Linux Audio

Check our new training course

Loading...
v4.17
 
   1/*
   2 * linux/kernel/time/clocksource.c
   3 *
   4 * This file contains the functions which manage clocksource drivers.
   5 *
   6 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
   7 *
   8 * This program is free software; you can redistribute it and/or modify
   9 * it under the terms of the GNU General Public License as published by
  10 * the Free Software Foundation; either version 2 of the License, or
  11 * (at your option) any later version.
  12 *
  13 * This program is distributed in the hope that it will be useful,
  14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16 * GNU General Public License for more details.
  17 *
  18 * You should have received a copy of the GNU General Public License
  19 * along with this program; if not, write to the Free Software
  20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  21 *
  22 * TODO WishList:
  23 *   o Allow clocksource drivers to be unregistered
  24 */
  25
  26#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  27
  28#include <linux/device.h>
  29#include <linux/clocksource.h>
  30#include <linux/init.h>
  31#include <linux/module.h>
  32#include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
  33#include <linux/tick.h>
  34#include <linux/kthread.h>
  35
  36#include "tick-internal.h"
  37#include "timekeeping_internal.h"
  38
  39/**
  40 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
  41 * @mult:	pointer to mult variable
  42 * @shift:	pointer to shift variable
  43 * @from:	frequency to convert from
  44 * @to:		frequency to convert to
  45 * @maxsec:	guaranteed runtime conversion range in seconds
  46 *
  47 * The function evaluates the shift/mult pair for the scaled math
  48 * operations of clocksources and clockevents.
  49 *
  50 * @to and @from are frequency values in HZ. For clock sources @to is
  51 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
  52 * event @to is the counter frequency and @from is NSEC_PER_SEC.
  53 *
  54 * The @maxsec conversion range argument controls the time frame in
  55 * seconds which must be covered by the runtime conversion with the
  56 * calculated mult and shift factors. This guarantees that no 64bit
  57 * overflow happens when the input value of the conversion is
  58 * multiplied with the calculated mult factor. Larger ranges may
  59 * reduce the conversion accuracy by chosing smaller mult and shift
  60 * factors.
  61 */
  62void
  63clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
  64{
  65	u64 tmp;
  66	u32 sft, sftacc= 32;
  67
  68	/*
  69	 * Calculate the shift factor which is limiting the conversion
  70	 * range:
  71	 */
  72	tmp = ((u64)maxsec * from) >> 32;
  73	while (tmp) {
  74		tmp >>=1;
  75		sftacc--;
  76	}
  77
  78	/*
  79	 * Find the conversion shift/mult pair which has the best
  80	 * accuracy and fits the maxsec conversion range:
  81	 */
  82	for (sft = 32; sft > 0; sft--) {
  83		tmp = (u64) to << sft;
  84		tmp += from / 2;
  85		do_div(tmp, from);
  86		if ((tmp >> sftacc) == 0)
  87			break;
  88	}
  89	*mult = tmp;
  90	*shift = sft;
  91}
  92EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
  93
  94/*[Clocksource internal variables]---------
  95 * curr_clocksource:
  96 *	currently selected clocksource.
 
 
  97 * clocksource_list:
  98 *	linked list with the registered clocksources
  99 * clocksource_mutex:
 100 *	protects manipulations to curr_clocksource and the clocksource_list
 101 * override_name:
 102 *	Name of the user-specified clocksource.
 103 */
 104static struct clocksource *curr_clocksource;
 
 105static LIST_HEAD(clocksource_list);
 106static DEFINE_MUTEX(clocksource_mutex);
 107static char override_name[CS_NAME_LEN];
 108static int finished_booting;
 
 109
 110#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
 111static void clocksource_watchdog_work(struct work_struct *work);
 112static void clocksource_select(void);
 113
 114static LIST_HEAD(watchdog_list);
 115static struct clocksource *watchdog;
 116static struct timer_list watchdog_timer;
 117static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
 118static DEFINE_SPINLOCK(watchdog_lock);
 119static int watchdog_running;
 120static atomic_t watchdog_reset_pending;
 121
 122static void inline clocksource_watchdog_lock(unsigned long *flags)
 123{
 124	spin_lock_irqsave(&watchdog_lock, *flags);
 125}
 126
 127static void inline clocksource_watchdog_unlock(unsigned long *flags)
 128{
 129	spin_unlock_irqrestore(&watchdog_lock, *flags);
 130}
 131
 132static int clocksource_watchdog_kthread(void *data);
 133static void __clocksource_change_rating(struct clocksource *cs, int rating);
 134
 135/*
 136 * Interval: 0.5sec Threshold: 0.0625s
 137 */
 138#define WATCHDOG_INTERVAL (HZ >> 1)
 139#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
 140
 141static void clocksource_watchdog_work(struct work_struct *work)
 142{
 143	/*
 
 
 
 
 
 
 
 
 
 144	 * If kthread_run fails the next watchdog scan over the
 145	 * watchdog_list will find the unstable clock again.
 146	 */
 147	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
 148}
 149
 150static void __clocksource_unstable(struct clocksource *cs)
 151{
 152	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
 153	cs->flags |= CLOCK_SOURCE_UNSTABLE;
 154
 155	/*
 156	 * If the clocksource is registered clocksource_watchdog_kthread() will
 157	 * re-rate and re-select.
 158	 */
 159	if (list_empty(&cs->list)) {
 160		cs->rating = 0;
 161		return;
 162	}
 163
 164	if (cs->mark_unstable)
 165		cs->mark_unstable(cs);
 166
 167	/* kick clocksource_watchdog_kthread() */
 168	if (finished_booting)
 169		schedule_work(&watchdog_work);
 170}
 171
 172/**
 173 * clocksource_mark_unstable - mark clocksource unstable via watchdog
 174 * @cs:		clocksource to be marked unstable
 175 *
 176 * This function is called by the x86 TSC code to mark clocksources as unstable;
 177 * it defers demotion and re-selection to a kthread.
 178 */
 179void clocksource_mark_unstable(struct clocksource *cs)
 180{
 181	unsigned long flags;
 182
 183	spin_lock_irqsave(&watchdog_lock, flags);
 184	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
 185		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
 186			list_add(&cs->wd_list, &watchdog_list);
 187		__clocksource_unstable(cs);
 188	}
 189	spin_unlock_irqrestore(&watchdog_lock, flags);
 190}
 191
 192static void clocksource_watchdog(struct timer_list *unused)
 193{
 194	struct clocksource *cs;
 195	u64 csnow, wdnow, cslast, wdlast, delta;
 196	int64_t wd_nsec, cs_nsec;
 197	int next_cpu, reset_pending;
 198
 199	spin_lock(&watchdog_lock);
 200	if (!watchdog_running)
 201		goto out;
 202
 203	reset_pending = atomic_read(&watchdog_reset_pending);
 204
 205	list_for_each_entry(cs, &watchdog_list, wd_list) {
 206
 207		/* Clocksource already marked unstable? */
 208		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 209			if (finished_booting)
 210				schedule_work(&watchdog_work);
 211			continue;
 212		}
 213
 214		local_irq_disable();
 215		csnow = cs->read(cs);
 216		wdnow = watchdog->read(watchdog);
 217		local_irq_enable();
 218
 219		/* Clocksource initialized ? */
 220		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
 221		    atomic_read(&watchdog_reset_pending)) {
 222			cs->flags |= CLOCK_SOURCE_WATCHDOG;
 223			cs->wd_last = wdnow;
 224			cs->cs_last = csnow;
 225			continue;
 226		}
 227
 228		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
 229		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
 230					     watchdog->shift);
 231
 232		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
 233		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
 234		wdlast = cs->wd_last; /* save these in case we print them */
 235		cslast = cs->cs_last;
 236		cs->cs_last = csnow;
 237		cs->wd_last = wdnow;
 238
 239		if (atomic_read(&watchdog_reset_pending))
 240			continue;
 241
 242		/* Check the deviation from the watchdog clocksource. */
 243		if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
 244			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
 245				smp_processor_id(), cs->name);
 246			pr_warn("                      '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
 247				watchdog->name, wdnow, wdlast, watchdog->mask);
 248			pr_warn("                      '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
 249				cs->name, csnow, cslast, cs->mask);
 250			__clocksource_unstable(cs);
 251			continue;
 252		}
 253
 254		if (cs == curr_clocksource && cs->tick_stable)
 255			cs->tick_stable(cs);
 256
 257		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
 258		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
 259		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
 260			/* Mark it valid for high-res. */
 261			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 262
 263			/*
 264			 * clocksource_done_booting() will sort it if
 265			 * finished_booting is not set yet.
 266			 */
 267			if (!finished_booting)
 268				continue;
 269
 270			/*
 271			 * If this is not the current clocksource let
 272			 * the watchdog thread reselect it. Due to the
 273			 * change to high res this clocksource might
 274			 * be preferred now. If it is the current
 275			 * clocksource let the tick code know about
 276			 * that change.
 277			 */
 278			if (cs != curr_clocksource) {
 279				cs->flags |= CLOCK_SOURCE_RESELECT;
 280				schedule_work(&watchdog_work);
 281			} else {
 282				tick_clock_notify();
 283			}
 284		}
 285	}
 286
 287	/*
 288	 * We only clear the watchdog_reset_pending, when we did a
 289	 * full cycle through all clocksources.
 290	 */
 291	if (reset_pending)
 292		atomic_dec(&watchdog_reset_pending);
 293
 294	/*
 295	 * Cycle through CPUs to check if the CPUs stay synchronized
 296	 * to each other.
 297	 */
 298	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
 299	if (next_cpu >= nr_cpu_ids)
 300		next_cpu = cpumask_first(cpu_online_mask);
 301	watchdog_timer.expires += WATCHDOG_INTERVAL;
 302	add_timer_on(&watchdog_timer, next_cpu);
 
