]>
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> | |
20 | #include <linux/profile.h> | |
21 | #include <linux/sched.h> | |
8083e4ad | 22 | #include <linux/module.h> |
00b42959 | 23 | #include <linux/irq_work.h> |
9014c45d FW |
24 | #include <linux/posix-timers.h> |
25 | #include <linux/perf_event.h> | |
2e709338 | 26 | #include <linux/context_tracking.h> |
79bf2bb3 | 27 | |
9e203bcc DM |
28 | #include <asm/irq_regs.h> |
29 | ||
79bf2bb3 TG |
30 | #include "tick-internal.h" |
31 | ||
cb41a290 FW |
32 | #include <trace/events/timer.h> |
33 | ||
79bf2bb3 TG |
34 | /* |
35 | * Per cpu nohz control structure | |
36 | */ | |
33a5f626 | 37 | DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched); |
79bf2bb3 TG |
38 | |
39 | /* | |
d6ad4187 | 40 | * The time, when the last jiffy update happened. Protected by jiffies_lock. |
79bf2bb3 TG |
41 | */ |
42 | static ktime_t last_jiffies_update; | |
43 | ||
289f480a IM |
44 | struct tick_sched *tick_get_tick_sched(int cpu) |
45 | { | |
46 | return &per_cpu(tick_cpu_sched, cpu); | |
47 | } | |
48 | ||
79bf2bb3 TG |
49 | /* |
50 | * Must be called with interrupts disabled ! | |
51 | */ | |
52 | static void tick_do_update_jiffies64(ktime_t now) | |
53 | { | |
54 | unsigned long ticks = 0; | |
55 | ktime_t delta; | |
56 | ||
7a14ce1d | 57 | /* |
d6ad4187 | 58 | * Do a quick check without holding jiffies_lock: |
7a14ce1d IM |
59 | */ |
60 | delta = ktime_sub(now, last_jiffies_update); | |
61 | if (delta.tv64 < tick_period.tv64) | |
62 | return; | |
63 | ||
d6ad4187 JS |
64 | /* Reevalute with jiffies_lock held */ |
65 | write_seqlock(&jiffies_lock); | |
79bf2bb3 TG |
66 | |
67 | delta = ktime_sub(now, last_jiffies_update); | |
68 | if (delta.tv64 >= tick_period.tv64) { | |
69 | ||
70 | delta = ktime_sub(delta, tick_period); | |
71 | last_jiffies_update = ktime_add(last_jiffies_update, | |
72 | tick_period); | |
73 | ||
74 | /* Slow path for long timeouts */ | |
75 | if (unlikely(delta.tv64 >= tick_period.tv64)) { | |
76 | s64 incr = ktime_to_ns(tick_period); | |
77 | ||
78 | ticks = ktime_divns(delta, incr); | |
79 | ||
80 | last_jiffies_update = ktime_add_ns(last_jiffies_update, | |
81 | incr * ticks); | |
82 | } | |
83 | do_timer(++ticks); | |
49d670fb TG |
84 | |
85 | /* Keep the tick_next_period variable up to date */ | |
86 | tick_next_period = ktime_add(last_jiffies_update, tick_period); | |
79bf2bb3 | 87 | } |
d6ad4187 | 88 | write_sequnlock(&jiffies_lock); |
47a1b796 | 89 | update_wall_time(); |
79bf2bb3 TG |
90 | } |
91 | ||
92 | /* | |
93 | * Initialize and return retrieve the jiffies update. | |
94 | */ | |
95 | static ktime_t tick_init_jiffy_update(void) | |
96 | { | |
97 | ktime_t period; | |
98 | ||
d6ad4187 | 99 | write_seqlock(&jiffies_lock); |
79bf2bb3 TG |
100 | /* Did we start the jiffies update yet ? */ |
101 | if (last_jiffies_update.tv64 == 0) | |
102 | last_jiffies_update = tick_next_period; | |
103 | period = last_jiffies_update; | |
d6ad4187 | 104 | write_sequnlock(&jiffies_lock); |
79bf2bb3 TG |
105 | return period; |
106 | } | |
107 | ||
5bb96226 FW |
108 | |
109 | static void tick_sched_do_timer(ktime_t now) | |
110 | { | |
111 | int cpu = smp_processor_id(); | |
112 | ||
3451d024 | 113 | #ifdef CONFIG_NO_HZ_COMMON |
5bb96226 FW |
114 | /* |
115 | * Check if the do_timer duty was dropped. We don't care about | |
116 | * concurrency: This happens only when the cpu in charge went | |
117 | * into a long sleep. If two cpus happen to assign themself to | |
118 | * this duty, then the jiffies update is still serialized by | |
9c3f9e28 | 119 | * jiffies_lock. |
5bb96226 | 120 | */ |
a382bf93 | 121 | if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE) |
c5bfece2 | 122 | && !tick_nohz_full_cpu(cpu)) |
5bb96226 FW |
123 | tick_do_timer_cpu = cpu; |
124 | #endif | |
125 | ||
126 | /* Check, if the jiffies need an update */ | |
127 | if (tick_do_timer_cpu == cpu) | |
128 | tick_do_update_jiffies64(now); | |
129 | } | |
130 | ||
9e8f559b FW |
131 | static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs) |
132 | { | |
3451d024 | 133 | #ifdef CONFIG_NO_HZ_COMMON |
9e8f559b FW |
134 | /* |
135 | * When we are idle and the tick is stopped, we have to touch | |
136 | * the watchdog as we might not schedule for a really long | |
137 | * time. This happens on complete idle SMP systems while | |
138 | * waiting on the login prompt. We also increment the "start of | |
139 | * idle" jiffy stamp so the idle accounting adjustment we do | |
140 | * when we go busy again does not account too much ticks. | |
141 | */ | |
142 | if (ts->tick_stopped) { | |
143 | touch_softlockup_watchdog(); | |
144 | if (is_idle_task(current)) | |
145 | ts->idle_jiffies++; | |
146 | } | |
94a57140 | 147 | #endif |
9e8f559b FW |
148 | update_process_times(user_mode(regs)); |
149 | profile_tick(CPU_PROFILING); | |
150 | } | |
151 | ||
c5bfece2 | 152 | #ifdef CONFIG_NO_HZ_FULL |
460775df | 153 | cpumask_var_t tick_nohz_full_mask; |
73867dcd | 154 | bool tick_nohz_full_running; |
a831881b | 155 | |
9014c45d FW |
156 | static bool can_stop_full_tick(void) |
157 | { | |
158 | WARN_ON_ONCE(!irqs_disabled()); | |
159 | ||
cb41a290 FW |
160 | if (!sched_can_stop_tick()) { |
161 | trace_tick_stop(0, "more than 1 task in runqueue\n"); | |
9014c45d | 162 | return false; |
cb41a290 | 163 | } |
9014c45d | 164 | |
cb41a290 FW |
165 | if (!posix_cpu_timers_can_stop_tick(current)) { |
166 | trace_tick_stop(0, "posix timers running\n"); | |
9014c45d | 167 | return false; |
cb41a290 | 168 | } |
