clocksource.c

   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
  23static void clocksource_enqueue(struct clocksource *cs);
  24
  25static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end)
  26{
  27	u64 delta = clocksource_delta(end, start, cs->mask, cs->max_raw_delta);
  28
  29	if (likely(delta < cs->max_cycles))
  30		return clocksource_cyc2ns(delta, cs->mult, cs->shift);
  31
  32	return mul_u64_u32_shr(delta, cs->mult, cs->shift);
  33}
  34
  35/**
  36 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
  37 * @mult:	pointer to mult variable
  38 * @shift:	pointer to shift variable
  39 * @from:	frequency to convert from
  40 * @to:		frequency to convert to
  41 * @maxsec:	guaranteed runtime conversion range in seconds
  42 *
  43 * The function evaluates the shift/mult pair for the scaled math
  44 * operations of clocksources and clockevents.
  45 *
  46 * @to and @from are frequency values in HZ. For clock sources @to is
  47 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
  48 * event @to is the counter frequency and @from is NSEC_PER_SEC.
  49 *
  50 * The @maxsec conversion range argument controls the time frame in
  51 * seconds which must be covered by the runtime conversion with the
  52 * calculated mult and shift factors. This guarantees that no 64bit
  53 * overflow happens when the input value of the conversion is
  54 * multiplied with the calculated mult factor. Larger ranges may
  55 * reduce the conversion accuracy by choosing smaller mult and shift
  56 * factors.
  57 */
  58void
  59clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
  60{
  61	u64 tmp;
  62	u32 sft, sftacc= 32;
  63
  64	/*
  65	 * Calculate the shift factor which is limiting the conversion
  66	 * range:
  67	 */
  68	tmp = ((u64)maxsec * from) >> 32;
  69	while (tmp) {
  70		tmp >>=1;
  71		sftacc--;
  72	}
  73
  74	/*
  75	 * Find the conversion shift/mult pair which has the best
  76	 * accuracy and fits the maxsec conversion range:
  77	 */
  78	for (sft = 32; sft > 0; sft--) {
  79		tmp = (u64) to << sft;
  80		tmp += from / 2;
  81		do_div(tmp, from);
  82		if ((tmp >> sftacc) == 0)
  83			break;
  84	}
  85	*mult = tmp;
  86	*shift = sft;
  87}
  88EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
  89
  90/*[Clocksource internal variables]---------
  91 * curr_clocksource:
  92 *	currently selected clocksource.
  93 * suspend_clocksource:
  94 *	used to calculate the suspend time.
  95 * clocksource_list:
  96 *	linked list with the registered clocksources
  97 * clocksource_mutex:
  98 *	protects manipulations to curr_clocksource and the clocksource_list
  99 * override_name:
 100 *	Name of the user-specified clocksource.
 101 */
 102static struct clocksource *curr_clocksource;
 103static struct clocksource *suspend_clocksource;
 104static LIST_HEAD(clocksource_list);
 105static DEFINE_MUTEX(clocksource_mutex);
 106static char override_name[CS_NAME_LEN];
 107static int finished_booting;
 108static u64 suspend_start;
 109
 110/*
 111 * Interval: 0.5sec.
 112 */
 113#define WATCHDOG_INTERVAL (HZ >> 1)
 114#define WATCHDOG_INTERVAL_MAX_NS ((2 * WATCHDOG_INTERVAL) * (NSEC_PER_SEC / HZ))
 115
 116/*
 117 * Threshold: 0.0312s, when doubled: 0.0625s.
 118 */
 119#define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5)
 120
 121/*
 122 * Maximum permissible delay between two readouts of the watchdog
 123 * clocksource surrounding a read of the clocksource being validated.
 124 * This delay could be due to SMIs, NMIs, or to VCPU preemptions.  Used as
 125 * a lower bound for cs->uncertainty_margin values when registering clocks.
 126 *
 127 * The default of 500 parts per million is based on NTP's limits.
 128 * If a clocksource is good enough for NTP, it is good enough for us!
 129 *
 130 * In other words, by default, even if a clocksource is extremely
 131 * precise (for example, with a sub-nanosecond period), the maximum
 132 * permissible skew between the clocksource watchdog and the clocksource
 133 * under test is not permitted to go below the 500ppm minimum defined
 134 * by MAX_SKEW_USEC.  This 500ppm minimum may be overridden using the
 135 * CLOCKSOURCE_WATCHDOG_MAX_SKEW_US Kconfig option.
 136 */
 137#ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
 138#define MAX_SKEW_USEC	CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
 139#else
 140#define MAX_SKEW_USEC	(125 * WATCHDOG_INTERVAL / HZ)
 141#endif
 142
 143/*
 144 * Default for maximum permissible skew when cs->uncertainty_margin is
 145 * not specified, and the lower bound even when cs->uncertainty_margin
 146 * is specified.  This is also the default that is used when registering
 147 * clocks with unspecified cs->uncertainty_margin, so this macro is used
 148 * even in CONFIG_CLOCKSOURCE_WATCHDOG=n kernels.
 149 */
 150#define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC)
 151
 152#ifdef CONFIG_CLOCKSOURCE_WATCHDOG
 153static void clocksource_watchdog_work(struct work_struct *work);
 154static void clocksource_select(void);
 155
 156static LIST_HEAD(watchdog_list);
 157static struct clocksource *watchdog;
 158static struct timer_list watchdog_timer;
 159static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
 160static DEFINE_SPINLOCK(watchdog_lock);
 161static int watchdog_running;
 162static atomic_t watchdog_reset_pending;
 163static int64_t watchdog_max_interval;
 164
 165static inline void clocksource_watchdog_lock(unsigned long *flags)
 166{
 167	spin_lock_irqsave(&watchdog_lock, *flags);
 168}
 169
 170static inline void clocksource_watchdog_unlock(unsigned long *flags)
 171{
 172	spin_unlock_irqrestore(&watchdog_lock, *flags);
 173}
 174
 175static int clocksource_watchdog_kthread(void *data);
 176
 177static void clocksource_watchdog_work(struct work_struct *work)
 178{
 179	/*
 180	 * We cannot directly run clocksource_watchdog_kthread() here, because
 181	 * clocksource_select() calls timekeeping_notify() which uses
 182	 * stop_machine(). One cannot use stop_machine() from a workqueue() due
 183	 * lock inversions wrt CPU hotplug.
 184	 *
 185	 * Also, we only ever run this work once or twice during the lifetime
 186	 * of the kernel, so there is no point in creating a more permanent
 187	 * kthread for this.