 
 
 
 
 
 
 303out:
 304	spin_unlock(&watchdog_lock);
 305}
 306
 307static inline void clocksource_start_watchdog(void)
 308{
 309	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
 310		return;
 311	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
 312	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
 313	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
 314	watchdog_running = 1;
 315}
 316
 317static inline void clocksource_stop_watchdog(void)
 318{
 319	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
 320		return;
 321	del_timer(&watchdog_timer);
 322	watchdog_running = 0;
 323}
 324
 325static inline void clocksource_reset_watchdog(void)
 326{
 327	struct clocksource *cs;
 328
 329	list_for_each_entry(cs, &watchdog_list, wd_list)
 330		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
 331}
 332
 333static void clocksource_resume_watchdog(void)
 334{
 335	atomic_inc(&watchdog_reset_pending);
 336}
 337
 338static void clocksource_enqueue_watchdog(struct clocksource *cs)
 339{
 340	INIT_LIST_HEAD(&cs->wd_list);
 341
 342	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 343		/* cs is a clocksource to be watched. */
 344		list_add(&cs->wd_list, &watchdog_list);
 345		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
 346	} else {
 347		/* cs is a watchdog. */
 348		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 349			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 350	}
 351}
 352
 353static void clocksource_select_watchdog(bool fallback)
 354{
 355	struct clocksource *cs, *old_wd;
 356	unsigned long flags;
 357
 358	spin_lock_irqsave(&watchdog_lock, flags);
 359	/* save current watchdog */
 360	old_wd = watchdog;
 361	if (fallback)
 362		watchdog = NULL;
 363
 364	list_for_each_entry(cs, &clocksource_list, list) {
 365		/* cs is a clocksource to be watched. */
 366		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
 367			continue;
 368
 369		/* Skip current if we were requested for a fallback. */
 370		if (fallback && cs == old_wd)
 371			continue;
 372
 373		/* Pick the best watchdog. */
 374		if (!watchdog || cs->rating > watchdog->rating)
 375			watchdog = cs;
 376	}
 377	/* If we failed to find a fallback restore the old one. */
 378	if (!watchdog)
 379		watchdog = old_wd;
 380
 381	/* If we changed the watchdog we need to reset cycles. */
 382	if (watchdog != old_wd)
 383		clocksource_reset_watchdog();
 384
 385	/* Check if the watchdog timer needs to be started. */
 386	clocksource_start_watchdog();
 387	spin_unlock_irqrestore(&watchdog_lock, flags);
 388}
 389
 390static void clocksource_dequeue_watchdog(struct clocksource *cs)
 391{
 392	if (cs != watchdog) {
 393		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 394			/* cs is a watched clocksource. */
 395			list_del_init(&cs->wd_list);
 396			/* Check if the watchdog timer needs to be stopped. */
 397			clocksource_stop_watchdog();
 398		}
 399	}
 400}
 401
 402static int __clocksource_watchdog_kthread(void)
 403{
 404	struct clocksource *cs, *tmp;
 405	unsigned long flags;
 406	int select = 0;
 407
 408	spin_lock_irqsave(&watchdog_lock, flags);
 409	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
 410		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 411			list_del_init(&cs->wd_list);
 412			__clocksource_change_rating(cs, 0);
 413			select = 1;
 414		}
 415		if (cs->flags & CLOCK_SOURCE_RESELECT) {
 416			cs->flags &= ~CLOCK_SOURCE_RESELECT;
 417			select = 1;
 418		}
 419	}
 420	/* Check if the watchdog timer needs to be stopped. */
 421	clocksource_stop_watchdog();
 422	spin_unlock_irqrestore(&watchdog_lock, flags);
 423
 424	return select;
 425}
 426
 427static int clocksource_watchdog_kthread(void *data)
 428{
 429	mutex_lock(&clocksource_mutex);
 430	if (__clocksource_watchdog_kthread())
 431		clocksource_select();
 432	mutex_unlock(&clocksource_mutex);
 433	return 0;
 434}
 435
 436static bool clocksource_is_watchdog(struct clocksource *cs)
 437{
 438	return cs == watchdog;
 439}
 440
 441#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
 442
 443static void clocksource_enqueue_watchdog(struct clocksource *cs)
 444{
 445	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 446		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 447}
 448
 449static void clocksource_select_watchdog(bool fallback) { }
 450static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
 451static inline void clocksource_resume_watchdog(void) { }
 452static inline int __clocksource_watchdog_kthread(void) { return 0; }
 453static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
 454void clocksource_mark_unstable(struct clocksource *cs) { }
 455
 456static void inline clocksource_watchdog_lock(unsigned long *flags) { }
 457static void inline clocksource_watchdog_unlock(unsigned long *flags) { }
 458
 459#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
 460
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 461/**
 462 * clocksource_suspend - suspend the clocksource(s)
 463 */
 464void clocksource_suspend(void)
 465{
 466	struct clocksource *cs;
 467
 468	list_for_each_entry_reverse(cs, &clocksource_list, list)
 469		if (cs->suspend)
 470			cs->suspend(cs);
 471}
 472
 473/**
 474 * clocksource_resume - resume the clocksource(s)
 475 */
 476void clocksource_resume(void)
 477{
 478	struct clocksource *cs;
 479
 480	list_for_each_entry(cs, &clocksource_list, list)
 481		if (cs->resume)
 482			cs->resume(cs);
 483
 484	clocksource_resume_watchdog();
 485}
 486
 487/**
 488 * clocksource_touch_watchdog - Update watchdog
 489 *
 490 * Update the watchdog after exception contexts such as kgdb so as not
 491 * to incorrectly trip the watchdog. This might fail when the kernel
 492 * was stopped in code which holds watchdog_lock.
 493 */
 494void clocksource_touch_watchdog(void)
 495{
 496	clocksource_resume_watchdog();
 497}
 498
 499/**
 500 * clocksource_max_adjustment- Returns max adjustment amount
 501 * @cs:         Pointer to clocksource
 502 *
 503 */
 504static u32 clocksource_max_adjustment(struct clocksource *cs)
 505{
 506	u64 ret;
 507	/*
 508	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
 509	 */
 510	ret = (u64)cs->mult * 11;
 511	do_div(ret,100);
 512	return (u32)ret;
 513}
 514
 515/**
 516 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
 517 * @mult:	cycle to nanosecond multiplier
 518 * @shift:	cycle to nanosecond divisor (power of two)
 519 * @maxadj:	maximum adjustment value to mult (~11%)
 520 * @mask:	bitmask for two's complement subtraction of non 64 bit counters
 521 * @max_cyc:	maximum cycle value before potential overflow (does not include
 522 *		any safety margin)
 523 *
 524 * NOTE: This function includes a safety margin of 50%, in other words, we
 525 * return half the number of nanoseconds the hardware counter can technically
 526 * cover. This is done so that we can potentially detect problems caused by
 527 * delayed timers or bad hardware, which might result in time intervals that
 528 * are larger than what the math used can handle without overflows.
 529 */
 530u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
 531{
 532	u64 max_nsecs, max_cycles;
 533
 534	/*
 535	 * Calculate the maximum number of cycles that we can pass to the
 536	 * cyc2ns() function without overflowing a 64-bit result.
 537	 */
 538	max_cycles = ULLONG_MAX;
 539	do_div(max_cycles, mult+maxadj);
 540
 541	/*
 542	 * The actual maximum number of cycles we can defer the clocksource is
 543	 * determined by the minimum of max_cycles and mask.
 544	 * Note: Here we subtract the maxadj to make sure we don't sleep for
 545	 * too long if there's a large negative adjustment.
 546	 */
 547	max_cycles = min(max_cycles, mask);
 548	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
 549
 550	/* return the max_cycles value as well if requested */
 551	if (max_cyc)
 552		*max_cyc = max_cycles;
 553
 554	/* Return 50% of the actual maximum, so we can detect bad values */
 555	max_nsecs >>= 1;
 556
 557	return max_nsecs;
 558}
 559
 560/**
 561 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
 562 * @cs:         Pointer to clocksource to be updated
 563 *
 564 */
 565static inline void clocksource_update_max_deferment(struct clocksource *cs)
 566{
 567	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
 568						cs->maxadj, cs->mask,
 569						&cs->max_cycles);
 570}
 571
 572#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
 573
 574static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
 575{
 576	struct clocksource *cs;
 577
 578	if (!finished_booting || list_empty(&clocksource_list))
 579		return NULL;
 580
 581	/*
 582	 * We pick the clocksource with the highest rating. If oneshot
 583	 * mode is active, we pick the highres valid clocksource with
 584	 * the best rating.
 585	 */
 586	list_for_each_entry(cs, &clocksource_list, list) {
 587		if (skipcur && cs == curr_clocksource)
 588			continue;
 589		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
 590			continue;
 591		return cs;
 592	}
 593	return NULL;
 594}
 595
 596static void __clocksource_select(bool skipcur)
 597{
 598	bool oneshot = tick_oneshot_mode_active();
 599	struct clocksource *best, *cs;
 600
 601	/* Find the best suitable clocksource */
 602	best = clocksource_find_best(oneshot, skipcur);
 603	if (!best)
 604		return;
 605
 606	if (!strlen(override_name))
 607		goto found;
 608
 609	/* Check for the override clocksource. */
 610	list_for_each_entry(cs, &clocksource_list, list) {
 611		if (skipcur && cs == curr_clocksource)
 612			continue;
 613		if (strcmp(cs->name, override_name) != 0)
 614			continue;
 615		/*
 616		 * Check to make sure we don't switch to a non-highres
 617		 * capable clocksource if the tick code is in oneshot
 618		 * mode (highres or nohz)
 619		 */
 620		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
 621			/* Override clocksource cannot be used. */
 622			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 623				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
 624					cs->name);
 625				override_name[0] = 0;
 626			} else {
 627				/*
 628				 * The override cannot be currently verified.
 629				 * Deferring to let the watchdog check.
 630				 */
 631				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
 632					cs->name);
 633			}
 634		} else
 635			/* Override clocksource can be used. */
 636			best = cs;
 637		break;
 638	}
 639
 640found:
 641	if (curr_clocksource != best && !timekeeping_notify(best)) {
 642		pr_info("Switched to clocksource %s\n", best->name);
 643		curr_clocksource = best;
 644	}
 645}
 646
 647/**
 648 * clocksource_select - Select the best clocksource available
 649 *
 650 * Private function. Must hold clocksource_mutex when called.
 651 *
 652 * Select the clocksource with the best rating, or the clocksource,
 653 * which is selected by userspace override.
 654 */
 655static void clocksource_select(void)
 656{
 657	__clocksource_select(false);
 658}
 659
 660static void clocksource_select_fallback(void)
 661{
 662	__clocksource_select(true);
 663}
 664
 665#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
 666static inline void clocksource_select(void) { }
 667static inline void clocksource_select_fallback(void) { }
 668
 669#endif
 670
 671/*
 672 * clocksource_done_booting - Called near the end of core bootup
 673 *
 674 * Hack to avoid lots of clocksource churn at boot time.
 675 * We use fs_initcall because we want this to start before
 676 * device_initcall but after subsys_initcall.
 677 */
 678static int __init clocksource_done_booting(void)
 679{
 680	mutex_lock(&clocksource_mutex);
 681	curr_clocksource = clocksource_default_clock();
 682	finished_booting = 1;
 683	/*
 684	 * Run the watchdog first to eliminate unstable clock sources
 685	 */
 686	__clocksource_watchdog_kthread();
 687	clocksource_select();
 688	mutex_unlock(&clocksource_mutex);
 689	return 0;
 690}
 691fs_initcall(clocksource_done_booting);
 692
 693/*
 694 * Enqueue the clocksource sorted by rating
 695 */
 696static void clocksource_enqueue(struct clocksource *cs)
 697{
 698	struct list_head *entry = &clocksource_list;
 699	struct clocksource *tmp;
 700
 701	list_for_each_entry(tmp, &clocksource_list, list) {
 702		/* Keep track of the place, where to insert */
 703		if (tmp->rating < cs->rating)
 704			break;
 705		entry = &tmp->list;
 706	}
 707	list_add(&cs->list, entry);
 708}
 709
 710/**
 711 * __clocksource_update_freq_scale - Used update clocksource with new freq
 712 * @cs:		clocksource to be registered
 713 * @scale:	Scale factor multiplied against freq to get clocksource hz
 714 * @freq:	clocksource frequency (cycles per second) divided by scale
 715 *
 716 * This should only be called from the clocksource->enable() method.
 717 *
 718 * This *SHOULD NOT* be called directly! Please use the
 719 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
 720 * functions.
 721 */
 722void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
 723{
 724	u64 sec;
 725
 726	/*
 727	 * Default clocksources are *special* and self-define their mult/shift.
 728	 * But, you're not special, so you should specify a freq value.
 729	 */
 730	if (freq) {
 731		/*
 732		 * Calc the maximum number of seconds which we can run before
 733		 * wrapping around. For clocksources which have a mask > 32-bit
 734		 * we need to limit the max sleep time to have a good
 735		 * conversion precision. 10 minutes is still a reasonable
 736		 * amount. That results in a shift value of 24 for a
 737		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
 738		 * ~ 0.06ppm granularity for NTP.
 739		 */
 740		sec = cs->mask;
 741		do_div(sec, freq);
 742		do_div(sec, scale);
 743		if (!sec)
 744			sec = 1;
 745		else if (sec > 600 && cs->mask > UINT_MAX)
 746			sec = 600;
 747
 748		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
 749				       NSEC_PER_SEC / scale, sec * scale);
 750	}
 751	/*
 752	 * Ensure clocksources that have large 'mult' values don't overflow
 753	 * when adjusted.
 754	 */
 755	cs->maxadj = clocksource_max_adjustment(cs);
 756	while (freq && ((cs->mult + cs->maxadj < cs->mult)
 757		|| (cs->mult - cs->maxadj > cs->mult))) {
 758		cs->mult >>= 1;
 759		cs->shift--;
 760		cs->maxadj = clocksource_max_adjustment(cs);
 761	}
 762
 763	/*
 764	 * Only warn for *special* clocksources that self-define
 765	 * their mult/shift values and don't specify a freq.
 766	 */
 767	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
 768		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
 769		cs->name);
 770
 771	clocksource_update_max_deferment(cs);
 772
 773	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
 774		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
 775}
 776EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
 777
 778/**
 779 * __clocksource_register_scale - Used to install new clocksources
 780 * @cs:		clocksource to be registered
 781 * @scale:	Scale factor multiplied against freq to get clocksource hz
 782 * @freq:	clocksource frequency (cycles per second) divided by scale
 783 *
 784 * Returns -EBUSY if registration fails, zero otherwise.
 785 *
 786 * This *SHOULD NOT* be called directly! Please use the
 787 * clocksource_register_hz() or clocksource_register_khz helper functions.
 788 */
 789int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
 790{
 791	unsigned long flags;
 792
 