9014c45d | 169 | |
cb41a290 FW |
170 | if (!perf_event_can_stop_tick()) { |
171 | trace_tick_stop(0, "perf events running\n"); | |
9014c45d | 172 | return false; |
cb41a290 | 173 | } |
9014c45d FW |
174 | |
175 | /* sched_clock_tick() needs us? */ | |
176 | #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK | |
177 | /* | |
178 | * TODO: kick full dynticks CPUs when | |
179 | * sched_clock_stable is set. | |
180 | */ | |
cb41a290 FW |
181 | if (!sched_clock_stable) { |
182 | trace_tick_stop(0, "unstable sched clock\n"); | |
e12d0271 SR |
183 | /* |
184 | * Don't allow the user to think they can get | |
185 | * full NO_HZ with this machine. | |
186 | */ | |
73867dcd | 187 | WARN_ONCE(tick_nohz_full_running, |
543487c7 | 188 | "NO_HZ FULL will not work with unstable sched clock"); |
9014c45d | 189 | return false; |
cb41a290 | 190 | } |
9014c45d FW |
191 | #endif |
192 | ||
193 | return true; | |
194 | } | |
195 | ||
196 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now); | |
197 | ||
76c24fb0 FW |
198 | /* |
199 | * Re-evaluate the need for the tick on the current CPU | |
200 | * and restart it if necessary. | |
201 | */ | |
d13508f9 | 202 | void __tick_nohz_full_check(void) |
76c24fb0 | 203 | { |
9014c45d FW |
204 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); |
205 | ||
206 | if (tick_nohz_full_cpu(smp_processor_id())) { | |
207 | if (ts->tick_stopped && !is_idle_task(current)) { | |
208 | if (!can_stop_full_tick()) | |
209 | tick_nohz_restart_sched_tick(ts, ktime_get()); | |
210 | } | |
211 | } | |
76c24fb0 FW |
212 | } |
213 | ||
214 | static void nohz_full_kick_work_func(struct irq_work *work) | |
215 | { | |
d13508f9 | 216 | __tick_nohz_full_check(); |
76c24fb0 FW |
217 | } |
218 | ||
219 | static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = { | |
220 | .func = nohz_full_kick_work_func, | |
221 | }; | |
222 | ||
223 | /* | |
224 | * Kick the current 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 | */ | |
227 | void tick_nohz_full_kick(void) | |
228 | { | |
229 | if (tick_nohz_full_cpu(smp_processor_id())) | |
230 | irq_work_queue(&__get_cpu_var(nohz_full_kick_work)); | |
231 | } | |
232 | ||
233 | static void nohz_full_kick_ipi(void *info) | |
234 | { | |
d13508f9 | 235 | __tick_nohz_full_check(); |
76c24fb0 FW |
236 | } |
237 | ||
238 | /* | |
239 | * Kick all full dynticks CPUs in order to force these to re-evaluate | |
240 | * their dependency on the tick and restart it if necessary. | |
241 | */ | |
242 | void tick_nohz_full_kick_all(void) | |
243 | { | |
73867dcd | 244 | if (!tick_nohz_full_running) |
76c24fb0 FW |
245 | return; |
246 | ||
247 | preempt_disable(); | |
73867dcd | 248 | smp_call_function_many(tick_nohz_full_mask, |
76c24fb0 | 249 | nohz_full_kick_ipi, NULL, false); |
c2e7fcf5 | 250 | tick_nohz_full_kick(); |
76c24fb0 FW |
251 | preempt_enable(); |
252 | } | |
253 | ||
99e5ada9 FW |
254 | /* |
255 | * Re-evaluate the need for the tick as we switch the current task. | |
256 | * It might need the tick due to per task/process properties: | |
257 | * perf events, posix cpu timers, ... | |
258 | */ | |
d13508f9 | 259 | void __tick_nohz_task_switch(struct task_struct *tsk) |
99e5ada9 FW |
260 | { |
261 | unsigned long flags; | |
262 | ||
99e5ada9 FW |
263 | local_irq_save(flags); |
264 | ||
6296ace4 LZ |
265 | if (!tick_nohz_full_cpu(smp_processor_id())) |
266 | goto out; | |
267 | ||
99e5ada9 FW |
268 | if (tick_nohz_tick_stopped() && !can_stop_full_tick()) |
269 | tick_nohz_full_kick(); | |
270 | ||
6296ace4 | 271 | out: |
99e5ada9 FW |
272 | local_irq_restore(flags); |
273 | } | |
274 | ||
a831881b | 275 | /* Parse the boot-time nohz CPU list from the kernel parameters. */ |
c5bfece2 | 276 | static int __init tick_nohz_full_setup(char *str) |
a831881b | 277 | { |
0453b435 FW |
278 | int cpu; |
279 | ||
73867dcd FW |
280 | alloc_bootmem_cpumask_var(&tick_nohz_full_mask); |
281 | if (cpulist_parse(str, tick_nohz_full_mask) < 0) { | |
c5bfece2 | 282 | pr_warning("NOHZ: Incorrect nohz_full cpumask\n"); |
0453b435 FW |
283 | return 1; |
284 | } | |
285 | ||
286 | cpu = smp_processor_id(); | |
73867dcd | 287 | if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) { |
0453b435 | 288 | pr_warning("NO_HZ: Clearing %d from nohz_full range for timekeeping\n", cpu); |
73867dcd | 289 | cpumask_clear_cpu(cpu, tick_nohz_full_mask); |
0453b435 | 290 | } |
73867dcd | 291 | tick_nohz_full_running = true; |
0453b435 | 292 | |
a831881b FW |
293 | return 1; |
294 | } | |
c5bfece2 | 295 | __setup("nohz_full=", tick_nohz_full_setup); |
a831881b | 296 | |
0db0628d | 297 | static int tick_nohz_cpu_down_callback(struct notifier_block *nfb, |
a382bf93 FW |
298 | unsigned long action, |
299 | void *hcpu) | |
300 | { | |
301 | unsigned int cpu = (unsigned long)hcpu; | |
302 | ||
303 | switch (action & ~CPU_TASKS_FROZEN) { | |
304 | case CPU_DOWN_PREPARE: | |
305 | /* | |
306 | * If we handle the timekeeping duty for full dynticks CPUs, | |
307 | * we can't safely shutdown that CPU. | |
308 | */ | |
73867dcd | 309 | if (tick_nohz_full_running && tick_do_timer_cpu == cpu) |
1a7f829f | 310 | return NOTIFY_BAD; |
a382bf93 FW |
311 | break; |
312 | } | |
313 | return NOTIFY_OK; | |
314 | } | |
315 | ||
1034fc2f FW |
316 | /* |
317 | * Worst case string length in chunks of CPU range seems 2 steps | |
318 | * separations: 0,2,4,6,... | |
319 | * This is NR_CPUS + sizeof('\0') | |
320 | */ | |
c5bfece2 | 321 | static char __initdata nohz_full_buf[NR_CPUS + 1]; |
