]>
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> |
8083e4ad | 25 | #include <linux/module.h> |
00b42959 | 26 | #include <linux/irq_work.h> |
9014c45d | 27 | #include <linux/posix-timers.h> |
2e709338 | 28 | #include <linux/context_tracking.h> |
79bf2bb3 | 29 | |
9e203bcc DM |
30 | #include <asm/irq_regs.h> |
31 | ||
79bf2bb3 TG |
32 | #include "tick-internal.h" |
33 | ||
cb41a290 FW |
34 | #include <trace/events/timer.h> |
35 | ||
79bf2bb3 | 36 | /* |
0de7611a | 37 | * Per-CPU nohz control structure |
79bf2bb3 | 38 | */ |
c1797baf | 39 | static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); |
79bf2bb3 | 40 | |
289f480a IM |
41 | struct tick_sched *tick_get_tick_sched(int cpu) |
42 | { | |
43 | return &per_cpu(tick_cpu_sched, cpu); | |
44 | } | |
45 | ||
7809998a AB |
46 | #if defined(CONFIG_NO_HZ_COMMON) || defined(CONFIG_HIGH_RES_TIMERS) |
47 | /* | |
48 | * The time, when the last jiffy update happened. Protected by jiffies_lock. | |
49 | */ | |
50 | static ktime_t last_jiffies_update; | |
51 | ||
79bf2bb3 TG |
52 | /* |
53 | * Must be called with interrupts disabled ! | |
54 | */ | |
55 | static void tick_do_update_jiffies64(ktime_t now) | |
56 | { | |
57 | unsigned long ticks = 0; | |
58 | ktime_t delta; | |
59 | ||
7a14ce1d | 60 | /* |
d6ad4187 | 61 | * Do a quick check without holding jiffies_lock: |
7a14ce1d IM |
62 | */ |
63 | delta = ktime_sub(now, last_jiffies_update); | |
2456e855 | 64 | if (delta < tick_period) |
7a14ce1d IM |
65 | return; |
66 | ||
6168f8ed | 67 | /* Reevaluate with jiffies_lock held */ |
d6ad4187 | 68 | write_seqlock(&jiffies_lock); |
79bf2bb3 TG |
69 | |
70 | delta = ktime_sub(now, last_jiffies_update); | |
2456e855 | 71 | if (delta >= tick_period) { |
79bf2bb3 TG |
72 | |
73 | delta = ktime_sub(delta, tick_period); | |
74 | last_jiffies_update = ktime_add(last_jiffies_update, | |
75 | tick_period); | |
76 | ||
77 | /* Slow path for long timeouts */ | |
2456e855 | 78 | if (unlikely(delta >= tick_period)) { |
79bf2bb3 TG |
79 | s64 incr = ktime_to_ns(tick_period); |
80 | ||
81 | ticks = ktime_divns(delta, incr); | |
82 | ||
83 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
84 | incr * ticks); | |
85 | } | |
86 | do_timer(++ticks); | |
49d670fb TG |
87 | |
88 | /* Keep the tick_next_period variable up to date */ | |
89 | tick_next_period = ktime_add(last_jiffies_update, tick_period); | |
03e6bdc5 VK |
90 | } else { |
91 | write_sequnlock(&jiffies_lock); | |
92 | return; | |
79bf2bb3 | 93 | } |
d6ad4187 | 94 | write_sequnlock(&jiffies_lock); |
47a1b796 | 95 | update_wall_time(); |
79bf2bb3 TG |
96 | } |
97 | ||
98 | /* | |
99 | * Initialize and return retrieve the jiffies update. | |
100 | */ | |
101 | static ktime_t tick_init_jiffy_update(void) | |
102 | { | |
103 | ktime_t period; | |
104 | ||
d6ad4187 | 105 | write_seqlock(&jiffies_lock); |
79bf2bb3 | 106 | /* Did we start the jiffies update yet ? */ |
2456e855 | 107 | if (last_jiffies_update == 0) |
79bf2bb3 TG |
108 | last_jiffies_update = tick_next_period; |
109 | period = last_jiffies_update; | |
d6ad4187 | 110 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
111 | return period; |
112 | } | |
113 | ||
5bb96226 FW |
114 | |
115 | static void tick_sched_do_timer(ktime_t now) | |
116 | { | |
117 | int cpu = smp_processor_id(); | |
118 | ||
3451d024 | 119 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
120 | /* |
121 | * Check if the do_timer duty was dropped. We don't care about | |
0de7611a IM |
122 | * concurrency: This happens only when the CPU in charge went |
123 | * into a long sleep. If two CPUs happen to assign themselves to | |
5bb96226 | 124 | * this duty, then the jiffies update is still serialized by |
9c3f9e28 | 125 | * jiffies_lock. |
5bb96226 | 126 | */ |
a382bf93 | 127 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE) |
c5bfece2 | 128 | && !tick_nohz_full_cpu(cpu)) |
5bb96226 FW |
129 | tick_do_timer_cpu = cpu; |
130 | #endif | |
131 | ||
132 | /* Check, if the jiffies need an update */ | |
133 | if (tick_do_timer_cpu == cpu) | |
134 | tick_do_update_jiffies64(now); | |
135 | } | |
136 | ||
9e8f559b FW |
137 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
138 | { | |
3451d024 | 139 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
140 | /* |
141 | * When we are idle and the tick is stopped, we have to touch | |
142 | * the watchdog as we might not schedule for a really long | |
143 | * time. This happens on complete idle SMP systems while | |
144 | * waiting on the login prompt. We also increment the "start of | |
145 | * idle" jiffy stamp so the idle accounting adjustment we do | |
146 | * when we go busy again does not account too much ticks. | |
147 | */ | |
148 | if (ts->tick_stopped) { | |
03e0d461 | 149 | touch_softlockup_watchdog_sched(); |
9e8f559b FW |
150 | if (is_idle_task(current)) |
151 | ts->idle_jiffies++; | |
152 | } | |
94a57140 | 153 | #endif |
9e8f559b FW |
154 | update_process_times(user_mode(regs)); |
155 | profile_tick(CPU_PROFILING); | |
156 | } | |
7809998a | 157 | #endif |
9e8f559b | 158 | |
c5bfece2 | 159 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 160 | cpumask_var_t tick_nohz_full_mask; |
c0f489d2 | 161 | cpumask_var_t housekeeping_mask; |
73867dcd | 162 | bool tick_nohz_full_running; |
f009a7a7 | 163 | static atomic_t tick_dep_mask; |
a831881b | 164 | |
f009a7a7 | 165 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 166 | { |
f009a7a7 FW |
167 | int val = atomic_read(dep); |
168 | ||
169 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 170 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 171 | return true; |
d027d45d FW |
172 | } |
173 | ||
