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#include <linux/prandom.h>
  18#include <linux/cpu.h>
  19
  20#include "tick-internal.h"
  21#include "timekeeping_internal.h"
  22
 
 
 
 
 
 
 
 
 
 
 
 
  23/**
  24 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
  25 * @mult:	pointer to mult variable
  26 * @shift:	pointer to shift variable
  27 * @from:	frequency to convert from
  28 * @to:		frequency to convert to
  29 * @maxsec:	guaranteed runtime conversion range in seconds
  30 *
  31 * The function evaluates the shift/mult pair for the scaled math
  32 * operations of clocksources and clockevents.
  33 *
  34 * @to and @from are frequency values in HZ. For clock sources @to is
  35 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
  36 * event @to is the counter frequency and @from is NSEC_PER_SEC.
  37 *
  38 * The @maxsec conversion range argument controls the time frame in
  39 * seconds which must be covered by the runtime conversion with the
  40 * calculated mult and shift factors. This guarantees that no 64bit
  41 * overflow happens when the input value of the conversion is
  42 * multiplied with the calculated mult factor. Larger ranges may
  43 * reduce the conversion accuracy by choosing smaller mult and shift
  44 * factors.
  45 */
  46void
  47clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
  48{
  49	u64 tmp;
  50	u32 sft, sftacc= 32;
  51
  52	/*
  53	 * Calculate the shift factor which is limiting the conversion
  54	 * range:
  55	 */
  56	tmp = ((u64)maxsec * from) >> 32;
  57	while (tmp) {
  58		tmp >>=1;
  59		sftacc--;
  60	}
  61
  62	/*
  63	 * Find the conversion shift/mult pair which has the best
  64	 * accuracy and fits the maxsec conversion range:
  65	 */
  66	for (sft = 32; sft > 0; sft--) {
  67		tmp = (u64) to << sft;
  68		tmp += from / 2;
  69		do_div(tmp, from);
  70		if ((tmp >> sftacc) == 0)
  71			break;
  72	}
  73	*mult = tmp;
  74	*shift = sft;
  75}
  76EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
  77
  78/*[Clocksource internal variables]---------
  79 * curr_clocksource:
  80 *	currently selected clocksource.
  81 * suspend_clocksource:
  82 *	used to calculate the suspend time.
  83 * clocksource_list:
  84 *	linked list with the registered clocksources
  85 * clocksource_mutex:
  86 *	protects manipulations to curr_clocksource and the clocksource_list
  87 * override_name:
  88 *	Name of the user-specified clocksource.
  89 */
  90static struct clocksource *curr_clocksource;
  91static struct clocksource *suspend_clocksource;
  92static LIST_HEAD(clocksource_list);
  93static DEFINE_MUTEX(clocksource_mutex);
  94static char override_name[CS_NAME_LEN];
  95static int finished_booting;
  96static u64 suspend_start;
  97
  98/*
  99 * Interval: 0.5sec.
 100 */
 101#define WATCHDOG_INTERVAL (HZ >> 1)
 102#define WATCHDOG_INTERVAL_MAX_NS ((2 * WATCHDOG_INTERVAL) * (NSEC_PER_SEC / HZ))
 103
 104/*
 105 * Threshold: 0.0312s, when doubled: 0.0625s.
 106 * Also a default for cs->uncertainty_margin when registering clocks.
 107 */
 108#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
 109
 110/*
 111 * Maximum permissible delay between two readouts of the watchdog
 112 * clocksource surrounding a read of the clocksource being validated.
 113 * This delay could be due to SMIs, NMIs, or to VCPU preemptions.  Used as
 114 * a lower bound for cs->uncertainty_margin values when registering clocks.
 115 *
 116 * The default of 500 parts per million is based on NTP's limits.
 117 * If a clocksource is good enough for NTP, it is good enough for us!
 
 
 
 
 
 
 
 118 */
 119#ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
 120#define MAX_SKEW_USEC	CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
 121#else
 122#define MAX_SKEW_USEC	(125 * WATCHDOG_INTERVAL / HZ)
 123#endif
 124
 
 
 
 
 
 
 
 125#define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
 126
 127#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
 128static void clocksource_watchdog_work(struct work_struct *work);
 129static void clocksource_select(void);
 130
 131static LIST_HEAD(watchdog_list);
 132static struct clocksource *watchdog;
 133static struct timer_list watchdog_timer;
 134static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
 135static DEFINE_SPINLOCK(watchdog_lock);
 136static int watchdog_running;
 137static atomic_t watchdog_reset_pending;
 138static int64_t watchdog_max_interval;
 139
 140static inline void clocksource_watchdog_lock(unsigned long *flags)
 141{
 142	spin_lock_irqsave(&watchdog_lock, *flags);
 143}
 144
 145static inline void clocksource_watchdog_unlock(unsigned long *flags)
 146{
 147	spin_unlock_irqrestore(&watchdog_lock, *flags);
 148}
 149
 150static int clocksource_watchdog_kthread(void *data);
 151static void __clocksource_change_rating(struct clocksource *cs, int rating);
 152
 153static void clocksource_watchdog_work(struct work_struct *work)
 154{
 155	/*
 156	 * We cannot directly run clocksource_watchdog_kthread() here, because
 157	 * clocksource_select() calls timekeeping_notify() which uses
 158	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
 159	 * lock inversions wrt CPU hotplug.
 160	 *
 161	 * Also, we only ever run this work once or twice during the lifetime
 162	 * of the kernel, so there is no point in creating a more permanent
 163	 * kthread for this.
 164	 *
 165	 * If kthread_run fails the next watchdog scan over the
 166	 * watchdog_list will find the unstable clock again.
 167	 */
 168	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
 169}
 170
 
 
 
 
 
 
 
