1/*
   2 *  linux/kernel/time/timekeeping.c
   3 *
   4 *  Kernel timekeeping code and accessor functions
   5 *
   6 *  This code was moved from linux/kernel/timer.c.
   7 *  Please see that file for copyright and history logs.
   8 *
   9 */
  10
  11#include <linux/timekeeper_internal.h>
  12#include <linux/module.h>
  13#include <linux/interrupt.h>
  14#include <linux/percpu.h>
  15#include <linux/init.h>
  16#include <linux/mm.h>
  17#include <linux/sched.h>
  18#include <linux/syscore_ops.h>
  19#include <linux/clocksource.h>
  20#include <linux/jiffies.h>
  21#include <linux/time.h>
  22#include <linux/tick.h>
  23#include <linux/stop_machine.h>
  24#include <linux/pvclock_gtod.h>
  25#include <linux/compiler.h>
  26
  27#include "tick-internal.h"
  28#include "ntp_internal.h"
  29#include "timekeeping_internal.h"
  30
  31#define TK_CLEAR_NTP		(1 << 0)
  32#define TK_MIRROR		(1 << 1)
  33#define TK_CLOCK_WAS_SET	(1 << 2)
  34
  35static struct timekeeper timekeeper;
  36static DEFINE_RAW_SPINLOCK(timekeeper_lock);
  37static seqcount_t timekeeper_seq;
  38static struct timekeeper shadow_timekeeper;
  39
  40/* flag for if timekeeping is suspended */
  41int __read_mostly timekeeping_suspended;
  42
  43/* Flag for if there is a persistent clock on this platform */
  44bool __read_mostly persistent_clock_exist = false;
  45
  46static inline void tk_normalize_xtime(struct timekeeper *tk)
  47{
  48	while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
  49		tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
  50		tk->xtime_sec++;
  51	}
  52}
  53
  54static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
  55{
  56	tk->xtime_sec = ts->tv_sec;
  57	tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
  58}
  59
  60static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
  61{
  62	tk->xtime_sec += ts->tv_sec;
  63	tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
  64	tk_normalize_xtime(tk);
  65}
  66
  67static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
  68{
  69	struct timespec tmp;
  70
  71	/*
  72	 * Verify consistency of: offset_real = -wall_to_monotonic
  73	 * before modifying anything
  74	 */
  75	set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
  76					-tk->wall_to_monotonic.tv_nsec);
  77	WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
  78	tk->wall_to_monotonic = wtm;
  79	set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
  80	tk->offs_real = timespec_to_ktime(tmp);
  81	tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
  82}
  83
  84static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
  85{
  86	/* Verify consistency before modifying */
  87	WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  88
  89	tk->total_sleep_time	= t;
  90	tk->offs_boot		= timespec_to_ktime(t);
  91}
  92
  93/**
  94 * tk_setup_internals - Set up internals to use clocksource clock.
  95 *
  96 * @tk:		The target timekeeper to setup.
  97 * @clock:		Pointer to clocksource.
  98 *
  99 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
 100 * pair and interval request.
 101 *
 102 * Unless you're the timekeeping code, you should not be using this!
 103 */
 104static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
 105{
 106	cycle_t interval;
 107	u64 tmp, ntpinterval;
 108	struct clocksource *old_clock;
 109
 110	old_clock = tk->clock;
 111	tk->clock = clock;
 112	tk->cycle_last = clock->cycle_last = clock->read(clock);
 113
 114	/* Do the ns -> cycle conversion first, using original mult */
 115	tmp = NTP_INTERVAL_LENGTH;
 116	tmp <<= clock->shift;
 117	ntpinterval = tmp;
 118	tmp += clock->mult/2;
 119	do_div(tmp, clock->mult);
 120	if (tmp == 0)
 121		tmp = 1;
 122
 123	interval = (cycle_t) tmp;
 124	tk->cycle_interval = interval;
 125
 126	/* Go back from cycles -> shifted ns */
 127	tk->xtime_interval = (u64) interval * clock->mult;
 128	tk->xtime_remainder = ntpinterval - tk->xtime_interval;
 129	tk->raw_interval =
 130		((u64) interval * clock->mult) >> clock->shift;
 131
 132	 /* if changing clocks, convert xtime_nsec shift units */
 133	if (old_clock) {
 134		int shift_change = clock->shift - old_clock->shift;
 135		if (shift_change < 0)
 136			tk->xtime_nsec >>= -shift_change;
 137		else
 138			tk->xtime_nsec <<= shift_change;
 139	}
 140	tk->shift = clock->shift;
 141
 142	tk->ntp_error = 0;
 143	tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
 144
 145	/*
 146	 * The timekeeper keeps its own mult values for the currently
 147	 * active clocksource. These value will be adjusted via NTP
 148	 * to counteract clock drifting.
 149	 */
 150	tk->mult = clock->mult;
 151}
 152
 153/* Timekeeper helper functions. */
 154
 155#ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
 156u32 (*arch_gettimeoffset)(void);
 157
 158u32 get_arch_timeoffset(void)
 159{
 160	if (likely(arch_gettimeoffset))
 161		return arch_gettimeoffset();
 162	return 0;
 163}
 164#else
 165static inline u32 get_arch_timeoffset(void) { return 0; }
 166#endif
 167
 168static inline s64 timekeeping_get_ns(struct timekeeper *tk)
 169{
 170	cycle_t cycle_now, cycle_delta;
 171	struct clocksource *clock;
 172	s64 nsec;
 173
 174	/* read clocksource: */
 175	clock = tk->clock;
 176	cycle_now = clock->read(clock);
 177
 178	/* calculate the delta since the last update_wall_time: */
 179	cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
 180
 181	nsec = cycle_delta * tk->mult + tk->xtime_nsec;
 182	nsec >>= tk->shift;
 183
 184	/* If arch requires, add in get_arch_timeoffset() */
 185	return nsec + get_arch_timeoffset();
 186}
 187
 188static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
 189{
 190	cycle_t cycle_now, cycle_delta;
 191	struct clocksource *clock;
 192	s64 nsec;
 193
 194	/* read clocksource: */
 195	clock = tk->clock;
 196	cycle_now = clock->read(clock);
 197
 198	/* calculate the delta since the last update_wall_time: */
 199	cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
 200
 201	/* convert delta to nanoseconds. */
 202	nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
 203
 204	/* If arch requires, add in get_arch_timeoffset() */
 205	return nsec + get_arch_timeoffset();
 206}
 207
 208static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
 209
 210static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
 211{
 212	raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
 213}
 214
 215/**
 216 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
 