 188	 *
 189	 * If kthread_run fails the next watchdog scan over the
 190	 * watchdog_list will find the unstable clock again.
 191	 */
 192	kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
 193}
 194
 195static void clocksource_change_rating(struct clocksource *cs, int rating)
 196{
 197	list_del(&cs->list);
 198	cs->rating = rating;
 199	clocksource_enqueue(cs);
 200}
 201
 202static void __clocksource_unstable(struct clocksource *cs)
 203{
 204	cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
 205	cs->flags |= CLOCK_SOURCE_UNSTABLE;
 206
 207	/*
 208	 * If the clocksource is registered clocksource_watchdog_kthread() will
 209	 * re-rate and re-select.
 210	 */
 211	if (list_empty(&cs->list)) {
 212		cs->rating = 0;
 213		return;
 214	}
 215
 216	if (cs->mark_unstable)
 217		cs->mark_unstable(cs);
 218
 219	/* kick clocksource_watchdog_kthread() */
 220	if (finished_booting)
 221		schedule_work(&watchdog_work);
 222}
 223
 224/**
 225 * clocksource_mark_unstable - mark clocksource unstable via watchdog
 226 * @cs:		clocksource to be marked unstable
 227 *
 228 * This function is called by the x86 TSC code to mark clocksources as unstable;
 229 * it defers demotion and re-selection to a kthread.
 230 */
 231void clocksource_mark_unstable(struct clocksource *cs)
 232{
 233	unsigned long flags;
 234
 235	spin_lock_irqsave(&watchdog_lock, flags);
 236	if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
 237		if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
 238			list_add(&cs->wd_list, &watchdog_list);
 239		__clocksource_unstable(cs);
 240	}
 241	spin_unlock_irqrestore(&watchdog_lock, flags);
 242}
 243
 244static int verify_n_cpus = 8;
 245module_param(verify_n_cpus, int, 0644);
 246
 247enum wd_read_status {
 248	WD_READ_SUCCESS,
 249	WD_READ_UNSTABLE,
 250	WD_READ_SKIP
 251};
 252
 253static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow)
 254{
 255	int64_t md = watchdog->uncertainty_margin;
 256	unsigned int nretries, max_retries;
 257	int64_t wd_delay, wd_seq_delay;
 258	u64 wd_end, wd_end2;
 259
 260	max_retries = clocksource_get_max_watchdog_retry();
 261	for (nretries = 0; nretries <= max_retries; nretries++) {
 262		local_irq_disable();
 263		*wdnow = watchdog->read(watchdog);
 264		*csnow = cs->read(cs);
 265		wd_end = watchdog->read(watchdog);
 266		wd_end2 = watchdog->read(watchdog);
 267		local_irq_enable();
 268
 269		wd_delay = cycles_to_nsec_safe(watchdog, *wdnow, wd_end);
 270		if (wd_delay <= md + cs->uncertainty_margin) {
 271			if (nretries > 1 && nretries >= max_retries) {
 272				pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n",
 273					smp_processor_id(), watchdog->name, nretries);
 274			}
 275			return WD_READ_SUCCESS;
 276		}
 277
 278		/*
 279		 * Now compute delay in consecutive watchdog read to see if
 280		 * there is too much external interferences that cause
 281		 * significant delay in reading both clocksource and watchdog.
 282		 *
 283		 * If consecutive WD read-back delay > md, report
 284		 * system busy, reinit the watchdog and skip the current
 285		 * watchdog test.
 286		 */
 287		wd_seq_delay = cycles_to_nsec_safe(watchdog, wd_end, wd_end2);
 288		if (wd_seq_delay > md)
 289			goto skip_test;
 290	}
 291
 292	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",
 293		smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name);
 294	return WD_READ_UNSTABLE;
 295
 296skip_test:
 297	pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n",
 298		smp_processor_id(), watchdog->name, wd_seq_delay);
 299	pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n",
 300		cs->name, wd_delay);
 301	return WD_READ_SKIP;
 302}
 303
 304static u64 csnow_mid;
 305static cpumask_t cpus_ahead;
 306static cpumask_t cpus_behind;
 307static cpumask_t cpus_chosen;
 308
 309static void clocksource_verify_choose_cpus(void)
 310{
 311	int cpu, i, n = verify_n_cpus;
 312
 313	if (n < 0 || n >= num_online_cpus()) {
 314		/* Check all of the CPUs. */
 315		cpumask_copy(&cpus_chosen, cpu_online_mask);
 316		cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
 317		return;
 318	}
 319
 320	/* If no checking desired, or no other CPU to check, leave. */
 321	cpumask_clear(&cpus_chosen);
 322	if (n == 0 || num_online_cpus() <= 1)
 323		return;
 324
 325	/* Make sure to select at least one CPU other than the current CPU. */
 326	cpu = cpumask_any_but(cpu_online_mask, smp_processor_id());
 327	if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
 328		return;
 329	cpumask_set_cpu(cpu, &cpus_chosen);
 330
 331	/* Force a sane value for the boot parameter. */
 332	if (n > nr_cpu_ids)
 333		n = nr_cpu_ids;
 334
 335	/*
 336	 * Randomly select the specified number of CPUs.  If the same
 337	 * CPU is selected multiple times, that CPU is checked only once,
 338	 * and no replacement CPU is selected.  This gracefully handles
 339	 * situations where verify_n_cpus is greater than the number of
 340	 * CPUs that are currently online.
