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 | /* | |
47 | * The time, when the last jiffy update happened. Protected by jiffies_lock. | |
48 | */ | |
49 | static ktime_t last_jiffies_update; | |
50 | ||
79bf2bb3 TG |
51 | /* |
52 | * Must be called with interrupts disabled ! | |
53 | */ | |
54 | static void tick_do_update_jiffies64(ktime_t now) | |
55 | { | |
56 | unsigned long ticks = 0; | |
57 | ktime_t delta; | |
58 | ||
7a14ce1d | 59 | /* |
d6ad4187 | 60 | * Do a quick check without holding jiffies_lock: |
7a14ce1d IM |
61 | */ |
62 | delta = ktime_sub(now, last_jiffies_update); | |
2456e855 | 63 | if (delta < tick_period) |
7a14ce1d IM |
64 | return; |
65 | ||
6168f8ed | 66 | /* Reevaluate with jiffies_lock held */ |
d6ad4187 | 67 | write_seqlock(&jiffies_lock); |
79bf2bb3 TG |
68 | |
69 | delta = ktime_sub(now, last_jiffies_update); | |
2456e855 | 70 | if (delta >= tick_period) { |
79bf2bb3 TG |
71 | |
72 | delta = ktime_sub(delta, tick_period); | |
73 | last_jiffies_update = ktime_add(last_jiffies_update, | |
74 | tick_period); | |
75 | ||
76 | /* Slow path for long timeouts */ | |
2456e855 | 77 | if (unlikely(delta >= tick_period)) { |
79bf2bb3 TG |
78 | s64 incr = ktime_to_ns(tick_period); |
79 | ||
80 | ticks = ktime_divns(delta, incr); | |
81 | ||
82 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
83 | incr * ticks); | |
84 | } | |
85 | do_timer(++ticks); | |
49d670fb TG |
86 | |
87 | /* Keep the tick_next_period variable up to date */ | |
88 | tick_next_period = ktime_add(last_jiffies_update, tick_period); | |
03e6bdc5 VK |
89 | } else { |
90 | write_sequnlock(&jiffies_lock); | |
91 | return; | |
79bf2bb3 | 92 | } |
d6ad4187 | 93 | write_sequnlock(&jiffies_lock); |
47a1b796 | 94 | update_wall_time(); |
79bf2bb3 TG |
95 | } |
96 | ||
97 | /* | |
98 | * Initialize and return retrieve the jiffies update. | |
99 | */ | |
100 | static ktime_t tick_init_jiffy_update(void) | |
101 | { | |
102 | ktime_t period; | |
103 | ||
d6ad4187 | 104 | write_seqlock(&jiffies_lock); |
79bf2bb3 | 105 | /* Did we start the jiffies update yet ? */ |
2456e855 | 106 | if (last_jiffies_update == 0) |
79bf2bb3 TG |
107 | last_jiffies_update = tick_next_period; |
108 | period = last_jiffies_update; | |
d6ad4187 | 109 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
110 | return period; |
111 | } | |
112 | ||
ff7de620 | 113 | static void tick_sched_do_timer(struct tick_sched *ts, ktime_t now) |
5bb96226 FW |
114 | { |
115 | int cpu = smp_processor_id(); | |
116 | ||
3451d024 | 117 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
118 | /* |
119 | * Check if the do_timer duty was dropped. We don't care about | |
0de7611a IM |
120 | * concurrency: This happens only when the CPU in charge went |
121 | * into a long sleep. If two CPUs happen to assign themselves to | |
5bb96226 | 122 | * this duty, then the jiffies update is still serialized by |
9c3f9e28 | 123 | * jiffies_lock. |
5bb96226 | 124 | */ |
a382bf93 | 125 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE) |
c5bfece2 | 126 | && !tick_nohz_full_cpu(cpu)) |
5bb96226 FW |
127 | tick_do_timer_cpu = cpu; |
128 | #endif | |
129 | ||
130 | /* Check, if the jiffies need an update */ | |
131 | if (tick_do_timer_cpu == cpu) | |
132 | tick_do_update_jiffies64(now); | |
ff7de620 RW |
133 | |
134 | if (ts->inidle) | |
135 | ts->got_idle_tick = 1; | |
5bb96226 FW |
136 | } |
137 | ||
9e8f559b FW |
138 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
139 | { | |
3451d024 | 140 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
141 | /* |
142 | * When we are idle and the tick is stopped, we have to touch | |
143 | * the watchdog as we might not schedule for a really long | |
144 | * time. This happens on complete idle SMP systems while | |
145 | * waiting on the login prompt. We also increment the "start of | |
146 | * idle" jiffy stamp so the idle accounting adjustment we do | |
147 | * when we go busy again does not account too much ticks. | |
148 | */ | |
149 | if (ts->tick_stopped) { | |
03e0d461 | 150 | touch_softlockup_watchdog_sched(); |
9e8f559b FW |
151 | if (is_idle_task(current)) |
152 | ts->idle_jiffies++; | |
411fe24e FW |
153 | /* |
154 | * In case the current tick fired too early past its expected | |
155 | * expiration, make sure we don't bypass the next clock reprogramming | |
156 | * to the same deadline. | |
157 | */ | |
158 | ts->next_tick = 0; | |
9e8f559b | 159 | } |
94a57140 | 160 | #endif |
9e8f559b FW |
161 | update_process_times(user_mode(regs)); |
162 | profile_tick(CPU_PROFILING); | |
163 | } | |
7809998a | 164 | #endif |
9e8f559b | 165 | |
c5bfece2 | 166 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 167 | cpumask_var_t tick_nohz_full_mask; |
73867dcd | 168 | bool tick_nohz_full_running; |
f009a7a7 | 169 | static atomic_t tick_dep_mask; |
a831881b | 170 | |
f009a7a7 | 171 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 172 | { |
f009a7a7 FW |
173 | int val = atomic_read(dep); |
174 | ||
175 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 176 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 177 | return true; |
d027d45d FW |
178 | } |
179 | ||
f009a7a7 | 180 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 181 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 182 | return true; |
d027d45d FW |
183 | } |
184 | ||
f009a7a7 | 185 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 186 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 187 | return true; |
d027d45d FW |
188 | } |
189 | ||
f009a7a7 | 190 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 191 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
192 | return true; |
193 | } | |
194 | ||
195 | return false; | |
d027d45d FW |
196 | } |
197 | ||
57ccdf44 | 198 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d | 199 | { |
ebf3adba | 200 | lockdep_assert_irqs_disabled(); |
9014c45d | 201 | |
57ccdf44 WL |
202 | if (unlikely(!cpu_online(cpu))) |
203 | return false; | |
204 | ||
f009a7a7 | 205 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 206 | return false; |
d027d45d | 207 | |
f009a7a7 | 208 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 209 | return false; |
