Merge tag 'v3.10.107' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / kernel / time / tick-broadcast.c
CommitLineData
f8381cba
TG
1/*
2 * linux/kernel/time/tick-broadcast.c
3 *
4 * This file contains functions which emulate a local clock-event
5 * device via a broadcast event source.
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>
f8381cba
TG
18#include <linux/percpu.h>
19#include <linux/profile.h>
20#include <linux/sched.h>
12ad1000 21#include <linux/smp.h>
409d4ffa 22#include <linux/module.h>
f8381cba
TG
23
24#include "tick-internal.h"
25
26/*
27 * Broadcast support for broken x86 hardware, where the local apic
28 * timer stops in C3 state.
29 */
30
a52f5c56 31static struct tick_device tick_broadcast_device;
b352bc1c 32static cpumask_var_t tick_broadcast_mask;
1c0d08e6 33static cpumask_var_t tick_broadcast_on;
b352bc1c 34static cpumask_var_t tmpmask;
b5f91da0 35static DEFINE_RAW_SPINLOCK(tick_broadcast_lock);
aa276e1c 36static int tick_broadcast_force;
f8381cba 37
5590a536
TG
38#ifdef CONFIG_TICK_ONESHOT
39static void tick_broadcast_clear_oneshot(int cpu);
40#else
41static inline void tick_broadcast_clear_oneshot(int cpu) { }
42#endif
43
289f480a
IM
44/*
45 * Debugging: see timer_list.c
46 */
47struct tick_device *tick_get_broadcast_device(void)
48{
49 return &tick_broadcast_device;
50}
51
6b954823 52struct cpumask *tick_get_broadcast_mask(void)
289f480a 53{
b352bc1c 54 return tick_broadcast_mask;
289f480a
IM
55}
56
f8381cba
TG
57/*
58 * Start the device in periodic mode
59 */
60static void tick_broadcast_start_periodic(struct clock_event_device *bc)
61{
18de5bc4 62 if (bc)
f8381cba
TG
63 tick_setup_periodic(bc, 1);
64}
65
66/*
67 * Check, if the device can be utilized as broadcast device:
68 */
9bae8ea0
TG
69static bool tick_check_broadcast_device(struct clock_event_device *curdev,
70 struct clock_event_device *newdev)
71{
72 if ((newdev->features & CLOCK_EVT_FEAT_DUMMY) ||
73 (newdev->features & CLOCK_EVT_FEAT_C3STOP))
74 return false;
75
76 if (tick_broadcast_device.mode == TICKDEV_MODE_ONESHOT &&
77 !(newdev->features & CLOCK_EVT_FEAT_ONESHOT))
78 return false;
79
80 return !curdev || newdev->rating > curdev->rating;
81}
82
83/*
84 * Conditionally install/replace broadcast device
85 */
e8d63033 86void tick_install_broadcast_device(struct clock_event_device *dev)
f8381cba 87{
6f7a05d7
TG
88 struct clock_event_device *cur = tick_broadcast_device.evtdev;
89
9bae8ea0 90 if (!tick_check_broadcast_device(cur, dev))
e8d63033 91 return;
9bae8ea0 92
409d4ffa
TG
93 if (!try_module_get(dev->owner))
94 return;
f8381cba 95
9bae8ea0 96 clockevents_exchange_device(cur, dev);
6f7a05d7
TG
97 if (cur)
98 cur->event_handler = clockevents_handle_noop;
f8381cba 99 tick_broadcast_device.evtdev = dev;
b352bc1c 100 if (!cpumask_empty(tick_broadcast_mask))
f8381cba 101 tick_broadcast_start_periodic(dev);
c038c1c4
SB
102 /*
103 * Inform all cpus about this. We might be in a situation
104 * where we did not switch to oneshot mode because the per cpu
105 * devices are affected by CLOCK_EVT_FEAT_C3STOP and the lack
106 * of a oneshot capable broadcast device. Without that
107 * notification the systems stays stuck in periodic mode
108 * forever.
