]>
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> |
79bf2bb3 | 30 | |
9e203bcc DM |
31 | #include <asm/irq_regs.h> |
32 | ||
79bf2bb3 TG |
33 | #include "tick-internal.h" |
34 | ||
cb41a290 FW |
35 | #include <trace/events/timer.h> |
36 | ||
79bf2bb3 | 37 | /* |
0de7611a | 38 | * Per-CPU nohz control structure |
79bf2bb3 | 39 | */ |
c1797baf | 40 | static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); |
79bf2bb3 | 41 | |
289f480a IM |
42 | struct tick_sched *tick_get_tick_sched(int cpu) |
43 | { | |
44 | return &per_cpu(tick_cpu_sched, cpu); | |
45 | } | |
46 | ||
7809998a AB |
47 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
48 | /* | |
49 | * The time, when the last jiffy update happened. Protected by jiffies_lock. | |
50 | */ | |
51 | static ktime_t last_jiffies_update; | |
52 | ||
79bf2bb3 TG |
53 | /* |
54 | * Must be called with interrupts disabled ! | |
55 | */ | |
56 | static void tick_do_update_jiffies64(ktime_t now) | |
57 | { | |
58 | unsigned long ticks = 0; | |
59 | ktime_t delta; | |
60 | ||
7a14ce1d | 61 | /* |
d6ad4187 | 62 | * Do a quick check without holding jiffies_lock: |
7a14ce1d IM |
63 | */ |
64 | delta = ktime_sub(now, last_jiffies_update); | |
2456e855 | 65 | if (delta < tick_period) |
7a14ce1d IM |
66 | return; |
67 | ||
6168f8ed | 68 | /* Reevaluate with jiffies_lock held */ |
d6ad4187 | 69 | write_seqlock(&jiffies_lock); |
79bf2bb3 TG |
70 | |
71 | delta = ktime_sub(now, last_jiffies_update); | |
2456e855 | 72 | if (delta >= tick_period) { |
79bf2bb3 TG |
73 | |
74 | delta = ktime_sub(delta, tick_period); | |
75 | last_jiffies_update = ktime_add(last_jiffies_update, | |
76 | tick_period); | |
77 | ||
78 | /* Slow path for long timeouts */ | |
2456e855 | 79 | if (unlikely(delta >= tick_period)) { |
79bf2bb3 TG |
80 | s64 incr = ktime_to_ns(tick_period); |
81 | ||
82 | ticks = ktime_divns(delta, incr); | |
83 | ||
84 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
85 | incr * ticks); | |
86 | } | |
87 | do_timer(++ticks); | |
49d670fb TG |
88 | |
89 | /* Keep the tick_next_period variable up to date */ | |
90 | tick_next_period = ktime_add(last_jiffies_update, tick_period); | |
03e6bdc5 VK |
91 | } else { |
92 | write_sequnlock(&jiffies_lock); | |
93 | return; | |
79bf2bb3 | 94 | } |
d6ad4187 | 95 | write_sequnlock(&jiffies_lock); |
47a1b796 | 96 | update_wall_time(); |
79bf2bb3 TG |
97 | } |
98 | ||
99 | /* | |
100 | * Initialize and return retrieve the jiffies update. | |
101 | */ | |
102 | static ktime_t tick_init_jiffy_update(void) | |
103 | { | |
104 | ktime_t period; | |
105 | ||
d6ad4187 | 106 | write_seqlock(&jiffies_lock); |
79bf2bb3 | 107 | /* Did we start the jiffies update yet ? */ |
2456e855 | 108 | if (last_jiffies_update == 0) |
79bf2bb3 TG |
109 | last_jiffies_update = tick_next_period; |
110 | period = last_jiffies_update; | |
d6ad4187 | 111 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
112 | return period; |
113 | } | |
114 | ||
5bb96226 FW |
115 | |
116 | static void tick_sched_do_timer(ktime_t now) | |
117 | { | |
118 | int cpu = smp_processor_id(); | |
119 | ||
3451d024 | 120 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
121 | /* |
122 | * Check if the do_timer duty was dropped. We don't care about | |
0de7611a IM |
123 | * concurrency: This happens only when the CPU in charge went |
124 | * into a long sleep. If two CPUs happen to assign themselves to | |
5bb96226 | 125 | * this duty, then the jiffies update is still serialized by |
9c3f9e28 | 126 | * jiffies_lock. |
5bb96226 | 127 | */ |
a382bf93 | 128 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE) |
c5bfece2 | 129 | && !tick_nohz_full_cpu(cpu)) |
5bb96226 FW |
130 | tick_do_timer_cpu = cpu; |
131 | #endif | |
132 | ||
133 | /* Check, if the jiffies need an update */ | |
134 | if (tick_do_timer_cpu == cpu) | |
135 | tick_do_update_jiffies64(now); | |
136 | } | |
137 | ||
9e8f559b FW |
138 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
139 | { | |
3451d024 | 140 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
141 | /* |
142 | * When we are idle and the tick is stopped, we have to touch | |
143 | * the watchdog as we might not schedule for a really long | |
144 | * time. This happens on complete idle SMP systems while | |
145 | * waiting on the login prompt. We also increment the "start of | |
146 | * idle" jiffy stamp so the idle accounting adjustment we do | |
147 | * when we go busy again does not account too much ticks. | |
148 | */ | |
149 | if (ts->tick_stopped) { | |
03e0d461 | 150 | touch_softlockup_watchdog_sched(); |
9e8f559b FW |
151 | if (is_idle_task(current)) |
152 | ts->idle_jiffies++; | |
153 | } | |
94a57140 | 154 | #endif |
9e8f559b FW |
155 | update_process_times(user_mode(regs)); |
156 | profile_tick(CPU_PROFILING); | |
157 | } | |
7809998a | 158 | #endif |
9e8f559b | 159 | |
c5bfece2 | 160 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 161 | cpumask_var_t tick_nohz_full_mask; |
c0f489d2 | 162 | cpumask_var_t housekeeping_mask; |
73867dcd | 163 | bool tick_nohz_full_running; |
f009a7a7 | 164 | static atomic_t tick_dep_mask; |
a831881b | 165 | |
f009a7a7 | 166 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 167 | { |
f009a7a7 FW |
168 | int val = atomic_read(dep); |
169 | ||
170 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 171 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 172 | return true; |
d027d45d FW |
173 | } |
174 | ||
f009a7a7 | 175 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 176 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 177 | return true; |
d027d45d FW |
178 | } |
179 | ||
