Commit | Line | Data |
---|---|---|
35728b82 | 1 | // SPDX-License-Identifier: GPL-2.0 |
79bf2bb3 | 2 | /* |
79bf2bb3 TG |
3 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> |
4 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar | |
5 | * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner | |
6 | * | |
7 | * No idle tick implementation for low and high resolution timers | |
8 | * | |
9 | * Started by: Thomas Gleixner and Ingo Molnar | |
79bf2bb3 TG |
10 | */ |
11 | #include <linux/cpu.h> | |
12 | #include <linux/err.h> | |
13 | #include <linux/hrtimer.h> | |
14 | #include <linux/interrupt.h> | |
15 | #include <linux/kernel_stat.h> | |
16 | #include <linux/percpu.h> | |
38b8d208 | 17 | #include <linux/nmi.h> |
79bf2bb3 | 18 | #include <linux/profile.h> |
3f07c014 | 19 | #include <linux/sched/signal.h> |
e6017571 | 20 | #include <linux/sched/clock.h> |
03441a34 | 21 | #include <linux/sched/stat.h> |
370c9135 | 22 | #include <linux/sched/nohz.h> |
896b969e | 23 | #include <linux/sched/loadavg.h> |
8083e4ad | 24 | #include <linux/module.h> |
00b42959 | 25 | #include <linux/irq_work.h> |
9014c45d | 26 | #include <linux/posix-timers.h> |
2e709338 | 27 | #include <linux/context_tracking.h> |
62cb1188 | 28 | #include <linux/mm.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 | /* | |
c398960c TG |
48 | * The time, when the last jiffy update happened. Write access must hold |
49 | * jiffies_lock and jiffies_seq. tick_nohz_next_event() needs to get a | |
50 | * consistent view of jiffies and last_jiffies_update. | |
7809998a AB |
51 | */ |
52 | static ktime_t last_jiffies_update; | |
53 | ||
79bf2bb3 TG |
54 | /* |
55 | * Must be called with interrupts disabled ! | |
56 | */ | |
57 | static void tick_do_update_jiffies64(ktime_t now) | |
58 | { | |
7a35bf2a | 59 | unsigned long ticks = 1; |
aa3b66f4 | 60 | ktime_t delta, nextp; |
79bf2bb3 | 61 | |
7a14ce1d | 62 | /* |
aa3b66f4 TG |
63 | * 64bit can do a quick check without holding jiffies lock and |
64 | * without looking at the sequence count. The smp_load_acquire() | |
372acbba TG |
65 | * pairs with the update done later in this function. |
66 | * | |
aa3b66f4 TG |
67 | * 32bit cannot do that because the store of tick_next_period |
68 | * consists of two 32bit stores and the first store could move it | |
69 | * to a random point in the future. | |
7a14ce1d | 70 | */ |
aa3b66f4 TG |
71 | if (IS_ENABLED(CONFIG_64BIT)) { |
72 | if (ktime_before(now, smp_load_acquire(&tick_next_period))) | |
73 | return; | |
74 | } else { | |
75 | unsigned int seq; | |
76 | ||
77 | /* | |
78 | * Avoid contention on jiffies_lock and protect the quick | |
79 | * check with the sequence count. | |
80 | */ | |
81 | do { | |
82 | seq = read_seqcount_begin(&jiffies_seq); | |
83 | nextp = tick_next_period; | |
84 | } while (read_seqcount_retry(&jiffies_seq, seq)); | |
85 | ||
86 | if (ktime_before(now, nextp)) | |
87 | return; | |
88 | } | |
7a14ce1d | 89 | |
aa3b66f4 | 90 | /* Quick check failed, i.e. update is required. */ |
e5d4d175 | 91 | raw_spin_lock(&jiffies_lock); |
aa3b66f4 TG |
92 | /* |
93 | * Reevaluate with the lock held. Another CPU might have done the | |
94 | * update already. | |
95 | */ | |
94ad2e3c | 96 | if (ktime_before(now, tick_next_period)) { |
e5d4d175 | 97 | raw_spin_unlock(&jiffies_lock); |
03e6bdc5 | 98 | return; |
79bf2bb3 | 99 | } |
94ad2e3c YY |
100 | |
101 | write_seqcount_begin(&jiffies_seq); | |
102 | ||
94ad2e3c | 103 | delta = ktime_sub(now, tick_next_period); |
b9965449 | 104 | if (unlikely(delta >= TICK_NSEC)) { |
94ad2e3c | 105 | /* Slow path for long idle sleep times */ |
b9965449 | 106 | s64 incr = TICK_NSEC; |
94ad2e3c | 107 | |
7a35bf2a | 108 | ticks += ktime_divns(delta, incr); |
94ad2e3c YY |
109 | |
110 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
111 | incr * ticks); | |
7a35bf2a | 112 | } else { |
b9965449 TG |
113 | last_jiffies_update = ktime_add_ns(last_jiffies_update, |
114 | TICK_NSEC); | |
94ad2e3c YY |
115 | } |
116 | ||
896b969e YY |
117 | /* Advance jiffies to complete the jiffies_seq protected job */ |
118 | jiffies_64 += ticks; | |
94ad2e3c YY |
119 | |
120 | /* | |
aa3b66f4 | 121 | * Keep the tick_next_period variable up to date. |
94ad2e3c | 122 | */ |
aa3b66f4 TG |
123 | nextp = ktime_add_ns(last_jiffies_update, TICK_NSEC); |
124 | ||
125 | if (IS_ENABLED(CONFIG_64BIT)) { | |
126 | /* | |
127 | * Pairs with smp_load_acquire() in the lockless quick | |
128 | * check above and ensures that the update to jiffies_64 is | |
129 | * not reordered vs. the store to tick_next_period, neither | |
130 | * by the compiler nor by the CPU. | |
131 | */ | |
132 | smp_store_release(&tick_next_period, nextp); | |
133 | } else { | |
134 | /* | |
135 | * A plain store is good enough on 32bit as the quick check | |
136 | * above is protected by the sequence count. | |
137 | */ | |
138 | tick_next_period = nextp; | |
139 | } | |
94ad2e3c | 140 | |
896b969e YY |
141 | /* |
142 | * Release the sequence count. calc_global_load() below is not | |
143 | * protected by it, but jiffies_lock needs to be held to prevent | |
144 | * concurrent invocations. | |
145 | */ | |
e5d4d175 | 146 | write_seqcount_end(&jiffies_seq); |
896b969e YY |
147 | |
148 | calc_global_load(); | |
149 | ||
e5d4d175 | 150 | raw_spin_unlock(&jiffies_lock); |
47a1b796 | 151 | update_wall_time(); |
79bf2bb3 TG |
152 | } |
153 | ||
154 | /* | |
155 | * Initialize and return retrieve the jiffies update. | |
156 | */ | |
157 | static ktime_t tick_init_jiffy_update(void) | |
158 | { | |
159 | ktime_t period; | |
160 | ||
e5d4d175 TG |
161 | raw_spin_lock(&jiffies_lock); |
162 | write_seqcount_begin(&jiffies_seq); | |
79bf2bb3 | 163 | /* Did we start the jiffies update yet ? */ |
2456e855 | 164 | if (last_jiffies_update == 0) |
79bf2bb3 TG |
165 | last_jiffies_update = tick_next_period; |
166 | period = last_jiffies_update; | |
e5d4d175 TG |
167 | write_seqcount_end(&jiffies_seq); |
168 | raw_spin_unlock(&jiffies_lock); | |
79bf2bb3 TG |
169 | return period; |
170 | } | |
171 | ||
a1ff03cd FW |
172 | #define MAX_STALLED_JIFFIES 5 |
173 | ||
ff7de620 | 174 | static void tick_sched_do_timer(struct tick_sched *ts, ktime_t now) |
5bb96226 FW |
175 | { |
176 | int cpu = smp_processor_id(); | |
177 | ||
3451d024 | 178 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
179 | /* |
180 | * Check if the do_timer duty was dropped. We don't care about | |
0de7611a IM |
181 | * concurrency: This happens only when the CPU in charge went |
182 | * into a long sleep. If two CPUs happen to assign themselves to | |
5bb96226 | 183 | * this duty, then the jiffies update is still serialized by |
9c3f9e28 | 184 | * jiffies_lock. |
08ae95f4 NP |
185 | * |
186 | * If nohz_full is enabled, this should not happen because the | |
187 | * tick_do_timer_cpu never relinquishes. | |
5bb96226 | 188 | */ |
08ae95f4 NP |
189 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) { |
190 | #ifdef CONFIG_NO_HZ_FULL | |
191 | WARN_ON(tick_nohz_full_running); | |
192 | #endif | |
5bb96226 | 193 | tick_do_timer_cpu = cpu; |
08ae95f4 | 194 | } |
5bb96226 FW |
195 | #endif |
196 | ||
197 | /* Check, if the jiffies need an update */ | |
198 | if (tick_do_timer_cpu == cpu) | |
199 | tick_do_update_jiffies64(now); | |
ff7de620 | 200 | |
a1ff03cd FW |
201 | /* |
202 | * If jiffies update stalled for too long (timekeeper in stop_machine() | |
