login prompt. We also increment the "start of | |
181 | * idle" jiffy stamp so the idle accounting adjustment we do | |
182 | * when we go busy again does not account too much ticks. | |
183 | */ | |
184 | if (ts->tick_stopped) { | |
03e0d461 | 185 | touch_softlockup_watchdog_sched(); |
9e8f559b FW |
186 | if (is_idle_task(current)) |
187 | ts->idle_jiffies++; | |
411fe24e FW |
188 | /* |
189 | * In case the current tick fired too early past its expected | |
190 | * expiration, make sure we don't bypass the next clock reprogramming | |
191 | * to the same deadline. | |
192 | */ | |
193 | ts->next_tick = 0; | |
9e8f559b | 194 | } |
94a57140 | 195 | #endif |
9e8f559b FW |
196 | update_process_times(user_mode(regs)); |
197 | profile_tick(CPU_PROFILING); | |
198 | } | |
7809998a | 199 | #endif |
9e8f559b | 200 | |
c5bfece2 | 201 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 202 | cpumask_var_t tick_nohz_full_mask; |
73867dcd | 203 | bool tick_nohz_full_running; |
ae9e557b | 204 | EXPORT_SYMBOL_GPL(tick_nohz_full_running); |
f009a7a7 | 205 | static atomic_t tick_dep_mask; |
a831881b | 206 | |
f009a7a7 | 207 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 208 | { |
f009a7a7 FW |
209 | int val = atomic_read(dep); |
210 | ||
211 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 212 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 213 | return true; |
d027d45d FW |
214 | } |
215 | ||
f009a7a7 | 216 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 217 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 218 | return true; |
d027d45d FW |
219 | } |
220 | ||
f009a7a7 | 221 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 222 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 223 | return true; |
d027d45d FW |
224 | } |
225 | ||
f009a7a7 | 226 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 227 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
228 | return true; |
229 | } | |
230 | ||
01b4c399 FW |
231 | if (val & TICK_DEP_MASK_RCU) { |
232 | trace_tick_stop(0, TICK_DEP_MASK_RCU); | |
233 | return true; | |
234 | } | |
235 | ||
f009a7a7 | 236 | return false; |
d027d45d FW |
237 | } |
238 | ||
57ccdf44 | 239 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d | 240 | { |
ebf3adba | 241 | lockdep_assert_irqs_disabled(); |
9014c45d | 242 | |
57ccdf44 WL |
243 | if (unlikely(!cpu_online(cpu))) |
244 | return false; | |
245 | ||
f009a7a7 | 246 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 247 | return false; |
d027d45d | 248 | |
f009a7a7 | 249 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 250 | return false; |
d027d45d | 251 | |
f009a7a7 | 252 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 253 | return false; |
d027d45d | 254 | |
f009a7a7 | 255 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 256 | return false; |
d027d45d | 257 | |
9014c45d FW |
258 | return true; |
259 | } | |
260 | ||
d027d45d | 261 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 262 | { |
73738a95 | 263 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
264 | } |
265 | ||
266 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = { | |
d027d45d | 267 | .func = nohz_full_kick_func, |
49915ac3 | 268 | .flags = ATOMIC_INIT(IRQ_WORK_HARD_IRQ), |
76c24fb0 FW |
269 | }; |
270 | ||
40bea039 FW |
271 | /* |
272 | * Kick this CPU if it's full dynticks in order to force it to | |
273 | * re-evaluate its dependency on the tick and restart it if necessary. | |
274 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
275 | * is NMI safe. | |
276 | */ | |
555e0c1e | 277 | static void tick_nohz_full_kick(void) |
40bea039 FW |
278 | { |
279 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
280 | return; | |
281 | ||
56e4dea8 | 282 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
283 | } |
284 | ||
76c24fb0 | 285 | /* |
3d36aebc | 286 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
287 | * re-evaluate its dependency on the tick and restart it if necessary. |
288 | */ | |
3d36aebc | 289 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 290 | { |
3d36aebc FW |
291 | if (!tick_nohz_full_cpu(cpu)) |
292 | return; | |
293 | ||
294 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
295 | } |
296 | ||
76c24fb0 FW |
297 | /* |
298 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
299 | * their dependency on the tick and restart it if necessary. | |
300 | */ | |
b7878300 | 301 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 302 | { |
8537bb95 FW |
303 | int cpu; |
304 | ||
73867dcd | 305 | if (!tick_nohz_full_running) |
76c24fb0 FW |
306 | return; |
307 | ||
308 | preempt_disable(); | |
8537bb95 FW |
309 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
310 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
311 | preempt_enable(); |
312 | } | |
313 | ||
f009a7a7 | 314 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
315 | enum tick_dep_bits bit) |
316 | { | |
f009a7a7 | 317 | int prev; |
d027d45d | 318 | |
a1cc5bcf | 319 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
320 | if (!prev) |
321 | tick_nohz_full_kick_all(); | |
322 | } | |
323 | ||
324 | /* | |
325 | * Set a global tick dependency. Used by perf events that rely on freq and | |
326 | * by unstable clock. | |
327 | */ | |
328 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
329 | { | |
330 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
331 | } | |
332 | ||
333 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
334 | { | |
f009a7a7 | 335 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
336 | } |
337 | ||
338 | /* | |
339 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
340 | * manage events throttling. | |
341 | */ | |
342 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
343 | { | |
f009a7a7 | 344 | int prev; |
d027d45d FW |
345 | struct tick_sched *ts; |
346 | ||
347 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
348 | ||
a1cc5bcf | 349 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