d027d45d | 210 | |
f009a7a7 | 211 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 212 | return false; |
d027d45d | 213 | |
f009a7a7 | 214 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 215 | return false; |
d027d45d | 216 | |
9014c45d FW |
217 | return true; |
218 | } | |
219 | ||
d027d45d | 220 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 221 | { |
73738a95 | 222 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
223 | } |
224 | ||
225 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = { | |
d027d45d | 226 | .func = nohz_full_kick_func, |
76c24fb0 FW |
227 | }; |
228 | ||
40bea039 FW |
229 | /* |
230 | * Kick this CPU if it's full dynticks in order to force it to | |
231 | * re-evaluate its dependency on the tick and restart it if necessary. | |
232 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
233 | * is NMI safe. | |
234 | */ | |
555e0c1e | 235 | static void tick_nohz_full_kick(void) |
40bea039 FW |
236 | { |
237 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
238 | return; | |
239 | ||
56e4dea8 | 240 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
241 | } |
242 | ||
76c24fb0 | 243 | /* |
3d36aebc | 244 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
245 | * re-evaluate its dependency on the tick and restart it if necessary. |
246 | */ | |
3d36aebc | 247 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 248 | { |
3d36aebc FW |
249 | if (!tick_nohz_full_cpu(cpu)) |
250 | return; | |
251 | ||
252 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
253 | } |
254 | ||
76c24fb0 FW |
255 | /* |
256 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
257 | * their dependency on the tick and restart it if necessary. | |
258 | */ | |
b7878300 | 259 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 260 | { |
8537bb95 FW |
261 | int cpu; |
262 | ||
73867dcd | 263 | if (!tick_nohz_full_running) |
76c24fb0 FW |
264 | return; |
265 | ||
266 | preempt_disable(); | |
8537bb95 FW |
267 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
268 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
269 | preempt_enable(); |
270 | } | |
271 | ||
f009a7a7 | 272 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
273 | enum tick_dep_bits bit) |
274 | { | |
f009a7a7 | 275 | int prev; |
d027d45d | 276 | |
a1cc5bcf | 277 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
278 | if (!prev) |
279 | tick_nohz_full_kick_all(); | |
280 | } | |
281 | ||
282 | /* | |
283 | * Set a global tick dependency. Used by perf events that rely on freq and | |
284 | * by unstable clock. | |
285 | */ | |
286 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
287 | { | |
288 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
289 | } | |
290 | ||
291 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
292 | { | |
f009a7a7 | 293 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
294 | } |
295 | ||
296 | /* | |
297 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
298 | * manage events throttling. | |
299 | */ | |
300 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
301 | { | |
f009a7a7 | 302 | int prev; |
d027d45d FW |
303 | struct tick_sched *ts; |
304 | ||
305 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
306 | ||
a1cc5bcf | 307 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
308 | if (!prev) { |
309 | preempt_disable(); | |
310 | /* Perf needs local kick that is NMI safe */ | |
311 | if (cpu == smp_processor_id()) { | |
312 | tick_nohz_full_kick(); | |
313 | } else { | |
314 | /* Remote irq work not NMI-safe */ | |
315 | if (!WARN_ON_ONCE(in_nmi())) | |
316 | tick_nohz_full_kick_cpu(cpu); | |
317 | } | |
318 | preempt_enable(); | |
319 | } | |
320 | } | |
321 | ||
322 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
323 | { | |
324 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
325 | ||
f009a7a7 | 326 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
327 | } |
328 | ||
329 | /* | |
330 | * Set a per-task tick dependency. Posix CPU timers need this in order to elapse | |
331 | * per task timers. | |
332 | */ | |
333 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
334 | { | |
335 | /* | |
336 | * We could optimize this with just kicking the target running the task | |
337 | * if that noise matters for nohz full users. | |
338 | */ | |
339 | tick_nohz_dep_set_all(&tsk->tick_dep_mask, bit); | |
340 | } | |
341 | ||
342 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
343 | { | |
f009a7a7 | 344 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d FW |
345 | } |
346 | ||
347 | /* | |
348 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
349 | * per process timers. | |
350 | */ | |
351 | void tick_nohz_dep_set_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
352 | { | |
353 | tick_nohz_dep_set_all(&sig->tick_dep_mask, bit); | |
354 | } | |
355 | ||
356 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
357 | { | |
f009a7a7 | 358 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
359 | } |
360 | ||
99e5ada9 FW |
361 | /* |
362 | * Re-evaluate the need for the tick as we switch the current task. | |
363 | * It might need the tick due to per task/process properties: | |
0de7611a | 364 | * perf events, posix CPU timers, ... |
99e5ada9 | 365 | */ |
de734f89 | 366 | void __tick_nohz_task_switch(void) |
99e5ada9 FW |
367 | { |
368 | unsigned long flags; | |
d027d45d | 369 | struct tick_sched *ts; |
99e5ada9 | 370 | |
99e5ada9 FW |
371 | local_irq_save(flags); |
372 | ||
6296ace4 LZ |
373 | if (!tick_nohz_full_cpu(smp_processor_id())) |
374 | goto out; | |
375 | ||
d027d45d | 376 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 377 | |
d027d45d | 378 | if (ts->tick_stopped) { |
f009a7a7 FW |
379 | if (atomic_read(¤t->tick_dep_mask) || |
380 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
381 | tick_nohz_full_kick(); |
382 | } | |
6296ace4 | 383 | out: |
99e5ada9 FW |
384 | local_irq_restore(flags); |
385 | } | |
386 | ||
6f1982fe FW |
387 | /* Get the boot-time nohz CPU list from the kernel parameters. */ |
388 | void __init tick_nohz_full_setup(cpumask_var_t cpumask) | |
a831881b | 389 | { |
73867dcd | 390 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