 341	 */
 342	for (i = 1; i < n; i++) {
 343		cpu = cpumask_random(cpu_online_mask);
 344		if (!WARN_ON_ONCE(cpu >= nr_cpu_ids))
 345			cpumask_set_cpu(cpu, &cpus_chosen);
 346	}
 347
 348	/* Don't verify ourselves. */
 349	cpumask_clear_cpu(smp_processor_id(), &cpus_chosen);
 350}
 351
 352static void clocksource_verify_one_cpu(void *csin)
 353{
 354	struct clocksource *cs = (struct clocksource *)csin;
 355
 356	csnow_mid = cs->read(cs);
 357}
 358
 359void clocksource_verify_percpu(struct clocksource *cs)
 360{
 361	int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX;
 362	u64 csnow_begin, csnow_end;
 363	int cpu, testcpu;
 364	s64 delta;
 365
 366	if (verify_n_cpus == 0)
 367		return;
 368	cpumask_clear(&cpus_ahead);
 369	cpumask_clear(&cpus_behind);
 370	cpus_read_lock();
 371	migrate_disable();
 372	clocksource_verify_choose_cpus();
 373	if (cpumask_empty(&cpus_chosen)) {
 374		migrate_enable();
 375		cpus_read_unlock();
 376		pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name);
 377		return;
 378	}
 379	testcpu = smp_processor_id();
 380	pr_info("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n",
 381		cs->name, testcpu, cpumask_pr_args(&cpus_chosen));
 382	preempt_disable();
 383	for_each_cpu(cpu, &cpus_chosen) {
 384		if (cpu == testcpu)
 385			continue;
 386		csnow_begin = cs->read(cs);
 387		smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1);
 388		csnow_end = cs->read(cs);
 389		delta = (s64)((csnow_mid - csnow_begin) & cs->mask);
 390		if (delta < 0)
 391			cpumask_set_cpu(cpu, &cpus_behind);
 392		delta = (csnow_end - csnow_mid) & cs->mask;
 393		if (delta < 0)
 394			cpumask_set_cpu(cpu, &cpus_ahead);
 395		cs_nsec = cycles_to_nsec_safe(cs, csnow_begin, csnow_end);
 396		if (cs_nsec > cs_nsec_max)
 397			cs_nsec_max = cs_nsec;
 398		if (cs_nsec < cs_nsec_min)
 399			cs_nsec_min = cs_nsec;
 400	}
 401	preempt_enable();
 402	migrate_enable();
 403	cpus_read_unlock();
 404	if (!cpumask_empty(&cpus_ahead))
 405		pr_warn("        CPUs %*pbl ahead of CPU %d for clocksource %s.\n",
 406			cpumask_pr_args(&cpus_ahead), testcpu, cs->name);
 407	if (!cpumask_empty(&cpus_behind))
 408		pr_warn("        CPUs %*pbl behind CPU %d for clocksource %s.\n",
 409			cpumask_pr_args(&cpus_behind), testcpu, cs->name);
 410	pr_info("        CPU %d check durations %lldns - %lldns for clocksource %s.\n",
 411		testcpu, cs_nsec_min, cs_nsec_max, cs->name);
 412}
 413EXPORT_SYMBOL_GPL(clocksource_verify_percpu);
 414
 415static inline void clocksource_reset_watchdog(void)
 416{
 417	struct clocksource *cs;
 418
 419	list_for_each_entry(cs, &watchdog_list, wd_list)
 420		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
 421}
 422
 423
 424static void clocksource_watchdog(struct timer_list *unused)
 425{
 426	int64_t wd_nsec, cs_nsec, interval;
 427	u64 csnow, wdnow, cslast, wdlast;
 428	int next_cpu, reset_pending;
 429	struct clocksource *cs;
 430	enum wd_read_status read_ret;
 431	unsigned long extra_wait = 0;
 432	u32 md;
 433
 434	spin_lock(&watchdog_lock);
 435	if (!watchdog_running)
 436		goto out;
 437
 438	reset_pending = atomic_read(&watchdog_reset_pending);
 439
 440	list_for_each_entry(cs, &watchdog_list, wd_list) {
 441
 442		/* Clocksource already marked unstable? */
 443		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 444			if (finished_booting)
 445				schedule_work(&watchdog_work);
 446			continue;
 447		}
 448
 449		read_ret = cs_watchdog_read(cs, &csnow, &wdnow);
 450
 451		if (read_ret == WD_READ_UNSTABLE) {
 452			/* Clock readout unreliable, so give it up. */
 453			__clocksource_unstable(cs);
 454			continue;
 455		}
 456
 457		/*
 458		 * When WD_READ_SKIP is returned, it means the system is likely
 459		 * under very heavy load, where the latency of reading
 460		 * watchdog/clocksource is very big, and affect the accuracy of
 461		 * watchdog check. So give system some space and suspend the
 462		 * watchdog check for 5 minutes.
 463		 */
 464		if (read_ret == WD_READ_SKIP) {
 465			/*
 466			 * As the watchdog timer will be suspended, and
 467			 * cs->last could keep unchanged for 5 minutes, reset
 468			 * the counters.
 469			 */
 470			clocksource_reset_watchdog();
 471			extra_wait = HZ * 300;
 472			break;
 473		}
 474
 475		/* Clocksource initialized ? */
 476		if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
 477		    atomic_read(&watchdog_reset_pending)) {
 478			cs->flags |= CLOCK_SOURCE_WATCHDOG;
 479			cs->wd_last = wdnow;
 480			cs->cs_last = csnow;
 481			continue;
 482		}
 483
 484		wd_nsec = cycles_to_nsec_safe(watchdog, cs->wd_last, wdnow);
 485		cs_nsec = cycles_to_nsec_safe(cs, cs->cs_last, csnow);
 486		wdlast = cs->wd_last; /* save these in case we print them */
 487		cslast = cs->cs_last;
 488		cs->cs_last = csnow;
 489		cs->wd_last = wdnow;
 490
 491		if (atomic_read(&watchdog_reset_pending))
 492			continue;
 493
 494		/*
 495		 * The processing of timer softirqs can get delayed (usually
 496		 * on account of ksoftirqd not getting to run in a timely
 497		 * manner), which causes the watchdog interval to stretch.
 498		 * Skew detection may fail for longer watchdog intervals
 499		 * on account of fixed margins being used.
 500		 * Some clocksources, e.g. acpi_pm, cannot tolerate
 501		 * watchdog intervals longer than a few seconds.
 502		 */
 503		interval = max(cs_nsec, wd_nsec);
 504		if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) {
 505			if (system_state > SYSTEM_SCHEDULING &&
 506			    interval > 2 * watchdog_max_interval) {
 507				watchdog_max_interval = interval;
 508				pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n",
 509					cs_nsec, wd_nsec);
 510			}
 511			watchdog_timer.expires = jiffies;
 512			continue;
 513		}
 514
 515		/* Check the deviation from the watchdog clocksource. */
 516		md = cs->uncertainty_margin + watchdog->uncertainty_margin;
 517		if (abs(cs_nsec - wd_nsec) > md) {
 518			s64 cs_wd_msec;
 519			s64 wd_msec;
 520			u32 wd_rem;
 521
 522			pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
 523				smp_processor_id(), cs->name);
 524			pr_warn("                      '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n",
 525				watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask);
 526			pr_warn("                      '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n",
 527				cs->name, cs_nsec, csnow, cslast, cs->mask);
 528			cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem);
 529			wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem);
 530			pr_warn("                      Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n",
 531				cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec);
 532			if (curr_clocksource == cs)
 533				pr_warn("                      '%s' is current clocksource.\n", cs->name);
 534			else if (curr_clocksource)
 535				pr_warn("                      '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name);
 536			else
 537				pr_warn("                      No current clocksource.\n");
 538			__clocksource_unstable(cs);
 539			continue;
 540		}
 541
 542		if (cs == curr_clocksource && cs->tick_stable)
 543			cs->tick_stable(cs);
 544
 545		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
 546		    (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
 547		    (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
 548			/* Mark it valid for high-res. */
 549			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 550
 551			/*
 552			 * clocksource_done_booting() will sort it if
 553			 * finished_booting is not set yet.