f009a7a7 | 174 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 175 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 176 | return true; |
d027d45d FW |
177 | } |
178 | ||
f009a7a7 | 179 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 180 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 181 | return true; |
d027d45d FW |
182 | } |
183 | ||
f009a7a7 | 184 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 185 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
186 | return true; |
187 | } | |
188 | ||
189 | return false; | |
d027d45d FW |
190 | } |
191 | ||
57ccdf44 | 192 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d FW |
193 | { |
194 | WARN_ON_ONCE(!irqs_disabled()); | |
195 | ||
57ccdf44 WL |
196 | if (unlikely(!cpu_online(cpu))) |
197 | return false; | |
198 | ||
f009a7a7 | 199 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 200 | return false; |
d027d45d | 201 | |
f009a7a7 | 202 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 203 | return false; |
d027d45d | 204 | |
f009a7a7 | 205 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 206 | return false; |
d027d45d | 207 | |
f009a7a7 | 208 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 209 | return false; |
d027d45d | 210 | |
9014c45d FW |
211 | return true; |
212 | } | |
213 | ||
d027d45d | 214 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 215 | { |
73738a95 | 216 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
217 | } |
218 | ||
219 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = { | |
d027d45d | 220 | .func = nohz_full_kick_func, |
76c24fb0 FW |
221 | }; |
222 | ||
40bea039 FW |
223 | /* |
224 | * Kick this CPU if it's full dynticks in order to force it to | |
225 | * re-evaluate its dependency on the tick and restart it if necessary. | |
226 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
227 | * is NMI safe. | |
228 | */ | |
555e0c1e | 229 | static void tick_nohz_full_kick(void) |
40bea039 FW |
230 | { |
231 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
232 | return; | |
233 | ||
56e4dea8 | 234 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
235 | } |
236 | ||
76c24fb0 | 237 | /* |
3d36aebc | 238 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
239 | * re-evaluate its dependency on the tick and restart it if necessary. |
240 | */ | |
3d36aebc | 241 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 242 | { |
3d36aebc FW |
243 | if (!tick_nohz_full_cpu(cpu)) |
244 | return; | |
245 | ||
246 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
247 | } |
248 | ||
76c24fb0 FW |
249 | /* |
250 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
251 | * their dependency on the tick and restart it if necessary. | |
252 | */ | |
b7878300 | 253 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 254 | { |
8537bb95 FW |
255 | int cpu; |
256 | ||
73867dcd | 257 | if (!tick_nohz_full_running) |
76c24fb0 FW |
258 | return; |
259 | ||
260 | preempt_disable(); | |
8537bb95 FW |
261 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
262 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
263 | preempt_enable(); |
264 | } | |
265 | ||
f009a7a7 | 266 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
267 | enum tick_dep_bits bit) |
268 | { | |
f009a7a7 | 269 | int prev; |
d027d45d | 270 | |
a1cc5bcf | 271 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
272 | if (!prev) |
273 | tick_nohz_full_kick_all(); | |
274 | } | |
275 | ||
276 | /* | |
277 | * Set a global tick dependency. Used by perf events that rely on freq and | |
278 | * by unstable clock. | |
279 | */ | |
280 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
281 | { | |
282 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
283 | } | |
284 | ||
285 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
286 | { | |
f009a7a7 | 287 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
288 | } |
289 | ||
290 | /* | |
291 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
292 | * manage events throttling. | |
293 | */ | |
294 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
295 | { | |
f009a7a7 | 296 | int prev; |
d027d45d FW |
297 | struct tick_sched *ts; |
298 | ||
299 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
300 | ||
a1cc5bcf | 301 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
302 | if (!prev) { |
303 | preempt_disable(); | |
304 | /* Perf needs local kick that is NMI safe */ | |
305 | if (cpu == smp_processor_id()) { | |
306 | tick_nohz_full_kick(); | |
307 | } else { | |
308 | /* Remote irq work not NMI-safe */ | |
309 | if (!WARN_ON_ONCE(in_nmi())) | |
310 | tick_nohz_full_kick_cpu(cpu); | |
311 | } | |
312 | preempt_enable(); | |
313 | } | |
314 | } | |
315 | ||
316 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
317 | { | |
318 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
319 | ||
f009a7a7 | 320 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
321 | } |
322 | ||
323 | /* | |
324 | * Set a per-task tick dependency. Posix CPU timers need this in order to elapse | |
325 | * per task timers. | |
326 | */ | |
327 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
328 | { | |
329 | /* | |
330 | * We could optimize this with just kicking the target running the task | |
331 | * if that noise matters for nohz full users. | |
332 | */ | |
333 | tick_nohz_dep_set_all(&tsk->tick_dep_mask, bit); | |
334 | } | |
335 | ||
336 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
337 | { | |
f009a7a7 | 338 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d FW |
339 | } |
340 | ||
341 | /* | |
342 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
343 | * per process timers. | |
344 | */ | |
345 | void tick_nohz_dep_set_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
346 | { | |
