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. |
08ae95f4 NP |
124 | * |
125 | * If nohz_full is enabled, this should not happen because the | |
126 | * tick_do_timer_cpu never relinquishes. | |
5bb96226 | 127 | */ |
08ae95f4 NP |
128 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) { |
129 | #ifdef CONFIG_NO_HZ_FULL | |
130 | WARN_ON(tick_nohz_full_running); | |
131 | #endif | |
5bb96226 | 132 | tick_do_timer_cpu = cpu; |
08ae95f4 | 133 | } |
5bb96226 FW |
134 | #endif |
135 | ||
136 | /* Check, if the jiffies need an update */ | |
137 | if (tick_do_timer_cpu == cpu) | |
138 | tick_do_update_jiffies64(now); | |
ff7de620 RW |
139 | |
140 | if (ts->inidle) | |
141 | ts->got_idle_tick = 1; | |
5bb96226 FW |
142 | } |
143 | ||
9e8f559b FW |
144 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
145 | { | |
3451d024 | 146 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
147 | /* |
148 | * When we are idle and the tick is stopped, we have to touch | |
149 | * the watchdog as we might not schedule for a really long | |
150 | * time. This happens on complete idle SMP systems while | |
151 | * waiting on the login prompt. We also increment the "start of | |
152 | * idle" jiffy stamp so the idle accounting adjustment we do | |
153 | * when we go busy again does not account too much ticks. | |
154 | */ | |
155 | if (ts->tick_stopped) { | |
03e0d461 | 156 | touch_softlockup_watchdog_sched(); |
9e8f559b FW |
157 | if (is_idle_task(current)) |
158 | ts->idle_jiffies++; | |
411fe24e FW |
159 | /* |
160 | * In case the current tick fired too early past its expected | |
161 | * expiration, make sure we don't bypass the next clock reprogramming | |
162 | * to the same deadline. | |
163 | */ | |
164 | ts->next_tick = 0; | |
9e8f559b | 165 | } |
94a57140 | 166 | #endif |
9e8f559b FW |
167 | update_process_times(user_mode(regs)); |
168 | profile_tick(CPU_PROFILING); | |
169 | } | |
7809998a | 170 | #endif |
9e8f559b | 171 | |
c5bfece2 | 172 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 173 | cpumask_var_t tick_nohz_full_mask; |
73867dcd | 174 | bool tick_nohz_full_running; |
f009a7a7 | 175 | static atomic_t tick_dep_mask; |
a831881b | 176 | |
f009a7a7 | 177 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 178 | { |
f009a7a7 FW |
179 | int val = atomic_read(dep); |
180 | ||
181 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 182 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 183 | return true; |
d027d45d FW |
184 | } |
185 | ||
f009a7a7 | 186 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 187 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 188 | return true; |
d027d45d FW |
189 | } |
190 | ||
f009a7a7 | 191 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 192 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 193 | return true; |
d027d45d FW |
194 | } |
195 | ||
f009a7a7 | 196 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 197 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
198 | return true; |
199 | } | |
200 | ||
201 | return false; | |
d027d45d FW |
202 | } |
203 | ||
57ccdf44 | 204 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d | 205 | { |
ebf3adba | 206 | lockdep_assert_irqs_disabled(); |
9014c45d | 207 | |
57ccdf44 WL |
208 | if (unlikely(!cpu_online(cpu))) |
209 | return false; | |
210 | ||
f009a7a7 | 211 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 212 | return false; |
d027d45d | 213 | |
f009a7a7 | 214 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 215 | return false; |
d027d45d | 216 | |
f009a7a7 | 217 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 218 | return false; |
d027d45d | 219 | |
f009a7a7 | 220 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 221 | return false; |
d027d45d | 222 | |
9014c45d FW |
223 | return true; |
224 | } | |
225 | ||
d027d45d | 226 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 227 | { |
73738a95 | 228 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
229 | } |
230 | ||
231 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = { | |
d027d45d | 232 | .func = nohz_full_kick_func, |
76c24fb0 FW |
233 | }; |
234 | ||
40bea039 FW |
235 | /* |
236 | * Kick this CPU if it's full dynticks in order to force it to | |
237 | * re-evaluate its dependency on the tick and restart it if necessary. | |
238 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
239 | * is NMI safe. | |
240 | */ | |
555e0c1e | 241 | static void tick_nohz_full_kick(void) |
40bea039 FW |
242 | { |
243 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
244 | return; | |
245 | ||
56e4dea8 | 246 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
247 | } |
248 | ||
76c24fb0 | 249 | /* |
3d36aebc | 250 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
251 | * re-evaluate its dependency on the tick and restart it if necessary. |
252 | */ | |
3d36aebc | 253 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 254 | { |
3d36aebc FW |
255 | if (!tick_nohz_full_cpu(cpu)) |
256 | return; | |
257 | ||
258 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
259 | } |
260 | ||
76c24fb0 FW |
261 | /* |
262 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
263 | * their dependency on the tick and restart it if necessary. | |
264 | */ | |
b7878300 | 265 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 266 | { |
8537bb95 FW |
267 | int cpu; |
268 | ||
73867dcd | 269 | if (!tick_nohz_full_running) |
76c24fb0 FW |
270 | return; |
271 | ||
272 | preempt_disable(); | |
8537bb95 FW |
273 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
274 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
275 | preempt_enable(); |
276 | } | |
277 | ||
f009a7a7 | 278 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
279 | enum tick_dep_bits bit) |
280 | { | |
f009a7a7 | 281 | int prev; |
d027d45d | 282 | |
a1cc5bcf | 283 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
284 | if (!prev) |
285 | tick_nohz_full_kick_all(); | |
286 | } | |
287 | ||
288 | /* | |
