tick-broadcast.c 16.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/*
 * linux/kernel/time/tick-broadcast.c
 *
 * This file contains functions which emulate a local clock-event
 * device via a broadcast event source.
 *
 * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
 * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
 * Copyright(C) 2006-2007, Timesys Corp., Thomas Gleixner
 *
 * This code is licenced under the GPL version 2. For details see
 * kernel-base/COPYING.
 */
#include <linux/cpu.h>
#include <linux/err.h>
#include <linux/hrtimer.h>
17
#include <linux/interrupt.h>
18 19 20
#include <linux/percpu.h>
#include <linux/profile.h>
#include <linux/sched.h>
21
#include <linux/smp.h>
22 23 24 25 26 27 28 29

#include "tick-internal.h"

/*
 * Broadcast support for broken x86 hardware, where the local apic
 * timer stops in C3 state.
 */

30
static struct tick_device tick_broadcast_device;
31 32
static cpumask_var_t tick_broadcast_mask;
static cpumask_var_t tmpmask;
33
static DEFINE_RAW_SPINLOCK(tick_broadcast_lock);
34
static int tick_broadcast_force;
35

36 37 38 39 40 41
#ifdef CONFIG_TICK_ONESHOT
static void tick_broadcast_clear_oneshot(int cpu);
#else
static inline void tick_broadcast_clear_oneshot(int cpu) { }
#endif

42 43 44 45 46 47 48 49
/*
 * Debugging: see timer_list.c
 */
struct tick_device *tick_get_broadcast_device(void)
{
	return &tick_broadcast_device;
}

50
struct cpumask *tick_get_broadcast_mask(void)
51
{
52
	return tick_broadcast_mask;
53 54
}

55 56 57 58 59
/*
 * Start the device in periodic mode
 */
static void tick_broadcast_start_periodic(struct clock_event_device *bc)
{
T
Thomas Gleixner 已提交
60
	if (bc)
61 62 63 64 65 66 67 68
		tick_setup_periodic(bc, 1);
}

/*
 * Check, if the device can be utilized as broadcast device:
 */
int tick_check_broadcast_device(struct clock_event_device *dev)
{
69 70 71
	if ((tick_broadcast_device.evtdev &&
	     tick_broadcast_device.evtdev->rating >= dev->rating) ||
	     (dev->features & CLOCK_EVT_FEAT_C3STOP))
72 73
		return 0;

74
	clockevents_exchange_device(tick_broadcast_device.evtdev, dev);
75
	tick_broadcast_device.evtdev = dev;
76
	if (!cpumask_empty(tick_broadcast_mask))
77 78 79 80 81 82 83 84 85 86 87 88
		tick_broadcast_start_periodic(dev);
	return 1;
}

/*
 * Check, if the device is the broadcast device
 */
int tick_is_broadcast_device(struct clock_event_device *dev)
{
	return (dev && tick_broadcast_device.evtdev == dev);
}

89 90 91 92 93
static void err_broadcast(const struct cpumask *mask)
{
	pr_crit_once("Failed to broadcast timer tick. Some CPUs may be unresponsive.\n");
}

94 95 96 97 98 99 100 101 102 103 104
static void tick_device_setup_broadcast_func(struct clock_event_device *dev)
{
	if (!dev->broadcast)
		dev->broadcast = tick_broadcast;
	if (!dev->broadcast) {
		pr_warn_once("%s depends on broadcast, but no broadcast function available\n",
			     dev->name);
		dev->broadcast = err_broadcast;
	}
}

105 106 107 108 109 110 111 112 113
/*
 * Check, if the device is disfunctional and a place holder, which
 * needs to be handled by the broadcast device.
 */
int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu)
{
	unsigned long flags;
	int ret = 0;

114
	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
115 116 117 118 119 120 121 122 123

	/*
	 * Devices might be registered with both periodic and oneshot
	 * mode disabled. This signals, that the device needs to be
	 * operated from the broadcast device and is a placeholder for
	 * the cpu local device.
	 */
	if (!tick_device_is_functional(dev)) {
		dev->event_handler = tick_handle_periodic;
124
		tick_device_setup_broadcast_func(dev);
125
		cpumask_set_cpu(cpu, tick_broadcast_mask);
126 127
		tick_broadcast_start_periodic(tick_broadcast_device.evtdev);
		ret = 1;
128 129 130 131 132 133 134 135
	} else {
		/*
		 * When the new device is not affected by the stop
		 * feature and the cpu is marked in the broadcast mask
		 * then clear the broadcast bit.
		 */
		if (!(dev->features & CLOCK_EVT_FEAT_C3STOP)) {
			int cpu = smp_processor_id();
136
			cpumask_clear_cpu(cpu, tick_broadcast_mask);
137
			tick_broadcast_clear_oneshot(cpu);
138 139
		} else {
			tick_device_setup_broadcast_func(dev);
140 141
		}
	}
142
	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
143 144 145
	return ret;
}