 217 */
 218int pvclock_gtod_register_notifier(struct notifier_block *nb)
 219{
 220	struct timekeeper *tk = &timekeeper;
 221	unsigned long flags;
 222	int ret;
 223
 224	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 225	ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
 226	update_pvclock_gtod(tk, true);
 227	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 228
 229	return ret;
 230}
 231EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 232
 233/**
 234 * pvclock_gtod_unregister_notifier - unregister a pvclock
 235 * timedata update listener
 236 */
 237int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
 238{
 239	unsigned long flags;
 240	int ret;
 241
 242	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 243	ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
 244	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 245
 246	return ret;
 247}
 248EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
 249
 250/* must hold timekeeper_lock */
 251static void timekeeping_update(struct timekeeper *tk, unsigned int action)
 252{
 253	if (action & TK_CLEAR_NTP) {
 254		tk->ntp_error = 0;
 255		ntp_clear();
 256	}
 257	update_vsyscall(tk);
 258	update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
 259
 260	if (action & TK_MIRROR)
 261		memcpy(&shadow_timekeeper, &timekeeper, sizeof(timekeeper));
 262}
 263
 264/**
 265 * timekeeping_forward_now - update clock to the current time
 266 *
 267 * Forward the current clock to update its state since the last call to
 268 * update_wall_time(). This is useful before significant clock changes,
 269 * as it avoids having to deal with this time offset explicitly.
 270 */
 271static void timekeeping_forward_now(struct timekeeper *tk)
 272{
 273	cycle_t cycle_now, cycle_delta;
 274	struct clocksource *clock;
 275	s64 nsec;
 276
 277	clock = tk->clock;
 278	cycle_now = clock->read(clock);
 279	cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
 280	tk->cycle_last = clock->cycle_last = cycle_now;
 281
 282	tk->xtime_nsec += cycle_delta * tk->mult;
 
 283
 284	/* If arch requires, add in get_arch_timeoffset() */
 285	tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift;
 286
 287	tk_normalize_xtime(tk);
 288
 289	nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
 290	timespec_add_ns(&tk->raw_time, nsec);
 291}
 292
 293/**
 294 * __getnstimeofday - Returns the time of day in a timespec.
 295 * @ts:		pointer to the timespec to be set
 296 *
 297 * Updates the time of day in the timespec.
 298 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
 299 */
 300int __getnstimeofday(struct timespec *ts)
 301{
 302	struct timekeeper *tk = &timekeeper;
 303	unsigned long seq;
 304	s64 nsecs = 0;
 
 
 305
 306	do {
 307		seq = read_seqcount_begin(&timekeeper_seq);
 308
 309		ts->tv_sec = tk->xtime_sec;
 310		nsecs = timekeeping_get_ns(tk);
 311
 312	} while (read_seqcount_retry(&timekeeper_seq, seq));
 
 313
 314	ts->tv_nsec = 0;
 315	timespec_add_ns(ts, nsecs);
 316
 317	/*
 318	 * Do not bail out early, in case there were callers still using
 319	 * the value, even in the face of the WARN_ON.
 320	 */
 321	if (unlikely(timekeeping_suspended))
 322		return -EAGAIN;
 323	return 0;
 324}
 325EXPORT_SYMBOL(__getnstimeofday);
 326
 327/**
 328 * getnstimeofday - Returns the time of day in a timespec.
 329 * @ts:		pointer to the timespec to be set
 330 *
 331 * Returns the time of day in a timespec (WARN if suspended).
 332 */
 333void getnstimeofday(struct timespec *ts)
 334{
 335	WARN_ON(__getnstimeofday(ts));
 336}
 337EXPORT_SYMBOL(getnstimeofday);
 338
 339ktime_t ktime_get(void)
 340{
 341	struct timekeeper *tk = &timekeeper;
 342	unsigned int seq;
 343	s64 secs, nsecs;
 344
 345	WARN_ON(timekeeping_suspended);
 346
 347	do {
 348		seq = read_seqcount_begin(&timekeeper_seq);
 349		secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
 350		nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
 
 351
 352	} while (read_seqcount_retry(&timekeeper_seq, seq));
 353	/*
 354	 * Use ktime_set/ktime_add_ns to create a proper ktime on
 355	 * 32-bit architectures without CONFIG_KTIME_SCALAR.
 356	 */
 357	return ktime_add_ns(ktime_set(secs, 0), nsecs);
 358}
 359EXPORT_SYMBOL_GPL(ktime_get);
 360
 361/**
 362 * ktime_get_ts - get the monotonic clock in timespec format
 363 * @ts:		pointer to timespec variable
 364 *
 365 * The function calculates the monotonic clock from the realtime
 366 * clock and the wall_to_monotonic offset and stores the result
 367 * in normalized timespec format in the variable pointed to by @ts.
 368 */
 369void ktime_get_ts(struct timespec *ts)
 370{
 371	struct timekeeper *tk = &timekeeper;
 372	struct timespec tomono;
 373	s64 nsec;
 374	unsigned int seq;
 