289 | * Set a global tick dependency. Used by perf events that rely on freq and | |
290 | * by unstable clock. | |
291 | */ | |
292 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
293 | { | |
294 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
295 | } | |
296 | ||
297 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
298 | { | |
f009a7a7 | 299 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
300 | } |
301 | ||
302 | /* | |
303 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
304 | * manage events throttling. | |
305 | */ | |
306 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
307 | { | |
f009a7a7 | 308 | int prev; |
d027d45d FW |
309 | struct tick_sched *ts; |
310 | ||
311 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
312 | ||
a1cc5bcf | 313 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
314 | if (!prev) { |
315 | preempt_disable(); | |
316 | /* Perf needs local kick that is NMI safe */ | |
317 | if (cpu == smp_processor_id()) { | |
318 | tick_nohz_full_kick(); | |
319 | } else { | |
320 | /* Remote irq work not NMI-safe */ | |
321 | if (!WARN_ON_ONCE(in_nmi())) | |
322 | tick_nohz_full_kick_cpu(cpu); | |
323 | } | |
324 | preempt_enable(); | |
325 | } | |
326 | } | |
327 | ||
328 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
329 | { | |
330 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
331 | ||
f009a7a7 | 332 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
333 | } |
334 | ||
335 | /* | |
336 | * Set a per-task tick dependency. Posix CPU timers need this in order to elapse | |
337 | * per task timers. | |
338 | */ | |
339 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
340 | { | |
341 | /* | |
342 | * We could optimize this with just kicking the target running the task | |
343 | * if that noise matters for nohz full users. | |
344 | */ | |
345 | tick_nohz_dep_set_all(&tsk->tick_dep_mask, bit); | |
346 | } | |
347 | ||
348 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
349 | { | |
f009a7a7 | 350 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d FW |
351 | } |
352 | ||
353 | /* | |
354 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
355 | * per process timers. | |
356 | */ | |
357 | void tick_nohz_dep_set_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
358 | { | |
359 | tick_nohz_dep_set_all(&sig->tick_dep_mask, bit); | |
360 | } | |
361 | ||
362 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
363 | { | |
f009a7a7 | 364 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
365 | } |
366 | ||
99e5ada9 FW |
367 | /* |
368 | * Re-evaluate the need for the tick as we switch the current task. | |
369 | * It might need the tick due to per task/process properties: | |
0de7611a | 370 | * perf events, posix CPU timers, ... |
99e5ada9 | 371 | */ |
de734f89 | 372 | void __tick_nohz_task_switch(void) |
99e5ada9 FW |
373 | { |
374 | unsigned long flags; | |
d027d45d | 375 | struct tick_sched *ts; |
99e5ada9 | 376 | |
99e5ada9 FW |
377 | local_irq_save(flags); |
378 | ||
6296ace4 LZ |
379 | if (!tick_nohz_full_cpu(smp_processor_id())) |
380 | goto out; | |
381 | ||
d027d45d | 382 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 383 | |
d027d45d | 384 | if (ts->tick_stopped) { |
f009a7a7 FW |
385 | if (atomic_read(¤t->tick_dep_mask) || |
386 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
387 | tick_nohz_full_kick(); |
388 | } | |
6296ace4 | 389 | out: |
99e5ada9 FW |
390 | local_irq_restore(flags); |
391 | } | |
392 | ||
6f1982fe FW |
393 | /* Get the boot-time nohz CPU list from the kernel parameters. */ |
394 | void __init tick_nohz_full_setup(cpumask_var_t cpumask) | |
a831881b | 395 | { |
73867dcd | 396 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
6f1982fe | 397 | cpumask_copy(tick_nohz_full_mask, cpumask); |
73867dcd | 398 | tick_nohz_full_running = true; |
a831881b | 399 | } |
a831881b | 400 | |
31eff243 | 401 | static int tick_nohz_cpu_down(unsigned int cpu) |
a382bf93 | 402 | { |
31eff243 | 403 | /* |
08ae95f4 NP |
404 | * The tick_do_timer_cpu CPU handles housekeeping duty (unbound |
405 | * timers, workqueues, timekeeping, ...) on behalf of full dynticks | |
31eff243 SAS |
406 | * CPUs. It must remain online when nohz full is enabled. |
407 | */ | |
408 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) | |
409 | return -EBUSY; | |
410 | return 0; | |
a382bf93 FW |
411 | } |
412 | ||
d1e43fa5 | 413 | void __init tick_nohz_init(void) |
a831881b | 414 | { |
31eff243 | 415 | int cpu, ret; |
d1e43fa5 | 416 | |
a7c8655b PM |
417 | if (!tick_nohz_full_running) |
418 | return; | |
d1e43fa5 | 419 | |
9b01f5bf FW |
420 | /* |
421 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
422 | * locking contexts. But then we need irq work to raise its own | |
423 | * interrupts to avoid circular dependency on the tick | |
424 | */ | |
425 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 426 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf | 427 | cpumask_clear(tick_nohz_full_mask); |
9b01f5bf FW |
428 | tick_nohz_full_running = false; |
429 | return; | |
430 | } | |
431 | ||
08ae95f4 NP |
432 | if (IS_ENABLED(CONFIG_PM_SLEEP_SMP) && |
433 | !IS_ENABLED(CONFIG_PM_SLEEP_SMP_NONZERO_CPU)) { | |
434 | cpu = smp_processor_id(); | |
4327b15f | 435 | |
08ae95f4 NP |
436 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { |
437 | pr_warn("NO_HZ: Clearing %d from nohz_full range " | |
438 | "for timekeeping\n", cpu); | |
439 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); | |
440 | } | |
4327b15f FW |
441 | } |
442 | ||
73867dcd | 443 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
444 | context_tracking_cpu_set(cpu); |
445 | ||
31eff243 SAS |
446 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
447 | "kernel/nohz:predown", NULL, | |
448 | tick_nohz_cpu_down); | |
449 | WARN_ON(ret < 0); | |
ffda22c1 TH |
450 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
451 | cpumask_pr_args(tick_nohz_full_mask)); | |
a831881b | 452 | } |