350 | if (!prev) { |
351 | preempt_disable(); | |
352 | /* Perf needs local kick that is NMI safe */ | |
353 | if (cpu == smp_processor_id()) { | |
354 | tick_nohz_full_kick(); | |
355 | } else { | |
356 | /* Remote irq work not NMI-safe */ | |
357 | if (!WARN_ON_ONCE(in_nmi())) | |
358 | tick_nohz_full_kick_cpu(cpu); | |
359 | } | |
360 | preempt_enable(); | |
361 | } | |
362 | } | |
01b4c399 | 363 | EXPORT_SYMBOL_GPL(tick_nohz_dep_set_cpu); |
d027d45d FW |
364 | |
365 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
366 | { | |
367 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
368 | ||
f009a7a7 | 369 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d | 370 | } |
01b4c399 | 371 | EXPORT_SYMBOL_GPL(tick_nohz_dep_clear_cpu); |
d027d45d FW |
372 | |
373 | /* | |
3c8920e2 FW |
374 | * Set a per-task tick dependency. RCU need this. Also posix CPU timers |
375 | * in order to elapse per task timers. | |
d027d45d FW |
376 | */ |
377 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
378 | { | |
3c8920e2 FW |
379 | if (!atomic_fetch_or(BIT(bit), &tsk->tick_dep_mask)) { |
380 | if (tsk == current) { | |
381 | preempt_disable(); | |
382 | tick_nohz_full_kick(); | |
383 | preempt_enable(); | |
384 | } else { | |
385 | /* | |
386 | * Some future tick_nohz_full_kick_task() | |
387 | * should optimize this. | |
388 | */ | |
389 | tick_nohz_full_kick_all(); | |
390 | } | |
391 | } | |
d027d45d | 392 | } |
ae9e557b | 393 | EXPORT_SYMBOL_GPL(tick_nohz_dep_set_task); |
d027d45d FW |
394 | |
395 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
396 | { | |
f009a7a7 | 397 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d | 398 | } |
ae9e557b | 399 | EXPORT_SYMBOL_GPL(tick_nohz_dep_clear_task); |
d027d45d FW |
400 | |
401 | /* | |
402 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
403 | * per process timers. | |
404 | */ | |
405 | void tick_nohz_dep_set_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
406 | { | |
407 | tick_nohz_dep_set_all(&sig->tick_dep_mask, bit); | |
408 | } | |
409 | ||
410 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
411 | { | |
f009a7a7 | 412 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
413 | } |
414 | ||
99e5ada9 FW |
415 | /* |
416 | * Re-evaluate the need for the tick as we switch the current task. | |
417 | * It might need the tick due to per task/process properties: | |
0de7611a | 418 | * perf events, posix CPU timers, ... |
99e5ada9 | 419 | */ |
de734f89 | 420 | void __tick_nohz_task_switch(void) |
99e5ada9 FW |
421 | { |
422 | unsigned long flags; | |
d027d45d | 423 | struct tick_sched *ts; |
99e5ada9 | 424 | |
99e5ada9 FW |
425 | local_irq_save(flags); |
426 | ||
6296ace4 LZ |
427 | if (!tick_nohz_full_cpu(smp_processor_id())) |
428 | goto out; | |
429 | ||
d027d45d | 430 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 431 | |
d027d45d | 432 | if (ts->tick_stopped) { |
f009a7a7 FW |
433 | if (atomic_read(¤t->tick_dep_mask) || |
434 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
435 | tick_nohz_full_kick(); |
436 | } | |
6296ace4 | 437 | out: |
99e5ada9 FW |
438 | local_irq_restore(flags); |
439 | } | |
440 | ||
6f1982fe FW |
441 | /* Get the boot-time nohz CPU list from the kernel parameters. */ |
442 | void __init tick_nohz_full_setup(cpumask_var_t cpumask) | |
a831881b | 443 | { |
73867dcd | 444 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
6f1982fe | 445 | cpumask_copy(tick_nohz_full_mask, cpumask); |
73867dcd | 446 | tick_nohz_full_running = true; |
a831881b | 447 | } |
ae9e557b | 448 | EXPORT_SYMBOL_GPL(tick_nohz_full_setup); |
a831881b | 449 | |
31eff243 | 450 | static int tick_nohz_cpu_down(unsigned int cpu) |
a382bf93 | 451 | { |
31eff243 | 452 | /* |
08ae95f4 NP |
453 | * The tick_do_timer_cpu CPU handles housekeeping duty (unbound |
454 | * timers, workqueues, timekeeping, ...) on behalf of full dynticks | |
31eff243 SAS |
455 | * CPUs. It must remain online when nohz full is enabled. |
456 | */ | |
457 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) | |
458 | return -EBUSY; | |
459 | return 0; | |
a382bf93 FW |
460 | } |
461 | ||
d1e43fa5 | 462 | void __init tick_nohz_init(void) |
a831881b | 463 | { |
31eff243 | 464 | int cpu, ret; |
d1e43fa5 | 465 | |
a7c8655b PM |
466 | if (!tick_nohz_full_running) |
467 | return; | |
d1e43fa5 | 468 | |
9b01f5bf FW |
469 | /* |
470 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
471 | * locking contexts. But then we need irq work to raise its own | |
472 | * interrupts to avoid circular dependency on the tick | |
473 | */ | |
474 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 475 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf | 476 | cpumask_clear(tick_nohz_full_mask); |
9b01f5bf FW |
477 | tick_nohz_full_running = false; |
478 | return; | |
479 | } | |
480 | ||
08ae95f4 NP |
481 | if (IS_ENABLED(CONFIG_PM_SLEEP_SMP) && |
482 | !IS_ENABLED(CONFIG_PM_SLEEP_SMP_NONZERO_CPU)) { | |
483 | cpu = smp_processor_id(); | |
4327b15f | 484 | |
08ae95f4 NP |
485 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { |
486 | pr_warn("NO_HZ: Clearing %d from nohz_full range " | |
487 | "for timekeeping\n", cpu); | |
488 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); | |
489 | } | |
4327b15f FW |
490 | } |
491 | ||
73867dcd | 492 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
493 | context_tracking_cpu_set(cpu); |
494 | ||
31eff243 SAS |
495 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
496 | "kernel/nohz:predown", NULL, | |
497 | tick_nohz_cpu_down); | |
498 | WARN_ON(ret < 0); | |
ffda22c1 TH |
499 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
500 | cpumask_pr_args(tick_nohz_full_mask)); | |
a831881b | 501 | } |
a831881b FW |
502 | #endif |
503 | ||
79bf2bb3 TG |
504 | /* |
505 | * NOHZ - aka dynamic tick functionality | |
506 | */ | |
3451d024 | 507 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
508 | /* |
509 | * NO HZ enabled ? | |