 171static void __clocksource_unstable(struct clocksource *cs)
 172{
 173	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
 174	cs->flags |= CLOCK_SOURCE_UNSTABLE;
 175
 176	/*
 177	 * If the clocksource is registered clocksource_watchdog_kthread() will
 178	 * re-rate and re-select.
 179	 */
 180	if (list_empty(&cs->list)) {
 181		cs->rating = 0;
 182		return;
 183	}
 184
 185	if (cs->mark_unstable)
 186		cs->mark_unstable(cs);
 187
 188	/* kick clocksource_watchdog_kthread() */
 189	if (finished_booting)
 190		schedule_work(&watchdog_work);
 191}
 192
 193/**
 194 * clocksource_mark_unstable - mark clocksource unstable via watchdog
 195 * @cs:		clocksource to be marked unstable
 196 *
 197 * This function is called by the x86 TSC code to mark clocksources as unstable;
 198 * it defers demotion and re-selection to a kthread.
 199 */
 200void clocksource_mark_unstable(struct clocksource *cs)
 201{
 202	unsigned long flags;
 203
 204	spin_lock_irqsave(&watchdog_lock, flags);
 205	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
 206		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
 207			list_add(&cs->wd_list, &watchdog_list);
 208		__clocksource_unstable(cs);
 209	}
 210	spin_unlock_irqrestore(&watchdog_lock, flags);
 211}
 212
 213ulong max_cswd_read_retries = 2;
 214module_param(max_cswd_read_retries, ulong, 0644);
 215EXPORT_SYMBOL_GPL(max_cswd_read_retries);
 216static int verify_n_cpus = 8;
 217module_param(verify_n_cpus, int, 0644);
 218
 219enum wd_read_status {
 220	WD_READ_SUCCESS,
 221	WD_READ_UNSTABLE,
 222	WD_READ_SKIP
 223};
 224
 225static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
 226{
 227	unsigned int nretries;
 228	u64 wd_end, wd_end2, wd_delta;
 229	int64_t wd_delay, wd_seq_delay;
 
 230
 231	for (nretries = 0; nretries <= max_cswd_read_retries; nretries++) {
 
 232		local_irq_disable();
 233		*wdnow = watchdog->read(watchdog);
 234		*csnow = cs->read(cs);
 235		wd_end = watchdog->read(watchdog);
 236		wd_end2 = watchdog->read(watchdog);
 237		local_irq_enable();
 238
 239		wd_delta = clocksource_delta(wd_end, *wdnow, watchdog->mask);
 240		wd_delay = clocksource_cyc2ns(wd_delta, watchdog->mult,
 241					      watchdog->shift);
 242		if (wd_delay <= WATCHDOG_MAX_SKEW) {
 243			if (nretries > 1 || nretries >= max_cswd_read_retries) {
 244				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
 245					smp_processor_id(), watchdog->name, nretries);
 246			}
 247			return WD_READ_SUCCESS;
 248		}
 249
 250		/*
 251		 * Now compute delay in consecutive watchdog read to see if
 252		 * there is too much external interferences that cause
 253		 * significant delay in reading both clocksource and watchdog.
 254		 *
 255		 * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2,
 256		 * report system busy, reinit the watchdog and skip the current
 257		 * watchdog test.
 258		 */
 259		wd_delta = clocksource_delta(wd_end2, wd_end, watchdog->mask);
 260		wd_seq_delay = clocksource_cyc2ns(wd_delta, watchdog->mult, watchdog->shift);
 261		if (wd_seq_delay > WATCHDOG_MAX_SKEW/2)
 262			goto skip_test;
 263	}
 264
 265	pr_warn("timekeeping watchdog on CPU%d: wd-%s-wd excessive read-back delay of %lldns vs. limit of %ldns, wd-wd read-back delay only %lldns, attempt %d, marking %s unstable\n",
 266		smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name);
 267	return WD_READ_UNSTABLE;
 268
 269skip_test:
 270	pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
 271		smp_processor_id(), watchdog->name, wd_seq_delay);
 272	pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
 273		cs->name, wd_delay);
 274	return WD_READ_SKIP;
 275}
 276
 277static u64 csnow_mid;
 278static cpumask_t cpus_ahead;
 279static cpumask_t cpus_behind;
 280static cpumask_t cpus_chosen;
 281
 282static void clocksource_verify_choose_cpus(void)
 283{
 284	int cpu, i, n = verify_n_cpus;
 285
 286	if (n < 0) {
 287		/* Check all of the CPUs. */
 288		cpumask_copy(&cpus_chosen, cpu_online_mask);
 289		cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
 290		return;
 291	}
 292
 293	/* If no checking desired, or no other CPU to check, leave. */
 294	cpumask_clear(&cpus_chosen);
 295	if (n == 0 || num_online_cpus() <= 1)
 296		return;
 297
 298	/* Make sure to select at least one CPU other than the current CPU. */
 299	cpu = cpumask_first(cpu_online_mask);
 300	if (cpu == smp_processor_id())
 301		cpu = cpumask_next(cpu, cpu_online_mask);
 302	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
 303		return;
 304	cpumask_set_cpu(cpu, &cpus_chosen);
 305
 306	/* Force a sane value for the boot parameter. */
 307	if (n > nr_cpu_ids)
 308		n = nr_cpu_ids;
 309
 310	/*
 311	 * Randomly select the specified number of CPUs.  If the same
 312	 * CPU is selected multiple times, that CPU is checked only once,
 313	 * and no replacement CPU is selected.  This gracefully handles
 314	 * situations where verify_n_cpus is greater than the number of
 315	 * CPUs that are currently online.
 316	 */
 317	for (i = 1; i < n; i++) {
 318		cpu = get_random_u32_below(nr_cpu_ids);
 319		cpu = cpumask_next(cpu - 1, cpu_online_mask);
 320		if (cpu >= nr_cpu_ids)
 321			cpu = cpumask_first(cpu_online_mask);
 322		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
 323			cpumask_set_cpu(cpu, &cpus_chosen);
 324	}
 325
 326	/* Don't verify ourselves. */
 327	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
 328}
 329
 330static void clocksource_verify_one_cpu(void *csin)
 331{
 332	struct clocksource *cs = (struct clocksource *)csin;
 333
 334	csnow_mid = cs->read(cs);
 335}
 336
 337void clocksource_verify_percpu(struct clocksource *cs)
 338{
 339	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
 340	u64 csnow_begin, csnow_end;
 341	int cpu, testcpu;
 342	s64 delta;
 343
 344	if (verify_n_cpus == 0)
 345		return;
 346	cpumask_clear(&cpus_ahead);
 347	cpumask_clear(&cpus_behind);
 348	cpus_read_lock();
 349	preempt_disable();
 350	clocksource_verify_choose_cpus();
 351	if (cpumask_empty(&cpus_chosen)) {
 352		preempt_enable();
 353		cpus_read_unlock();
 354		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
 355		return;
 356	}
 357	testcpu = smp_processor_id();
 358	pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
 