a831881b FW |
453 | #endif |
454 | ||
79bf2bb3 TG |
455 | /* |
456 | * NOHZ - aka dynamic tick functionality | |
457 | */ | |
3451d024 | 458 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
459 | /* |
460 | * NO HZ enabled ? | |
461 | */ | |
4cc7ecb7 | 462 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 463 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
464 | /* |
465 | * Enable / Disable tickless mode | |
466 | */ | |
467 | static int __init setup_tick_nohz(char *str) | |
468 | { | |
4cc7ecb7 | 469 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
470 | } |
471 | ||
472 | __setup("nohz=", setup_tick_nohz); | |
473 | ||
a3642983 | 474 | bool tick_nohz_tick_stopped(void) |
c1797baf | 475 | { |
2bc629a6 FW |
476 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
477 | ||
478 | return ts->tick_stopped; | |
c1797baf TG |
479 | } |
480 | ||
22ab8bc0 FW |
481 | bool tick_nohz_tick_stopped_cpu(int cpu) |
482 | { | |
483 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
484 | ||
485 | return ts->tick_stopped; | |
486 | } | |
487 | ||
79bf2bb3 TG |
488 | /** |
489 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
490 | * | |
491 | * Called from interrupt entry when the CPU was idle | |
492 | * | |
493 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
494 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
495 | * value. We do this unconditionally on any CPU, as we don't know whether the |
496 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 497 | */ |
eed3b9cf | 498 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 499 | { |
79bf2bb3 | 500 | unsigned long flags; |
79bf2bb3 | 501 | |
e8fcaa5c | 502 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
503 | |
504 | local_irq_save(flags); | |
505 | tick_do_update_jiffies64(now); | |
506 | local_irq_restore(flags); | |
02ff3755 | 507 | |
03e0d461 | 508 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
509 | } |
510 | ||
595aac48 | 511 | /* |
0de7611a | 512 | * Updates the per-CPU time idle statistics counters |
595aac48 | 513 | */ |
8d63bf94 | 514 | static void |
8c215bd3 | 515 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 516 | { |
eed3b9cf | 517 | ktime_t delta; |
6378ddb5 | 518 | |
595aac48 AV |
519 | if (ts->idle_active) { |
520 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 521 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 522 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
523 | else |
524 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 525 | ts->idle_entrytime = now; |
595aac48 | 526 | } |
8d63bf94 | 527 | |
e0e37c20 | 528 | if (last_update_time) |
8d63bf94 AV |
529 | *last_update_time = ktime_to_us(now); |
530 | ||
595aac48 AV |
531 | } |
532 | ||
e8fcaa5c | 533 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 534 | { |
e8fcaa5c | 535 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 536 | ts->idle_active = 0; |
56c7426b | 537 | |
ac1e843f | 538 | sched_clock_idle_wakeup_event(); |
6378ddb5 VP |
539 | } |
540 | ||
0e776768 | 541 | static void tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 542 | { |
0e776768 | 543 | ts->idle_entrytime = ktime_get(); |
6378ddb5 | 544 | ts->idle_active = 1; |
56c7426b | 545 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
546 | } |
547 | ||
b1f724c3 | 548 | /** |
0de7611a | 549 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 550 | * @cpu: CPU number to query |
09a1d34f MH |
551 | * @last_update_time: variable to store update time in. Do not update |
552 | * counters if NULL. | |
b1f724c3 | 553 | * |
6168f8ed | 554 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 555 | * CPU, in microseconds. |
b1f724c3 AV |
556 | * |
557 | * This time is measured via accounting rather than sampling, | |
558 | * and is as accurate as ktime_get() is. | |
559 | * | |
560 | * This function returns -1 if NOHZ is not enabled. | |
561 | */ | |
6378ddb5 VP |
562 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
563 | { | |
564 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 565 | ktime_t now, idle; |
6378ddb5 | 566 | |
d689fe22 | 567 | if (!tick_nohz_active) |
8083e4ad | 568 | return -1; |
569 | ||
09a1d34f MH |
570 | now = ktime_get(); |
571 | if (last_update_time) { | |
572 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
573 | idle = ts->idle_sleeptime; | |
574 | } else { | |
575 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
576 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
577 | ||
578 | idle = ktime_add(ts->idle_sleeptime, delta); | |
579 | } else { | |
580 | idle = ts->idle_sleeptime; | |
581 | } | |
582 | } | |
583 | ||
584 | return ktime_to_us(idle); | |
8083e4ad | 585 | |
6378ddb5 | 586 | } |
8083e4ad | 587 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 588 | |
6beea0cd | 589 | /** |
0de7611a | 590 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 591 | * @cpu: CPU number to query |
09a1d34f MH |
592 | * @last_update_time: variable to store update time in. Do not update |
593 | * counters if NULL. | |
0224cf4c | 594 | * |
6168f8ed | 595 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
596 | * CPU, in microseconds. |
597 | * | |
598 | * This time is measured via accounting rather than sampling, | |
599 | * and is as accurate as ktime_get() is. | |
600 | * | |
601 | * This function returns -1 if NOHZ is not enabled. | |
602 | */ | |
603 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
604 | { | |
605 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 606 | ktime_t now, iowait; |
0224cf4c | 607 | |
d689fe22 | 608 | if (!tick_nohz_active) |
0224cf4c AV |
609 | return -1; |
610 | ||
09a1d34f MH |
611 | now = ktime_get(); |
612 | if (last_update_time) { | |
613 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
614 | iowait = ts->iowait_sleeptime; | |
615 | } else { | |
616 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
617 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 618 | |
09a1d34f MH |
619 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