146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
int tick_receive_broadcast(void)
{
	struct tick_device *td = this_cpu_ptr(&tick_cpu_device);
	struct clock_event_device *evt = td->evtdev;

	if (!evt)
		return -ENODEV;

	if (!evt->event_handler)
		return -EINVAL;

	evt->event_handler(evt);
	return 0;
}
#endif

163
/*
164
 * Broadcast the event to the cpus, which are set in the mask (mangled).
165
 */
166
static void tick_do_broadcast(struct cpumask *mask)
167
{
168
	int cpu = smp_processor_id();
169 170 171 172 173
	struct tick_device *td;

	/*
	 * Check, if the current cpu is in the mask
	 */
174 175
	if (cpumask_test_cpu(cpu, mask)) {
		cpumask_clear_cpu(cpu, mask);
176 177 178 179
		td = &per_cpu(tick_cpu_device, cpu);
		td->evtdev->event_handler(td->evtdev);
	}

180
	if (!cpumask_empty(mask)) {
181 182 183 184 185 186
		/*
		 * It might be necessary to actually check whether the devices
		 * have different broadcast functions. For now, just use the
		 * one of the first device. This works as long as we have this
		 * misfeature only on x86 (lapic)
		 */
187 188
		td = &per_cpu(tick_cpu_device, cpumask_first(mask));
		td->evtdev->broadcast(mask);
189 190 191 192 193 194 195 196 197
	}
}

/*
 * Periodic broadcast:
 * - invoke the broadcast handlers
 */
static void tick_do_periodic_broadcast(void)
{
198
	raw_spin_lock(&tick_broadcast_lock);
199

200 201
	cpumask_and(tmpmask, cpu_online_mask, tick_broadcast_mask);
	tick_do_broadcast(tmpmask);
202

203
	raw_spin_unlock(&tick_broadcast_lock);
204 205 206 207 208 209 210
}

/*
 * Event handler for periodic broadcast ticks
 */
static void tick_handle_periodic_broadcast(struct clock_event_device *dev)
{
211 212
	ktime_t next;

213 214 215 216 217 218 219 220 221 222
	tick_do_periodic_broadcast();

	/*
	 * The device is in periodic mode. No reprogramming necessary:
	 */
	if (dev->mode == CLOCK_EVT_MODE_PERIODIC)
		return;

	/*
	 * Setup the next period for devices, which do not have
223
	 * periodic mode. We read dev->next_event first and add to it
224
	 * when the event already expired. clockevents_program_event()
225 226
	 * sets dev->next_event only when the event is really
	 * programmed to the device.
227
	 */
228 229
	for (next = dev->next_event; ;) {
		next = ktime_add(next, tick_period);
230

231
		if (!clockevents_program_event(dev, next, false))
232 233 234 235 236 237 238 239 240
			return;
		tick_do_periodic_broadcast();
	}
}

/*
 * Powerstate information: The system enters/leaves a state, where
 * affected devices might stop
 */
241
static void tick_do_broadcast_on_off(unsigned long *reason)
242 243 244
{
	struct clock_event_device *bc, *dev;
	struct tick_device *td;
245
	unsigned long flags;
246
	int cpu, bc_stopped;
247

248
	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
249 250 251 252 253 254 255

	cpu = smp_processor_id();
	td = &per_cpu(tick_cpu_device, cpu);
	dev = td->evtdev;
	bc = tick_broadcast_device.evtdev;

	/*
256
	 * Is the device not affected by the powerstate ?
257
	 */
258
	if (!dev || !(dev->features & CLOCK_EVT_FEAT_C3STOP))
259 260
		goto out;