6f1982fe | 391 | cpumask_copy(tick_nohz_full_mask, cpumask); |
73867dcd | 392 | tick_nohz_full_running = true; |
a831881b | 393 | } |
a831881b | 394 | |
31eff243 | 395 | static int tick_nohz_cpu_down(unsigned int cpu) |
a382bf93 | 396 | { |
31eff243 SAS |
397 | /* |
398 | * The boot CPU handles housekeeping duty (unbound timers, | |
399 | * workqueues, timekeeping, ...) on behalf of full dynticks | |
400 | * CPUs. It must remain online when nohz full is enabled. | |
401 | */ | |
402 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) | |
403 | return -EBUSY; | |
404 | return 0; | |
a382bf93 FW |
405 | } |
406 | ||
d1e43fa5 | 407 | void __init tick_nohz_init(void) |
a831881b | 408 | { |
31eff243 | 409 | int cpu, ret; |
d1e43fa5 | 410 | |
a7c8655b PM |
411 | if (!tick_nohz_full_running) |
412 | return; | |
d1e43fa5 | 413 | |
9b01f5bf FW |
414 | /* |
415 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
416 | * locking contexts. But then we need irq work to raise its own | |
417 | * interrupts to avoid circular dependency on the tick | |
418 | */ | |
419 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 420 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf | 421 | cpumask_clear(tick_nohz_full_mask); |
9b01f5bf FW |
422 | tick_nohz_full_running = false; |
423 | return; | |
424 | } | |
425 | ||
4327b15f FW |
426 | cpu = smp_processor_id(); |
427 | ||
428 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { | |
a395d6a7 JP |
429 | pr_warn("NO_HZ: Clearing %d from nohz_full range for timekeeping\n", |
430 | cpu); | |
4327b15f FW |
431 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); |
432 | } | |
433 | ||
73867dcd | 434 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
435 | context_tracking_cpu_set(cpu); |
436 | ||
31eff243 SAS |
437 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
438 | "kernel/nohz:predown", NULL, | |
439 | tick_nohz_cpu_down); | |
440 | WARN_ON(ret < 0); | |
ffda22c1 TH |
441 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
442 | cpumask_pr_args(tick_nohz_full_mask)); | |
a831881b | 443 | } |
a831881b FW |
444 | #endif |
445 | ||
79bf2bb3 TG |
446 | /* |
447 | * NOHZ - aka dynamic tick functionality | |
448 | */ | |
3451d024 | 449 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
450 | /* |
451 | * NO HZ enabled ? | |
452 | */ | |
4cc7ecb7 | 453 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 454 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
455 | /* |
456 | * Enable / Disable tickless mode | |
457 | */ | |
458 | static int __init setup_tick_nohz(char *str) | |
459 | { | |
4cc7ecb7 | 460 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
461 | } |
462 | ||
463 | __setup("nohz=", setup_tick_nohz); | |
464 | ||
a3642983 | 465 | bool tick_nohz_tick_stopped(void) |
c1797baf | 466 | { |
2bc629a6 FW |
467 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
468 | ||
469 | return ts->tick_stopped; | |
c1797baf TG |
470 | } |
471 | ||
22ab8bc0 FW |
472 | bool tick_nohz_tick_stopped_cpu(int cpu) |
473 | { | |
474 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
475 | ||
476 | return ts->tick_stopped; | |
477 | } | |
478 | ||
79bf2bb3 TG |
479 | /** |
480 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
481 | * | |
482 | * Called from interrupt entry when the CPU was idle | |
483 | * | |
484 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
485 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
486 | * value. We do this unconditionally on any CPU, as we don't know whether the |
487 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 488 | */ |
eed3b9cf | 489 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 490 | { |
79bf2bb3 | 491 | unsigned long flags; |
79bf2bb3 | 492 | |
e8fcaa5c | 493 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
494 | |
495 | local_irq_save(flags); | |
496 | tick_do_update_jiffies64(now); | |
497 | local_irq_restore(flags); | |
02ff3755 | 498 | |
03e0d461 | 499 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
500 | } |
501 | ||
595aac48 | 502 | /* |
0de7611a | 503 | * Updates the per-CPU time idle statistics counters |
595aac48 | 504 | */ |
8d63bf94 | 505 | static void |
8c215bd3 | 506 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 507 | { |
eed3b9cf | 508 | ktime_t delta; |
6378ddb5 | 509 | |
595aac48 AV |
510 | if (ts->idle_active) { |
511 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 512 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 513 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
514 | else |
515 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 516 | ts->idle_entrytime = now; |
595aac48 | 517 | } |
8d63bf94 | 518 | |
e0e37c20 | 519 | if (last_update_time) |
8d63bf94 AV |
520 | *last_update_time = ktime_to_us(now); |
521 | ||
595aac48 AV |
522 | } |
523 | ||
e8fcaa5c | 524 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 525 | { |
e8fcaa5c | 526 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 527 | ts->idle_active = 0; |
56c7426b | 528 | |
ac1e843f | 529 | sched_clock_idle_wakeup_event(); |
6378ddb5 VP |
530 | } |
531 | ||
0e776768 | 532 | static void tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 533 | { |
0e776768 | 534 | ts->idle_entrytime = ktime_get(); |
6378ddb5 | 535 | ts->idle_active = 1; |
56c7426b | 536 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
537 | } |
538 | ||
b1f724c3 | 539 | /** |
0de7611a | 540 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 541 | * @cpu: CPU number to query |
09a1d34f MH |
542 | * @last_update_time: variable to store update time in. Do not update |
543 | * counters if NULL. | |
b1f724c3 | 544 | * |
6168f8ed | 545 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 546 | * CPU, in microseconds. |
b1f724c3 AV |
547 | * |
548 | * This time is measured via accounting rather than sampling, | |
549 | * and is as accurate as ktime_get() is. | |
550 | * | |
551 | * This function returns -1 if NOHZ is not enabled. | |
552 | */ | |
6378ddb5 VP |
553 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
554 | { | |
555 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 556 | ktime_t now, idle; |
6378ddb5 | 557 | |