203 | * or VMEXIT'ed for several msecs), force an update. | |
204 | */ | |
205 | if (ts->last_tick_jiffies != jiffies) { | |
206 | ts->stalled_jiffies = 0; | |
207 | ts->last_tick_jiffies = READ_ONCE(jiffies); | |
208 | } else { | |
209 | if (++ts->stalled_jiffies == MAX_STALLED_JIFFIES) { | |
210 | tick_do_update_jiffies64(now); | |
211 | ts->stalled_jiffies = 0; | |
212 | ts->last_tick_jiffies = READ_ONCE(jiffies); | |
213 | } | |
214 | } | |
215 | ||
ff7de620 RW |
216 | if (ts->inidle) |
217 | ts->got_idle_tick = 1; | |
5bb96226 FW |
218 | } |
219 | ||
9e8f559b FW |
220 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
221 | { | |
3451d024 | 222 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
223 | /* |
224 | * When we are idle and the tick is stopped, we have to touch | |
225 | * the watchdog as we might not schedule for a really long | |
226 | * time. This happens on complete idle SMP systems while | |
227 | * waiting on the login prompt. We also increment the "start of | |
228 | * idle" jiffy stamp so the idle accounting adjustment we do | |
229 | * when we go busy again does not account too much ticks. | |
230 | */ | |
231 | if (ts->tick_stopped) { | |
03e0d461 | 232 | touch_softlockup_watchdog_sched(); |
9e8f559b FW |
233 | if (is_idle_task(current)) |
234 | ts->idle_jiffies++; | |
411fe24e FW |
235 | /* |
236 | * In case the current tick fired too early past its expected | |
237 | * expiration, make sure we don't bypass the next clock reprogramming | |
238 | * to the same deadline. | |
239 | */ | |
240 | ts->next_tick = 0; | |
9e8f559b | 241 | } |
94a57140 | 242 | #endif |
9e8f559b FW |
243 | update_process_times(user_mode(regs)); |
244 | profile_tick(CPU_PROFILING); | |
245 | } | |
7809998a | 246 | #endif |
9e8f559b | 247 | |
c5bfece2 | 248 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 249 | cpumask_var_t tick_nohz_full_mask; |
f268c373 | 250 | EXPORT_SYMBOL_GPL(tick_nohz_full_mask); |
73867dcd | 251 | bool tick_nohz_full_running; |
ae9e557b | 252 | EXPORT_SYMBOL_GPL(tick_nohz_full_running); |
f009a7a7 | 253 | static atomic_t tick_dep_mask; |
a831881b | 254 | |
f009a7a7 | 255 | static bool check_tick_dependency(atomic_t *dep) |
d027d45d | 256 | { |
f009a7a7 FW |
257 | int val = atomic_read(dep); |
258 | ||
259 | if (val & TICK_DEP_MASK_POSIX_TIMER) { | |
e6e6cc22 | 260 | trace_tick_stop(0, TICK_DEP_MASK_POSIX_TIMER); |
f009a7a7 | 261 | return true; |
d027d45d FW |
262 | } |
263 | ||
f009a7a7 | 264 | if (val & TICK_DEP_MASK_PERF_EVENTS) { |
e6e6cc22 | 265 | trace_tick_stop(0, TICK_DEP_MASK_PERF_EVENTS); |
f009a7a7 | 266 | return true; |
d027d45d FW |
267 | } |
268 | ||
f009a7a7 | 269 | if (val & TICK_DEP_MASK_SCHED) { |
e6e6cc22 | 270 | trace_tick_stop(0, TICK_DEP_MASK_SCHED); |
f009a7a7 | 271 | return true; |
d027d45d FW |
272 | } |
273 | ||
f009a7a7 | 274 | if (val & TICK_DEP_MASK_CLOCK_UNSTABLE) { |
e6e6cc22 | 275 | trace_tick_stop(0, TICK_DEP_MASK_CLOCK_UNSTABLE); |
f009a7a7 FW |
276 | return true; |
277 | } | |
278 | ||
01b4c399 FW |
279 | if (val & TICK_DEP_MASK_RCU) { |
280 | trace_tick_stop(0, TICK_DEP_MASK_RCU); | |
281 | return true; | |
282 | } | |
283 | ||
f009a7a7 | 284 | return false; |
d027d45d FW |
285 | } |
286 | ||
57ccdf44 | 287 | static bool can_stop_full_tick(int cpu, struct tick_sched *ts) |
9014c45d | 288 | { |
ebf3adba | 289 | lockdep_assert_irqs_disabled(); |
9014c45d | 290 | |
57ccdf44 WL |
291 | if (unlikely(!cpu_online(cpu))) |
292 | return false; | |
293 | ||
f009a7a7 | 294 | if (check_tick_dependency(&tick_dep_mask)) |
d027d45d | 295 | return false; |
d027d45d | 296 | |
f009a7a7 | 297 | if (check_tick_dependency(&ts->tick_dep_mask)) |
d027d45d | 298 | return false; |
d027d45d | 299 | |
f009a7a7 | 300 | if (check_tick_dependency(¤t->tick_dep_mask)) |
d027d45d | 301 | return false; |
d027d45d | 302 | |
f009a7a7 | 303 | if (check_tick_dependency(¤t->signal->tick_dep_mask)) |
d027d45d | 304 | return false; |
d027d45d | 305 | |
9014c45d FW |
306 | return true; |
307 | } | |
308 | ||
d027d45d | 309 | static void nohz_full_kick_func(struct irq_work *work) |
76c24fb0 | 310 | { |
73738a95 | 311 | /* Empty, the tick restart happens on tick_nohz_irq_exit() */ |
76c24fb0 FW |
312 | } |
313 | ||
7a9f50a0 PZ |
314 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = |
315 | IRQ_WORK_INIT_HARD(nohz_full_kick_func); | |
76c24fb0 | 316 | |
40bea039 FW |
317 | /* |
318 | * Kick this CPU if it's full dynticks in order to force it to | |
319 | * re-evaluate its dependency on the tick and restart it if necessary. | |
320 | * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(), | |
321 | * is NMI safe. | |
322 | */ | |
555e0c1e | 323 | static void tick_nohz_full_kick(void) |
40bea039 FW |
324 | { |
325 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
326 | return; | |
327 | ||
56e4dea8 | 328 | irq_work_queue(this_cpu_ptr(&nohz_full_kick_work)); |
40bea039 FW |
329 | } |
330 | ||
76c24fb0 | 331 | /* |
3d36aebc | 332 | * Kick the CPU if it's full dynticks in order to force it to |
76c24fb0 FW |
333 | * re-evaluate its dependency on the tick and restart it if necessary. |
334 | */ | |
3d36aebc | 335 | void tick_nohz_full_kick_cpu(int cpu) |
76c24fb0 | 336 | { |
3d36aebc FW |
337 | if (!tick_nohz_full_cpu(cpu)) |
338 | return; | |
339 | ||
340 | irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu); | |
76c24fb0 FW |
341 | } |
342 | ||
29721b85 FW |
343 | static void tick_nohz_kick_task(struct task_struct *tsk) |
344 | { | |
a1dfb631 MT |
345 | int cpu; |
346 | ||
347 | /* | |
348 | * If the task is not running, run_posix_cpu_timers() | |
349 | * has nothing to elapse, IPI can then be spared. | |
350 | * | |
351 | * activate_task() STORE p->tick_dep_mask | |
352 | * STORE p->on_rq | |
353 | * __schedule() (switch to task 'p') smp_mb() (atomic_fetch_or()) | |
354 | * LOCK rq->lock LOAD p->on_rq | |
355 | * smp_mb__after_spin_lock() | |
356 | * tick_nohz_task_switch() | |
357 | * LOAD p->tick_dep_mask | |
358 | */ | |
359 | if (!sched_task_on_rq(tsk)) | |
360 | return; | |
29721b85 FW |
361 | |
362 | /* | |
363 | * If the task concurrently migrates to another CPU, | |
364 | * we guarantee it sees the new tick dependency upon | |
365 | * schedule. | |
366 | * | |
29721b85 FW |
367 | * set_task_cpu(p, cpu); |
368 | * STORE p->cpu = @cpu | |
369 | * __schedule() (switch to task 'p') | |
370 | * LOCK rq->lock | |
371 | * smp_mb__after_spin_lock() STORE p->tick_dep_mask | |
372 | * tick_nohz_task_switch() smp_mb() (atomic_fetch_or()) | |
373 | * LOAD p->tick_dep_mask LOAD p->cpu | |
374 | */ | |
a1dfb631 | 375 | cpu = task_cpu(tsk); |
29721b85 FW |
376 | |
377 | preempt_disable(); | |
378 | if (cpu_online(cpu)) | |
379 | tick_nohz_full_kick_cpu(cpu); | |
380 | preempt_enable(); | |
381 | } | |
382 | ||
76c24fb0 FW |
383 | /* |
384 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
385 | * their dependency on the tick and restart it if necessary. | |
386 | */ | |
b7878300 | 387 | static void tick_nohz_full_kick_all(void) |
76c24fb0 | 388 | { |
8537bb95 FW |
389 | int cpu; |
390 | ||
73867dcd | 391 | if (!tick_nohz_full_running) |
76c24fb0 FW |
392 | return; |
393 | ||
394 | preempt_disable(); | |
8537bb95 FW |
395 | for_each_cpu_and(cpu, tick_nohz_full_mask, cpu_online_mask) |
396 | tick_nohz_full_kick_cpu(cpu); | |
76c24fb0 FW |
397 | preempt_enable(); |
398 | } | |
399 | ||
f009a7a7 | 400 | static void tick_nohz_dep_set_all(atomic_t *dep, |
d027d45d FW |
