]>
Commit | Line | Data |
---|---|---|
79bf2bb3 TG |
1 | /* |
2 | * linux/kernel/time/tick-sched.c | |
3 | * | |
4 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> | |
5 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar | |
6 | * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner | |
7 | * | |
8 | * No idle tick implementation for low and high resolution timers | |
9 | * | |
10 | * Started by: Thomas Gleixner and Ingo Molnar | |
11 | * | |
b10db7f0 | 12 | * Distribute under GPLv2. |
79bf2bb3 TG |
13 | */ |
14 | #include <linux/cpu.h> | |
15 | #include <linux/err.h> | |
16 | #include <linux/hrtimer.h> | |
17 | #include <linux/interrupt.h> | |
18 | #include <linux/kernel_stat.h> | |
19 | #include <linux/percpu.h> | |
38b8d208 | 20 | #include <linux/nmi.h> |
79bf2bb3 | 21 | #include <linux/profile.h> |
3f07c014 | 22 | #include <linux/sched/signal.h> |
e6017571 | 23 | #include <linux/sched/clock.h> |
03441a34 | 24 | #include <linux/sched/stat.h> |
370c9135 | 25 | #include <linux/sched/nohz.h> |
8083e4ad | 26 | #include <linux/module.h> |
00b42959 | 27 | #include <linux/irq_work.h> |
9014c45d | 28 | #include <linux/posix-timers.h> |
2e709338 | 29 | #include <linux/context_tracking.h> |
62cb1188 | 30 | #include <linux/mm.h> |
79bf2bb3 | 31 | |
9e203bcc DM |
32 | #include <asm/irq_regs.h> |
33 | ||
79bf2bb3 TG |
34 | #include "tick-internal.h" |
35 | ||
cb41a290 FW |
36 | #include <trace/events/timer.h> |
37 | ||
79bf2bb3 | 38 | /* |
0de7611a | 39 | * Per-CPU nohz control structure |
79bf2bb3 | 40 | */ |
c1797baf | 41 | static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); |
79bf2bb3 | 42 | |
289f480a IM |
43 | struct tick_sched *tick_get_tick_sched(int cpu) |
44 | { | |
45 | return &per_cpu(tick_cpu_sched, cpu); | |
46 | } | |
47 | ||
7809998a AB |
48 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
49 | /* | |
50 | * The time, when the last jiffy update happened. Protected by jiffies_lock. | |
51 | */ | |
52 | static ktime_t last_jiffies_update; | |
53 | ||
79bf2bb3 TG |
54 | /* |
55 | * Must be called with interrupts disabled ! | |
56 | */ | |
57 | static void tick_do_update_jiffies64(ktime_t now) | |
58 | { | |
59 | unsigned long ticks = 0; | |
60 | ktime_t delta; | |
61 | ||
7a14ce1d | 62 | /* |
d6ad4187 | 63 | * Do a quick check without holding jiffies_lock: |
7a14ce1d IM |
64 | */ |
65 | delta = ktime_sub(now, last_jiffies_update); | |
2456e855 | 66 | if (delta < tick_period) |
7a14ce1d IM |
67 | return; |
68 | ||
6168f8ed | 69 | /* Reevaluate with jiffies_lock held */ |
d6ad4187 | 70 | write_seqlock(&jiffies_lock); |
79bf2bb3 TG |
71 | |
72 | delta = ktime_sub(now, last_jiffies_update); | |
2456e855 | 73 | if (delta >= tick_period) { |
79bf2bb3 TG |
74 | |
75 | delta = ktime_sub(delta, tick_period); | |
76 | last_jiffies_update = ktime_add(last_jiffies_update, | |
77 | tick_period); | |
78 | ||
79 | /* Slow path for long timeouts */ | |
2456e855 | 80 | if (unlikely(delta >= tick_period)) { |
79bf2bb3 TG |
81 | s64 incr = ktime_to_ns(tick_period); |
82 | ||
83 | ticks = ktime_divns(delta, incr); | |
84 | ||
85 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
86 | incr * ticks); | |
87 | } | |
88 | do_timer(++ticks); | |
49d670fb TG |
89 | |
90 | /* Keep the tick_next_period variable up to date */ | |
91 | tick_next_period = ktime_add(last_jiffies_update, tick_period); | |
03e6bdc5 VK |
92 | } else { |
93 | write_sequnlock(&jiffies_lock); | |
94 | return; | |
79bf2bb3 | 95 | } |
d6ad4187 | 96 | write_sequnlock(&jiffies_lock); |
47a1b796 | 97 | update_wall_time(); |
79bf2bb3 TG |
98 | } |
99 | ||
100 | /* | |
101 | * Initialize and return retrieve the jiffies update. | |
102 | */ | |
103 | static ktime_t tick_init_jiffy_update(void) | |
104 | { | |
105 | ktime_t period; | |
106 | ||
d6ad4187 | 107 | write_seqlock(&jiffies_lock); |
79bf2bb3 | 108 | /* Did we start the jiffies update yet ? */ |
2456e855 | 109 | if (last_jiffies_update == 0) |
79bf2bb3 TG |
110 | last_jiffies_update = tick_next_period; |
111 | period = last_jiffies_update; | |
d6ad4187 | 112 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
113 | return period; |
114 | } | |
115 | ||
5bb96226 FW |
116 | |
117 | static void tick_sched_do_timer(ktime_t now) | |
118 | { | |
119 | int cpu = smp_processor_id(); | |
120 | ||
3451d024 | 121 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
122 | /* |
123 | * Check if the do_timer duty was dropped. We don't care about | |
0de7611a IM |
124 | * concurrency: This happens only when the CPU in charge went |
125 | * into a long sleep. If two CPUs happen to assign themselves to | |
5bb96226 | 126 | * this duty, then the jiffies update is still serialized by |
9c3f9e28 | 127 | * jiffies_lock. |
5bb96226 | 128 | */ |
a382bf93 | 129 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE) |
c5bfece2 | 130 | && !tick_nohz_full_cpu(cpu)) |
5bb96226 FW |
131 | tick_do_timer_cpu = cpu; |
132 | #endif | |
133 | ||
134 | /* Check, if the jiffies need an update */ | |
135 | if (tick_do_timer_cpu == cpu) | |
136 | tick_do_update_jiffies64(now); | |
137 | } | |
138 | ||
9e8f559b FW |
139 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
140 | { | |
3451d024 | 141 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
142 | /* |
143 | * When we are idle and the tick is stopped, we have to touch | |
144 | * the watchdog as we might not schedule for a really long | |
145 | * time. This happens on complete idle SMP systems while | |
146 | * waiting on the login prompt. We also increment the "start of | |
147 | * idle" jiffy stamp so the idle accounting adjustment we do | |
148 | * when we go busy again does not account too much ticks. | |
149 | */ | |
150 | if (ts->tick_stopped) { | |
03e0d461 | 151 | touch_softlockup_watchdog_sched(); |
9e8f559b FW |
152 | if (is_idle_task(current)) |
153 | ts->idle_jiffies++; | |
411fe24e FW |
154 | /* |
155 | * In case the current tick fired too early past its expected | |