 375
 376	WARN_ON(timekeeping_suspended);
 377
 378	do {
 379		seq = read_seqcount_begin(&timekeeper_seq);
 380		ts->tv_sec = tk->xtime_sec;
 381		nsec = timekeeping_get_ns(tk);
 382		tomono = tk->wall_to_monotonic;
 383
 384	} while (read_seqcount_retry(&timekeeper_seq, seq));
 385
 386	ts->tv_sec += tomono.tv_sec;
 387	ts->tv_nsec = 0;
 388	timespec_add_ns(ts, nsec + tomono.tv_nsec);
 389}
 390EXPORT_SYMBOL_GPL(ktime_get_ts);
 391
 392
 393/**
 394 * timekeeping_clocktai - Returns the TAI time of day in a timespec
 395 * @ts:		pointer to the timespec to be set
 396 *
 397 * Returns the time of day in a timespec.
 398 */
 399void timekeeping_clocktai(struct timespec *ts)
 400{
 401	struct timekeeper *tk = &timekeeper;
 402	unsigned long seq;
 403	u64 nsecs;
 404
 405	WARN_ON(timekeeping_suspended);
 406
 407	do {
 408		seq = read_seqcount_begin(&timekeeper_seq);
 409
 410		ts->tv_sec = tk->xtime_sec + tk->tai_offset;
 411		nsecs = timekeeping_get_ns(tk);
 412
 413	} while (read_seqcount_retry(&timekeeper_seq, seq));
 414
 415	ts->tv_nsec = 0;
 416	timespec_add_ns(ts, nsecs);
 417
 418}
 419EXPORT_SYMBOL(timekeeping_clocktai);
 420
 421
 422/**
 423 * ktime_get_clocktai - Returns the TAI time of day in a ktime
 424 *
 425 * Returns the time of day in a ktime.
 426 */
 427ktime_t ktime_get_clocktai(void)
 428{
 429	struct timespec ts;
 430
 431	timekeeping_clocktai(&ts);
 432	return timespec_to_ktime(ts);
 433}
 434EXPORT_SYMBOL(ktime_get_clocktai);
 435
 436#ifdef CONFIG_NTP_PPS
 437
 438/**
 439 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
 440 * @ts_raw:	pointer to the timespec to be set to raw monotonic time
 441 * @ts_real:	pointer to the timespec to be set to the time of day
 442 *
 443 * This function reads both the time of day and raw monotonic time at the
 444 * same time atomically and stores the resulting timestamps in timespec
 445 * format.
 446 */
 447void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
 448{
 449	struct timekeeper *tk = &timekeeper;
 450	unsigned long seq;
 451	s64 nsecs_raw, nsecs_real;
 452
 453	WARN_ON_ONCE(timekeeping_suspended);
 454
 455	do {
 456		seq = read_seqcount_begin(&timekeeper_seq);
 
 
 
 
 
 457
 458		*ts_raw = tk->raw_time;
 459		ts_real->tv_sec = tk->xtime_sec;
 460		ts_real->tv_nsec = 0;
 461
 462		nsecs_raw = timekeeping_get_ns_raw(tk);
 463		nsecs_real = timekeeping_get_ns(tk);
 
 
 464
 465	} while (read_seqcount_retry(&timekeeper_seq, seq));
 466
 467	timespec_add_ns(ts_raw, nsecs_raw);
 468	timespec_add_ns(ts_real, nsecs_real);
 469}
 470EXPORT_SYMBOL(getnstime_raw_and_real);
 471
 472#endif /* CONFIG_NTP_PPS */
 473
 474/**
 475 * do_gettimeofday - Returns the time of day in a timeval
 476 * @tv:		pointer to the timeval to be set
 477 *
 478 * NOTE: Users should be converted to using getnstimeofday()
 479 */
 480void do_gettimeofday(struct timeval *tv)
 481{
 482	struct timespec now;
 483
 484	getnstimeofday(&now);
 485	tv->tv_sec = now.tv_sec;
 486	tv->tv_usec = now.tv_nsec/1000;
 487}
 488EXPORT_SYMBOL(do_gettimeofday);
 489
 
 490/**
 491 * do_settimeofday - Sets the time of day
 492 * @tv:		pointer to the timespec variable containing the new time
 493 *
 494 * Sets the time of day to the new time and update NTP and notify hrtimers
 495 */
 496int do_settimeofday(const struct timespec *tv)
 497{
 498	struct timekeeper *tk = &timekeeper;
 499	struct timespec ts_delta, xt;
 500	unsigned long flags;
 501
 502	if (!timespec_valid_strict(tv))
 503		return -EINVAL;
 504
 505	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 506	write_seqcount_begin(&timekeeper_seq);
 507
 508	timekeeping_forward_now(tk);
 509
 510	xt = tk_xtime(tk);
 511	ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
 512	ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
 513
 514	tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
 515
 516	tk_set_xtime(tk, tv);
 
 517
 518	timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
 
 519
 520	write_seqcount_end(&timekeeper_seq);
 521	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 522
 523	/* signal hrtimers about time change */
 524	clock_was_set();
 525
 526	return 0;
 527}
 