f009a7a7 | 180 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 181 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 182 | return true; |
d027d45d FW |
183 | } |
184 | ||
f009a7a7 | 185 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 186 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
187 | return true; |
188 | } | |
189 | ||
190 | return false; | |
d027d45d FW |
191 | } |
192 | ||
57ccdf44 | 193 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d FW |
194 | { |
195 | WARN_ON_ONCE(!irqs_disabled()); | |
196 | ||
57ccdf44 WL |
197 | if (unlikely(!cpu_online(cpu))) |
198 | return false; | |
199 | ||
f009a7a7 | 200 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 201 | return false; |
d027d45d | 202 | |
f009a7a7 | 203 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 204 | return false; |
d027d45d | 205 | |
f009a7a7 | 206 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 207 | return false; |
d027d45d | 208 | |
f009a7a7 | 209 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 210 | return false; |
d027d45d | 211 | |
9014c45d FW |
212 | return true; |
213 | } | |
214 | ||
d027d45d | 215 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 216 | { |
73738a95 | 217 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
218 | } |
219 | ||
220 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = { | |
d027d45d | 221 | .func = nohz_full_kick_func, |
76c24fb0 FW |
222 | }; |
223 | ||
40bea039 FW |
224 | /* |
225 | * Kick this CPU if it's full dynticks in order to force it to | |
226 | * re-evaluate its dependency on the tick and restart it if necessary. | |
227 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
228 | * is NMI safe. | |
229 | */ | |
555e0c1e | 230 | static void tick_nohz_full_kick(void) |
40bea039 FW |
231 | { |
232 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
233 | return; | |
234 | ||
56e4dea8 | 235 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
236 | } |
237 | ||
76c24fb0 | 238 | /* |
3d36aebc | 239 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
240 | * re-evaluate its dependency on the tick and restart it if necessary. |
241 | */ | |
3d36aebc | 242 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 243 | { |
3d36aebc FW |
244 | if (!tick_nohz_full_cpu(cpu)) |
245 | return; | |
246 | ||
247 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
248 | } |
249 | ||
76c24fb0 FW |
250 | /* |
251 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
252 | * their dependency on the tick and restart it if necessary. | |
253 | */ | |
b7878300 | 254 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 255 | { |
8537bb95 FW |
256 | int cpu; |
257 | ||
73867dcd | 258 | if (!tick_nohz_full_running) |
76c24fb0 FW |
259 | return; |
260 | ||
261 | preempt_disable(); | |
8537bb95 FW |
262 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
263 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
264 | preempt_enable(); |
265 | } | |
266 | ||
f009a7a7 | 267 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
268 | enum tick_dep_bits bit) |
269 | { | |
f009a7a7 | 270 | int prev; |
d027d45d | 271 | |
a1cc5bcf | 272 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
273 | if (!prev) |
274 | tick_nohz_full_kick_all(); | |
275 | } | |
276 | ||
277 | /* | |
278 | * Set a global tick dependency. Used by perf events that rely on freq and | |
279 | * by unstable clock. | |
280 | */ | |
281 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
282 | { | |
283 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
284 | } | |
285 | ||
286 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
287 | { | |
f009a7a7 | 288 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
289 | } |
290 | ||
291 | /* | |
292 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
293 | * manage events throttling. | |
294 | */ | |
295 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
296 | { | |
f009a7a7 | 297 | int prev; |
d027d45d FW |
298 | struct tick_sched *ts; |
299 | ||
300 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
301 | ||
a1cc5bcf | 302 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
303 | if (!prev) { |
304 | preempt_disable(); | |
305 | /* Perf needs local kick that is NMI safe */ | |
306 | if (cpu == smp_processor_id()) { | |
307 | tick_nohz_full_kick(); | |
308 | } else { | |
309 | /* Remote irq work not NMI-safe */ | |
310 | if (!WARN_ON_ONCE(in_nmi())) | |
311 | tick_nohz_full_kick_cpu(cpu); | |
312 | } | |
313 | preempt_enable(); | |
314 | } | |
315 | } | |
316 | ||
317 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
318 | { | |
319 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
320 | ||
f009a7a7 | 321 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
322 | } |
323 | ||
324 | /* | |
325 | * Set a per-task tick dependency. Posix CPU timers need this in order to elapse | |
326 | * per task timers. | |
327 | */ | |
328 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
329 | { | |
330 | /* | |
331 | * We could optimize this with just kicking the target running the task | |
332 | * if that noise matters for nohz full users. | |
333 | */ | |
334 | tick_nohz_dep_set_all(&tsk->tick_dep_mask, bit); | |
335 | } | |
336 | ||
337 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
338 | { | |
f009a7a7 | 339 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d FW |
340 | } |
341 | ||
342 | /* | |
343 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
344 | * per process timers. | |
345 | */ | |
346 | void tick_nohz_dep_set_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
347 | { | |
348 | tick_nohz_dep_set_all(&sig->tick_dep_mask, bit); | |
349 | } | |
350 | ||
351 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