401 | enum tick_dep_bits bit) |
402 | { | |
f009a7a7 | 403 | int prev; |
d027d45d | 404 | |
a1cc5bcf | 405 | prev = atomic_fetch_or(BIT(bit), dep); |
d027d45d FW |
406 | if (!prev) |
407 | tick_nohz_full_kick_all(); | |
408 | } | |
409 | ||
410 | /* | |
411 | * Set a global tick dependency. Used by perf events that rely on freq and | |
412 | * by unstable clock. | |
413 | */ | |
414 | void tick_nohz_dep_set(enum tick_dep_bits bit) | |
415 | { | |
416 | tick_nohz_dep_set_all(&tick_dep_mask, bit); | |
417 | } | |
418 | ||
419 | void tick_nohz_dep_clear(enum tick_dep_bits bit) | |
420 | { | |
f009a7a7 | 421 | atomic_andnot(BIT(bit), &tick_dep_mask); |
d027d45d FW |
422 | } |
423 | ||
424 | /* | |
425 | * Set per-CPU tick dependency. Used by scheduler and perf events in order to | |
426 | * manage events throttling. | |
427 | */ | |
428 | void tick_nohz_dep_set_cpu(int cpu, enum tick_dep_bits bit) | |
429 | { | |
f009a7a7 | 430 | int prev; |
d027d45d FW |
431 | struct tick_sched *ts; |
432 | ||
433 | ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
434 | ||
a1cc5bcf | 435 | prev = atomic_fetch_or(BIT(bit), &ts->tick_dep_mask); |
d027d45d FW |
436 | if (!prev) { |
437 | preempt_disable(); | |
438 | /* Perf needs local kick that is NMI safe */ | |
439 | if (cpu == smp_processor_id()) { | |
440 | tick_nohz_full_kick(); | |
441 | } else { | |
442 | /* Remote irq work not NMI-safe */ | |
443 | if (!WARN_ON_ONCE(in_nmi())) | |
444 | tick_nohz_full_kick_cpu(cpu); | |
445 | } | |
446 | preempt_enable(); | |
447 | } | |
448 | } | |
01b4c399 | 449 | EXPORT_SYMBOL_GPL(tick_nohz_dep_set_cpu); |
d027d45d FW |
450 | |
451 | void tick_nohz_dep_clear_cpu(int cpu, enum tick_dep_bits bit) | |
452 | { | |
453 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
454 | ||
f009a7a7 | 455 | atomic_andnot(BIT(bit), &ts->tick_dep_mask); |
d027d45d | 456 | } |
01b4c399 | 457 | EXPORT_SYMBOL_GPL(tick_nohz_dep_clear_cpu); |
d027d45d FW |
458 | |
459 | /* | |
3c8920e2 FW |
460 | * Set a per-task tick dependency. RCU need this. Also posix CPU timers |
461 | * in order to elapse per task timers. | |
d027d45d FW |
462 | */ |
463 | void tick_nohz_dep_set_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
464 | { | |
29721b85 FW |
465 | if (!atomic_fetch_or(BIT(bit), &tsk->tick_dep_mask)) |
466 | tick_nohz_kick_task(tsk); | |
d027d45d | 467 | } |
ae9e557b | 468 | EXPORT_SYMBOL_GPL(tick_nohz_dep_set_task); |
d027d45d FW |
469 | |
470 | void tick_nohz_dep_clear_task(struct task_struct *tsk, enum tick_dep_bits bit) | |
471 | { | |
f009a7a7 | 472 | atomic_andnot(BIT(bit), &tsk->tick_dep_mask); |
d027d45d | 473 | } |
ae9e557b | 474 | EXPORT_SYMBOL_GPL(tick_nohz_dep_clear_task); |
d027d45d FW |
475 | |
476 | /* | |
477 | * Set a per-taskgroup tick dependency. Posix CPU timers need this in order to elapse | |
478 | * per process timers. | |
479 | */ | |
1e4ca26d MT |
480 | void tick_nohz_dep_set_signal(struct task_struct *tsk, |
481 | enum tick_dep_bits bit) | |
d027d45d | 482 | { |
1e4ca26d MT |
483 | int prev; |
484 | struct signal_struct *sig = tsk->signal; | |
485 | ||
486 | prev = atomic_fetch_or(BIT(bit), &sig->tick_dep_mask); | |
487 | if (!prev) { | |
488 | struct task_struct *t; | |
489 | ||
490 | lockdep_assert_held(&tsk->sighand->siglock); | |
491 | __for_each_thread(sig, t) | |
492 | tick_nohz_kick_task(t); | |
493 | } | |
d027d45d FW |
494 | } |
495 | ||
496 | void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bit) | |
497 | { | |
f009a7a7 | 498 | atomic_andnot(BIT(bit), &sig->tick_dep_mask); |
d027d45d FW |
499 | } |
500 | ||
99e5ada9 FW |
501 | /* |
502 | * Re-evaluate the need for the tick as we switch the current task. | |
503 | * It might need the tick due to per task/process properties: | |
0de7611a | 504 | * perf events, posix CPU timers, ... |
99e5ada9 | 505 | */ |
de734f89 | 506 | void __tick_nohz_task_switch(void) |
99e5ada9 | 507 | { |
d027d45d | 508 | struct tick_sched *ts; |
99e5ada9 | 509 | |
6296ace4 | 510 | if (!tick_nohz_full_cpu(smp_processor_id())) |
0fdcccfa | 511 | return; |
6296ace4 | 512 | |
d027d45d | 513 | ts = this_cpu_ptr(&tick_cpu_sched); |
99e5ada9 | 514 | |
d027d45d | 515 | if (ts->tick_stopped) { |
f009a7a7 FW |
516 | if (atomic_read(¤t->tick_dep_mask) || |
517 | atomic_read(¤t->signal->tick_dep_mask)) | |
d027d45d FW |
518 | tick_nohz_full_kick(); |
519 | } | |
99e5ada9 FW |
520 | } |
521 | ||
6f1982fe FW |
522 | /* Get the boot-time nohz CPU list from the kernel parameters. */ |
523 | void __init tick_nohz_full_setup(cpumask_var_t cpumask) | |
a831881b | 524 | { |
73867dcd | 525 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
6f1982fe | 526 | cpumask_copy(tick_nohz_full_mask, cpumask); |
73867dcd | 527 | tick_nohz_full_running = true; |
a831881b | 528 | } |
ae9e557b | 529 | EXPORT_SYMBOL_GPL(tick_nohz_full_setup); |
a831881b | 530 | |
31eff243 | 531 | static int tick_nohz_cpu_down(unsigned int cpu) |
a382bf93 | 532 | { |
31eff243 | 533 | /* |
08ae95f4 NP |
534 | * The tick_do_timer_cpu CPU handles housekeeping duty (unbound |
535 | * timers, workqueues, timekeeping, ...) on behalf of full dynticks | |
31eff243 SAS |
536 | * CPUs. It must remain online when nohz full is enabled. |
537 | */ | |
538 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) | |
539 | return -EBUSY; | |
540 | return 0; | |
a382bf93 FW |
541 | } |
542 | ||
d1e43fa5 | 543 | void __init tick_nohz_init(void) |
a831881b | 544 | { |
31eff243 | 545 | int cpu, ret; |
d1e43fa5 | 546 | |
a7c8655b PM |
547 | if (!tick_nohz_full_running) |
548 | return; | |
d1e43fa5 | 549 | |
9b01f5bf FW |
550 | /* |
551 | * Full dynticks uses irq work to drive the tick rescheduling on safe | |
552 | * locking contexts. But then we need irq work to raise its own | |
553 | * interrupts to avoid circular dependency on the tick | |
554 | */ | |
555 | if (!arch_irq_work_has_interrupt()) { | |
a395d6a7 | 556 | pr_warn("NO_HZ: Can't run full dynticks because arch doesn't support irq work self-IPIs\n"); |
9b01f5bf | 557 | cpumask_clear(tick_nohz_full_mask); |
9b01f5bf FW |
558 | tick_nohz_full_running = false; |
559 | return; | |
560 | } | |
561 | ||
08ae95f4 NP |
562 | if (IS_ENABLED(CONFIG_PM_SLEEP_SMP) && |
563 | !IS_ENABLED(CONFIG_PM_SLEEP_SMP_NONZERO_CPU)) { | |
564 | cpu = smp_processor_id(); | |
4327b15f | 565 | |
08ae95f4 NP |
566 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { |
567 | pr_warn("NO_HZ: Clearing %d from nohz_full range " | |
568 | "for timekeeping\n", cpu); | |
569 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); | |
570 | } | |
4327b15f FW |
571 | } |
572 | ||
73867dcd | 573 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
574 | context_tracking_cpu_set(cpu); |
575 | ||
31eff243 SAS |
576 | ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, |
577 | "kernel/nohz:predown", NULL, | |
578 | tick_nohz_cpu_down); | |
579 | WARN_ON(ret < 0); | |
ffda22c1 TH |
580 | pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n", |
581 | cpumask_pr_args(tick_nohz_full_mask)); | |
a831881b | 582 | } |
a831881b FW |
583 | #endif |
584 | ||
79bf2bb3 TG |
585 | /* |
586 | * NOHZ - aka dynamic tick functionality | |
587 | */ | |
3451d024 | 588 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
589 | /* |
590 | * NO HZ enabled ? | |
591 | */ | |
4cc7ecb7 | 592 | bool tick_nohz_enabled __read_mostly = true; |
bc7a34b8 | 593 | unsigned long tick_nohz_active __read_mostly; |
79bf2bb3 TG |
594 | /* |