 528EXPORT_SYMBOL(do_settimeofday);
 529
 
 530/**
 531 * timekeeping_inject_offset - Adds or subtracts from the current time.
 532 * @tv:		pointer to the timespec variable containing the offset
 533 *
 534 * Adds or subtracts an offset value from the current time.
 535 */
 536int timekeeping_inject_offset(struct timespec *ts)
 537{
 538	struct timekeeper *tk = &timekeeper;
 539	unsigned long flags;
 540	struct timespec tmp;
 541	int ret = 0;
 542
 543	if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
 544		return -EINVAL;
 545
 546	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 547	write_seqcount_begin(&timekeeper_seq);
 548
 549	timekeeping_forward_now(tk);
 550
 551	/* Make sure the proposed value is valid */
 552	tmp = timespec_add(tk_xtime(tk),  *ts);
 553	if (!timespec_valid_strict(&tmp)) {
 554		ret = -EINVAL;
 555		goto error;
 556	}
 557
 558	tk_xtime_add(tk, ts);
 559	tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
 560
 561error: /* even if we error out, we forwarded the time, so call update */
 562	timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
 563
 564	write_seqcount_end(&timekeeper_seq);
 565	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 566
 567	/* signal hrtimers about time change */
 568	clock_was_set();
 569
 570	return ret;
 571}
 572EXPORT_SYMBOL(timekeeping_inject_offset);
 573
 574
 575/**
 576 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
 577 *
 578 */
 579s32 timekeeping_get_tai_offset(void)
 580{
 581	struct timekeeper *tk = &timekeeper;
 582	unsigned int seq;
 583	s32 ret;
 584
 585	do {
 586		seq = read_seqcount_begin(&timekeeper_seq);
 587		ret = tk->tai_offset;
 588	} while (read_seqcount_retry(&timekeeper_seq, seq));
 589
 590	return ret;
 591}
 592
 593/**
 594 * __timekeeping_set_tai_offset - Lock free worker function
 595 *
 596 */
 597static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
 598{
 599	tk->tai_offset = tai_offset;
 600	tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
 601}
 602
 603/**
 604 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
 605 *
 606 */
 607void timekeeping_set_tai_offset(s32 tai_offset)
 608{
 609	struct timekeeper *tk = &timekeeper;
 610	unsigned long flags;
 611
 612	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 613	write_seqcount_begin(&timekeeper_seq);
 614	__timekeeping_set_tai_offset(tk, tai_offset);
 615	timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
 616	write_seqcount_end(&timekeeper_seq);
 617	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 618	clock_was_set();
 619}
 620
 621/**
 622 * change_clocksource - Swaps clocksources if a new one is available
 623 *
 624 * Accumulates current time interval and initializes new clocksource
 625 */
 626static int change_clocksource(void *data)
 627{
 628	struct timekeeper *tk = &timekeeper;
 629	struct clocksource *new, *old;
 630	unsigned long flags;
 631
 632	new = (struct clocksource *) data;
 633
 634	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 635	write_seqcount_begin(&timekeeper_seq);
 636
 637	timekeeping_forward_now(tk);
 638	/*
 639	 * If the cs is in module, get a module reference. Succeeds
 640	 * for built-in code (owner == NULL) as well.
 641	 */
 642	if (try_module_get(new->owner)) {
 643		if (!new->enable || new->enable(new) == 0) {
 644			old = tk->clock;
 645			tk_setup_internals(tk, new);
 646			if (old->disable)
 647				old->disable(old);
 648			module_put(old->owner);
 649		} else {
 650			module_put(new->owner);
 651		}
 652	}
 653	timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
 654
 655	write_seqcount_end(&timekeeper_seq);
 656	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 657
 658	return 0;
 659}
 660
 661/**
 662 * timekeeping_notify - Install a new clock source
 663 * @clock:		pointer to the clock source
 664 *
 665 * This function is called from clocksource.c after a new, better clock
 666 * source has been registered. The caller holds the clocksource_mutex.
 667 */
 668int timekeeping_notify(struct clocksource *clock)
 669{
 670	struct timekeeper *tk = &timekeeper;
 671
 672	if (tk->clock == clock)
 673		return 0;
 674	stop_machine(change_clocksource, clock, NULL);
 675	tick_clock_notify();
 676	return tk->clock == clock ? 0 : -1;
 677}
 678
 679/**
 680 * ktime_get_real - get the real (wall-) time in ktime_t format
 681 *
 682 * returns the time in ktime_t format
 683 */
 684ktime_t ktime_get_real(void)
 685{
 686	struct timespec now;
 687
 688	getnstimeofday(&now);
 689
 690	return timespec_to_ktime(now);
 691}
 692EXPORT_SYMBOL_GPL(ktime_get_real);
 693
 694/**
 695 * getrawmonotonic - Returns the raw monotonic time in a timespec
 696 * @ts:		pointer to the timespec to be set
 697 *
 698 * Returns the raw monotonic time (completely un-modified by ntp)
 699 */
 700void getrawmonotonic(struct timespec *ts)
 701{
 702	struct timekeeper *tk = &timekeeper;
 703	unsigned long seq;
 704	s64 nsecs;
 705
 706	do {
 707		seq = read_seqcount_begin(&timekeeper_seq);
 708		nsecs = timekeeping_get_ns_raw(tk);
 709		*ts = tk->raw_time;
 710
 711	} while (read_seqcount_retry(&timekeeper_seq, seq));
 712
 713	timespec_add_ns(ts, nsecs);
 714}
 715EXPORT_SYMBOL(getrawmonotonic);
 716
 
 717/**
 718 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
 719 */
 720int timekeeping_valid_for_hres(void)
 721{
 722	struct timekeeper *tk = &timekeeper;
 723	unsigned long seq;
 724	int ret;
 725
 726	do {
 727		seq = read_seqcount_begin(&timekeeper_seq);
 728
 729		ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
 730
 731	} while (read_seqcount_retry(&timekeeper_seq, seq));
 732
 733	return ret;
 734}
 735
 736/**
 737 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
 
 
 