 
 
 
 
 
 
 
 
 793	/* Initialize mult/shift and max_idle_ns */
 794	__clocksource_update_freq_scale(cs, scale, freq);
 795
 796	/* Add clocksource to the clocksource list */
 797	mutex_lock(&clocksource_mutex);
 798
 799	clocksource_watchdog_lock(&flags);
 800	clocksource_enqueue(cs);
 801	clocksource_enqueue_watchdog(cs);
 802	clocksource_watchdog_unlock(&flags);
 803
 804	clocksource_select();
 805	clocksource_select_watchdog(false);
 
 806	mutex_unlock(&clocksource_mutex);
 807	return 0;
 808}
 809EXPORT_SYMBOL_GPL(__clocksource_register_scale);
 810
 811static void __clocksource_change_rating(struct clocksource *cs, int rating)
 812{
 813	list_del(&cs->list);
 814	cs->rating = rating;
 815	clocksource_enqueue(cs);
 816}
 817
 818/**
 819 * clocksource_change_rating - Change the rating of a registered clocksource
 820 * @cs:		clocksource to be changed
 821 * @rating:	new rating
 822 */
 823void clocksource_change_rating(struct clocksource *cs, int rating)
 824{
 825	unsigned long flags;
 826
 827	mutex_lock(&clocksource_mutex);
 828	clocksource_watchdog_lock(&flags);
 829	__clocksource_change_rating(cs, rating);
 830	clocksource_watchdog_unlock(&flags);
 831
 832	clocksource_select();
 833	clocksource_select_watchdog(false);
 
 834	mutex_unlock(&clocksource_mutex);
 835}
 836EXPORT_SYMBOL(clocksource_change_rating);
 837
 838/*
 839 * Unbind clocksource @cs. Called with clocksource_mutex held
 840 */
 841static int clocksource_unbind(struct clocksource *cs)
 842{
 843	unsigned long flags;
 844
 845	if (clocksource_is_watchdog(cs)) {
 846		/* Select and try to install a replacement watchdog. */
 847		clocksource_select_watchdog(true);
 848		if (clocksource_is_watchdog(cs))
 849			return -EBUSY;
 850	}
 851
 852	if (cs == curr_clocksource) {
 853		/* Select and try to install a replacement clock source */
 854		clocksource_select_fallback();
 855		if (curr_clocksource == cs)
 856			return -EBUSY;
 
 
 
 
 
 
 
 
 