620 | } else { | |
621 | iowait = ts->iowait_sleeptime; | |
622 | } | |
623 | } | |
0224cf4c | 624 | |
09a1d34f | 625 | return ktime_to_us(iowait); |
0224cf4c AV |
626 | } |
627 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
628 | ||
0ff53d09 TG |
629 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
630 | { | |
631 | hrtimer_cancel(&ts->sched_timer); | |
632 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
633 | ||
634 | /* Forward the time to expire in the future */ | |
635 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
636 | ||
902a9f9c SAS |
637 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
638 | hrtimer_start_expires(&ts->sched_timer, | |
639 | HRTIMER_MODE_ABS_PINNED_HARD); | |
640 | } else { | |
0ff53d09 | 641 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
902a9f9c | 642 | } |
411fe24e FW |
643 | |
644 | /* | |
645 | * Reset to make sure next tick stop doesn't get fooled by past | |
646 | * cached clock deadline. | |
647 | */ | |
648 | ts->next_tick = 0; | |
0ff53d09 TG |
649 | } |
650 | ||
5d62c183 TG |
651 | static inline bool local_timer_softirq_pending(void) |
652 | { | |
80d20d35 | 653 | return local_softirq_pending() & BIT(TIMER_SOFTIRQ); |
5d62c183 TG |
654 | } |
655 | ||
23a8d888 | 656 | static ktime_t tick_nohz_next_event(struct tick_sched *ts, int cpu) |
79bf2bb3 | 657 | { |
c1ad348b | 658 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
e1e41b6c RV |
659 | unsigned long basejiff; |
660 | unsigned int seq; | |
855a0fc3 | 661 | |
79bf2bb3 TG |
662 | /* Read jiffies and the time when jiffies were updated last */ |
663 | do { | |
d6ad4187 | 664 | seq = read_seqbegin(&jiffies_lock); |
2456e855 | 665 | basemono = last_jiffies_update; |
c1ad348b | 666 | basejiff = jiffies; |
d6ad4187 | 667 | } while (read_seqretry(&jiffies_lock, seq)); |
c1ad348b | 668 | ts->last_jiffies = basejiff; |
23a8d888 | 669 | ts->timer_expires_base = basemono; |
79bf2bb3 | 670 | |
5d62c183 TG |
671 | /* |
672 | * Keep the periodic tick, when RCU, architecture or irq_work | |
673 | * requests it. | |
674 | * Aside of that check whether the local timer softirq is | |
675 | * pending. If so its a bad idea to call get_next_timer_interrupt() | |
676 | * because there is an already expired timer, so it will request | |
677 | * immeditate expiry, which rearms the hardware timer with a | |
678 | * minimal delta which brings us back to this place | |
679 | * immediately. Lather, rinse and repeat... | |
680 | */ | |
681 | if (rcu_needs_cpu(basemono, &next_rcu) || arch_needs_cpu() || | |
682 | irq_work_needs_cpu() || local_timer_softirq_pending()) { | |
c1ad348b | 683 | next_tick = basemono + TICK_NSEC; |
3c5d92a0 | 684 | } else { |
c1ad348b TG |
685 | /* |
686 | * Get the next pending timer. If high resolution | |
687 | * timers are enabled this only takes the timer wheel | |
688 | * timers into account. If high resolution timers are | |
689 | * disabled this also looks at the next expiring | |
690 | * hrtimer. | |
691 | */ | |
692 | next_tmr = get_next_timer_interrupt(basejiff, basemono); | |
693 | ts->next_timer = next_tmr; | |
694 | /* Take the next rcu event into account */ | |
695 | next_tick = next_rcu < next_tmr ? next_rcu : next_tmr; | |
3c5d92a0 | 696 | } |
47aa8b6c | 697 | |
c1ad348b TG |
698 | /* |
699 | * If the tick is due in the next period, keep it ticking or | |
82bbe34b | 700 | * force prod the timer. |
c1ad348b TG |
701 | */ |
702 | delta = next_tick - basemono; | |
703 | if (delta <= (u64)TICK_NSEC) { | |
a683f390 TG |
704 | /* |
705 | * Tell the timer code that the base is not idle, i.e. undo | |
706 | * the effect of get_next_timer_interrupt(): | |
707 | */ | |
708 | timer_clear_idle(); | |
82bbe34b PZ |
709 | /* |
710 | * We've not stopped the tick yet, and there's a timer in the | |
711 | * next period, so no point in stopping it either, bail. | |
712 | */ | |
f99973e1 | 713 | if (!ts->tick_stopped) { |
23a8d888 | 714 | ts->timer_expires = 0; |
157d29e1 TG |
715 | goto out; |
716 | } | |
717 | } | |
718 | ||
23a8d888 RW |
719 | /* |
720 | * If this CPU is the one which had the do_timer() duty last, we limit | |
721 | * the sleep time to the timekeeping max_deferment value. | |
722 | * Otherwise we can sleep as long as we want. | |
723 | */ | |
724 | delta = timekeeping_max_deferment(); | |
725 | if (cpu != tick_do_timer_cpu && | |
726 | (tick_do_timer_cpu != TICK_DO_TIMER_NONE || !ts->do_timer_last)) | |
727 | delta = KTIME_MAX; | |
728 | ||
729 | /* Calculate the next expiry time */ | |
730 | if (delta < (KTIME_MAX - basemono)) | |
731 | expires = basemono + delta; | |
732 | else | |
733 | expires = KTIME_MAX; | |
734 | ||
735 | ts->timer_expires = min_t(u64, expires, next_tick); | |
736 | ||
737 | out: | |
738 | return ts->timer_expires; | |
739 | } | |
740 | ||
741 | static void tick_nohz_stop_tick(struct tick_sched *ts, int cpu) | |
742 | { | |
743 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); | |
744 | u64 basemono = ts->timer_expires_base; | |
745 | u64 expires = ts->timer_expires; | |
746 | ktime_t tick = expires; | |
747 | ||
748 | /* Make sure we won't be trying to stop it twice in a row. */ | |
749 | ts->timer_expires_base = 0; | |
750 | ||
79bf2bb3 | 751 | /* |
0de7611a IM |
752 | * If this CPU is the one which updates jiffies, then give up |
753 | * the assignment and let it be taken by the CPU which runs | |
754 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
755 | * don't drop this here the jiffies might be stale and |
756 | * do_timer() never invoked. Keep track of the fact that it | |
23a8d888 | 757 | * was the one which had the do_timer() duty last. |
79bf2bb3 | 758 | */ |
157d29e1 TG |
759 | if (cpu == tick_do_timer_cpu) { |
760 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
761 | ts->do_timer_last = 1; | |
762 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
157d29e1 | 763 | ts->do_timer_last = 0; |
157d29e1 | 764 | } |
27185016 | 765 | |