347 | tick_nohz_dep_set_all(&sig->tick_dep_mask, bit); | |
348 | } | |
349 | ||
350 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
351 | { | |
f009a7a7 | 352 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
353 | } |
354 | ||
99e5ada9 FW |
355 | /* |
356 | * Re-evaluate the need for the tick as we switch the current task. | |
357 | * It might need the tick due to per task/process properties: | |
0de7611a | 358 | * perf events, posix CPU timers, ... |
99e5ada9 | 359 | */ |
de734f89 | 360 | void __tick_nohz_task_switch(void) |
99e5ada9 FW |
361 | { |
362 | unsigned long flags; | |
d027d45d | 363 | struct tick_sched *ts; |
99e5ada9 | 364 | |
99e5ada9 FW |
365 | local_irq_save(flags); |
366 | ||
6296ace4 LZ |
367 | if (!tick_nohz_full_cpu(smp_processor_id())) |
368 | goto out; | |
369 | ||
d027d45d | 370 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 371 | |
d027d45d | 372 | if (ts->tick_stopped) { |
f009a7a7 FW |
373 | if (atomic_read(¤t->tick_dep_mask) || |
374 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
375 | tick_nohz_full_kick(); |
376 | } | |
6296ace4 | 377 | out: |
99e5ada9 FW |
378 | local_irq_restore(flags); |
379 | } | |
380 | ||
a831881b | 381 | /* Parse the boot-time nohz CPU list from the kernel parameters. */ |
c5bfece2 | 382 | static int __init tick_nohz_full_setup(char *str) |
a831881b | 383 | { |
73867dcd FW |
384 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
385 | if (cpulist_parse(str, tick_nohz_full_mask) < 0) { | |
a395d6a7 | 386 | pr_warn("NO_HZ: Incorrect nohz_full cpumask\n"); |
4327b15f | 387 | free_bootmem_cpumask_var(tick_nohz_full_mask); |
0453b435 FW |
388 | return 1; |
389 | } | |
73867dcd | 390 | tick_nohz_full_running = true; |
0453b435 | 391 | |
a831881b FW |
392 | return 1; |
393 | } | |
c5bfece2 | 394 | __setup("nohz_full=", tick_nohz_full_setup); |
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 | ||
f98823ac FW |
408 | static int tick_nohz_init_all(void) |
409 | { | |
410 | int err = -1; | |
411 | ||
412 | #ifdef CONFIG_NO_HZ_FULL_ALL | |
73867dcd | 413 | if (!alloc_cpumask_var(&tick_nohz_full_mask, GFP_KERNEL)) { |
4327b15f | 414 | WARN(1, "NO_HZ: Can't allocate full dynticks cpumask\n"); |
c0f489d2 PM |
415 | return err; |
416 | } | |
f98823ac | 417 | err = 0; |
73867dcd | 418 | cpumask_setall(tick_nohz_full_mask); |
73867dcd | 419 | tick_nohz_full_running = true; |
f98823ac FW |
420 | #endif |
421 | return err; | |
422 | } | |
423 | ||
d1e43fa5 | 424 | void __init tick_nohz_init(void) |
a831881b | 425 | { |
31eff243 | 426 | int cpu, ret; |
d1e43fa5 | 427 | |
73867dcd | 428 | if (!tick_nohz_full_running) { |
f98823ac FW |
429 | if (tick_nohz_init_all() < 0) |
430 | return; | |
431 | } | |
d1e43fa5 | 432 | |
4327b15f FW |
433 | if (!alloc_cpumask_var(&housekeeping_mask, GFP_KERNEL)) { |
434 | WARN(1, "NO_HZ: Can't allocate not-full dynticks cpumask\n"); | |
435 | cpumask_clear(tick_nohz_full_mask); | |
436 | tick_nohz_full_running = false; | |
437 | return; | |
438 | } | |
439 | ||
9b01f5bf FW |
440 | /* |
441 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
442 | * locking contexts. But then we need irq work to raise its own | |
443 | * interrupts to avoid circular dependency on the tick | |
444 | */ | |
445 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 446 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf FW |
447 | cpumask_clear(tick_nohz_full_mask); |
448 | cpumask_copy(housekeeping_mask, cpu_possible_mask); | |
449 | tick_nohz_full_running = false; | |
450 | return; | |
451 | } | |
452 | ||
4327b15f FW |
453 | cpu = smp_processor_id(); |
454 | ||
455 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { | |
a395d6a7 JP |
456 | pr_warn("NO_HZ: Clearing %d from nohz_full range for timekeeping\n", |
457 | cpu); | |
4327b15f FW |
458 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); |
459 | } | |
460 | ||
461 | cpumask_andnot(housekeeping_mask, | |
462 | cpu_possible_mask, tick_nohz_full_mask); | |
463 | ||
73867dcd | 464 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
465 | context_tracking_cpu_set(cpu); |
466 | ||
31eff243 SAS |
467 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
468 | "kernel/nohz:predown", NULL, | |
469 | tick_nohz_cpu_down); | |
470 | WARN_ON(ret < 0); | |
ffda22c1 TH |
471 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
472 | cpumask_pr_args(tick_nohz_full_mask)); | |
7c8bb6cb FW |
473 | |
474 | /* | |
475 | * We need at least one CPU to handle housekeeping work such | |
476 | * as timekeeping, unbound timers, workqueues, ... | |
477 | */ | |
478 | WARN_ON_ONCE(cpumask_empty(housekeeping_mask)); | |
a831881b | 479 | } |
a831881b FW |
480 | #endif |
481 | ||
79bf2bb3 TG |
482 | /* |
483 | * NOHZ - aka dynamic tick functionality | |
484 | */ | |
3451d024 | 485 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
486 | /* |
487 | * NO HZ enabled ? | |
488 | */ | |
4cc7ecb7 | 489 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 490 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
491 | /* |
492 | * Enable / Disable tickless mode | |
493 | */ | |
494 | static int __init setup_tick_nohz(char *str) | |
495 | { | |
4cc7ecb7 | 496 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
497 | } |
498 | ||
499 | __setup("nohz=", setup_tick_nohz); | |
500 | ||
c1797baf TG |
501 | int tick_nohz_tick_stopped(void) |
502 | { | |
503 | return __this_cpu_read(tick_cpu_sched.tick_stopped); | |
504 | } | |
505 | ||
79bf2bb3 TG |
506 | /** |
507 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
508 | * | |
509 | * Called from interrupt entry when the CPU was idle | |
510 | * | |
511 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
512 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