261 262
	if (!tick_device_is_functional(dev))
		goto out;
263

264
	bc_stopped = cpumask_empty(tick_broadcast_mask);
265

266 267 268
	switch (*reason) {
	case CLOCK_EVT_NOTIFY_BROADCAST_ON:
	case CLOCK_EVT_NOTIFY_BROADCAST_FORCE:
269
		if (!cpumask_test_and_set_cpu(cpu, tick_broadcast_mask)) {
270 271
			if (tick_broadcast_device.mode ==
			    TICKDEV_MODE_PERIODIC)
272
				clockevents_shutdown(dev);
273
		}
274
		if (*reason == CLOCK_EVT_NOTIFY_BROADCAST_FORCE)
275
			tick_broadcast_force = 1;
276 277
		break;
	case CLOCK_EVT_NOTIFY_BROADCAST_OFF:
278
		if (!tick_broadcast_force &&
279
		    cpumask_test_and_clear_cpu(cpu, tick_broadcast_mask)) {
280 281
			if (tick_broadcast_device.mode ==
			    TICKDEV_MODE_PERIODIC)
282 283
				tick_setup_periodic(dev, 0);
		}
284
		break;
285 286
	}

287
	if (cpumask_empty(tick_broadcast_mask)) {
288
		if (!bc_stopped)
289
			clockevents_shutdown(bc);
290
	} else if (bc_stopped) {
291 292
		if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC)
			tick_broadcast_start_periodic(bc);
293 294
		else
			tick_broadcast_setup_oneshot(bc);
295 296
	}
out:
297
	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
298 299 300 301 302 303 304 305
}

/*
 * Powerstate information: The system enters/leaves a state, where
 * affected devices might stop.
 */
void tick_broadcast_on_off(unsigned long reason, int *oncpu)
{
306
	if (!cpumask_test_cpu(*oncpu, cpu_online_mask))
307
		printk(KERN_ERR "tick-broadcast: ignoring broadcast for "
308
		       "offline CPU #%d\n", *oncpu);
309
	else
310
		tick_do_broadcast_on_off(&reason);
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332
}

/*
 * Set the periodic handler depending on broadcast on/off
 */
void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
{
	if (!broadcast)
		dev->event_handler = tick_handle_periodic;
	else
		dev->event_handler = tick_handle_periodic_broadcast;
}

/*
 * Remove a CPU from broadcasting
 */
void tick_shutdown_broadcast(unsigned int *cpup)
{
	struct clock_event_device *bc;
	unsigned long flags;
	unsigned int cpu = *cpup;

333
	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
334 335

	bc = tick_broadcast_device.evtdev;
336
	cpumask_clear_cpu(cpu, tick_broadcast_mask);
337 338

	if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) {
339
		if (bc && cpumask_empty(tick_broadcast_mask))
340
			clockevents_shutdown(bc);
341 342
	}

343
	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
344
}
345

346 347 348 349 350
void tick_suspend_broadcast(void)
{
	struct clock_event_device *bc;
	unsigned long flags;

351
	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
352 353

	bc = tick_broadcast_device.evtdev;
T
Thomas Gleixner 已提交
354
	if (bc)
355
		clockevents_shutdown(bc);
356

357
	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
358 359 360 361 362 363 364 365
}

int tick_resume_broadcast(void)
{
	struct clock_event_device *bc;
	unsigned long flags;
	int broadcast = 0;

366
	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
367 368 369

	bc = tick_broadcast_device.evtdev;

370
	if (bc) {
T
Thomas Gleixner 已提交
371 372
		clockevents_set_mode(bc, CLOCK_EVT_MODE_RESUME);

373 374
		switch (tick_broadcast_device.mode) {
		case TICKDEV_MODE_PERIODIC:
375
			if (!cpumask_empty(tick_broadcast_mask))
376
				tick_broadcast_start_periodic(bc);
377
			broadcast = cpumask_test_cpu(smp_processor_id(),
378
						     tick_broadcast_mask);
379 380
			break;
		case TICKDEV_MODE_ONESHOT:
381
			if (!cpumask_empty(tick_broadcast_mask))
382
				broadcast = tick_resume_broadcast_oneshot(bc);
383 384
			break;
		}
385
	}
386
	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
387 388 389 390 391

	return broadcast;
}


392 393
#ifdef CONFIG_TICK_ONESHOT

394
static cpumask_var_t tick_broadcast_oneshot_mask;
395

396
/*
397
 * Exposed for debugging: see timer_list.c
398
 */
399
struct cpumask *tick_get_broadcast_oneshot_mask(void)
400
{
401
	return tick_broadcast_oneshot_mask;
402 403
}

404 405
static int tick_broadcast_set_event(struct clock_event_device *bc,
				    ktime_t expires, int force)
406
{
407 408 409
	if (bc->mode != CLOCK_EVT_MODE_ONESHOT)
		clockevents_set_mode(bc, CLOCK_EVT_MODE_ONESHOT);