156 | * expiration, make sure we don't bypass the next clock reprogramming | |
157 | * to the same deadline. | |
158 | */ | |
159 | ts->next_tick = 0; | |
9e8f559b | 160 | } |
94a57140 | 161 | #endif |
9e8f559b FW |
162 | update_process_times(user_mode(regs)); |
163 | profile_tick(CPU_PROFILING); | |
164 | } | |
7809998a | 165 | #endif |
9e8f559b | 166 | |
c5bfece2 | 167 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 168 | cpumask_var_t tick_nohz_full_mask; |
73867dcd | 169 | bool tick_nohz_full_running; |
f009a7a7 | 170 | static atomic_t tick_dep_mask; |
a831881b | 171 | |
f009a7a7 | 172 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 173 | { |
f009a7a7 FW |
174 | int val = atomic_read(dep); |
175 | ||
176 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 177 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 178 | return true; |
d027d45d FW |
179 | } |
180 | ||
f009a7a7 | 181 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 182 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 183 | return true; |
d027d45d FW |
184 | } |
185 | ||
f009a7a7 | 186 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 187 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 188 | return true; |
d027d45d FW |
189 | } |
190 | ||
f009a7a7 | 191 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 192 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
193 | return true; |
194 | } | |
195 | ||
196 | return false; | |
d027d45d FW |
197 | } |
198 | ||
57ccdf44 | 199 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d | 200 | { |
ebf3adba | 201 | lockdep_assert_irqs_disabled(); |
9014c45d | 202 | |
57ccdf44 WL |
203 | if (unlikely(!cpu_online(cpu))) |
204 | return false; | |
205 | ||
f009a7a7 | 206 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 207 | return false; |
d027d45d | 208 | |
f009a7a7 | 209 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 210 | return false; |
d027d45d | 211 | |
f009a7a7 | 212 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 213 | return false; |
d027d45d | 214 | |
f009a7a7 | 215 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 216 | return false; |
d027d45d | 217 | |
9014c45d FW |
218 | return true; |
219 | } | |
220 | ||
d027d45d | 221 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 222 | { |
73738a95 | 223 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
224 | } |
225 | ||
226 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = { | |
d027d45d | 227 | .func = nohz_full_kick_func, |
76c24fb0 FW |
228 | }; |
229 | ||
40bea039 FW |
230 | /* |
231 | * Kick this CPU if it's full dynticks in order to force it to | |
232 | * re-evaluate its dependency on the tick and restart it if necessary. | |
233 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
234 | * is NMI safe. | |
235 | */ | |
555e0c1e | 236 | static void tick_nohz_full_kick(void) |
40bea039 FW |
237 | { |
238 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
239 | return; | |
240 | ||
56e4dea8 | 241 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
242 | } |
243 | ||
76c24fb0 | 244 | /* |
3d36aebc | 245 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
246 | * re-evaluate its dependency on the tick and restart it if necessary. |
247 | */ | |
3d36aebc | 248 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 249 | { |
3d36aebc FW |
250 | if (!tick_nohz_full_cpu(cpu)) |
251 | return; | |
252 | ||
253 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
254 | } |
255 | ||
76c24fb0 FW |
256 | /* |
257 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
258 | * their dependency on the tick and restart it if necessary. | |
259 | */ | |
b7878300 | 260 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 261 | { |
8537bb95 FW |
262 | int cpu; |
263 | ||
73867dcd | 264 | if (!tick_nohz_full_running) |
76c24fb0 FW |
265 | return; |
266 | ||
267 | preempt_disable(); | |
8537bb95 FW |
268 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
269 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
270 | preempt_enable(); |
271 | } | |
272 | ||
f009a7a7 | 273 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
274 | enum tick_dep_bits bit) |
275 | { | |
f009a7a7 | 276 | int prev; |
d027d45d | 277 | |
a1cc5bcf | 278 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
279 | if (!prev) |
280 | tick_nohz_full_kick_all(); | |
281 | } | |
282 | ||
283 | /* | |
284 | * Set a global tick dependency. Used by perf events that rely on freq and | |
285 | * by unstable clock. | |
286 | */ | |
287 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
288 | { | |
289 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
290 | } | |
291 | ||
292 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
293 | { | |
f009a7a7 | 294 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
295 | } |
296 | ||
297 | /* | |
298 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
299 | * manage events throttling. | |
300 | */ | |
301 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
302 | { | |
f009a7a7 | 303 | int prev; |
d027d45d FW |
304 | struct tick_sched *ts; |
305 | ||
306 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
307 | ||
a1cc5bcf | 308 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
309 | if (!prev) { |
310 | preempt_disable(); | |
311 | /* Perf needs local kick that is NMI safe */ | |
312 | if (cpu == smp_processor_id()) { | |
313 | tick_nohz_full_kick(); | |
314 | } else { | |
315 | /* Remote irq work not NMI-safe */ | |
316 | if (!WARN_ON_ONCE(in_nmi())) | |
317 | tick_nohz_full_kick_cpu(cpu); | |
318 | } | |
319 | preempt_enable(); | |
320 | } | |
321 | } | |
322 | ||
323 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
324 | { | |
325 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
326 | ||
f009a7a7 | 327 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
328 | } |
329 | ||
330 | /* | |
331 | * Set a per-task tick dependency. Posix CPU timers need this in order to elapse | |