595 | * Enable / Disable tickless mode | |
596 | */ | |
597 | static int __init setup_tick_nohz(char *str) | |
598 | { | |
4cc7ecb7 | 599 | return (kstrtobool(str, &tick_nohz_enabled) == 0); |
79bf2bb3 TG |
600 | } |
601 | ||
602 | __setup("nohz=", setup_tick_nohz); | |
603 | ||
a3642983 | 604 | bool tick_nohz_tick_stopped(void) |
c1797baf | 605 | { |
2bc629a6 FW |
606 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
607 | ||
608 | return ts->tick_stopped; | |
c1797baf TG |
609 | } |
610 | ||
22ab8bc0 FW |
611 | bool tick_nohz_tick_stopped_cpu(int cpu) |
612 | { | |
613 | struct tick_sched *ts = per_cpu_ptr(&tick_cpu_sched, cpu); | |
614 | ||
615 | return ts->tick_stopped; | |
616 | } | |
617 | ||
79bf2bb3 TG |
618 | /** |
619 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
620 | * | |
621 | * Called from interrupt entry when the CPU was idle | |
622 | * | |
623 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
624 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
0de7611a IM |
625 | * value. We do this unconditionally on any CPU, as we don't know whether the |
626 | * CPU, which has the update task assigned is in a long sleep. | |
79bf2bb3 | 627 | */ |
eed3b9cf | 628 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 629 | { |
79bf2bb3 | 630 | unsigned long flags; |
79bf2bb3 | 631 | |
e8fcaa5c | 632 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
633 | |
634 | local_irq_save(flags); | |
635 | tick_do_update_jiffies64(now); | |
636 | local_irq_restore(flags); | |
02ff3755 | 637 | |
03e0d461 | 638 | touch_softlockup_watchdog_sched(); |
79bf2bb3 TG |
639 | } |
640 | ||
595aac48 | 641 | /* |
0de7611a | 642 | * Updates the per-CPU time idle statistics counters |
595aac48 | 643 | */ |
8d63bf94 | 644 | static void |
8c215bd3 | 645 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 646 | { |
eed3b9cf | 647 | ktime_t delta; |
6378ddb5 | 648 | |
595aac48 AV |
649 | if (ts->idle_active) { |
650 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 651 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 652 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
653 | else |
654 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 655 | ts->idle_entrytime = now; |
595aac48 | 656 | } |
8d63bf94 | 657 | |
e0e37c20 | 658 | if (last_update_time) |
8d63bf94 AV |
659 | *last_update_time = ktime_to_us(now); |
660 | ||
595aac48 AV |
661 | } |
662 | ||
e8fcaa5c | 663 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 664 | { |
e8fcaa5c | 665 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 666 | ts->idle_active = 0; |
56c7426b | 667 | |
ac1e843f | 668 | sched_clock_idle_wakeup_event(); |
6378ddb5 VP |
669 | } |
670 | ||
0e776768 | 671 | static void tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 672 | { |
0e776768 | 673 | ts->idle_entrytime = ktime_get(); |
6378ddb5 | 674 | ts->idle_active = 1; |
56c7426b | 675 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
676 | } |
677 | ||
b1f724c3 | 678 | /** |
0de7611a | 679 | * get_cpu_idle_time_us - get the total idle time of a CPU |
b1f724c3 | 680 | * @cpu: CPU number to query |
09a1d34f MH |
681 | * @last_update_time: variable to store update time in. Do not update |
682 | * counters if NULL. | |
b1f724c3 | 683 | * |
6168f8ed | 684 | * Return the cumulative idle time (since boot) for a given |
6beea0cd | 685 | * CPU, in microseconds. |
b1f724c3 AV |
686 | * |
687 | * This time is measured via accounting rather than sampling, | |
688 | * and is as accurate as ktime_get() is. | |
689 | * | |
690 | * This function returns -1 if NOHZ is not enabled. | |
691 | */ | |
6378ddb5 VP |
692 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
693 | { | |
694 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 695 | ktime_t now, idle; |
6378ddb5 | 696 | |
d689fe22 | 697 | if (!tick_nohz_active) |
8083e4ad | 698 | return -1; |
699 | ||
09a1d34f MH |
700 | now = ktime_get(); |
701 | if (last_update_time) { | |
702 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
703 | idle = ts->idle_sleeptime; | |
704 | } else { | |
705 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
706 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
707 | ||
708 | idle = ktime_add(ts->idle_sleeptime, delta); | |
709 | } else { | |
710 | idle = ts->idle_sleeptime; | |
711 | } | |
712 | } | |
713 | ||
714 | return ktime_to_us(idle); | |
8083e4ad | 715 | |
6378ddb5 | 716 | } |
8083e4ad | 717 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 718 | |
6beea0cd | 719 | /** |
0de7611a | 720 | * get_cpu_iowait_time_us - get the total iowait time of a CPU |
0224cf4c | 721 | * @cpu: CPU number to query |
09a1d34f MH |
722 | * @last_update_time: variable to store update time in. Do not update |
723 | * counters if NULL. | |
0224cf4c | 724 | * |
6168f8ed | 725 | * Return the cumulative iowait time (since boot) for a given |
0224cf4c AV |
726 | * CPU, in microseconds. |
727 | * | |
728 | * This time is measured via accounting rather than sampling, | |
729 | * and is as accurate as ktime_get() is. | |
730 | * | |
731 | * This function returns -1 if NOHZ is not enabled. | |
732 | */ | |
733 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
734 | { | |
735 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 736 | ktime_t now, iowait; |
0224cf4c | 737 | |
d689fe22 | 738 | if (!tick_nohz_active) |
0224cf4c AV |
739 | return -1; |
740 | ||
09a1d34f MH |
741 | now = ktime_get(); |
742 | if (last_update_time) { | |
743 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
744 | iowait = ts->iowait_sleeptime; | |
745 | } else { | |
746 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
747 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 748 | |
09a1d34f MH |
749 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
750 | } else { | |
751 | iowait = ts->iowait_sleeptime; | |
752 | } | |
753 | } | |
0224cf4c | 754 | |
09a1d34f | 755 | return ktime_to_us(iowait); |
0224cf4c AV |
756 | } |
757 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
758 | ||
0ff53d09 TG |
759 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
760 | { | |
761 | hrtimer_cancel(&ts->sched_timer); | |
762 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); | |
763 | ||
764 | /* Forward the time to expire in the future */ | |
b9965449 | 765 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
0ff53d09 | 766 | |
902a9f9c SAS |
767 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
768 | hrtimer_start_expires(&ts->sched_timer, | |
769 | HRTIMER_MODE_ABS_PINNED_HARD); | |
770 | } else { | |
0ff53d09 | 771 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
902a9f9c | 772 | } |
411fe24e FW |
773 | |
774 | /* | |
775 | * Reset to make sure next tick stop doesn't get fooled by past | |
776 | * cached clock deadline. | |
777 | */ | |
778 | ts->next_tick = 0; | |
0ff53d09 TG |
779 | } |
780 | ||
5d62c183 TG |
781 | static inline bool local_timer_softirq_pending(void) |
782 | { | |
80d20d35 | 783 | return local_softirq_pending() & BIT(TIMER_SOFTIRQ); |
5d62c183 TG |
784 | } |
785 | ||
23a8d888 | 786 | static ktime_t tick_nohz_next_event(struct tick_sched *ts, int cpu) |
79bf2bb3 | 787 | { |
c1ad348b | 788 | u64 basemono, next_tick, next_tmr, next_rcu, delta, expires; |
e1e41b6c RV |
789 | unsigned long basejiff; |
790 | unsigned int seq; | |
855a0fc3 | 791 | |
79bf2bb3 TG |
792 | /* Read jiffies and the time when jiffies were updated last */ |