1034fc2f | 322 | |
f98823ac FW |
323 | static int tick_nohz_init_all(void) |
324 | { | |
325 | int err = -1; | |
326 | ||
327 | #ifdef CONFIG_NO_HZ_FULL_ALL | |
73867dcd | 328 | if (!alloc_cpumask_var(&tick_nohz_full_mask, GFP_KERNEL)) { |
f98823ac FW |
329 | pr_err("NO_HZ: Can't allocate full dynticks cpumask\n"); |
330 | return err; | |
331 | } | |
332 | err = 0; | |
73867dcd FW |
333 | cpumask_setall(tick_nohz_full_mask); |
334 | cpumask_clear_cpu(smp_processor_id(), tick_nohz_full_mask); | |
335 | tick_nohz_full_running = true; | |
f98823ac FW |
336 | #endif |
337 | return err; | |
338 | } | |
339 | ||
d1e43fa5 | 340 | void __init tick_nohz_init(void) |
a831881b | 341 | { |
d1e43fa5 FW |
342 | int cpu; |
343 | ||
73867dcd | 344 | if (!tick_nohz_full_running) { |
f98823ac FW |
345 | if (tick_nohz_init_all() < 0) |
346 | return; | |
347 | } | |
d1e43fa5 | 348 | |
73867dcd | 349 | for_each_cpu(cpu, tick_nohz_full_mask) |
2e709338 FW |
350 | context_tracking_cpu_set(cpu); |
351 | ||
d1e43fa5 | 352 | cpu_notifier(tick_nohz_cpu_down_callback, 0); |
73867dcd | 353 | cpulist_scnprintf(nohz_full_buf, sizeof(nohz_full_buf), tick_nohz_full_mask); |
c5bfece2 | 354 | pr_info("NO_HZ: Full dynticks CPUs: %s.\n", nohz_full_buf); |
a831881b | 355 | } |
a831881b FW |
356 | #endif |
357 | ||
79bf2bb3 TG |
358 | /* |
359 | * NOHZ - aka dynamic tick functionality | |
360 | */ | |
3451d024 | 361 | #ifdef CONFIG_NO_HZ_COMMON |
79bf2bb3 TG |
362 | /* |
363 | * NO HZ enabled ? | |
364 | */ | |
d689fe22 TG |
365 | static int tick_nohz_enabled __read_mostly = 1; |
366 | int tick_nohz_active __read_mostly; | |
79bf2bb3 TG |
367 | /* |
368 | * Enable / Disable tickless mode | |
369 | */ | |
370 | static int __init setup_tick_nohz(char *str) | |
371 | { | |
372 | if (!strcmp(str, "off")) | |
373 | tick_nohz_enabled = 0; | |
374 | else if (!strcmp(str, "on")) | |
375 | tick_nohz_enabled = 1; | |
376 | else | |
377 | return 0; | |
378 | return 1; | |
379 | } | |
380 | ||
381 | __setup("nohz=", setup_tick_nohz); | |
382 | ||
383 | /** | |
384 | * tick_nohz_update_jiffies - update jiffies when idle was interrupted | |
385 | * | |
386 | * Called from interrupt entry when the CPU was idle | |
387 | * | |
388 | * In case the sched_tick was stopped on this CPU, we have to check if jiffies | |
389 | * must be updated. Otherwise an interrupt handler could use a stale jiffy | |
390 | * value. We do this unconditionally on any cpu, as we don't know whether the | |
391 | * cpu, which has the update task assigned is in a long sleep. | |
392 | */ | |
eed3b9cf | 393 | static void tick_nohz_update_jiffies(ktime_t now) |
79bf2bb3 | 394 | { |
79bf2bb3 | 395 | unsigned long flags; |
79bf2bb3 | 396 | |
e8fcaa5c | 397 | __this_cpu_write(tick_cpu_sched.idle_waketime, now); |
79bf2bb3 TG |
398 | |
399 | local_irq_save(flags); | |
400 | tick_do_update_jiffies64(now); | |
401 | local_irq_restore(flags); | |
02ff3755 IM |
402 | |
403 | touch_softlockup_watchdog(); | |
79bf2bb3 TG |
404 | } |
405 | ||
595aac48 AV |
406 | /* |
407 | * Updates the per cpu time idle statistics counters | |
408 | */ | |
8d63bf94 | 409 | static void |
8c215bd3 | 410 | update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time) |
6378ddb5 | 411 | { |
eed3b9cf | 412 | ktime_t delta; |
6378ddb5 | 413 | |
595aac48 AV |
414 | if (ts->idle_active) { |
415 | delta = ktime_sub(now, ts->idle_entrytime); | |
8c215bd3 | 416 | if (nr_iowait_cpu(cpu) > 0) |
0224cf4c | 417 | ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta); |
6beea0cd MH |
418 | else |
419 | ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta); | |
8c7b09f4 | 420 | ts->idle_entrytime = now; |
595aac48 | 421 | } |
8d63bf94 | 422 | |
e0e37c20 | 423 | if (last_update_time) |
8d63bf94 AV |
424 | *last_update_time = ktime_to_us(now); |
425 | ||
595aac48 AV |
426 | } |
427 | ||
e8fcaa5c | 428 | static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now) |
595aac48 | 429 | { |
e8fcaa5c | 430 | update_ts_time_stats(smp_processor_id(), ts, now, NULL); |
eed3b9cf | 431 | ts->idle_active = 0; |
56c7426b | 432 | |
eed3b9cf | 433 | sched_clock_idle_wakeup_event(0); |
6378ddb5 VP |
434 | } |
435 | ||
e8fcaa5c | 436 | static ktime_t tick_nohz_start_idle(struct tick_sched *ts) |
6378ddb5 | 437 | { |
430ee881 | 438 | ktime_t now = ktime_get(); |
595aac48 | 439 | |
6378ddb5 VP |
440 | ts->idle_entrytime = now; |
441 | ts->idle_active = 1; | |
56c7426b | 442 | sched_clock_idle_sleep_event(); |
6378ddb5 VP |
443 | return now; |
444 | } | |
445 | ||
b1f724c3 AV |
446 | /** |
447 | * get_cpu_idle_time_us - get the total idle time of a cpu | |
448 | * @cpu: CPU number to query | |
09a1d34f MH |
449 | * @last_update_time: variable to store update time in. Do not update |
450 | * counters if NULL. | |
b1f724c3 AV |
451 | * |
452 | * Return the cummulative idle time (since boot) for a given | |
6beea0cd | 453 | * CPU, in microseconds. |
b1f724c3 AV |
454 | * |
455 | * This time is measured via accounting rather than sampling, | |
456 | * and is as accurate as ktime_get() is. | |
457 | * | |
458 | * This function returns -1 if NOHZ is not enabled. | |
459 | */ | |
6378ddb5 VP |
460 | u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time) |
461 | { | |
462 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 463 | ktime_t now, idle; |
6378ddb5 | 464 | |
d689fe22 | 465 | if (!tick_nohz_active) |
8083e4ad | 466 | return -1; |
467 | ||
09a1d34f MH |
468 | now = ktime_get(); |
469 | if (last_update_time) { | |