332 | * per task timers. | |
333 | */ | |
334 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
335 | { | |
336 | /* | |
337 | * We could optimize this with just kicking the target running the task | |
338 | * if that noise matters for nohz full users. | |
339 | */ | |
340 | tick_nohz_dep_set_all(&tsk->tick_dep_mask, bit); | |
341 | } | |
342 | ||
343 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
344 | { | |
f009a7a7 | 345 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d FW |
346 | } |
347 | ||
348 | /* | |
349 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
350 | * per process timers. | |
351 | */ | |
352 | void tick_nohz_dep_set_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
353 | { | |
354 | tick_nohz_dep_set_all(&sig->tick_dep_mask, bit); | |
355 | } | |
356 | ||
357 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
358 | { | |
f009a7a7 | 359 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
360 | } |
361 | ||
99e5ada9 FW |
362 | /* |
363 | * Re-evaluate the need for the tick as we switch the current task. | |
364 | * It might need the tick due to per task/process properties: | |
0de7611a | 365 | * perf events, posix CPU timers, ... |
99e5ada9 | 366 | */ |
de734f89 | 367 | void __tick_nohz_task_switch(void) |
99e5ada9 FW |
368 | { |
369 | unsigned long flags; | |
d027d45d | 370 | struct tick_sched *ts; |
99e5ada9 | 371 | |
99e5ada9 FW |
372 | local_irq_save(flags); |
373 | ||
6296ace4 LZ |
374 | if (!tick_nohz_full_cpu(smp_processor_id())) |
375 | goto out; | |
376 | ||
d027d45d | 377 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 378 | |
d027d45d | 379 | if (ts->tick_stopped) { |
f009a7a7 FW |
380 | if (atomic_read(¤t->tick_dep_mask) || |
381 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
382 | tick_nohz_full_kick(); |
383 | } | |
6296ace4 | 384 | out: |
99e5ada9 FW |
385 | local_irq_restore(flags); |
386 | } | |
387 | ||
6f1982fe FW |
388 | /* Get the boot-time nohz CPU list from the kernel parameters. */ |
389 | void __init tick_nohz_full_setup(cpumask_var_t cpumask) | |
a831881b | 390 | { |
73867dcd | 391 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
6f1982fe | 392 | cpumask_copy(tick_nohz_full_mask, cpumask); |
73867dcd | 393 | tick_nohz_full_running = true; |
a831881b | 394 | } |
a831881b | 395 | |
31eff243 | 396 | static int tick_nohz_cpu_down(unsigned int cpu) |
a382bf93 | 397 | { |
31eff243 SAS |
398 | /* |
399 | * The boot CPU handles housekeeping duty (unbound timers, | |
400 | * workqueues, timekeeping, ...) on behalf of full dynticks | |
401 | * CPUs. It must remain online when nohz full is enabled. | |
402 | */ | |
403 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) | |
404 | return -EBUSY; | |
405 | return 0; | |
a382bf93 FW |
406 | } |
407 | ||
d1e43fa5 | 408 | void __init tick_nohz_init(void) |
a831881b | 409 | { |
31eff243 | 410 | int cpu, ret; |
d1e43fa5 | 411 | |
a7c8655b PM |
412 | if (!tick_nohz_full_running) |
413 | return; | |
d1e43fa5 | 414 | |
9b01f5bf FW |
415 | /* |
416 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
417 | * locking contexts. But then we need irq work to raise its own | |
418 | * interrupts to avoid circular dependency on the tick | |
419 | */ | |
420 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 421 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf | 422 | cpumask_clear(tick_nohz_full_mask); |
9b01f5bf FW |
423 | tick_nohz_full_running = false; |
424 | return; | |
425 | } | |
426 | ||
4327b15f FW |
427 | cpu = smp_processor_id(); |
428 | ||
429 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { | |
a395d6a7 JP |
430 | pr_warn("NO_HZ: Clearing %d from nohz_full range for timekeeping\n", |
431 | cpu); | |
4327b15f FW |
432 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); |
433 | } | |
434 | ||
73867dcd | 435 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
436 | context_tracking_cpu_set(cpu); |
437 | ||
31eff243 SAS |
438 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
439 | "kernel/nohz:predown", NULL, | |
440 | tick_nohz_cpu_down); | |
441 | WARN_ON(ret < 0); | |
ffda22c1 TH |
442 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
443 | cpumask_pr_args(tick_nohz_full_mask)); | |
a831881b | 444 | } |
a831881b FW |
445 | #endif |
446 | ||
79bf2bb3 TG |
447 | /* |
448 | * NOHZ - aka dynamic tick functionality | |
449 | */ | |
3451d024 | 450 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
451 | /* |
452 | * NO HZ enabled ? | |
453 | */ | |
4cc7ecb7 | 454 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 455 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
456 | /* |
457 | * Enable / Disable tickless mode | |
458 | */ | |
459 | static int __init setup_tick_nohz(char *str) | |
460 | { | |
4cc7ecb7 | 461 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
462 | } |
463 | ||
464 | __setup("nohz=", setup_tick_nohz); | |
465 | ||
a3642983 | 466 | bool tick_nohz_tick_stopped(void) |
c1797baf TG |
467 | { |
468 | return __this_cpu_read(tick_cpu_sched.tick_stopped); | |
469 | } | |
470 | ||
22ab8bc0 FW |
471 | bool tick_nohz_tick_stopped_cpu(int cpu) |
472 | { | |
473 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
474 | ||
475 | return ts->tick_stopped; | |
476 | } | |
477 | ||
79bf2bb3 TG |
478 | /** |
479 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
480 | * | |
481 | * Called from interrupt entry when the CPU was idle | |
482 | * | |
483 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
484 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
485 | * value. We do this unconditionally on any CPU, as we don't know whether the |
486 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 487 | */ |
eed3b9cf | 488 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 489 | { |
79bf2bb3 | 490 | unsigned long flags; |
79bf2bb3 | 491 | |
e8fcaa5c | 492 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