793 | do { | |
e5d4d175 | 794 | seq = read_seqcount_begin(&jiffies_seq); |
2456e855 | 795 | basemono = last_jiffies_update; |
c1ad348b | 796 | basejiff = jiffies; |
e5d4d175 | 797 | } while (read_seqcount_retry(&jiffies_seq, seq)); |
c1ad348b | 798 | ts->last_jiffies = basejiff; |
23a8d888 | 799 | ts->timer_expires_base = basemono; |
79bf2bb3 | 800 | |
5d62c183 TG |
801 | /* |
802 | * Keep the periodic tick, when RCU, architecture or irq_work | |
803 | * requests it. | |
804 | * Aside of that check whether the local timer softirq is | |
805 | * pending. If so its a bad idea to call get_next_timer_interrupt() | |
806 | * because there is an already expired timer, so it will request | |
4bf07f65 | 807 | * immediate expiry, which rearms the hardware timer with a |
5d62c183 TG |
808 | * minimal delta which brings us back to this place |
809 | * immediately. Lather, rinse and repeat... | |
810 | */ | |
811 | if (rcu_needs_cpu(basemono, &next_rcu) || arch_needs_cpu() || | |
812 | irq_work_needs_cpu() || local_timer_softirq_pending()) { | |
c1ad348b | 813 | next_tick = basemono + TICK_NSEC; |
3c5d92a0 | 814 | } else { |
c1ad348b TG |
815 | /* |
816 | * Get the next pending timer. If high resolution | |
817 | * timers are enabled this only takes the timer wheel | |
818 | * timers into account. If high resolution timers are | |
819 | * disabled this also looks at the next expiring | |
820 | * hrtimer. | |
821 | */ | |
822 | next_tmr = get_next_timer_interrupt(basejiff, basemono); | |
823 | ts->next_timer = next_tmr; | |
824 | /* Take the next rcu event into account */ | |
825 | next_tick = next_rcu < next_tmr ? next_rcu : next_tmr; | |
3c5d92a0 | 826 | } |
47aa8b6c | 827 | |
c1ad348b TG |
828 | /* |
829 | * If the tick is due in the next period, keep it ticking or | |
82bbe34b | 830 | * force prod the timer. |
c1ad348b TG |
831 | */ |
832 | delta = next_tick - basemono; | |
833 | if (delta <= (u64)TICK_NSEC) { | |
a683f390 TG |
834 | /* |
835 | * Tell the timer code that the base is not idle, i.e. undo | |
836 | * the effect of get_next_timer_interrupt(): | |
837 | */ | |
838 | timer_clear_idle(); | |
82bbe34b PZ |
839 | /* |
840 | * We've not stopped the tick yet, and there's a timer in the | |
841 | * next period, so no point in stopping it either, bail. | |
842 | */ | |
f99973e1 | 843 | if (!ts->tick_stopped) { |
23a8d888 | 844 | ts->timer_expires = 0; |
157d29e1 TG |
845 | goto out; |
846 | } | |
847 | } | |
848 | ||
23a8d888 RW |
849 | /* |
850 | * If this CPU is the one which had the do_timer() duty last, we limit | |
851 | * the sleep time to the timekeeping max_deferment value. | |
852 | * Otherwise we can sleep as long as we want. | |
853 | */ | |
854 | delta = timekeeping_max_deferment(); | |
855 | if (cpu != tick_do_timer_cpu && | |
856 | (tick_do_timer_cpu != TICK_DO_TIMER_NONE || !ts->do_timer_last)) | |
857 | delta = KTIME_MAX; | |
858 | ||
859 | /* Calculate the next expiry time */ | |
860 | if (delta < (KTIME_MAX - basemono)) | |
861 | expires = basemono + delta; | |
862 | else | |
863 | expires = KTIME_MAX; | |
864 | ||
865 | ts->timer_expires = min_t(u64, expires, next_tick); | |
866 | ||
867 | out: | |
868 | return ts->timer_expires; | |
869 | } | |
870 | ||
871 | static void tick_nohz_stop_tick(struct tick_sched *ts, int cpu) | |
872 | { | |
873 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); | |
874 | u64 basemono = ts->timer_expires_base; | |
875 | u64 expires = ts->timer_expires; | |
876 | ktime_t tick = expires; | |
877 | ||
878 | /* Make sure we won't be trying to stop it twice in a row. */ | |
879 | ts->timer_expires_base = 0; | |
880 | ||
79bf2bb3 | 881 | /* |
0de7611a IM |
882 | * If this CPU is the one which updates jiffies, then give up |
883 | * the assignment and let it be taken by the CPU which runs | |
884 | * the tick timer next, which might be this CPU as well. If we | |
157d29e1 TG |
885 | * don't drop this here the jiffies might be stale and |
886 | * do_timer() never invoked. Keep track of the fact that it | |
23a8d888 | 887 | * was the one which had the do_timer() duty last. |
79bf2bb3 | 888 | */ |
157d29e1 TG |
889 | if (cpu == tick_do_timer_cpu) { |
890 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
891 | ts->do_timer_last = 1; | |
892 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
157d29e1 | 893 | ts->do_timer_last = 0; |
157d29e1 | 894 | } |
27185016 | 895 | |
157d29e1 | 896 | /* Skip reprogram of event if its not changed */ |
411fe24e FW |
897 | if (ts->tick_stopped && (expires == ts->next_tick)) { |
898 | /* Sanity check: make sure clockevent is actually programmed */ | |
d4af6d93 | 899 | if (tick == KTIME_MAX || ts->next_tick == hrtimer_get_expires(&ts->sched_timer)) |
23a8d888 | 900 | return; |
411fe24e FW |
901 | |
902 | WARN_ON_ONCE(1); | |
903 | printk_once("basemono: %llu ts->next_tick: %llu dev->next_event: %llu timer->active: %d timer->expires: %llu\n", | |
904 | basemono, ts->next_tick, dev->next_event, | |
905 | hrtimer_active(&ts->sched_timer), hrtimer_get_expires(&ts->sched_timer)); | |
ce6cf9a1 | 906 | } |
84bf1bcc | 907 | |
157d29e1 TG |
908 | /* |
909 | * nohz_stop_sched_tick can be called several times before | |
910 | * the nohz_restart_sched_tick is called. This happens when | |
911 | * interrupts arrive which do not cause a reschedule. In the | |
912 | * first call we save the current tick time, so we can restart | |
913 | * the scheduler tick in nohz_restart_sched_tick. | |
914 | */ | |
915 | if (!ts->tick_stopped) { | |
3c85d6db | 916 | calc_load_nohz_start(); |
62cb1188 | 917 | quiet_vmstat(); |
d3ed7824 | 918 | |
157d29e1 TG |
919 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
920 | ts->tick_stopped = 1; | |
e6e6cc22 | 921 | trace_tick_stop(1, TICK_DEP_MASK_NONE); |
157d29e1 | 922 | } |
eaad084b | 923 | |
411fe24e FW |
924 | ts->next_tick = tick; |
925 | ||
157d29e1 | 926 | /* |
c1ad348b TG |
927 | * If the expiration time == KTIME_MAX, then we simply stop |
928 | * the tick timer. | |
157d29e1 | 929 | */ |
c1ad348b | 930 | if (unlikely(expires == KTIME_MAX)) { |
157d29e1 TG |
931 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
932 | hrtimer_cancel(&ts->sched_timer); | |
23a8d888 | 933 | return; |
79bf2bb3 | 934 | } |
0ff53d09 | 935 | |
1f71addd | 936 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
902a9f9c SAS |
937 | hrtimer_start(&ts->sched_timer, tick, |
938 | HRTIMER_MODE_ABS_PINNED_HARD); | |
1f71addd TG |
939 | } else { |
940 | hrtimer_set_expires(&ts->sched_timer, tick); | |
c1ad348b | 941 | tick_program_event(tick, 1); |
1f71addd | 942 | } |
280f0677 FW |
943 | } |
944 | ||
23a8d888 RW |
945 | static void tick_nohz_retain_tick(struct tick_sched *ts) |
946 | { | |
947 | ts->timer_expires_base = 0; | |
948 | } | |
949 | ||
950 | #ifdef CONFIG_NO_HZ_FULL | |
951 | static void tick_nohz_stop_sched_tick(struct tick_sched *ts, int cpu) | |
952 | { | |
953 | if (tick_nohz_next_event(ts, cpu)) | |
954 | tick_nohz_stop_tick(ts, cpu); | |
955 | else | |
956 | tick_nohz_retain_tick(ts); | |
957 | } | |
958 | #endif /* CONFIG_NO_HZ_FULL */ | |
959 | ||
1f41906a | 960 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
59d2c7ca FW |
961 | { |
962 | /* Update jiffies first */ | |
963 | tick_do_update_jiffies64(now); | |
59d2c7ca | 964 | |
a683f390 TG |
965 | /* |
966 | * Clear the timer idle flag, so we avoid IPIs on remote queueing and | |
967 | * the clock forward checks in the enqueue path: | |