 738 */
 739u64 timekeeping_max_deferment(void)
 740{
 741	struct timekeeper *tk = &timekeeper;
 742	unsigned long seq;
 743	u64 ret;
 744
 745	do {
 746		seq = read_seqcount_begin(&timekeeper_seq);
 747
 748		ret = tk->clock->max_idle_ns;
 749
 750	} while (read_seqcount_retry(&timekeeper_seq, seq));
 751
 752	return ret;
 753}
 754
 755/**
 756 * read_persistent_clock -  Return time from the persistent clock.
 757 *
 758 * Weak dummy function for arches that do not yet support it.
 759 * Reads the time from the battery backed persistent clock.
 760 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
 761 *
 762 *  XXX - Do be sure to remove it once all arches implement it.
 763 */
 764void __weak read_persistent_clock(struct timespec *ts)
 765{
 766	ts->tv_sec = 0;
 767	ts->tv_nsec = 0;
 768}
 769
 770/**
 771 * read_boot_clock -  Return time of the system start.
 772 *
 773 * Weak dummy function for arches that do not yet support it.
 774 * Function to read the exact time the system has been started.
 775 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
 776 *
 777 *  XXX - Do be sure to remove it once all arches implement it.
 778 */
 779void __weak read_boot_clock(struct timespec *ts)
 780{
 781	ts->tv_sec = 0;
 782	ts->tv_nsec = 0;
 783}
 784
 785/*
 786 * timekeeping_init - Initializes the clocksource and common timekeeping values
 787 */
 788void __init timekeeping_init(void)
 789{
 790	struct timekeeper *tk = &timekeeper;
 791	struct clocksource *clock;
 792	unsigned long flags;
 793	struct timespec now, boot, tmp;
 794
 795	read_persistent_clock(&now);
 796
 797	if (!timespec_valid_strict(&now)) {
 798		pr_warn("WARNING: Persistent clock returned invalid value!\n"
 799			"         Check your CMOS/BIOS settings.\n");
 800		now.tv_sec = 0;
 801		now.tv_nsec = 0;
 802	} else if (now.tv_sec || now.tv_nsec)
 803		persistent_clock_exist = true;
 804
 805	read_boot_clock(&boot);
 806	if (!timespec_valid_strict(&boot)) {
 807		pr_warn("WARNING: Boot clock returned invalid value!\n"
 808			"         Check your CMOS/BIOS settings.\n");
 809		boot.tv_sec = 0;
 810		boot.tv_nsec = 0;
 811	}
 812
 813	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 814	write_seqcount_begin(&timekeeper_seq);
 815	ntp_init();
 816
 817	clock = clocksource_default_clock();
 818	if (clock->enable)
 819		clock->enable(clock);
 820	tk_setup_internals(tk, clock);
 821
 822	tk_set_xtime(tk, &now);
 823	tk->raw_time.tv_sec = 0;
 824	tk->raw_time.tv_nsec = 0;
 825	if (boot.tv_sec == 0 && boot.tv_nsec == 0)
 826		boot = tk_xtime(tk);
 827
 828	set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
 829	tk_set_wall_to_mono(tk, tmp);
 830
 831	tmp.tv_sec = 0;
 832	tmp.tv_nsec = 0;
 833	tk_set_sleep_time(tk, tmp);
 834
 835	memcpy(&shadow_timekeeper, &timekeeper, sizeof(timekeeper));
 836
 837	write_seqcount_end(&timekeeper_seq);
 838	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 
 
 
 
 
 
 
 
 
 839}
 840
 841/* time in seconds when suspend began */
 842static struct timespec timekeeping_suspend_time;
 843
 844/**
 845 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
 846 * @delta: pointer to a timespec delta value
 847 *
 848 * Takes a timespec offset measuring a suspend interval and properly
 849 * adds the sleep offset to the timekeeping variables.
 850 */
 851static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
 852							struct timespec *delta)
 853{
 854	if (!timespec_valid_strict(delta)) {
 855		printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
 856					"sleep delta value!\n");
 857		return;
 858	}
 859	tk_xtime_add(tk, delta);
 860	tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
 861	tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
 862	tk_debug_account_sleep_time(delta);
 863}
 864
 
 865/**
 866 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
 867 * @delta: pointer to a timespec delta value
 868 *
 869 * This hook is for architectures that cannot support read_persistent_clock
 870 * because their RTC/persistent clock is only accessible when irqs are enabled.
 871 *
 872 * This function should only be called by rtc_resume(), and allows
 873 * a suspend offset to be injected into the timekeeping values.
 874 */
 875void timekeeping_inject_sleeptime(struct timespec *delta)
 876{
 877	struct timekeeper *tk = &timekeeper;
 878	unsigned long flags;
 
 879
 880	/*
 881	 * Make sure we don't set the clock twice, as timekeeping_resume()
 882	 * already did it
 883	 */
 884	if (has_persistent_clock())
 885		return;
 886
 887	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 888	write_seqcount_begin(&timekeeper_seq);
 889
 890	timekeeping_forward_now(tk);
 891
 892	__timekeeping_inject_sleeptime(tk, delta);
 893
 894	timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
 
 
 
 895
 896	write_seqcount_end(&timekeeper_seq);
 897	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 898
 899	/* signal hrtimers about time change */
 900	clock_was_set();
 901}
 902
 