 857	}
 858
 859	clocksource_watchdog_lock(&flags);
 860	clocksource_dequeue_watchdog(cs);
 861	list_del_init(&cs->list);
 862	clocksource_watchdog_unlock(&flags);
 863
 864	return 0;
 865}
 866
 867/**
 868 * clocksource_unregister - remove a registered clocksource
 869 * @cs:	clocksource to be unregistered
 870 */
 871int clocksource_unregister(struct clocksource *cs)
 872{
 873	int ret = 0;
 874
 875	mutex_lock(&clocksource_mutex);
 876	if (!list_empty(&cs->list))
 877		ret = clocksource_unbind(cs);
 878	mutex_unlock(&clocksource_mutex);
 879	return ret;
 880}
 881EXPORT_SYMBOL(clocksource_unregister);
 882
 883#ifdef CONFIG_SYSFS
 884/**
 885 * current_clocksource_show - sysfs interface for current clocksource
 886 * @dev:	unused
 887 * @attr:	unused
 888 * @buf:	char buffer to be filled with clocksource list
 889 *
 890 * Provides sysfs interface for listing current clocksource.
 891 */
 892static ssize_t current_clocksource_show(struct device *dev,
 893					struct device_attribute *attr,
 894					char *buf)
 895{
 896	ssize_t count = 0;
 897
 898	mutex_lock(&clocksource_mutex);
 899	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
 900	mutex_unlock(&clocksource_mutex);
 901
 902	return count;
 903}
 904
 905ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
 906{
 907	size_t ret = cnt;
 908
 909	/* strings from sysfs write are not 0 terminated! */
 910	if (!cnt || cnt >= CS_NAME_LEN)
 911		return -EINVAL;
 912
 913	/* strip of \n: */
 914	if (buf[cnt-1] == '\n')
 915		cnt--;
 916	if (cnt > 0)
 917		memcpy(dst, buf, cnt);
 918	dst[cnt] = 0;
 919	return ret;
 920}
 921
 922/**
 923 * current_clocksource_store - interface for manually overriding clocksource
 924 * @dev:	unused
 925 * @attr:	unused
 926 * @buf:	name of override clocksource
 927 * @count:	length of buffer
 928 *
 929 * Takes input from sysfs interface for manually overriding the default
 930 * clocksource selection.
 931 */
 932static ssize_t current_clocksource_store(struct device *dev,
 933					 struct device_attribute *attr,
 934					 const char *buf, size_t count)
 935{
 936	ssize_t ret;
 937
 938	mutex_lock(&clocksource_mutex);
 939
 940	ret = sysfs_get_uname(buf, override_name, count);
 941	if (ret >= 0)
 942		clocksource_select();
 943
 944	mutex_unlock(&clocksource_mutex);
 945
 946	return ret;
 947}
 948static DEVICE_ATTR_RW(current_clocksource);
 949
 950/**
 951 * unbind_clocksource_store - interface for manually unbinding clocksource
 952 * @dev:	unused
 953 * @attr:	unused
 954 * @buf:	unused
 955 * @count:	length of buffer
 956 *
 957 * Takes input from sysfs interface for manually unbinding a clocksource.
 958 */
 959static ssize_t unbind_clocksource_store(struct device *dev,
 960					struct device_attribute *attr,
 961					const char *buf, size_t count)
 962{
 963	struct clocksource *cs;
 964	char name[CS_NAME_LEN];
 965	ssize_t ret;
 966
 967	ret = sysfs_get_uname(buf, name, count);
 968	if (ret < 0)
 969		return ret;
 970
 971	ret = -ENODEV;
 972	mutex_lock(&clocksource_mutex);
 973	list_for_each_entry(cs, &clocksource_list, list) {
 974		if (strcmp(cs->name, name))
 975			continue;
 976		ret = clocksource_unbind(cs);
 977		break;
 978	}
 979	mutex_unlock(&clocksource_mutex);
 980
 981	return ret ? ret : count;
 982}
 983static DEVICE_ATTR_WO(unbind_clocksource);
 984
 985/**
 986 * available_clocksource_show - sysfs interface for listing clocksource
 987 * @dev:	unused
 988 * @attr:	unused
 989 * @buf:	char buffer to be filled with clocksource list
 990 *
 991 * Provides sysfs interface for listing registered clocksources
 992 */
 993static ssize_t available_clocksource_show(struct device *dev,
 994					  struct device_attribute *attr,
 995					  char *buf)
 996{
 997	struct clocksource *src;
 998	ssize_t count = 0;
 999
1000	mutex_lock(&clocksource_mutex);
1001	list_for_each_entry(src, &clocksource_list, list) {
1002		/*
1003		 * Don't show non-HRES clocksource if the tick code is
1004		 * in one shot mode (highres=on or nohz=on)
1005		 */
1006		if (!tick_oneshot_mode_active() ||
1007		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1008			count += snprintf(buf + count,
1009				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1010				  "%s ", src->name);
1011	}
1012	mutex_unlock(&clocksource_mutex);
1013
1014	count += snprintf(buf + count,
1015			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1016
1017	return count;
1018}
1019static DEVICE_ATTR_RO(available_clocksource);
1020
1021static struct attribute *clocksource_attrs[] = {
1022	&dev_attr_current_clocksource.attr,
1023	&dev_attr_unbind_clocksource.attr,
1024	&dev_attr_available_clocksource.attr,
1025	NULL
1026};
1027ATTRIBUTE_GROUPS(clocksource);
1028
1029static struct bus_type clocksource_subsys = {
1030	.name = "clocksource",
1031	.dev_name = "clocksource",
1032};
1033
1034static struct device device_clocksource = {
1035	.id	= 0,
1036	.bus	= &clocksource_subsys,
1037	.groups	= clocksource_groups,
1038};
1039
1040static int __init init_clocksource_sysfs(void)
1041{
1042	int error = subsys_system_register(&clocksource_subsys, NULL);
1043
1044	if (!error)
1045		error = device_register(&device_clocksource);
1046
1047	return error;
1048}
1049
1050device_initcall(init_clocksource_sysfs);
1051#endif /* CONFIG_SYSFS */
1052
1053/**
1054 * boot_override_clocksource - boot clock override
1055 * @str:	override name
1056 *
1057 * Takes a clocksource= boot argument and uses it
1058 * as the clocksource override name.
1059 */
1060static int __init boot_override_clocksource(char* str)
1061{
1062	mutex_lock(&clocksource_mutex);
1063	if (str)
1064		strlcpy(override_name, str, sizeof(override_name));
1065	mutex_unlock(&clocksource_mutex);
1066	return 1;
1067}
1068
1069__setup("clocksource=", boot_override_clocksource);
1070
1071/**
1072 * boot_override_clock - Compatibility layer for deprecated boot option
1073 * @str:	override name
1074 *
1075 * DEPRECATED! Takes a clock= boot argument and uses it
1076 * as the clocksource override name
1077 */
1078static int __init boot_override_clock(char* str)
1079{
1080	if (!strcmp(str, "pmtmr")) {
1081		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1082		return boot_override_clocksource("acpi_pm");
1083	}
1084	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1085	return boot_override_clocksource(str);
1086}
1087
1088__setup("clock=", boot_override_clock);
v5.9
   1// SPDX-License-Identifier: GPL-2.0+
   2/*
 