968 | */ | |
969 | timer_clear_idle(); | |
970 | ||
3c85d6db | 971 | calc_load_nohz_stop(); |
03e0d461 | 972 | touch_softlockup_watchdog_sched(); |
59d2c7ca FW |
973 | /* |
974 | * Cancel the scheduled timer and restore the tick | |
975 | */ | |
976 | ts->tick_stopped = 0; | |
59d2c7ca FW |
977 | tick_nohz_restart(ts, now); |
978 | } | |
73738a95 | 979 | |
a5183862 YY |
980 | static void __tick_nohz_full_update_tick(struct tick_sched *ts, |
981 | ktime_t now) | |
5811d996 FW |
982 | { |
983 | #ifdef CONFIG_NO_HZ_FULL | |
e9a2eb40 | 984 | int cpu = smp_processor_id(); |
5811d996 | 985 | |
a5183862 YY |
986 | if (can_stop_full_tick(cpu, ts)) |
987 | tick_nohz_stop_sched_tick(ts, cpu); | |
988 | else if (ts->tick_stopped) | |
989 | tick_nohz_restart_sched_tick(ts, now); | |
990 | #endif | |
991 | } | |
992 | ||
993 | static void tick_nohz_full_update_tick(struct tick_sched *ts) | |
994 | { | |
995 | if (!tick_nohz_full_cpu(smp_processor_id())) | |
e9a2eb40 | 996 | return; |
5811d996 | 997 | |
e9a2eb40 AS |
998 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) |
999 | return; | |
5811d996 | 1000 | |
a5183862 | 1001 | __tick_nohz_full_update_tick(ts, ktime_get()); |
5811d996 FW |
1002 | } |
1003 | ||
5b39939a FW |
1004 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
1005 | { | |
1006 | /* | |
0de7611a | 1007 | * If this CPU is offline and it is the one which updates |
5b39939a | 1008 | * jiffies, then give up the assignment and let it be taken by |
0de7611a | 1009 | * the CPU which runs the tick timer next. If we don't drop |
5b39939a FW |
1010 | * this here the jiffies might be stale and do_timer() never |
1011 | * invoked. | |
1012 | */ | |
1013 | if (unlikely(!cpu_online(cpu))) { | |
1014 | if (cpu == tick_do_timer_cpu) | |
1015 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
411fe24e FW |
1016 | /* |
1017 | * Make sure the CPU doesn't get fooled by obsolete tick | |
1018 | * deadline if it comes back online later. | |
1019 | */ | |
1020 | ts->next_tick = 0; | |
f7ea0fd6 | 1021 | return false; |
5b39939a FW |
1022 | } |
1023 | ||
23a8d888 | 1024 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) |
5b39939a FW |
1025 | return false; |
1026 | ||
1027 | if (need_resched()) | |
1028 | return false; | |
1029 | ||
d59e0ba1 | 1030 | if (unlikely(local_softirq_pending())) { |
5b39939a FW |
1031 | static int ratelimit; |
1032 | ||
47c218dc | 1033 | if (ratelimit < 10 && !local_bh_blocked() && |
803b0eba | 1034 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { |
bca37119 | 1035 | pr_warn("NOHZ tick-stop error: Non-RCU local softirq work is pending, handler #%02x!!!\n", |
cfea7d7e | 1036 | (unsigned int) local_softirq_pending()); |
5b39939a FW |
1037 | ratelimit++; |
1038 | } | |
1039 | return false; | |
1040 | } | |
1041 | ||
460775df | 1042 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
1043 | /* |
1044 | * Keep the tick alive to guarantee timekeeping progression | |
1045 | * if there are full dynticks CPUs around | |
1046 | */ | |
1047 | if (tick_do_timer_cpu == cpu) | |
1048 | return false; | |
08ae95f4 NP |
1049 | |
1050 | /* Should not happen for nohz-full */ | |
1051 | if (WARN_ON_ONCE(tick_do_timer_cpu == TICK_DO_TIMER_NONE)) | |
a382bf93 FW |
1052 | return false; |
1053 | } | |
1054 | ||
5b39939a FW |
1055 | return true; |
1056 | } | |
1057 | ||
0e776768 | 1058 | static void __tick_nohz_idle_stop_tick(struct tick_sched *ts) |
19f5f736 | 1059 | { |
0e776768 | 1060 | ktime_t expires; |
5b39939a | 1061 | int cpu = smp_processor_id(); |
19f5f736 | 1062 | |
554c8aa8 RW |
1063 | /* |
1064 | * If tick_nohz_get_sleep_length() ran tick_nohz_next_event(), the | |
1065 | * tick timer expiration time is known already. | |
1066 | */ | |
1067 | if (ts->timer_expires_base) | |
1068 | expires = ts->timer_expires; | |
1069 | else if (can_stop_idle_tick(cpu, ts)) | |
1070 | expires = tick_nohz_next_event(ts, cpu); | |
1071 | else | |
1072 | return; | |
23a8d888 RW |
1073 | |
1074 | ts->idle_calls++; | |
08d07259 | 1075 | |
23a8d888 | 1076 | if (expires > 0LL) { |
5b39939a FW |
1077 | int was_stopped = ts->tick_stopped; |
1078 | ||
23a8d888 | 1079 | tick_nohz_stop_tick(ts, cpu); |
84bf1bcc | 1080 | |
23a8d888 RW |
1081 | ts->idle_sleeps++; |
1082 | ts->idle_expires = expires; | |
5b39939a | 1083 | |
a0db971e | 1084 | if (!was_stopped && ts->tick_stopped) { |
5b39939a | 1085 | ts->idle_jiffies = ts->last_jiffies; |
a0db971e FW |
1086 | nohz_balance_enter_idle(cpu); |
1087 | } | |
23a8d888 RW |
1088 | } else { |
1089 | tick_nohz_retain_tick(ts); | |
5b39939a | 1090 | } |
280f0677 FW |
1091 | } |
1092 | ||
1093 | /** | |
0e776768 | 1094 | * tick_nohz_idle_stop_tick - stop the idle tick from the idle task |
280f0677 FW |
1095 | * |
1096 | * When the next event is more than a tick into the future, stop the idle tick | |
0e776768 RW |
1097 | */ |
1098 | void tick_nohz_idle_stop_tick(void) | |
1099 | { | |
1100 | __tick_nohz_idle_stop_tick(this_cpu_ptr(&tick_cpu_sched)); | |
1101 | } | |
1102 | ||
554c8aa8 RW |
1103 | void tick_nohz_idle_retain_tick(void) |
1104 | { | |
1105 | tick_nohz_retain_tick(this_cpu_ptr(&tick_cpu_sched)); | |
1106 | /* | |
1107 | * Undo the effect of get_next_timer_interrupt() called from | |
1108 | * tick_nohz_next_event(). | |
1109 | */ | |
1110 | timer_clear_idle(); | |
1111 | } | |
1112 | ||
0e776768 RW |
1113 | /** |
1114 | * tick_nohz_idle_enter - prepare for entering idle on the current CPU | |
2bbb6817 | 1115 | * |
0e776768 | 1116 | * Called when we start the idle loop. |
280f0677 | 1117 | */ |
1268fbc7 | 1118 | void tick_nohz_idle_enter(void) |
280f0677 FW |
1119 | { |
1120 | struct tick_sched *ts; | |
1121 | ||
ebf3adba | 1122 | lockdep_assert_irqs_enabled(); |
0db49b72 | 1123 | |
1268fbc7 FW |
1124 | local_irq_disable(); |
1125 | ||
22127e93 | 1126 | ts = this_cpu_ptr(&tick_cpu_sched); |
23a8d888 RW |
1127 | |
1128 | WARN_ON_ONCE(ts->timer_expires_base); | |
1129 | ||
280f0677 | 1130 | ts->inidle = 1; |
0e776768 | 1131 | tick_nohz_start_idle(ts); |
1268fbc7 FW |
1132 | |
1133 | local_irq_enable(); | |
280f0677 FW |
1134 | } |
1135 | ||
1136 | /** | |
1137 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
1138 | * | |
1139 | * When an interrupt fires while we are idle and it doesn't cause | |
1140 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
1141 | * an RCU callback, etc... | |
1142 | * So we need to re-calculate and reprogram the next tick event. | |
1143 | */ | |
1144 | void tick_nohz_irq_exit(void) | |
1145 | { | |
22127e93 | 1146 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
280f0677 | 1147 | |
14851912 | 1148 | if (ts->inidle) |
0e776768 | 1149 | tick_nohz_start_idle(ts); |
14851912 | 1150 | else |
73738a95 | 1151 | tick_nohz_full_update_tick(ts); |
79bf2bb3 TG |
1152 | } |
1153 | ||
4f86d3a8 | 1154 | /** |
45f1ff59 RW |
1155 | * tick_nohz_idle_got_tick - Check whether or not the tick handler has run |
1156 | */ | |
1157 | bool tick_nohz_idle_got_tick(void) | |
1158 | { | |
1159 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1160 | ||
2bc629a6 FW |
1161 | if (ts->got_idle_tick) { |
1162 | ts->got_idle_tick = 0; | |
45f1ff59 RW |
1163 | return true; |
1164 | } | |
1165 | return false; | |
1166 | } | |
1167 | ||
6f9b83ac UH |
1168 | /** |
1169 | * tick_nohz_get_next_hrtimer - return the next expiration time for the hrtimer | |
1170 | * or the tick, whatever that expires first. Note that, if the tick has been | |