513 | * value. We do this unconditionally on any CPU, as we don't know whether the |
514 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 515 | */ |
eed3b9cf | 516 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 517 | { |
79bf2bb3 | 518 | unsigned long flags; |
79bf2bb3 | 519 | |
e8fcaa5c | 520 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
521 | |
522 | local_irq_save(flags); | |
523 | tick_do_update_jiffies64(now); | |
524 | local_irq_restore(flags); | |
02ff3755 | 525 | |
03e0d461 | 526 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
527 | } |
528 | ||
595aac48 | 529 | /* |
0de7611a | 530 | * Updates the per-CPU time idle statistics counters |
595aac48 | 531 | */ |
8d63bf94 | 532 | static void |
8c215bd3 | 533 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 534 | { |
eed3b9cf | 535 | ktime_t delta; |
6378ddb5 | 536 | |
595aac48 AV |
537 | if (ts->idle_active) { |
538 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 539 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 540 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
541 | else |
542 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 543 | ts->idle_entrytime = now; |
595aac48 | 544 | } |
8d63bf94 | 545 | |
e0e37c20 | 546 | if (last_update_time) |
8d63bf94 AV |
547 | *last_update_time = ktime_to_us(now); |
548 | ||
595aac48 AV |
549 | } |
550 | ||
e8fcaa5c | 551 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 552 | { |
e8fcaa5c | 553 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 554 | ts->idle_active = 0; |
56c7426b | 555 | |
eed3b9cf | 556 | sched_clock_idle_wakeup_event(0); |
6378ddb5 VP |
557 | } |
558 | ||
e8fcaa5c | 559 | static ktime_t tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 560 | { |
430ee881 | 561 | ktime_t now = ktime_get(); |
595aac48 | 562 | |
6378ddb5 VP |
563 | ts->idle_entrytime = now; |
564 | ts->idle_active = 1; | |
56c7426b | 565 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
566 | return now; |
567 | } | |
568 | ||
b1f724c3 | 569 | /** |
0de7611a | 570 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 571 | * @cpu: CPU number to query |
09a1d34f MH |
572 | * @last_update_time: variable to store update time in. Do not update |
573 | * counters if NULL. | |
b1f724c3 | 574 | * |
6168f8ed | 575 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 576 | * CPU, in microseconds. |
b1f724c3 AV |
577 | * |
578 | * This time is measured via accounting rather than sampling, | |
579 | * and is as accurate as ktime_get() is. | |
580 | * | |
581 | * This function returns -1 if NOHZ is not enabled. | |
582 | */ | |
6378ddb5 VP |
583 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
584 | { | |
585 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 586 | ktime_t now, idle; |
6378ddb5 | 587 | |
d689fe22 | 588 | if (!tick_nohz_active) |
8083e4ad | 589 | return -1; |
590 | ||
09a1d34f MH |
591 | now = ktime_get(); |
592 | if (last_update_time) { | |
593 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
594 | idle = ts->idle_sleeptime; | |
595 | } else { | |
596 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
597 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
598 | ||
599 | idle = ktime_add(ts->idle_sleeptime, delta); | |
600 | } else { | |
601 | idle = ts->idle_sleeptime; | |
602 | } | |
603 | } | |
604 | ||
605 | return ktime_to_us(idle); | |
8083e4ad | 606 | |
6378ddb5 | 607 | } |
8083e4ad | 608 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 609 | |
6beea0cd | 610 | /** |
0de7611a | 611 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 612 | * @cpu: CPU number to query |
09a1d34f MH |
613 | * @last_update_time: variable to store update time in. Do not update |
614 | * counters if NULL. | |
0224cf4c | 615 | * |
6168f8ed | 616 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
617 | * CPU, in microseconds. |
618 | * | |
619 | * This time is measured via accounting rather than sampling, | |
620 | * and is as accurate as ktime_get() is. | |
621 | * | |
622 | * This function returns -1 if NOHZ is not enabled. | |
623 | */ | |
624 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
625 | { | |
626 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 627 | ktime_t now, iowait; |
0224cf4c | 628 | |
d689fe22 | 629 | if (!tick_nohz_active) |
0224cf4c AV |
630 | return -1; |
631 | ||
09a1d34f MH |
632 | now = ktime_get(); |
633 | if (last_update_time) { | |
634 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
635 | iowait = ts->iowait_sleeptime; | |
636 | } else { | |
637 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
638 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 639 | |
09a1d34f MH |
640 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
641 | } else { | |
642 | iowait = ts->iowait_sleeptime; | |
643 | } | |
644 | } | |
0224cf4c | 645 | |
09a1d34f | 646 | return ktime_to_us(iowait); |
0224cf4c AV |
647 | } |
648 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
649 | ||
0ff53d09 TG |
650 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
651 | { | |
652 | hrtimer_cancel(&ts->sched_timer); | |
653 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
654 | ||
655 | /* Forward the time to expire in the future */ | |
656 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
657 | ||
658 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) | |
659 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); | |
660 | else | |
661 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
662 | } | |
663 | ||
84bf1bcc FW |
664 | static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts, |
665 | ktime_t now, int cpu) | |
79bf2bb3 | 666 | { |
22127e93 | 667 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