510 | */ | |
4cc7ecb7 | 511 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 512 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
513 | /* |
514 | * Enable / Disable tickless mode | |
515 | */ | |
516 | static int __init setup_tick_nohz(char *str) | |
517 | { | |
4cc7ecb7 | 518 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
519 | } |
520 | ||
521 | __setup("nohz=", setup_tick_nohz); | |
522 | ||
a3642983 | 523 | bool tick_nohz_tick_stopped(void) |
c1797baf | 524 | { |
2bc629a6 FW |
525 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
526 | ||
527 | return ts->tick_stopped; | |
c1797baf TG |
528 | } |
529 | ||
22ab8bc0 FW |
530 | bool tick_nohz_tick_stopped_cpu(int cpu) |
531 | { | |
532 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
533 | ||
534 | return ts->tick_stopped; | |
535 | } | |
536 | ||
79bf2bb3 TG |
537 | /** |
538 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
539 | * | |
540 | * Called from interrupt entry when the CPU was idle | |
541 | * | |
542 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
543 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
544 | * value. We do this unconditionally on any CPU, as we don't know whether the |
545 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 546 | */ |
eed3b9cf | 547 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 548 | { |
79bf2bb3 | 549 | unsigned long flags; |
79bf2bb3 | 550 | |
e8fcaa5c | 551 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
552 | |
553 | local_irq_save(flags); | |
554 | tick_do_update_jiffies64(now); | |
555 | local_irq_restore(flags); | |
02ff3755 | 556 | |
03e0d461 | 557 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
558 | } |
559 | ||
595aac48 | 560 | /* |
0de7611a | 561 | * Updates the per-CPU time idle statistics counters |
595aac48 | 562 | */ |
8d63bf94 | 563 | static void |
8c215bd3 | 564 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 565 | { |
eed3b9cf | 566 | ktime_t delta; |
6378ddb5 | 567 | |
595aac48 AV |
568 | if (ts->idle_active) { |
569 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 570 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 571 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
572 | else |
573 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 574 | ts->idle_entrytime = now; |
595aac48 | 575 | } |
8d63bf94 | 576 | |
e0e37c20 | 577 | if (last_update_time) |
8d63bf94 AV |
578 | *last_update_time = ktime_to_us(now); |
579 | ||
595aac48 AV |
580 | } |
581 | ||
e8fcaa5c | 582 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 583 | { |
e8fcaa5c | 584 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 585 | ts->idle_active = 0; |
56c7426b | 586 | |
ac1e843f | 587 | sched_clock_idle_wakeup_event(); |
6378ddb5 VP |
588 | } |
589 | ||
0e776768 | 590 | static void tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 591 | { |
0e776768 | 592 | ts->idle_entrytime = ktime_get(); |
6378ddb5 | 593 | ts->idle_active = 1; |
56c7426b | 594 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
595 | } |
596 | ||
b1f724c3 | 597 | /** |
0de7611a | 598 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 599 | * @cpu: CPU number to query |
09a1d34f MH |
600 | * @last_update_time: variable to store update time in. Do not update |
601 | * counters if NULL. | |
b1f724c3 | 602 | * |
6168f8ed | 603 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 604 | * CPU, in microseconds. |
b1f724c3 AV |
605 | * |
606 | * This time is measured via accounting rather than sampling, | |
607 | * and is as accurate as ktime_get() is. | |
608 | * | |
609 | * This function returns -1 if NOHZ is not enabled. | |
610 | */ | |
6378ddb5 VP |
611 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
612 | { | |
613 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 614 | ktime_t now, idle; |
6378ddb5 | 615 | |
d689fe22 | 616 | if (!tick_nohz_active) |
8083e4ad | 617 | return -1; |
618 | ||
09a1d34f MH |
619 | now = ktime_get(); |
620 | if (last_update_time) { | |
621 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
622 | idle = ts->idle_sleeptime; | |
623 | } else { | |
624 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
625 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
626 | ||
627 | idle = ktime_add(ts->idle_sleeptime, delta); | |
628 | } else { | |
629 | idle = ts->idle_sleeptime; | |
630 | } | |
631 | } | |
632 | ||
633 | return ktime_to_us(idle); | |
8083e4ad | 634 | |
6378ddb5 | 635 | } |
8083e4ad | 636 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 637 | |
6beea0cd | 638 | /** |
0de7611a | 639 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 640 | * @cpu: CPU number to query |
09a1d34f MH |
641 | * @last_update_time: variable to store update time in. Do not update |
642 | * counters if NULL. | |
0224cf4c | 643 | * |
6168f8ed | 644 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
645 | * CPU, in microseconds. |
646 | * | |
647 | * This time is measured via accounting rather than sampling, | |
648 | * and is as accurate as ktime_get() is. | |
649 | * | |
650 | * This function returns -1 if NOHZ is not enabled. | |
651 | */ | |
652 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
653 | { | |
654 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 655 | ktime_t now, iowait; |
0224cf4c | 656 | |
d689fe22 | 657 | if (!tick_nohz_active) |
0224cf4c AV |
658 | return -1; |
659 | ||
09a1d34f MH |
660 | now = ktime_get(); |
661 | if (last_update_time) { | |
662 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
663 | iowait = ts->iowait_sleeptime; | |
664 | } else { | |
665 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
666 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 667 | |
09a1d34f MH |
668 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
669 | } else { | |
670 | iowait = ts->iowait_sleeptime; | |
671 | } | |
672 | } | |
0224cf4c | 673 | |
09a1d34f | 674 | return ktime_to_us(iowait); |
0224cf4c AV |
675 | } |