109 */
110 if (dev->features & CLOCK_EVT_FEAT_ONESHOT)
111 tick_clock_notify();
f8381cba
TG
112}
113
114/*
115 * Check, if the device is the broadcast device
116 */
117int tick_is_broadcast_device(struct clock_event_device *dev)
118{
119 return (dev && tick_broadcast_device.evtdev == dev);
120}
121
12ad1000
MR
122static void err_broadcast(const struct cpumask *mask)
123{
124 pr_crit_once("Failed to broadcast timer tick. Some CPUs may be unresponsive.\n");
125}
126
5d1d9a29
MR
127static void tick_device_setup_broadcast_func(struct clock_event_device *dev)
128{
129 if (!dev->broadcast)
130 dev->broadcast = tick_broadcast;
131 if (!dev->broadcast) {
132 pr_warn_once("%s depends on broadcast, but no broadcast function available\n",
133 dev->name);
134 dev->broadcast = err_broadcast;
135 }
136}
137
f8381cba
TG
138/*
139 * Check, if the device is disfunctional and a place holder, which
140 * needs to be handled by the broadcast device.
141 */
142int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu)
143{
1c0d08e6 144 struct clock_event_device *bc = tick_broadcast_device.evtdev;
f8381cba 145 unsigned long flags;
1c0d08e6 146 int ret;
f8381cba 147
b5f91da0 148 raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
f8381cba
TG
149
150 /*
151 * Devices might be registered with both periodic and oneshot
152 * mode disabled. This signals, that the device needs to be
153 * operated from the broadcast device and is a placeholder for
154 * the cpu local device.
155 */
156 if (!tick_device_is_functional(dev)) {
157 dev->event_handler = tick_handle_periodic;
5d1d9a29 158 tick_device_setup_broadcast_func(dev);
b352bc1c 159 cpumask_set_cpu(cpu, tick_broadcast_mask);
1c0d08e6 160 tick_broadcast_start_periodic(bc);
f8381cba 161 ret = 1;
5590a536
TG
162 } else {
163 /*
1c0d08e6
TG
164 * Clear the broadcast bit for this cpu if the
165 * device is not power state affected.
5590a536 166 */
1c0d08e6 167 if (!(dev->features & CLOCK_EVT_FEAT_C3STOP))
b352bc1c 168 cpumask_clear_cpu(cpu, tick_broadcast_mask);
1c0d08e6 169 else
5d1d9a29 170 tick_device_setup_broadcast_func(dev);
1c0d08e6
TG
171
172 /*
173 * Clear the broadcast bit if the CPU is not in
174 * periodic broadcast on state.
175 */
176 if (!cpumask_test_cpu(cpu, tick_broadcast_on))
177 cpumask_clear_cpu(cpu, tick_broadcast_mask);
178
179 switch (tick_broadcast_device.mode) {
180 case TICKDEV_MODE_ONESHOT:
181 /*
182 * If the system is in oneshot mode we can
183 * unconditionally clear the oneshot mask bit,
184 * because the CPU is running and therefore
185 * not in an idle state which causes the power
186 * state affected device to stop. Let the
187 * caller initialize the device.
188 */
189 tick_broadcast_clear_oneshot(cpu);
190 ret = 0;
191 break;
192
193 case TICKDEV_MODE_PERIODIC:
194 /*
195 * If the system is in periodic mode, check
196 * whether the broadcast device can be
197 * switched off now.
198 */
199 if (cpumask_empty(tick_broadcast_mask) && bc)
200 clockevents_shutdown(bc);
201 /*
202 * If we kept the cpu in the broadcast mask,
203 * tell the caller to leave the per cpu device
204 * in shutdown state. The periodic interrupt
205 * is delivered by the broadcast device.
206 */
207 ret = cpumask_test_cpu(cpu, tick_broadcast_mask);
208 break;
209 default:
210 /* Nothing to do */
211 ret = 0;
212 break;
5590a536
TG
213 }
214 }
b5f91da0 215 raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
f8381cba
TG
216 return ret;
217}
218
12572dbb
MR
219#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
220int tick_receive_broadcast(void)
221{
222 struct tick_device *td = this_cpu_ptr(&tick_cpu_device);
223 struct clock_event_device *evt = td->evtdev;
224
225 if (!evt)
226 return -ENODEV;
227
228 if (!evt->event_handler)
229 return -EINVAL;
230
231 evt->event_handler(evt);
232 return 0;
233}
234#endif
235
f8381cba 236/*
6b954823 237 * Broadcast the event to the cpus, which are set in the mask (mangled).
