]>
Commit | Line | Data |
---|---|---|
906568c9 TG |
1 | /* |
2 | * linux/kernel/time/tick-common.c | |
3 | * | |
4 | * This file contains the base functions to manage periodic tick | |
5 | * related events. | |
6 | * | |
7 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> | |
8 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar | |
9 | * Copyright(C) 2006-2007, Timesys Corp., Thomas Gleixner | |
10 | * | |
11 | * This code is licenced under the GPL version 2. For details see | |
12 | * kernel-base/COPYING. | |
13 | */ | |
14 | #include <linux/cpu.h> | |
15 | #include <linux/err.h> | |
16 | #include <linux/hrtimer.h> | |
d7b90689 | 17 | #include <linux/interrupt.h> |
906568c9 TG |
18 | #include <linux/percpu.h> |
19 | #include <linux/profile.h> | |
20 | #include <linux/sched.h> | |
ccf33d68 | 21 | #include <linux/module.h> |
906568c9 | 22 | |
d7b90689 RK |
23 | #include <asm/irq_regs.h> |
24 | ||
f8381cba TG |
25 | #include "tick-internal.h" |
26 | ||
906568c9 TG |
27 | /* |
28 | * Tick devices | |
29 | */ | |
f8381cba | 30 | DEFINE_PER_CPU(struct tick_device, tick_cpu_device); |
906568c9 TG |
31 | /* |
32 | * Tick next event: keeps track of the tick time | |
33 | */ | |
f8381cba TG |
34 | ktime_t tick_next_period; |
35 | ktime_t tick_period; | |
050ded1b AM |
36 | |
37 | /* | |
38 | * tick_do_timer_cpu is a timer core internal variable which holds the CPU NR | |
39 | * which is responsible for calling do_timer(), i.e. the timekeeping stuff. This | |
40 | * variable has two functions: | |
41 | * | |
42 | * 1) Prevent a thundering herd issue of a gazillion of CPUs trying to grab the | |
43 | * timekeeping lock all at once. Only the CPU which is assigned to do the | |
44 | * update is handling it. | |
45 | * | |
46 | * 2) Hand off the duty in the NOHZ idle case by setting the value to | |
47 | * TICK_DO_TIMER_NONE, i.e. a non existing CPU. So the next cpu which looks | |
48 | * at it will take over and keep the time keeping alive. The handover | |
49 | * procedure also covers cpu hotplug. | |
50 | */ | |
6441402b | 51 | int tick_do_timer_cpu __read_mostly = TICK_DO_TIMER_BOOT; |
906568c9 | 52 | |
289f480a IM |
53 | /* |
54 | * Debugging: see timer_list.c | |
55 | */ | |
56 | struct tick_device *tick_get_device(int cpu) | |
57 | { | |
58 | return &per_cpu(tick_cpu_device, cpu); | |
59 | } | |
60 | ||
79bf2bb3 TG |
61 | /** |
62 | * tick_is_oneshot_available - check for a oneshot capable event device | |
63 | */ | |
64 | int tick_is_oneshot_available(void) | |
65 | { | |
909ea964 | 66 | struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev); |
79bf2bb3 | 67 | |
3a142a06 TG |
68 | if (!dev || !(dev->features & CLOCK_EVT_FEAT_ONESHOT)) |
69 | return 0; | |
70 | if (!(dev->features & CLOCK_EVT_FEAT_C3STOP)) | |
71 | return 1; | |
72 | return tick_broadcast_oneshot_available(); | |
79bf2bb3 TG |
73 | } |
74 | ||
906568c9 TG |
75 | /* |
76 | * Periodic tick | |
77 | */ | |
78 | static void tick_periodic(int cpu) | |
79 | { | |
80 | if (tick_do_timer_cpu == cpu) { | |
d6ad4187 | 81 | write_seqlock(&jiffies_lock); |
906568c9 TG |