493 | |
494 | local_irq_save(flags); | |
495 | tick_do_update_jiffies64(now); | |
496 | local_irq_restore(flags); | |
02ff3755 | 497 | |
03e0d461 | 498 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
499 | } |
500 | ||
595aac48 | 501 | /* |
0de7611a | 502 | * Updates the per-CPU time idle statistics counters |
595aac48 | 503 | */ |
8d63bf94 | 504 | static void |
8c215bd3 | 505 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 506 | { |
eed3b9cf | 507 | ktime_t delta; |
6378ddb5 | 508 | |
595aac48 AV |
509 | if (ts->idle_active) { |
510 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 511 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 512 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
513 | else |
514 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 515 | ts->idle_entrytime = now; |
595aac48 | 516 | } |
8d63bf94 | 517 | |
e0e37c20 | 518 | if (last_update_time) |
8d63bf94 AV |
519 | *last_update_time = ktime_to_us(now); |
520 | ||
595aac48 AV |
521 | } |
522 | ||
e8fcaa5c | 523 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 524 | { |
e8fcaa5c | 525 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 526 | ts->idle_active = 0; |
56c7426b | 527 | |
ac1e843f | 528 | sched_clock_idle_wakeup_event(); |
6378ddb5 VP |
529 | } |
530 | ||
e8fcaa5c | 531 | static ktime_t tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 532 | { |
430ee881 | 533 | ktime_t now = ktime_get(); |
595aac48 | 534 | |
6378ddb5 VP |
535 | ts->idle_entrytime = now; |
536 | ts->idle_active = 1; | |
56c7426b | 537 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
538 | return now; |
539 | } | |
540 | ||
b1f724c3 | 541 | /** |
0de7611a | 542 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 543 | * @cpu: CPU number to query |
09a1d34f MH |
544 | * @last_update_time: variable to store update time in. Do not update |
545 | * counters if NULL. | |
b1f724c3 | 546 | * |
6168f8ed | 547 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 548 | * CPU, in microseconds. |
b1f724c3 AV |
549 | * |
550 | * This time is measured via accounting rather than sampling, | |
551 | * and is as accurate as ktime_get() is. | |
552 | * | |
553 | * This function returns -1 if NOHZ is not enabled. | |
554 | */ | |
6378ddb5 VP |
555 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
556 | { | |
557 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 558 | ktime_t now, idle; |
6378ddb5 | 559 | |
d689fe22 | 560 | if (!tick_nohz_active) |
8083e4ad | 561 | return -1; |
562 | ||
09a1d34f MH |
563 | now = ktime_get(); |
564 | if (last_update_time) { | |
565 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
566 | idle = ts->idle_sleeptime; | |
567 | } else { | |
568 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
569 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
570 | ||
571 | idle = ktime_add(ts->idle_sleeptime, delta); | |
572 | } else { | |
573 | idle = ts->idle_sleeptime; | |
574 | } | |
575 | } | |
576 | ||
577 | return ktime_to_us(idle); | |
8083e4ad | 578 | |
6378ddb5 | 579 | } |
8083e4ad | 580 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 581 | |
6beea0cd | 582 | /** |
0de7611a | 583 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 584 | * @cpu: CPU number to query |
09a1d34f MH |
585 | * @last_update_time: variable to store update time in. Do not update |
586 | * counters if NULL. | |
0224cf4c | 587 | * |
6168f8ed | 588 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
589 | * CPU, in microseconds. |
590 | * | |
591 | * This time is measured via accounting rather than sampling, | |
592 | * and is as accurate as ktime_get() is. | |
593 | * | |
594 | * This function returns -1 if NOHZ is not enabled. | |
595 | */ | |
596 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
597 | { | |
598 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 599 | ktime_t now, iowait; |
0224cf4c | 600 | |
d689fe22 | 601 | if (!tick_nohz_active) |
0224cf4c AV |
602 | return -1; |
603 | ||
09a1d34f MH |
604 | now = ktime_get(); |
605 | if (last_update_time) { | |
606 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
607 | iowait = ts->iowait_sleeptime; | |
608 | } else { | |
609 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
610 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 611 | |
09a1d34f MH |
612 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
613 | } else { | |
614 | iowait = ts->iowait_sleeptime; | |
615 | } | |
616 | } | |
0224cf4c | 617 | |
09a1d34f | 618 | return ktime_to_us(iowait); |
0224cf4c AV |
619 | } |
620 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
621 | ||
0ff53d09 TG |
622 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
623 | { | |
624 | hrtimer_cancel(&ts->sched_timer); | |
625 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
626 | ||
627 | /* Forward the time to expire in the future */ | |
628 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
629 | ||
630 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) | |
631 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); | |
632 | else | |
633 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
411fe24e FW |
634 | |
635 | /* | |
636 | * Reset to make sure next tick stop doesn't get fooled by past | |
637 | * cached clock deadline. | |
638 | */ | |
639 | ts->next_tick = 0; | |
0ff53d09 TG |
640 | } |
641 | ||
5d62c183 TG |
642 | static inline bool local_timer_softirq_pending(void) |
643 | { | |
644 | return local_softirq_pending() & TIMER_SOFTIRQ; | |
645 | } | |
646 | ||
84bf1bcc FW |
647 | static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts, |
648 | ktime_t now, int cpu) | |
79bf2bb3 | 649 | { |
22127e93 | 650 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
c1ad348b TG |
651 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
652 | unsigned long seq, basejiff; | |
653 | ktime_t tick; | |
855a0fc3 | 654 | |
79bf2bb3 TG |
655 | /* Read jiffies and the time when jiffies were updated last */ |