 
 359	for_each_cpu(cpu, &cpus_chosen) {
 360		if (cpu == testcpu)
 361			continue;
 362		csnow_begin = cs->read(cs);
 363		smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
 364		csnow_end = cs->read(cs);
 365		delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
 366		if (delta < 0)
 367			cpumask_set_cpu(cpu, &cpus_behind);
 368		delta = (csnow_end - csnow_mid) & cs->mask;
 369		if (delta < 0)
 370			cpumask_set_cpu(cpu, &cpus_ahead);
 371		delta = clocksource_delta(csnow_end, csnow_begin, cs->mask);
 372		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
 373		if (cs_nsec > cs_nsec_max)
 374			cs_nsec_max = cs_nsec;
 375		if (cs_nsec < cs_nsec_min)
 376			cs_nsec_min = cs_nsec;
 377	}
 378	preempt_enable();
 
 379	cpus_read_unlock();
 380	if (!cpumask_empty(&cpus_ahead))
 381		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
 382			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
 383	if (!cpumask_empty(&cpus_behind))
 384		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
 385			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
 386	if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind))
 387		pr_warn("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
 388			testcpu, cs_nsec_min, cs_nsec_max, cs->name);
 389}
 390EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
 391
 392static inline void clocksource_reset_watchdog(void)
 393{
 394	struct clocksource *cs;
 395
 396	list_for_each_entry(cs, &watchdog_list, wd_list)
 397		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
 398}
 399
 400
 401static void clocksource_watchdog(struct timer_list *unused)
 402{
 403	u64 csnow, wdnow, cslast, wdlast, delta;
 404	int64_t wd_nsec, cs_nsec, interval;
 