82 | |
83 | /* Keep track of the next tick event */ | |
84 | tick_next_period = ktime_add(tick_next_period, tick_period); | |
85 | ||
86 | do_timer(1); | |
d6ad4187 | 87 | write_sequnlock(&jiffies_lock); |
47a1b796 | 88 | update_wall_time(); |
906568c9 TG |
89 | } |
90 | ||
91 | update_process_times(user_mode(get_irq_regs())); | |
92 | profile_tick(CPU_PROFILING); | |
93 | } | |
94 | ||
95 | /* | |
96 | * Event handler for periodic ticks | |
97 | */ | |
98 | void tick_handle_periodic(struct clock_event_device *dev) | |
99 | { | |
100 | int cpu = smp_processor_id(); | |
b97f0291 | 101 | ktime_t next = dev->next_event; |
906568c9 TG |
102 | |
103 | tick_periodic(cpu); | |
104 | ||
77e32c89 | 105 | if (dev->state != CLOCK_EVT_STATE_ONESHOT) |
906568c9 | 106 | return; |
906568c9 | 107 | for (;;) { |
b97f0291 VK |
108 | /* |
109 | * Setup the next period for devices, which do not have | |
110 | * periodic mode: | |
111 | */ | |
112 | next = ktime_add(next, tick_period); | |
113 | ||
d1748302 | 114 | if (!clockevents_program_event(dev, next, false)) |
906568c9 | 115 | return; |
74a03b69 JS |
116 | /* |
117 | * Have to be careful here. If we're in oneshot mode, | |
118 | * before we call tick_periodic() in a loop, we need | |
119 | * to be sure we're using a real hardware clocksource. | |
120 | * Otherwise we could get trapped in an infinite | |
121 | * loop, as the tick_periodic() increments jiffies, | |
cacb3c76 | 122 | * which then will increment time, possibly causing |
74a03b69 JS |
123 | * the loop to trigger again and again. |
124 | */ | |
125 | if (timekeeping_valid_for_hres()) | |
126 | tick_periodic(cpu); | |
906568c9 TG |
127 | } |
128 | } | |
129 | ||
130 | /* | |
131 | * Setup the device for a periodic tick | |
132 | */ | |
f8381cba | 133 | void tick_setup_periodic(struct clock_event_device *dev, int broadcast) |
906568c9 | 134 | { |
f8381cba TG |
135 | tick_set_periodic_handler(dev, broadcast); |
136 | ||
137 | /* Broadcast setup ? */ | |
138 | if (!tick_device_is_functional(dev)) | |
139 | return; | |
906568c9 | 140 | |
27ce4cb4 TG |
141 | if ((dev->features & CLOCK_EVT_FEAT_PERIODIC) && |
142 | !tick_broadcast_oneshot_active()) { | |
77e32c89 | 143 | clockevents_set_state(dev, CLOCK_EVT_STATE_PERIODIC); |
906568c9 TG |
144 | } else { |
145 | unsigned long seq; | |
146 | ktime_t next; | |
147 | ||
148 | do { | |
d6ad4187 | 149 | seq = read_seqbegin(&jiffies_lock); |
906568c9 | 150 | next = tick_next_period; |
d6ad4187 | 151 | } while (read_seqretry(&jiffies_lock, seq)); |
906568c9 | 152 | |
77e32c89 | 153 | clockevents_set_state(dev, CLOCK_EVT_STATE_ONESHOT); |
906568c9 TG |
154 | |
155 | for (;;) { | |
d1748302 | 156 | if (!clockevents_program_event(dev, next, false)) |
906568c9 TG |
157 | return; |
158 | next = ktime_add(next, tick_period); | |
159 | } | |
160 | } | |
161 | } | |
162 | ||
163 | /* | |
164 | * Setup the tick device | |
165 | */ | |
166 | static void tick_setup_device(struct tick_device *td, | |
167 | struct clock_event_device *newdev, int cpu, | |
0de26520 | 168 | const struct cpumask *cpumask) |