470 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
471 | idle = ts->idle_sleeptime; | |
472 | } else { | |
473 | if (ts->idle_active && !nr_iowait_cpu(cpu)) { | |
474 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
475 | ||
476 | idle = ktime_add(ts->idle_sleeptime, delta); | |
477 | } else { | |
478 | idle = ts->idle_sleeptime; | |
479 | } | |
480 | } | |
481 | ||
482 | return ktime_to_us(idle); | |
8083e4ad | 483 | |
6378ddb5 | 484 | } |
8083e4ad | 485 | EXPORT_SYMBOL_GPL(get_cpu_idle_time_us); |
6378ddb5 | 486 | |
6beea0cd | 487 | /** |
0224cf4c AV |
488 | * get_cpu_iowait_time_us - get the total iowait time of a cpu |
489 | * @cpu: CPU number to query | |
09a1d34f MH |
490 | * @last_update_time: variable to store update time in. Do not update |
491 | * counters if NULL. | |
0224cf4c AV |
492 | * |
493 | * Return the cummulative iowait time (since boot) for a given | |
494 | * CPU, in microseconds. | |
495 | * | |
496 | * This time is measured via accounting rather than sampling, | |
497 | * and is as accurate as ktime_get() is. | |
498 | * | |
499 | * This function returns -1 if NOHZ is not enabled. | |
500 | */ | |
501 | u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time) | |
502 | { | |
503 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
09a1d34f | 504 | ktime_t now, iowait; |
0224cf4c | 505 | |
d689fe22 | 506 | if (!tick_nohz_active) |
0224cf4c AV |
507 | return -1; |
508 | ||
09a1d34f MH |
509 | now = ktime_get(); |
510 | if (last_update_time) { | |
511 | update_ts_time_stats(cpu, ts, now, last_update_time); | |
512 | iowait = ts->iowait_sleeptime; | |
513 | } else { | |
514 | if (ts->idle_active && nr_iowait_cpu(cpu) > 0) { | |
515 | ktime_t delta = ktime_sub(now, ts->idle_entrytime); | |
0224cf4c | 516 | |
09a1d34f MH |
517 | iowait = ktime_add(ts->iowait_sleeptime, delta); |
518 | } else { | |
519 | iowait = ts->iowait_sleeptime; | |
520 | } | |
521 | } | |
0224cf4c | 522 | |
09a1d34f | 523 | return ktime_to_us(iowait); |
0224cf4c AV |
524 | } |
525 | EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us); | |
526 | ||
84bf1bcc FW |
527 | static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts, |
528 | ktime_t now, int cpu) | |
79bf2bb3 | 529 | { |
280f0677 | 530 | unsigned long seq, last_jiffies, next_jiffies, delta_jiffies; |
84bf1bcc | 531 | ktime_t last_update, expires, ret = { .tv64 = 0 }; |
aa9b1630 | 532 | unsigned long rcu_delta_jiffies; |
4f86d3a8 | 533 | struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev; |
98962465 | 534 | u64 time_delta; |
79bf2bb3 | 535 | |
79bf2bb3 TG |
536 | /* Read jiffies and the time when jiffies were updated last */ |
537 | do { | |
d6ad4187 | 538 | seq = read_seqbegin(&jiffies_lock); |
79bf2bb3 TG |
539 | last_update = last_jiffies_update; |
540 | last_jiffies = jiffies; | |
27185016 | 541 | time_delta = timekeeping_max_deferment(); |
d6ad4187 | 542 | } while (read_seqretry(&jiffies_lock, seq)); |
79bf2bb3 | 543 | |
74876a98 | 544 | if (rcu_needs_cpu(cpu, &rcu_delta_jiffies) || |
00b42959 | 545 | arch_needs_cpu(cpu) || irq_work_needs_cpu()) { |
3c5d92a0 | 546 | next_jiffies = last_jiffies + 1; |
6ba9b346 | 547 | delta_jiffies = 1; |
3c5d92a0 MS |
548 | } else { |
549 | /* Get the next timer wheel timer */ | |
550 | next_jiffies = get_next_timer_interrupt(last_jiffies); | |
551 | delta_jiffies = next_jiffies - last_jiffies; | |
aa9b1630 PM |
552 | if (rcu_delta_jiffies < delta_jiffies) { |
553 | next_jiffies = last_jiffies + rcu_delta_jiffies; | |
554 | delta_jiffies = rcu_delta_jiffies; | |
555 | } | |
3c5d92a0 | 556 | } |
47aa8b6c | 557 | |
79bf2bb3 | 558 | /* |
47aa8b6c IM |
559 | * Do not stop the tick, if we are only one off (or less) |
560 | * or if the cpu is required for RCU: | |
79bf2bb3 | 561 | */ |
47aa8b6c | 562 | if (!ts->tick_stopped && delta_jiffies <= 1) |
79bf2bb3 TG |
563 | goto out; |
564 | ||
565 | /* Schedule the tick, if we are at least one jiffie off */ | |
566 | if ((long)delta_jiffies >= 1) { | |
567 | ||
00147449 WR |
568 | /* |
569 | * If this cpu is the one which updates jiffies, then | |
570 | * give up the assignment and let it be taken by the | |
571 | * cpu which runs the tick timer next, which might be | |
572 | * this cpu as well. If we don't drop this here the | |
573 | * jiffies might be stale and do_timer() never | |
27185016 TG |
574 | * invoked. Keep track of the fact that it was the one |
575 | * which had the do_timer() duty last. If this cpu is | |
576 | * the one which had the do_timer() duty last, we | |
577 | * limit the sleep time to the timekeeping | |
578 | * max_deferement value which we retrieved | |
579 | * above. Otherwise we can sleep as long as we want. | |
00147449 | 580 | */ |
27185016 | 581 | if (cpu == tick_do_timer_cpu) { |
00147449 | 582 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; |
27185016 TG |
583 | ts->do_timer_last = 1; |
584 | } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) { | |
585 | time_delta = KTIME_MAX; | |
586 | ts->do_timer_last = 0; | |
587 | } else if (!ts->do_timer_last) { | |
588 | time_delta = KTIME_MAX; | |
589 | } | |
590 | ||
265f22a9 FW |
591 | #ifdef CONFIG_NO_HZ_FULL |
592 | if (!ts->inidle) { | |
593 | time_delta = min(time_delta, | |
594 | scheduler_tick_max_deferment()); | |
595 | } | |
596 | #endif | |
597 | ||
00147449 | 598 | /* |
98962465 JH |
599 | * calculate the expiry time for the next timer wheel |
600 | * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals | |
601 | * that there is no timer pending or at least extremely | |