d689fe22 | 558 | if (!tick_nohz_active) |
8083e4ad | 559 | return -1; |
560 | ||
09a1d34f MH |
561 | now = ktime_get(); |
562 | if (last_update_time) { | |
563 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
564 | idle = ts->idle_sleeptime; | |
565 | } else { | |
566 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
567 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
568 | ||
569 | idle = ktime_add(ts->idle_sleeptime, delta); | |
570 | } else { | |
571 | idle = ts->idle_sleeptime; | |
572 | } | |
573 | } | |
574 | ||
575 | return ktime_to_us(idle); | |
8083e4ad | 576 | |
6378ddb5 | 577 | } |
8083e4ad | 578 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 579 | |
6beea0cd | 580 | /** |
0de7611a | 581 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 582 | * @cpu: CPU number to query |
09a1d34f MH |
583 | * @last_update_time: variable to store update time in. Do not update |
584 | * counters if NULL. | |
0224cf4c | 585 | * |
6168f8ed | 586 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
587 | * CPU, in microseconds. |
588 | * | |
589 | * This time is measured via accounting rather than sampling, | |
590 | * and is as accurate as ktime_get() is. | |
591 | * | |
592 | * This function returns -1 if NOHZ is not enabled. | |
593 | */ | |
594 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
595 | { | |
596 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 597 | ktime_t now, iowait; |
0224cf4c | 598 | |
d689fe22 | 599 | if (!tick_nohz_active) |
0224cf4c AV |
600 | return -1; |
601 | ||
09a1d34f MH |
602 | now = ktime_get(); |
603 | if (last_update_time) { | |
604 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
605 | iowait = ts->iowait_sleeptime; | |
606 | } else { | |
607 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
608 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 609 | |
09a1d34f MH |
610 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
611 | } else { | |
612 | iowait = ts->iowait_sleeptime; | |
613 | } | |
614 | } | |
0224cf4c | 615 | |
09a1d34f | 616 | return ktime_to_us(iowait); |
0224cf4c AV |
617 | } |
618 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
619 | ||
0ff53d09 TG |
620 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
621 | { | |
622 | hrtimer_cancel(&ts->sched_timer); | |
623 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
624 | ||
625 | /* Forward the time to expire in the future */ | |
626 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
627 | ||
628 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) | |
629 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); | |
630 | else | |
631 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
411fe24e FW |
632 | |
633 | /* | |
634 | * Reset to make sure next tick stop doesn't get fooled by past | |
635 | * cached clock deadline. | |
636 | */ | |
637 | ts->next_tick = 0; | |
0ff53d09 TG |
638 | } |
639 | ||
5d62c183 TG |
640 | static inline bool local_timer_softirq_pending(void) |
641 | { | |
80d20d35 | 642 | return local_softirq_pending() & BIT(TIMER_SOFTIRQ); |
5d62c183 TG |
643 | } |
644 | ||
23a8d888 | 645 | static ktime_t tick_nohz_next_event(struct tick_sched *ts, int cpu) |
79bf2bb3 | 646 | { |
c1ad348b TG |
647 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
648 | unsigned long seq, basejiff; | |
855a0fc3 | 649 | |
79bf2bb3 TG |
650 | /* Read jiffies and the time when jiffies were updated last */ |
651 | do { | |
d6ad4187 | 652 | seq = read_seqbegin(&jiffies_lock); |
2456e855 | 653 | basemono = last_jiffies_update; |
c1ad348b | 654 | basejiff = jiffies; |
d6ad4187 | 655 | } while (read_seqretry(&jiffies_lock, seq)); |
c1ad348b | 656 | ts->last_jiffies = basejiff; |
23a8d888 | 657 | ts->timer_expires_base = basemono; |
79bf2bb3 | 658 | |
5d62c183 TG |
659 | /* |
660 | * Keep the periodic tick, when RCU, architecture or irq_work | |
661 | * requests it. | |
662 | * Aside of that check whether the local timer softirq is | |
663 | * pending. If so its a bad idea to call get_next_timer_interrupt() | |
664 | * because there is an already expired timer, so it will request | |
665 | * immeditate expiry, which rearms the hardware timer with a | |
666 | * minimal delta which brings us back to this place | |
667 | * immediately. Lather, rinse and repeat... | |
668 | */ | |
669 | if (rcu_needs_cpu(basemono, &next_rcu) || arch_needs_cpu() || | |
670 | irq_work_needs_cpu() || local_timer_softirq_pending()) { | |
c1ad348b | 671 | next_tick = basemono + TICK_NSEC; |
3c5d92a0 | 672 | } else { |
c1ad348b TG |
673 | /* |
674 | * Get the next pending timer. If high resolution | |
675 | * timers are enabled this only takes the timer wheel | |
676 | * timers into account. If high resolution timers are | |
677 | * disabled this also looks at the next expiring | |
678 | * hrtimer. | |
679 | */ | |
680 | next_tmr = get_next_timer_interrupt(basejiff, basemono); | |
681 | ts->next_timer = next_tmr; | |
682 | /* Take the next rcu event into account */ | |
683 | next_tick = next_rcu < next_tmr ? next_rcu : next_tmr; | |
3c5d92a0 | 684 | } |
47aa8b6c | 685 | |
c1ad348b TG |
686 | /* |
687 | * If the tick is due in the next period, keep it ticking or | |
82bbe34b | 688 | * force prod the timer. |
c1ad348b TG |
689 | */ |
690 | delta = next_tick - basemono; | |
691 | if (delta <= (u64)TICK_NSEC) { | |
a683f390 TG |
692 | /* |
693 | * Tell the timer code that the base is not idle, i.e. undo | |
694 | * the effect of get_next_timer_interrupt(): | |
695 | */ | |
696 | timer_clear_idle(); | |
82bbe34b PZ |
697 | /* |
698 | * We've not stopped the tick yet, and there's a timer in the | |
699 | * next period, so no point in stopping it either, bail. | |
700 | */ | |
f99973e1 | 701 | if (!ts->tick_stopped) { |
23a8d888 | 702 | ts->timer_expires = 0; |
157d29e1 TG |
703 | goto out; |
704 | } | |
705 | } | |
706 | ||
23a8d888 RW |
707 | /* |
708 | * If this CPU is the one which had the do_timer() duty last, we limit | |
709 | * the sleep time to the timekeeping max_deferment value. | |
710 | * Otherwise we can sleep as long as we want. | |
711 | */ | |
712 | delta = timekeeping_max_deferment(); | |
713 | if (cpu != tick_do_timer_cpu && | |