 554			 */
 555			if (!finished_booting)
 556				continue;
 557
 558			/*
 559			 * If this is not the current clocksource let
 560			 * the watchdog thread reselect it. Due to the
 561			 * change to high res this clocksource might
 562			 * be preferred now. If it is the current
 563			 * clocksource let the tick code know about
 564			 * that change.
 565			 */
 566			if (cs != curr_clocksource) {
 567				cs->flags |= CLOCK_SOURCE_RESELECT;
 568				schedule_work(&watchdog_work);
 569			} else {
 570				tick_clock_notify();
 571			}
 572		}
 573	}
 574
 575	/*
 576	 * We only clear the watchdog_reset_pending, when we did a
 577	 * full cycle through all clocksources.
 578	 */
 579	if (reset_pending)
 580		atomic_dec(&watchdog_reset_pending);
 581
 582	/*
 583	 * Cycle through CPUs to check if the CPUs stay synchronized
 584	 * to each other.
 585	 */
 586	next_cpu = cpumask_next_wrap(raw_smp_processor_id(), cpu_online_mask);
 587
 588	/*
 589	 * Arm timer if not already pending: could race with concurrent
 590	 * pair clocksource_stop_watchdog() clocksource_start_watchdog().
 591	 */
 592	if (!timer_pending(&watchdog_timer)) {
 593		watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait;
 594		add_timer_on(&watchdog_timer, next_cpu);
 595	}
 596out:
 597	spin_unlock(&watchdog_lock);
 598}
 599
 600static inline void clocksource_start_watchdog(void)
 601{
 602	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
 603		return;
 604	timer_setup(&watchdog_timer, clocksource_watchdog, 0);
 605	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
 606	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
 607	watchdog_running = 1;
 608}
 609
 610static inline void clocksource_stop_watchdog(void)
 611{
 612	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
 613		return;
 614	timer_delete(&watchdog_timer);
 615	watchdog_running = 0;
 616}
 617
 618static void clocksource_resume_watchdog(void)
 619{
 620	atomic_inc(&watchdog_reset_pending);
 621}
 622
 623static void clocksource_enqueue_watchdog(struct clocksource *cs)
 624{
 625	INIT_LIST_HEAD(&cs->wd_list);
 626
 627	if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 628		/* cs is a clocksource to be watched. */
 629		list_add(&cs->wd_list, &watchdog_list);
 630		cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
 631	} else {
 632		/* cs is a watchdog. */
 633		if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 634			cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 635	}
 636}
 637
 638static void clocksource_select_watchdog(bool fallback)
 639{
 640	struct clocksource *cs, *old_wd;
 641	unsigned long flags;
 642
 643	spin_lock_irqsave(&watchdog_lock, flags);
 644	/* save current watchdog */
 645	old_wd = watchdog;
 646	if (fallback)
 647		watchdog = NULL;
 648
 649	list_for_each_entry(cs, &clocksource_list, list) {
 650		/* cs is a clocksource to be watched. */
 651		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
 652			continue;
 653
 654		/* Skip current if we were requested for a fallback. */
 655		if (fallback && cs == old_wd)
 656			continue;
 657
 658		/* Pick the best watchdog. */
 659		if (!watchdog || cs->rating > watchdog->rating)
 660			watchdog = cs;
 661	}
 662	/* If we failed to find a fallback restore the old one. */
 663	if (!watchdog)
 664		watchdog = old_wd;
 665
 666	/* If we changed the watchdog we need to reset cycles. */
 667	if (watchdog != old_wd)
 668		clocksource_reset_watchdog();
 669
 670	/* Check if the watchdog timer needs to be started. */
 671	clocksource_start_watchdog();
 672	spin_unlock_irqrestore(&watchdog_lock, flags);
 673}
 674
 675static void clocksource_dequeue_watchdog(struct clocksource *cs)
 676{
 677	if (cs != watchdog) {
 678		if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 679			/* cs is a watched clocksource. */
 680			list_del_init(&cs->wd_list);
 681			/* Check if the watchdog timer needs to be stopped. */
 682			clocksource_stop_watchdog();
 683		}
 684	}
 685}
 686
 687static int __clocksource_watchdog_kthread(void)
 688{
 689	struct clocksource *cs, *tmp;
 690	unsigned long flags;
 691	int select = 0;
 692
 693	/* Do any required per-CPU skew verification. */
 694	if (curr_clocksource &&
 695	    curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE &&
 696	    curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU)
 697		clocksource_verify_percpu(curr_clocksource);
 698
 699	spin_lock_irqsave(&watchdog_lock, flags);
 700	list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
 701		if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
 702			list_del_init(&cs->wd_list);
 703			clocksource_change_rating(cs, 0);
 704			select = 1;
 705		}
 706		if (cs->flags & CLOCK_SOURCE_RESELECT) {
 707			cs->flags &= ~CLOCK_SOURCE_RESELECT;
 708			select = 1;
 709		}
 710	}
 711	/* Check if the watchdog timer needs to be stopped. */
 712	clocksource_stop_watchdog();
 713	spin_unlock_irqrestore(&watchdog_lock, flags);
 714
 715	return select;
 716}
 717
 718static int clocksource_watchdog_kthread(void *data)
 719{
 720	mutex_lock(&clocksource_mutex);
 721	if (__clocksource_watchdog_kthread())
 722		clocksource_select();
 723	mutex_unlock(&clocksource_mutex);
 724	return 0;
 725}
 726
 727static bool clocksource_is_watchdog(struct clocksource *cs)
 728{
 729	return cs == watchdog;
 730}
 731
 732#else /* CONFIG_CLOCKSOURCE_WATCHDOG */
 733
 734static void clocksource_enqueue_watchdog(struct clocksource *cs)
 735{
 736	if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 737		cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 738}
 739
 740static void clocksource_select_watchdog(bool fallback) { }
 741static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
 742static inline void clocksource_resume_watchdog(void) { }
 743static inline int __clocksource_watchdog_kthread(void) { return 0; }
 744static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
 745void clocksource_mark_unstable(struct clocksource *cs) { }
 746
 747static inline void clocksource_watchdog_lock(unsigned long *flags) { }
 748static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
 749
 750#endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
 751
 752static bool clocksource_is_suspend(struct clocksource *cs)
 753{
 754	return cs == suspend_clocksource;
 755}
 756
 757static void __clocksource_suspend_select(struct clocksource *cs)
 758{
 759	/*
 760	 * Skip the clocksource which will be stopped in suspend state.