656 | do { | |
d6ad4187 | 657 | seq = read_seqbegin(&jiffies_lock); |
2456e855 | 658 | basemono = last_jiffies_update; |
c1ad348b | 659 | basejiff = jiffies; |
d6ad4187 | 660 | } while (read_seqretry(&jiffies_lock, seq)); |
c1ad348b | 661 | ts->last_jiffies = basejiff; |
79bf2bb3 | 662 | |
5d62c183 TG |
663 | /* |
664 | * Keep the periodic tick, when RCU, architecture or irq_work | |
665 | * requests it. | |
666 | * Aside of that check whether the local timer softirq is | |
667 | * pending. If so its a bad idea to call get_next_timer_interrupt() | |
668 | * because there is an already expired timer, so it will request | |
669 | * immeditate expiry, which rearms the hardware timer with a | |
670 | * minimal delta which brings us back to this place | |
671 | * immediately. Lather, rinse and repeat... | |
672 | */ | |
673 | if (rcu_needs_cpu(basemono, &next_rcu) || arch_needs_cpu() || | |
674 | irq_work_needs_cpu() || local_timer_softirq_pending()) { | |
c1ad348b | 675 | next_tick = basemono + TICK_NSEC; |
3c5d92a0 | 676 | } else { |
c1ad348b TG |
677 | /* |
678 | * Get the next pending timer. If high resolution | |
679 | * timers are enabled this only takes the timer wheel | |
680 | * timers into account. If high resolution timers are | |
681 | * disabled this also looks at the next expiring | |
682 | * hrtimer. | |
683 | */ | |
684 | next_tmr = get_next_timer_interrupt(basejiff, basemono); | |
685 | ts->next_timer = next_tmr; | |
686 | /* Take the next rcu event into account */ | |
687 | next_tick = next_rcu < next_tmr ? next_rcu : next_tmr; | |
3c5d92a0 | 688 | } |
47aa8b6c | 689 | |
c1ad348b TG |
690 | /* |
691 | * If the tick is due in the next period, keep it ticking or | |
82bbe34b | 692 | * force prod the timer. |
c1ad348b TG |
693 | */ |
694 | delta = next_tick - basemono; | |
695 | if (delta <= (u64)TICK_NSEC) { | |
a683f390 TG |
696 | /* |
697 | * Tell the timer code that the base is not idle, i.e. undo | |
698 | * the effect of get_next_timer_interrupt(): | |
699 | */ | |
700 | timer_clear_idle(); | |
82bbe34b PZ |
701 | /* |
702 | * We've not stopped the tick yet, and there's a timer in the | |
703 | * next period, so no point in stopping it either, bail. | |
704 | */ | |
f99973e1 FW |
705 | if (!ts->tick_stopped) { |
706 | tick = 0; | |
157d29e1 TG |
707 | goto out; |
708 | } | |
709 | } | |
710 | ||
79bf2bb3 | 711 | /* |
0de7611a IM |
712 | * If this CPU is the one which updates jiffies, then give up |
713 | * the assignment and let it be taken by the CPU which runs | |
714 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
715 | * don't drop this here the jiffies might be stale and |
716 | * do_timer() never invoked. Keep track of the fact that it | |
0de7611a | 717 | * was the one which had the do_timer() duty last. If this CPU |
157d29e1 | 718 | * is the one which had the do_timer() duty last, we limit the |
6168f8ed | 719 | * sleep time to the timekeeping max_deferment value. |
c1ad348b | 720 | * Otherwise we can sleep as long as we want. |
79bf2bb3 | 721 | */ |
c1ad348b | 722 | delta = timekeeping_max_deferment(); |
157d29e1 TG |
723 | if (cpu == tick_do_timer_cpu) { |
724 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
725 | ts->do_timer_last = 1; | |
726 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
c1ad348b | 727 | delta = KTIME_MAX; |
157d29e1 TG |
728 | ts->do_timer_last = 0; |
729 | } else if (!ts->do_timer_last) { | |
c1ad348b | 730 | delta = KTIME_MAX; |
157d29e1 | 731 | } |
27185016 | 732 | |
c1ad348b TG |
733 | /* Calculate the next expiry time */ |
734 | if (delta < (KTIME_MAX - basemono)) | |
735 | expires = basemono + delta; | |
157d29e1 | 736 | else |
c1ad348b TG |
737 | expires = KTIME_MAX; |
738 | ||
739 | expires = min_t(u64, expires, next_tick); | |
2456e855 | 740 | tick = expires; |
00147449 | 741 | |
157d29e1 | 742 | /* Skip reprogram of event if its not changed */ |
411fe24e FW |
743 | if (ts->tick_stopped && (expires == ts->next_tick)) { |
744 | /* Sanity check: make sure clockevent is actually programmed */ | |
d4af6d93 | 745 | if (tick == KTIME_MAX || ts->next_tick == hrtimer_get_expires(&ts->sched_timer)) |
ce6cf9a1 | 746 | goto out; |
411fe24e FW |
747 | |
748 | WARN_ON_ONCE(1); | |
749 | printk_once("basemono: %llu ts->next_tick: %llu dev->next_event: %llu timer->active: %d timer->expires: %llu\n", | |
750 | basemono, ts->next_tick, dev->next_event, | |
751 | hrtimer_active(&ts->sched_timer), hrtimer_get_expires(&ts->sched_timer)); | |
ce6cf9a1 | 752 | } |
84bf1bcc | 753 | |
157d29e1 TG |
754 | /* |
755 | * nohz_stop_sched_tick can be called several times before | |
756 | * the nohz_restart_sched_tick is called. This happens when | |
757 | * interrupts arrive which do not cause a reschedule. In the | |
758 | * first call we save the current tick time, so we can restart | |
759 | * the scheduler tick in nohz_restart_sched_tick. | |
760 | */ | |
761 | if (!ts->tick_stopped) { | |
3c85d6db | 762 | calc_load_nohz_start(); |
1f41906a | 763 | cpu_load_update_nohz_start(); |
62cb1188 | 764 | quiet_vmstat(); |
d3ed7824 | 765 | |
157d29e1 TG |
766 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
767 | ts->tick_stopped = 1; | |
e6e6cc22 | 768 | trace_tick_stop(1, TICK_DEP_MASK_NONE); |
157d29e1 | 769 | } |
eaad084b | 770 | |
411fe24e FW |
771 | ts->next_tick = tick; |
772 | ||
157d29e1 | 773 | /* |
c1ad348b TG |
774 | * If the expiration time == KTIME_MAX, then we simply stop |
775 | * the tick timer. | |
157d29e1 | 776 | */ |
c1ad348b | 777 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
778 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
779 | hrtimer_cancel(&ts->sched_timer); | |
780 | goto out; | |
79bf2bb3 | 781 | } |
0ff53d09 | 782 | |
d4af6d93 FW |
783 | hrtimer_set_expires(&ts->sched_timer, tick); |
784 | ||
157d29e1 | 785 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
d4af6d93 | 786 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); |