602 | * far into the future (12 days for HZ=1000). In this | |
603 | * case we set the expiry to the end of time. | |
604 | */ | |
605 | if (likely(delta_jiffies < NEXT_TIMER_MAX_DELTA)) { | |
606 | /* | |
607 | * Calculate the time delta for the next timer event. | |
608 | * If the time delta exceeds the maximum time delta | |
609 | * permitted by the current clocksource then adjust | |
610 | * the time delta accordingly to ensure the | |
611 | * clocksource does not wrap. | |
612 | */ | |
613 | time_delta = min_t(u64, time_delta, | |
614 | tick_period.tv64 * delta_jiffies); | |
98962465 | 615 | } |
00147449 | 616 | |
27185016 TG |
617 | if (time_delta < KTIME_MAX) |
618 | expires = ktime_add_ns(last_update, time_delta); | |
619 | else | |
620 | expires.tv64 = KTIME_MAX; | |
00147449 | 621 | |
00147449 WR |
622 | /* Skip reprogram of event if its not changed */ |
623 | if (ts->tick_stopped && ktime_equal(expires, dev->next_event)) | |
624 | goto out; | |
625 | ||
84bf1bcc FW |
626 | ret = expires; |
627 | ||
79bf2bb3 TG |
628 | /* |
629 | * nohz_stop_sched_tick can be called several times before | |
630 | * the nohz_restart_sched_tick is called. This happens when | |
631 | * interrupts arrive which do not cause a reschedule. In the | |
632 | * first call we save the current tick time, so we can restart | |
633 | * the scheduler tick in nohz_restart_sched_tick. | |
634 | */ | |
635 | if (!ts->tick_stopped) { | |
c1cc017c | 636 | nohz_balance_enter_idle(cpu); |
5167e8d5 | 637 | calc_load_enter_idle(); |
46cb4b7c | 638 | |
f5d411c9 | 639 | ts->last_tick = hrtimer_get_expires(&ts->sched_timer); |
79bf2bb3 | 640 | ts->tick_stopped = 1; |
cb41a290 | 641 | trace_tick_stop(1, " "); |
79bf2bb3 | 642 | } |
d3ed7824 | 643 | |
eaad084b | 644 | /* |
98962465 JH |
645 | * If the expiration time == KTIME_MAX, then |
646 | * in this case we simply stop the tick timer. | |
eaad084b | 647 | */ |
98962465 | 648 | if (unlikely(expires.tv64 == KTIME_MAX)) { |
eaad084b TG |
649 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) |
650 | hrtimer_cancel(&ts->sched_timer); | |
651 | goto out; | |
652 | } | |
653 | ||
79bf2bb3 TG |
654 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { |
655 | hrtimer_start(&ts->sched_timer, expires, | |
5c333864 | 656 | HRTIMER_MODE_ABS_PINNED); |
79bf2bb3 TG |
657 | /* Check, if the timer was already in the past */ |
658 | if (hrtimer_active(&ts->sched_timer)) | |
659 | goto out; | |
4c9dc641 | 660 | } else if (!tick_program_event(expires, 0)) |
79bf2bb3 TG |
661 | goto out; |
662 | /* | |
663 | * We are past the event already. So we crossed a | |
664 | * jiffie boundary. Update jiffies and raise the | |
665 | * softirq. | |
666 | */ | |
667 | tick_do_update_jiffies64(ktime_get()); | |
79bf2bb3 TG |
668 | } |
669 | raise_softirq_irqoff(TIMER_SOFTIRQ); | |
670 | out: | |
671 | ts->next_jiffies = next_jiffies; | |
672 | ts->last_jiffies = last_jiffies; | |
4f86d3a8 | 673 | ts->sleep_length = ktime_sub(dev->next_event, now); |
84bf1bcc FW |
674 | |
675 | return ret; | |
280f0677 FW |
676 | } |
677 | ||
5811d996 FW |
678 | static void tick_nohz_full_stop_tick(struct tick_sched *ts) |
679 | { | |
680 | #ifdef CONFIG_NO_HZ_FULL | |
681 | int cpu = smp_processor_id(); | |
682 | ||
683 | if (!tick_nohz_full_cpu(cpu) || is_idle_task(current)) | |
684 | return; | |
685 | ||
686 | if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE) | |
687 | return; | |
688 | ||
689 | if (!can_stop_full_tick()) | |
690 | return; | |
691 | ||
692 | tick_nohz_stop_sched_tick(ts, ktime_get(), cpu); | |
693 | #endif | |
694 | } | |
695 | ||
5b39939a FW |
696 | static bool can_stop_idle_tick(int cpu, struct tick_sched *ts) |
697 | { | |
698 | /* | |
699 | * If this cpu is offline and it is the one which updates | |
700 | * jiffies, then give up the assignment and let it be taken by | |
701 | * the cpu which runs the tick timer next. If we don't drop | |
702 | * this here the jiffies might be stale and do_timer() never | |
703 | * invoked. | |
704 | */ | |
705 | if (unlikely(!cpu_online(cpu))) { | |
706 | if (cpu == tick_do_timer_cpu) | |
707 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
f7ea0fd6 | 708 | return false; |
5b39939a FW |
709 | } |
710 | ||
0e576acb TG |
711 | if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) { |
712 | ts->sleep_length = (ktime_t) { .tv64 = NSEC_PER_SEC/HZ }; | |
5b39939a | 713 | return false; |
0e576acb | 714 | } |
5b39939a FW |
715 | |
716 | if (need_resched()) | |
717 | return false; | |
718 | ||
719 | if (unlikely(local_softirq_pending() && cpu_online(cpu))) { | |
720 | static int ratelimit; | |
721 | ||
803b0eba PM |
722 | if (ratelimit < 10 && |
723 | (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) { | |
cfea7d7e RV |
724 | pr_warn("NOHZ: local_softirq_pending %02x\n", |
725 | (unsigned int) local_softirq_pending()); | |
5b39939a FW |
726 | ratelimit++; |
727 | } | |
728 | return false; | |
729 | } | |
730 | ||
460775df | 731 | if (tick_nohz_full_enabled()) { |
a382bf93 FW |
732 | /* |
733 | * Keep the tick alive to guarantee timekeeping progression | |
734 | * if there are full dynticks CPUs around | |
735 | */ | |
736 | if (tick_do_timer_cpu == cpu) | |
737 | return false; | |
738 | /* | |
739 | * Boot safety: make sure the timekeeping duty has been | |
740 | * assigned before entering dyntick-idle mode, | |
741 | */ | |
742 | if (tick_do_timer_cpu == TICK_DO_TIMER_NONE) | |
743 | return false; | |
744 | } | |
745 | ||
5b39939a FW |
746 | return true; |