f8381cba 238 */
6b954823 239static void tick_do_broadcast(struct cpumask *mask)
f8381cba 240{
186e3cb8 241 int cpu = smp_processor_id();
f8381cba
TG
242 struct tick_device *td;
243
244 /*
245 * Check, if the current cpu is in the mask
246 */
6b954823
RR
247 if (cpumask_test_cpu(cpu, mask)) {
248 cpumask_clear_cpu(cpu, mask);
f8381cba
TG
249 td = &per_cpu(tick_cpu_device, cpu);
250 td->evtdev->event_handler(td->evtdev);
f8381cba
TG
251 }
252
6b954823 253 if (!cpumask_empty(mask)) {
f8381cba
TG
254 /*
255 * It might be necessary to actually check whether the devices
256 * have different broadcast functions. For now, just use the
257 * one of the first device. This works as long as we have this
258 * misfeature only on x86 (lapic)
259 */
6b954823
RR
260 td = &per_cpu(tick_cpu_device, cpumask_first(mask));
261 td->evtdev->broadcast(mask);
f8381cba 262 }
f8381cba
TG
263}
264
265/*
266 * Periodic broadcast:
267 * - invoke the broadcast handlers
268 */
269static void tick_do_periodic_broadcast(void)
270{
b5f91da0 271 raw_spin_lock(&tick_broadcast_lock);
f8381cba 272
b352bc1c
TG
273 cpumask_and(tmpmask, cpu_online_mask, tick_broadcast_mask);
274 tick_do_broadcast(tmpmask);
f8381cba 275
b5f91da0 276 raw_spin_unlock(&tick_broadcast_lock);
f8381cba
TG
277}
278
279/*
280 * Event handler for periodic broadcast ticks
281 */
282static void tick_handle_periodic_broadcast(struct clock_event_device *dev)
283{
d4496b39
TG
284 ktime_t next;
285
f8381cba
TG
286 tick_do_periodic_broadcast();
287
288 /*
289 * The device is in periodic mode. No reprogramming necessary:
290 */
291 if (dev->mode == CLOCK_EVT_MODE_PERIODIC)
292 return;
293
294 /*
295 * Setup the next period for devices, which do not have
d4496b39 296 * periodic mode. We read dev->next_event first and add to it
698f9315 297 * when the event already expired. clockevents_program_event()
d4496b39
TG
298 * sets dev->next_event only when the event is really
299 * programmed to the device.
f8381cba 300 */
d4496b39
TG
301 for (next = dev->next_event; ;) {
302 next = ktime_add(next, tick_period);
f8381cba 303
d1748302 304 if (!clockevents_program_event(dev, next, false))
f8381cba
TG
305 return;
306 tick_do_periodic_broadcast();
307 }
308}
309
310/*
311 * Powerstate information: The system enters/leaves a state, where
312 * affected devices might stop
313 */
f833bab8 314static void tick_do_broadcast_on_off(unsigned long *reason)
f8381cba
TG
315{
316 struct clock_event_device *bc, *dev;
317 struct tick_device *td;
f833bab8 318 unsigned long flags;
9c17bcda 319 int cpu, bc_stopped;
f8381cba 320
b5f91da0 321 raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
f8381cba
TG
322
323 cpu = smp_processor_id();
324 td = &per_cpu(tick_cpu_device, cpu);
325 dev = td->evtdev;
326 bc = tick_broadcast_device.evtdev;
327
328 /*
1595f452 329 * Is the device not affected by the powerstate ?
f8381cba 330 */
1595f452 331 if (!dev || !(dev->features & CLOCK_EVT_FEAT_C3STOP))
f8381cba
TG
332 goto out;
333
3dfbc884
TG
334 if (!tick_device_is_functional(dev))
335 goto out;
1595f452 336
b352bc1c 337 bc_stopped = cpumask_empty(tick_broadcast_mask);
9c17bcda 338
1595f452
TG
339 switch (*reason) {
340 case CLOCK_EVT_NOTIFY_BROADCAST_ON:
341 case CLOCK_EVT_NOTIFY_BROADCAST_FORCE:
1c0d08e6 342 cpumask_set_cpu(cpu, tick_broadcast_on);
b352bc1c 343 if (!cpumask_test_and_set_cpu(cpu, tick_broadcast_mask)) {
07454bff
TG
344 if (tick_broadcast_device.mode ==
345 TICKDEV_MODE_PERIODIC)
2344abbc 346 clockevents_shutdown(dev);
f8381cba 347 }
3dfbc884 348 if (*reason == CLOCK_EVT_NOTIFY_BROADCAST_FORCE)
aa276e1c 349 tick_broadcast_force = 1;
1595f452
TG
350 break;
351 case CLOCK_EVT_NOTIFY_BROADCAST_OFF:
1c0d08e6
TG
352 if (tick_broadcast_force)
353 break;
354 cpumask_clear_cpu(cpu, tick_broadcast_on);
355 if (!tick_device_is_functional(dev))
356 break;
357 if (cpumask_test_and_clear_cpu(cpu, tick_broadcast_mask)) {
07454bff
TG
358 if (tick_broadcast_device.mode ==
359 TICKDEV_MODE_PERIODIC)
f8381cba
TG
360 tick_setup_periodic(dev, 0);
361 }
1595f452 362 break;
f8381cba
TG
363 }
364
b352bc1c 365 if (cpumask_empty(tick_broadcast_mask)) {
9c17bcda 366 if (!bc_stopped)
2344abbc 367 clockevents_shutdown(bc);
9c17bcda 368 } else if (bc_stopped) {
f8381cba
TG
369 if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC)
370 tick_broadcast_start_periodic(bc);
79bf2bb3
TG
371 else
372 tick_broadcast_setup_oneshot(bc);
f8381cba
TG
373 }
374out:
b5f91da0 375 raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
f8381cba
TG
376}
377
378/*
379 * Powerstate information: The system enters/leaves a state, where
380 * affected devices might stop.