 761	 */
 762	if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP))
 763		return;
 764
 765	/*
 766	 * The nonstop clocksource can be selected as the suspend clocksource to
 767	 * calculate the suspend time, so it should not supply suspend/resume
 768	 * interfaces to suspend the nonstop clocksource when system suspends.
 769	 */
 770	if (cs->suspend || cs->resume) {
 771		pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n",
 772			cs->name);
 773	}
 774
 775	/* Pick the best rating. */
 776	if (!suspend_clocksource || cs->rating > suspend_clocksource->rating)
 777		suspend_clocksource = cs;
 778}
 779
 780/**
 781 * clocksource_suspend_select - Select the best clocksource for suspend timing
 782 * @fallback:	if select a fallback clocksource
 783 */
 784static void clocksource_suspend_select(bool fallback)
 785{
 786	struct clocksource *cs, *old_suspend;
 787
 788	old_suspend = suspend_clocksource;
 789	if (fallback)
 790		suspend_clocksource = NULL;
 791
 792	list_for_each_entry(cs, &clocksource_list, list) {
 793		/* Skip current if we were requested for a fallback. */
 794		if (fallback && cs == old_suspend)
 795			continue;
 796
 797		__clocksource_suspend_select(cs);
 798	}
 799}
 800
 801/**
 802 * clocksource_start_suspend_timing - Start measuring the suspend timing
 803 * @cs:			current clocksource from timekeeping
 804 * @start_cycles:	current cycles from timekeeping
 805 *
 806 * This function will save the start cycle values of suspend timer to calculate
 807 * the suspend time when resuming system.
 808 *
 809 * This function is called late in the suspend process from timekeeping_suspend(),
 810 * that means processes are frozen, non-boot cpus and interrupts are disabled
 811 * now. It is therefore possible to start the suspend timer without taking the
 812 * clocksource mutex.
 813 */
 814void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles)
 815{
 816	if (!suspend_clocksource)
 817		return;
 818
 819	/*
 820	 * If current clocksource is the suspend timer, we should use the
 821	 * tkr_mono.cycle_last value as suspend_start to avoid same reading
 822	 * from suspend timer.
 823	 */
 824	if (clocksource_is_suspend(cs)) {
 825		suspend_start = start_cycles;
 826		return;
 827	}
 828
 829	if (suspend_clocksource->enable &&
 830	    suspend_clocksource->enable(suspend_clocksource)) {
 831		pr_warn_once("Failed to enable the non-suspend-able clocksource.\n");
 832		return;
 833	}
 834
 835	suspend_start = suspend_clocksource->read(suspend_clocksource);
 836}
 837
 838/**
 839 * clocksource_stop_suspend_timing - Stop measuring the suspend timing
 840 * @cs:		current clocksource from timekeeping
 841 * @cycle_now:	current cycles from timekeeping
 842 *
 843 * This function will calculate the suspend time from suspend timer.
 844 *
 845 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource.
 846 *
 847 * This function is called early in the resume process from timekeeping_resume(),
 848 * that means there is only one cpu, no processes are running and the interrupts
 849 * are disabled. It is therefore possible to stop the suspend timer without
 850 * taking the clocksource mutex.
 851 */
 852u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now)
 853{
 854	u64 now, nsec = 0;
 855
 856	if (!suspend_clocksource)
 857		return 0;
 858
 859	/*
 860	 * If current clocksource is the suspend timer, we should use the
 861	 * tkr_mono.cycle_last value from timekeeping as current cycle to
 862	 * avoid same reading from suspend timer.
 863	 */
 864	if (clocksource_is_suspend(cs))
 865		now = cycle_now;
 866	else
 867		now = suspend_clocksource->read(suspend_clocksource);
 868
 869	if (now > suspend_start)
 870		nsec = cycles_to_nsec_safe(suspend_clocksource, suspend_start, now);
 871
 872	/*
 873	 * Disable the suspend timer to save power if current clocksource is
 874	 * not the suspend timer.
 875	 */
 876	if (!clocksource_is_suspend(cs) && suspend_clocksource->disable)
 877		suspend_clocksource->disable(suspend_clocksource);
 878
 879	return nsec;
 880}
 881
 882/**
 883 * clocksource_suspend - suspend the clocksource(s)
 884 */
 885void clocksource_suspend(void)
 886{
 887	struct clocksource *cs;
 888
 889	list_for_each_entry_reverse(cs, &clocksource_list, list)
 890		if (cs->suspend)
 891			cs->suspend(cs);
 892}
 893
 894/**
 895 * clocksource_resume - resume the clocksource(s)
 896 */
 897void clocksource_resume(void)
 898{
 899	struct clocksource *cs;
 900
 901	list_for_each_entry(cs, &clocksource_list, list)
 902		if (cs->resume)
 903			cs->resume(cs);
 904
 905	clocksource_resume_watchdog();
 906}
 907
 908/**
 909 * clocksource_touch_watchdog - Update watchdog
 910 *
 911 * Update the watchdog after exception contexts such as kgdb so as not
 912 * to incorrectly trip the watchdog. This might fail when the kernel
 913 * was stopped in code which holds watchdog_lock.
 914 */
 915void clocksource_touch_watchdog(void)
 916{
 917	clocksource_resume_watchdog();
 918}
 919
 920/**
 921 * clocksource_max_adjustment- Returns max adjustment amount
 922 * @cs:         Pointer to clocksource
 923 *
 924 */
 925static u32 clocksource_max_adjustment(struct clocksource *cs)
 926{
 927	u64 ret;
 928	/*
 929	 * We won't try to correct for more than 11% adjustments (110,000 ppm),
 930	 */
 931	ret = (u64)cs->mult * 11;
 932	do_div(ret,100);
 933	return (u32)ret;
 934}
 935
 936/**
 937 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
 938 * @mult:	cycle to nanosecond multiplier
 939 * @shift:	cycle to nanosecond divisor (power of two)
 940 * @maxadj:	maximum adjustment value to mult (~11%)
 941 * @mask:	bitmask for two's complement subtraction of non 64 bit counters
 942 * @max_cyc:	maximum cycle value before potential overflow (does not include
 943 *		any safety margin)
 944 *
 945 * NOTE: This function includes a safety margin of 50%, in other words, we
 946 * return half the number of nanoseconds the hardware counter can technically
 947 * cover. This is done so that we can potentially detect problems caused by
 948 * delayed timers or bad hardware, which might result in time intervals that
 949 * are larger than what the math used can handle without overflows.