 405	int next_cpu, reset_pending;
 406	struct clocksource *cs;
 407	enum wd_read_status read_ret;
 408	unsigned long extra_wait = 0;
 409	u32 md;
 410
 411	spin_lock(&watchdog_lock);
 412	if (!watchdog_running)
 413		goto out;
 414
 415	reset_pending = atomic_read(&watchdog_reset_pending);
 416
 417	list_for_each_entry(cs, &watchdog_list, wd_list) {
 418
 419		/* Clocksource already marked unstable? */
 420		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 421			if (finished_booting)
 422				schedule_work(&watchdog_work);
 423			continue;
 424		}
 425
 426		read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
 427
 428		if (read_ret == WD_READ_UNSTABLE) {
 429			/* Clock readout unreliable, so give it up. */
 430			__clocksource_unstable(cs);
 431			continue;
 432		}
 433
 434		/*
 435		 * When WD_READ_SKIP is returned, it means the system is likely
 436		 * under very heavy load, where the latency of reading
 437		 * watchdog/clocksource is very big, and affect the accuracy of
 438		 * watchdog check. So give system some space and suspend the
 439		 * watchdog check for 5 minutes.
 440		 */
 441		if (read_ret == WD_READ_SKIP) {
 442			/*
 443			 * As the watchdog timer will be suspended, and
 444			 * cs->last could keep unchanged for 5 minutes, reset
 445			 * the counters.
 446			 */
 447			clocksource_reset_watchdog();
 448			extra_wait = HZ * 300;
 449			break;
 450		}
 451
 452		/* Clocksource initialized ? */
 453		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
 454		    atomic_read(&watchdog_reset_pending)) {
 455			cs->flags |= CLOCK_SOURCE_WATCHDOG;
 456			cs->wd_last = wdnow;
 457			cs->cs_last = csnow;
 458			continue;
 459		}
 460
 461		delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
 462		wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
 463					     watchdog->shift);
 464
 465		delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
 466		cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
 467		wdlast = cs->wd_last; /* save these in case we print them */
 468		cslast = cs->cs_last;
 469		cs->cs_last = csnow;
 470		cs->wd_last = wdnow;
 471
 472		if (atomic_read(&watchdog_reset_pending))
 473			continue;
 474
 475		/*
 476		 * The processing of timer softirqs can get delayed (usually
 477		 * on account of ksoftirqd not getting to run in a timely
 478		 * manner), which causes the watchdog interval to stretch.
 479		 * Skew detection may fail for longer watchdog intervals
 480		 * on account of fixed margins being used.
 481		 * Some clocksources, e.g. acpi_pm, cannot tolerate
 482		 * watchdog intervals longer than a few seconds.
 483		 */
 484		interval = max(cs_nsec, wd_nsec);
 485		if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) {
 486			if (system_state > SYSTEM_SCHEDULING &&
 487			    interval > 2 * watchdog_max_interval) {
 488				watchdog_max_interval = interval;
 489				pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n",
 490					cs_nsec, wd_nsec);
 491			}
 492			watchdog_timer.expires = jiffies;
 493			continue;
 494		}
 495
 496		/* Check the deviation from the watchdog clocksource. */
 497		md = cs->uncertainty_margin + watchdog->uncertainty_margin;
 498		if (abs(cs_nsec - wd_nsec) > md) {
 499			s64 cs_wd_msec;
 500			s64 wd_msec;
 501			u32 wd_rem;
 502
 503			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
 504				smp_processor_id(), cs->name);
 505			pr_warn("                      '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
 506				watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
 507			pr_warn("                      '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
 508				cs->name, cs_nsec, csnow, cslast, cs->mask);
 509			cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
 510			wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
 511			pr_warn("                      Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
 512				cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
 513			if (curr_clocksource == cs)
 514				pr_warn("                      '%s' is current clocksource.\n", cs->name);
 515			else if (curr_clocksource)
 516				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
 517			else
 518				pr_warn("                      No current clocksource.\n");
 519			__clocksource_unstable(cs);
 520			continue;
 521		}
 522
 523		if (cs == curr_clocksource && cs->tick_stable)
 524			cs->tick_stable(cs);
 525
 526		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
 527		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
 528		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
 529			/* Mark it valid for high-res. */
 530			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 531
 532			/*
 533			 * clocksource_done_booting() will sort it if
 534			 * finished_booting is not set yet.
 535			 */
 536			if (!finished_booting)
 537				continue;
 538
 539			/*
 540			 * If this is not the current clocksource let
 541			 * the watchdog thread reselect it. Due to the
 542			 * change to high res this clocksource might
 543			 * be preferred now. If it is the current
 544			 * clocksource let the tick code know about
 545			 * that change.
 546			 */
 547			if (cs != curr_clocksource) {
 548				cs->flags |= CLOCK_SOURCE_RESELECT;
 549				schedule_work(&watchdog_work);
 550			} else {
 551				tick_clock_notify();
 552			}
 553		}
 554	}
 555
 556	/*
 557	 * We only clear the watchdog_reset_pending, when we did a
 558	 * full cycle through all clocksources.
 559	 */
 560	if (reset_pending)
 561		atomic_dec(&watchdog_reset_pending);
 562
 563	/*
 564	 * Cycle through CPUs to check if the CPUs stay synchronized
 565	 * to each other.
 566	 */
 567	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
 568	if (next_cpu >= nr_cpu_ids)
 569		next_cpu = cpumask_first(cpu_online_mask);
 570
 571	/*
 572	 * Arm timer if not already pending: could race with concurrent
 573	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
 574	 */
 575	if (!timer_pending(&watchdog_timer)) {
 576		watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
 577		add_timer_on(&watchdog_timer, next_cpu);
 578	}
 579out:
 580	spin_unlock(&watchdog_lock);
 581}
 582
 583static inline void clocksource_start_watchdog(void)
 584{
 585	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