157d29e1 | 766 | /* Skip reprogram of event if its not changed */ |
411fe24e FW |
767 | if (ts->tick_stopped && (expires == ts->next_tick)) { |
768 | /* Sanity check: make sure clockevent is actually programmed */ | |
d4af6d93 | 769 | if (tick == KTIME_MAX || ts->next_tick == hrtimer_get_expires(&ts->sched_timer)) |
23a8d888 | 770 | return; |
411fe24e FW |
771 | |
772 | WARN_ON_ONCE(1); | |
773 | printk_once("basemono: %llu ts->next_tick: %llu dev->next_event: %llu timer->active: %d timer->expires: %llu\n", | |
774 | basemono, ts->next_tick, dev->next_event, | |
775 | hrtimer_active(&ts->sched_timer), hrtimer_get_expires(&ts->sched_timer)); | |
ce6cf9a1 | 776 | } |
84bf1bcc | 777 | |
157d29e1 TG |
778 | /* |
779 | * nohz_stop_sched_tick can be called several times before | |
780 | * the nohz_restart_sched_tick is called. This happens when | |
781 | * interrupts arrive which do not cause a reschedule. In the | |
782 | * first call we save the current tick time, so we can restart | |
783 | * the scheduler tick in nohz_restart_sched_tick. | |
784 | */ | |
785 | if (!ts->tick_stopped) { | |
3c85d6db | 786 | calc_load_nohz_start(); |
62cb1188 | 787 | quiet_vmstat(); |
d3ed7824 | 788 | |
157d29e1 TG |
789 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
790 | ts->tick_stopped = 1; | |
e6e6cc22 | 791 | trace_tick_stop(1, TICK_DEP_MASK_NONE); |
157d29e1 | 792 | } |
eaad084b | 793 | |
411fe24e FW |
794 | ts->next_tick = tick; |
795 | ||
157d29e1 | 796 | /* |
c1ad348b TG |
797 | * If the expiration time == KTIME_MAX, then we simply stop |
798 | * the tick timer. | |
157d29e1 | 799 | */ |
c1ad348b | 800 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
801 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
802 | hrtimer_cancel(&ts->sched_timer); | |
23a8d888 | 803 | return; |
79bf2bb3 | 804 | } |
0ff53d09 | 805 | |
1f71addd | 806 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
902a9f9c SAS |
807 | hrtimer_start(&ts->sched_timer, tick, |
808 | HRTIMER_MODE_ABS_PINNED_HARD); | |
1f71addd TG |
809 | } else { |
810 | hrtimer_set_expires(&ts->sched_timer, tick); | |
c1ad348b | 811 | tick_program_event(tick, 1); |
1f71addd | 812 | } |
280f0677 FW |
813 | } |
814 | ||
23a8d888 RW |
815 | static void tick_nohz_retain_tick(struct tick_sched *ts) |
816 | { | |
817 | ts->timer_expires_base = 0; | |
818 | } | |
819 | ||
820 | #ifdef CONFIG_NO_HZ_FULL | |
821 | static void tick_nohz_stop_sched_tick(struct tick_sched *ts, int cpu) | |
822 | { | |
823 | if (tick_nohz_next_event(ts, cpu)) | |
824 | tick_nohz_stop_tick(ts, cpu); | |
825 | else | |
826 | tick_nohz_retain_tick(ts); | |
827 | } | |
828 | #endif /* CONFIG_NO_HZ_FULL */ | |
829 | ||
1f41906a | 830 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
831 | { |
832 | /* Update jiffies first */ | |
833 | tick_do_update_jiffies64(now); | |
59d2c7ca | 834 | |
a683f390 TG |
835 | /* |
836 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
837 | * the clock forward checks in the enqueue path: | |
838 | */ | |
839 | timer_clear_idle(); | |
840 | ||
3c85d6db | 841 | calc_load_nohz_stop(); |
03e0d461 | 842 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
843 | /* |
844 | * Cancel the scheduled timer and restore the tick | |
845 | */ | |
846 | ts->tick_stopped = 0; | |
847 | ts->idle_exittime = now; | |
848 | ||
849 | tick_nohz_restart(ts, now); | |
850 | } | |
73738a95 FW |
851 | |
852 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
5811d996 FW |
853 | { |
854 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 855 | int cpu = smp_processor_id(); |
5811d996 | 856 | |
59449359 | 857 | if (!tick_nohz_full_cpu(cpu)) |
e9a2eb40 | 858 | return; |
5811d996 | 859 | |
e9a2eb40 AS |
860 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
861 | return; | |
5811d996 | 862 | |
57ccdf44 | 863 | if (can_stop_full_tick(cpu, ts)) |
23a8d888 | 864 | tick_nohz_stop_sched_tick(ts, cpu); |
73738a95 | 865 | else if (ts->tick_stopped) |
1f41906a | 866 | tick_nohz_restart_sched_tick(ts, ktime_get()); |
5811d996 FW |
867 | #endif |
868 | } | |
869 | ||
5b39939a FW |
870 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
871 | { | |
872 | /* | |
0de7611a | 873 | * If this CPU is offline and it is the one which updates |
5b39939a | 874 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 875 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
876 | * this here the jiffies might be stale and do_timer() never |
877 | * invoked. | |
878 | */ | |
879 | if (unlikely(!cpu_online(cpu))) { | |
880 | if (cpu == tick_do_timer_cpu) | |
881 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
411fe24e FW |
882 | /* |
883 | * Make sure the CPU doesn't get fooled by obsolete tick | |
884 | * deadline if it comes back online later. | |
885 | */ | |
886 | ts->next_tick = 0; | |
f7ea0fd6 | 887 | return false; |
5b39939a FW |
888 | } |
889 | ||
23a8d888 | 890 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) |
5b39939a FW |
891 | return false; |
892 | ||
893 | if (need_resched()) | |
894 | return false; | |
895 | ||
d59e0ba1 | 896 | if (unlikely(local_softirq_pending())) { |
5b39939a FW |
897 | static int ratelimit; |
898 | ||
803b0eba PM |
899 | if (ratelimit < 10 && |
900 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
cfea7d7e RV |
901 | pr_warn("NOHZ: local_softirq_pending %02x\n", |
902 | (unsigned int) local_softirq_pending()); | |
5b39939a FW |
903 | ratelimit++; |
904 | } | |
905 | return false; | |
906 | } | |
907 | ||
460775df | 908 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
909 | /* |
910 | * Keep the tick alive to guarantee timekeeping progression | |
911 | * if there are full dynticks CPUs around | |
912 | */ | |
913 | if (tick_do_timer_cpu == cpu) | |
914 | return false; | |
915 | /* | |
916 | * Boot safety: make sure the timekeeping duty has been | |
917 | * assigned before entering dyntick-idle mode, | |
08ae95f4 | 918 | * tick_do_timer_cpu is TICK_DO_TIMER_BOOT |
a382bf93 | 919 | */ |
08ae95f4 NP |