410
	return clockevents_program_event(bc, expires, force);
411 412
}

413 414 415
int tick_resume_broadcast_oneshot(struct clock_event_device *bc)
{
	clockevents_set_mode(bc, CLOCK_EVT_MODE_ONESHOT);
416
	return 0;
417 418
}

419 420 421 422 423 424
/*
 * Called from irq_enter() when idle was interrupted to reenable the
 * per cpu device.
 */
void tick_check_oneshot_broadcast(int cpu)
{
425
	if (cpumask_test_cpu(cpu, tick_broadcast_oneshot_mask)) {
426 427 428 429 430 431
		struct tick_device *td = &per_cpu(tick_cpu_device, cpu);

		clockevents_set_mode(td->evtdev, CLOCK_EVT_MODE_ONESHOT);
	}
}

432 433 434 435 436 437
/*
 * Handle oneshot mode broadcasting
 */
static void tick_handle_oneshot_broadcast(struct clock_event_device *dev)
{
	struct tick_device *td;
438
	ktime_t now, next_event;
439 440
	int cpu;

441
	raw_spin_lock(&tick_broadcast_lock);
442 443
again:
	dev->next_event.tv64 = KTIME_MAX;
444
	next_event.tv64 = KTIME_MAX;
445
	cpumask_clear(tmpmask);
446 447
	now = ktime_get();
	/* Find all expired events */
448
	for_each_cpu(cpu, tick_broadcast_oneshot_mask) {
449 450
		td = &per_cpu(tick_cpu_device, cpu);
		if (td->evtdev->next_event.tv64 <= now.tv64)
451
			cpumask_set_cpu(cpu, tmpmask);
452 453
		else if (td->evtdev->next_event.tv64 < next_event.tv64)
			next_event.tv64 = td->evtdev->next_event.tv64;
454 455 456
	}

	/*
457 458
	 * Wakeup the cpus which have an expired event.
	 */
459
	tick_do_broadcast(tmpmask);
460 461 462 463 464 465 466 467 468 469

	/*
	 * Two reasons for reprogram:
	 *
	 * - The global event did not expire any CPU local
	 * events. This happens in dyntick mode, as the maximum PIT
	 * delta is quite small.
	 *
	 * - There are pending events on sleeping CPUs which were not
	 * in the event mask
470
	 */
471
	if (next_event.tv64 != KTIME_MAX) {
472
		/*
473 474
		 * Rearm the broadcast device. If event expired,
		 * repeat the above
475
		 */
476
		if (tick_broadcast_set_event(dev, next_event, 0))
477 478
			goto again;
	}
479
	raw_spin_unlock(&tick_broadcast_lock);
480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497
}

/*
 * Powerstate information: The system enters/leaves a state, where
 * affected devices might stop
 */
void tick_broadcast_oneshot_control(unsigned long reason)
{
	struct clock_event_device *bc, *dev;
	struct tick_device *td;
	unsigned long flags;
	int cpu;

	/*
	 * Periodic mode does not care about the enter/exit of power
	 * states
	 */
	if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC)
498
		return;
499

500 501 502 503
	/*
	 * We are called with preemtion disabled from the depth of the
	 * idle code, so we can't be moved away.
	 */
504 505 506 507 508
	cpu = smp_processor_id();
	td = &per_cpu(tick_cpu_device, cpu);
	dev = td->evtdev;

	if (!(dev->features & CLOCK_EVT_FEAT_C3STOP))
509 510 511
		return;

	bc = tick_broadcast_device.evtdev;
512

513
	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
514
	if (reason == CLOCK_EVT_NOTIFY_BROADCAST_ENTER) {
515
		if (!cpumask_test_and_set_cpu(cpu, tick_broadcast_oneshot_mask)) {
516 517
			clockevents_set_mode(dev, CLOCK_EVT_MODE_SHUTDOWN);
			if (dev->next_event.tv64 < bc->next_event.tv64)
518
				tick_broadcast_set_event(bc, dev->next_event, 1);
519 520
		}
	} else {
521
		if (cpumask_test_and_clear_cpu(cpu, tick_broadcast_oneshot_mask)) {
522 523 524 525 526
			clockevents_set_mode(dev, CLOCK_EVT_MODE_ONESHOT);
			if (dev->next_event.tv64 != KTIME_MAX)
				tick_program_event(dev->next_event, 1);
		}
	}
527
	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
528 529
}

530 531 532 533 534 535 536
/*
 * Reset the one shot broadcast for a cpu
 *
 * Called with tick_broadcast_lock held
 */
static void tick_broadcast_clear_oneshot(int cpu)
{
537
	cpumask_clear_cpu(cpu, tick_broadcast_oneshot_mask);
538 539
}