 950 */
 951u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
 952{
 953	u64 max_nsecs, max_cycles;
 954
 955	/*
 956	 * Calculate the maximum number of cycles that we can pass to the
 957	 * cyc2ns() function without overflowing a 64-bit result.
 958	 */
 959	max_cycles = ULLONG_MAX;
 960	do_div(max_cycles, mult+maxadj);
 961
 962	/*
 963	 * The actual maximum number of cycles we can defer the clocksource is
 964	 * determined by the minimum of max_cycles and mask.
 965	 * Note: Here we subtract the maxadj to make sure we don't sleep for
 966	 * too long if there's a large negative adjustment.
 967	 */
 968	max_cycles = min(max_cycles, mask);
 969	max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
 970
 971	/* return the max_cycles value as well if requested */
 972	if (max_cyc)
 973		*max_cyc = max_cycles;
 974
 975	/* Return 50% of the actual maximum, so we can detect bad values */
 976	max_nsecs >>= 1;
 977
 978	return max_nsecs;
 979}
 980
 981/**
 982 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
 983 * @cs:         Pointer to clocksource to be updated
 984 *
 985 */
 986static inline void clocksource_update_max_deferment(struct clocksource *cs)
 987{
 988	cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
 989						cs->maxadj, cs->mask,
 990						&cs->max_cycles);
 991
 992	/*
 993	 * Threshold for detecting negative motion in clocksource_delta().
 994	 *
 995	 * Allow for 0.875 of the counter width so that overly long idle
 996	 * sleeps, which go slightly over mask/2, do not trigger the
 997	 * negative motion detection.
 998	 */
 999	cs->max_raw_delta = (cs->mask >> 1) + (cs->mask >> 2) + (cs->mask >> 3);
1000}
1001
1002static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
1003{
1004	struct clocksource *cs;
1005
1006	if (!finished_booting || list_empty(&clocksource_list))
1007		return NULL;
1008
1009	/*
1010	 * We pick the clocksource with the highest rating. If oneshot
1011	 * mode is active, we pick the highres valid clocksource with
1012	 * the best rating.
1013	 */
1014	list_for_each_entry(cs, &clocksource_list, list) {
1015		if (skipcur && cs == curr_clocksource)
1016			continue;
1017		if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1018			continue;
1019		return cs;
1020	}
1021	return NULL;
1022}
1023
1024static void __clocksource_select(bool skipcur)
1025{
1026	bool oneshot = tick_oneshot_mode_active();
1027	struct clocksource *best, *cs;
1028
1029	/* Find the best suitable clocksource */
1030	best = clocksource_find_best(oneshot, skipcur);
1031	if (!best)
1032		return;
1033
1034	if (!strlen(override_name))
1035		goto found;
1036
1037	/* Check for the override clocksource. */
1038	list_for_each_entry(cs, &clocksource_list, list) {
1039		if (skipcur && cs == curr_clocksource)
1040			continue;
1041		if (strcmp(cs->name, override_name) != 0)
1042			continue;
1043		/*
1044		 * Check to make sure we don't switch to a non-highres
1045		 * capable clocksource if the tick code is in oneshot
1046		 * mode (highres or nohz)
1047		 */
1048		if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
1049			/* Override clocksource cannot be used. */
1050			if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
1051				pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
1052					cs->name);
1053				override_name[0] = 0;
1054			} else {
1055				/*
1056				 * The override cannot be currently verified.
1057				 * Deferring to let the watchdog check.
1058				 */
1059				pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
1060					cs->name);
1061			}
1062		} else
1063			/* Override clocksource can be used. */
1064			best = cs;
1065		break;
1066	}
1067
1068found:
1069	if (curr_clocksource != best && !timekeeping_notify(best)) {
1070		pr_info("Switched to clocksource %s\n", best->name);
1071		curr_clocksource = best;
1072	}
1073}
1074
1075/**
1076 * clocksource_select - Select the best clocksource available
1077 *
1078 * Private function. Must hold clocksource_mutex when called.
1079 *
1080 * Select the clocksource with the best rating, or the clocksource,
1081 * which is selected by userspace override.
1082 */
1083static void clocksource_select(void)
1084{
1085	__clocksource_select(false);
1086}
1087
1088static void clocksource_select_fallback(void)
1089{
1090	__clocksource_select(true);
1091}
1092
1093/*
1094 * clocksource_done_booting - Called near the end of core bootup
1095 *
1096 * Hack to avoid lots of clocksource churn at boot time.
1097 * We use fs_initcall because we want this to start before
1098 * device_initcall but after subsys_initcall.
1099 */
1100static int __init clocksource_done_booting(void)
1101{
1102	mutex_lock(&clocksource_mutex);
1103	curr_clocksource = clocksource_default_clock();
1104	finished_booting = 1;
1105	/*
1106	 * Run the watchdog first to eliminate unstable clock sources
1107	 */
1108	__clocksource_watchdog_kthread();
1109	clocksource_select();
1110	mutex_unlock(&clocksource_mutex);
1111	return 0;
1112}
1113fs_initcall(clocksource_done_booting);
1114
1115/*
1116 * Enqueue the clocksource sorted by rating
1117 */
1118static void clocksource_enqueue(struct clocksource *cs)
1119{
1120	struct list_head *entry = &clocksource_list;
1121	struct clocksource *tmp;
1122
1123	list_for_each_entry(tmp, &clocksource_list, list) {
1124		/* Keep track of the place, where to insert */
1125		if (tmp->rating < cs->rating)
1126			break;
1127		entry = &tmp->list;
1128	}
1129	list_add(&cs->list, entry);
1130}
1131
1132/**
1133 * __clocksource_update_freq_scale - Used update clocksource with new freq
1134 * @cs:		clocksource to be registered
1135 * @scale:	Scale factor multiplied against freq to get clocksource hz
1136 * @freq:	clocksource frequency (cycles per second) divided by scale
1137 *
1138 * This should only be called from the clocksource->enable() method.