920 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_BOOT)) |
921 | return false; | |
922 | ||
923 | /* Should not happen for nohz-full */ | |
924 | if (WARN_ON_ONCE(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) | |
a382bf93 FW |
925 | return false; |
926 | } | |
927 | ||
5b39939a FW |
928 | return true; |
929 | } | |
930 | ||
0e776768 | 931 | static void __tick_nohz_idle_stop_tick(struct tick_sched *ts) |
19f5f736 | 932 | { |
0e776768 | 933 | ktime_t expires; |
5b39939a | 934 | int cpu = smp_processor_id(); |
19f5f736 | 935 | |
554c8aa8 RW |
936 | /* |
937 | * If tick_nohz_get_sleep_length() ran tick_nohz_next_event(), the | |
938 | * tick timer expiration time is known already. | |
939 | */ | |
940 | if (ts->timer_expires_base) | |
941 | expires = ts->timer_expires; | |
942 | else if (can_stop_idle_tick(cpu, ts)) | |
943 | expires = tick_nohz_next_event(ts, cpu); | |
944 | else | |
945 | return; | |
23a8d888 RW |
946 | |
947 | ts->idle_calls++; | |
08d07259 | 948 | |
23a8d888 | 949 | if (expires > 0LL) { |
5b39939a FW |
950 | int was_stopped = ts->tick_stopped; |
951 | ||
23a8d888 | 952 | tick_nohz_stop_tick(ts, cpu); |
84bf1bcc | 953 | |
23a8d888 RW |
954 | ts->idle_sleeps++; |
955 | ts->idle_expires = expires; | |
5b39939a | 956 | |
a0db971e | 957 | if (!was_stopped && ts->tick_stopped) { |
5b39939a | 958 | ts->idle_jiffies = ts->last_jiffies; |
a0db971e FW |
959 | nohz_balance_enter_idle(cpu); |
960 | } | |
23a8d888 RW |
961 | } else { |
962 | tick_nohz_retain_tick(ts); | |
5b39939a | 963 | } |
280f0677 FW |
964 | } |
965 | ||
966 | /** | |
0e776768 | 967 | * tick_nohz_idle_stop_tick - stop the idle tick from the idle task |
280f0677 FW |
968 | * |
969 | * When the next event is more than a tick into the future, stop the idle tick | |
0e776768 RW |
970 | */ |
971 | void tick_nohz_idle_stop_tick(void) | |
972 | { | |
973 | __tick_nohz_idle_stop_tick(this_cpu_ptr(&tick_cpu_sched)); | |
974 | } | |
975 | ||
554c8aa8 RW |
976 | void tick_nohz_idle_retain_tick(void) |
977 | { | |
978 | tick_nohz_retain_tick(this_cpu_ptr(&tick_cpu_sched)); | |
979 | /* | |
980 | * Undo the effect of get_next_timer_interrupt() called from | |
981 | * tick_nohz_next_event(). | |
982 | */ | |
983 | timer_clear_idle(); | |
984 | } | |
985 | ||
0e776768 RW |
986 | /** |
987 | * tick_nohz_idle_enter - prepare for entering idle on the current CPU | |
2bbb6817 | 988 | * |
0e776768 | 989 | * Called when we start the idle loop. |
280f0677 | 990 | */ |
1268fbc7 | 991 | void tick_nohz_idle_enter(void) |
280f0677 FW |
992 | { |
993 | struct tick_sched *ts; | |
994 | ||
ebf3adba | 995 | lockdep_assert_irqs_enabled(); |
0db49b72 | 996 | |
1268fbc7 FW |
997 | local_irq_disable(); |
998 | ||
22127e93 | 999 | ts = this_cpu_ptr(&tick_cpu_sched); |
23a8d888 RW |
1000 | |
1001 | WARN_ON_ONCE(ts->timer_expires_base); | |
1002 | ||
280f0677 | 1003 | ts->inidle = 1; |
0e776768 | 1004 | tick_nohz_start_idle(ts); |
1268fbc7 FW |
1005 | |
1006 | local_irq_enable(); | |
280f0677 FW |
1007 | } |
1008 | ||
1009 | /** | |
1010 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
1011 | * | |
1012 | * When an interrupt fires while we are idle and it doesn't cause | |
1013 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
1014 | * an RCU callback, etc... | |
1015 | * So we need to re-calculate and reprogram the next tick event. | |
1016 | */ | |
1017 | void tick_nohz_irq_exit(void) | |
1018 | { | |
22127e93 | 1019 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 1020 | |
14851912 | 1021 | if (ts->inidle) |
0e776768 | 1022 | tick_nohz_start_idle(ts); |
14851912 | 1023 | else |
73738a95 | 1024 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
1025 | } |
1026 | ||
4f86d3a8 | 1027 | /** |
45f1ff59 RW |
1028 | * tick_nohz_idle_got_tick - Check whether or not the tick handler has run |
1029 | */ | |
1030 | bool tick_nohz_idle_got_tick(void) | |
1031 | { | |
1032 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1033 | ||
2bc629a6 FW |
1034 | if (ts->got_idle_tick) { |
1035 | ts->got_idle_tick = 0; | |
45f1ff59 RW |
1036 | return true; |
1037 | } | |
1038 | return false; | |
1039 | } | |
1040 | ||
6f9b83ac UH |
1041 | /** |
1042 | * tick_nohz_get_next_hrtimer - return the next expiration time for the hrtimer | |
1043 | * or the tick, whatever that expires first. Note that, if the tick has been | |
1044 | * stopped, it returns the next hrtimer. | |
1045 | * | |
1046 | * Called from power state control code with interrupts disabled | |
1047 | */ | |
1048 | ktime_t tick_nohz_get_next_hrtimer(void) | |
1049 | { | |
1050 | return __this_cpu_read(tick_cpu_device.evtdev)->next_event; | |
1051 | } | |
1052 | ||
4f86d3a8 | 1053 | /** |
554c8aa8 | 1054 | * tick_nohz_get_sleep_length - return the expected length of the current sleep |
296bb1e5 | 1055 | * @delta_next: duration until the next event if the tick cannot be stopped |
4f86d3a8 LB |
1056 | * |
1057 | * Called from power state control code with interrupts disabled | |
1058 | */ | |
296bb1e5 | 1059 | ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next) |
4f86d3a8 | 1060 | { |
554c8aa8 | 1061 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
22127e93 | 1062 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
554c8aa8 RW |
1063 | int cpu = smp_processor_id(); |
1064 | /* | |
1065 | * The idle entry time is expected to be a sufficient approximation of | |
1066 | * the current time at this point. | |
1067 | */ | |
1068 | ktime_t now = ts->idle_entrytime; | |
1069 | ktime_t next_event; | |
1070 | ||
1071 | WARN_ON_ONCE(!ts->inidle); | |
1072 | ||
296bb1e5 RW |
1073 | *delta_next = ktime_sub(dev->next_event, now); |
1074 | ||
554c8aa8 | 1075 | if (!can_stop_idle_tick(cpu, ts)) |
296bb1e5 | 1076 | return *delta_next; |
554c8aa8 RW |
1077 | |
1078 | next_event = tick_nohz_next_event(ts, cpu); | |
1079 | if (!next_event) | |
296bb1e5 | 1080 | return *delta_next; |
554c8aa8 RW |
1081 | |
1082 | /* | |
1083 | * If the next highres timer to expire is earlier than next_event, the | |