747 | } | |
748 | ||
19f5f736 FW |
749 | static void __tick_nohz_idle_enter(struct tick_sched *ts) |
750 | { | |
84bf1bcc | 751 | ktime_t now, expires; |
5b39939a | 752 | int cpu = smp_processor_id(); |
19f5f736 | 753 | |
e8fcaa5c | 754 | now = tick_nohz_start_idle(ts); |
2ac0d98f | 755 | |
5b39939a FW |
756 | if (can_stop_idle_tick(cpu, ts)) { |
757 | int was_stopped = ts->tick_stopped; | |
758 | ||
759 | ts->idle_calls++; | |
84bf1bcc FW |
760 | |
761 | expires = tick_nohz_stop_sched_tick(ts, now, cpu); | |
762 | if (expires.tv64 > 0LL) { | |
763 | ts->idle_sleeps++; | |
764 | ts->idle_expires = expires; | |
765 | } | |
5b39939a FW |
766 | |
767 | if (!was_stopped && ts->tick_stopped) | |
768 | ts->idle_jiffies = ts->last_jiffies; | |
769 | } | |
280f0677 FW |
770 | } |
771 | ||
772 | /** | |
773 | * tick_nohz_idle_enter - stop the idle tick from the idle task | |
774 | * | |
775 | * When the next event is more than a tick into the future, stop the idle tick | |
776 | * Called when we start the idle loop. | |
2bbb6817 | 777 | * |
1268fbc7 | 778 | * The arch is responsible of calling: |
2bbb6817 FW |
779 | * |
780 | * - rcu_idle_enter() after its last use of RCU before the CPU is put | |
781 | * to sleep. | |
782 | * - rcu_idle_exit() before the first use of RCU after the CPU is woken up. | |
280f0677 | 783 | */ |
1268fbc7 | 784 | void tick_nohz_idle_enter(void) |
280f0677 FW |
785 | { |
786 | struct tick_sched *ts; | |
787 | ||
1268fbc7 FW |
788 | WARN_ON_ONCE(irqs_disabled()); |
789 | ||
0db49b72 LT |
790 | /* |
791 | * Update the idle state in the scheduler domain hierarchy | |
792 | * when tick_nohz_stop_sched_tick() is called from the idle loop. | |
793 | * State will be updated to busy during the first busy tick after | |
794 | * exiting idle. | |
795 | */ | |
796 | set_cpu_sd_state_idle(); | |
797 | ||
1268fbc7 FW |
798 | local_irq_disable(); |
799 | ||
280f0677 | 800 | ts = &__get_cpu_var(tick_cpu_sched); |
280f0677 | 801 | ts->inidle = 1; |
19f5f736 | 802 | __tick_nohz_idle_enter(ts); |
1268fbc7 FW |
803 | |
804 | local_irq_enable(); | |
280f0677 | 805 | } |
4dbd2771 | 806 | EXPORT_SYMBOL_GPL(tick_nohz_idle_enter); |
280f0677 FW |
807 | |
808 | /** | |
809 | * tick_nohz_irq_exit - update next tick event from interrupt exit | |
810 | * | |
811 | * When an interrupt fires while we are idle and it doesn't cause | |
812 | * a reschedule, it may still add, modify or delete a timer, enqueue | |
813 | * an RCU callback, etc... | |
814 | * So we need to re-calculate and reprogram the next tick event. | |
815 | */ | |
816 | void tick_nohz_irq_exit(void) | |
817 | { | |
818 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
819 | ||
14851912 | 820 | if (ts->inidle) |
5811d996 | 821 | __tick_nohz_idle_enter(ts); |
14851912 | 822 | else |
5811d996 | 823 | tick_nohz_full_stop_tick(ts); |
79bf2bb3 TG |
824 | } |
825 | ||
4f86d3a8 LB |
826 | /** |
827 | * tick_nohz_get_sleep_length - return the length of the current sleep | |
828 | * | |
829 | * Called from power state control code with interrupts disabled | |
830 | */ | |
831 | ktime_t tick_nohz_get_sleep_length(void) | |
832 | { | |
833 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
834 | ||
835 | return ts->sleep_length; | |
836 | } | |
837 | ||
c34bec5a TG |
838 | static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) |
839 | { | |
840 | hrtimer_cancel(&ts->sched_timer); | |
f5d411c9 | 841 | hrtimer_set_expires(&ts->sched_timer, ts->last_tick); |
c34bec5a TG |
842 | |
843 | while (1) { | |
844 | /* Forward the time to expire in the future */ | |
845 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
846 | ||
847 | if (ts->nohz_mode == NOHZ_MODE_HIGHRES) { | |
268a3dcf | 848 | hrtimer_start_expires(&ts->sched_timer, |
5c333864 | 849 | HRTIMER_MODE_ABS_PINNED); |
c34bec5a TG |
850 | /* Check, if the timer was already in the past */ |
851 | if (hrtimer_active(&ts->sched_timer)) | |
852 | break; | |
853 | } else { | |
268a3dcf TG |
854 | if (!tick_program_event( |
855 | hrtimer_get_expires(&ts->sched_timer), 0)) | |
c34bec5a TG |
856 | break; |
857 | } | |
6f103929 | 858 | /* Reread time and update jiffies */ |
c34bec5a | 859 | now = ktime_get(); |
6f103929 | 860 | tick_do_update_jiffies64(now); |
c34bec5a TG |
861 | } |
862 | } | |
863 | ||
19f5f736 | 864 | static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) |
79bf2bb3 | 865 | { |
79bf2bb3 | 866 | /* Update jiffies first */ |
79bf2bb3 | 867 | tick_do_update_jiffies64(now); |
5aaa0b7a | 868 | update_cpu_load_nohz(); |
79bf2bb3 | 869 | |
749c8814 | 870 | calc_load_exit_idle(); |
2ac0d98f FW |
871 | touch_softlockup_watchdog(); |
872 | /* | |
873 | * Cancel the scheduled timer and restore the tick | |
874 | */ | |
875 | ts->tick_stopped = 0; | |
876 | ts->idle_exittime = now; | |
877 | ||
878 | tick_nohz_restart(ts, now); | |
879 | } | |
880 | ||
881 | static void tick_nohz_account_idle_ticks(struct tick_sched *ts) | |
882 | { | |
3f4724ea | 883 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
2ac0d98f | 884 | unsigned long ticks; |
3f4724ea FW |
885 | |
886 | if (vtime_accounting_enabled()) | |
887 | return; | |
79bf2bb3 TG |
888 | /* |
889 | * We stopped the tick in idle. Update process times would miss the | |
890 | * time we slept as update_process_times does only a 1 tick | |
891 | * accounting. Enforce that this is accounted to idle ! | |
892 | */ | |
893 | ticks = jiffies - ts->idle_jiffies; | |
894 | /* | |
895 | * We might be one off. Do not randomly account a huge number of ticks! | |