906568c9 TG |
169 | { |
170 | ktime_t next_event; | |
171 | void (*handler)(struct clock_event_device *) = NULL; | |
172 | ||
173 | /* | |
174 | * First device setup ? | |
175 | */ | |
176 | if (!td->evtdev) { | |
177 | /* | |
178 | * If no cpu took the do_timer update, assign it to | |
179 | * this cpu: | |
180 | */ | |
6441402b | 181 | if (tick_do_timer_cpu == TICK_DO_TIMER_BOOT) { |
c5bfece2 | 182 | if (!tick_nohz_full_cpu(cpu)) |
a382bf93 FW |
183 | tick_do_timer_cpu = cpu; |
184 | else | |
185 | tick_do_timer_cpu = TICK_DO_TIMER_NONE; | |
906568c9 TG |
186 | tick_next_period = ktime_get(); |
187 | tick_period = ktime_set(0, NSEC_PER_SEC / HZ); | |
188 | } | |
189 | ||
190 | /* | |
191 | * Startup in periodic mode first. | |
192 | */ | |
193 | td->mode = TICKDEV_MODE_PERIODIC; | |
194 | } else { | |
195 | handler = td->evtdev->event_handler; | |
196 | next_event = td->evtdev->next_event; | |
7c1e7689 | 197 | td->evtdev->event_handler = clockevents_handle_noop; |
906568c9 TG |
198 | } |
199 | ||
200 | td->evtdev = newdev; | |
201 | ||
202 | /* | |
203 | * When the device is not per cpu, pin the interrupt to the | |
204 | * current cpu: | |
205 | */ | |
320ab2b0 | 206 | if (!cpumask_equal(newdev->cpumask, cpumask)) |
0de26520 | 207 | irq_set_affinity(newdev->irq, cpumask); |
906568c9 | 208 | |
f8381cba TG |
209 | /* |
210 | * When global broadcasting is active, check if the current | |
211 | * device is registered as a placeholder for broadcast mode. | |
212 | * This allows us to handle this x86 misfeature in a generic | |
07bd1172 TG |
213 | * way. This function also returns !=0 when we keep the |
214 | * current active broadcast state for this CPU. | |
f8381cba TG |
215 | */ |
216 | if (tick_device_uses_broadcast(newdev, cpu)) | |
217 | return; | |
218 | ||
906568c9 TG |
219 | if (td->mode == TICKDEV_MODE_PERIODIC) |
220 | tick_setup_periodic(newdev, 0); | |
79bf2bb3 TG |
221 | else |
222 | tick_setup_oneshot(newdev, handler, next_event); | |
906568c9 TG |
223 | } |
224 | ||
03e13cf5 TG |
225 | void tick_install_replacement(struct clock_event_device *newdev) |
226 | { | |
22127e93 | 227 | struct tick_device *td = this_cpu_ptr(&tick_cpu_device); |
03e13cf5 TG |
228 | int cpu = smp_processor_id(); |
229 | ||
230 | clockevents_exchange_device(td->evtdev, newdev); | |
231 | tick_setup_device(td, newdev, cpu, cpumask_of(cpu)); | |
232 | if (newdev->features & CLOCK_EVT_FEAT_ONESHOT) | |
233 | tick_oneshot_notify(); | |
234 | } | |
235 | ||
45cb8e01 TG |
236 | static bool tick_check_percpu(struct clock_event_device *curdev, |
237 | struct clock_event_device *newdev, int cpu) | |
238 | { | |
239 | if (!cpumask_test_cpu(cpu, newdev->cpumask)) | |
240 | return false; | |
241 | if (cpumask_equal(newdev->cpumask, cpumask_of(cpu))) | |
242 | return true; | |
243 | /* Check if irq affinity can be set */ | |
244 | if (newdev->irq >= 0 && !irq_can_set_affinity(newdev->irq)) | |
245 | return false; | |
246 | /* Prefer an existing cpu local device */ | |
247 | if (curdev && cpumask_equal(curdev->cpumask, cpumask_of(cpu))) | |
248 | return false; | |