1084 | * idle governor needs to know that. | |
1085 | */ | |
1086 | next_event = min_t(u64, next_event, | |
1087 | hrtimer_next_event_without(&ts->sched_timer)); | |
4f86d3a8 | 1088 | |
554c8aa8 | 1089 | return ktime_sub(next_event, now); |
4f86d3a8 LB |
1090 | } |
1091 | ||
466a2b42 JF |
1092 | /** |
1093 | * tick_nohz_get_idle_calls_cpu - return the current idle calls counter value | |
1094 | * for a particular CPU. | |
1095 | * | |
1096 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1097 | */ | |
1098 | unsigned long tick_nohz_get_idle_calls_cpu(int cpu) | |
1099 | { | |
1100 | struct tick_sched *ts = tick_get_tick_sched(cpu); | |
1101 | ||
1102 | return ts->idle_calls; | |
1103 | } | |
1104 | ||
b7eaf1aa RW |
1105 | /** |
1106 | * tick_nohz_get_idle_calls - return the current idle calls counter value | |
1107 | * | |
1108 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1109 | */ | |
1110 | unsigned long tick_nohz_get_idle_calls(void) | |
1111 | { | |
1112 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1113 | ||
1114 | return ts->idle_calls; | |
1115 | } | |
1116 | ||
2ac0d98f FW |
1117 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) |
1118 | { | |
3f4724ea | 1119 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 1120 | unsigned long ticks; |
3f4724ea | 1121 | |
55dbdcfa | 1122 | if (vtime_accounting_cpu_enabled()) |
3f4724ea | 1123 | return; |
79bf2bb3 TG |
1124 | /* |
1125 | * We stopped the tick in idle. Update process times would miss the | |
1126 | * time we slept as update_process_times does only a 1 tick | |
1127 | * accounting. Enforce that this is accounted to idle ! | |
1128 | */ | |
1129 | ticks = jiffies - ts->idle_jiffies; | |
1130 | /* | |
1131 | * We might be one off. Do not randomly account a huge number of ticks! | |
1132 | */ | |
79741dd3 MS |
1133 | if (ticks && ticks < LONG_MAX) |
1134 | account_idle_ticks(ticks); | |
1135 | #endif | |
19f5f736 FW |
1136 | } |
1137 | ||
2aaf709a RW |
1138 | static void __tick_nohz_idle_restart_tick(struct tick_sched *ts, ktime_t now) |
1139 | { | |
1140 | tick_nohz_restart_sched_tick(ts, now); | |
1141 | tick_nohz_account_idle_ticks(ts); | |
1142 | } | |
1143 | ||
1144 | void tick_nohz_idle_restart_tick(void) | |
1145 | { | |
1146 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1147 | ||
1148 | if (ts->tick_stopped) | |
1149 | __tick_nohz_idle_restart_tick(ts, ktime_get()); | |
1150 | } | |
1151 | ||
79bf2bb3 | 1152 | /** |
280f0677 | 1153 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1154 | * |
1155 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1156 | * This also exit the RCU extended quiescent state. The CPU |
1157 | * can use RCU again after this function is called. | |
79bf2bb3 | 1158 | */ |
280f0677 | 1159 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1160 | { |
4a32fea9 | 1161 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
bbe9a70a | 1162 | bool idle_active, tick_stopped; |
6378ddb5 | 1163 | ktime_t now; |
79bf2bb3 | 1164 | |
6378ddb5 | 1165 | local_irq_disable(); |
2bbb6817 | 1166 | |
15f827be | 1167 | WARN_ON_ONCE(!ts->inidle); |
23a8d888 | 1168 | WARN_ON_ONCE(ts->timer_expires_base); |
15f827be FW |
1169 | |
1170 | ts->inidle = 0; | |
bbe9a70a AB |
1171 | idle_active = ts->idle_active; |
1172 | tick_stopped = ts->tick_stopped; | |
15f827be | 1173 | |
bbe9a70a | 1174 | if (idle_active || tick_stopped) |
eed3b9cf MS |
1175 | now = ktime_get(); |
1176 | ||
bbe9a70a | 1177 | if (idle_active) |
e8fcaa5c | 1178 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1179 | |
bbe9a70a | 1180 | if (tick_stopped) |
2aaf709a | 1181 | __tick_nohz_idle_restart_tick(ts, now); |
79bf2bb3 | 1182 | |
79bf2bb3 TG |
1183 | local_irq_enable(); |
1184 | } | |
1185 | ||
79bf2bb3 TG |
1186 | /* |
1187 | * The nohz low res interrupt handler | |
1188 | */ | |
1189 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1190 | { | |
22127e93 | 1191 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1192 | struct pt_regs *regs = get_irq_regs(); |
1193 | ktime_t now = ktime_get(); | |
1194 | ||
2456e855 | 1195 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1196 | |
ff7de620 | 1197 | tick_sched_do_timer(ts, now); |
9e8f559b | 1198 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1199 | |
b5e995e6 VK |
1200 | /* No need to reprogram if we are running tickless */ |
1201 | if (unlikely(ts->tick_stopped)) | |
1202 | return; | |
1203 | ||
0ff53d09 TG |
1204 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1205 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
79bf2bb3 TG |
1206 | } |
1207 | ||
bc7a34b8 TG |
1208 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1209 | { | |
1210 | if (!tick_nohz_enabled) | |
1211 | return; | |
1212 | ts->nohz_mode = mode; | |
1213 | /* One update is enough */ | |
1214 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
ae67bada | 1215 | timers_update_nohz(); |
bc7a34b8 TG |
1216 | } |
1217 | ||
79bf2bb3 TG |
1218 | /** |
1219 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1220 | */ | |
1221 | static void tick_nohz_switch_to_nohz(void) | |
1222 | { | |
22127e93 | 1223 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1224 | ktime_t next; |
1225 | ||
27630532 | 1226 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1227 | return; |
1228 | ||
6b442bc8 | 1229 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1230 | return; |
6b442bc8 | 1231 | |
79bf2bb3 TG |
1232 | /* |
1233 | * Recycle the hrtimer in ts, so we can share the | |
1234 | * hrtimer_forward with the highres code. | |
1235 | */ | |
71fed982 | 1236 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 TG |
1237 | /* Get the next period */ |
1238 | next = tick_init_jiffy_update(); | |
1239 | ||
0ff53d09 | 1240 | hrtimer_set_expires(&ts->sched_timer, next); |
1ca8ec53 WL |
1241 | hrtimer_forward_now(&ts->sched_timer, tick_period); |
1242 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