c1ad348b TG |
668 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
669 | unsigned long seq, basejiff; | |
670 | ktime_t tick; | |
855a0fc3 | 671 | |
79bf2bb3 TG |
672 | /* Read jiffies and the time when jiffies were updated last */ |
673 | do { | |
d6ad4187 | 674 | seq = read_seqbegin(&jiffies_lock); |
2456e855 | 675 | basemono = last_jiffies_update; |
c1ad348b | 676 | basejiff = jiffies; |
d6ad4187 | 677 | } while (read_seqretry(&jiffies_lock, seq)); |
c1ad348b | 678 | ts->last_jiffies = basejiff; |
79bf2bb3 | 679 | |
c1ad348b | 680 | if (rcu_needs_cpu(basemono, &next_rcu) || |
fe0f4976 | 681 | arch_needs_cpu() || irq_work_needs_cpu()) { |
c1ad348b | 682 | next_tick = basemono + TICK_NSEC; |
3c5d92a0 | 683 | } else { |
c1ad348b TG |
684 | /* |
685 | * Get the next pending timer. If high resolution | |
686 | * timers are enabled this only takes the timer wheel | |
687 | * timers into account. If high resolution timers are | |
688 | * disabled this also looks at the next expiring | |
689 | * hrtimer. | |
690 | */ | |
691 | next_tmr = get_next_timer_interrupt(basejiff, basemono); | |
692 | ts->next_timer = next_tmr; | |
693 | /* Take the next rcu event into account */ | |
694 | next_tick = next_rcu < next_tmr ? next_rcu : next_tmr; | |
3c5d92a0 | 695 | } |
47aa8b6c | 696 | |
c1ad348b TG |
697 | /* |
698 | * If the tick is due in the next period, keep it ticking or | |
82bbe34b | 699 | * force prod the timer. |
c1ad348b TG |
700 | */ |
701 | delta = next_tick - basemono; | |
702 | if (delta <= (u64)TICK_NSEC) { | |
2456e855 | 703 | tick = 0; |
a683f390 TG |
704 | |
705 | /* | |
706 | * Tell the timer code that the base is not idle, i.e. undo | |
707 | * the effect of get_next_timer_interrupt(): | |
708 | */ | |
709 | timer_clear_idle(); | |
82bbe34b PZ |
710 | /* |
711 | * We've not stopped the tick yet, and there's a timer in the | |
712 | * next period, so no point in stopping it either, bail. | |
713 | */ | |
157d29e1 TG |
714 | if (!ts->tick_stopped) |
715 | goto out; | |
82bbe34b PZ |
716 | |
717 | /* | |
718 | * If, OTOH, we did stop it, but there's a pending (expired) | |
719 | * timer reprogram the timer hardware to fire now. | |
720 | * | |
721 | * We will not restart the tick proper, just prod the timer | |
722 | * hardware into firing an interrupt to process the pending | |
723 | * timers. Just like tick_irq_exit() will not restart the tick | |
724 | * for 'normal' interrupts. | |
725 | * | |
726 | * Only once we exit the idle loop will we re-enable the tick, | |
727 | * see tick_nohz_idle_exit(). | |
728 | */ | |
c1ad348b | 729 | if (delta == 0) { |
157d29e1 TG |
730 | tick_nohz_restart(ts, now); |
731 | goto out; | |
732 | } | |
733 | } | |
734 | ||
79bf2bb3 | 735 | /* |
0de7611a IM |
736 | * If this CPU is the one which updates jiffies, then give up |
737 | * the assignment and let it be taken by the CPU which runs | |
738 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
739 | * don't drop this here the jiffies might be stale and |
740 | * do_timer() never invoked. Keep track of the fact that it | |
0de7611a | 741 | * was the one which had the do_timer() duty last. If this CPU |
157d29e1 | 742 | * is the one which had the do_timer() duty last, we limit the |
6168f8ed | 743 | * sleep time to the timekeeping max_deferment value. |
c1ad348b | 744 | * Otherwise we can sleep as long as we want. |
79bf2bb3 | 745 | */ |
c1ad348b | 746 | delta = timekeeping_max_deferment(); |
157d29e1 TG |
747 | if (cpu == tick_do_timer_cpu) { |
748 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
749 | ts->do_timer_last = 1; | |
750 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
c1ad348b | 751 | delta = KTIME_MAX; |
157d29e1 TG |
752 | ts->do_timer_last = 0; |
753 | } else if (!ts->do_timer_last) { | |
c1ad348b | 754 | delta = KTIME_MAX; |
157d29e1 | 755 | } |
27185016 | 756 | |
265f22a9 | 757 | #ifdef CONFIG_NO_HZ_FULL |
c1ad348b | 758 | /* Limit the tick delta to the maximum scheduler deferment */ |
157d29e1 | 759 | if (!ts->inidle) |
c1ad348b | 760 | delta = min(delta, scheduler_tick_max_deferment()); |
265f22a9 FW |
761 | #endif |
762 | ||
c1ad348b TG |
763 | /* Calculate the next expiry time */ |
764 | if (delta < (KTIME_MAX - basemono)) | |
765 | expires = basemono + delta; | |
157d29e1 | 766 | else |
c1ad348b TG |
767 | expires = KTIME_MAX; |
768 | ||
769 | expires = min_t(u64, expires, next_tick); | |
2456e855 | 770 | tick = expires; |
00147449 | 771 | |
157d29e1 | 772 | /* Skip reprogram of event if its not changed */ |
558e8e27 | 773 | if (ts->tick_stopped && (expires == dev->next_event)) |
157d29e1 | 774 | goto out; |
84bf1bcc | 775 | |
157d29e1 TG |
776 | /* |
777 | * nohz_stop_sched_tick can be called several times before | |
778 | * the nohz_restart_sched_tick is called. This happens when | |
779 | * interrupts arrive which do not cause a reschedule. In the | |
780 | * first call we save the current tick time, so we can restart | |
781 | * the scheduler tick in nohz_restart_sched_tick. | |
782 | */ | |
783 | if (!ts->tick_stopped) { | |
784 | nohz_balance_enter_idle(cpu); | |
785 | calc_load_enter_idle(); | |
1f41906a | 786 | cpu_load_update_nohz_start(); |
d3ed7824 | 787 | |
157d29e1 TG |
788 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
789 | ts->tick_stopped = 1; | |
e6e6cc22 | 790 | trace_tick_stop(1, TICK_DEP_MASK_NONE); |
157d29e1 | 791 | } |
eaad084b | 792 | |
157d29e1 | 793 | /* |
c1ad348b TG |
794 | * If the expiration time == KTIME_MAX, then we simply stop |
795 | * the tick timer. | |
157d29e1 | 796 | */ |
c1ad348b | 797 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
798 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