676 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
677 | ||
0ff53d09 TG |
678 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
679 | { | |
680 | hrtimer_cancel(&ts->sched_timer); | |
681 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
682 | ||
683 | /* Forward the time to expire in the future */ | |
684 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
685 | ||
902a9f9c SAS |
686 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
687 | hrtimer_start_expires(&ts->sched_timer, | |
688 | HRTIMER_MODE_ABS_PINNED_HARD); | |
689 | } else { | |
0ff53d09 | 690 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
902a9f9c | 691 | } |
411fe24e FW |
692 | |
693 | /* | |
694 | * Reset to make sure next tick stop doesn't get fooled by past | |
695 | * cached clock deadline. | |
696 | */ | |
697 | ts->next_tick = 0; | |
0ff53d09 TG |
698 | } |
699 | ||
5d62c183 TG |
700 | static inline bool local_timer_softirq_pending(void) |
701 | { | |
80d20d35 | 702 | return local_softirq_pending() & BIT(TIMER_SOFTIRQ); |
5d62c183 TG |
703 | } |
704 | ||
23a8d888 | 705 | static ktime_t tick_nohz_next_event(struct tick_sched *ts, int cpu) |
79bf2bb3 | 706 | { |
c1ad348b | 707 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
e1e41b6c RV |
708 | unsigned long basejiff; |
709 | unsigned int seq; | |
855a0fc3 | 710 | |
79bf2bb3 TG |
711 | /* Read jiffies and the time when jiffies were updated last */ |
712 | do { | |
e5d4d175 | 713 | seq = read_seqcount_begin(&jiffies_seq); |
2456e855 | 714 | basemono = last_jiffies_update; |
c1ad348b | 715 | basejiff = jiffies; |
e5d4d175 | 716 | } while (read_seqcount_retry(&jiffies_seq, seq)); |
c1ad348b | 717 | ts->last_jiffies = basejiff; |
23a8d888 | 718 | ts->timer_expires_base = basemono; |
79bf2bb3 | 719 | |
5d62c183 TG |
720 | /* |
721 | * Keep the periodic tick, when RCU, architecture or irq_work | |
722 | * requests it. | |
723 | * Aside of that check whether the local timer softirq is | |
724 | * pending. If so its a bad idea to call get_next_timer_interrupt() | |
725 | * because there is an already expired timer, so it will request | |
726 | * immeditate expiry, which rearms the hardware timer with a | |
727 | * minimal delta which brings us back to this place | |
728 | * immediately. Lather, rinse and repeat... | |
729 | */ | |
730 | if (rcu_needs_cpu(basemono, &next_rcu) || arch_needs_cpu() || | |
731 | irq_work_needs_cpu() || local_timer_softirq_pending()) { | |
c1ad348b | 732 | next_tick = basemono + TICK_NSEC; |
3c5d92a0 | 733 | } else { |
c1ad348b TG |
734 | /* |
735 | * Get the next pending timer. If high resolution | |
736 | * timers are enabled this only takes the timer wheel | |
737 | * timers into account. If high resolution timers are | |
738 | * disabled this also looks at the next expiring | |
739 | * hrtimer. | |
740 | */ | |
741 | next_tmr = get_next_timer_interrupt(basejiff, basemono); | |
742 | ts->next_timer = next_tmr; | |
743 | /* Take the next rcu event into account */ | |
744 | next_tick = next_rcu < next_tmr ? next_rcu : next_tmr; | |
3c5d92a0 | 745 | } |
47aa8b6c | 746 | |
c1ad348b TG |
747 | /* |
748 | * If the tick is due in the next period, keep it ticking or | |
82bbe34b | 749 | * force prod the timer. |
c1ad348b TG |
750 | */ |
751 | delta = next_tick - basemono; | |
752 | if (delta <= (u64)TICK_NSEC) { | |
a683f390 TG |
753 | /* |
754 | * Tell the timer code that the base is not idle, i.e. undo | |
755 | * the effect of get_next_timer_interrupt(): | |
756 | */ | |
757 | timer_clear_idle(); | |
82bbe34b PZ |
758 | /* |
759 | * We've not stopped the tick yet, and there's a timer in the | |
760 | * next period, so no point in stopping it either, bail. | |
761 | */ | |
f99973e1 | 762 | if (!ts->tick_stopped) { |
23a8d888 | 763 | ts->timer_expires = 0; |
157d29e1 TG |
764 | goto out; |
765 | } | |
766 | } | |
767 | ||
23a8d888 RW |
768 | /* |
769 | * If this CPU is the one which had the do_timer() duty last, we limit | |
770 | * the sleep time to the timekeeping max_deferment value. | |
771 | * Otherwise we can sleep as long as we want. | |
772 | */ | |
773 | delta = timekeeping_max_deferment(); | |
774 | if (cpu != tick_do_timer_cpu && | |
775 | (tick_do_timer_cpu != TICK_DO_TIMER_NONE || !ts->do_timer_last)) | |
776 | delta = KTIME_MAX; | |
777 | ||
778 | /* Calculate the next expiry time */ | |
779 | if (delta < (KTIME_MAX - basemono)) | |
780 | expires = basemono + delta; | |
781 | else | |
782 | expires = KTIME_MAX; | |
783 | ||
784 | ts->timer_expires = min_t(u64, expires, next_tick); | |
785 | ||
786 | out: | |
787 | return ts->timer_expires; | |
788 | } | |
789 | ||
790 | static void tick_nohz_stop_tick(struct tick_sched *ts, int cpu) | |
791 | { | |
792 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); | |
793 | u64 basemono = ts->timer_expires_base; | |
794 | u64 expires = ts->timer_expires; | |
795 | ktime_t tick = expires; | |
796 | ||
797 | /* Make sure we won't be trying to stop it twice in a row. */ | |
798 | ts->timer_expires_base = 0; | |
799 | ||
79bf2bb3 | 800 | /* |
0de7611a IM |
801 | * If this CPU is the one which updates jiffies, then give up |
802 | * the assignment and let it be taken by the CPU which runs | |
803 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
804 | * don't drop this here the jiffies might be stale and |
805 | * do_timer() never invoked. Keep track of the fact that it | |
23a8d888 | 806 | * was the one which had the do_timer() duty last. |
79bf2bb3 | 807 | */ |
157d29e1 TG |
808 | if (cpu == tick_do_timer_cpu) { |
809 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
810 | ts->do_timer_last = 1; | |
811 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
157d29e1 | 812 | ts->do_timer_last = 0; |
157d29e1 | 813 | } |
27185016 | 814 | |
157d29e1 | 815 | /* Skip reprogram of event if its not changed */ |
411fe24e FW |
816 | if (ts->tick_stopped && (expires == ts->next_tick)) { |