 
   3 * This file contains the functions which manage clocksource drivers.
   4 *
   5 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   6 */
   7
   8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10#include <linux/device.h>
  11#include <linux/clocksource.h>
  12#include <linux/init.h>
  13#include <linux/module.h>
  14#include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
  15#include <linux/tick.h>
  16#include <linux/kthread.h>
  17
  18#include "tick-internal.h"
  19#include "timekeeping_internal.h"
  20
  21/**
  22 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
  23 * @mult:	pointer to mult variable
  24 * @shift:	pointer to shift variable
  25 * @from:	frequency to convert from
  26 * @to:		frequency to convert to
  27 * @maxsec:	guaranteed runtime conversion range in seconds
  28 *
  29 * The function evaluates the shift/mult pair for the scaled math
  30 * operations of clocksources and clockevents.
  31 *
  32 * @to and @from are frequency values in HZ. For clock sources @to is
  33 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
  34 * event @to is the counter frequency and @from is NSEC_PER_SEC.
  35 *
  36 * The @maxsec conversion range argument controls the time frame in
  37 * seconds which must be covered by the runtime conversion with the
  38 * calculated mult and shift factors. This guarantees that no 64bit
  39 * overflow happens when the input value of the conversion is
  40 * multiplied with the calculated mult factor. Larger ranges may
  41 * reduce the conversion accuracy by chosing smaller mult and shift
  42 * factors.
  43 */
  44void
  45clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
  46{
  47	u64 tmp;
  48	u32 sft, sftacc= 32;
  49
  50	/*
  51	 * Calculate the shift factor which is limiting the conversion
  52	 * range:
  53	 */
  54	tmp = ((u64)maxsec * from) >> 32;
  55	while (tmp) {
  56		tmp >>=1;
  57		sftacc--;
  58	}
  59
  60	/*
  61	 * Find the conversion shift/mult pair which has the best
  62	 * accuracy and fits the maxsec conversion range:
  63	 */
  64	for (sft = 32; sft > 0; sft--) {
  65		tmp = (u64) to << sft;
  66		tmp += from / 2;
  67		do_div(tmp, from);
  68		if ((tmp >> sftacc) == 0)
  69			break;
  70	}
  71	*mult = tmp;
  72	*shift = sft;
  73}
  74EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
  75
  76/*[Clocksource internal variables]---------
  77 * curr_clocksource:
  78 *	currently selected clocksource.
  79 * suspend_clocksource:
  80 *	used to calculate the suspend time.
  81 * clocksource_list:
  82 *	linked list with the registered clocksources
  83 * clocksource_mutex:
  84 *	protects manipulations to curr_clocksource and the clocksource_list
  85 * override_name:
  86 *	Name of the user-specified clocksource.
  87 */
  88static struct clocksource *curr_clocksource;
  89static struct clocksource *suspend_clocksource;
  90static LIST_HEAD(clocksource_list);
  91static DEFINE_MUTEX(clocksource_mutex);
  92static char override_name[CS_NAME_LEN];
  93static int finished_booting;
  94static u64 suspend_start;
  95
  96#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
  97static void clocksource_watchdog_work(struct work_struct *work);
  98static void clocksource_select(void);
  99
 100static LIST_HEAD(watchdog_list);
 101static struct clocksource *watchdog;
 102static struct timer_list watchdog_timer;
 103static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
 104static DEFINE_SPINLOCK(watchdog_lock);
 105static int watchdog_running;
 106static atomic_t watchdog_reset_pending;
 107
 108static inline void clocksource_watchdog_lock(unsigned long *flags)
 109{
 110	spin_lock_irqsave(&watchdog_lock, *flags);
 111}
 112
 113static inline void clocksource_watchdog_unlock(unsigned long *flags)
 114{
 115	spin_unlock_irqrestore(&watchdog_lock, *flags);
 116}
 117
 118static int clocksource_watchdog_kthread(void *data);
 119static void __clocksource_change_rating(struct clocksource *cs, int rating);
 120
 121/*
 122 * Interval: 0.5sec Threshold: 0.0625s
 123 */
 124#define WATCHDOG_INTERVAL (HZ >> 1)
 125#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
 126
 127static void clocksource_watchdog_work(struct work_struct *work)
 128{
 129	/*
 130	 * We cannot directly run clocksource_watchdog_kthread() here, because
 131	 * clocksource_select() calls timekeeping_notify() which uses
 132	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
 133	 * lock inversions wrt CPU hotplug.
 134	 *
 135	 * Also, we only ever run this work once or twice during the lifetime
 136	 * of the kernel, so there is no point in creating a more permanent
 137	 * kthread for this.
 138	 *
 139	 * If kthread_run fails the next watchdog scan over the
 140	 * watchdog_list will find the unstable clock again.
 141	 */
 142	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
 143}
 144
 145static void __clocksource_unstable(struct clocksource *cs)
 146{
 147	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
 148	cs->flags |= CLOCK_SOURCE_UNSTABLE;
 149
 150	/*
 151	 * If the clocksource is registered clocksource_watchdog_kthread() will
 152	 * re-rate and re-select.
 153	 */
 154	if (list_empty(&cs->list)) {
 155		cs->rating = 0;
 156		return;
 157	}
 158
 159	if (cs->mark_unstable)
 160		cs->mark_unstable(cs);
 161
 162	/* kick clocksource_watchdog_kthread() */
 163	if (finished_booting)
 164		schedule_work(&watchdog_work);
 165}
 166
 167/**
 168 * clocksource_mark_unstable - mark clocksource unstable via watchdog
 169 * @cs:		clocksource to be marked unstable
 170 *
 171 * This function is called by the x86 TSC code to mark clocksources as unstable;
 172 * it defers demotion and re-selection to a kthread.
 173 */
 174void clocksource_mark_unstable(struct clocksource *cs)
 175{
 176	unsigned long flags;
 177
 178	spin_lock_irqsave(&watchdog_lock, flags);
 179	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
 180		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
 181			list_add(&cs->wd_list, &watchdog_list);
 182		__clocksource_unstable(cs);
 183	}
 184	spin_unlock_irqrestore(&watchdog_lock, flags);
 185}
 186
 187static void clocksource_watchdog(struct timer_list *unused)
 188{
 189	struct clocksource *cs;
 190	u64 csnow, wdnow, cslast, wdlast, delta;
 191	int64_t wd_nsec, cs_nsec;
 192	int next_cpu, reset_pending;
 193
 194	spin_lock(&watchdog_lock);
 195	if (!watchdog_running)
 196		goto out;
 197
 198	reset_pending = atomic_read(&watchdog_reset_pending);
 199
 200	list_for_each_entry(cs, &watchdog_list, wd_list) {
 201
 202		/* Clocksource already marked unstable? */
 203		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 204			if (finished_booting)
 205				schedule_work(&watchdog_work);
 206			continue;
 207		}
 208
 209		local_irq_disable();
 210		csnow = cs->read(cs);
 211		wdnow = watchdog->read(watchdog);
 212		local_irq_enable();
 213
 214		/* Clocksource initialized ? */
 215		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
 216		    atomic_read(&watchdog_reset_pending)) {
 217			cs->flags |= CLOCK_SOURCE_WATCHDOG;
 218			cs->wd_last = wdnow;
 219			cs->cs_last = csnow;
 220			continue;
 221		}
 222
 223		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
 224		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
 225					     watchdog->shift);
 226
 227		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
 228		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
 229		wdlast = cs->wd_last; /* save these in case we print them */
 230		cslast = cs->cs_last;
 231		cs->cs_last = csnow;
 232		cs->wd_last = wdnow;
 233
 234		if (atomic_read(&watchdog_reset_pending))
 235			continue;
 236
 237		/* Check the deviation from the watchdog clocksource. */
 238		if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
 239			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
 240				smp_processor_id(), cs->name);
 241			pr_warn("                      '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
 242				watchdog->name, wdnow, wdlast, watchdog->mask);
 243			pr_warn("                      '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
 244				cs->name, csnow, cslast, cs->mask);
 245			__clocksource_unstable(cs);
 246			continue;
 247		}
 248
 249		if (cs == curr_clocksource && cs->tick_stable)
 250			cs->tick_stable(cs);
 251
 252		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
 253		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
 254		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
 255			/* Mark it valid for high-res. */
 256			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 257
 258			/*
 259			 * clocksource_done_booting() will sort it if
 260			 * finished_booting is not set yet.
 261			 */
 262			if (!finished_booting)
 263				continue;
 264
 265			/*
 266			 * If this is not the current clocksource let
 267			 * the watchdog thread reselect it. Due to the
 268			 * change to high res this clocksource might
 269			 * be preferred now. If it is the current
 270			 * clocksource let the tick code know about
 271			 * that change.
 272			 */
 273			if (cs != curr_clocksource) {
 274				cs->flags |= CLOCK_SOURCE_RESELECT;
 275				schedule_work(&watchdog_work);
 276			} else {
 277				tick_clock_notify();
 278			}
 279		}
 280	}
 281
 282	/*
 283	 * We only clear the watchdog_reset_pending, when we did a
 284	 * full cycle through all clocksources.
 285	 */
 286	if (reset_pending)
 287		atomic_dec(&watchdog_reset_pending);
 288
 289	/*
 290	 * Cycle through CPUs to check if the CPUs stay synchronized
 291	 * to each other.
 292	 */
 293	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
 294	if (next_cpu >= nr_cpu_ids)