 903/**
 904 * timekeeping_resume - Resumes the generic timekeeping subsystem.
 905 *
 906 * This is for the generic clocksource timekeeping.
 907 * xtime/wall_to_monotonic/jiffies/etc are
 908 * still managed by arch specific suspend/resume code.
 909 */
 910static void timekeeping_resume(void)
 911{
 912	struct timekeeper *tk = &timekeeper;
 913	struct clocksource *clock = tk->clock;
 914	unsigned long flags;
 915	struct timespec ts_new, ts_delta;
 916	cycle_t cycle_now, cycle_delta;
 917	bool suspendtime_found = false;
 918
 919	read_persistent_clock(&ts_new);
 920
 921	clockevents_resume();
 922	clocksource_resume();
 923
 924	raw_spin_lock_irqsave(&timekeeper_lock, flags);
 925	write_seqcount_begin(&timekeeper_seq);
 926
 927	/*
 928	 * After system resumes, we need to calculate the suspended time and
 929	 * compensate it for the OS time. There are 3 sources that could be
 930	 * used: Nonstop clocksource during suspend, persistent clock and rtc
 931	 * device.
 932	 *
 933	 * One specific platform may have 1 or 2 or all of them, and the
 934	 * preference will be:
 935	 *	suspend-nonstop clocksource -> persistent clock -> rtc
 936	 * The less preferred source will only be tried if there is no better
 937	 * usable source. The rtc part is handled separately in rtc core code.
 938	 */
 939	cycle_now = clock->read(clock);
 940	if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
 941		cycle_now > clock->cycle_last) {
 942		u64 num, max = ULLONG_MAX;
 943		u32 mult = clock->mult;
 944		u32 shift = clock->shift;
 945		s64 nsec = 0;
 946
 947		cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
 948
 949		/*
 950		 * "cycle_delta * mutl" may cause 64 bits overflow, if the
 951		 * suspended time is too long. In that case we need do the
 952		 * 64 bits math carefully
 953		 */
 954		do_div(max, mult);
 955		if (cycle_delta > max) {
 956			num = div64_u64(cycle_delta, max);
 957			nsec = (((u64) max * mult) >> shift) * num;
 958			cycle_delta -= num * max;
 959		}
 960		nsec += ((u64) cycle_delta * mult) >> shift;
 961
 962		ts_delta = ns_to_timespec(nsec);
 963		suspendtime_found = true;
 964	} else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) {
 965		ts_delta = timespec_sub(ts_new, timekeeping_suspend_time);
 966		suspendtime_found = true;
 967	}
 968
 969	if (suspendtime_found)
 970		__timekeeping_inject_sleeptime(tk, &ts_delta);
 971
 972	/* Re-base the last cycle value */
 973	tk->cycle_last = clock->cycle_last = cycle_now;
 974	tk->ntp_error = 0;
 975	timekeeping_suspended = 0;
 976	timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
 977	write_seqcount_end(&timekeeper_seq);
 978	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
 979
 980	touch_softlockup_watchdog();
 981
 982	clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
 983
 984	/* Resume hrtimers */
 985	hrtimers_resume();
 986}
 987
 988static int timekeeping_suspend(void)
 989{
 990	struct timekeeper *tk = &timekeeper;
 991	unsigned long flags;
 992	struct timespec		delta, delta_delta;
 993	static struct timespec	old_delta;
 994
 995	read_persistent_clock(&timekeeping_suspend_time);
 996
 997	/*
 998	 * On some systems the persistent_clock can not be detected at
 999	 * timekeeping_init by its return value, so if we see a valid
1000	 * value returned, update the persistent_clock_exists flag.
1001	 */
1002	if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1003		persistent_clock_exist = true;
1004
1005	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1006	write_seqcount_begin(&timekeeper_seq);
1007	timekeeping_forward_now(tk);
1008	timekeeping_suspended = 1;
1009
1010	/*
1011	 * To avoid drift caused by repeated suspend/resumes,
1012	 * which each can add ~1 second drift error,
1013	 * try to compensate so the difference in system time
1014	 * and persistent_clock time stays close to constant.
1015	 */
1016	delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
1017	delta_delta = timespec_sub(delta, old_delta);
1018	if (abs(delta_delta.tv_sec)  >= 2) {
1019		/*
1020		 * if delta_delta is too large, assume time correction
1021		 * has occured and set old_delta to the current delta.
1022		 */
1023		old_delta = delta;
1024	} else {
1025		/* Otherwise try to adjust old_system to compensate */
1026		timekeeping_suspend_time =
1027			timespec_add(timekeeping_suspend_time, delta_delta);
1028	}
1029
1030	timekeeping_update(tk, TK_MIRROR);
1031	write_seqcount_end(&timekeeper_seq);
1032	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1033
1034	clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
1035	clocksource_suspend();
1036	clockevents_suspend();
1037
1038	return 0;
1039}
1040
1041/* sysfs resume/suspend bits for timekeeping */
1042static struct syscore_ops timekeeping_syscore_ops = {
1043	.resume		= timekeeping_resume,
1044	.suspend	= timekeeping_suspend,
1045};
1046
1047static int __init timekeeping_init_ops(void)
1048{
1049	register_syscore_ops(&timekeeping_syscore_ops);
1050	return 0;
1051}
1052
1053device_initcall(timekeeping_init_ops);
1054
1055/*
1056 * If the error is already larger, we look ahead even further
1057 * to compensate for late or lost adjustments.
1058 */
1059static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
1060						 s64 error, s64 *interval,
1061						 s64 *offset)
1062{
1063	s64 tick_error, i;
1064	u32 look_ahead, adj;
1065	s32 error2, mult;
1066
1067	/*
1068	 * Use the current error value to determine how much to look ahead.
1069	 * The larger the error the slower we adjust for it to avoid problems
1070	 * with losing too many ticks, otherwise we would overadjust and
1071	 * produce an even larger error.  The smaller the adjustment the
1072	 * faster we try to adjust for it, as lost ticks can do less harm
1073	 * here.  This is tuned so that an error of about 1 msec is adjusted
1074	 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1075	 */
1076	error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
1077	error2 = abs(error2);
1078	for (look_ahead = 0; error2 > 0; look_ahead++)
1079		error2 >>= 2;
1080
1081	/*
1082	 * Now calculate the error in (1 << look_ahead) ticks, but first
1083	 * remove the single look ahead already included in the error.
1084	 */
1085	tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
1086	tick_error -= tk->xtime_interval >> 1;
1087	error = ((error - tick_error) >> look_ahead) + tick_error;
1088
1089	/* Finally calculate the adjustment shift value.  */
1090	i = *interval;
1091	mult = 1;
1092	if (error < 0) {
1093		error = -error;
1094		*interval = -*interval;
1095		*offset = -*offset;
1096		mult = -1;
1097	}
1098	for (adj = 0; error > i; adj++)
1099		error >>= 1;
1100
1101	*interval <<= adj;
1102	*offset <<= adj;
1103	return mult << adj;
1104}
1105
1106/*
1107 * Adjust the multiplier to reduce the error value,
1108 * this is optimized for the most common adjustments of -1,0,1,
1109 * for other values we can do a bit more work.
1110 */
1111static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1112{
1113	s64 error, interval = tk->cycle_interval;
1114	int adj;
1115
1116	/*
1117	 * The point of this is to check if the error is greater than half
1118	 * an interval.
1119	 *
1120	 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
1121	 *
1122	 * Note we subtract one in the shift, so that error is really error*2.
1123	 * This "saves" dividing(shifting) interval twice, but keeps the
1124	 * (error > interval) comparison as still measuring if error is
1125	 * larger than half an interval.
1126	 *
1127	 * Note: It does not "save" on aggravation when reading the code.
1128	 */
1129	error = tk->ntp_error >> (tk->ntp_error_shift - 1);
1130	if (error > interval) {
1131		/*
1132		 * We now divide error by 4(via shift), which checks if
1133		 * the error is greater than twice the interval.
1134		 * If it is greater, we need a bigadjust, if its smaller,
1135		 * we can adjust by 1.
1136		 */
1137		error >>= 2;
1138		if (likely(error <= interval))
1139			adj = 1;
1140		else
1141			adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1142	} else {
1143		if (error < -interval) {
1144			/* See comment above, this is just switched for the negative */
1145			error >>= 2;
1146			if (likely(error >= -interval)) {
1147				adj = -1;
1148				interval = -interval;
1149				offset = -offset;
1150			} else {
1151				adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1152			}
1153		} else {
1154			goto out_adjust;
1155		}
1156	}
1157
1158	if (unlikely(tk->clock->maxadj &&
1159		(tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
1160		printk_once(KERN_WARNING
1161			"Adjusting %s more than 11%% (%ld vs %ld)\n",
1162			tk->clock->name, (long)tk->mult + adj,
1163			(long)tk->clock->mult + tk->clock->maxadj);
1164	}
1165	/*
1166	 * So the following can be confusing.
1167	 *
1168	 * To keep things simple, lets assume adj == 1 for now.
1169	 *
1170	 * When adj != 1, remember that the interval and offset values
1171	 * have been appropriately scaled so the math is the same.
1172	 *
1173	 * The basic idea here is that we're increasing the multiplier
1174	 * by one, this causes the xtime_interval to be incremented by
1175	 * one cycle_interval. This is because:
1176	 *	xtime_interval = cycle_interval * mult
1177	 * So if mult is being incremented by one:
1178	 *	xtime_interval = cycle_interval * (mult + 1)
1179	 * Its the same as:
1180	 *	xtime_interval = (cycle_interval * mult) + cycle_interval
1181	 * Which can be shortened to:
1182	 *	xtime_interval += cycle_interval
1183	 *
1184	 * So offset stores the non-accumulated cycles. Thus the current
1185	 * time (in shifted nanoseconds) is:
1186	 *	now = (offset * adj) + xtime_nsec
1187	 * Now, even though we're adjusting the clock frequency, we have
1188	 * to keep time consistent. In other words, we can't jump back
1189	 * in time, and we also want to avoid jumping forward in time.
1190	 *
1191	 * So given the same offset value, we need the time to be the same
1192	 * both before and after the freq adjustment.
1193	 *	now = (offset * adj_1) + xtime_nsec_1
1194	 *	now = (offset * adj_2) + xtime_nsec_2
1195	 * So:
1196	 *	(offset * adj_1) + xtime_nsec_1 =
1197	 *		(offset * adj_2) + xtime_nsec_2
1198	 * And we know:
1199	 *	adj_2 = adj_1 + 1
1200	 * So:
1201	 *	(offset * adj_1) + xtime_nsec_1 =
1202	 *		(offset * (adj_1+1)) + xtime_nsec_2
1203	 *	(offset * adj_1) + xtime_nsec_1 =
1204	 *		(offset * adj_1) + offset + xtime_nsec_2
1205	 * Canceling the sides:
1206	 *	xtime_nsec_1 = offset + xtime_nsec_2
1207	 * Which gives us:
1208	 *	xtime_nsec_2 = xtime_nsec_1 - offset
1209	 * Which simplfies to:
1210	 *	xtime_nsec -= offset
1211	 *
1212	 * XXX - TODO: Doc ntp_error calculation.
1213	 */
1214	tk->mult += adj;
1215	tk->xtime_interval += interval;
1216	tk->xtime_nsec -= offset;
1217	tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1218
1219out_adjust:
1220	/*
1221	 * It may be possible that when we entered this function, xtime_nsec
1222	 * was very small.  Further, if we're slightly speeding the clocksource
1223	 * in the code above, its possible the required corrective factor to
1224	 * xtime_nsec could cause it to underflow.
1225	 *
1226	 * Now, since we already accumulated the second, cannot simply roll
1227	 * the accumulated second back, since the NTP subsystem has been
1228	 * notified via second_overflow. So instead we push xtime_nsec forward
1229	 * by the amount we underflowed, and add that amount into the error.
1230	 *
1231	 * We'll correct this error next time through this function, when
1232	 * xtime_nsec is not as small.
1233	 */
1234	if (unlikely((s64)tk->xtime_nsec < 0)) {
1235		s64 neg = -(s64)tk->xtime_nsec;
1236		tk->xtime_nsec = 0;
1237		tk->ntp_error += neg << tk->ntp_error_shift;
1238	}
1239
 