799 | hrtimer_cancel(&ts->sched_timer); | |
800 | goto out; | |
79bf2bb3 | 801 | } |
0ff53d09 | 802 | |
157d29e1 | 803 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
c1ad348b | 804 | hrtimer_start(&ts->sched_timer, tick, HRTIMER_MODE_ABS_PINNED); |
157d29e1 | 805 | else |
c1ad348b | 806 | tick_program_event(tick, 1); |
79bf2bb3 | 807 | out: |
558e8e27 | 808 | /* Update the estimated sleep length */ |
4f86d3a8 | 809 | ts->sleep_length = ktime_sub(dev->next_event, now); |
c1ad348b | 810 | return tick; |
280f0677 FW |
811 | } |
812 | ||
1f41906a | 813 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
814 | { |
815 | /* Update jiffies first */ | |
816 | tick_do_update_jiffies64(now); | |
1f41906a | 817 | cpu_load_update_nohz_stop(); |
59d2c7ca | 818 | |
a683f390 TG |
819 | /* |
820 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
821 | * the clock forward checks in the enqueue path: | |
822 | */ | |
823 | timer_clear_idle(); | |
824 | ||
59d2c7ca | 825 | calc_load_exit_idle(); |
03e0d461 | 826 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
827 | /* |
828 | * Cancel the scheduled timer and restore the tick | |
829 | */ | |
830 | ts->tick_stopped = 0; | |
831 | ts->idle_exittime = now; | |
832 | ||
833 | tick_nohz_restart(ts, now); | |
834 | } | |
73738a95 FW |
835 | |
836 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
5811d996 FW |
837 | { |
838 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 839 | int cpu = smp_processor_id(); |
5811d996 | 840 | |
59449359 | 841 | if (!tick_nohz_full_cpu(cpu)) |
e9a2eb40 | 842 | return; |
5811d996 | 843 | |
e9a2eb40 AS |
844 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
845 | return; | |
5811d996 | 846 | |
57ccdf44 | 847 | if (can_stop_full_tick(cpu, ts)) |
73738a95 FW |
848 | tick_nohz_stop_sched_tick(ts, ktime_get(), cpu); |
849 | else if (ts->tick_stopped) | |
1f41906a | 850 | tick_nohz_restart_sched_tick(ts, ktime_get()); |
5811d996 FW |
851 | #endif |
852 | } | |
853 | ||
5b39939a FW |
854 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
855 | { | |
856 | /* | |
0de7611a | 857 | * If this CPU is offline and it is the one which updates |
5b39939a | 858 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 859 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
860 | * this here the jiffies might be stale and do_timer() never |
861 | * invoked. | |
862 | */ | |
863 | if (unlikely(!cpu_online(cpu))) { | |
864 | if (cpu == tick_do_timer_cpu) | |
865 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
f7ea0fd6 | 866 | return false; |
5b39939a FW |
867 | } |
868 | ||
0e576acb | 869 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) { |
2456e855 | 870 | ts->sleep_length = NSEC_PER_SEC / HZ; |
5b39939a | 871 | return false; |
0e576acb | 872 | } |
5b39939a FW |
873 | |
874 | if (need_resched()) | |
875 | return false; | |
876 | ||
877 | if (unlikely(local_softirq_pending() && cpu_online(cpu))) { | |
878 | static int ratelimit; | |
879 | ||
803b0eba PM |
880 | if (ratelimit < 10 && |
881 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
cfea7d7e RV |
882 | pr_warn("NOHZ: local_softirq_pending %02x\n", |
883 | (unsigned int) local_softirq_pending()); | |
5b39939a FW |
884 | ratelimit++; |
885 | } | |
886 | return false; | |
887 | } | |
888 | ||
460775df | 889 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
890 | /* |
891 | * Keep the tick alive to guarantee timekeeping progression | |
892 | * if there are full dynticks CPUs around | |
893 | */ | |
894 | if (tick_do_timer_cpu == cpu) | |
895 | return false; | |
896 | /* | |
897 | * Boot safety: make sure the timekeeping duty has been | |
898 | * assigned before entering dyntick-idle mode, | |
899 | */ | |
900 | if (tick_do_timer_cpu == TICK_DO_TIMER_NONE) | |
901 | return false; | |
902 | } | |
903 | ||
5b39939a FW |
904 | return true; |
905 | } | |
906 | ||
19f5f736 FW |
907 | static void __tick_nohz_idle_enter(struct tick_sched *ts) |
908 | { | |
84bf1bcc | 909 | ktime_t now, expires; |
5b39939a | 910 | int cpu = smp_processor_id(); |
19f5f736 | 911 | |
08d07259 WL |
912 | now = tick_nohz_start_idle(ts); |
913 | ||
5b39939a FW |
914 | if (can_stop_idle_tick(cpu, ts)) { |
915 | int was_stopped = ts->tick_stopped; | |
916 | ||
917 | ts->idle_calls++; | |
84bf1bcc FW |
918 | |
919 | expires = tick_nohz_stop_sched_tick(ts, now, cpu); | |
2456e855 | 920 | if (expires > 0LL) { |
84bf1bcc FW |
921 | ts->idle_sleeps++; |
922 | ts->idle_expires = expires; | |
923 | } | |
5b39939a FW |
924 | |
925 | if (!was_stopped && ts->tick_stopped) | |
926 | ts->idle_jiffies = ts->last_jiffies; | |
927 | } | |
280f0677 FW |
928 | } |
929 | ||
930 | /** | |
931 | * tick_nohz_idle_enter - stop the idle tick from the idle task | |
932 | * | |
933 | * When the next event is more than a tick into the future, stop the idle tick | |
934 | * Called when we start the idle loop. | |
2bbb6817 | 935 | * |
1268fbc7 | 936 | * The arch is responsible of calling: |
2bbb6817 FW |
937 | * |
938 | * - rcu_idle_enter() after its last use of RCU before the CPU is put | |
939 | * to sleep. | |
940 | * - rcu_idle_exit() before the first use of RCU after the CPU is woken up. | |
280f0677 | 941 | */ |
1268fbc7 | 942 | void tick_nohz_idle_enter(void) |
280f0677 FW |
943 | { |
944 | struct tick_sched *ts; | |
945 | ||
1268fbc7 FW |
946 | WARN_ON_ONCE(irqs_disabled()); |
947 | ||
0db49b72 | 948 | /* |
0de7611a IM |
949 | * Update the idle state in the scheduler domain hierarchy |
950 | * when tick_nohz_stop_sched_tick() is called from the idle loop. | |
951 | * State will be updated to busy during the first busy tick after | |
952 | * exiting idle. | |
953 | */ | |
0db49b72 LT |
954 | set_cpu_sd_state_idle(); |
955 | ||