817 | /* Sanity check: make sure clockevent is actually programmed */ | |
d4af6d93 | 818 | if (tick == KTIME_MAX || ts->next_tick == hrtimer_get_expires(&ts->sched_timer)) |
23a8d888 | 819 | return; |
411fe24e FW |
820 | |
821 | WARN_ON_ONCE(1); | |
822 | printk_once("basemono: %llu ts->next_tick: %llu dev->next_event: %llu timer->active: %d timer->expires: %llu\n", | |
823 | basemono, ts->next_tick, dev->next_event, | |
824 | hrtimer_active(&ts->sched_timer), hrtimer_get_expires(&ts->sched_timer)); | |
ce6cf9a1 | 825 | } |
84bf1bcc | 826 | |
157d29e1 TG |
827 | /* |
828 | * nohz_stop_sched_tick can be called several times before | |
829 | * the nohz_restart_sched_tick is called. This happens when | |
830 | * interrupts arrive which do not cause a reschedule. In the | |
831 | * first call we save the current tick time, so we can restart | |
832 | * the scheduler tick in nohz_restart_sched_tick. | |
833 | */ | |
834 | if (!ts->tick_stopped) { | |
3c85d6db | 835 | calc_load_nohz_start(); |
62cb1188 | 836 | quiet_vmstat(); |
d3ed7824 | 837 | |
157d29e1 TG |
838 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
839 | ts->tick_stopped = 1; | |
e6e6cc22 | 840 | trace_tick_stop(1, TICK_DEP_MASK_NONE); |
157d29e1 | 841 | } |
eaad084b | 842 | |
411fe24e FW |
843 | ts->next_tick = tick; |
844 | ||
157d29e1 | 845 | /* |
c1ad348b TG |
846 | * If the expiration time == KTIME_MAX, then we simply stop |
847 | * the tick timer. | |
157d29e1 | 848 | */ |
c1ad348b | 849 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
850 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
851 | hrtimer_cancel(&ts->sched_timer); | |
23a8d888 | 852 | return; |
79bf2bb3 | 853 | } |
0ff53d09 | 854 | |
1f71addd | 855 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
902a9f9c SAS |
856 | hrtimer_start(&ts->sched_timer, tick, |
857 | HRTIMER_MODE_ABS_PINNED_HARD); | |
1f71addd TG |
858 | } else { |
859 | hrtimer_set_expires(&ts->sched_timer, tick); | |
c1ad348b | 860 | tick_program_event(tick, 1); |
1f71addd | 861 | } |
280f0677 FW |
862 | } |
863 | ||
23a8d888 RW |
864 | static void tick_nohz_retain_tick(struct tick_sched *ts) |
865 | { | |
866 | ts->timer_expires_base = 0; | |
867 | } | |
868 | ||
869 | #ifdef CONFIG_NO_HZ_FULL | |
870 | static void tick_nohz_stop_sched_tick(struct tick_sched *ts, int cpu) | |
871 | { | |
872 | if (tick_nohz_next_event(ts, cpu)) | |
873 | tick_nohz_stop_tick(ts, cpu); | |
874 | else | |
875 | tick_nohz_retain_tick(ts); | |
876 | } | |
877 | #endif /* CONFIG_NO_HZ_FULL */ | |
878 | ||
1f41906a | 879 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
880 | { |
881 | /* Update jiffies first */ | |
882 | tick_do_update_jiffies64(now); | |
59d2c7ca | 883 | |
a683f390 TG |
884 | /* |
885 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
886 | * the clock forward checks in the enqueue path: | |
887 | */ | |
888 | timer_clear_idle(); | |
889 | ||
3c85d6db | 890 | calc_load_nohz_stop(); |
03e0d461 | 891 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
892 | /* |
893 | * Cancel the scheduled timer and restore the tick | |
894 | */ | |
895 | ts->tick_stopped = 0; | |
896 | ts->idle_exittime = now; | |
897 | ||
898 | tick_nohz_restart(ts, now); | |
899 | } | |
73738a95 FW |
900 | |
901 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
5811d996 FW |
902 | { |
903 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 904 | int cpu = smp_processor_id(); |
5811d996 | 905 | |
59449359 | 906 | if (!tick_nohz_full_cpu(cpu)) |
e9a2eb40 | 907 | return; |
5811d996 | 908 | |
e9a2eb40 AS |
909 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
910 | return; | |
5811d996 | 911 | |
57ccdf44 | 912 | if (can_stop_full_tick(cpu, ts)) |
23a8d888 | 913 | tick_nohz_stop_sched_tick(ts, cpu); |
73738a95 | 914 | else if (ts->tick_stopped) |
1f41906a | 915 | tick_nohz_restart_sched_tick(ts, ktime_get()); |
5811d996 FW |
916 | #endif |
917 | } | |
918 | ||
5b39939a FW |
919 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
920 | { | |
921 | /* | |
0de7611a | 922 | * If this CPU is offline and it is the one which updates |
5b39939a | 923 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 924 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
925 | * this here the jiffies might be stale and do_timer() never |
926 | * invoked. | |
927 | */ | |
928 | if (unlikely(!cpu_online(cpu))) { | |
929 | if (cpu == tick_do_timer_cpu) | |
930 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
411fe24e FW |
931 | /* |
932 | * Make sure the CPU doesn't get fooled by obsolete tick | |
933 | * deadline if it comes back online later. | |
934 | */ | |
935 | ts->next_tick = 0; | |
f7ea0fd6 | 936 | return false; |
5b39939a FW |
937 | } |
938 | ||
23a8d888 | 939 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) |
5b39939a FW |
940 | return false; |
941 | ||
942 | if (need_resched()) | |
943 | return false; | |
944 | ||
d59e0ba1 | 945 | if (unlikely(local_softirq_pending())) { |
5b39939a FW |
946 | static int ratelimit; |
947 | ||
803b0eba PM |
948 | if (ratelimit < 10 && |
949 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
bca37119 | 950 | pr_warn("NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #%02x!!!\n", |
cfea7d7e | 951 | (unsigned int) local_softirq_pending()); |
5b39939a FW |
952 | ratelimit++; |
953 | } | |
954 | return false; | |
955 | } | |
956 | ||
460775df | 957 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
958 | /* |
959 | * Keep the tick alive to guarantee timekeeping progression | |
960 | * if there are full dynticks CPUs around | |
961 | */ | |
962 | if (tick_do_timer_cpu == cpu) | |
963 | return false; | |
964 | /* | |
965 | * Boot safety: make sure the timekeeping duty has been | |
966 | * assigned before entering dyntick-idle mode, | |
08ae95f4 | 967 | * tick_do_timer_cpu is TICK_DO_TIMER_BOOT |
a382bf93 | 968 | */ |
08ae95f4 NP |
969 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_BOOT)) |
970 | return false; | |
971 | ||