540 541
static void tick_broadcast_init_next_event(struct cpumask *mask,
					   ktime_t expires)
542 543 544 545
{
	struct tick_device *td;
	int cpu;

546
	for_each_cpu(cpu, mask) {
547 548 549 550 551 552
		td = &per_cpu(tick_cpu_device, cpu);
		if (td->evtdev)
			td->evtdev->next_event = expires;
	}
}

553
/**
554
 * tick_broadcast_setup_oneshot - setup the broadcast device
555 556 557
 */
void tick_broadcast_setup_oneshot(struct clock_event_device *bc)
{
558 559
	int cpu = smp_processor_id();

560 561
	/* Set it up only once ! */
	if (bc->event_handler != tick_handle_oneshot_broadcast) {
562 563
		int was_periodic = bc->mode == CLOCK_EVT_MODE_PERIODIC;

564
		bc->event_handler = tick_handle_oneshot_broadcast;
565 566 567 568 569 570 571 572 573 574

		/* Take the do_timer update */
		tick_do_timer_cpu = cpu;

		/*
		 * We must be careful here. There might be other CPUs
		 * waiting for periodic broadcast. We need to set the
		 * oneshot_mask bits for those and program the
		 * broadcast device to fire.
		 */
575 576 577 578
		cpumask_copy(tmpmask, tick_broadcast_mask);
		cpumask_clear_cpu(cpu, tmpmask);
		cpumask_or(tick_broadcast_oneshot_mask,
			   tick_broadcast_oneshot_mask, tmpmask);
579

580
		if (was_periodic && !cpumask_empty(tmpmask)) {
581
			clockevents_set_mode(bc, CLOCK_EVT_MODE_ONESHOT);
582
			tick_broadcast_init_next_event(tmpmask,
583
						       tick_next_period);
584
			tick_broadcast_set_event(bc, tick_next_period, 1);
585 586
		} else
			bc->next_event.tv64 = KTIME_MAX;
587 588 589 590 591 592 593 594 595
	} else {
		/*
		 * The first cpu which switches to oneshot mode sets
		 * the bit for all other cpus which are in the general
		 * (periodic) broadcast mask. So the bit is set and
		 * would prevent the first broadcast enter after this
		 * to program the bc device.
		 */
		tick_broadcast_clear_oneshot(cpu);
596
	}
597 598 599 600 601 602 603 604 605 606
}

/*
 * Select oneshot operating mode for the broadcast device
 */
void tick_broadcast_switch_to_oneshot(void)
{
	struct clock_event_device *bc;
	unsigned long flags;

607
	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
608 609

	tick_broadcast_device.mode = TICKDEV_MODE_ONESHOT;
610 611 612
	bc = tick_broadcast_device.evtdev;
	if (bc)
		tick_broadcast_setup_oneshot(bc);
613

614
	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
615 616 617 618 619 620 621 622 623 624 625
}


/*
 * Remove a dead CPU from broadcasting
 */
void tick_shutdown_broadcast_oneshot(unsigned int *cpup)
{
	unsigned long flags;
	unsigned int cpu = *cpup;

626
	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
627

628 629 630 631
	/*
	 * Clear the broadcast mask flag for the dead cpu, but do not
	 * stop the broadcast device!
	 */
632
	cpumask_clear_cpu(cpu, tick_broadcast_oneshot_mask);
633

634
	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
635 636
}

637 638 639 640 641 642 643 644
/*
 * Check, whether the broadcast device is in one shot mode
 */
int tick_broadcast_oneshot_active(void)
{
	return tick_broadcast_device.mode == TICKDEV_MODE_ONESHOT;
}

645 646 647 648 649 650 651 652 653 654
/*
 * Check whether the broadcast device supports oneshot.
 */
bool tick_broadcast_oneshot_available(void)
{
	struct clock_event_device *bc = tick_broadcast_device.evtdev;

	return bc ? bc->features & CLOCK_EVT_FEAT_ONESHOT : false;
}

655
#endif
656 657 658 659 660 661 662 663 664

void __init tick_broadcast_init(void)
{
	alloc_cpumask_var(&tick_broadcast_mask, GFP_NOWAIT);
	alloc_cpumask_var(&tmpmask, GFP_NOWAIT);
#ifdef CONFIG_TICK_ONESHOT
	alloc_cpumask_var(&tick_broadcast_oneshot_mask, GFP_NOWAIT);
#endif
}