381 */
382void tick_broadcast_on_off(unsigned long reason, int *oncpu)
383{
6b954823 384 if (!cpumask_test_cpu(*oncpu, cpu_online_mask))
833df317 385 printk(KERN_ERR "tick-broadcast: ignoring broadcast for "
72fcde96 386 "offline CPU #%d\n", *oncpu);
bf020cb7 387 else
f833bab8 388 tick_do_broadcast_on_off(&reason);
f8381cba
TG
389}
390
391/*
392 * Set the periodic handler depending on broadcast on/off
393 */
394void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
395{
396 if (!broadcast)
397 dev->event_handler = tick_handle_periodic;
398 else
399 dev->event_handler = tick_handle_periodic_broadcast;
400}
401
402/*
403 * Remove a CPU from broadcasting
404 */
405void tick_shutdown_broadcast(unsigned int *cpup)
406{
407 struct clock_event_device *bc;
408 unsigned long flags;
409 unsigned int cpu = *cpup;
410
b5f91da0 411 raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
f8381cba
TG
412
413 bc = tick_broadcast_device.evtdev;
b352bc1c 414 cpumask_clear_cpu(cpu, tick_broadcast_mask);
1c0d08e6 415 cpumask_clear_cpu(cpu, tick_broadcast_on);
f8381cba
TG
416
417 if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) {
b352bc1c 418 if (bc && cpumask_empty(tick_broadcast_mask))
2344abbc 419 clockevents_shutdown(bc);
f8381cba
TG
420 }
421
b5f91da0 422 raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
f8381cba 423}
79bf2bb3 424
6321dd60
TG
425void tick_suspend_broadcast(void)
426{
427 struct clock_event_device *bc;
428 unsigned long flags;
429
b5f91da0 430 raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
6321dd60
TG
431
432 bc = tick_broadcast_device.evtdev;
18de5bc4 433 if (bc)
2344abbc 434 clockevents_shutdown(bc);
6321dd60 435
b5f91da0 436 raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
6321dd60
TG
437}
438
439int tick_resume_broadcast(void)
440{
441 struct clock_event_device *bc;
442 unsigned long flags;
443 int broadcast = 0;
444
b5f91da0 445 raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
6321dd60
TG
446
447 bc = tick_broadcast_device.evtdev;
6321dd60 448
cd05a1f8 449 if (bc) {
18de5bc4
TG
450 clockevents_set_mode(bc, CLOCK_EVT_MODE_RESUME);
451
cd05a1f8
TG
452 switch (tick_broadcast_device.mode) {
453 case TICKDEV_MODE_PERIODIC:
b352bc1c 454 if (!cpumask_empty(tick_broadcast_mask))
cd05a1f8 455 tick_broadcast_start_periodic(bc);
6b954823 456 broadcast = cpumask_test_cpu(smp_processor_id(),
b352bc1c 457 tick_broadcast_mask);
cd05a1f8
TG
458 break;
459 case TICKDEV_MODE_ONESHOT:
b352bc1c 460 if (!cpumask_empty(tick_broadcast_mask))
a6371f80 461 broadcast = tick_resume_broadcast_oneshot(bc);
cd05a1f8
TG
462 break;
463 }
6321dd60 464 }
b5f91da0 465 raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
6321dd60
TG
466
467 return broadcast;
468}
469
470
79bf2bb3
TG
471#ifdef CONFIG_TICK_ONESHOT
472
b352bc1c 473static cpumask_var_t tick_broadcast_oneshot_mask;
26517f3e 474static cpumask_var_t tick_broadcast_pending_mask;
989dcb64 475static cpumask_var_t tick_broadcast_force_mask;
79bf2bb3 476
289f480a 477/*
6b954823 478 * Exposed for debugging: see timer_list.c
289f480a 479 */
6b954823 480struct cpumask *tick_get_broadcast_oneshot_mask(void)
289f480a 481{
b352bc1c 482 return tick_broadcast_oneshot_mask;
289f480a
IM
483}
484
eaa907c5
TG
485/*
486 * Called before going idle with interrupts disabled. Checks whether a
487 * broadcast event from the other core is about to happen. We detected
488 * that in tick_broadcast_oneshot_control(). The callsite can use this
489 * to avoid a deep idle transition as we are about to get the
490 * broadcast IPI right away.