1171 | * stopped, it returns the next hrtimer. | |
1172 | * | |
1173 | * Called from power state control code with interrupts disabled | |
1174 | */ | |
1175 | ktime_t tick_nohz_get_next_hrtimer(void) | |
1176 | { | |
1177 | return __this_cpu_read(tick_cpu_device.evtdev)->next_event; | |
1178 | } | |
1179 | ||
4f86d3a8 | 1180 | /** |
554c8aa8 | 1181 | * tick_nohz_get_sleep_length - return the expected length of the current sleep |
296bb1e5 | 1182 | * @delta_next: duration until the next event if the tick cannot be stopped |
4f86d3a8 | 1183 | * |
4c81cb7e RW |
1184 | * Called from power state control code with interrupts disabled. |
1185 | * | |
1186 | * The return value of this function and/or the value returned by it through the | |
1187 | * @delta_next pointer can be negative which must be taken into account by its | |
1188 | * callers. | |
4f86d3a8 | 1189 | */ |
296bb1e5 | 1190 | ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next) |
4f86d3a8 | 1191 | { |
554c8aa8 | 1192 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
22127e93 | 1193 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
554c8aa8 RW |
1194 | int cpu = smp_processor_id(); |
1195 | /* | |
1196 | * The idle entry time is expected to be a sufficient approximation of | |
1197 | * the current time at this point. | |
1198 | */ | |
1199 | ktime_t now = ts->idle_entrytime; | |
1200 | ktime_t next_event; | |
1201 | ||
1202 | WARN_ON_ONCE(!ts->inidle); | |
1203 | ||
296bb1e5 RW |
1204 | *delta_next = ktime_sub(dev->next_event, now); |
1205 | ||
554c8aa8 | 1206 | if (!can_stop_idle_tick(cpu, ts)) |
296bb1e5 | 1207 | return *delta_next; |
554c8aa8 RW |
1208 | |
1209 | next_event = tick_nohz_next_event(ts, cpu); | |
1210 | if (!next_event) | |
296bb1e5 | 1211 | return *delta_next; |
554c8aa8 RW |
1212 | |
1213 | /* | |
1214 | * If the next highres timer to expire is earlier than next_event, the | |
1215 | * idle governor needs to know that. | |
1216 | */ | |
1217 | next_event = min_t(u64, next_event, | |
1218 | hrtimer_next_event_without(&ts->sched_timer)); | |
4f86d3a8 | 1219 | |
554c8aa8 | 1220 | return ktime_sub(next_event, now); |
4f86d3a8 LB |
1221 | } |
1222 | ||
466a2b42 JF |
1223 | /** |
1224 | * tick_nohz_get_idle_calls_cpu - return the current idle calls counter value | |
1225 | * for a particular CPU. | |
1226 | * | |
1227 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1228 | */ | |
1229 | unsigned long tick_nohz_get_idle_calls_cpu(int cpu) | |
1230 | { | |
1231 | struct tick_sched *ts = tick_get_tick_sched(cpu); | |
1232 | ||
1233 | return ts->idle_calls; | |
1234 | } | |
1235 | ||
b7eaf1aa RW |
1236 | /** |
1237 | * tick_nohz_get_idle_calls - return the current idle calls counter value | |
1238 | * | |
1239 | * Called from the schedutil frequency scaling governor in scheduler context. | |
1240 | */ | |
1241 | unsigned long tick_nohz_get_idle_calls(void) | |
1242 | { | |
1243 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); | |
1244 | ||
1245 | return ts->idle_calls; | |
1246 | } | |
1247 | ||
96c9b903 YY |
1248 | static void tick_nohz_account_idle_time(struct tick_sched *ts, |
1249 | ktime_t now) | |
2ac0d98f | 1250 | { |
2ac0d98f | 1251 | unsigned long ticks; |
3f4724ea | 1252 | |
96c9b903 YY |
1253 | ts->idle_exittime = now; |
1254 | ||
e44fcb4b | 1255 | if (vtime_accounting_enabled_this_cpu()) |
3f4724ea | 1256 | return; |
79bf2bb3 TG |
1257 | /* |
1258 | * We stopped the tick in idle. Update process times would miss the | |
1259 | * time we slept as update_process_times does only a 1 tick | |
1260 | * accounting. Enforce that this is accounted to idle ! | |
1261 | */ | |
1262 | ticks = jiffies - ts->idle_jiffies; | |
1263 | /* | |
1264 | * We might be one off. Do not randomly account a huge number of ticks! | |
1265 | */ | |
79741dd3 MS |
1266 | if (ticks && ticks < LONG_MAX) |
1267 | account_idle_ticks(ticks); | |
19f5f736 FW |
1268 | } |
1269 | ||
a5183862 | 1270 | void tick_nohz_idle_restart_tick(void) |
2aaf709a | 1271 | { |
a5183862 YY |
1272 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
1273 | ||
1274 | if (ts->tick_stopped) { | |
96c9b903 YY |
1275 | ktime_t now = ktime_get(); |
1276 | tick_nohz_restart_sched_tick(ts, now); | |
1277 | tick_nohz_account_idle_time(ts, now); | |
a5183862 | 1278 | } |
2aaf709a RW |
1279 | } |
1280 | ||
a5183862 | 1281 | static void tick_nohz_idle_update_tick(struct tick_sched *ts, ktime_t now) |
2aaf709a | 1282 | { |
a5183862 YY |
1283 | if (tick_nohz_full_cpu(smp_processor_id())) |
1284 | __tick_nohz_full_update_tick(ts, now); | |
1285 | else | |
1286 | tick_nohz_restart_sched_tick(ts, now); | |
2aaf709a | 1287 | |
96c9b903 | 1288 | tick_nohz_account_idle_time(ts, now); |
2aaf709a RW |
1289 | } |
1290 | ||
79bf2bb3 | 1291 | /** |
280f0677 | 1292 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
1293 | * |
1294 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
1295 | * This also exit the RCU extended quiescent state. The CPU |
1296 | * can use RCU again after this function is called. | |
79bf2bb3 | 1297 | */ |
280f0677 | 1298 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 1299 | { |
4a32fea9 | 1300 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
bbe9a70a | 1301 | bool idle_active, tick_stopped; |
6378ddb5 | 1302 | ktime_t now; |
79bf2bb3 | 1303 | |
6378ddb5 | 1304 | local_irq_disable(); |
2bbb6817 | 1305 | |
15f827be | 1306 | WARN_ON_ONCE(!ts->inidle); |
23a8d888 | 1307 | WARN_ON_ONCE(ts->timer_expires_base); |
15f827be FW |
1308 | |
1309 | ts->inidle = 0; | |
bbe9a70a AB |
1310 | idle_active = ts->idle_active; |
1311 | tick_stopped = ts->tick_stopped; | |
15f827be | 1312 | |
bbe9a70a | 1313 | if (idle_active || tick_stopped) |
eed3b9cf MS |
1314 | now = ktime_get(); |
1315 | ||
bbe9a70a | 1316 | if (idle_active) |
e8fcaa5c | 1317 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 1318 | |
bbe9a70a | 1319 | if (tick_stopped) |
a5183862 | 1320 | tick_nohz_idle_update_tick(ts, now); |
79bf2bb3 | 1321 | |
79bf2bb3 TG |
1322 | local_irq_enable(); |
1323 | } | |
1324 | ||
79bf2bb3 TG |
1325 | /* |
1326 | * The nohz low res interrupt handler | |
1327 | */ | |
1328 | static void tick_nohz_handler(struct clock_event_device *dev) | |
1329 | { | |
22127e93 | 1330 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1331 | struct pt_regs *regs = get_irq_regs(); |
1332 | ktime_t now = ktime_get(); | |
1333 | ||
2456e855 | 1334 | dev->next_event = KTIME_MAX; |
79bf2bb3 | 1335 | |
ff7de620 | 1336 | tick_sched_do_timer(ts, now); |
9e8f559b | 1337 | tick_sched_handle(ts, regs); |
79bf2bb3 | 1338 | |
b5e995e6 VK |
1339 | /* No need to reprogram if we are running tickless */ |
1340 | if (unlikely(ts->tick_stopped)) | |
1341 | return; | |
1342 | ||
b9965449 | 1343 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
0ff53d09 | 1344 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
79bf2bb3 TG |
1345 | } |
1346 | ||
bc7a34b8 TG |
1347 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) |
1348 | { | |
1349 | if (!tick_nohz_enabled) | |
1350 | return; | |
1351 | ts->nohz_mode = mode; | |
1352 | /* One update is enough */ | |
1353 | if (!test_and_set_bit(0, &tick_nohz_active)) | |
ae67bada | 1354 | timers_update_nohz(); |
bc7a34b8 TG |
1355 | } |
1356 | ||
79bf2bb3 TG |
1357 | /** |
1358 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
1359 | */ | |
1360 | static void tick_nohz_switch_to_nohz(void) | |