714 | (tick_do_timer_cpu != TICK_DO_TIMER_NONE || !ts->do_timer_last)) | |
715 | delta = KTIME_MAX; | |
716 | ||
717 | /* Calculate the next expiry time */ | |
718 | if (delta < (KTIME_MAX - basemono)) | |
719 | expires = basemono + delta; | |
720 | else | |
721 | expires = KTIME_MAX; | |
722 | ||
723 | ts->timer_expires = min_t(u64, expires, next_tick); | |
724 | ||
725 | out: | |
726 | return ts->timer_expires; | |
727 | } | |
728 | ||
729 | static void tick_nohz_stop_tick(struct tick_sched *ts, int cpu) | |
730 | { | |
731 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); | |
732 | u64 basemono = ts->timer_expires_base; | |
733 | u64 expires = ts->timer_expires; | |
734 | ktime_t tick = expires; | |
735 | ||
736 | /* Make sure we won't be trying to stop it twice in a row. */ | |
737 | ts->timer_expires_base = 0; | |
738 | ||
79bf2bb3 | 739 | /* |
0de7611a IM |
740 | * If this CPU is the one which updates jiffies, then give up |
741 | * the assignment and let it be taken by the CPU which runs | |
742 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
743 | * don't drop this here the jiffies might be stale and |
744 | * do_timer() never invoked. Keep track of the fact that it | |
23a8d888 | 745 | * was the one which had the do_timer() duty last. |
79bf2bb3 | 746 | */ |
157d29e1 TG |
747 | if (cpu == tick_do_timer_cpu) { |
748 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
749 | ts->do_timer_last = 1; | |
750 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
157d29e1 | 751 | ts->do_timer_last = 0; |
157d29e1 | 752 | } |
27185016 | 753 | |
157d29e1 | 754 | /* Skip reprogram of event if its not changed */ |
411fe24e FW |
755 | if (ts->tick_stopped && (expires == ts->next_tick)) { |
756 | /* Sanity check: make sure clockevent is actually programmed */ | |
d4af6d93 | 757 | if (tick == KTIME_MAX || ts->next_tick == hrtimer_get_expires(&ts->sched_timer)) |
23a8d888 | 758 | return; |
411fe24e FW |
759 | |
760 | WARN_ON_ONCE(1); | |
761 | printk_once("basemono: %llu ts->next_tick: %llu dev->next_event: %llu timer->active: %d timer->expires: %llu\n", | |
762 | basemono, ts->next_tick, dev->next_event, | |
763 | hrtimer_active(&ts->sched_timer), hrtimer_get_expires(&ts->sched_timer)); | |
ce6cf9a1 | 764 | } |
84bf1bcc | 765 | |
157d29e1 TG |
766 | /* |
767 | * nohz_stop_sched_tick can be called several times before | |
768 | * the nohz_restart_sched_tick is called. This happens when | |
769 | * interrupts arrive which do not cause a reschedule. In the | |
770 | * first call we save the current tick time, so we can restart | |
771 | * the scheduler tick in nohz_restart_sched_tick. | |
772 | */ | |
773 | if (!ts->tick_stopped) { | |
3c85d6db | 774 | calc_load_nohz_start(); |
1f41906a | 775 | cpu_load_update_nohz_start(); |
62cb1188 | 776 | quiet_vmstat(); |
d3ed7824 | 777 | |
157d29e1 TG |
778 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
779 | ts->tick_stopped = 1; | |
e6e6cc22 | 780 | trace_tick_stop(1, TICK_DEP_MASK_NONE); |
157d29e1 | 781 | } |
eaad084b | 782 | |
411fe24e FW |
783 | ts->next_tick = tick; |
784 | ||
157d29e1 | 785 | /* |
c1ad348b TG |
786 | * If the expiration time == KTIME_MAX, then we simply stop |
787 | * the tick timer. | |
157d29e1 | 788 | */ |
c1ad348b | 789 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
790 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
791 | hrtimer_cancel(&ts->sched_timer); | |
23a8d888 | 792 | return; |
79bf2bb3 | 793 | } |
0ff53d09 | 794 | |
1f71addd TG |
795 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
796 | hrtimer_start(&ts->sched_timer, tick, HRTIMER_MODE_ABS_PINNED); | |
797 | } else { | |
798 | hrtimer_set_expires(&ts->sched_timer, tick); | |
c1ad348b | 799 | tick_program_event(tick, 1); |
1f71addd | 800 | } |
280f0677 FW |
801 | } |
802 | ||
23a8d888 RW |
803 | static void tick_nohz_retain_tick(struct tick_sched *ts) |
804 | { | |
805 | ts->timer_expires_base = 0; | |
806 | } | |
807 | ||
808 | #ifdef CONFIG_NO_HZ_FULL | |
809 | static void tick_nohz_stop_sched_tick(struct tick_sched *ts, int cpu) | |
810 | { | |
811 | if (tick_nohz_next_event(ts, cpu)) | |
812 | tick_nohz_stop_tick(ts, cpu); | |
813 | else | |
814 | tick_nohz_retain_tick(ts); | |
815 | } | |
816 | #endif /* CONFIG_NO_HZ_FULL */ | |
817 | ||
1f41906a | 818 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
819 | { |
820 | /* Update jiffies first */ | |
821 | tick_do_update_jiffies64(now); | |
1f41906a | 822 | cpu_load_update_nohz_stop(); |
59d2c7ca | 823 | |
a683f390 TG |
824 | /* |
825 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
826 | * the clock forward checks in the enqueue path: | |
827 | */ | |
828 | timer_clear_idle(); | |
829 | ||
3c85d6db | 830 | calc_load_nohz_stop(); |
03e0d461 | 831 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
832 | /* |
833 | * Cancel the scheduled timer and restore the tick | |
834 | */ | |
835 | ts->tick_stopped = 0; | |
836 | ts->idle_exittime = now; | |
837 | ||
838 | tick_nohz_restart(ts, now); | |
839 | } | |
73738a95 FW |
840 | |
841 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
5811d996 FW |
842 | { |
843 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 844 | int cpu = smp_processor_id(); |
5811d996 | 845 | |
59449359 | 846 | if (!tick_nohz_full_cpu(cpu)) |
e9a2eb40 | 847 | return; |
5811d996 | 848 | |
e9a2eb40 AS |
849 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
850 | return; | |
5811d996 | 851 | |
57ccdf44 | 852 | if (can_stop_full_tick(cpu, ts)) |
23a8d888 | 853 | tick_nohz_stop_sched_tick(ts, cpu); |
73738a95 | 854 | else if (ts->tick_stopped) |
1f41906a | 855 | tick_nohz_restart_sched_tick(ts, ktime_get()); |
5811d996 FW |
856 | #endif |
857 | } | |
858 | ||
5b39939a FW |
859 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
860 | { | |
861 | /* | |
0de7611a | 862 | * If this CPU is offline and it is the one which updates |
5b39939a | 863 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 864 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
865 | * this here the jiffies might be stale and do_timer() never |
866 | * invoked. | |
867 | */ | |
868 | if (unlikely(!cpu_online(cpu))) { | |