1139 *
1140 * This *SHOULD NOT* be called directly! Please use the
1141 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
1142 * functions.
1143 */
1144void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
1145{
1146	u64 sec;
1147
1148	/*
1149	 * Default clocksources are *special* and self-define their mult/shift.
1150	 * But, you're not special, so you should specify a freq value.
1151	 */
1152	if (freq) {
1153		/*
1154		 * Calc the maximum number of seconds which we can run before
1155		 * wrapping around. For clocksources which have a mask > 32-bit
1156		 * we need to limit the max sleep time to have a good
1157		 * conversion precision. 10 minutes is still a reasonable
1158		 * amount. That results in a shift value of 24 for a
1159		 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
1160		 * ~ 0.06ppm granularity for NTP.
1161		 */
1162		sec = cs->mask;
1163		do_div(sec, freq);
1164		do_div(sec, scale);
1165		if (!sec)
1166			sec = 1;
1167		else if (sec > 600 && cs->mask > UINT_MAX)
1168			sec = 600;
1169
1170		clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
1171				       NSEC_PER_SEC / scale, sec * scale);
1172	}
1173
1174	/*
1175	 * If the uncertainty margin is not specified, calculate it.  If
1176	 * both scale and freq are non-zero, calculate the clock period, but
1177	 * bound below at 2*WATCHDOG_MAX_SKEW, that is, 500ppm by default.
1178	 * However, if either of scale or freq is zero, be very conservative
1179	 * and take the tens-of-milliseconds WATCHDOG_THRESHOLD value
1180	 * for the uncertainty margin.  Allow stupidly small uncertainty
1181	 * margins to be specified by the caller for testing purposes,
1182	 * but warn to discourage production use of this capability.
1183	 *
1184	 * Bottom line:  The sum of the uncertainty margins of the
1185	 * watchdog clocksource and the clocksource under test will be at
1186	 * least 500ppm by default.  For more information, please see the
1187	 * comment preceding CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US above.
1188	 */
1189	if (scale && freq && !cs->uncertainty_margin) {
1190		cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq);
1191		if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW)
1192			cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW;
1193	} else if (!cs->uncertainty_margin) {
1194		cs->uncertainty_margin = WATCHDOG_THRESHOLD;
1195	}
1196	WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW);
1197
1198	/*
1199	 * Ensure clocksources that have large 'mult' values don't overflow
1200	 * when adjusted.
1201	 */
1202	cs->maxadj = clocksource_max_adjustment(cs);
1203	while (freq && ((cs->mult + cs->maxadj < cs->mult)
1204		|| (cs->mult - cs->maxadj > cs->mult))) {
1205		cs->mult >>= 1;
1206		cs->shift--;
1207		cs->maxadj = clocksource_max_adjustment(cs);
1208	}
1209
1210	/*
1211	 * Only warn for *special* clocksources that self-define
1212	 * their mult/shift values and don't specify a freq.
1213	 */
1214	WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
1215		"timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
1216		cs->name);
1217
1218	clocksource_update_max_deferment(cs);
1219
1220	pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
1221		cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
1222}
1223EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
1224
1225/**
1226 * __clocksource_register_scale - Used to install new clocksources
1227 * @cs:		clocksource to be registered
1228 * @scale:	Scale factor multiplied against freq to get clocksource hz
1229 * @freq:	clocksource frequency (cycles per second) divided by scale
1230 *
1231 * Returns -EBUSY if registration fails, zero otherwise.
1232 *
1233 * This *SHOULD NOT* be called directly! Please use the
1234 * clocksource_register_hz() or clocksource_register_khz helper functions.
1235 */
1236int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
1237{
1238	unsigned long flags;
1239
1240	clocksource_arch_init(cs);
1241
1242	if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX))
1243		cs->id = CSID_GENERIC;
1244	if (cs->vdso_clock_mode < 0 ||
1245	    cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) {
1246		pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n",
1247			cs->name, cs->vdso_clock_mode);
1248		cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE;
1249	}
1250
1251	/* Initialize mult/shift and max_idle_ns */
1252	__clocksource_update_freq_scale(cs, scale, freq);
1253
1254	/* Add clocksource to the clocksource list */
1255	mutex_lock(&clocksource_mutex);
1256
1257	clocksource_watchdog_lock(&flags);
1258	clocksource_enqueue(cs);
1259	clocksource_enqueue_watchdog(cs);
1260	clocksource_watchdog_unlock(&flags);
1261
1262	clocksource_select();
1263	clocksource_select_watchdog(false);
1264	__clocksource_suspend_select(cs);
1265	mutex_unlock(&clocksource_mutex);
1266	return 0;
1267}
1268EXPORT_SYMBOL_GPL(__clocksource_register_scale);
1269
1270/*
1271 * Unbind clocksource @cs. Called with clocksource_mutex held
1272 */
1273static int clocksource_unbind(struct clocksource *cs)
1274{
1275	unsigned long flags;
1276
1277	if (clocksource_is_watchdog(cs)) {
1278		/* Select and try to install a replacement watchdog. */
1279		clocksource_select_watchdog(true);
1280		if (clocksource_is_watchdog(cs))
1281			return -EBUSY;
1282	}
1283
1284	if (cs == curr_clocksource) {
1285		/* Select and try to install a replacement clock source */
1286		clocksource_select_fallback();
1287		if (curr_clocksource == cs)
1288			return -EBUSY;
1289	}
1290
1291	if (clocksource_is_suspend(cs)) {
1292		/*
1293		 * Select and try to install a replacement suspend clocksource.
1294		 * If no replacement suspend clocksource, we will just let the
1295		 * clocksource go and have no suspend clocksource.
1296		 */
1297		clocksource_suspend_select(true);
1298	}
1299
1300	clocksource_watchdog_lock(&flags);
1301	clocksource_dequeue_watchdog(cs);
1302	list_del_init(&cs->list);
1303	clocksource_watchdog_unlock(&flags);
1304
1305	return 0;
1306}
1307
1308/**
1309 * clocksource_unregister - remove a registered clocksource
1310 * @cs:	clocksource to be unregistered
1311 */
1312int clocksource_unregister(struct clocksource *cs)
1313{
1314	int ret = 0;
1315
1316	mutex_lock(&clocksource_mutex);
1317	if (!list_empty(&cs->list))
1318		ret = clocksource_unbind(cs);
1319	mutex_unlock(&clocksource_mutex);
1320	return ret;
1321}
1322EXPORT_SYMBOL(clocksource_unregister);
1323
1324#ifdef CONFIG_SYSFS
1325/**
1326 * current_clocksource_show - sysfs interface for current clocksource
1327 * @dev:	unused
1328 * @attr:	unused
1329 * @buf:	char buffer to be filled with clocksource list
1330 *
1331 * Provides sysfs interface for listing current clocksource.