bc7a34b8 | 1243 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1244 | } |
1245 | ||
5acac1be | 1246 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1247 | { |
4a32fea9 | 1248 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1249 | ktime_t now; |
1250 | ||
1251 | if (!ts->idle_active && !ts->tick_stopped) | |
1252 | return; | |
1253 | now = ktime_get(); | |
1254 | if (ts->idle_active) | |
e8fcaa5c | 1255 | tick_nohz_stop_idle(ts, now); |
ff006732 | 1256 | if (ts->tick_stopped) |
eed3b9cf | 1257 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1258 | } |
1259 | ||
79bf2bb3 TG |
1260 | #else |
1261 | ||
1262 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1263 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1264 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1265 | |
3451d024 | 1266 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1267 | |
719254fa TG |
1268 | /* |
1269 | * Called from irq_enter to notify about the possible interruption of idle() | |
1270 | */ | |
5acac1be | 1271 | void tick_irq_enter(void) |
719254fa | 1272 | { |
e8fcaa5c | 1273 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1274 | tick_nohz_irq_enter(); |
719254fa TG |
1275 | } |
1276 | ||
79bf2bb3 TG |
1277 | /* |
1278 | * High resolution timer specific code | |
1279 | */ | |
1280 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1281 | /* | |
4c9dc641 | 1282 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1283 | * Called with interrupts disabled. |
79bf2bb3 TG |
1284 | */ |
1285 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1286 | { | |
1287 | struct tick_sched *ts = | |
1288 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1289 | struct pt_regs *regs = get_irq_regs(); |
1290 | ktime_t now = ktime_get(); | |
d3ed7824 | 1291 | |
ff7de620 | 1292 | tick_sched_do_timer(ts, now); |
79bf2bb3 TG |
1293 | |
1294 | /* | |
1295 | * Do not call, when we are not in irq context and have | |
1296 | * no valid regs pointer | |
1297 | */ | |
9e8f559b FW |
1298 | if (regs) |
1299 | tick_sched_handle(ts, regs); | |
7c259045 FW |
1300 | else |
1301 | ts->next_tick = 0; | |
79bf2bb3 | 1302 | |
2a16fc93 VK |
1303 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1304 | if (unlikely(ts->tick_stopped)) | |
1305 | return HRTIMER_NORESTART; | |
1306 | ||
79bf2bb3 TG |
1307 | hrtimer_forward(timer, now, tick_period); |
1308 | ||
1309 | return HRTIMER_RESTART; | |
1310 | } | |
1311 | ||
5307c955 MG |
1312 | static int sched_skew_tick; |
1313 | ||
62cf20b3 TG |
1314 | static int __init skew_tick(char *str) |
1315 | { | |
1316 | get_option(&str, &sched_skew_tick); | |
1317 | ||
1318 | return 0; | |
1319 | } | |
1320 | early_param("skew_tick", skew_tick); | |
1321 | ||
79bf2bb3 TG |
1322 | /** |
1323 | * tick_setup_sched_timer - setup the tick emulation timer | |
1324 | */ | |
1325 | void tick_setup_sched_timer(void) | |
1326 | { | |
22127e93 | 1327 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1328 | ktime_t now = ktime_get(); |
1329 | ||
1330 | /* | |
1331 | * Emulate tick processing via per-CPU hrtimers: | |
1332 | */ | |
902a9f9c | 1333 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 | 1334 | ts->sched_timer.function = tick_sched_timer; |
79bf2bb3 | 1335 | |
0de7611a | 1336 | /* Get the next period (per-CPU) */ |
cc584b21 | 1337 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1338 | |
9c3f9e28 | 1339 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
1340 | if (sched_skew_tick) { |
1341 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
1342 | do_div(offset, num_possible_cpus()); | |
1343 | offset *= smp_processor_id(); | |
1344 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1345 | } | |
1346 | ||
afc08b15 | 1347 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
902a9f9c | 1348 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED_HARD); |
bc7a34b8 | 1349 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1350 | } |
3c4fbe5e | 1351 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1352 | |
3451d024 | 1353 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1354 | void tick_cancel_sched_timer(int cpu) |
1355 | { | |
1356 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1357 | ||
3c4fbe5e | 1358 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1359 | if (ts->sched_timer.base) |
1360 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1361 | # endif |
a7901766 | 1362 | |
4b0c0f29 | 1363 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1364 | } |
3c4fbe5e | 1365 | #endif |
79bf2bb3 TG |
1366 | |
1367 | /** | |
1368 | * Async notification about clocksource changes | |
1369 | */ | |
1370 | void tick_clock_notify(void) | |
1371 | { | |
1372 | int cpu; | |
1373 | ||
1374 | for_each_possible_cpu(cpu) | |
1375 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1376 | } | |
1377 | ||
1378 | /* | |
1379 | * Async notification about clock event changes | |
1380 | */ | |
1381 | void tick_oneshot_notify(void) | |
1382 | { | |
22127e93 | 1383 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1384 | |
1385 | set_bit(0, &ts->check_clocks); | |
1386 | } | |
1387 | ||
1388 | /** | |
1389 | * Check, if a change happened, which makes oneshot possible. | |
1390 | * | |
1391 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1392 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1393 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1394 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1395 | */ |
1396 | int tick_check_oneshot_change(int allow_nohz) | |
1397 | { | |
22127e93 | 1398 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1399 | |
1400 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1401 | return 0; | |
1402 | ||
1403 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1404 | return 0; | |
1405 | ||
cf4fc6cb | 1406 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1407 | return 0; |
1408 | ||
1409 | if (!allow_nohz) | |
1410 | return 1; | |
1411 | ||
1412 | tick_nohz_switch_to_nohz(); | |
1413 | return 0; | |
1414 | } |