352 | { | |
f009a7a7 | 353 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
354 | } |
355 | ||
99e5ada9 FW |
356 | /* |
357 | * Re-evaluate the need for the tick as we switch the current task. | |
358 | * It might need the tick due to per task/process properties: | |
0de7611a | 359 | * perf events, posix CPU timers, ... |
99e5ada9 | 360 | */ |
de734f89 | 361 | void __tick_nohz_task_switch(void) |
99e5ada9 FW |
362 | { |
363 | unsigned long flags; | |
d027d45d | 364 | struct tick_sched *ts; |
99e5ada9 | 365 | |
99e5ada9 FW |
366 | local_irq_save(flags); |
367 | ||
6296ace4 LZ |
368 | if (!tick_nohz_full_cpu(smp_processor_id())) |
369 | goto out; | |
370 | ||
d027d45d | 371 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 372 | |
d027d45d | 373 | if (ts->tick_stopped) { |
f009a7a7 FW |
374 | if (atomic_read(¤t->tick_dep_mask) || |
375 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
376 | tick_nohz_full_kick(); |
377 | } | |
6296ace4 | 378 | out: |
99e5ada9 FW |
379 | local_irq_restore(flags); |
380 | } | |
381 | ||
a831881b | 382 | /* Parse the boot-time nohz CPU list from the kernel parameters. */ |
c5bfece2 | 383 | static int __init tick_nohz_full_setup(char *str) |
a831881b | 384 | { |
73867dcd FW |
385 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
386 | if (cpulist_parse(str, tick_nohz_full_mask) < 0) { | |
a395d6a7 | 387 | pr_warn("NO_HZ: Incorrect nohz_full cpumask\n"); |
4327b15f | 388 | free_bootmem_cpumask_var(tick_nohz_full_mask); |
0453b435 FW |
389 | return 1; |
390 | } | |
73867dcd | 391 | tick_nohz_full_running = true; |
0453b435 | 392 | |
a831881b FW |
393 | return 1; |
394 | } | |
c5bfece2 | 395 | __setup("nohz_full=", tick_nohz_full_setup); |
a831881b | 396 | |
31eff243 | 397 | static int tick_nohz_cpu_down(unsigned int cpu) |
a382bf93 | 398 | { |
31eff243 SAS |
399 | /* |
400 | * The boot CPU handles housekeeping duty (unbound timers, | |
401 | * workqueues, timekeeping, ...) on behalf of full dynticks | |
402 | * CPUs. It must remain online when nohz full is enabled. | |
403 | */ | |
404 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) | |
405 | return -EBUSY; | |
406 | return 0; | |
a382bf93 FW |
407 | } |
408 | ||
f98823ac FW |
409 | static int tick_nohz_init_all(void) |
410 | { | |
411 | int err = -1; | |
412 | ||
413 | #ifdef CONFIG_NO_HZ_FULL_ALL | |
73867dcd | 414 | if (!alloc_cpumask_var(&tick_nohz_full_mask, GFP_KERNEL)) { |
4327b15f | 415 | WARN(1, "NO_HZ: Can't allocate full dynticks cpumask\n"); |
c0f489d2 PM |
416 | return err; |
417 | } | |
f98823ac | 418 | err = 0; |
73867dcd | 419 | cpumask_setall(tick_nohz_full_mask); |
73867dcd | 420 | tick_nohz_full_running = true; |
f98823ac FW |
421 | #endif |
422 | return err; | |
423 | } | |
424 | ||
d1e43fa5 | 425 | void __init tick_nohz_init(void) |
a831881b | 426 | { |
31eff243 | 427 | int cpu, ret; |
d1e43fa5 | 428 | |
73867dcd | 429 | if (!tick_nohz_full_running) { |
f98823ac FW |
430 | if (tick_nohz_init_all() < 0) |
431 | return; | |
432 | } | |
d1e43fa5 | 433 | |
4327b15f FW |
434 | if (!alloc_cpumask_var(&housekeeping_mask, GFP_KERNEL)) { |
435 | WARN(1, "NO_HZ: Can't allocate not-full dynticks cpumask\n"); | |
436 | cpumask_clear(tick_nohz_full_mask); | |
437 | tick_nohz_full_running = false; | |
438 | return; | |
439 | } | |
440 | ||
9b01f5bf FW |
441 | /* |
442 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
443 | * locking contexts. But then we need irq work to raise its own | |
444 | * interrupts to avoid circular dependency on the tick | |
445 | */ | |
446 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 447 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf FW |
448 | cpumask_clear(tick_nohz_full_mask); |
449 | cpumask_copy(housekeeping_mask, cpu_possible_mask); | |
450 | tick_nohz_full_running = false; | |
451 | return; | |
452 | } | |
453 | ||
4327b15f FW |
454 | cpu = smp_processor_id(); |
455 | ||
456 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { | |
a395d6a7 JP |
457 | pr_warn("NO_HZ: Clearing %d from nohz_full range for timekeeping\n", |
458 | cpu); | |
4327b15f FW |
459 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); |
460 | } | |
461 | ||
462 | cpumask_andnot(housekeeping_mask, | |
463 | cpu_possible_mask, tick_nohz_full_mask); | |
464 | ||
73867dcd | 465 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
466 | context_tracking_cpu_set(cpu); |
467 | ||
31eff243 SAS |
468 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
469 | "kernel/nohz:predown", NULL, | |
470 | tick_nohz_cpu_down); | |
471 | WARN_ON(ret < 0); | |
ffda22c1 TH |
472 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
473 | cpumask_pr_args(tick_nohz_full_mask)); | |
7c8bb6cb FW |
474 | |
475 | /* | |
476 | * We need at least one CPU to handle housekeeping work such | |
477 | * as timekeeping, unbound timers, workqueues, ... | |
478 | */ | |
479 | WARN_ON_ONCE(cpumask_empty(housekeeping_mask)); | |
a831881b | 480 | } |
a831881b FW |
481 | #endif |
482 | ||
79bf2bb3 TG |
483 | /* |
484 | * NOHZ - aka dynamic tick functionality | |
485 | */ | |
3451d024 | 486 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
487 | /* |
488 | * NO HZ enabled ? | |
489 | */ | |
4cc7ecb7 | 490 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 491 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
492 | /* |
493 | * Enable / Disable tickless mode | |
494 | */ | |
495 | static int __init setup_tick_nohz(char *str) | |
496 | { | |
4cc7ecb7 | 497 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
498 | } |
499 | ||
500 | __setup("nohz=", setup_tick_nohz); | |
501 | ||
c1797baf TG |
502 | int tick_nohz_tick_stopped(void) |