 
 
 
 
1240}
1241
1242/**
1243 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1244 *
1245 * Helper function that accumulates a the nsecs greater then a second
1246 * from the xtime_nsec field to the xtime_secs field.
1247 * It also calls into the NTP code to handle leapsecond processing.
1248 *
1249 */
1250static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
1251{
1252	u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1253	unsigned int clock_set = 0;
1254
1255	while (tk->xtime_nsec >= nsecps) {
1256		int leap;
1257
1258		tk->xtime_nsec -= nsecps;
1259		tk->xtime_sec++;
1260
1261		/* Figure out if its a leap sec and apply if needed */
1262		leap = second_overflow(tk->xtime_sec);
1263		if (unlikely(leap)) {
1264			struct timespec ts;
1265
1266			tk->xtime_sec += leap;
1267
1268			ts.tv_sec = leap;
1269			ts.tv_nsec = 0;
1270			tk_set_wall_to_mono(tk,
1271				timespec_sub(tk->wall_to_monotonic, ts));
1272
1273			__timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1274
1275			clock_set = TK_CLOCK_WAS_SET;
1276		}
1277	}
1278	return clock_set;
1279}
1280
1281/**
1282 * logarithmic_accumulation - shifted accumulation of cycles
1283 *
1284 * This functions accumulates a shifted interval of cycles into
1285 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1286 * loop.
1287 *
1288 * Returns the unconsumed cycles.
1289 */
1290static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1291						u32 shift,
1292						unsigned int *clock_set)
1293{
1294	cycle_t interval = tk->cycle_interval << shift;
1295	u64 raw_nsecs;
1296
1297	/* If the offset is smaller then a shifted interval, do nothing */
1298	if (offset < interval)
1299		return offset;
1300
1301	/* Accumulate one shifted interval */
1302	offset -= interval;
1303	tk->cycle_last += interval;
1304
1305	tk->xtime_nsec += tk->xtime_interval << shift;
1306	*clock_set |= accumulate_nsecs_to_secs(tk);
 