869 | if (cpu == tick_do_timer_cpu) | |
870 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
411fe24e FW |
871 | /* |
872 | * Make sure the CPU doesn't get fooled by obsolete tick | |
873 | * deadline if it comes back online later. | |
874 | */ | |
875 | ts->next_tick = 0; | |
f7ea0fd6 | 876 | return false; |
5b39939a FW |
877 | } |
878 | ||
23a8d888 | 879 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) |
5b39939a FW |
880 | return false; |
881 | ||
882 | if (need_resched()) | |
883 | return false; | |
884 | ||
d59e0ba1 | 885 | if (unlikely(local_softirq_pending())) { |
5b39939a FW |
886 | static int ratelimit; |
887 | ||
803b0eba PM |
888 | if (ratelimit < 10 && |
889 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
cfea7d7e RV |
890 | pr_warn("NOHZ: local_softirq_pending %02x\n", |
891 | (unsigned int) local_softirq_pending()); | |
5b39939a FW |
892 | ratelimit++; |
893 | } | |
894 | return false; | |
895 | } | |
896 | ||
460775df | 897 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
898 | /* |
899 | * Keep the tick alive to guarantee timekeeping progression | |
900 | * if there are full dynticks CPUs around | |
901 | */ | |
902 | if (tick_do_timer_cpu == cpu) | |
903 | return false; | |
904 | /* | |
905 | * Boot safety: make sure the timekeeping duty has been | |
906 | * assigned before entering dyntick-idle mode, | |
907 | */ | |
908 | if (tick_do_timer_cpu == TICK_DO_TIMER_NONE) | |
909 | return false; | |
910 | } | |
911 | ||
5b39939a FW |
912 | return true; |
913 | } | |
914 | ||
0e776768 | 915 | static void __tick_nohz_idle_stop_tick(struct tick_sched *ts) |
19f5f736 | 916 | { |
0e776768 | 917 | ktime_t expires; |
5b39939a | 918 | int cpu = smp_processor_id(); |
19f5f736 | 919 | |
554c8aa8 RW |
920 | /* |
921 | * If tick_nohz_get_sleep_length() ran tick_nohz_next_event(), the | |
922 | * tick timer expiration time is known already. | |
923 | */ | |
924 | if (ts->timer_expires_base) | |
925 | expires = ts->timer_expires; | |
926 | else if (can_stop_idle_tick(cpu, ts)) | |
927 | expires = tick_nohz_next_event(ts, cpu); | |
928 | else | |
929 | return; | |
23a8d888 RW |
930 | |
931 | ts->idle_calls++; | |
08d07259 | 932 | |
23a8d888 | 933 | if (expires > 0LL) { |
5b39939a FW |
934 | int was_stopped = ts->tick_stopped; |
935 | ||
23a8d888 | 936 | tick_nohz_stop_tick(ts, cpu); |
84bf1bcc | 937 | |
23a8d888 RW |
938 | ts->idle_sleeps++; |
939 | ts->idle_expires = expires; | |
5b39939a | 940 | |
a0db971e | 941 | if (!was_stopped && ts->tick_stopped) { |
5b39939a | 942 | ts->idle_jiffies = ts->last_jiffies; |
a0db971e FW |
943 | nohz_balance_enter_idle(cpu); |
944 | } | |
23a8d888 RW |
945 | } else { |
946 | tick_nohz_retain_tick(ts); | |
5b39939a | 947 | } |
280f0677 FW |
948 | } |
949 | ||
950 | /** | |
0e776768 | 951 | * tick_nohz_idle_stop_tick - stop the idle tick from the idle task |
280f0677 FW |
952 | * |
953 | * When the next event is more than a tick into the future, stop the idle tick | |
0e776768 RW |
954 | */ |
955 | void tick_nohz_idle_stop_tick(void) | |
956 | { | |
957 | __tick_nohz_idle_stop_tick(this_cpu_ptr(&tick_cpu_sched)); | |
958 | } | |
959 | ||
554c8aa8 RW |
960 | void tick_nohz_idle_retain_tick(void) |
961 | { | |
962 | tick_nohz_retain_tick(this_cpu_ptr(&tick_cpu_sched)); | |
963 | /* | |
964 | * Undo the effect of get_next_timer_interrupt() called from | |
965 | * tick_nohz_next_event(). | |
966 | */ | |
967 | timer_clear_idle(); | |
968 | } | |
969 | ||
0e776768 RW |
970 | /** |
971 | * tick_nohz_idle_enter - prepare for entering idle on the current CPU | |
2bbb6817 | 972 | * |
0e776768 | 973 | * Called when we start the idle loop. |
280f0677 | 974 | */ |
1268fbc7 | 975 | void tick_nohz_idle_enter(void) |
280f0677 FW |
976 | { |
977 | struct tick_sched *ts; | |
978 | ||
ebf3adba | 979 | lockdep_assert_irqs_enabled(); |
0db49b72 | 980 | |
1268fbc7 FW |
981 | local_irq_disable(); |
982 | ||
22127e93 | 983 | ts = this_cpu_ptr(&tick_cpu_sched); |
23a8d888 RW |
984 | |
985 | WARN_ON_ONCE(ts->timer_expires_base); | |
986 | ||
280f0677 | 987 | ts->inidle = 1; |
0e776768 | 988 | tick_nohz_start_idle(ts); |
1268fbc7 FW |
989 | |
990 | local_irq_enable(); | |
280f0677 FW |
991 | } |
992 | ||
993 | /** | |
994 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
995 | * | |
996 | * When an interrupt fires while we are idle and it doesn't cause | |
997 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
998 | * an RCU callback, etc... | |
999 | * So we need to re-calculate and reprogram the next tick event. | |
1000 | */ | |
1001 | void tick_nohz_irq_exit(void) | |
1002 | { | |
22127e93 | 1003 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 1004 | |
14851912 | 1005 | if (ts->inidle) |
0e776768 | 1006 | tick_nohz_start_idle(ts); |
14851912 | 1007 | else |
73738a95 | 1008 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
1009 | } |
1010 | ||
4f86d3a8 | 1011 | /** |
45f1ff59 RW |
1012 | * tick_nohz_idle_got_tick - Check whether or not the tick handler has run |
1013 | */ | |
1014 | bool tick_nohz_idle_got_tick(void) | |
1015 | { | |
1016 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1017 | ||
2bc629a6 FW |
1018 | if (ts->got_idle_tick) { |
1019 | ts->got_idle_tick = 0; | |
45f1ff59 RW |
1020 | return true; |
1021 | } | |
1022 | return false; | |
1023 | } | |
1024 | ||
4f86d3a8 | 1025 | /** |
554c8aa8 | 1026 | * tick_nohz_get_sleep_length - return the expected length of the current sleep |
296bb1e5 | 1027 | * @delta_next: duration until the next event if the tick cannot be stopped |
4f86d3a8 LB |
1028 | * |
1029 | * Called from power state control code with interrupts disabled | |
1030 | */ | |
296bb1e5 | 1031 | ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next) |
4f86d3a8 | 1032 | { |
554c8aa8 | 1033 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
22127e93 | 1034 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
554c8aa8 RW |
1035 | int cpu = smp_processor_id(); |
1036 | /* | |
1037 | * The idle entry time is expected to be a sufficient approximation of | |
1038 | * the current time at this point. | |
1039 | */ | |
1040 | ktime_t now = ts->idle_entrytime; | |
1041 | ktime_t next_event; | |