 586		return;
 587	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
 588	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
 589	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
 590	watchdog_running = 1;
 591}
 592
 593static inline void clocksource_stop_watchdog(void)
 594{
 595	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
 596		return;
 597	del_timer(&watchdog_timer);
 598	watchdog_running = 0;
 599}
 600
 601static void clocksource_resume_watchdog(void)
 602{
 603	atomic_inc(&watchdog_reset_pending);
 604}
 605
 606static void clocksource_enqueue_watchdog(struct clocksource *cs)
 607{
 608	INIT_LIST_HEAD(&cs->wd_list);
 609
 610	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 611		/* cs is a clocksource to be watched. */
 612		list_add(&cs->wd_list, &watchdog_list);
 613		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
 614	} else {
 615		/* cs is a watchdog. */
 616		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 617			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 618	}
 619}
 620
 621static void clocksource_select_watchdog(bool fallback)
 622{
 623	struct clocksource *cs, *old_wd;
 624	unsigned long flags;
 625
 626	spin_lock_irqsave(&watchdog_lock, flags);
 627	/* save current watchdog */
 628	old_wd = watchdog;
 629	if (fallback)
 630		watchdog = NULL;
 631
 632	list_for_each_entry(cs, &clocksource_list, list) {
 633		/* cs is a clocksource to be watched. */
 634		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
 635			continue;
 636
 637		/* Skip current if we were requested for a fallback. */
 638		if (fallback && cs == old_wd)
 639			continue;
 640
 641		/* Pick the best watchdog. */
 642		if (!watchdog || cs->rating > watchdog->rating)
 643			watchdog = cs;
 644	}
 645	/* If we failed to find a fallback restore the old one. */
 646	if (!watchdog)
 647		watchdog = old_wd;
 648
 649	/* If we changed the watchdog we need to reset cycles. */
 650	if (watchdog != old_wd)
 651		clocksource_reset_watchdog();
 652
 653	/* Check if the watchdog timer needs to be started. */
 654	clocksource_start_watchdog();
 655	spin_unlock_irqrestore(&watchdog_lock, flags);
 656}
 657
 658static void clocksource_dequeue_watchdog(struct clocksource *cs)
 659{
 660	if (cs != watchdog) {
 661		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 662			/* cs is a watched clocksource. */
 663			list_del_init(&cs->wd_list);
 664			/* Check if the watchdog timer needs to be stopped. */
 665			clocksource_stop_watchdog();
 666		}
 667	}
 668}
 669
 670static int __clocksource_watchdog_kthread(void)
 671{
 672	struct clocksource *cs, *tmp;
 673	unsigned long flags;
 674	int select = 0;
 675
 676	/* Do any required per-CPU skew verification. */
 677	if (curr_clocksource &&
 678	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
 679	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
 680		clocksource_verify_percpu(curr_clocksource);
 681
 682	spin_lock_irqsave(&watchdog_lock, flags);
 683	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
 684		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 685			list_del_init(&cs->wd_list);
 686			__clocksource_change_rating(cs, 0);
 687			select = 1;
 688		}
 689		if (cs->flags & CLOCK_SOURCE_RESELECT) {
 690			cs->flags &= ~CLOCK_SOURCE_RESELECT;
 691			select = 1;
 692		}
 693	}
 694	/* Check if the watchdog timer needs to be stopped. */
 695	clocksource_stop_watchdog();
 696	spin_unlock_irqrestore(&watchdog_lock, flags);
 697
 698	return select;
 699}
 700
 701static int clocksource_watchdog_kthread(void *data)
 702{
 703	mutex_lock(&clocksource_mutex);
 704	if (__clocksource_watchdog_kthread())
 705		clocksource_select();
 706	mutex_unlock(&clocksource_mutex);
 707	return 0;
 708}
 709
 710static bool clocksource_is_watchdog(struct clocksource *cs)
 711{
 712	return cs == watchdog;
 713}
 714
 715#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
 716
 717static void clocksource_enqueue_watchdog(struct clocksource *cs)
 718{
 719	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 720		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 721}
 722
 723static void clocksource_select_watchdog(bool fallback) { }
 724static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
 725static inline void clocksource_resume_watchdog(void) { }
 726static inline int __clocksource_watchdog_kthread(void) { return 0; }
 727static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
 728void clocksource_mark_unstable(struct clocksource *cs) { }
 729
 730static inline void clocksource_watchdog_lock(unsigned long *flags) { }
 731static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
 732
 733#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
 734
 735static bool clocksource_is_suspend(struct clocksource *cs)
 736{
 737	return cs == suspend_clocksource;
 738}
 739
 740static void __clocksource_suspend_select(struct clocksource *cs)
 741{
 742	/*
 743	 * Skip the clocksource which will be stopped in suspend state.
 744	 */
 745	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
 746		return;
 747
 748	/*
 749	 * The nonstop clocksource can be selected as the suspend clocksource to
 750	 * calculate the suspend time, so it should not supply suspend/resume
 751	 * interfaces to suspend the nonstop clocksource when system suspends.
 752	 */
 753	if (cs->suspend || cs->resume) {
 754		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
 755			cs->name);
 756	}
 757
 758	/* Pick the best rating. */
 759	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
 760		suspend_clocksource = cs;
 761}
 762
 763/**
 764 * clocksource_suspend_select - Select the best clocksource for suspend timing
 765 * @fallback:	if select a fallback clocksource
 766 */
 767static void clocksource_suspend_select(bool fallback)
 768{
 769	struct clocksource *cs, *old_suspend;
 770
 771	old_suspend = suspend_clocksource;
 772	if (fallback)
 773		suspend_clocksource = NULL;
 774
 775	list_for_each_entry(cs, &clocksource_list, list) {
 776		/* Skip current if we were requested for a fallback. */
 777		if (fallback && cs == old_suspend)
 778			continue;
 779
 780		__clocksource_suspend_select(cs);
 781	}
 782}
 783
 784/**
 785 * clocksource_start_suspend_timing - Start measuring the suspend timing