249 | return true; | |
250 | } | |
251 | ||
252 | static bool tick_check_preferred(struct clock_event_device *curdev, | |
253 | struct clock_event_device *newdev) | |
254 | { | |
255 | /* Prefer oneshot capable device */ | |
256 | if (!(newdev->features & CLOCK_EVT_FEAT_ONESHOT)) { | |
257 | if (curdev && (curdev->features & CLOCK_EVT_FEAT_ONESHOT)) | |
258 | return false; | |
259 | if (tick_oneshot_mode_active()) | |
260 | return false; | |
261 | } | |
262 | ||
70e5975d SB |
263 | /* |
264 | * Use the higher rated one, but prefer a CPU local device with a lower | |
265 | * rating than a non-CPU local device | |
266 | */ | |
267 | return !curdev || | |
268 | newdev->rating > curdev->rating || | |
269 | !cpumask_equal(curdev->cpumask, newdev->cpumask); | |
45cb8e01 TG |
270 | } |
271 | ||
03e13cf5 TG |
272 | /* |
273 | * Check whether the new device is a better fit than curdev. curdev | |
274 | * can be NULL ! | |
275 | */ | |
276 | bool tick_check_replacement(struct clock_event_device *curdev, | |
277 | struct clock_event_device *newdev) | |
278 | { | |
521c4299 | 279 | if (!tick_check_percpu(curdev, newdev, smp_processor_id())) |
03e13cf5 TG |
280 | return false; |
281 | ||
282 | return tick_check_preferred(curdev, newdev); | |
283 | } | |
284 | ||
906568c9 | 285 | /* |
7126cac4 TG |
286 | * Check, if the new registered device should be used. Called with |
287 | * clockevents_lock held and interrupts disabled. | |
906568c9 | 288 | */ |
7172a286 | 289 | void tick_check_new_device(struct clock_event_device *newdev) |
906568c9 TG |
290 | { |
291 | struct clock_event_device *curdev; | |
292 | struct tick_device *td; | |
7172a286 | 293 | int cpu; |
906568c9 TG |
294 | |
295 | cpu = smp_processor_id(); | |
320ab2b0 | 296 | if (!cpumask_test_cpu(cpu, newdev->cpumask)) |
4a93232d | 297 | goto out_bc; |
906568c9 TG |
298 | |
299 | td = &per_cpu(tick_cpu_device, cpu); | |
300 | curdev = td->evtdev; | |
906568c9 TG |
301 | |
302 | /* cpu local device ? */ | |
45cb8e01 TG |
303 | if (!tick_check_percpu(curdev, newdev, cpu)) |
304 | goto out_bc; | |
906568c9 | 305 | |
45cb8e01 TG |
306 | /* Preference decision */ |
307 | if (!tick_check_preferred(curdev, newdev)) | |
308 | goto out_bc; | |
906568c9 | 309 | |
ccf33d68 TG |
310 | if (!try_module_get(newdev->owner)) |
311 | return; | |
312 | ||
906568c9 TG |
313 | /* |
314 | * Replace the eventually existing device by the new | |
f8381cba TG |
315 | * device. If the current device is the broadcast device, do |
316 | * not give it back to the clockevents layer ! | |
906568c9 | 317 | */ |
f8381cba | 318 | if (tick_is_broadcast_device(curdev)) { |
2344abbc | 319 | clockevents_shutdown(curdev); |
f8381cba TG |
320 | curdev = NULL; |
321 | } | |
906568c9 | 322 | clockevents_exchange_device(curdev, newdev); |
6b954823 | 323 | tick_setup_device(td, newdev, cpu, cpumask_of(cpu)); |
79bf2bb3 TG |
324 | if (newdev->features & CLOCK_EVT_FEAT_ONESHOT) |
325 | tick_oneshot_notify(); | |
7172a286 | 326 | return; |
f8381cba TG |
327 | |
328 | out_bc: | |
329 | /* | |
330 | * Can the new device be used as a broadcast device ? | |