503 | { | |
504 | return __this_cpu_read(tick_cpu_sched.tick_stopped); | |
505 | } | |
506 | ||
79bf2bb3 TG |
507 | /** |
508 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
509 | * | |
510 | * Called from interrupt entry when the CPU was idle | |
511 | * | |
512 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
513 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
514 | * value. We do this unconditionally on any CPU, as we don't know whether the |
515 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 516 | */ |
eed3b9cf | 517 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 518 | { |
79bf2bb3 | 519 | unsigned long flags; |
79bf2bb3 | 520 | |
e8fcaa5c | 521 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
522 | |
523 | local_irq_save(flags); | |
524 | tick_do_update_jiffies64(now); | |
525 | local_irq_restore(flags); | |
02ff3755 | 526 | |
03e0d461 | 527 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
528 | } |
529 | ||
595aac48 | 530 | /* |
0de7611a | 531 | * Updates the per-CPU time idle statistics counters |
595aac48 | 532 | */ |
8d63bf94 | 533 | static void |
8c215bd3 | 534 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 535 | { |
eed3b9cf | 536 | ktime_t delta; |
6378ddb5 | 537 | |
595aac48 AV |
538 | if (ts->idle_active) { |
539 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 540 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 541 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
542 | else |
543 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 544 | ts->idle_entrytime = now; |
595aac48 | 545 | } |
8d63bf94 | 546 | |
e0e37c20 | 547 | if (last_update_time) |
8d63bf94 AV |
548 | *last_update_time = ktime_to_us(now); |
549 | ||
595aac48 AV |
550 | } |
551 | ||
e8fcaa5c | 552 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 553 | { |
e8fcaa5c | 554 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 555 | ts->idle_active = 0; |
56c7426b | 556 | |
eed3b9cf | 557 | sched_clock_idle_wakeup_event(0); |
6378ddb5 VP |
558 | } |
559 | ||
e8fcaa5c | 560 | static ktime_t tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 561 | { |
430ee881 | 562 | ktime_t now = ktime_get(); |
595aac48 | 563 | |
6378ddb5 VP |
564 | ts->idle_entrytime = now; |
565 | ts->idle_active = 1; | |
56c7426b | 566 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
567 | return now; |
568 | } | |
569 | ||
b1f724c3 | 570 | /** |
0de7611a | 571 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 572 | * @cpu: CPU number to query |
09a1d34f MH |
573 | * @last_update_time: variable to store update time in. Do not update |
574 | * counters if NULL. | |
b1f724c3 | 575 | * |
6168f8ed | 576 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 577 | * CPU, in microseconds. |
b1f724c3 AV |
578 | * |
579 | * This time is measured via accounting rather than sampling, | |
580 | * and is as accurate as ktime_get() is. | |
581 | * | |
582 | * This function returns -1 if NOHZ is not enabled. | |
583 | */ | |
6378ddb5 VP |
584 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
585 | { | |
586 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 587 | ktime_t now, idle; |
6378ddb5 | 588 | |
d689fe22 | 589 | if (!tick_nohz_active) |
8083e4ad | 590 | return -1; |
591 | ||
09a1d34f MH |
592 | now = ktime_get(); |
593 | if (last_update_time) { | |
594 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
595 | idle = ts->idle_sleeptime; | |
596 | } else { | |
597 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
598 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
599 | ||
600 | idle = ktime_add(ts->idle_sleeptime, delta); | |
601 | } else { | |
602 | idle = ts->idle_sleeptime; | |
603 | } | |
604 | } | |
605 | ||
606 | return ktime_to_us(idle); | |
8083e4ad | 607 | |
6378ddb5 | 608 | } |
8083e4ad | 609 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 610 | |
6beea0cd | 611 | /** |
0de7611a | 612 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 613 | * @cpu: CPU number to query |
09a1d34f MH |
614 | * @last_update_time: variable to store update time in. Do not update |
615 | * counters if NULL. | |
0224cf4c | 616 | * |
6168f8ed | 617 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
618 | * CPU, in microseconds. |
619 | * | |
620 | * This time is measured via accounting rather than sampling, | |
621 | * and is as accurate as ktime_get() is. | |
622 | * | |
623 | * This function returns -1 if NOHZ is not enabled. | |
624 | */ | |
625 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
626 | { | |
627 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 628 | ktime_t now, iowait; |
0224cf4c | 629 | |
d689fe22 | 630 | if (!tick_nohz_active) |
0224cf4c AV |
631 | return -1; |
632 | ||
09a1d34f MH |
633 | now = ktime_get(); |
634 | if (last_update_time) { | |
635 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
636 | iowait = ts->iowait_sleeptime; | |
637 | } else { | |
638 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
639 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 640 | |
09a1d34f MH |
641 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
642 | } else { | |
643 | iowait = ts->iowait_sleeptime; | |
644 | } | |
645 | } | |
0224cf4c | 646 | |
09a1d34f | 647 | return ktime_to_us(iowait); |
0224cf4c AV |
648 | } |
649 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
650 | ||
0ff53d09 TG |
651 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
652 | { | |
653 | hrtimer_cancel(&ts->sched_timer); | |
654 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
655 | ||
656 | /* Forward the time to expire in the future */ | |
657 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