972 | /* Should not happen for nohz-full */ | |
973 | if (WARN_ON_ONCE(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) | |
a382bf93 FW |
974 | return false; |
975 | } | |
976 | ||
5b39939a FW |
977 | return true; |
978 | } | |
979 | ||
0e776768 | 980 | static void __tick_nohz_idle_stop_tick(struct tick_sched *ts) |
19f5f736 | 981 | { |
0e776768 | 982 | ktime_t expires; |
5b39939a | 983 | int cpu = smp_processor_id(); |
19f5f736 | 984 | |
554c8aa8 RW |
985 | /* |
986 | * If tick_nohz_get_sleep_length() ran tick_nohz_next_event(), the | |
987 | * tick timer expiration time is known already. | |
988 | */ | |
989 | if (ts->timer_expires_base) | |
990 | expires = ts->timer_expires; | |
991 | else if (can_stop_idle_tick(cpu, ts)) | |
992 | expires = tick_nohz_next_event(ts, cpu); | |
993 | else | |
994 | return; | |
23a8d888 RW |
995 | |
996 | ts->idle_calls++; | |
08d07259 | 997 | |
23a8d888 | 998 | if (expires > 0LL) { |
5b39939a FW |
999 | int was_stopped = ts->tick_stopped; |
1000 | ||
23a8d888 | 1001 | tick_nohz_stop_tick(ts, cpu); |
84bf1bcc | 1002 | |
23a8d888 RW |
1003 | ts->idle_sleeps++; |
1004 | ts->idle_expires = expires; | |
5b39939a | 1005 | |
a0db971e | 1006 | if (!was_stopped && ts->tick_stopped) { |
5b39939a | 1007 | ts->idle_jiffies = ts->last_jiffies; |
a0db971e FW |
1008 | nohz_balance_enter_idle(cpu); |
1009 | } | |
23a8d888 RW |
1010 | } else { |
1011 | tick_nohz_retain_tick(ts); | |
5b39939a | 1012 | } |
280f0677 FW |
1013 | } |
1014 | ||
1015 | /** | |
0e776768 | 1016 | * tick_nohz_idle_stop_tick - stop the idle tick from the idle task |
280f0677 FW |
1017 | * |
1018 | * When the next event is more than a tick into the future, stop the idle tick | |
0e776768 RW |
1019 | */ |
1020 | void tick_nohz_idle_stop_tick(void) | |
1021 | { | |
1022 | __tick_nohz_idle_stop_tick(this_cpu_ptr(&tick_cpu_sched)); | |
1023 | } | |
1024 | ||
554c8aa8 RW |
1025 | void tick_nohz_idle_retain_tick(void) |
1026 | { | |
1027 | tick_nohz_retain_tick(this_cpu_ptr(&tick_cpu_sched)); | |
1028 | /* | |
1029 | * Undo the effect of get_next_timer_interrupt() called from | |
1030 | * tick_nohz_next_event(). | |
1031 | */ | |
1032 | timer_clear_idle(); | |
1033 | } | |
1034 | ||
0e776768 RW |
1035 | /** |
1036 | * tick_nohz_idle_enter - prepare for entering idle on the current CPU | |
2bbb6817 | 1037 | * |
0e776768 | 1038 | * Called when we start the idle loop. |
280f0677 | 1039 | */ |
1268fbc7 | 1040 | void tick_nohz_idle_enter(void) |
280f0677 FW |
1041 | { |
1042 | struct tick_sched *ts; | |
1043 | ||
ebf3adba | 1044 | lockdep_assert_irqs_enabled(); |
0db49b72 | 1045 | |
1268fbc7 FW |
1046 | local_irq_disable(); |
1047 | ||
22127e93 | 1048 | ts = this_cpu_ptr(&tick_cpu_sched); |
23a8d888 RW |
1049 | |
1050 | WARN_ON_ONCE(ts->timer_expires_base); | |
1051 | ||
280f0677 | 1052 | ts->inidle = 1; |
0e776768 | 1053 | tick_nohz_start_idle(ts); |
1268fbc7 FW |
1054 | |
1055 | local_irq_enable(); | |
280f0677 FW |
1056 | } |
1057 | ||
1058 | /** | |
1059 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
1060 | * | |
1061 | * When an interrupt fires while we are idle and it doesn't cause | |
1062 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
1063 | * an RCU callback, etc... | |
1064 | * So we need to re-calculate and reprogram the next tick event. | |
1065 | */ | |
1066 | void tick_nohz_irq_exit(void) | |
1067 | { | |
22127e93 | 1068 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 1069 | |
14851912 | 1070 | if (ts->inidle) |
0e776768 | 1071 | tick_nohz_start_idle(ts); |
14851912 | 1072 | else |
73738a95 | 1073 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
1074 | } |
1075 | ||
4f86d3a8 | 1076 | /** |
45f1ff59 RW |
1077 | * tick_nohz_idle_got_tick - Check whether or not the tick handler has run |
1078 | */ | |
1079 | bool tick_nohz_idle_got_tick(void) | |
1080 | { | |
1081 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1082 | ||
2bc629a6 FW |
1083 | if (ts->got_idle_tick) { |
1084 | ts->got_idle_tick = 0; | |
45f1ff59 RW |
1085 | return true; |
1086 | } | |
1087 | return false; | |
1088 | } | |
1089 | ||
6f9b83ac UH |
1090 | /** |
1091 | * tick_nohz_get_next_hrtimer - return the next expiration time for the hrtimer | |
1092 | * or the tick, whatever that expires first. Note that, if the tick has been | |
1093 | * stopped, it returns the next hrtimer. | |
1094 | * | |
1095 | * Called from power state control code with interrupts disabled | |
1096 | */ | |
1097 | ktime_t tick_nohz_get_next_hrtimer(void) | |
1098 | { | |
1099 | return __this_cpu_read(tick_cpu_device.evtdev)->next_event; | |
1100 | } | |
1101 | ||
4f86d3a8 | 1102 | /** |
554c8aa8 | 1103 | * tick_nohz_get_sleep_length - return the expected length of the current sleep |
296bb1e5 | 1104 | * @delta_next: duration until the next event if the tick cannot be stopped |
4f86d3a8 LB |
1105 | * |
1106 | * Called from power state control code with interrupts disabled | |
1107 | */ | |
296bb1e5 | 1108 | ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next) |
4f86d3a8 | 1109 | { |
554c8aa8 | 1110 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
22127e93 | 1111 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
554c8aa8 RW |
1112 | int cpu = smp_processor_id(); |
1113 | /* | |
1114 | * The idle entry time is expected to be a sufficient approximation of | |
1115 | * the current time at this point. | |
1116 | */ | |
1117 | ktime_t now = ts->idle_entrytime; | |
1118 | ktime_t next_event; | |
1119 | ||
1120 | WARN_ON_ONCE(!ts->inidle); | |
1121 | ||
296bb1e5 RW |
1122 | *delta_next = ktime_sub(dev->next_event, now); |
1123 | ||
554c8aa8 | 1124 | if (!can_stop_idle_tick(cpu, ts)) |
296bb1e5 | 1125 | return *delta_next; |
554c8aa8 RW |
1126 | |
1127 | next_event = tick_nohz_next_event(ts, cpu); | |
1128 | if (!next_event) | |
296bb1e5 | 1129 | return *delta_next; |
554c8aa8 RW |
1130 | |
1131 | /* | |
1132 | * If the next highres timer to expire is earlier than next_event, the | |