331 | */ | |
7172a286 | 332 | tick_install_broadcast_device(newdev); |
906568c9 TG |
333 | } |
334 | ||
52c063d1 | 335 | #ifdef CONFIG_HOTPLUG_CPU |
94df7de0 SD |
336 | /* |
337 | * Transfer the do_timer job away from a dying cpu. | |
338 | * | |
52c063d1 TG |
339 | * Called with interrupts disabled. Not locking required. If |
340 | * tick_do_timer_cpu is owned by this cpu, nothing can change it. | |
94df7de0 | 341 | */ |
52c063d1 | 342 | void tick_handover_do_timer(void) |
94df7de0 | 343 | { |
52c063d1 | 344 | if (tick_do_timer_cpu == smp_processor_id()) { |
94df7de0 SD |
345 | int cpu = cpumask_first(cpu_online_mask); |
346 | ||
347 | tick_do_timer_cpu = (cpu < nr_cpu_ids) ? cpu : | |
348 | TICK_DO_TIMER_NONE; | |
349 | } | |
350 | } | |
351 | ||
906568c9 TG |
352 | /* |
353 | * Shutdown an event device on a given cpu: | |
354 | * | |
355 | * This is called on a life CPU, when a CPU is dead. So we cannot | |
356 | * access the hardware device itself. | |
357 | * We just set the mode and remove it from the lists. | |
358 | */ | |
a49b116d | 359 | void tick_shutdown(unsigned int cpu) |
906568c9 | 360 | { |
a49b116d | 361 | struct tick_device *td = &per_cpu(tick_cpu_device, cpu); |
906568c9 | 362 | struct clock_event_device *dev = td->evtdev; |
906568c9 | 363 | |
906568c9 TG |
364 | td->mode = TICKDEV_MODE_PERIODIC; |
365 | if (dev) { | |
366 | /* | |
367 | * Prevent that the clock events layer tries to call | |
368 | * the set mode function! | |
369 | */ | |
77e32c89 | 370 | dev->state = CLOCK_EVT_STATE_DETACHED; |
906568c9 TG |
371 | dev->mode = CLOCK_EVT_MODE_UNUSED; |
372 | clockevents_exchange_device(dev, NULL); | |
6f7a05d7 | 373 | dev->event_handler = clockevents_handle_noop; |
906568c9 TG |
374 | td->evtdev = NULL; |
375 | } | |
906568c9 | 376 | } |
a49b116d | 377 | #endif |
906568c9 | 378 | |
4ffee521 | 379 | /** |
f46481d0 | 380 | * tick_suspend_local - Suspend the local tick device |
4ffee521 | 381 | * |
f46481d0 | 382 | * Called from the local cpu for freeze with interrupts disabled. |
4ffee521 TG |
383 | * |
384 | * No locks required. Nothing can change the per cpu device. | |
385 | */ | |
7270d11c | 386 | void tick_suspend_local(void) |
6321dd60 | 387 | { |
22127e93 | 388 | struct tick_device *td = this_cpu_ptr(&tick_cpu_device); |
6321dd60 | 389 | |
2344abbc | 390 | clockevents_shutdown(td->evtdev); |
6321dd60 TG |
391 | } |
392 | ||
4ffee521 | 393 | /** |
f46481d0 | 394 | * tick_resume_local - Resume the local tick device |
4ffee521 | 395 | * |
f46481d0 | 396 | * Called from the local CPU for unfreeze or XEN resume magic. |
4ffee521 TG |
397 | * |
398 | * No locks required. Nothing can change the per cpu device. | |
399 | */ | |
f46481d0 | 400 | void tick_resume_local(void) |
6321dd60 | 401 | { |
f46481d0 TG |
402 | struct tick_device *td = this_cpu_ptr(&tick_cpu_device); |
403 | bool broadcast = tick_resume_check_broadcast(); | |
6321dd60 | 404 | |
554ef387 | 405 | clockevents_tick_resume(td->evtdev); |
18de5bc4 TG |
406 | if (!broadcast) { |
407 | if (td->mode == TICKDEV_MODE_PERIODIC) | |