 295		next_cpu = cpumask_first(cpu_online_mask);
 296
 297	/*
 298	 * Arm timer if not already pending: could race with concurrent
 299	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
 300	 */
 301	if (!timer_pending(&watchdog_timer)) {
 302		watchdog_timer.expires += WATCHDOG_INTERVAL;
 303		add_timer_on(&watchdog_timer, next_cpu);
 304	}
 305out:
 306	spin_unlock(&watchdog_lock);
 307}
 308
 309static inline void clocksource_start_watchdog(void)
 310{
 311	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
 312		return;
 313	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
 314	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
 315	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
 316	watchdog_running = 1;
 317}
 318
 319static inline void clocksource_stop_watchdog(void)
 320{
 321	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
 322		return;
 323	del_timer(&watchdog_timer);
 324	watchdog_running = 0;
 325}
 326
 327static inline void clocksource_reset_watchdog(void)
 328{
 329	struct clocksource *cs;
 330
 331	list_for_each_entry(cs, &watchdog_list, wd_list)
 332		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
 333}
 334
 335static void clocksource_resume_watchdog(void)
 336{
 337	atomic_inc(&watchdog_reset_pending);
 338}
 339
 340static void clocksource_enqueue_watchdog(struct clocksource *cs)
 341{
 342	INIT_LIST_HEAD(&cs->wd_list);
 343
 344	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 345		/* cs is a clocksource to be watched. */
 346		list_add(&cs->wd_list, &watchdog_list);
 347		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
 348	} else {
 349		/* cs is a watchdog. */
 350		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 351			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 352	}
 353}
 354
 355static void clocksource_select_watchdog(bool fallback)
 356{
 357	struct clocksource *cs, *old_wd;
 358	unsigned long flags;
 359
 360	spin_lock_irqsave(&watchdog_lock, flags);
 361	/* save current watchdog */
 362	old_wd = watchdog;
 363	if (fallback)
 364		watchdog = NULL;
 365
 366	list_for_each_entry(cs, &clocksource_list, list) {
 367		/* cs is a clocksource to be watched. */
 368		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
 369			continue;
 370
 371		/* Skip current if we were requested for a fallback. */
 372		if (fallback && cs == old_wd)
 373			continue;
 374
 375		/* Pick the best watchdog. */
 376		if (!watchdog || cs->rating > watchdog->rating)
 377			watchdog = cs;
 378	}
 379	/* If we failed to find a fallback restore the old one. */
 380	if (!watchdog)
 381		watchdog = old_wd;
 382
 383	/* If we changed the watchdog we need to reset cycles. */
 384	if (watchdog != old_wd)
 385		clocksource_reset_watchdog();
 386
 387	/* Check if the watchdog timer needs to be started. */
 388	clocksource_start_watchdog();
 389	spin_unlock_irqrestore(&watchdog_lock, flags);
 390}
 391
 392static void clocksource_dequeue_watchdog(struct clocksource *cs)
 393{
 394	if (cs != watchdog) {
 395		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 396			/* cs is a watched clocksource. */
 397			list_del_init(&cs->wd_list);
 398			/* Check if the watchdog timer needs to be stopped. */
 399			clocksource_stop_watchdog();
 400		}
 401	}
 402}
 403
 404static int __clocksource_watchdog_kthread(void)
 405{
 406	struct clocksource *cs, *tmp;
 407	unsigned long flags;
 408	int select = 0;
 409
 410	spin_lock_irqsave(&watchdog_lock, flags);
 411	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
 412		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 413			list_del_init(&cs->wd_list);
 414			__clocksource_change_rating(cs, 0);
 415			select = 1;
 416		}
 417		if (cs->flags & CLOCK_SOURCE_RESELECT) {
 418			cs->flags &= ~CLOCK_SOURCE_RESELECT;
 419			select = 1;
 420		}
 421	}
 422	/* Check if the watchdog timer needs to be stopped. */
 423	clocksource_stop_watchdog();
 424	spin_unlock_irqrestore(&watchdog_lock, flags);
 425
 426	return select;
 427}
 428
 429static int clocksource_watchdog_kthread(void *data)
 430{
 431	mutex_lock(&clocksource_mutex);
 432	if (__clocksource_watchdog_kthread())
 433		clocksource_select();
 434	mutex_unlock(&clocksource_mutex);
 435	return 0;
 436}
 437
 438static bool clocksource_is_watchdog(struct clocksource *cs)
 439{
 440	return cs == watchdog;
 441}
 442
 443#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
 444
 445static void clocksource_enqueue_watchdog(struct clocksource *cs)
 446{
 447	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 448		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 449}
 450
 451static void clocksource_select_watchdog(bool fallback) { }
 452static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
 453static inline void clocksource_resume_watchdog(void) { }
 454static inline int __clocksource_watchdog_kthread(void) { return 0; }
 455static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
 456void clocksource_mark_unstable(struct clocksource *cs) { }
 457
 458static inline void clocksource_watchdog_lock(unsigned long *flags) { }
 459static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
 460
 461#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
 462
 463static bool clocksource_is_suspend(struct clocksource *cs)
 464{
 465	return cs == suspend_clocksource;
 466}
 467
 468static void __clocksource_suspend_select(struct clocksource *cs)
 469{
 470	/*
 471	 * Skip the clocksource which will be stopped in suspend state.
 472	 */
 473	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
 474		return;
 475
 476	/*
 477	 * The nonstop clocksource can be selected as the suspend clocksource to
 478	 * calculate the suspend time, so it should not supply suspend/resume
 479	 * interfaces to suspend the nonstop clocksource when system suspends.
 480	 */
 481	if (cs->suspend || cs->resume) {
 482		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
 483			cs->name);
 484	}
 485
 486	/* Pick the best rating. */
 487	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
 488		suspend_clocksource = cs;
 489}
 490
 491/**
 492 * clocksource_suspend_select - Select the best clocksource for suspend timing
 493 * @fallback:	if select a fallback clocksource
 494 */
 495static void clocksource_suspend_select(bool fallback)
 496{
 497	struct clocksource *cs, *old_suspend;
 498
 499	old_suspend = suspend_clocksource;
 500	if (fallback)
 501		suspend_clocksource = NULL;
 502
 503	list_for_each_entry(cs, &clocksource_list, list) {
 504		/* Skip current if we were requested for a fallback. */
 505		if (fallback && cs == old_suspend)
 506			continue;
 507
 508		__clocksource_suspend_select(cs);
 509	}
 510}
 511
 512/**
 513 * clocksource_start_suspend_timing - Start measuring the suspend timing
 514 * @cs:			current clocksource from timekeeping
 515 * @start_cycles:	current cycles from timekeeping
 516 *
 517 * This function will save the start cycle values of suspend timer to calculate
 518 * the suspend time when resuming system.
 519 *
 520 * This function is called late in the suspend process from timekeeping_suspend(),
 521 * that means processes are freezed, non-boot cpus and interrupts are disabled
 522 * now. It is therefore possible to start the suspend timer without taking the
 523 * clocksource mutex.
 524 */
 525void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
 526{
 527	if (!suspend_clocksource)
 528		return;
 529
 530	/*
 531	 * If current clocksource is the suspend timer, we should use the
 532	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
 533	 * from suspend timer.
 534	 */
 535	if (clocksource_is_suspend(cs)) {
 536		suspend_start = start_cycles;
 537		return;
 538	}
 539
 540	if (suspend_clocksource->enable &&
 541	    suspend_clocksource->enable(suspend_clocksource)) {
 542		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
 543		return;
 544	}
 545
 546	suspend_start = suspend_clocksource->read(suspend_clocksource);
 547}
 548
 549/**
 550 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
 551 * @cs:		current clocksource from timekeeping
 552 * @cycle_now:	current cycles from timekeeping
 553 *
 554 * This function will calculate the suspend time from suspend timer.
 555 *
 556 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
 557 *
 558 * This function is called early in the resume process from timekeeping_resume(),
 559 * that means there is only one cpu, no processes are running and the interrupts
 560 * are disabled. It is therefore possible to stop the suspend timer without
 561 * taking the clocksource mutex.
 562 */
 563u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
 564{
 565	u64 now, delta, nsec = 0;
 566
 567	if (!suspend_clocksource)
 568		return 0;
 569
 570	/*
 571	 * If current clocksource is the suspend timer, we should use the
 572	 * tkr_mono.cycle_last value from timekeeping as current cycle to
 573	 * avoid same reading from suspend timer.
 574	 */
 575	if (clocksource_is_suspend(cs))
 576		now = cycle_now;
 577	else
 578		now = suspend_clocksource->read(suspend_clocksource);
 579
 580	if (now > suspend_start) {
 581		delta = clocksource_delta(now, suspend_start,
 582					  suspend_clocksource->mask);
 583		nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
 584				       suspend_clocksource->shift);
 585	}
 586
 587	/*
 588	 * Disable the suspend timer to save power if current clocksource is
 589	 * not the suspend timer.
 590	 */
 591	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
 592		suspend_clocksource->disable(suspend_clocksource);
 593
 594	return nsec;
 595}
 596
 597/**
 598 * clocksource_suspend - suspend the clocksource(s)
 599 */
 600void clocksource_suspend(void)