1268fbc7 FW |
956 | local_irq_disable(); |
957 | ||
22127e93 | 958 | ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 959 | ts->inidle = 1; |
19f5f736 | 960 | __tick_nohz_idle_enter(ts); |
1268fbc7 FW |
961 | |
962 | local_irq_enable(); | |
280f0677 FW |
963 | } |
964 | ||
965 | /** | |
966 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
967 | * | |
968 | * When an interrupt fires while we are idle and it doesn't cause | |
969 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
970 | * an RCU callback, etc... | |
971 | * So we need to re-calculate and reprogram the next tick event. | |
972 | */ | |
973 | void tick_nohz_irq_exit(void) | |
974 | { | |
22127e93 | 975 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 976 | |
14851912 | 977 | if (ts->inidle) |
5811d996 | 978 | __tick_nohz_idle_enter(ts); |
14851912 | 979 | else |
73738a95 | 980 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
981 | } |
982 | ||
4f86d3a8 LB |
983 | /** |
984 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
985 | * | |
986 | * Called from power state control code with interrupts disabled | |
987 | */ | |
988 | ktime_t tick_nohz_get_sleep_length(void) | |
989 | { | |
22127e93 | 990 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
4f86d3a8 LB |
991 | |
992 | return ts->sleep_length; | |
993 | } | |
994 | ||
2ac0d98f FW |
995 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) |
996 | { | |
3f4724ea | 997 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 998 | unsigned long ticks; |
3f4724ea | 999 | |
55dbdcfa | 1000 | if (vtime_accounting_cpu_enabled()) |
3f4724ea | 1001 | return; |
79bf2bb3 TG |
1002 | /* |
1003 | * We stopped the tick in idle. Update process times would miss the | |
1004 | * time we slept as update_process_times does only a 1 tick | |
1005 | * accounting. Enforce that this is accounted to idle ! | |
1006 | */ | |
1007 | ticks = jiffies - ts->idle_jiffies; | |
1008 | /* | |
1009 | * We might be one off. Do not randomly account a huge number of ticks! | |
1010 | */ | |
79741dd3 MS |
1011 | if (ticks && ticks < LONG_MAX) |
1012 | account_idle_ticks(ticks); | |
1013 | #endif | |
19f5f736 FW |
1014 | } |
1015 | ||
79bf2bb3 | 1016 | /** |
280f0677 | 1017 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1018 | * |
1019 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1020 | * This also exit the RCU extended quiescent state. The CPU |
1021 | * can use RCU again after this function is called. | |
79bf2bb3 | 1022 | */ |
280f0677 | 1023 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1024 | { |
4a32fea9 | 1025 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
6378ddb5 | 1026 | ktime_t now; |
79bf2bb3 | 1027 | |
6378ddb5 | 1028 | local_irq_disable(); |
2bbb6817 | 1029 | |
15f827be FW |
1030 | WARN_ON_ONCE(!ts->inidle); |
1031 | ||
1032 | ts->inidle = 0; | |
1033 | ||
1034 | if (ts->idle_active || ts->tick_stopped) | |
eed3b9cf MS |
1035 | now = ktime_get(); |
1036 | ||
1037 | if (ts->idle_active) | |
e8fcaa5c | 1038 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1039 | |
2ac0d98f | 1040 | if (ts->tick_stopped) { |
1f41906a | 1041 | tick_nohz_restart_sched_tick(ts, now); |
2ac0d98f | 1042 | tick_nohz_account_idle_ticks(ts); |
6378ddb5 | 1043 | } |
79bf2bb3 | 1044 | |
79bf2bb3 TG |
1045 | local_irq_enable(); |
1046 | } | |
1047 | ||
79bf2bb3 TG |
1048 | /* |
1049 | * The nohz low res interrupt handler | |
1050 | */ | |
1051 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1052 | { | |
22127e93 | 1053 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1054 | struct pt_regs *regs = get_irq_regs(); |
1055 | ktime_t now = ktime_get(); | |
1056 | ||
2456e855 | 1057 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1058 | |
5bb96226 | 1059 | tick_sched_do_timer(now); |
9e8f559b | 1060 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1061 | |
b5e995e6 VK |
1062 | /* No need to reprogram if we are running tickless */ |
1063 | if (unlikely(ts->tick_stopped)) | |
1064 | return; | |
1065 | ||
0ff53d09 TG |
1066 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1067 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
79bf2bb3 TG |
1068 | } |
1069 | ||
bc7a34b8 TG |
1070 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1071 | { | |
1072 | if (!tick_nohz_enabled) | |
1073 | return; | |
1074 | ts->nohz_mode = mode; | |
1075 | /* One update is enough */ | |
1076 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
683be13a | 1077 | timers_update_migration(true); |
bc7a34b8 TG |
1078 | } |
1079 | ||
79bf2bb3 TG |
1080 | /** |
1081 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1082 | */ | |
1083 | static void tick_nohz_switch_to_nohz(void) | |
1084 | { | |
22127e93 | 1085 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1086 | ktime_t next; |
1087 | ||
27630532 | 1088 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1089 | return; |
1090 | ||
6b442bc8 | 1091 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1092 | return; |
6b442bc8 | 1093 | |
79bf2bb3 TG |
1094 | /* |
1095 | * Recycle the hrtimer in ts, so we can share the | |
1096 | * hrtimer_forward with the highres code. | |
1097 | */ | |
1098 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1099 | /* Get the next period */ | |
1100 | next = tick_init_jiffy_update(); | |
1101 | ||
0ff53d09 | 1102 | hrtimer_set_expires(&ts->sched_timer, next); |
1ca8ec53 WL |
1103 | hrtimer_forward_now(&ts->sched_timer, tick_period); |
1104 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); | |
bc7a34b8 | 1105 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1106 | } |
1107 | ||
5acac1be | 1108 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1109 | { |