658 | ||
659 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) | |
660 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); | |
661 | else | |
662 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
663 | } | |
664 | ||
84bf1bcc FW |
665 | static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts, |
666 | ktime_t now, int cpu) | |
79bf2bb3 | 667 | { |
22127e93 | 668 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
c1ad348b TG |
669 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
670 | unsigned long seq, basejiff; | |
671 | ktime_t tick; | |
855a0fc3 | 672 | |
79bf2bb3 TG |
673 | /* Read jiffies and the time when jiffies were updated last */ |
674 | do { | |
d6ad4187 | 675 | seq = read_seqbegin(&jiffies_lock); |
2456e855 | 676 | basemono = last_jiffies_update; |
c1ad348b | 677 | basejiff = jiffies; |
d6ad4187 | 678 | } while (read_seqretry(&jiffies_lock, seq)); |
c1ad348b | 679 | ts->last_jiffies = basejiff; |
79bf2bb3 | 680 | |
c1ad348b | 681 | if (rcu_needs_cpu(basemono, &next_rcu) || |
fe0f4976 | 682 | arch_needs_cpu() || irq_work_needs_cpu()) { |
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) { | |
2456e855 | 704 | tick = 0; |
a683f390 TG |
705 | |
706 | /* | |
707 | * Tell the timer code that the base is not idle, i.e. undo | |
708 | * the effect of get_next_timer_interrupt(): | |
709 | */ | |
710 | timer_clear_idle(); | |
82bbe34b PZ |
711 | /* |
712 | * We've not stopped the tick yet, and there's a timer in the | |
713 | * next period, so no point in stopping it either, bail. | |
714 | */ | |
157d29e1 TG |
715 | if (!ts->tick_stopped) |
716 | goto out; | |
82bbe34b PZ |
717 | |
718 | /* | |
719 | * If, OTOH, we did stop it, but there's a pending (expired) | |
720 | * timer reprogram the timer hardware to fire now. | |
721 | * | |
722 | * We will not restart the tick proper, just prod the timer | |
723 | * hardware into firing an interrupt to process the pending | |
724 | * timers. Just like tick_irq_exit() will not restart the tick | |
725 | * for 'normal' interrupts. | |
726 | * | |
727 | * Only once we exit the idle loop will we re-enable the tick, | |
728 | * see tick_nohz_idle_exit(). | |
729 | */ | |
c1ad348b | 730 | if (delta == 0) { |
157d29e1 TG |
731 | tick_nohz_restart(ts, now); |
732 | goto out; | |
733 | } | |
734 | } | |
735 | ||
79bf2bb3 | 736 | /* |
0de7611a IM |
737 | * If this CPU is the one which updates jiffies, then give up |
738 | * the assignment and let it be taken by the CPU which runs | |
739 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
740 | * don't drop this here the jiffies might be stale and |
741 | * do_timer() never invoked. Keep track of the fact that it | |
0de7611a | 742 | * was the one which had the do_timer() duty last. If this CPU |
157d29e1 | 743 | * is the one which had the do_timer() duty last, we limit the |
6168f8ed | 744 | * sleep time to the timekeeping max_deferment value. |
c1ad348b | 745 | * Otherwise we can sleep as long as we want. |
79bf2bb3 | 746 | */ |
c1ad348b | 747 | delta = timekeeping_max_deferment(); |
157d29e1 TG |
748 | if (cpu == tick_do_timer_cpu) { |
749 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
750 | ts->do_timer_last = 1; | |
751 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
c1ad348b | 752 | delta = KTIME_MAX; |
157d29e1 TG |
753 | ts->do_timer_last = 0; |
754 | } else if (!ts->do_timer_last) { | |
c1ad348b | 755 | delta = KTIME_MAX; |
157d29e1 | 756 | } |
27185016 | 757 | |
265f22a9 | 758 | #ifdef CONFIG_NO_HZ_FULL |
c1ad348b | 759 | /* Limit the tick delta to the maximum scheduler deferment */ |
157d29e1 | 760 | if (!ts->inidle) |
c1ad348b | 761 | delta = min(delta, scheduler_tick_max_deferment()); |
265f22a9 FW |
762 | #endif |
763 | ||
c1ad348b TG |
764 | /* Calculate the next expiry time */ |
765 | if (delta < (KTIME_MAX - basemono)) | |
766 | expires = basemono + delta; | |
157d29e1 | 767 | else |
c1ad348b TG |
768 | expires = KTIME_MAX; |
769 | ||
770 | expires = min_t(u64, expires, next_tick); | |
2456e855 | 771 | tick = expires; |
00147449 | 772 | |
157d29e1 | 773 | /* Skip reprogram of event if its not changed */ |
558e8e27 | 774 | if (ts->tick_stopped && (expires == dev->next_event)) |
157d29e1 | 775 | goto out; |
84bf1bcc | 776 | |
157d29e1 TG |
777 | /* |
778 | * nohz_stop_sched_tick can be called several times before | |
779 | * the nohz_restart_sched_tick is called. This happens when | |
780 | * interrupts arrive which do not cause a reschedule. In the | |
781 | * first call we save the current tick time, so we can restart | |
782 | * the scheduler tick in nohz_restart_sched_tick. | |
783 | */ | |
784 | if (!ts->tick_stopped) { | |
785 | nohz_balance_enter_idle(cpu); | |
786 | calc_load_enter_idle(); | |
1f41906a | 787 | cpu_load_update_nohz_start(); |
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 | |
157d29e1 | 794 | /* |
c1ad348b TG |
795 | * If the expiration time == KTIME_MAX, then we simply stop |
796 | * the tick timer. | |
157d29e1 | 797 | */ |
c1ad348b | 798 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
799 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
800 | hrtimer_cancel(&ts->sched_timer); | |
801 | goto out; | |
79bf2bb3 | 802 | } |
0ff53d09 | 803 | |
157d29e1 | 804 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
c1ad348b | 805 | hrtimer_start(&ts->sched_timer, tick, HRTIMER_MODE_ABS_PINNED); |
157d29e1 | 806 | else |
c1ad348b | 807 | tick_program_event(tick, 1); |
79bf2bb3 | 808 | out: |
558e8e27 | 809 | /* Update the estimated sleep length */ |
4f86d3a8 | 810 | ts->sleep_length = ktime_sub(dev->next_event, now); |
c1ad348b | 811 | return tick; |
280f0677 FW |
812 | } |