1133 | * idle governor needs to know that. | |
1134 | */ | |
1135 | next_event = min_t(u64, next_event, | |
1136 | hrtimer_next_event_without(&ts->sched_timer)); | |
4f86d3a8 | 1137 | |
554c8aa8 | 1138 | return ktime_sub(next_event, now); |
4f86d3a8 LB |
1139 | } |
1140 | ||
466a2b42 JF |
1141 | /** |
1142 | * tick_nohz_get_idle_calls_cpu - return the current idle calls counter value | |
1143 | * for a particular CPU. | |
1144 | * | |
1145 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1146 | */ | |
1147 | unsigned long tick_nohz_get_idle_calls_cpu(int cpu) | |
1148 | { | |
1149 | struct tick_sched *ts = tick_get_tick_sched(cpu); | |
1150 | ||
1151 | return ts->idle_calls; | |
1152 | } | |
1153 | ||
b7eaf1aa RW |
1154 | /** |
1155 | * tick_nohz_get_idle_calls - return the current idle calls counter value | |
1156 | * | |
1157 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1158 | */ | |
1159 | unsigned long tick_nohz_get_idle_calls(void) | |
1160 | { | |
1161 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1162 | ||
1163 | return ts->idle_calls; | |
1164 | } | |
1165 | ||
2ac0d98f FW |
1166 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) |
1167 | { | |
3f4724ea | 1168 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 1169 | unsigned long ticks; |
3f4724ea | 1170 | |
e44fcb4b | 1171 | if (vtime_accounting_enabled_this_cpu()) |
3f4724ea | 1172 | return; |
79bf2bb3 TG |
1173 | /* |
1174 | * We stopped the tick in idle. Update process times would miss the | |
1175 | * time we slept as update_process_times does only a 1 tick | |
1176 | * accounting. Enforce that this is accounted to idle ! | |
1177 | */ | |
1178 | ticks = jiffies - ts->idle_jiffies; | |
1179 | /* | |
1180 | * We might be one off. Do not randomly account a huge number of ticks! | |
1181 | */ | |
79741dd3 MS |
1182 | if (ticks && ticks < LONG_MAX) |
1183 | account_idle_ticks(ticks); | |
1184 | #endif | |
19f5f736 FW |
1185 | } |
1186 | ||
2aaf709a RW |
1187 | static void __tick_nohz_idle_restart_tick(struct tick_sched *ts, ktime_t now) |
1188 | { | |
1189 | tick_nohz_restart_sched_tick(ts, now); | |
1190 | tick_nohz_account_idle_ticks(ts); | |
1191 | } | |
1192 | ||
1193 | void tick_nohz_idle_restart_tick(void) | |
1194 | { | |
1195 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1196 | ||
1197 | if (ts->tick_stopped) | |
1198 | __tick_nohz_idle_restart_tick(ts, ktime_get()); | |
1199 | } | |
1200 | ||
79bf2bb3 | 1201 | /** |
280f0677 | 1202 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1203 | * |
1204 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1205 | * This also exit the RCU extended quiescent state. The CPU |
1206 | * can use RCU again after this function is called. | |
79bf2bb3 | 1207 | */ |
280f0677 | 1208 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1209 | { |
4a32fea9 | 1210 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
bbe9a70a | 1211 | bool idle_active, tick_stopped; |
6378ddb5 | 1212 | ktime_t now; |
79bf2bb3 | 1213 | |
6378ddb5 | 1214 | local_irq_disable(); |
2bbb6817 | 1215 | |
15f827be | 1216 | WARN_ON_ONCE(!ts->inidle); |
23a8d888 | 1217 | WARN_ON_ONCE(ts->timer_expires_base); |
15f827be FW |
1218 | |
1219 | ts->inidle = 0; | |
bbe9a70a AB |
1220 | idle_active = ts->idle_active; |
1221 | tick_stopped = ts->tick_stopped; | |
15f827be | 1222 | |
bbe9a70a | 1223 | if (idle_active || tick_stopped) |
eed3b9cf MS |
1224 | now = ktime_get(); |
1225 | ||
bbe9a70a | 1226 | if (idle_active) |
e8fcaa5c | 1227 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1228 | |
bbe9a70a | 1229 | if (tick_stopped) |
2aaf709a | 1230 | __tick_nohz_idle_restart_tick(ts, now); |
79bf2bb3 | 1231 | |
79bf2bb3 TG |
1232 | local_irq_enable(); |
1233 | } | |
1234 | ||
79bf2bb3 TG |
1235 | /* |
1236 | * The nohz low res interrupt handler | |
1237 | */ | |
1238 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1239 | { | |
22127e93 | 1240 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1241 | struct pt_regs *regs = get_irq_regs(); |
1242 | ktime_t now = ktime_get(); | |
1243 | ||
2456e855 | 1244 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1245 | |
ff7de620 | 1246 | tick_sched_do_timer(ts, now); |
9e8f559b | 1247 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1248 | |
b5e995e6 VK |
1249 | /* No need to reprogram if we are running tickless */ |
1250 | if (unlikely(ts->tick_stopped)) | |
1251 | return; | |
1252 | ||
0ff53d09 TG |
1253 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1254 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
79bf2bb3 TG |
1255 | } |
1256 | ||
bc7a34b8 TG |
1257 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1258 | { | |
1259 | if (!tick_nohz_enabled) | |
1260 | return; | |
1261 | ts->nohz_mode = mode; | |
1262 | /* One update is enough */ | |
1263 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
ae67bada | 1264 | timers_update_nohz(); |
bc7a34b8 TG |
1265 | } |
1266 | ||
79bf2bb3 TG |
1267 | /** |
1268 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1269 | */ | |
1270 | static void tick_nohz_switch_to_nohz(void) | |
1271 | { | |
22127e93 | 1272 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1273 | ktime_t next; |
1274 | ||
27630532 | 1275 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1276 | return; |
1277 | ||
6b442bc8 | 1278 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1279 | return; |
6b442bc8 | 1280 | |
79bf2bb3 TG |
1281 | /* |
1282 | * Recycle the hrtimer in ts, so we can share the | |
1283 | * hrtimer_forward with the highres code. | |
1284 | */ | |
71fed982 | 1285 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 TG |
1286 | /* Get the next period */ |
1287 | next = tick_init_jiffy_update(); | |
1288 | ||
0ff53d09 | 1289 | hrtimer_set_expires(&ts->sched_timer, next); |
1ca8ec53 WL |
1290 | hrtimer_forward_now(&ts->sched_timer, tick_period); |
1291 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