1361 | { | |
22127e93 | 1362 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1363 | ktime_t next; |
1364 | ||
27630532 | 1365 | if (!tick_nohz_enabled) |
79bf2bb3 TG |
1366 | return; |
1367 | ||
6b442bc8 | 1368 | if (tick_switch_to_oneshot(tick_nohz_handler)) |
79bf2bb3 | 1369 | return; |
6b442bc8 | 1370 | |
79bf2bb3 TG |
1371 | /* |
1372 | * Recycle the hrtimer in ts, so we can share the | |
1373 | * hrtimer_forward with the highres code. | |
1374 | */ | |
71fed982 | 1375 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 TG |
1376 | /* Get the next period */ |
1377 | next = tick_init_jiffy_update(); | |
1378 | ||
0ff53d09 | 1379 | hrtimer_set_expires(&ts->sched_timer, next); |
b9965449 | 1380 | hrtimer_forward_now(&ts->sched_timer, TICK_NSEC); |
1ca8ec53 | 1381 | tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1); |
bc7a34b8 | 1382 | tick_nohz_activate(ts, NOHZ_MODE_LOWRES); |
79bf2bb3 TG |
1383 | } |
1384 | ||
5acac1be | 1385 | static inline void tick_nohz_irq_enter(void) |
eed3b9cf | 1386 | { |
4a32fea9 | 1387 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
eed3b9cf MS |
1388 | ktime_t now; |
1389 | ||
1390 | if (!ts->idle_active && !ts->tick_stopped) | |
1391 | return; | |
1392 | now = ktime_get(); | |
1393 | if (ts->idle_active) | |
e8fcaa5c | 1394 | tick_nohz_stop_idle(ts, now); |
53e87e3c FW |
1395 | /* |
1396 | * If all CPUs are idle. We may need to update a stale jiffies value. | |
1397 | * Note nohz_full is a special case: a timekeeper is guaranteed to stay | |
1398 | * alive but it might be busy looping with interrupts disabled in some | |
1399 | * rare case (typically stop machine). So we must make sure we have a | |
1400 | * last resort. | |
1401 | */ | |
ff006732 | 1402 | if (ts->tick_stopped) |
eed3b9cf | 1403 | tick_nohz_update_jiffies(now); |
eed3b9cf MS |
1404 | } |
1405 | ||
79bf2bb3 TG |
1406 | #else |
1407 | ||
1408 | static inline void tick_nohz_switch_to_nohz(void) { } | |
5acac1be | 1409 | static inline void tick_nohz_irq_enter(void) { } |
bc7a34b8 | 1410 | static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { } |
79bf2bb3 | 1411 | |
3451d024 | 1412 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1413 | |
719254fa TG |
1414 | /* |
1415 | * Called from irq_enter to notify about the possible interruption of idle() | |
1416 | */ | |
5acac1be | 1417 | void tick_irq_enter(void) |
719254fa | 1418 | { |
e8fcaa5c | 1419 | tick_check_oneshot_broadcast_this_cpu(); |
5acac1be | 1420 | tick_nohz_irq_enter(); |
719254fa TG |
1421 | } |
1422 | ||
79bf2bb3 TG |
1423 | /* |
1424 | * High resolution timer specific code | |
1425 | */ | |
1426 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1427 | /* | |
4c9dc641 | 1428 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1429 | * Called with interrupts disabled. |
79bf2bb3 TG |
1430 | */ |
1431 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1432 | { | |
1433 | struct tick_sched *ts = | |
1434 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1435 | struct pt_regs *regs = get_irq_regs(); |
1436 | ktime_t now = ktime_get(); | |
d3ed7824 | 1437 | |
ff7de620 | 1438 | tick_sched_do_timer(ts, now); |
79bf2bb3 TG |
1439 | |
1440 | /* | |
1441 | * Do not call, when we are not in irq context and have | |
1442 | * no valid regs pointer | |
1443 | */ | |
9e8f559b FW |
1444 | if (regs) |
1445 | tick_sched_handle(ts, regs); | |
7c259045 FW |
1446 | else |
1447 | ts->next_tick = 0; | |
79bf2bb3 | 1448 | |
2a16fc93 VK |
1449 | /* No need to reprogram if we are in idle or full dynticks mode */ |
1450 | if (unlikely(ts->tick_stopped)) | |
1451 | return HRTIMER_NORESTART; | |
1452 | ||
b9965449 | 1453 | hrtimer_forward(timer, now, TICK_NSEC); |
79bf2bb3 TG |
1454 | |
1455 | return HRTIMER_RESTART; | |
1456 | } | |
1457 | ||
5307c955 MG |
1458 | static int sched_skew_tick; |
1459 | ||
62cf20b3 TG |
1460 | static int __init skew_tick(char *str) |
1461 | { | |
1462 | get_option(&str, &sched_skew_tick); | |
1463 | ||
1464 | return 0; | |
1465 | } | |
1466 | early_param("skew_tick", skew_tick); | |
1467 | ||
79bf2bb3 TG |
1468 | /** |
1469 | * tick_setup_sched_timer - setup the tick emulation timer | |
1470 | */ | |
1471 | void tick_setup_sched_timer(void) | |
1472 | { | |
22127e93 | 1473 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1474 | ktime_t now = ktime_get(); |
1475 | ||
1476 | /* | |
1477 | * Emulate tick processing via per-CPU hrtimers: | |
1478 | */ | |
902a9f9c | 1479 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_HARD); |
79bf2bb3 | 1480 | ts->sched_timer.function = tick_sched_timer; |
79bf2bb3 | 1481 | |
0de7611a | 1482 | /* Get the next period (per-CPU) */ |
cc584b21 | 1483 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1484 | |
9c3f9e28 | 1485 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 | 1486 | if (sched_skew_tick) { |
b9965449 | 1487 | u64 offset = TICK_NSEC >> 1; |
5307c955 MG |
1488 | do_div(offset, num_possible_cpus()); |
1489 | offset *= smp_processor_id(); | |
1490 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1491 | } | |
1492 | ||
b9965449 | 1493 | hrtimer_forward(&ts->sched_timer, now, TICK_NSEC); |
902a9f9c | 1494 | hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED_HARD); |
bc7a34b8 | 1495 | tick_nohz_activate(ts, NOHZ_MODE_HIGHRES); |
79bf2bb3 | 1496 | } |
3c4fbe5e | 1497 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1498 | |
3451d024 | 1499 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1500 | void tick_cancel_sched_timer(int cpu) |
1501 | { | |
1502 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1503 | ||
3c4fbe5e | 1504 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1505 | if (ts->sched_timer.base) |
1506 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1507 | # endif |
a7901766 | 1508 | |
4b0c0f29 | 1509 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1510 | } |
3c4fbe5e | 1511 | #endif |
79bf2bb3 TG |
1512 | |
1513 | /** | |
1514 | * Async notification about clocksource changes | |
1515 | */ | |
1516 | void tick_clock_notify(void) | |
1517 | { | |
1518 | int cpu; | |
1519 | ||
1520 | for_each_possible_cpu(cpu) | |
1521 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * Async notification about clock event changes | |
1526 | */ | |
1527 | void tick_oneshot_notify(void) | |
1528 | { | |
22127e93 | 1529 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1530 | |
1531 | set_bit(0, &ts->check_clocks); | |
1532 | } | |
1533 | ||
1534 | /** | |
1535 | * Check, if a change happened, which makes oneshot possible. | |
1536 | * | |
1537 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1538 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1539 | * mode, because high resolution timers are disabled (either compile | |
6b442bc8 | 1540 | * or runtime). Called with interrupts disabled. |
79bf2bb3 TG |
1541 | */ |
1542 | int tick_check_oneshot_change(int allow_nohz) | |
1543 | { | |
22127e93 | 1544 | struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched); |
79bf2bb3 TG |
1545 | |
1546 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1547 | return 0; | |
1548 | ||
1549 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1550 | return 0; | |
1551 | ||
cf4fc6cb | 1552 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1553 | return 0; |
1554 | ||
1555 | if (!allow_nohz) | |
1556 | return 1; | |
1557 | ||
1558 | tick_nohz_switch_to_nohz(); | |
1559 | return 0; | |
1560 | } |