157d29e1 | 787 | else |
c1ad348b | 788 | tick_program_event(tick, 1); |
79bf2bb3 | 789 | out: |
411fe24e FW |
790 | /* |
791 | * Update the estimated sleep length until the next timer | |
792 | * (not only the tick). | |
793 | */ | |
4f86d3a8 | 794 | ts->sleep_length = ktime_sub(dev->next_event, now); |
c1ad348b | 795 | return tick; |
280f0677 FW |
796 | } |
797 | ||
1f41906a | 798 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
799 | { |
800 | /* Update jiffies first */ | |
801 | tick_do_update_jiffies64(now); | |
1f41906a | 802 | cpu_load_update_nohz_stop(); |
59d2c7ca | 803 | |
a683f390 TG |
804 | /* |
805 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
806 | * the clock forward checks in the enqueue path: | |
807 | */ | |
808 | timer_clear_idle(); | |
809 | ||
3c85d6db | 810 | calc_load_nohz_stop(); |
03e0d461 | 811 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
812 | /* |
813 | * Cancel the scheduled timer and restore the tick | |
814 | */ | |
815 | ts->tick_stopped = 0; | |
816 | ts->idle_exittime = now; | |
817 | ||
818 | tick_nohz_restart(ts, now); | |
819 | } | |
73738a95 FW |
820 | |
821 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
5811d996 FW |
822 | { |
823 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 824 | int cpu = smp_processor_id(); |
5811d996 | 825 | |
59449359 | 826 | if (!tick_nohz_full_cpu(cpu)) |
e9a2eb40 | 827 | return; |
5811d996 | 828 | |
e9a2eb40 AS |
829 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
830 | return; | |
5811d996 | 831 | |
57ccdf44 | 832 | if (can_stop_full_tick(cpu, ts)) |
73738a95 FW |
833 | tick_nohz_stop_sched_tick(ts, ktime_get(), cpu); |
834 | else if (ts->tick_stopped) | |
1f41906a | 835 | tick_nohz_restart_sched_tick(ts, ktime_get()); |
5811d996 FW |
836 | #endif |
837 | } | |
838 | ||
5b39939a FW |
839 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
840 | { | |
841 | /* | |
0de7611a | 842 | * If this CPU is offline and it is the one which updates |
5b39939a | 843 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 844 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
845 | * this here the jiffies might be stale and do_timer() never |
846 | * invoked. | |
847 | */ | |
848 | if (unlikely(!cpu_online(cpu))) { | |
849 | if (cpu == tick_do_timer_cpu) | |
850 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
411fe24e FW |
851 | /* |
852 | * Make sure the CPU doesn't get fooled by obsolete tick | |
853 | * deadline if it comes back online later. | |
854 | */ | |
855 | ts->next_tick = 0; | |
f7ea0fd6 | 856 | return false; |
5b39939a FW |
857 | } |
858 | ||
0e576acb | 859 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) { |
2456e855 | 860 | ts->sleep_length = NSEC_PER_SEC / HZ; |
5b39939a | 861 | return false; |
0e576acb | 862 | } |
5b39939a FW |
863 | |
864 | if (need_resched()) | |
865 | return false; | |
866 | ||
867 | if (unlikely(local_softirq_pending() && cpu_online(cpu))) { | |
868 | static int ratelimit; | |
869 | ||
803b0eba PM |
870 | if (ratelimit < 10 && |
871 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
cfea7d7e RV |
872 | pr_warn("NOHZ: local_softirq_pending %02x\n", |
873 | (unsigned int) local_softirq_pending()); | |
5b39939a FW |
874 | ratelimit++; |
875 | } | |
876 | return false; | |
877 | } | |
878 | ||
460775df | 879 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
880 | /* |
881 | * Keep the tick alive to guarantee timekeeping progression | |
882 | * if there are full dynticks CPUs around | |
883 | */ | |
884 | if (tick_do_timer_cpu == cpu) | |
885 | return false; | |
886 | /* | |
887 | * Boot safety: make sure the timekeeping duty has been | |
888 | * assigned before entering dyntick-idle mode, | |
889 | */ | |
890 | if (tick_do_timer_cpu == TICK_DO_TIMER_NONE) | |
891 | return false; | |
892 | } | |
893 | ||
5b39939a FW |
894 | return true; |
895 | } | |
896 | ||
19f5f736 FW |
897 | static void __tick_nohz_idle_enter(struct tick_sched *ts) |
898 | { | |
84bf1bcc | 899 | ktime_t now, expires; |
5b39939a | 900 | int cpu = smp_processor_id(); |
19f5f736 | 901 | |
08d07259 WL |
902 | now = tick_nohz_start_idle(ts); |
903 | ||
5b39939a FW |
904 | if (can_stop_idle_tick(cpu, ts)) { |
905 | int was_stopped = ts->tick_stopped; | |
906 | ||
907 | ts->idle_calls++; | |
84bf1bcc FW |
908 | |
909 | expires = tick_nohz_stop_sched_tick(ts, now, cpu); | |
2456e855 | 910 | if (expires > 0LL) { |
84bf1bcc FW |
911 | ts->idle_sleeps++; |
912 | ts->idle_expires = expires; | |
913 | } | |
5b39939a | 914 | |
a0db971e | 915 | if (!was_stopped && ts->tick_stopped) { |
5b39939a | 916 | ts->idle_jiffies = ts->last_jiffies; |
a0db971e FW |
917 | nohz_balance_enter_idle(cpu); |
918 | } | |
5b39939a | 919 | } |
280f0677 FW |
920 | } |
921 | ||
922 | /** | |
923 | * tick_nohz_idle_enter - stop the idle tick from the idle task | |
924 | * | |
925 | * When the next event is more than a tick into the future, stop the idle tick | |
926 | * Called when we start the idle loop. | |
2bbb6817 | 927 | * |
1268fbc7 | 928 | * The arch is responsible of calling: |
2bbb6817 FW |
929 | * |
930 | * - rcu_idle_enter() after its last use of RCU before the CPU is put | |
931 | * to sleep. | |
932 | * - rcu_idle_exit() before the first use of RCU after the CPU is woken up. | |
280f0677 | 933 | */ |
1268fbc7 | 934 | void tick_nohz_idle_enter(void) |
280f0677 FW |
935 | { |
936 | struct tick_sched *ts; | |
937 | ||
ebf3adba | 938 | lockdep_assert_irqs_enabled(); |
0db49b72 | 939 | |
1268fbc7 FW |
940 | local_irq_disable(); |
941 | ||
22127e93 | 942 | ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 943 | ts->inidle = 1; |
19f5f736 | 944 | __tick_nohz_idle_enter(ts); |
1268fbc7 FW |
945 | |
946 | local_irq_enable(); | |
280f0677 FW |
947 | } |
948 | ||
949 | /** | |
950 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
951 | * | |
952 | * When an interrupt fires while we are idle and it doesn't cause | |