813 | ||
1f41906a | 814 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
815 | { |
816 | /* Update jiffies first */ | |
817 | tick_do_update_jiffies64(now); | |
1f41906a | 818 | cpu_load_update_nohz_stop(); |
59d2c7ca | 819 | |
a683f390 TG |
820 | /* |
821 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
822 | * the clock forward checks in the enqueue path: | |
823 | */ | |
824 | timer_clear_idle(); | |
825 | ||
59d2c7ca | 826 | calc_load_exit_idle(); |
03e0d461 | 827 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
828 | /* |
829 | * Cancel the scheduled timer and restore the tick | |
830 | */ | |
831 | ts->tick_stopped = 0; | |
832 | ts->idle_exittime = now; | |
833 | ||
834 | tick_nohz_restart(ts, now); | |
835 | } | |
73738a95 FW |
836 | |
837 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
5811d996 FW |
838 | { |
839 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 840 | int cpu = smp_processor_id(); |
5811d996 | 841 | |
59449359 | 842 | if (!tick_nohz_full_cpu(cpu)) |
e9a2eb40 | 843 | return; |
5811d996 | 844 | |
e9a2eb40 AS |
845 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
846 | return; | |
5811d996 | 847 | |
57ccdf44 | 848 | if (can_stop_full_tick(cpu, ts)) |
73738a95 FW |
849 | tick_nohz_stop_sched_tick(ts, ktime_get(), cpu); |
850 | else if (ts->tick_stopped) | |
1f41906a | 851 | tick_nohz_restart_sched_tick(ts, ktime_get()); |
5811d996 FW |
852 | #endif |
853 | } | |
854 | ||
5b39939a FW |
855 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
856 | { | |
857 | /* | |
0de7611a | 858 | * If this CPU is offline and it is the one which updates |
5b39939a | 859 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 860 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
861 | * this here the jiffies might be stale and do_timer() never |
862 | * invoked. | |
863 | */ | |
864 | if (unlikely(!cpu_online(cpu))) { | |
865 | if (cpu == tick_do_timer_cpu) | |
866 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
f7ea0fd6 | 867 | return false; |
5b39939a FW |
868 | } |
869 | ||
0e576acb | 870 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) { |
2456e855 | 871 | ts->sleep_length = NSEC_PER_SEC / HZ; |
5b39939a | 872 | return false; |
0e576acb | 873 | } |
5b39939a FW |
874 | |
875 | if (need_resched()) | |
876 | return false; | |
877 | ||
878 | if (unlikely(local_softirq_pending() && cpu_online(cpu))) { | |
879 | static int ratelimit; | |
880 | ||
803b0eba PM |
881 | if (ratelimit < 10 && |
882 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
cfea7d7e RV |
883 | pr_warn("NOHZ: local_softirq_pending %02x\n", |
884 | (unsigned int) local_softirq_pending()); | |
5b39939a FW |
885 | ratelimit++; |
886 | } | |
887 | return false; | |
888 | } | |
889 | ||
460775df | 890 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
891 | /* |
892 | * Keep the tick alive to guarantee timekeeping progression | |
893 | * if there are full dynticks CPUs around | |
894 | */ | |
895 | if (tick_do_timer_cpu == cpu) | |
896 | return false; | |
897 | /* | |
898 | * Boot safety: make sure the timekeeping duty has been | |
899 | * assigned before entering dyntick-idle mode, | |
900 | */ | |
901 | if (tick_do_timer_cpu == TICK_DO_TIMER_NONE) | |
902 | return false; | |
903 | } | |
904 | ||
5b39939a FW |
905 | return true; |
906 | } | |
907 | ||
19f5f736 FW |
908 | static void __tick_nohz_idle_enter(struct tick_sched *ts) |
909 | { | |
84bf1bcc | 910 | ktime_t now, expires; |
5b39939a | 911 | int cpu = smp_processor_id(); |
19f5f736 | 912 | |
08d07259 WL |
913 | now = tick_nohz_start_idle(ts); |
914 | ||
5b39939a FW |
915 | if (can_stop_idle_tick(cpu, ts)) { |
916 | int was_stopped = ts->tick_stopped; | |
917 | ||
918 | ts->idle_calls++; | |
84bf1bcc FW |
919 | |
920 | expires = tick_nohz_stop_sched_tick(ts, now, cpu); | |
2456e855 | 921 | if (expires > 0LL) { |
84bf1bcc FW |
922 | ts->idle_sleeps++; |
923 | ts->idle_expires = expires; | |
924 | } | |
5b39939a FW |
925 | |
926 | if (!was_stopped && ts->tick_stopped) | |
927 | ts->idle_jiffies = ts->last_jiffies; | |
928 | } | |
280f0677 FW |
929 | } |
930 | ||
931 | /** | |
932 | * tick_nohz_idle_enter - stop the idle tick from the idle task | |
933 | * | |
934 | * When the next event is more than a tick into the future, stop the idle tick | |
935 | * Called when we start the idle loop. | |
2bbb6817 | 936 | * |
1268fbc7 | 937 | * The arch is responsible of calling: |
2bbb6817 FW |
938 | * |
939 | * - rcu_idle_enter() after its last use of RCU before the CPU is put | |
940 | * to sleep. | |
941 | * - rcu_idle_exit() before the first use of RCU after the CPU is woken up. | |
280f0677 | 942 | */ |
1268fbc7 | 943 | void tick_nohz_idle_enter(void) |
280f0677 FW |
944 | { |
945 | struct tick_sched *ts; | |
946 | ||
1268fbc7 FW |
947 | WARN_ON_ONCE(irqs_disabled()); |
948 | ||
0db49b72 | 949 | /* |
0de7611a IM |
950 | * Update the idle state in the scheduler domain hierarchy |
951 | * when tick_nohz_stop_sched_tick() is called from the idle loop. | |
952 | * State will be updated to busy during the first busy tick after | |
953 | * exiting idle. | |
954 | */ | |
0db49b72 LT |
955 | set_cpu_sd_state_idle(); |
956 | ||
1268fbc7 FW |
957 | local_irq_disable(); |
958 | ||
22127e93 | 959 | ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 960 | ts->inidle = 1; |
19f5f736 | 961 | __tick_nohz_idle_enter(ts); |
1268fbc7 FW |
962 | |
963 | local_irq_enable(); | |
280f0677 FW |
964 | } |
965 | ||
966 | /** | |
967 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
968 | * | |
969 | * When an interrupt fires while we are idle and it doesn't cause | |
970 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