896 | */ | |
79741dd3 MS |
897 | if (ticks && ticks < LONG_MAX) |
898 | account_idle_ticks(ticks); | |
899 | #endif | |
19f5f736 FW |
900 | } |
901 | ||
79bf2bb3 | 902 | /** |
280f0677 | 903 | * tick_nohz_idle_exit - restart the idle tick from the idle task |
79bf2bb3 TG |
904 | * |
905 | * Restart the idle tick when the CPU is woken up from idle | |
280f0677 FW |
906 | * This also exit the RCU extended quiescent state. The CPU |
907 | * can use RCU again after this function is called. | |
79bf2bb3 | 908 | */ |
280f0677 | 909 | void tick_nohz_idle_exit(void) |
79bf2bb3 | 910 | { |
e8fcaa5c | 911 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); |
6378ddb5 | 912 | ktime_t now; |
79bf2bb3 | 913 | |
6378ddb5 | 914 | local_irq_disable(); |
2bbb6817 | 915 | |
15f827be FW |
916 | WARN_ON_ONCE(!ts->inidle); |
917 | ||
918 | ts->inidle = 0; | |
919 | ||
920 | if (ts->idle_active || ts->tick_stopped) | |
eed3b9cf MS |
921 | now = ktime_get(); |
922 | ||
923 | if (ts->idle_active) | |
e8fcaa5c | 924 | tick_nohz_stop_idle(ts, now); |
6378ddb5 | 925 | |
2ac0d98f | 926 | if (ts->tick_stopped) { |
19f5f736 | 927 | tick_nohz_restart_sched_tick(ts, now); |
2ac0d98f | 928 | tick_nohz_account_idle_ticks(ts); |
6378ddb5 | 929 | } |
79bf2bb3 | 930 | |
79bf2bb3 TG |
931 | local_irq_enable(); |
932 | } | |
4dbd2771 | 933 | EXPORT_SYMBOL_GPL(tick_nohz_idle_exit); |
79bf2bb3 TG |
934 | |
935 | static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now) | |
936 | { | |
937 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
cc584b21 | 938 | return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0); |
79bf2bb3 TG |
939 | } |
940 | ||
941 | /* | |
942 | * The nohz low res interrupt handler | |
943 | */ | |
944 | static void tick_nohz_handler(struct clock_event_device *dev) | |
945 | { | |
946 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
947 | struct pt_regs *regs = get_irq_regs(); | |
948 | ktime_t now = ktime_get(); | |
949 | ||
950 | dev->next_event.tv64 = KTIME_MAX; | |
951 | ||
5bb96226 | 952 | tick_sched_do_timer(now); |
9e8f559b | 953 | tick_sched_handle(ts, regs); |
79bf2bb3 | 954 | |
79bf2bb3 TG |
955 | while (tick_nohz_reprogram(ts, now)) { |
956 | now = ktime_get(); | |
957 | tick_do_update_jiffies64(now); | |
958 | } | |
959 | } | |
960 | ||
961 | /** | |
962 | * tick_nohz_switch_to_nohz - switch to nohz mode | |
963 | */ | |
964 | static void tick_nohz_switch_to_nohz(void) | |
965 | { | |
966 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
967 | ktime_t next; | |
968 | ||
d689fe22 | 969 | if (!tick_nohz_active) |
79bf2bb3 TG |
970 | return; |
971 | ||
972 | local_irq_disable(); | |
973 | if (tick_switch_to_oneshot(tick_nohz_handler)) { | |
974 | local_irq_enable(); | |
975 | return; | |
976 | } | |
d689fe22 | 977 | tick_nohz_active = 1; |
79bf2bb3 TG |
978 | ts->nohz_mode = NOHZ_MODE_LOWRES; |
979 | ||
980 | /* | |
981 | * Recycle the hrtimer in ts, so we can share the | |
982 | * hrtimer_forward with the highres code. | |
983 | */ | |
984 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
985 | /* Get the next period */ | |
986 | next = tick_init_jiffy_update(); | |
987 | ||
988 | for (;;) { | |
cc584b21 | 989 | hrtimer_set_expires(&ts->sched_timer, next); |
79bf2bb3 TG |
990 | if (!tick_program_event(next, 0)) |
991 | break; | |
992 | next = ktime_add(next, tick_period); | |
993 | } | |
994 | local_irq_enable(); | |
79bf2bb3 TG |
995 | } |
996 | ||
fb02fbc1 TG |
997 | /* |
998 | * When NOHZ is enabled and the tick is stopped, we need to kick the | |
999 | * tick timer from irq_enter() so that the jiffies update is kept | |
1000 | * alive during long running softirqs. That's ugly as hell, but | |
1001 | * correctness is key even if we need to fix the offending softirq in | |
1002 | * the first place. | |
1003 | * | |
1004 | * Note, this is different to tick_nohz_restart. We just kick the | |
1005 | * timer and do not touch the other magic bits which need to be done | |
1006 | * when idle is left. | |
1007 | */ | |
e8fcaa5c | 1008 | static void tick_nohz_kick_tick(struct tick_sched *ts, ktime_t now) |
fb02fbc1 | 1009 | { |
ae99286b TG |
1010 | #if 0 |
1011 | /* Switch back to 2.6.27 behaviour */ | |
eed3b9cf | 1012 | ktime_t delta; |
fb02fbc1 | 1013 | |
c4bd822e TG |
1014 | /* |
1015 | * Do not touch the tick device, when the next expiry is either | |
1016 | * already reached or less/equal than the tick period. | |
1017 | */ | |
268a3dcf | 1018 | delta = ktime_sub(hrtimer_get_expires(&ts->sched_timer), now); |
c4bd822e TG |
1019 | if (delta.tv64 <= tick_period.tv64) |
1020 | return; | |
1021 | ||
1022 | tick_nohz_restart(ts, now); | |
ae99286b | 1023 | #endif |
fb02fbc1 TG |
1024 | } |
1025 | ||
e8fcaa5c | 1026 | static inline void tick_check_nohz_this_cpu(void) |
eed3b9cf | 1027 | { |
e8fcaa5c | 1028 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); |
eed3b9cf MS |
1029 | ktime_t now; |
1030 | ||
1031 | if (!ts->idle_active && !ts->tick_stopped) | |
1032 | return; | |
1033 | now = ktime_get(); | |
1034 | if (ts->idle_active) | |
e8fcaa5c | 1035 | tick_nohz_stop_idle(ts, now); |
eed3b9cf MS |
1036 | if (ts->tick_stopped) { |
1037 | tick_nohz_update_jiffies(now); | |
e8fcaa5c | 1038 | tick_nohz_kick_tick(ts, now); |
eed3b9cf MS |
1039 | } |
1040 | } | |
1041 | ||
79bf2bb3 TG |
1042 | #else |
1043 | ||
1044 | static inline void tick_nohz_switch_to_nohz(void) { } | |
e8fcaa5c | 1045 | static inline void tick_check_nohz_this_cpu(void) { } |
79bf2bb3 | 1046 | |