 786 * @cs:			current clocksource from timekeeping
 787 * @start_cycles:	current cycles from timekeeping
 788 *
 789 * This function will save the start cycle values of suspend timer to calculate
 790 * the suspend time when resuming system.
 791 *
 792 * This function is called late in the suspend process from timekeeping_suspend(),
 793 * that means processes are frozen, non-boot cpus and interrupts are disabled
 794 * now. It is therefore possible to start the suspend timer without taking the
 795 * clocksource mutex.
 796 */
 797void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
 798{
 799	if (!suspend_clocksource)
 800		return;
 801
 802	/*
 803	 * If current clocksource is the suspend timer, we should use the
 804	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
 805	 * from suspend timer.
 806	 */
 807	if (clocksource_is_suspend(cs)) {
 808		suspend_start = start_cycles;
 809		return;
 810	}
 811
 812	if (suspend_clocksource->enable &&
 813	    suspend_clocksource->enable(suspend_clocksource)) {
 814		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
 815		return;
 816	}
 817
 818	suspend_start = suspend_clocksource->read(suspend_clocksource);
 819}
 820
 821/**
 822 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
 823 * @cs:		current clocksource from timekeeping
 824 * @cycle_now:	current cycles from timekeeping
 825 *
 826 * This function will calculate the suspend time from suspend timer.
 827 *
 828 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
 829 *
 830 * This function is called early in the resume process from timekeeping_resume(),
 831 * that means there is only one cpu, no processes are running and the interrupts
 832 * are disabled. It is therefore possible to stop the suspend timer without
 833 * taking the clocksource mutex.
 834 */
 835u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
 836{
 837	u64 now, delta, nsec = 0;
 838
 839	if (!suspend_clocksource)
 840		return 0;
 841
 842	/*
 843	 * If current clocksource is the suspend timer, we should use the
 844	 * tkr_mono.cycle_last value from timekeeping as current cycle to
 845	 * avoid same reading from suspend timer.
 846	 */
 847	if (clocksource_is_suspend(cs))
 848		now = cycle_now;
 849	else
 850		now = suspend_clocksource->read(suspend_clocksource);
 851
 852	if (now > suspend_start) {
 853		delta = clocksource_delta(now, suspend_start,
 854					  suspend_clocksource->mask);
 855		nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult,
 856				       suspend_clocksource->shift);
 857	}
 858
 859	/*
 860	 * Disable the suspend timer to save power if current clocksource is
 861	 * not the suspend timer.
 862	 */
 863	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
 864		suspend_clocksource->disable(suspend_clocksource);
 865
 866	return nsec;
 867}
 868
 869/**
 870 * clocksource_suspend - suspend the clocksource(s)
 871 */
 872void clocksource_suspend(void)
 873{
 874	struct clocksource *cs;
 875
 876	list_for_each_entry_reverse(cs, &clocksource_list, list)
 877		if (cs->suspend)
 878			cs->suspend(cs);
 879}
 880
 881/**
 882 * clocksource_resume - resume the clocksource(s)
 883 */
 884void clocksource_resume(void)
 885{
 886	struct clocksource *cs;
 887
 888	list_for_each_entry(cs, &clocksource_list, list)
 889		if (cs->resume)
 890			cs->resume(cs);
 891
 892	clocksource_resume_watchdog();
 893}
 894
 895/**
 896 * clocksource_touch_watchdog - Update watchdog
 897 *
 898 * Update the watchdog after exception contexts such as kgdb so as not
 899 * to incorrectly trip the watchdog. This might fail when the kernel
 900 * was stopped in code which holds watchdog_lock.
 901 */
 902void clocksource_touch_watchdog(void)
 903{
 904	clocksource_resume_watchdog();
 905}
 906
 907/**
 908 * clocksource_max_adjustment- Returns max adjustment amount
 909 * @cs:         Pointer to clocksource
 910 *
 911 */
 912static u32 clocksource_max_adjustment(struct clocksource *cs)
 913{
 914	u64 ret;
 915	/*
 916	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
 917	 */
 918	ret = (u64)cs->mult * 11;
 919	do_div(ret,100);
 920	return (u32)ret;
 921}
 922
 923/**
 924 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
 925 * @mult:	cycle to nanosecond multiplier
 926 * @shift:	cycle to nanosecond divisor (power of two)
 927 * @maxadj:	maximum adjustment value to mult (~11%)
 928 * @mask:	bitmask for two's complement subtraction of non 64 bit counters
 929 * @max_cyc:	maximum cycle value before potential overflow (does not include
 930 *		any safety margin)
 931 *
 932 * NOTE: This function includes a safety margin of 50%, in other words, we
 933 * return half the number of nanoseconds the hardware counter can technically
 934 * cover. This is done so that we can potentially detect problems caused by
 935 * delayed timers or bad hardware, which might result in time intervals that
 936 * are larger than what the math used can handle without overflows.
 937 */
 938u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
 939{
 940	u64 max_nsecs, max_cycles;
 941
 942	/*
 943	 * Calculate the maximum number of cycles that we can pass to the
 944	 * cyc2ns() function without overflowing a 64-bit result.
 945	 */
 946	max_cycles = ULLONG_MAX;
 947	do_div(max_cycles, mult+maxadj);
 948
 949	/*
 950	 * The actual maximum number of cycles we can defer the clocksource is
 951	 * determined by the minimum of max_cycles and mask.
 952	 * Note: Here we subtract the maxadj to make sure we don't sleep for
 953	 * too long if there's a large negative adjustment.
 954	 */
 955	max_cycles = min(max_cycles, mask);
 956	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
 957
 958	/* return the max_cycles value as well if requested */
 959	if (max_cyc)
 960		*max_cyc = max_cycles;
 961
 962	/* Return 50% of the actual maximum, so we can detect bad values */
 963	max_nsecs >>= 1;
 964
 965	return max_nsecs;
 966}
 967
 968/**
 969 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
 970 * @cs:         Pointer to clocksource to be updated
 971 *
 972 */
 973static inline void clocksource_update_max_deferment(struct clocksource *cs)
 974{
 975	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
 976						cs->maxadj, cs->mask,
 977						&cs->max_cycles);
 