953 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
954 | * an RCU callback, etc... | |
955 | * So we need to re-calculate and reprogram the next tick event. | |
956 | */ | |
957 | void tick_nohz_irq_exit(void) | |
958 | { | |
22127e93 | 959 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 960 | |
14851912 | 961 | if (ts->inidle) |
5811d996 | 962 | __tick_nohz_idle_enter(ts); |
14851912 | 963 | else |
73738a95 | 964 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
965 | } |
966 | ||
4f86d3a8 LB |
967 | /** |
968 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
969 | * | |
970 | * Called from power state control code with interrupts disabled | |
971 | */ | |
972 | ktime_t tick_nohz_get_sleep_length(void) | |
973 | { | |
22127e93 | 974 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
4f86d3a8 LB |
975 | |
976 | return ts->sleep_length; | |
977 | } | |
978 | ||
466a2b42 JF |
979 | /** |
980 | * tick_nohz_get_idle_calls_cpu - return the current idle calls counter value | |
981 | * for a particular CPU. | |
982 | * | |
983 | * Called from the schedutil frequency scaling governor in scheduler context. | |
984 | */ | |
985 | unsigned long tick_nohz_get_idle_calls_cpu(int cpu) | |
986 | { | |
987 | struct tick_sched *ts = tick_get_tick_sched(cpu); | |
988 | ||
989 | return ts->idle_calls; | |
990 | } | |
991 | ||
b7eaf1aa RW |
992 | /** |
993 | * tick_nohz_get_idle_calls - return the current idle calls counter value | |
994 | * | |
995 | * Called from the schedutil frequency scaling governor in scheduler context. | |
996 | */ | |
997 | unsigned long tick_nohz_get_idle_calls(void) | |
998 | { | |
999 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1000 | ||
1001 | return ts->idle_calls; | |
1002 | } | |
1003 | ||
2ac0d98f FW |
1004 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) |
1005 | { | |
3f4724ea | 1006 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 1007 | unsigned long ticks; |
3f4724ea | 1008 | |
55dbdcfa | 1009 | if (vtime_accounting_cpu_enabled()) |
3f4724ea | 1010 | return; |
79bf2bb3 TG |
1011 | /* |
1012 | * We stopped the tick in idle. Update process times would miss the | |
1013 | * time we slept as update_process_times does only a 1 tick | |
1014 | * accounting. Enforce that this is accounted to idle ! | |
1015 | */ | |
1016 | ticks = jiffies - ts->idle_jiffies; | |
1017 | /* | |
1018 | * We might be one off. Do not randomly account a huge number of ticks! | |
1019 | */ | |
79741dd3 MS |
1020 | if (ticks && ticks < LONG_MAX) |
1021 | account_idle_ticks(ticks); | |
1022 | #endif | |
19f5f736 FW |
1023 | } |
1024 | ||
79bf2bb3 | 1025 | /** |
280f0677 | 1026 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1027 | * |
1028 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1029 | * This also exit the RCU extended quiescent state. The CPU |
1030 | * can use RCU again after this function is called. | |
79bf2bb3 | 1031 | */ |
280f0677 | 1032 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1033 | { |
4a32fea9 | 1034 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
6378ddb5 | 1035 | ktime_t now; |
79bf2bb3 | 1036 | |
6378ddb5 | 1037 | local_irq_disable(); |
2bbb6817 | 1038 | |
15f827be FW |
1039 | WARN_ON_ONCE(!ts->inidle); |
1040 | ||
1041 | ts->inidle = 0; | |
1042 | ||
1043 | if (ts->idle_active || ts->tick_stopped) | |
eed3b9cf MS |
1044 | now = ktime_get(); |
1045 | ||
1046 | if (ts->idle_active) | |
e8fcaa5c | 1047 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1048 | |
2ac0d98f | 1049 | if (ts->tick_stopped) { |
1f41906a | 1050 | tick_nohz_restart_sched_tick(ts, now); |
2ac0d98f | 1051 | tick_nohz_account_idle_ticks(ts); |
6378ddb5 | 1052 | } |
79bf2bb3 | 1053 | |
79bf2bb3 TG |
1054 | local_irq_enable(); |
1055 | } | |
1056 | ||
79bf2bb3 TG |
1057 | /* |
1058 | * The nohz low res interrupt handler | |
1059 | */ | |
1060 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1061 | { | |
22127e93 | 1062 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1063 | struct pt_regs *regs = get_irq_regs(); |
1064 | ktime_t now = ktime_get(); | |
1065 | ||
2456e855 | 1066 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1067 | |
5bb96226 | 1068 | tick_sched_do_timer(now); |
9e8f559b | 1069 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1070 | |
b5e995e6 VK |
1071 | /* No need to reprogram if we are running tickless */ |
1072 | if (unlikely(ts->tick_stopped)) | |
1073 | return; | |
1074 | ||
0ff53d09 TG |
1075 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1076 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
79bf2bb3 TG |
1077 | } |
1078 | ||
bc7a34b8 TG |
1079 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1080 | { | |
1081 | if (!tick_nohz_enabled) | |
1082 | return; | |
1083 | ts->nohz_mode = mode; | |
1084 | /* One update is enough */ | |
1085 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
ae67bada | 1086 | timers_update_nohz(); |
bc7a34b8 TG |
1087 | } |
1088 | ||
79bf2bb3 TG |
1089 | /** |
1090 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1091 | */ | |
1092 | static void tick_nohz_switch_to_nohz(void) | |
1093 | { | |
22127e93 | 1094 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1095 | ktime_t next; |
1096 | ||
27630532 | 1097 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1098 | return; |
1099 | ||
6b442bc8 | 1100 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1101 | return; |
6b442bc8 | 1102 | |
79bf2bb3 TG |
1103 | /* |
1104 | * Recycle the hrtimer in ts, so we can share the | |
1105 | * hrtimer_forward with the highres code. | |
1106 | */ | |
1107 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1108 | /* Get the next period */ | |
1109 | next = tick_init_jiffy_update(); | |
1110 | ||
0ff53d09 | 1111 | hrtimer_set_expires(&ts->sched_timer, next); |
1ca8ec53 WL |
1112 | hrtimer_forward_now(&ts->sched_timer, tick_period); |
1113 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