971 | * an RCU callback, etc... | |
972 | * So we need to re-calculate and reprogram the next tick event. | |
973 | */ | |
974 | void tick_nohz_irq_exit(void) | |
975 | { | |
22127e93 | 976 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 977 | |
14851912 | 978 | if (ts->inidle) |
5811d996 | 979 | __tick_nohz_idle_enter(ts); |
14851912 | 980 | else |
73738a95 | 981 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
982 | } |
983 | ||
4f86d3a8 LB |
984 | /** |
985 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
986 | * | |
987 | * Called from power state control code with interrupts disabled | |
988 | */ | |
989 | ktime_t tick_nohz_get_sleep_length(void) | |
990 | { | |
22127e93 | 991 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
4f86d3a8 LB |
992 | |
993 | return ts->sleep_length; | |
994 | } | |
995 | ||
b7eaf1aa RW |
996 | /** |
997 | * tick_nohz_get_idle_calls - return the current idle calls counter value | |
998 | * | |
999 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1000 | */ | |
1001 | unsigned long tick_nohz_get_idle_calls(void) | |
1002 | { | |
1003 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1004 | ||
1005 | return ts->idle_calls; | |
1006 | } | |
1007 | ||
2ac0d98f FW |
1008 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) |
1009 | { | |
3f4724ea | 1010 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 1011 | unsigned long ticks; |
3f4724ea | 1012 | |
55dbdcfa | 1013 | if (vtime_accounting_cpu_enabled()) |
3f4724ea | 1014 | return; |
79bf2bb3 TG |
1015 | /* |
1016 | * We stopped the tick in idle. Update process times would miss the | |
1017 | * time we slept as update_process_times does only a 1 tick | |
1018 | * accounting. Enforce that this is accounted to idle ! | |
1019 | */ | |
1020 | ticks = jiffies - ts->idle_jiffies; | |
1021 | /* | |
1022 | * We might be one off. Do not randomly account a huge number of ticks! | |
1023 | */ | |
79741dd3 MS |
1024 | if (ticks && ticks < LONG_MAX) |
1025 | account_idle_ticks(ticks); | |
1026 | #endif | |
19f5f736 FW |
1027 | } |
1028 | ||
79bf2bb3 | 1029 | /** |
280f0677 | 1030 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1031 | * |
1032 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1033 | * This also exit the RCU extended quiescent state. The CPU |
1034 | * can use RCU again after this function is called. | |
79bf2bb3 | 1035 | */ |
280f0677 | 1036 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1037 | { |
4a32fea9 | 1038 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
6378ddb5 | 1039 | ktime_t now; |
79bf2bb3 | 1040 | |
6378ddb5 | 1041 | local_irq_disable(); |
2bbb6817 | 1042 | |
15f827be FW |
1043 | WARN_ON_ONCE(!ts->inidle); |
1044 | ||
1045 | ts->inidle = 0; | |
1046 | ||
1047 | if (ts->idle_active || ts->tick_stopped) | |
eed3b9cf MS |
1048 | now = ktime_get(); |
1049 | ||
1050 | if (ts->idle_active) | |
e8fcaa5c | 1051 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1052 | |
2ac0d98f | 1053 | if (ts->tick_stopped) { |
1f41906a | 1054 | tick_nohz_restart_sched_tick(ts, now); |
2ac0d98f | 1055 | tick_nohz_account_idle_ticks(ts); |
6378ddb5 | 1056 | } |
79bf2bb3 | 1057 | |
79bf2bb3 TG |
1058 | local_irq_enable(); |
1059 | } | |
1060 | ||
79bf2bb3 TG |
1061 | /* |
1062 | * The nohz low res interrupt handler | |
1063 | */ | |
1064 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1065 | { | |
22127e93 | 1066 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1067 | struct pt_regs *regs = get_irq_regs(); |
1068 | ktime_t now = ktime_get(); | |
1069 | ||
2456e855 | 1070 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1071 | |
5bb96226 | 1072 | tick_sched_do_timer(now); |
9e8f559b | 1073 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1074 | |
b5e995e6 VK |
1075 | /* No need to reprogram if we are running tickless */ |
1076 | if (unlikely(ts->tick_stopped)) | |
1077 | return; | |
1078 | ||
0ff53d09 TG |
1079 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1080 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
79bf2bb3 TG |
1081 | } |
1082 | ||
bc7a34b8 TG |
1083 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1084 | { | |
1085 | if (!tick_nohz_enabled) | |
1086 | return; | |
1087 | ts->nohz_mode = mode; | |
1088 | /* One update is enough */ | |
1089 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
683be13a | 1090 | timers_update_migration(true); |
bc7a34b8 TG |
1091 | } |
1092 | ||
79bf2bb3 TG |
1093 | /** |
1094 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1095 | */ | |
1096 | static void tick_nohz_switch_to_nohz(void) | |
1097 | { | |
22127e93 | 1098 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1099 | ktime_t next; |
1100 | ||
27630532 | 1101 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1102 | return; |
1103 | ||
6b442bc8 | 1104 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1105 | return; |
6b442bc8 | 1106 | |
79bf2bb3 TG |
1107 | /* |
1108 | * Recycle the hrtimer in ts, so we can share the | |
1109 | * hrtimer_forward with the highres code. | |
1110 | */ | |
1111 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1112 | /* Get the next period */ | |
1113 | next = tick_init_jiffy_update(); | |
1114 | ||
0ff53d09 | 1115 | hrtimer_set_expires(&ts->sched_timer, next); |
1ca8ec53 WL |
1116 | hrtimer_forward_now(&ts->sched_timer, tick_period); |
1117 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
bc7a34b8 | 1118 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1119 | } |
1120 | ||
5acac1be | 1121 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1122 | { |
4a32fea9 | 1123 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1124 | ktime_t now; |