 
 
 
 
 
 
 
 
 978}
 979
 980static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
 981{
 982	struct clocksource *cs;
 983
 984	if (!finished_booting || list_empty(&clocksource_list))
 985		return NULL;
 986
 987	/*
 988	 * We pick the clocksource with the highest rating. If oneshot
 989	 * mode is active, we pick the highres valid clocksource with
 990	 * the best rating.
 991	 */
 992	list_for_each_entry(cs, &clocksource_list, list) {
 993		if (skipcur && cs == curr_clocksource)
 994			continue;
 995		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
 996			continue;
 997		return cs;
 998	}
 999	return NULL;
1000}
1001
1002static void __clocksource_select(bool skipcur)
1003{
1004	bool oneshot = tick_oneshot_mode_active();
1005	struct clocksource *best, *cs;
1006
1007	/* Find the best suitable clocksource */
1008	best = clocksource_find_best(oneshot, skipcur);
1009	if (!best)
1010		return;
1011
1012	if (!strlen(override_name))
1013		goto found;
1014
1015	/* Check for the override clocksource. */
1016	list_for_each_entry(cs, &clocksource_list, list) {
1017		if (skipcur && cs == curr_clocksource)
1018			continue;
1019		if (strcmp(cs->name, override_name) != 0)
1020			continue;
1021		/*
1022		 * Check to make sure we don't switch to a non-highres
1023		 * capable clocksource if the tick code is in oneshot
1024		 * mode (highres or nohz)
1025		 */
1026		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1027			/* Override clocksource cannot be used. */
1028			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1029				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1030					cs->name);
1031				override_name[0] = 0;
1032			} else {
1033				/*
1034				 * The override cannot be currently verified.
1035				 * Deferring to let the watchdog check.
1036				 */
1037				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1038					cs->name);
1039			}
1040		} else
1041			/* Override clocksource can be used. */
1042			best = cs;
1043		break;
1044	}
1045
1046found:
1047	if (curr_clocksource != best && !timekeeping_notify(best)) {
1048		pr_info("Switched to clocksource %s\n", best->name);
1049		curr_clocksource = best;
1050	}
1051}
1052
1053/**
1054 * clocksource_select - Select the best clocksource available
1055 *
1056 * Private function. Must hold clocksource_mutex when called.
1057 *
1058 * Select the clocksource with the best rating, or the clocksource,
1059 * which is selected by userspace override.
1060 */
1061static void clocksource_select(void)
1062{
1063	__clocksource_select(false);
1064}
1065
1066static void clocksource_select_fallback(void)
1067{
1068	__clocksource_select(true);
1069}
1070
1071/*
1072 * clocksource_done_booting - Called near the end of core bootup
1073 *
1074 * Hack to avoid lots of clocksource churn at boot time.
1075 * We use fs_initcall because we want this to start before
1076 * device_initcall but after subsys_initcall.
1077 */
1078static int __init clocksource_done_booting(void)
1079{
1080	mutex_lock(&clocksource_mutex);
1081	curr_clocksource = clocksource_default_clock();
1082	finished_booting = 1;
1083	/*
1084	 * Run the watchdog first to eliminate unstable clock sources
1085	 */
1086	__clocksource_watchdog_kthread();
1087	clocksource_select();
1088	mutex_unlock(&clocksource_mutex);
1089	return 0;
1090}
1091fs_initcall(clocksource_done_booting);
1092
1093/*
1094 * Enqueue the clocksource sorted by rating
1095 */
1096static void clocksource_enqueue(struct clocksource *cs)
1097{
1098	struct list_head *entry = &clocksource_list;
1099	struct clocksource *tmp;
1100
1101	list_for_each_entry(tmp, &clocksource_list, list) {
1102		/* Keep track of the place, where to insert */
1103		if (tmp->rating < cs->rating)
1104			break;
1105		entry = &tmp->list;
1106	}
1107	list_add(&cs->list, entry);
1108}
1109
1110/**
1111 * __clocksource_update_freq_scale - Used update clocksource with new freq
1112 * @cs:		clocksource to be registered
1113 * @scale:	Scale factor multiplied against freq to get clocksource hz
1114 * @freq:	clocksource frequency (cycles per second) divided by scale
1115 *
1116 * This should only be called from the clocksource->enable() method.
1117 *
1118 * This *SHOULD NOT* be called directly! Please use the
1119 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1120 * functions.
1121 */
1122void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1123{
1124	u64 sec;
1125
1126	/*
1127	 * Default clocksources are *special* and self-define their mult/shift.
1128	 * But, you're not special, so you should specify a freq value.
1129	 */
1130	if (freq) {
1131		/*
1132		 * Calc the maximum number of seconds which we can run before
1133		 * wrapping around. For clocksources which have a mask > 32-bit
1134		 * we need to limit the max sleep time to have a good
1135		 * conversion precision. 10 minutes is still a reasonable
1136		 * amount. That results in a shift value of 24 for a
1137		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1138		 * ~ 0.06ppm granularity for NTP.
1139		 */
1140		sec = cs->mask;
1141		do_div(sec, freq);
1142		do_div(sec, scale);
1143		if (!sec)
1144			sec = 1;
1145		else if (sec > 600 && cs->mask > UINT_MAX)
1146			sec = 600;
1147
1148		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1149				       NSEC_PER_SEC / scale, sec * scale);
1150	}
1151
1152	/*
1153	 * If the uncertainty margin is not specified, calculate it.
1154	 * If both scale and freq are non-zero, calculate the clock
1155	 * period, but bound below at 2*WATCHDOG_MAX_SKEW.  However,
1156	 * if either of scale or freq is zero, be very conservative and
1157	 * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the
1158	 * uncertainty margin.  Allow stupidly small uncertainty margins
1159	 * to be specified by the caller for testing purposes, but warn
1160	 * to discourage production use of this capability.
 