 601{
 602	struct clocksource *cs;
 603
 604	list_for_each_entry_reverse(cs, &clocksource_list, list)
 605		if (cs->suspend)
 606			cs->suspend(cs);
 607}
 608
 609/**
 610 * clocksource_resume - resume the clocksource(s)
 611 */
 612void clocksource_resume(void)
 613{
 614	struct clocksource *cs;
 615
 616	list_for_each_entry(cs, &clocksource_list, list)
 617		if (cs->resume)
 618			cs->resume(cs);
 619
 620	clocksource_resume_watchdog();
 621}
 622
 623/**
 624 * clocksource_touch_watchdog - Update watchdog
 625 *
 626 * Update the watchdog after exception contexts such as kgdb so as not
 627 * to incorrectly trip the watchdog. This might fail when the kernel
 628 * was stopped in code which holds watchdog_lock.
 629 */
 630void clocksource_touch_watchdog(void)
 631{
 632	clocksource_resume_watchdog();
 633}
 634
 635/**
 636 * clocksource_max_adjustment- Returns max adjustment amount
 637 * @cs:         Pointer to clocksource
 638 *
 639 */
 640static u32 clocksource_max_adjustment(struct clocksource *cs)
 641{
 642	u64 ret;
 643	/*
 644	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
 645	 */
 646	ret = (u64)cs->mult * 11;
 647	do_div(ret,100);
 648	return (u32)ret;
 649}
 650
 651/**
 652 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
 653 * @mult:	cycle to nanosecond multiplier
 654 * @shift:	cycle to nanosecond divisor (power of two)
 655 * @maxadj:	maximum adjustment value to mult (~11%)
 656 * @mask:	bitmask for two's complement subtraction of non 64 bit counters
 657 * @max_cyc:	maximum cycle value before potential overflow (does not include
 658 *		any safety margin)
 659 *
 660 * NOTE: This function includes a safety margin of 50%, in other words, we
 661 * return half the number of nanoseconds the hardware counter can technically
 662 * cover. This is done so that we can potentially detect problems caused by
 663 * delayed timers or bad hardware, which might result in time intervals that
 664 * are larger than what the math used can handle without overflows.
 665 */
 666u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
 667{
 668	u64 max_nsecs, max_cycles;
 669
 670	/*
 671	 * Calculate the maximum number of cycles that we can pass to the
 672	 * cyc2ns() function without overflowing a 64-bit result.
 673	 */
 674	max_cycles = ULLONG_MAX;
 675	do_div(max_cycles, mult+maxadj);
 676
 677	/*
 678	 * The actual maximum number of cycles we can defer the clocksource is
 679	 * determined by the minimum of max_cycles and mask.
 680	 * Note: Here we subtract the maxadj to make sure we don't sleep for
 681	 * too long if there's a large negative adjustment.
 682	 */
 683	max_cycles = min(max_cycles, mask);
 684	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
 685
 686	/* return the max_cycles value as well if requested */
 687	if (max_cyc)
 688		*max_cyc = max_cycles;
 689
 690	/* Return 50% of the actual maximum, so we can detect bad values */
 691	max_nsecs >>= 1;
 692
 693	return max_nsecs;
 694}
 695
 696/**
 697 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
 698 * @cs:         Pointer to clocksource to be updated
 699 *
 700 */
 701static inline void clocksource_update_max_deferment(struct clocksource *cs)
 702{
 703	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
 704						cs->maxadj, cs->mask,
 705						&cs->max_cycles);
 706}
 707
 708#ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
 709
 710static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
 711{
 712	struct clocksource *cs;
 713
 714	if (!finished_booting || list_empty(&clocksource_list))
 715		return NULL;
 716
 717	/*
 718	 * We pick the clocksource with the highest rating. If oneshot
 719	 * mode is active, we pick the highres valid clocksource with
 720	 * the best rating.
 721	 */
 722	list_for_each_entry(cs, &clocksource_list, list) {
 723		if (skipcur && cs == curr_clocksource)
 724			continue;
 725		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
 726			continue;
 727		return cs;
 728	}
 729	return NULL;
 730}
 731
 732static void __clocksource_select(bool skipcur)
 733{
 734	bool oneshot = tick_oneshot_mode_active();
 735	struct clocksource *best, *cs;
 736
 737	/* Find the best suitable clocksource */
 738	best = clocksource_find_best(oneshot, skipcur);
 739	if (!best)
 740		return;
 741
 742	if (!strlen(override_name))
 743		goto found;
 744
 745	/* Check for the override clocksource. */
 746	list_for_each_entry(cs, &clocksource_list, list) {
 747		if (skipcur && cs == curr_clocksource)
 748			continue;
 749		if (strcmp(cs->name, override_name) != 0)
 750			continue;
 751		/*
 752		 * Check to make sure we don't switch to a non-highres
 753		 * capable clocksource if the tick code is in oneshot
 754		 * mode (highres or nohz)
 755		 */
 756		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
 757			/* Override clocksource cannot be used. */
 758			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 759				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
 760					cs->name);
 761				override_name[0] = 0;
 762			} else {
 763				/*
 764				 * The override cannot be currently verified.
 765				 * Deferring to let the watchdog check.
 766				 */
 767				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
 768					cs->name);
 769			}
 770		} else
 771			/* Override clocksource can be used. */
 772			best = cs;
 773		break;
 774	}
 775
 776found:
 777	if (curr_clocksource != best && !timekeeping_notify(best)) {
 778		pr_info("Switched to clocksource %s\n", best->name);
 779		curr_clocksource = best;
 780	}
 781}
 782
 783/**
 784 * clocksource_select - Select the best clocksource available
 785 *
 786 * Private function. Must hold clocksource_mutex when called.
 787 *
 788 * Select the clocksource with the best rating, or the clocksource,
 789 * which is selected by userspace override.
 790 */
 791static void clocksource_select(void)
 792{
 793	__clocksource_select(false);
 794}
 795
 796static void clocksource_select_fallback(void)
 797{
 798	__clocksource_select(true);
 799}
 800
 801#else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
 802static inline void clocksource_select(void) { }
 803static inline void clocksource_select_fallback(void) { }
 804
 805#endif
 806
 807/*
 808 * clocksource_done_booting - Called near the end of core bootup
 809 *
 810 * Hack to avoid lots of clocksource churn at boot time.
 811 * We use fs_initcall because we want this to start before
 812 * device_initcall but after subsys_initcall.
 813 */
 814static int __init clocksource_done_booting(void)
 815{
 816	mutex_lock(&clocksource_mutex);
 817	curr_clocksource = clocksource_default_clock();
 818	finished_booting = 1;
 819	/*
 820	 * Run the watchdog first to eliminate unstable clock sources
 821	 */
 822	__clocksource_watchdog_kthread();
 823	clocksource_select();
 824	mutex_unlock(&clocksource_mutex);
 825	return 0;
 826}
 827fs_initcall(clocksource_done_booting);
 828
 829/*
 830 * Enqueue the clocksource sorted by rating
 831 */
 832static void clocksource_enqueue(struct clocksource *cs)
 833{
 834	struct list_head *entry = &clocksource_list;
 835	struct clocksource *tmp;
 836
 837	list_for_each_entry(tmp, &clocksource_list, list) {
 838		/* Keep track of the place, where to insert */
 839		if (tmp->rating < cs->rating)
 840			break;
 841		entry = &tmp->list;
 842	}
 843	list_add(&cs->list, entry);
 844}
 845
 846/**
 847 * __clocksource_update_freq_scale - Used update clocksource with new freq
 848 * @cs:		clocksource to be registered
 849 * @scale:	Scale factor multiplied against freq to get clocksource hz
 850 * @freq:	clocksource frequency (cycles per second) divided by scale
 851 *
 852 * This should only be called from the clocksource->enable() method.
 853 *
 854 * This *SHOULD NOT* be called directly! Please use the
 855 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
 856 * functions.
 857 */
 858void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
 859{
 860	u64 sec;
 861
 862	/*
 863	 * Default clocksources are *special* and self-define their mult/shift.
 864	 * But, you're not special, so you should specify a freq value.
 865	 */
 866	if (freq) {
 867		/*
 868		 * Calc the maximum number of seconds which we can run before
 869		 * wrapping around. For clocksources which have a mask > 32-bit
 870		 * we need to limit the max sleep time to have a good
 871		 * conversion precision. 10 minutes is still a reasonable
 872		 * amount. That results in a shift value of 24 for a
 873		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
 874		 * ~ 0.06ppm granularity for NTP.
 875		 */
 876		sec = cs->mask;
 877		do_div(sec, freq);
 878		do_div(sec, scale);
 879		if (!sec)
 880			sec = 1;
 881		else if (sec > 600 && cs->mask > UINT_MAX)
 882			sec = 600;
 883
 884		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
 885				       NSEC_PER_SEC / scale, sec * scale);
 886	}
 887	/*
 888	 * Ensure clocksources that have large 'mult' values don't overflow
 889	 * when adjusted.
 890	 */
 891	cs->maxadj = clocksource_max_adjustment(cs);
 892	while (freq && ((cs->mult + cs->maxadj < cs->mult)
 893		|| (cs->mult - cs->maxadj > cs->mult))) {
 894		cs->mult >>= 1;
 895		cs->shift--;
 896		cs->maxadj = clocksource_max_adjustment(cs);
 897	}
 898
 899	/*
 900	 * Only warn for *special* clocksources that self-define
 901	 * their mult/shift values and don't specify a freq.
 902	 */
 903	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
 904		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
 905		cs->name);
 906
 907	clocksource_update_max_deferment(cs);
 908
 909	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