bc7a34b8 | 1114 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1115 | } |
1116 | ||
5acac1be | 1117 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1118 | { |
4a32fea9 | 1119 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1120 | ktime_t now; |
1121 | ||
1122 | if (!ts->idle_active && !ts->tick_stopped) | |
1123 | return; | |
1124 | now = ktime_get(); | |
1125 | if (ts->idle_active) | |
e8fcaa5c | 1126 | tick_nohz_stop_idle(ts, now); |
ff006732 | 1127 | if (ts->tick_stopped) |
eed3b9cf | 1128 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1129 | } |
1130 | ||
79bf2bb3 TG |
1131 | #else |
1132 | ||
1133 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1134 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1135 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1136 | |
3451d024 | 1137 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1138 | |
719254fa TG |
1139 | /* |
1140 | * Called from irq_enter to notify about the possible interruption of idle() | |
1141 | */ | |
5acac1be | 1142 | void tick_irq_enter(void) |
719254fa | 1143 | { |
e8fcaa5c | 1144 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1145 | tick_nohz_irq_enter(); |
719254fa TG |
1146 | } |
1147 | ||
79bf2bb3 TG |
1148 | /* |
1149 | * High resolution timer specific code | |
1150 | */ | |
1151 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1152 | /* | |
4c9dc641 | 1153 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1154 | * Called with interrupts disabled. |
79bf2bb3 TG |
1155 | */ |
1156 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1157 | { | |
1158 | struct tick_sched *ts = | |
1159 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1160 | struct pt_regs *regs = get_irq_regs(); |
1161 | ktime_t now = ktime_get(); | |
d3ed7824 | 1162 | |
5bb96226 | 1163 | tick_sched_do_timer(now); |
79bf2bb3 TG |
1164 | |
1165 | /* | |
1166 | * Do not call, when we are not in irq context and have | |
1167 | * no valid regs pointer | |
1168 | */ | |
9e8f559b FW |
1169 | if (regs) |
1170 | tick_sched_handle(ts, regs); | |
7c259045 FW |
1171 | else |
1172 | ts->next_tick = 0; | |
79bf2bb3 | 1173 | |
2a16fc93 VK |
1174 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1175 | if (unlikely(ts->tick_stopped)) | |
1176 | return HRTIMER_NORESTART; | |
1177 | ||
79bf2bb3 TG |
1178 | hrtimer_forward(timer, now, tick_period); |
1179 | ||
1180 | return HRTIMER_RESTART; | |
1181 | } | |
1182 | ||
5307c955 MG |
1183 | static int sched_skew_tick; |
1184 | ||
62cf20b3 TG |
1185 | static int __init skew_tick(char *str) |
1186 | { | |
1187 | get_option(&str, &sched_skew_tick); | |
1188 | ||
1189 | return 0; | |
1190 | } | |
1191 | early_param("skew_tick", skew_tick); | |
1192 | ||
79bf2bb3 TG |
1193 | /** |
1194 | * tick_setup_sched_timer - setup the tick emulation timer | |
1195 | */ | |
1196 | void tick_setup_sched_timer(void) | |
1197 | { | |
22127e93 | 1198 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1199 | ktime_t now = ktime_get(); |
1200 | ||
1201 | /* | |
1202 | * Emulate tick processing via per-CPU hrtimers: | |
1203 | */ | |
1204 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1205 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 1206 | |
0de7611a | 1207 | /* Get the next period (per-CPU) */ |
cc584b21 | 1208 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1209 | |
9c3f9e28 | 1210 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
1211 | if (sched_skew_tick) { |
1212 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
1213 | do_div(offset, num_possible_cpus()); | |
1214 | offset *= smp_processor_id(); | |
1215 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1216 | } | |
1217 | ||
afc08b15 TG |
1218 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1219 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); | |
bc7a34b8 | 1220 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1221 | } |
3c4fbe5e | 1222 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1223 | |
3451d024 | 1224 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1225 | void tick_cancel_sched_timer(int cpu) |
1226 | { | |
1227 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1228 | ||
3c4fbe5e | 1229 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1230 | if (ts->sched_timer.base) |
1231 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1232 | # endif |
a7901766 | 1233 | |
4b0c0f29 | 1234 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1235 | } |
3c4fbe5e | 1236 | #endif |
79bf2bb3 TG |
1237 | |
1238 | /** | |
1239 | * Async notification about clocksource changes | |
1240 | */ | |
1241 | void tick_clock_notify(void) | |
1242 | { | |
1243 | int cpu; | |
1244 | ||
1245 | for_each_possible_cpu(cpu) | |
1246 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * Async notification about clock event changes | |
1251 | */ | |
1252 | void tick_oneshot_notify(void) | |
1253 | { | |
22127e93 | 1254 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1255 | |
1256 | set_bit(0, &ts->check_clocks); | |
1257 | } | |
1258 | ||
1259 | /** | |
1260 | * Check, if a change happened, which makes oneshot possible. | |
1261 | * | |
1262 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1263 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1264 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1265 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1266 | */ |
1267 | int tick_check_oneshot_change(int allow_nohz) | |
1268 | { | |
22127e93 | 1269 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1270 | |
1271 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1272 | return 0; | |
1273 | ||
1274 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1275 | return 0; | |
1276 | ||
cf4fc6cb | 1277 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1278 | return 0; |
1279 | ||
1280 | if (!allow_nohz) | |
1281 | return 1; | |
1282 | ||
1283 | tick_nohz_switch_to_nohz(); | |
1284 | return 0; | |
1285 | } |