 
 
 
 
1161	 */
1162	if (scale && freq && !cs->uncertainty_margin) {
1163		cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1164		if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1165			cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1166	} else if (!cs->uncertainty_margin) {
1167		cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1168	}
1169	WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1170
1171	/*
1172	 * Ensure clocksources that have large 'mult' values don't overflow
1173	 * when adjusted.
1174	 */
1175	cs->maxadj = clocksource_max_adjustment(cs);
1176	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1177		|| (cs->mult - cs->maxadj > cs->mult))) {
1178		cs->mult >>= 1;
1179		cs->shift--;
1180		cs->maxadj = clocksource_max_adjustment(cs);
1181	}
1182
1183	/*
1184	 * Only warn for *special* clocksources that self-define
1185	 * their mult/shift values and don't specify a freq.
1186	 */
1187	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1188		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1189		cs->name);
1190
1191	clocksource_update_max_deferment(cs);
1192
1193	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1194		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1195}
1196EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1197
1198/**
1199 * __clocksource_register_scale - Used to install new clocksources
1200 * @cs:		clocksource to be registered
1201 * @scale:	Scale factor multiplied against freq to get clocksource hz
1202 * @freq:	clocksource frequency (cycles per second) divided by scale
1203 *
1204 * Returns -EBUSY if registration fails, zero otherwise.
1205 *
1206 * This *SHOULD NOT* be called directly! Please use the
1207 * clocksource_register_hz() or clocksource_register_khz helper functions.
1208 */
1209int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1210{
1211	unsigned long flags;
1212
1213	clocksource_arch_init(cs);
1214
1215	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1216		cs->id = CSID_GENERIC;
1217	if (cs->vdso_clock_mode < 0 ||
1218	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1219		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1220			cs->name, cs->vdso_clock_mode);
1221		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1222	}
1223
1224	/* Initialize mult/shift and max_idle_ns */
1225	__clocksource_update_freq_scale(cs, scale, freq);
1226
1227	/* Add clocksource to the clocksource list */
1228	mutex_lock(&clocksource_mutex);
1229
1230	clocksource_watchdog_lock(&flags);
1231	clocksource_enqueue(cs);
1232	clocksource_enqueue_watchdog(cs);
1233	clocksource_watchdog_unlock(&flags);
1234
1235	clocksource_select();
1236	clocksource_select_watchdog(false);
1237	__clocksource_suspend_select(cs);
1238	mutex_unlock(&clocksource_mutex);
1239	return 0;
1240}
1241EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1242
1243static void __clocksource_change_rating(struct clocksource *cs, int rating)
1244{
1245	list_del(&cs->list);
1246	cs->rating = rating;
1247	clocksource_enqueue(cs);
1248}
1249
1250/**
1251 * clocksource_change_rating - Change the rating of a registered clocksource
1252 * @cs:		clocksource to be changed
1253 * @rating:	new rating
1254 */
1255void clocksource_change_rating(struct clocksource *cs, int rating)
1256{
1257	unsigned long flags;
1258
1259	mutex_lock(&clocksource_mutex);
1260	clocksource_watchdog_lock(&flags);
1261	__clocksource_change_rating(cs, rating);
1262	clocksource_watchdog_unlock(&flags);
1263
1264	clocksource_select();
1265	clocksource_select_watchdog(false);
1266	clocksource_suspend_select(false);
1267	mutex_unlock(&clocksource_mutex);
1268}
1269EXPORT_SYMBOL(clocksource_change_rating);
1270
1271/*
1272 * Unbind clocksource @cs. Called with clocksource_mutex held
1273 */
1274static int clocksource_unbind(struct clocksource *cs)
1275{
1276	unsigned long flags;
1277
1278	if (clocksource_is_watchdog(cs)) {
1279		/* Select and try to install a replacement watchdog. */
1280		clocksource_select_watchdog(true);
1281		if (clocksource_is_watchdog(cs))
1282			return -EBUSY;
1283	}
1284
1285	if (cs == curr_clocksource) {
1286		/* Select and try to install a replacement clock source */
1287		clocksource_select_fallback();
1288		if (curr_clocksource == cs)
1289			return -EBUSY;
1290	}
1291
1292	if (clocksource_is_suspend(cs)) {
1293		/*
1294		 * Select and try to install a replacement suspend clocksource.
1295		 * If no replacement suspend clocksource, we will just let the
1296		 * clocksource go and have no suspend clocksource.
1297		 */
1298		clocksource_suspend_select(true);
1299	}
1300
1301	clocksource_watchdog_lock(&flags);
1302	clocksource_dequeue_watchdog(cs);
1303	list_del_init(&cs->list);
1304	clocksource_watchdog_unlock(&flags);
1305
1306	return 0;
1307}
1308
1309/**
1310 * clocksource_unregister - remove a registered clocksource
1311 * @cs:	clocksource to be unregistered
1312 */
1313int clocksource_unregister(struct clocksource *cs)
1314{
1315	int ret = 0;
1316
1317	mutex_lock(&clocksource_mutex);
1318	if (!list_empty(&cs->list))
1319		ret = clocksource_unbind(cs);
1320	mutex_unlock(&clocksource_mutex);
1321	return ret;
1322}
1323EXPORT_SYMBOL(clocksource_unregister);
1324
1325#ifdef CONFIG_SYSFS
1326/**
1327 * current_clocksource_show - sysfs interface for current clocksource
1328 * @dev:	unused
1329 * @attr:	unused
1330 * @buf:	char buffer to be filled with clocksource list
1331 *
1332 * Provides sysfs interface for listing current clocksource.
1333 */
1334static ssize_t current_clocksource_show(struct device *dev,
1335					struct device_attribute *attr,
1336					char *buf)
1337{
1338	ssize_t count = 0;
1339
1340	mutex_lock(&clocksource_mutex);
1341	count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
1342	mutex_unlock(&clocksource_mutex);
1343
1344	return count;
1345}
1346
1347ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1348{
1349	size_t ret = cnt;
1350
1351	/* strings from sysfs write are not 0 terminated! */
1352	if (!cnt || cnt >= CS_NAME_LEN)
1353		return -EINVAL;
1354
1355	/* strip of \n: */
1356	if (buf[cnt-1] == '\n')
1357		cnt--;
1358	if (cnt > 0)
1359		memcpy(dst, buf, cnt);
1360	dst[cnt] = 0;
1361	return ret;
1362}
1363
1364/**
1365 * current_clocksource_store - interface for manually overriding clocksource
1366 * @dev:	unused
1367 * @attr:	unused
1368 * @buf:	name of override clocksource
1369 * @count:	length of buffer
1370 *
1371 * Takes input from sysfs interface for manually overriding the default
1372 * clocksource selection.
1373 */
1374static ssize_t current_clocksource_store(struct device *dev,
1375					 struct device_attribute *attr,
1376					 const char *buf, size_t count)
1377{
1378	ssize_t ret;
1379
1380	mutex_lock(&clocksource_mutex);
1381
1382	ret = sysfs_get_uname(buf, override_name, count);
1383	if (ret >= 0)
1384		clocksource_select();
1385
1386	mutex_unlock(&clocksource_mutex);
1387
1388	return ret;
1389}
1390static DEVICE_ATTR_RW(current_clocksource);
1391
1392/**
1393 * unbind_clocksource_store - interface for manually unbinding clocksource
1394 * @dev:	unused
1395 * @attr:	unused
1396 * @buf:	unused
1397 * @count:	length of buffer
1398 *
1399 * Takes input from sysfs interface for manually unbinding a clocksource.
1400 */
1401static ssize_t unbind_clocksource_store(struct device *dev,
1402					struct device_attribute *attr,
1403					const char *buf, size_t count)
1404{
1405	struct clocksource *cs;
1406	char name[CS_NAME_LEN];
1407	ssize_t ret;
1408
1409	ret = sysfs_get_uname(buf, name, count);
1410	if (ret < 0)
1411		return ret;
1412
1413	ret = -ENODEV;
1414	mutex_lock(&clocksource_mutex);
1415	list_for_each_entry(cs, &clocksource_list, list) {
1416		if (strcmp(cs->name, name))
1417			continue;
1418		ret = clocksource_unbind(cs);
1419		break;
1420	}
1421	mutex_unlock(&clocksource_mutex);
1422
1423	return ret ? ret : count;
1424}
1425static DEVICE_ATTR_WO(unbind_clocksource);
1426
1427/**
1428 * available_clocksource_show - sysfs interface for listing clocksource
1429 * @dev:	unused
1430 * @attr:	unused
1431 * @buf:	char buffer to be filled with clocksource list
1432 *
1433 * Provides sysfs interface for listing registered clocksources
1434 */
1435static ssize_t available_clocksource_show(struct device *dev,
1436					  struct device_attribute *attr,
1437					  char *buf)
1438{
1439	struct clocksource *src;
1440	ssize_t count = 0;
1441
1442	mutex_lock(&clocksource_mutex);
1443	list_for_each_entry(src, &clocksource_list, list) {
1444		/*
1445		 * Don't show non-HRES clocksource if the tick code is
1446		 * in one shot mode (highres=on or nohz=on)
1447		 */
1448		if (!tick_oneshot_mode_active() ||
1449		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1450			count += snprintf(buf + count,
1451				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1452				  "%s ", src->name);
1453	}
1454	mutex_unlock(&clocksource_mutex);
1455
1456	count += snprintf(buf + count,
1457			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1458
1459	return count;
1460}
1461static DEVICE_ATTR_RO(available_clocksource);
1462
1463static struct attribute *clocksource_attrs[] = {
1464	&dev_attr_current_clocksource.attr,
1465	&dev_attr_unbind_clocksource.attr,
1466	&dev_attr_available_clocksource.attr,
1467	NULL
1468};
1469ATTRIBUTE_GROUPS(clocksource);
1470
1471static struct bus_type clocksource_subsys = {
1472	.name = "clocksource",
1473	.dev_name = "clocksource",
1474};
1475
1476static struct device device_clocksource = {
1477	.id	= 0,
1478	.bus	= &clocksource_subsys,
1479	.groups	= clocksource_groups,
1480};
1481
1482static int __init init_clocksource_sysfs(void)
1483{
1484	int error = subsys_system_register(&clocksource_subsys, NULL);
1485
1486	if (!error)
1487		error = device_register(&device_clocksource);
1488
1489	return error;
1490}
1491
1492device_initcall(init_clocksource_sysfs);
1493#endif /* CONFIG_SYSFS */
1494
1495/**
1496 * boot_override_clocksource - boot clock override
1497 * @str:	override name
1498 *
1499 * Takes a clocksource= boot argument and uses it
1500 * as the clocksource override name.
1501 */
1502static int __init boot_override_clocksource(char* str)
1503{
1504	mutex_lock(&clocksource_mutex);
1505	if (str)
1506		strscpy(override_name, str, sizeof(override_name));
1507	mutex_unlock(&clocksource_mutex);
1508	return 1;
1509}
1510
1511__setup("clocksource=", boot_override_clocksource);
1512
1513/**
1514 * boot_override_clock - Compatibility layer for deprecated boot option
1515 * @str:	override name
1516 *
1517 * DEPRECATED! Takes a clock= boot argument and uses it
1518 * as the clocksource override name
1519 */
1520static int __init boot_override_clock(char* str)
1521{
1522	if (!strcmp(str, "pmtmr")) {
1523		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1524		return boot_override_clocksource("acpi_pm");
1525	}
1526	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1527	return boot_override_clocksource(str);
1528}
1529
1530__setup("clock=", boot_override_clock);