bc7a34b8 | 1292 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1293 | } |
1294 | ||
5acac1be | 1295 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1296 | { |
4a32fea9 | 1297 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1298 | ktime_t now; |
1299 | ||
1300 | if (!ts->idle_active && !ts->tick_stopped) | |
1301 | return; | |
1302 | now = ktime_get(); | |
1303 | if (ts->idle_active) | |
e8fcaa5c | 1304 | tick_nohz_stop_idle(ts, now); |
ff006732 | 1305 | if (ts->tick_stopped) |
eed3b9cf | 1306 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1307 | } |
1308 | ||
79bf2bb3 TG |
1309 | #else |
1310 | ||
1311 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1312 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1313 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1314 | |
3451d024 | 1315 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1316 | |
719254fa TG |
1317 | /* |
1318 | * Called from irq_enter to notify about the possible interruption of idle() | |
1319 | */ | |
5acac1be | 1320 | void tick_irq_enter(void) |
719254fa | 1321 | { |
e8fcaa5c | 1322 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1323 | tick_nohz_irq_enter(); |
719254fa TG |
1324 | } |
1325 | ||
79bf2bb3 TG |
1326 | /* |
1327 | * High resolution timer specific code | |
1328 | */ | |
1329 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1330 | /* | |
4c9dc641 | 1331 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1332 | * Called with interrupts disabled. |
79bf2bb3 TG |
1333 | */ |
1334 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1335 | { | |
1336 | struct tick_sched *ts = | |
1337 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1338 | struct pt_regs *regs = get_irq_regs(); |
1339 | ktime_t now = ktime_get(); | |
d3ed7824 | 1340 | |
ff7de620 | 1341 | tick_sched_do_timer(ts, now); |
79bf2bb3 TG |
1342 | |
1343 | /* | |
1344 | * Do not call, when we are not in irq context and have | |
1345 | * no valid regs pointer | |
1346 | */ | |
9e8f559b FW |
1347 | if (regs) |
1348 | tick_sched_handle(ts, regs); | |
7c259045 FW |
1349 | else |
1350 | ts->next_tick = 0; | |
79bf2bb3 | 1351 | |
2a16fc93 VK |
1352 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1353 | if (unlikely(ts->tick_stopped)) | |
1354 | return HRTIMER_NORESTART; | |
1355 | ||
79bf2bb3 TG |
1356 | hrtimer_forward(timer, now, tick_period); |
1357 | ||
1358 | return HRTIMER_RESTART; | |
1359 | } | |
1360 | ||
5307c955 MG |
1361 | static int sched_skew_tick; |
1362 | ||
62cf20b3 TG |
1363 | static int __init skew_tick(char *str) |
1364 | { | |
1365 | get_option(&str, &sched_skew_tick); | |
1366 | ||
1367 | return 0; | |
1368 | } | |
1369 | early_param("skew_tick", skew_tick); | |
1370 | ||
79bf2bb3 TG |
1371 | /** |
1372 | * tick_setup_sched_timer - setup the tick emulation timer | |
1373 | */ | |
1374 | void tick_setup_sched_timer(void) | |
1375 | { | |
22127e93 | 1376 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1377 | ktime_t now = ktime_get(); |
1378 | ||
1379 | /* | |
1380 | * Emulate tick processing via per-CPU hrtimers: | |
1381 | */ | |
902a9f9c | 1382 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 | 1383 | ts->sched_timer.function = tick_sched_timer; |
79bf2bb3 | 1384 | |
0de7611a | 1385 | /* Get the next period (per-CPU) */ |
cc584b21 | 1386 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1387 | |
9c3f9e28 | 1388 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
1389 | if (sched_skew_tick) { |
1390 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
1391 | do_div(offset, num_possible_cpus()); | |
1392 | offset *= smp_processor_id(); | |
1393 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1394 | } | |
1395 | ||
afc08b15 | 1396 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
902a9f9c | 1397 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED_HARD); |
bc7a34b8 | 1398 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1399 | } |
3c4fbe5e | 1400 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1401 | |
3451d024 | 1402 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1403 | void tick_cancel_sched_timer(int cpu) |
1404 | { | |
1405 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1406 | ||
3c4fbe5e | 1407 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1408 | if (ts->sched_timer.base) |
1409 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1410 | # endif |
a7901766 | 1411 | |
4b0c0f29 | 1412 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1413 | } |
3c4fbe5e | 1414 | #endif |
79bf2bb3 TG |
1415 | |
1416 | /** | |
1417 | * Async notification about clocksource changes | |
1418 | */ | |
1419 | void tick_clock_notify(void) | |
1420 | { | |
1421 | int cpu; | |
1422 | ||
1423 | for_each_possible_cpu(cpu) | |
1424 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * Async notification about clock event changes | |
1429 | */ | |
1430 | void tick_oneshot_notify(void) | |
1431 | { | |
22127e93 | 1432 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1433 | |
1434 | set_bit(0, &ts->check_clocks); | |
1435 | } | |
1436 | ||
1437 | /** | |
1438 | * Check, if a change happened, which makes oneshot possible. | |
1439 | * | |
1440 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1441 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1442 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1443 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1444 | */ |
1445 | int tick_check_oneshot_change(int allow_nohz) | |
1446 | { | |
22127e93 | 1447 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1448 | |
1449 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1450 | return 0; | |
1451 | ||
1452 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1453 | return 0; | |
1454 | ||
cf4fc6cb | 1455 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1456 | return 0; |
1457 | ||
1458 | if (!allow_nohz) | |
1459 | return 1; | |
1460 | ||
1461 | tick_nohz_switch_to_nohz(); | |
1462 | return 0; | |
1463 | } |