4a32fea9 | 1110 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1111 | ktime_t now; |
1112 | ||
1113 | if (!ts->idle_active && !ts->tick_stopped) | |
1114 | return; | |
1115 | now = ktime_get(); | |
1116 | if (ts->idle_active) | |
e8fcaa5c | 1117 | tick_nohz_stop_idle(ts, now); |
ff006732 | 1118 | if (ts->tick_stopped) |
eed3b9cf | 1119 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1120 | } |
1121 | ||
79bf2bb3 TG |
1122 | #else |
1123 | ||
1124 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1125 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1126 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1127 | |
3451d024 | 1128 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1129 | |
719254fa TG |
1130 | /* |
1131 | * Called from irq_enter to notify about the possible interruption of idle() | |
1132 | */ | |
5acac1be | 1133 | void tick_irq_enter(void) |
719254fa | 1134 | { |
e8fcaa5c | 1135 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1136 | tick_nohz_irq_enter(); |
719254fa TG |
1137 | } |
1138 | ||
79bf2bb3 TG |
1139 | /* |
1140 | * High resolution timer specific code | |
1141 | */ | |
1142 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1143 | /* | |
4c9dc641 | 1144 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1145 | * Called with interrupts disabled. |
79bf2bb3 TG |
1146 | */ |
1147 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1148 | { | |
1149 | struct tick_sched *ts = | |
1150 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1151 | struct pt_regs *regs = get_irq_regs(); |
1152 | ktime_t now = ktime_get(); | |
d3ed7824 | 1153 | |
5bb96226 | 1154 | tick_sched_do_timer(now); |
79bf2bb3 TG |
1155 | |
1156 | /* | |
1157 | * Do not call, when we are not in irq context and have | |
1158 | * no valid regs pointer | |
1159 | */ | |
9e8f559b FW |
1160 | if (regs) |
1161 | tick_sched_handle(ts, regs); | |
79bf2bb3 | 1162 | |
2a16fc93 VK |
1163 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1164 | if (unlikely(ts->tick_stopped)) | |
1165 | return HRTIMER_NORESTART; | |
1166 | ||
79bf2bb3 TG |
1167 | hrtimer_forward(timer, now, tick_period); |
1168 | ||
1169 | return HRTIMER_RESTART; | |
1170 | } | |
1171 | ||
5307c955 MG |
1172 | static int sched_skew_tick; |
1173 | ||
62cf20b3 TG |
1174 | static int __init skew_tick(char *str) |
1175 | { | |
1176 | get_option(&str, &sched_skew_tick); | |
1177 | ||
1178 | return 0; | |
1179 | } | |
1180 | early_param("skew_tick", skew_tick); | |
1181 | ||
79bf2bb3 TG |
1182 | /** |
1183 | * tick_setup_sched_timer - setup the tick emulation timer | |
1184 | */ | |
1185 | void tick_setup_sched_timer(void) | |
1186 | { | |
22127e93 | 1187 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1188 | ktime_t now = ktime_get(); |
1189 | ||
1190 | /* | |
1191 | * Emulate tick processing via per-CPU hrtimers: | |
1192 | */ | |
1193 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1194 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 1195 | |
0de7611a | 1196 | /* Get the next period (per-CPU) */ |
cc584b21 | 1197 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1198 | |
9c3f9e28 | 1199 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
1200 | if (sched_skew_tick) { |
1201 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
1202 | do_div(offset, num_possible_cpus()); | |
1203 | offset *= smp_processor_id(); | |
1204 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1205 | } | |
1206 | ||
afc08b15 TG |
1207 | hrtimer_forward(&ts->sched_timer, now, tick_period); |
1208 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED); | |
bc7a34b8 | 1209 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1210 | } |
3c4fbe5e | 1211 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1212 | |
3451d024 | 1213 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1214 | void tick_cancel_sched_timer(int cpu) |
1215 | { | |
1216 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1217 | ||
3c4fbe5e | 1218 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1219 | if (ts->sched_timer.base) |
1220 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1221 | # endif |
a7901766 | 1222 | |
4b0c0f29 | 1223 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1224 | } |
3c4fbe5e | 1225 | #endif |
79bf2bb3 TG |
1226 | |
1227 | /** | |
1228 | * Async notification about clocksource changes | |
1229 | */ | |
1230 | void tick_clock_notify(void) | |
1231 | { | |
1232 | int cpu; | |
1233 | ||
1234 | for_each_possible_cpu(cpu) | |
1235 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1236 | } | |
1237 | ||
1238 | /* | |
1239 | * Async notification about clock event changes | |
1240 | */ | |
1241 | void tick_oneshot_notify(void) | |
1242 | { | |
22127e93 | 1243 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1244 | |
1245 | set_bit(0, &ts->check_clocks); | |
1246 | } | |
1247 | ||
1248 | /** | |
1249 | * Check, if a change happened, which makes oneshot possible. | |
1250 | * | |
1251 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1252 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1253 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1254 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1255 | */ |
1256 | int tick_check_oneshot_change(int allow_nohz) | |
1257 | { | |
22127e93 | 1258 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1259 | |
1260 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1261 | return 0; | |
1262 | ||
1263 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1264 | return 0; | |
1265 | ||
cf4fc6cb | 1266 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1267 | return 0; |
1268 | ||
1269 | if (!allow_nohz) | |
1270 | return 1; | |
1271 | ||
1272 | tick_nohz_switch_to_nohz(); | |
1273 | return 0; | |
1274 | } |