1332 */
1333static ssize_t current_clocksource_show(struct device *dev,
1334					struct device_attribute *attr,
1335					char *buf)
1336{
1337	ssize_t count = 0;
1338
1339	mutex_lock(&clocksource_mutex);
1340	count = sysfs_emit(buf, "%s\n", curr_clocksource->name);
1341	mutex_unlock(&clocksource_mutex);
1342
1343	return count;
1344}
1345
1346ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
1347{
1348	size_t ret = cnt;
1349
1350	/* strings from sysfs write are not 0 terminated! */
1351	if (!cnt || cnt >= CS_NAME_LEN)
1352		return -EINVAL;
1353
1354	/* strip of \n: */
1355	if (buf[cnt-1] == '\n')
1356		cnt--;
1357	if (cnt > 0)
1358		memcpy(dst, buf, cnt);
1359	dst[cnt] = 0;
1360	return ret;
1361}
1362
1363/**
1364 * current_clocksource_store - interface for manually overriding clocksource
1365 * @dev:	unused
1366 * @attr:	unused
1367 * @buf:	name of override clocksource
1368 * @count:	length of buffer
1369 *
1370 * Takes input from sysfs interface for manually overriding the default
1371 * clocksource selection.
1372 */
1373static ssize_t current_clocksource_store(struct device *dev,
1374					 struct device_attribute *attr,
1375					 const char *buf, size_t count)
1376{
1377	ssize_t ret;
1378
1379	mutex_lock(&clocksource_mutex);
1380
1381	ret = sysfs_get_uname(buf, override_name, count);
1382	if (ret >= 0)
1383		clocksource_select();
1384
1385	mutex_unlock(&clocksource_mutex);
1386
1387	return ret;
1388}
1389static DEVICE_ATTR_RW(current_clocksource);
1390
1391/**
1392 * unbind_clocksource_store - interface for manually unbinding clocksource
1393 * @dev:	unused
1394 * @attr:	unused
1395 * @buf:	unused
1396 * @count:	length of buffer
1397 *
1398 * Takes input from sysfs interface for manually unbinding a clocksource.
1399 */
1400static ssize_t unbind_clocksource_store(struct device *dev,
1401					struct device_attribute *attr,
1402					const char *buf, size_t count)
1403{
1404	struct clocksource *cs;
1405	char name[CS_NAME_LEN];
1406	ssize_t ret;
1407
1408	ret = sysfs_get_uname(buf, name, count);
1409	if (ret < 0)
1410		return ret;
1411
1412	ret = -ENODEV;
1413	mutex_lock(&clocksource_mutex);
1414	list_for_each_entry(cs, &clocksource_list, list) {
1415		if (strcmp(cs->name, name))
1416			continue;
1417		ret = clocksource_unbind(cs);
1418		break;
1419	}
1420	mutex_unlock(&clocksource_mutex);
1421
1422	return ret ? ret : count;
1423}
1424static DEVICE_ATTR_WO(unbind_clocksource);
1425
1426/**
1427 * available_clocksource_show - sysfs interface for listing clocksource
1428 * @dev:	unused
1429 * @attr:	unused
1430 * @buf:	char buffer to be filled with clocksource list
1431 *
1432 * Provides sysfs interface for listing registered clocksources
1433 */
1434static ssize_t available_clocksource_show(struct device *dev,
1435					  struct device_attribute *attr,
1436					  char *buf)
1437{
1438	struct clocksource *src;
1439	ssize_t count = 0;
1440
1441	mutex_lock(&clocksource_mutex);
1442	list_for_each_entry(src, &clocksource_list, list) {
1443		/*
1444		 * Don't show non-HRES clocksource if the tick code is
1445		 * in one shot mode (highres=on or nohz=on)
1446		 */
1447		if (!tick_oneshot_mode_active() ||
1448		    (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1449			count += snprintf(buf + count,
1450				  max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1451				  "%s ", src->name);
1452	}
1453	mutex_unlock(&clocksource_mutex);
1454
1455	count += snprintf(buf + count,
1456			  max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1457
1458	return count;
1459}
1460static DEVICE_ATTR_RO(available_clocksource);
1461
1462static struct attribute *clocksource_attrs[] = {
1463	&dev_attr_current_clocksource.attr,
1464	&dev_attr_unbind_clocksource.attr,
1465	&dev_attr_available_clocksource.attr,
1466	NULL
1467};
1468ATTRIBUTE_GROUPS(clocksource);
1469
1470static const struct bus_type clocksource_subsys = {
1471	.name = "clocksource",
1472	.dev_name = "clocksource",
1473};
1474
1475static struct device device_clocksource = {
1476	.id	= 0,
1477	.bus	= &clocksource_subsys,
1478	.groups	= clocksource_groups,
1479};
1480
1481static int __init init_clocksource_sysfs(void)
1482{
1483	int error = subsys_system_register(&clocksource_subsys, NULL);
1484
1485	if (!error)
1486		error = device_register(&device_clocksource);
1487
1488	return error;
1489}
1490
1491device_initcall(init_clocksource_sysfs);
1492#endif /* CONFIG_SYSFS */
1493
1494/**
1495 * boot_override_clocksource - boot clock override
1496 * @str:	override name
1497 *
1498 * Takes a clocksource= boot argument and uses it
1499 * as the clocksource override name.
1500 */
1501static int __init boot_override_clocksource(char* str)
1502{
1503	mutex_lock(&clocksource_mutex);
1504	if (str)
1505		strscpy(override_name, str);
1506	mutex_unlock(&clocksource_mutex);
1507	return 1;
1508}
1509
1510__setup("clocksource=", boot_override_clocksource);
1511
1512/**
1513 * boot_override_clock - Compatibility layer for deprecated boot option
1514 * @str:	override name
1515 *
1516 * DEPRECATED! Takes a clock= boot argument and uses it
1517 * as the clocksource override name
1518 */
1519static int __init boot_override_clock(char* str)
1520{
1521	if (!strcmp(str, "pmtmr")) {
1522		pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
1523		return boot_override_clocksource("acpi_pm");
1524	}
1525	pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
1526	return boot_override_clocksource(str);
1527}
1528
1529__setup("clock=", boot_override_clock);