1042 | ||
1043 | WARN_ON_ONCE(!ts->inidle); | |
1044 | ||
296bb1e5 RW |
1045 | *delta_next = ktime_sub(dev->next_event, now); |
1046 | ||
554c8aa8 | 1047 | if (!can_stop_idle_tick(cpu, ts)) |
296bb1e5 | 1048 | return *delta_next; |
554c8aa8 RW |
1049 | |
1050 | next_event = tick_nohz_next_event(ts, cpu); | |
1051 | if (!next_event) | |
296bb1e5 | 1052 | return *delta_next; |
554c8aa8 RW |
1053 | |
1054 | /* | |
1055 | * If the next highres timer to expire is earlier than next_event, the | |
1056 | * idle governor needs to know that. | |
1057 | */ | |
1058 | next_event = min_t(u64, next_event, | |
1059 | hrtimer_next_event_without(&ts->sched_timer)); | |
4f86d3a8 | 1060 | |
554c8aa8 | 1061 | return ktime_sub(next_event, now); |
4f86d3a8 LB |
1062 | } |
1063 | ||
466a2b42 JF |
1064 | /** |
1065 | * tick_nohz_get_idle_calls_cpu - return the current idle calls counter value | |
1066 | * for a particular CPU. | |
1067 | * | |
1068 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1069 | */ | |
1070 | unsigned long tick_nohz_get_idle_calls_cpu(int cpu) | |
1071 | { | |
1072 | struct tick_sched *ts = tick_get_tick_sched(cpu); | |
1073 | ||
1074 | return ts->idle_calls; | |
1075 | } | |
1076 | ||
b7eaf1aa RW |
1077 | /** |
1078 | * tick_nohz_get_idle_calls - return the current idle calls counter value | |
1079 | * | |
1080 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1081 | */ | |
1082 | unsigned long tick_nohz_get_idle_calls(void) | |
1083 | { | |
1084 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1085 | ||
1086 | return ts->idle_calls; | |
1087 | } | |
1088 | ||
2ac0d98f FW |
1089 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) |
1090 | { | |
3f4724ea | 1091 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 1092 | unsigned long ticks; |
3f4724ea | 1093 | |
55dbdcfa | 1094 | if (vtime_accounting_cpu_enabled()) |
3f4724ea | 1095 | return; |
79bf2bb3 TG |
1096 | /* |
1097 | * We stopped the tick in idle. Update process times would miss the | |
1098 | * time we slept as update_process_times does only a 1 tick | |
1099 | * accounting. Enforce that this is accounted to idle ! | |
1100 | */ | |
1101 | ticks = jiffies - ts->idle_jiffies; | |
1102 | /* | |
1103 | * We might be one off. Do not randomly account a huge number of ticks! | |
1104 | */ | |
79741dd3 MS |
1105 | if (ticks && ticks < LONG_MAX) |
1106 | account_idle_ticks(ticks); | |
1107 | #endif | |
19f5f736 FW |
1108 | } |
1109 | ||
2aaf709a RW |
1110 | static void __tick_nohz_idle_restart_tick(struct tick_sched *ts, ktime_t now) |
1111 | { | |
1112 | tick_nohz_restart_sched_tick(ts, now); | |
1113 | tick_nohz_account_idle_ticks(ts); | |
1114 | } | |
1115 | ||
1116 | void tick_nohz_idle_restart_tick(void) | |
1117 | { | |
1118 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1119 | ||
1120 | if (ts->tick_stopped) | |
1121 | __tick_nohz_idle_restart_tick(ts, ktime_get()); | |
1122 | } | |
1123 | ||
79bf2bb3 | 1124 | /** |
280f0677 | 1125 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1126 | * |
1127 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1128 | * This also exit the RCU extended quiescent state. The CPU |
1129 | * can use RCU again after this function is called. | |
79bf2bb3 | 1130 | */ |
280f0677 | 1131 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1132 | { |
4a32fea9 | 1133 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
bbe9a70a | 1134 | bool idle_active, tick_stopped; |
6378ddb5 | 1135 | ktime_t now; |
79bf2bb3 | 1136 | |
6378ddb5 | 1137 | local_irq_disable(); |
2bbb6817 | 1138 | |
15f827be | 1139 | WARN_ON_ONCE(!ts->inidle); |
23a8d888 | 1140 | WARN_ON_ONCE(ts->timer_expires_base); |
15f827be FW |
1141 | |
1142 | ts->inidle = 0; | |
bbe9a70a AB |
1143 | idle_active = ts->idle_active; |
1144 | tick_stopped = ts->tick_stopped; | |
15f827be | 1145 | |
bbe9a70a | 1146 | if (idle_active || tick_stopped) |
eed3b9cf MS |
1147 | now = ktime_get(); |
1148 | ||
bbe9a70a | 1149 | if (idle_active) |
e8fcaa5c | 1150 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1151 | |
bbe9a70a | 1152 | if (tick_stopped) |
2aaf709a | 1153 | __tick_nohz_idle_restart_tick(ts, now); |
79bf2bb3 | 1154 | |
79bf2bb3 TG |
1155 | local_irq_enable(); |
1156 | } | |
1157 | ||
79bf2bb3 TG |
1158 | /* |
1159 | * The nohz low res interrupt handler | |
1160 | */ | |
1161 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1162 | { | |
22127e93 | 1163 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1164 | struct pt_regs *regs = get_irq_regs(); |
1165 | ktime_t now = ktime_get(); | |
1166 | ||
2456e855 | 1167 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1168 | |
ff7de620 | 1169 | tick_sched_do_timer(ts, now); |
9e8f559b | 1170 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1171 | |
b5e995e6 VK |
1172 | /* No need to reprogram if we are running tickless */ |
1173 | if (unlikely(ts->tick_stopped)) | |
1174 | return; | |
1175 | ||
0ff53d09 TG |
1176 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1177 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
79bf2bb3 TG |
1178 | } |
1179 | ||
bc7a34b8 TG |
1180 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1181 | { | |
1182 | if (!tick_nohz_enabled) | |
1183 | return; | |
1184 | ts->nohz_mode = mode; | |
1185 | /* One update is enough */ | |
1186 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
ae67bada | 1187 | timers_update_nohz(); |
bc7a34b8 TG |
1188 | } |
1189 | ||
79bf2bb3 TG |
1190 | /** |
1191 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1192 | */ | |
1193 | static void tick_nohz_switch_to_nohz(void) | |
1194 | { | |
22127e93 | 1195 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1196 | ktime_t next; |
1197 | ||
27630532 | 1198 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1199 | return; |
1200 | ||
6b442bc8 | 1201 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1202 | return; |
6b442bc8 | 1203 | |
79bf2bb3 TG |
1204 | /* |
1205 | * Recycle the hrtimer in ts, so we can share the | |
1206 | * hrtimer_forward with the highres code. | |
1207 | */ | |
1208 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1209 | /* Get the next period */ | |