1125 | ||
1126 | if (!ts->idle_active && !ts->tick_stopped) | |
1127 | return; | |
1128 | now = ktime_get(); | |
1129 | if (ts->idle_active) | |
e8fcaa5c | 1130 | tick_nohz_stop_idle(ts, now); |
ff006732 | 1131 | if (ts->tick_stopped) |
eed3b9cf | 1132 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1133 | } |
1134 | ||
79bf2bb3 TG |
1135 | #else |
1136 | ||
1137 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1138 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1139 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1140 | |
3451d024 | 1141 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1142 | |
719254fa TG |
1143 | /* |
1144 | * Called from irq_enter to notify about the possible interruption of idle() | |
1145 | */ | |
5acac1be | 1146 | void tick_irq_enter(void) |
719254fa | 1147 | { |
e8fcaa5c | 1148 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1149 | tick_nohz_irq_enter(); |
719254fa TG |
1150 | } |
1151 | ||
79bf2bb3 TG |
1152 | /* |
1153 | * High resolution timer specific code | |
1154 | */ | |
1155 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1156 | /* | |
4c9dc641 | 1157 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1158 | * Called with interrupts disabled. |
79bf2bb3 TG |
1159 | */ |
1160 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1161 | { | |
1162 | struct tick_sched *ts = | |
1163 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1164 | struct pt_regs *regs = get_irq_regs(); |
1165 | ktime_t now = ktime_get(); | |
d3ed7824 | 1166 | |
5bb96226 | 1167 | tick_sched_do_timer(now); |
79bf2bb3 TG |
1168 | |
1169 | /* | |
1170 | * Do not call, when we are not in irq context and have | |
1171 | * no valid regs pointer | |
1172 | */ | |
9e8f559b FW |
1173 | if (regs) |
1174 | tick_sched_handle(ts, regs); | |
79bf2bb3 | 1175 | |
2a16fc93 VK |
1176 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1177 | if (unlikely(ts->tick_stopped)) | |
1178 | return HRTIMER_NORESTART; | |
1179 | ||
79bf2bb3 TG |
1180 | hrtimer_forward(timer, now, tick_period); |
1181 | ||
1182 | return HRTIMER_RESTART; | |
1183 | } | |
1184 | ||
5307c955 MG |
1185 | static int sched_skew_tick; |
1186 | ||
62cf20b3 TG |
1187 | static int __init skew_tick(char *str) |
1188 | { | |
1189 | get_option(&str, &sched_skew_tick); | |
1190 | ||
1191 | return 0; | |
1192 | } | |
1193 | early_param("skew_tick", skew_tick); | |
1194 | ||
79bf2bb3 TG |
1195 | /** |
1196 | * tick_setup_sched_timer - setup the tick emulation timer | |
1197 | */ | |
1198 | void tick_setup_sched_timer(void) | |
1199 | { | |
22127e93 | 1200 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1201 | ktime_t now = ktime_get(); |
1202 | ||
1203 | /* | |
1204 | * Emulate tick processing via per-CPU hrtimers: | |
1205 | */ | |
1206 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1207 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 1208 | |
0de7611a | 1209 | /* Get the next period (per-CPU) */ |
cc584b21 | 1210 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1211 | |
9c3f9e28 | 1212 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
1213 | if (sched_skew_tick) { |
1214 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
1215 | do_div(offset, num_possible_cpus()); | |
1216 | offset *= smp_processor_id(); | |
1217 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1218 | } | |
1219 | ||
afc08b15 TG |
1220 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1221 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); | |
bc7a34b8 | 1222 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1223 | } |
3c4fbe5e | 1224 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1225 | |
3451d024 | 1226 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1227 | void tick_cancel_sched_timer(int cpu) |
1228 | { | |
1229 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1230 | ||
3c4fbe5e | 1231 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1232 | if (ts->sched_timer.base) |
1233 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1234 | # endif |
a7901766 | 1235 | |
4b0c0f29 | 1236 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1237 | } |
3c4fbe5e | 1238 | #endif |
79bf2bb3 TG |
1239 | |
1240 | /** | |
1241 | * Async notification about clocksource changes | |
1242 | */ | |
1243 | void tick_clock_notify(void) | |
1244 | { | |
1245 | int cpu; | |
1246 | ||
1247 | for_each_possible_cpu(cpu) | |
1248 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1249 | } | |
1250 | ||
1251 | /* | |
1252 | * Async notification about clock event changes | |
1253 | */ | |
1254 | void tick_oneshot_notify(void) | |
1255 | { | |
22127e93 | 1256 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1257 | |
1258 | set_bit(0, &ts->check_clocks); | |
1259 | } | |
1260 | ||
1261 | /** | |
1262 | * Check, if a change happened, which makes oneshot possible. | |
1263 | * | |
1264 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1265 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1266 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1267 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1268 | */ |
1269 | int tick_check_oneshot_change(int allow_nohz) | |
1270 | { | |
22127e93 | 1271 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1272 | |
1273 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1274 | return 0; | |
1275 | ||
1276 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1277 | return 0; | |
1278 | ||
cf4fc6cb | 1279 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1280 | return 0; |
1281 | ||
1282 | if (!allow_nohz) | |
1283 | return 1; | |
1284 | ||
1285 | tick_nohz_switch_to_nohz(); | |
1286 | return 0; | |
1287 | } |