491 */
492int tick_check_broadcast_expired(void)
493{
494 return cpumask_test_cpu(smp_processor_id(), tick_broadcast_force_mask);
495}
496
d2348fb6
DL
497/*
498 * Set broadcast interrupt affinity
499 */
500static void tick_broadcast_set_affinity(struct clock_event_device *bc,
501 const struct cpumask *cpumask)
502{
503 if (!(bc->features & CLOCK_EVT_FEAT_DYNIRQ))
504 return;
505
506 if (cpumask_equal(bc->cpumask, cpumask))
507 return;
508
509 bc->cpumask = cpumask;
510 irq_set_affinity(bc->irq, bc->cpumask);
511}
512
513static int tick_broadcast_set_event(struct clock_event_device *bc, int cpu,
f9ae39d0 514 ktime_t expires, int force)
79bf2bb3 515{
d2348fb6
DL
516 int ret;
517
b9a6a235
TG
518 if (bc->mode != CLOCK_EVT_MODE_ONESHOT)
519 clockevents_set_mode(bc, CLOCK_EVT_MODE_ONESHOT);
520
d2348fb6
DL
521 ret = clockevents_program_event(bc, expires, force);
522 if (!ret)
523 tick_broadcast_set_affinity(bc, cpumask_of(cpu));
524 return ret;
79bf2bb3
TG
525}
526
cd05a1f8
TG
527int tick_resume_broadcast_oneshot(struct clock_event_device *bc)
528{
529 clockevents_set_mode(bc, CLOCK_EVT_MODE_ONESHOT);
b7e113dc 530 return 0;
cd05a1f8
TG
531}
532
fb02fbc1
TG
533/*
534 * Called from irq_enter() when idle was interrupted to reenable the
535 * per cpu device.
536 */
537void tick_check_oneshot_broadcast(int cpu)
538{
b352bc1c 539 if (cpumask_test_cpu(cpu, tick_broadcast_oneshot_mask)) {
fb02fbc1
TG
540 struct tick_device *td = &per_cpu(tick_cpu_device, cpu);
541
084c895d
TG
542 /*
543 * We might be in the middle of switching over from
544 * periodic to oneshot. If the CPU has not yet
545 * switched over, leave the device alone.
546 */
547 if (td->mode == TICKDEV_MODE_ONESHOT) {
548 clockevents_set_mode(td->evtdev,
549 CLOCK_EVT_MODE_ONESHOT);
550 }
fb02fbc1
TG
551 }
552}
553
79bf2bb3
TG
554/*
555 * Handle oneshot mode broadcasting
556 */
557static void tick_handle_oneshot_broadcast(struct clock_event_device *dev)
558{
559 struct tick_device *td;
cdc6f27d 560 ktime_t now, next_event;
d2348fb6 561 int cpu, next_cpu = 0;
79bf2bb3 562
b5f91da0 563 raw_spin_lock(&tick_broadcast_lock);
79bf2bb3
TG
564again:
565 dev->next_event.tv64 = KTIME_MAX;
cdc6f27d 566 next_event.tv64 = KTIME_MAX;
b352bc1c 567 cpumask_clear(tmpmask);
79bf2bb3
TG
568 now = ktime_get();
569 /* Find all expired events */
b352bc1c 570 for_each_cpu(cpu, tick_broadcast_oneshot_mask) {
79bf2bb3 571 td = &per_cpu(tick_cpu_device, cpu);
d2348fb6 572 if (td->evtdev->next_event.tv64 <= now.tv64) {
b352bc1c 573 cpumask_set_cpu(cpu, tmpmask);
26517f3e
TG
574 /*
575 * Mark the remote cpu in the pending mask, so
576 * it can avoid reprogramming the cpu local
577 * timer in tick_broadcast_oneshot_control().