 910		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
 911}
 912EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
 913
 914/**
 915 * __clocksource_register_scale - Used to install new clocksources
 916 * @cs:		clocksource to be registered
 917 * @scale:	Scale factor multiplied against freq to get clocksource hz
 918 * @freq:	clocksource frequency (cycles per second) divided by scale
 919 *
 920 * Returns -EBUSY if registration fails, zero otherwise.
 921 *
 922 * This *SHOULD NOT* be called directly! Please use the
 923 * clocksource_register_hz() or clocksource_register_khz helper functions.
 924 */
 925int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
 926{
 927	unsigned long flags;
 928
 929	clocksource_arch_init(cs);
 930
 931	if (cs->vdso_clock_mode < 0 ||
 932	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
 933		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
 934			cs->name, cs->vdso_clock_mode);
 935		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
 936	}
 937
 938	/* Initialize mult/shift and max_idle_ns */
 939	__clocksource_update_freq_scale(cs, scale, freq);
 940
 941	/* Add clocksource to the clocksource list */
 942	mutex_lock(&clocksource_mutex);
 943
 944	clocksource_watchdog_lock(&flags);
 945	clocksource_enqueue(cs);
 946	clocksource_enqueue_watchdog(cs);
 947	clocksource_watchdog_unlock(&flags);
 948
 949	clocksource_select();
 950	clocksource_select_watchdog(false);
 951	__clocksource_suspend_select(cs);
 952	mutex_unlock(&clocksource_mutex);
 953	return 0;
 954}
 955EXPORT_SYMBOL_GPL(__clocksource_register_scale);
 956
 957static void __clocksource_change_rating(struct clocksource *cs, int rating)
 958{
 959	list_del(&cs->list);
 960	cs->rating = rating;
 961	clocksource_enqueue(cs);
 962}
 963
 964/**
 965 * clocksource_change_rating - Change the rating of a registered clocksource
 966 * @cs:		clocksource to be changed
 967 * @rating:	new rating
 968 */
 969void clocksource_change_rating(struct clocksource *cs, int rating)
 970{
 971	unsigned long flags;
 972
 973	mutex_lock(&clocksource_mutex);
 974	clocksource_watchdog_lock(&flags);
 975	__clocksource_change_rating(cs, rating);
 976	clocksource_watchdog_unlock(&flags);
 977
 978	clocksource_select();
 979	clocksource_select_watchdog(false);
 980	clocksource_suspend_select(false);
 981	mutex_unlock(&clocksource_mutex);
 982}
 983EXPORT_SYMBOL(clocksource_change_rating);
 984
 985/*
 986 * Unbind clocksource @cs. Called with clocksource_mutex held
 987 */
 988static int clocksource_unbind(struct clocksource *cs)
 989{
 990	unsigned long flags;
 991
 992	if (clocksource_is_watchdog(cs)) {
 993		/* Select and try to install a replacement watchdog. */
 994		clocksource_select_watchdog(true);
 995		if (clocksource_is_watchdog(cs))
 996			return -EBUSY;
 997	}
 998
 999	if (cs == curr_clocksource) {
1000		/* Select and try to install a replacement clock source */
1001		clocksource_select_fallback();
1002		if (curr_clocksource == cs)
1003			return -EBUSY;
1004	}
1005
1006	if (clocksource_is_suspend(cs)) {
1007		/*
1008		 * Select and try to install a replacement suspend clocksource.
1009		 * If no replacement suspend clocksource, we will just let the
1010		 * clocksource go and have no suspend clocksource.
1011		 */
1012		clocksource_suspend_select(true);
1013	}
1014
1015	clocksource_watchdog_lock(&flags);
1016	clocksource_dequeue_watchdog(cs);
1017	list_del_init(&cs->list);
1018	clocksource_watchdog_unlock(&flags);
1019
1020	return 0;
1021}
1022
1023/**
1024 * clocksource_unregister - remove a registered clocksource
1025 * @cs:	clocksource to be unregistered
1026 */
1027int clocksource_unregister(struct clocksource *cs)
1028{
1029	int ret = 0;
1030
1031	mutex_lock(&clocksource_mutex);
1032	if (!list_empty(&cs->list))
1033		ret = clocksource_unbind(cs);
1034	mutex_unlock(&clocksource_mutex);
1035	return ret;
1036}
1037EXPORT_SYMBOL(clocksource_unregister);
1038
1039#ifdef CONFIG_SYSFS
1040/**
1041 * current_clocksource_show - sysfs interface for current clocksource
1042 * @dev:	unused
1043 * @attr:	unused
1044 * @buf:	char buffer to be filled with clocksource list
1045 *
1046 * Provides sysfs interface for listing current clocksource.
1047 */
1048static ssize_t current_clocksource_show(struct device *dev,
1049					struct device_attribute *attr,
1050					char *buf)
1051{
1052	ssize_t count = 0;
1053
1054	mutex_lock(&clocksource_mutex);
1055	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1056	mutex_unlock(&clocksource_mutex);
1057
1058	return count;
1059}
1060
1061ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1062{
1063	size_t ret = cnt;
1064
1065	/* strings from sysfs write are not 0 terminated! */
1066	if (!cnt || cnt >= CS_NAME_LEN)
1067		return -EINVAL;
1068
1069	/* strip of \n: */
1070	if (buf[cnt-1] == '\n')
1071		cnt--;
1072	if (cnt > 0)
1073		memcpy(dst, buf, cnt);
1074	dst[cnt] = 0;
1075	return ret;
1076}
1077
1078/**
1079 * current_clocksource_store - interface for manually overriding clocksource
1080 * @dev:	unused
1081 * @attr:	unused
1082 * @buf:	name of override clocksource
1083 * @count:	length of buffer
1084 *
1085 * Takes input from sysfs interface for manually overriding the default
1086 * clocksource selection.
1087 */
1088static ssize_t current_clocksource_store(struct device *dev,
1089					 struct device_attribute *attr,
1090					 const char *buf, size_t count)
1091{
1092	ssize_t ret;
1093
1094	mutex_lock(&clocksource_mutex);
1095
1096	ret = sysfs_get_uname(buf, override_name, count);
1097	if (ret >= 0)
1098		clocksource_select();
1099
1100	mutex_unlock(&clocksource_mutex);
1101
1102	return ret;
1103}
1104static DEVICE_ATTR_RW(current_clocksource);
1105
1106/**
1107 * unbind_clocksource_store - interface for manually unbinding clocksource
1108 * @dev:	unused
1109 * @attr:	unused
1110 * @buf:	unused
1111 * @count:	length of buffer
1112 *
1113 * Takes input from sysfs interface for manually unbinding a clocksource.
1114 */
1115static ssize_t unbind_clocksource_store(struct device *dev,
1116					struct device_attribute *attr,
1117					const char *buf, size_t count)
1118{
1119	struct clocksource *cs;
1120	char name[CS_NAME_LEN];
1121	ssize_t ret;
1122
1123	ret = sysfs_get_uname(buf, name, count);
1124	if (ret < 0)
1125		return ret;
1126
1127	ret = -ENODEV;
1128	mutex_lock(&clocksource_mutex);
1129	list_for_each_entry(cs, &clocksource_list, list) {
1130		if (strcmp(cs->name, name))
1131			continue;
1132		ret = clocksource_unbind(cs);
1133		break;
1134	}
1135	mutex_unlock(&clocksource_mutex);
1136
1137	return ret ? ret : count;
1138}
1139static DEVICE_ATTR_WO(unbind_clocksource);
1140
1141/**
1142 * available_clocksource_show - sysfs interface for listing clocksource
1143 * @dev:	unused
1144 * @attr:	unused
1145 * @buf:	char buffer to be filled with clocksource list
1146 *
1147 * Provides sysfs interface for listing registered clocksources
1148 */
1149static ssize_t available_clocksource_show(struct device *dev,
1150					  struct device_attribute *attr,
1151					  char *buf)
1152{
1153	struct clocksource *src;
1154	ssize_t count = 0;
1155
1156	mutex_lock(&clocksource_mutex);
1157	list_for_each_entry(src, &clocksource_list, list) {
1158		/*
1159		 * Don't show non-HRES clocksource if the tick code is
1160		 * in one shot mode (highres=on or nohz=on)
1161		 */
1162		if (!tick_oneshot_mode_active() ||
1163		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1164			count += snprintf(buf + count,
1165				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1166				  "%s ", src->name);
1167	}
1168	mutex_unlock(&clocksource_mutex);
1169
1170	count += snprintf(buf + count,
1171			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1172
1173	return count;
1174}
1175static DEVICE_ATTR_RO(available_clocksource);
1176
1177static struct attribute *clocksource_attrs[] = {
1178	&dev_attr_current_clocksource.attr,
1179	&dev_attr_unbind_clocksource.attr,
1180	&dev_attr_available_clocksource.attr,
1181	NULL
1182};
1183ATTRIBUTE_GROUPS(clocksource);
1184
1185static struct bus_type clocksource_subsys = {
1186	.name = "clocksource",
1187	.dev_name = "clocksource",
1188};
1189
1190static struct device device_clocksource = {
1191	.id	= 0,
1192	.bus	= &clocksource_subsys,
1193	.groups	= clocksource_groups,
1194};
1195
1196static int __init init_clocksource_sysfs(void)
1197{
1198	int error = subsys_system_register(&clocksource_subsys, NULL);
1199
1200	if (!error)
1201		error = device_register(&device_clocksource);
1202
1203	return error;
1204}
1205
1206device_initcall(init_clocksource_sysfs);
1207#endif /* CONFIG_SYSFS */
1208
1209/**
1210 * boot_override_clocksource - boot clock override
1211 * @str:	override name
1212 *
1213 * Takes a clocksource= boot argument and uses it
1214 * as the clocksource override name.
1215 */
1216static int __init boot_override_clocksource(char* str)
1217{
1218	mutex_lock(&clocksource_mutex);
1219	if (str)
1220		strlcpy(override_name, str, sizeof(override_name));
1221	mutex_unlock(&clocksource_mutex);
1222	return 1;
1223}
1224
1225__setup("clocksource=", boot_override_clocksource);
1226
1227/**
1228 * boot_override_clock - Compatibility layer for deprecated boot option
1229 * @str:	override name
1230 *
1231 * DEPRECATED! Takes a clock= boot argument and uses it
1232 * as the clocksource override name
1233 */
1234static int __init boot_override_clock(char* str)
1235{
1236	if (!strcmp(str, "pmtmr")) {
1237		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1238		return boot_override_clocksource("acpi_pm");
1239	}
1240	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1241	return boot_override_clocksource(str);
1242}
1243
1244__setup("clock=", boot_override_clock);