408 | tick_setup_periodic(td->evtdev, 0); | |
409 | else | |
410 | tick_resume_oneshot(); | |
411 | } | |
6321dd60 TG |
412 | } |
413 | ||
f46481d0 TG |
414 | /** |
415 | * tick_suspend - Suspend the tick and the broadcast device | |
416 | * | |
417 | * Called from syscore_suspend() via timekeeping_suspend with only one | |
418 | * CPU online and interrupts disabled or from tick_unfreeze() under | |
419 | * tick_freeze_lock. | |
420 | * | |
421 | * No locks required. Nothing can change the per cpu device. | |
422 | */ | |
423 | void tick_suspend(void) | |
424 | { | |
425 | tick_suspend_local(); | |
426 | tick_suspend_broadcast(); | |
427 | } | |
428 | ||
429 | /** | |
430 | * tick_resume - Resume the tick and the broadcast device | |
431 | * | |
432 | * Called from syscore_resume() via timekeeping_resume with only one | |
433 | * CPU online and interrupts disabled. | |
434 | * | |
435 | * No locks required. Nothing can change the per cpu device. | |
436 | */ | |
437 | void tick_resume(void) | |
438 | { | |
439 | tick_resume_broadcast(); | |
440 | tick_resume_local(); | |
441 | } | |
442 | ||
124cf911 RW |
443 | static DEFINE_RAW_SPINLOCK(tick_freeze_lock); |
444 | static unsigned int tick_freeze_depth; | |
445 | ||
446 | /** | |
447 | * tick_freeze - Suspend the local tick and (possibly) timekeeping. | |
448 | * | |
449 | * Check if this is the last online CPU executing the function and if so, | |
450 | * suspend timekeeping. Otherwise suspend the local tick. | |
451 | * | |
452 | * Call with interrupts disabled. Must be balanced with %tick_unfreeze(). | |
453 | * Interrupts must not be enabled before the subsequent %tick_unfreeze(). | |
454 | */ | |
455 | void tick_freeze(void) | |
456 | { | |
457 | raw_spin_lock(&tick_freeze_lock); | |
458 | ||
459 | tick_freeze_depth++; | |
460 | if (tick_freeze_depth == num_online_cpus()) { | |
461 | timekeeping_suspend(); | |
462 | } else { | |
f46481d0 | 463 | tick_suspend_local(); |
124cf911 RW |
464 | } |
465 | ||
466 | raw_spin_unlock(&tick_freeze_lock); | |
467 | } | |
468 | ||
469 | /** | |
470 | * tick_unfreeze - Resume the local tick and (possibly) timekeeping. | |
471 | * | |
472 | * Check if this is the first CPU executing the function and if so, resume | |
473 | * timekeeping. Otherwise resume the local tick. | |
474 | * | |
475 | * Call with interrupts disabled. Must be balanced with %tick_freeze(). | |
476 | * Interrupts must not be enabled after the preceding %tick_freeze(). | |
477 | */ | |
478 | void tick_unfreeze(void) | |
479 | { | |
480 | raw_spin_lock(&tick_freeze_lock); | |
481 | ||
482 | if (tick_freeze_depth == num_online_cpus()) | |
483 | timekeeping_resume(); | |
484 | else | |
485 | tick_resume(); | |
486 | ||
487 | tick_freeze_depth--; | |
488 | ||
489 | raw_spin_unlock(&tick_freeze_lock); | |
490 | } | |
491 | ||
906568c9 TG |
492 | /** |
493 | * tick_init - initialize the tick control | |
906568c9 TG |
494 | */ |
495 | void __init tick_init(void) | |
496 | { | |
b352bc1c | 497 | tick_broadcast_init(); |
a80e49e2 | 498 | tick_nohz_init(); |
906568c9 | 499 | } |