578 */
579 cpumask_set_cpu(cpu, tick_broadcast_pending_mask);
d2348fb6 580 } else if (td->evtdev->next_event.tv64 < next_event.tv64) {
cdc6f27d 581 next_event.tv64 = td->evtdev->next_event.tv64;
d2348fb6
DL
582 next_cpu = cpu;
583 }
79bf2bb3
TG
584 }
585
2938d275
TG
586 /*
587 * Remove the current cpu from the pending mask. The event is
588 * delivered immediately in tick_do_broadcast() !
589 */
590 cpumask_clear_cpu(smp_processor_id(), tick_broadcast_pending_mask);
591
989dcb64
TG
592 /* Take care of enforced broadcast requests */
593 cpumask_or(tmpmask, tmpmask, tick_broadcast_force_mask);
594 cpumask_clear(tick_broadcast_force_mask);
595
a191212a
TG
596 /*
597 * Sanity check. Catch the case where we try to broadcast to
598 * offline cpus.
599 */
600 if (WARN_ON_ONCE(!cpumask_subset(tmpmask, cpu_online_mask)))
601 cpumask_and(tmpmask, tmpmask, cpu_online_mask);
602
79bf2bb3 603 /*
cdc6f27d
TG
604 * Wakeup the cpus which have an expired event.
605 */
b352bc1c 606 tick_do_broadcast(tmpmask);
cdc6f27d
TG
607
608 /*
609 * Two reasons for reprogram:
610 *
611 * - The global event did not expire any CPU local
612 * events. This happens in dyntick mode, as the maximum PIT
613 * delta is quite small.
614 *
615 * - There are pending events on sleeping CPUs which were not
616 * in the event mask
79bf2bb3 617 */
cdc6f27d 618 if (next_event.tv64 != KTIME_MAX) {
79bf2bb3 619 /*
cdc6f27d
TG
620 * Rearm the broadcast device. If event expired,
621 * repeat the above
79bf2bb3 622 */
d2348fb6 623 if (tick_broadcast_set_event(dev, next_cpu, next_event, 0))
79bf2bb3
TG
624 goto again;
625 }
b5f91da0 626 raw_spin_unlock(&tick_broadcast_lock);
79bf2bb3
TG
627}
628
629/*
630 * Powerstate information: The system enters/leaves a state, where
631 * affected devices might stop
632 */
633void tick_broadcast_oneshot_control(unsigned long reason)
634{
635 struct clock_event_device *bc, *dev;
636 struct tick_device *td;
637 unsigned long flags;
989dcb64 638 ktime_t now;
79bf2bb3
TG
639 int cpu;
640
79bf2bb3
TG
641 /*
642 * Periodic mode does not care about the enter/exit of power
643 * states
644 */
645 if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC)
7372b0b1 646 return;
79bf2bb3 647
7372b0b1
AK
648 /*
649 * We are called with preemtion disabled from the depth of the
650 * idle code, so we can't be moved away.
651 */
79bf2bb3
TG
652 cpu = smp_processor_id();
653 td = &per_cpu(tick_cpu_device, cpu);
654 dev = td->evtdev;
655
656 if (!(dev->features & CLOCK_EVT_FEAT_C3STOP))
7372b0b1
AK
657 return;
658
659 bc = tick_broadcast_device.evtdev;
79bf2bb3 660
7372b0b1 661 raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
79bf2bb3 662 if (reason == CLOCK_EVT_NOTIFY_BROADCAST_ENTER) {
b352bc1c 663 if (!cpumask_test_and_set_cpu(cpu, tick_broadcast_oneshot_mask)) {
2938d275 664 WARN_ON_ONCE(cpumask_test_cpu(cpu, tick_broadcast_pending_mask));
79bf2bb3 665 clockevents_set_mode(dev, CLOCK_EVT_MODE_SHUTDOWN);
989dcb64
TG
666 /*
667 * We only reprogram the broadcast timer if we
668 * did not mark ourself in the force mask and
669 * if the cpu local event is earlier than the
670 * broadcast event. If the current CPU is in
671 * the force mask, then we are going to be
672 * woken by the IPI right away.