3451d024 | 1047 | #endif /* CONFIG_NO_HZ_COMMON */ |
79bf2bb3 | 1048 | |
719254fa TG |
1049 | /* |
1050 | * Called from irq_enter to notify about the possible interruption of idle() | |
1051 | */ | |
e8fcaa5c | 1052 | void tick_check_idle(void) |
719254fa | 1053 | { |
e8fcaa5c FW |
1054 | tick_check_oneshot_broadcast_this_cpu(); |
1055 | tick_check_nohz_this_cpu(); | |
719254fa TG |
1056 | } |
1057 | ||
79bf2bb3 TG |
1058 | /* |
1059 | * High resolution timer specific code | |
1060 | */ | |
1061 | #ifdef CONFIG_HIGH_RES_TIMERS | |
1062 | /* | |
4c9dc641 | 1063 | * We rearm the timer until we get disabled by the idle code. |
351f181f | 1064 | * Called with interrupts disabled. |
79bf2bb3 TG |
1065 | */ |
1066 | static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer) | |
1067 | { | |
1068 | struct tick_sched *ts = | |
1069 | container_of(timer, struct tick_sched, sched_timer); | |
79bf2bb3 TG |
1070 | struct pt_regs *regs = get_irq_regs(); |
1071 | ktime_t now = ktime_get(); | |
d3ed7824 | 1072 | |
5bb96226 | 1073 | tick_sched_do_timer(now); |
79bf2bb3 TG |
1074 | |
1075 | /* | |
1076 | * Do not call, when we are not in irq context and have | |
1077 | * no valid regs pointer | |
1078 | */ | |
9e8f559b FW |
1079 | if (regs) |
1080 | tick_sched_handle(ts, regs); | |
79bf2bb3 | 1081 | |
79bf2bb3 TG |
1082 | hrtimer_forward(timer, now, tick_period); |
1083 | ||
1084 | return HRTIMER_RESTART; | |
1085 | } | |
1086 | ||
5307c955 MG |
1087 | static int sched_skew_tick; |
1088 | ||
62cf20b3 TG |
1089 | static int __init skew_tick(char *str) |
1090 | { | |
1091 | get_option(&str, &sched_skew_tick); | |
1092 | ||
1093 | return 0; | |
1094 | } | |
1095 | early_param("skew_tick", skew_tick); | |
1096 | ||
79bf2bb3 TG |
1097 | /** |
1098 | * tick_setup_sched_timer - setup the tick emulation timer | |
1099 | */ | |
1100 | void tick_setup_sched_timer(void) | |
1101 | { | |
1102 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1103 | ktime_t now = ktime_get(); | |
1104 | ||
1105 | /* | |
1106 | * Emulate tick processing via per-CPU hrtimers: | |
1107 | */ | |
1108 | hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
1109 | ts->sched_timer.function = tick_sched_timer; | |
79bf2bb3 | 1110 | |
3704540b | 1111 | /* Get the next period (per cpu) */ |
cc584b21 | 1112 | hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update()); |
79bf2bb3 | 1113 | |
9c3f9e28 | 1114 | /* Offset the tick to avert jiffies_lock contention. */ |
5307c955 MG |
1115 | if (sched_skew_tick) { |
1116 | u64 offset = ktime_to_ns(tick_period) >> 1; | |
1117 | do_div(offset, num_possible_cpus()); | |
1118 | offset *= smp_processor_id(); | |
1119 | hrtimer_add_expires_ns(&ts->sched_timer, offset); | |
1120 | } | |
1121 | ||
79bf2bb3 TG |
1122 | for (;;) { |
1123 | hrtimer_forward(&ts->sched_timer, now, tick_period); | |
5c333864 AB |
1124 | hrtimer_start_expires(&ts->sched_timer, |
1125 | HRTIMER_MODE_ABS_PINNED); | |
79bf2bb3 TG |
1126 | /* Check, if the timer was already in the past */ |
1127 | if (hrtimer_active(&ts->sched_timer)) | |
1128 | break; | |
1129 | now = ktime_get(); | |
1130 | } | |
1131 | ||
3451d024 | 1132 | #ifdef CONFIG_NO_HZ_COMMON |
d689fe22 | 1133 | if (tick_nohz_enabled) { |
79bf2bb3 | 1134 | ts->nohz_mode = NOHZ_MODE_HIGHRES; |
d689fe22 TG |
1135 | tick_nohz_active = 1; |
1136 | } | |
79bf2bb3 TG |
1137 | #endif |
1138 | } | |
3c4fbe5e | 1139 | #endif /* HIGH_RES_TIMERS */ |
79bf2bb3 | 1140 | |
3451d024 | 1141 | #if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1142 | void tick_cancel_sched_timer(int cpu) |
1143 | { | |
1144 | struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu); | |
1145 | ||
3c4fbe5e | 1146 | # ifdef CONFIG_HIGH_RES_TIMERS |
79bf2bb3 TG |
1147 | if (ts->sched_timer.base) |
1148 | hrtimer_cancel(&ts->sched_timer); | |
3c4fbe5e | 1149 | # endif |
a7901766 | 1150 | |
4b0c0f29 | 1151 | memset(ts, 0, sizeof(*ts)); |
79bf2bb3 | 1152 | } |
3c4fbe5e | 1153 | #endif |
79bf2bb3 TG |
1154 | |
1155 | /** | |
1156 | * Async notification about clocksource changes | |
1157 | */ | |
1158 | void tick_clock_notify(void) | |
1159 | { | |
1160 | int cpu; | |
1161 | ||
1162 | for_each_possible_cpu(cpu) | |
1163 | set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks); | |
1164 | } | |
1165 | ||
1166 | /* | |
1167 | * Async notification about clock event changes | |
1168 | */ | |
1169 | void tick_oneshot_notify(void) | |
1170 | { | |
1171 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1172 | ||
1173 | set_bit(0, &ts->check_clocks); | |
1174 | } | |
1175 | ||
1176 | /** | |
1177 | * Check, if a change happened, which makes oneshot possible. | |
1178 | * | |
1179 | * Called cyclic from the hrtimer softirq (driven by the timer | |
1180 | * softirq) allow_nohz signals, that we can switch into low-res nohz | |
1181 | * mode, because high resolution timers are disabled (either compile | |
1182 | * or runtime). | |
1183 | */ | |
1184 | int tick_check_oneshot_change(int allow_nohz) | |
1185 | { | |
1186 | struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched); | |
1187 | ||
1188 | if (!test_and_clear_bit(0, &ts->check_clocks)) | |
1189 | return 0; | |
1190 | ||
1191 | if (ts->nohz_mode != NOHZ_MODE_INACTIVE) | |
1192 | return 0; | |
1193 | ||
cf4fc6cb | 1194 | if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available()) |
79bf2bb3 TG |
1195 | return 0; |
1196 | ||
1197 | if (!allow_nohz) | |
1198 | return 1; | |
1199 | ||
1200 | tick_nohz_switch_to_nohz(); | |
1201 | return 0; | |
1202 | } |