1210 | next = tick_init_jiffy_update(); | |
1211 | ||
0ff53d09 | 1212 | hrtimer_set_expires(&ts->sched_timer, next); |
1ca8ec53 WL |
1213 | hrtimer_forward_now(&ts->sched_timer, tick_period); |
1214 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
bc7a34b8 | 1215 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1216 | } |
1217 | ||
5acac1be | 1218 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1219 | { |
4a32fea9 | 1220 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1221 | ktime_t now; |
1222 | ||
1223 | if (!ts->idle_active && !ts->tick_stopped) | |
1224 | return; | |
1225 | now = ktime_get(); | |
1226 | if (ts->idle_active) | |
e8fcaa5c | 1227 | tick_nohz_stop_idle(ts, now); |
ff006732 | 1228 | if (ts->tick_stopped) |
eed3b9cf | 1229 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1230 | } |
1231 | ||
79bf2bb3 TG |
1232 | #else |
1233 | ||
1234 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1235 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1236 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1237 | |
3451d024 | 1238 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1239 | |
719254fa TG |
1240 | /* |
1241 | * Called from irq_enter to notify about the possible interruption of idle() | |
1242 | */ | |
5acac1be | 1243 | void tick_irq_enter(void) |
719254fa | 1244 | { |
e8fcaa5c | 1245 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1246 | tick_nohz_irq_enter(); |
719254fa TG |
1247 | } |
1248 | ||
79bf2bb3 TG |
1249 | /* |
1250 | * High resolution timer specific code | |
1251 | */ | |
1252 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1253 | /* | |
4c9dc641 | 1254 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1255 | * Called with interrupts disabled. |
79bf2bb3 TG |
1256 | */ |
1257 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1258 | { | |
1259 | struct tick_sched *ts = | |
1260 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1261 | struct pt_regs *regs = get_irq_regs(); |
1262 | ktime_t now = ktime_get(); | |
d3ed7824 | 1263 | |
ff7de620 | 1264 | tick_sched_do_timer(ts, now); |
79bf2bb3 TG |
1265 | |
1266 | /* | |
1267 | * Do not call, when we are not in irq context and have | |
1268 | * no valid regs pointer | |
1269 | */ | |
9e8f559b FW |
1270 | if (regs) |
1271 | tick_sched_handle(ts, regs); | |
7c259045 FW |
1272 | else |
1273 | ts->next_tick = 0; | |
79bf2bb3 | 1274 | |
2a16fc93 VK |
1275 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1276 | if (unlikely(ts->tick_stopped)) | |
1277 | return HRTIMER_NORESTART; | |
1278 | ||
79bf2bb3 TG |
1279 | hrtimer_forward(timer, now, tick_period); |
1280 | ||
1281 | return HRTIMER_RESTART; | |
1282 | } | |
1283 | ||
5307c955 MG |
1284 | static int sched_skew_tick; |
1285 | ||
62cf20b3 TG |
1286 | static int __init skew_tick(char *str) |
1287 | { | |
1288 | get_option(&str, &sched_skew_tick); | |
1289 | ||
1290 | return 0; | |
1291 | } | |
1292 | early_param("skew_tick", skew_tick); | |
1293 | ||
79bf2bb3 TG |
1294 | /** |
1295 | * tick_setup_sched_timer - setup the tick emulation timer | |
1296 | */ | |
1297 | void tick_setup_sched_timer(void) | |
1298 | { | |
22127e93 | 1299 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1300 | ktime_t now = ktime_get(); |
1301 | ||
1302 | /* | |
1303 | * Emulate tick processing via per-CPU hrtimers: | |
1304 | */ | |
1305 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1306 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 1307 | |
0de7611a | 1308 | /* Get the next period (per-CPU) */ |
cc584b21 | 1309 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1310 | |
9c3f9e28 | 1311 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
1312 | if (sched_skew_tick) { |
1313 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
1314 | do_div(offset, num_possible_cpus()); | |
1315 | offset *= smp_processor_id(); | |
1316 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1317 | } | |
1318 | ||
afc08b15 TG |
1319 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1320 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); | |
bc7a34b8 | 1321 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1322 | } |
3c4fbe5e | 1323 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1324 | |
3451d024 | 1325 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1326 | void tick_cancel_sched_timer(int cpu) |
1327 | { | |
1328 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1329 | ||
3c4fbe5e | 1330 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1331 | if (ts->sched_timer.base) |
1332 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1333 | # endif |
a7901766 | 1334 | |
4b0c0f29 | 1335 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1336 | } |
3c4fbe5e | 1337 | #endif |
79bf2bb3 TG |
1338 | |
1339 | /** | |
1340 | * Async notification about clocksource changes | |
1341 | */ | |
1342 | void tick_clock_notify(void) | |
1343 | { | |
1344 | int cpu; | |
1345 | ||
1346 | for_each_possible_cpu(cpu) | |
1347 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * Async notification about clock event changes | |
1352 | */ | |
1353 | void tick_oneshot_notify(void) | |
1354 | { | |
22127e93 | 1355 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1356 | |
1357 | set_bit(0, &ts->check_clocks); | |
1358 | } | |
1359 | ||
1360 | /** | |
1361 | * Check, if a change happened, which makes oneshot possible. | |
1362 | * | |
1363 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1364 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1365 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1366 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1367 | */ |
1368 | int tick_check_oneshot_change(int allow_nohz) | |
1369 | { | |
22127e93 | 1370 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1371 | |
1372 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1373 | return 0; | |
1374 | ||
1375 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1376 | return 0; | |
1377 | ||
cf4fc6cb | 1378 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1379 | return 0; |
1380 | ||
1381 | if (!allow_nohz) | |
1382 | return 1; | |
1383 | ||
1384 | tick_nohz_switch_to_nohz(); | |
1385 | return 0; | |
1386 | } |