673 */
674 if (!cpumask_test_cpu(cpu, tick_broadcast_force_mask) &&
675 dev->next_event.tv64 < bc->next_event.tv64)
d2348fb6 676 tick_broadcast_set_event(bc, cpu, dev->next_event, 1);
79bf2bb3
TG
677 }
678 } else {
b352bc1c 679 if (cpumask_test_and_clear_cpu(cpu, tick_broadcast_oneshot_mask)) {
79bf2bb3 680 clockevents_set_mode(dev, CLOCK_EVT_MODE_ONESHOT);
26517f3e
TG
681 /*
682 * The cpu which was handling the broadcast
683 * timer marked this cpu in the broadcast
684 * pending mask and fired the broadcast
685 * IPI. So we are going to handle the expired
686 * event anyway via the broadcast IPI
687 * handler. No need to reprogram the timer
688 * with an already expired event.
689 */
690 if (cpumask_test_and_clear_cpu(cpu,
691 tick_broadcast_pending_mask))
692 goto out;
693
ea8deb8d
DL
694 /*
695 * Bail out if there is no next event.
696 */
697 if (dev->next_event.tv64 == KTIME_MAX)
698 goto out;
989dcb64
TG
699 /*
700 * If the pending bit is not set, then we are
701 * either the CPU handling the broadcast
702 * interrupt or we got woken by something else.
703 *
704 * We are not longer in the broadcast mask, so
705 * if the cpu local expiry time is already
706 * reached, we would reprogram the cpu local
707 * timer with an already expired event.
708 *
709 * This can lead to a ping-pong when we return
710 * to idle and therefor rearm the broadcast
711 * timer before the cpu local timer was able
712 * to fire. This happens because the forced
713 * reprogramming makes sure that the event
714 * will happen in the future and depending on
715 * the min_delta setting this might be far
716 * enough out that the ping-pong starts.
717 *
718 * If the cpu local next_event has expired
719 * then we know that the broadcast timer
720 * next_event has expired as well and
721 * broadcast is about to be handled. So we
722 * avoid reprogramming and enforce that the
723 * broadcast handler, which did not run yet,
724 * will invoke the cpu local handler.
725 *
726 * We cannot call the handler directly from
727 * here, because we might be in a NOHZ phase
728 * and we did not go through the irq_enter()
729 * nohz fixups.
730 */
731 now = ktime_get();
732 if (dev->next_event.tv64 <= now.tv64) {
733 cpumask_set_cpu(cpu, tick_broadcast_force_mask);
734 goto out;
735 }
736 /*
737 * We got woken by something else. Reprogram
738 * the cpu local timer device.
739 */
26517f3e 740 tick_program_event(dev->next_event, 1);
79bf2bb3
TG
741 }
742 }
26517f3e 743out:
b5f91da0 744 raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
79bf2bb3
TG
745}
746
5590a536
TG
747/*
748 * Reset the one shot broadcast for a cpu
749 *
750 * Called with tick_broadcast_lock held
751 */
752static void tick_broadcast_clear_oneshot(int cpu)
753{
b352bc1c 754 cpumask_clear_cpu(cpu, tick_broadcast_oneshot_mask);
dbd51587 755 cpumask_clear_cpu(cpu, tick_broadcast_pending_mask);
5590a536
TG
756}
757
6b954823
RR
758static void tick_broadcast_init_next_event(struct cpumask *mask,
759 ktime_t expires)
7300711e
TG
760{
761 struct tick_device *td;
762 int cpu;
763
5db0e1e9 764 for_each_cpu(cpu, mask) {
7300711e
TG
765 td = &per_cpu(tick_cpu_device, cpu);
766 if (td->evtdev)
767 td->evtdev->next_event = expires;
768 }
769}
770
79bf2bb3 771/**
8dce39c2 772 * tick_broadcast_setup_oneshot - setup the broadcast device
79bf2bb3
TG
773 */
774void tick_broadcast_setup_oneshot(struct clock_event_device *bc)
775{
07f4beb0
TG
776 int cpu = smp_processor_id();
777
468316bf
TG
778 if (!bc)
779 return;
780
9c17bcda
TG
781 /* Set it up only once ! */
782 if (bc->event_handler != tick_handle_oneshot_broadcast) {
7300711e 783 int was_periodic = bc->mode == CLOCK_EVT_MODE_PERIODIC;
7300711e 784
9c17bcda 785 bc->event_handler = tick_handle_oneshot_broadcast;
7300711e 786
7300711e
TG
787 /*
788 * We must be careful here. There might be other CPUs
789 * waiting for periodic broadcast. We need to set the
790 * oneshot_mask bits for those and program the
791 * broadcast device to fire.
792 */
b352bc1c
TG
793 cpumask_copy(tmpmask, tick_broadcast_mask);
794 cpumask_clear_cpu(cpu, tmpmask);
795 cpumask_or(tick_broadcast_oneshot_mask,
796 tick_broadcast_oneshot_mask, tmpmask);
6b954823 797
b352bc1c 798 if (was_periodic && !cpumask_empty(tmpmask)) {
b435092f 799 clockevents_set_mode(bc, CLOCK_EVT_MODE_ONESHOT);
b352bc1c 800 tick_broadcast_init_next_event(tmpmask,
6b954823 801 tick_next_period);
d2348fb6 802 tick_broadcast_set_event(bc, cpu, tick_next_period, 1);
7300711e
TG
803 } else
804 bc->next_event.tv64 = KTIME_MAX;
07f4beb0
TG
805 } else {
806 /*
807 * The first cpu which switches to oneshot mode sets
808 * the bit for all other cpus which are in the general
809 * (periodic) broadcast mask. So the bit is set and
810 * would prevent the first broadcast enter after this
811 * to program the bc device.
812 */
813 tick_broadcast_clear_oneshot(cpu);
9c17bcda 814 }
79bf2bb3
TG
815}
816
817/*
818 * Select oneshot operating mode for the broadcast device
819 */
820void tick_broadcast_switch_to_oneshot(void)
821{
822 struct clock_event_device *bc;
823 unsigned long flags;
824
b5f91da0 825 raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
fa4da365
SS
826
827 tick_broadcast_device.mode = TICKDEV_MODE_ONESHOT;
79bf2bb3
TG
828 bc = tick_broadcast_device.evtdev;
829 if (bc)
830 tick_broadcast_setup_oneshot(bc);
77b0d60c 831
b5f91da0 832 raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
79bf2bb3
TG
833}
834
835
836/*
837 * Remove a dead CPU from broadcasting
838 */
839void tick_shutdown_broadcast_oneshot(unsigned int *cpup)
840{
79bf2bb3
TG
841 unsigned long flags;
842 unsigned int cpu = *cpup;
843
b5f91da0 844 raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
79bf2bb3 845
31d9b393 846 /*
a191212a
TG
847 * Clear the broadcast masks for the dead cpu, but do not stop
848 * the broadcast device!
31d9b393 849 */
b352bc1c 850 cpumask_clear_cpu(cpu, tick_broadcast_oneshot_mask);
a191212a
TG
851 cpumask_clear_cpu(cpu, tick_broadcast_pending_mask);
852 cpumask_clear_cpu(cpu, tick_broadcast_force_mask);
79bf2bb3 853
b5f91da0 854 raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
79bf2bb3
TG
855}
856
27ce4cb4
TG
857/*
858 * Check, whether the broadcast device is in one shot mode
859 */
860int tick_broadcast_oneshot_active(void)
861{
862 return tick_broadcast_device.mode == TICKDEV_MODE_ONESHOT;
863}
864
3a142a06
TG
865/*
866 * Check whether the broadcast device supports oneshot.
867 */
868bool tick_broadcast_oneshot_available(void)
869{
870 struct clock_event_device *bc = tick_broadcast_device.evtdev;
871
872 return bc ? bc->features & CLOCK_EVT_FEAT_ONESHOT : false;
873}
874
79bf2bb3 875#endif
b352bc1c
TG
876
877void __init tick_broadcast_init(void)
878{
fbd44a60 879 zalloc_cpumask_var(&tick_broadcast_mask, GFP_NOWAIT);
1c0d08e6 880 zalloc_cpumask_var(&tick_broadcast_on, GFP_NOWAIT);
fbd44a60 881 zalloc_cpumask_var(&tmpmask, GFP_NOWAIT);
b352bc1c 882#ifdef CONFIG_TICK_ONESHOT
fbd44a60
TG
883 zalloc_cpumask_var(&tick_broadcast_oneshot_mask, GFP_NOWAIT);
884 zalloc_cpumask_var(&tick_broadcast_pending_mask, GFP_NOWAIT);
885 zalloc_cpumask_var(&tick_broadcast_force_mask, GFP_NOWAIT);
b352bc1c
TG
886#endif
887}