 
 
 
1307
1308	/* Accumulate raw time */
1309	raw_nsecs = (u64)tk->raw_interval << shift;
1310	raw_nsecs += tk->raw_time.tv_nsec;
1311	if (raw_nsecs >= NSEC_PER_SEC) {
1312		u64 raw_secs = raw_nsecs;
1313		raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1314		tk->raw_time.tv_sec += raw_secs;
1315	}
1316	tk->raw_time.tv_nsec = raw_nsecs;
1317
1318	/* Accumulate error between NTP and clock interval */
1319	tk->ntp_error += ntp_tick_length() << shift;
1320	tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1321						(tk->ntp_error_shift + shift);
 
1322
1323	return offset;
1324}
1325
1326#ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1327static inline void old_vsyscall_fixup(struct timekeeper *tk)
1328{
1329	s64 remainder;
1330
1331	/*
1332	* Store only full nanoseconds into xtime_nsec after rounding
1333	* it up and add the remainder to the error difference.
1334	* XXX - This is necessary to avoid small 1ns inconsistnecies caused
1335	* by truncating the remainder in vsyscalls. However, it causes
1336	* additional work to be done in timekeeping_adjust(). Once
1337	* the vsyscall implementations are converted to use xtime_nsec
1338	* (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1339	* users are removed, this can be killed.
1340	*/
1341	remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1342	tk->xtime_nsec -= remainder;
1343	tk->xtime_nsec += 1ULL << tk->shift;
1344	tk->ntp_error += remainder << tk->ntp_error_shift;
1345	tk->ntp_error -= (1ULL << tk->shift) << tk->ntp_error_shift;
1346}
1347#else
1348#define old_vsyscall_fixup(tk)
1349#endif
1350
1351
1352
1353/**
1354 * update_wall_time - Uses the current clocksource to increment the wall time
1355 *
 
1356 */
1357void update_wall_time(void)
1358{
1359	struct clocksource *clock;
1360	struct timekeeper *real_tk = &timekeeper;
1361	struct timekeeper *tk = &shadow_timekeeper;
1362	cycle_t offset;
1363	int shift = 0, maxshift;
1364	unsigned int clock_set = 0;
1365	unsigned long flags;
1366
1367	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1368
1369	/* Make sure we're fully resumed: */
1370	if (unlikely(timekeeping_suspended))
1371		goto out;
1372
1373	clock = real_tk->clock;
1374
1375#ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1376	offset = real_tk->cycle_interval;
1377#else
1378	offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1379#endif
1380
1381	/* Check if there's really nothing to do */
1382	if (offset < real_tk->cycle_interval)
1383		goto out;
1384
1385	/*
1386	 * With NO_HZ we may have to accumulate many cycle_intervals
1387	 * (think "ticks") worth of time at once. To do this efficiently,
1388	 * we calculate the largest doubling multiple of cycle_intervals
1389	 * that is smaller than the offset.  We then accumulate that
1390	 * chunk in one go, and then try to consume the next smaller
1391	 * doubled multiple.
1392	 */
1393	shift = ilog2(offset) - ilog2(tk->cycle_interval);
1394	shift = max(0, shift);
1395	/* Bound shift to one less than what overflows tick_length */
1396	maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1397	shift = min(shift, maxshift);
1398	while (offset >= tk->cycle_interval) {
1399		offset = logarithmic_accumulation(tk, offset, shift,
1400							&clock_set);
1401		if (offset < tk->cycle_interval<<shift)
1402			shift--;
1403	}
1404
1405	/* correct the clock when NTP error is too big */
1406	timekeeping_adjust(tk, offset);
1407
1408	/*
1409	 * XXX This can be killed once everyone converts
1410	 * to the new update_vsyscall.
 
 
 
 
 
 
 
 
 
 
 
 
1411	 */
1412	old_vsyscall_fixup(tk);
 
 
 
 
 
1413
1414	/*
1415	 * Finally, make sure that after the rounding
1416	 * xtime_nsec isn't larger than NSEC_PER_SEC
1417	 */
1418	clock_set |= accumulate_nsecs_to_secs(tk);
 
 
 
1419
1420	write_seqcount_begin(&timekeeper_seq);
1421	/* Update clock->cycle_last with the new value */
1422	clock->cycle_last = tk->cycle_last;
1423	/*
1424	 * Update the real timekeeper.
1425	 *
1426	 * We could avoid this memcpy by switching pointers, but that
1427	 * requires changes to all other timekeeper usage sites as
1428	 * well, i.e. move the timekeeper pointer getter into the
1429	 * spinlocked/seqcount protected sections. And we trade this
1430	 * memcpy under the timekeeper_seq against one before we start
1431	 * updating.
1432	 */
1433	memcpy(real_tk, tk, sizeof(*tk));
1434	timekeeping_update(real_tk, clock_set);
1435	write_seqcount_end(&timekeeper_seq);
1436out:
1437	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1438	if (clock_set)
1439		/* Have to call _delayed version, since in irq context*/
1440		clock_was_set_delayed();
 
1441}
1442
1443/**
1444 * getboottime - Return the real time of system boot.
1445 * @ts:		pointer to the timespec to be set
1446 *
1447 * Returns the wall-time of boot in a timespec.
1448 *
1449 * This is based on the wall_to_monotonic offset and the total suspend
1450 * time. Calls to settimeofday will affect the value returned (which
1451 * basically means that however wrong your real time clock is at boot time,
1452 * you get the right time here).
1453 */
1454void getboottime(struct timespec *ts)
1455{
1456	struct timekeeper *tk = &timekeeper;
1457	struct timespec boottime = {
1458		.tv_sec = tk->wall_to_monotonic.tv_sec +
1459				tk->total_sleep_time.tv_sec,
1460		.tv_nsec = tk->wall_to_monotonic.tv_nsec +
1461				tk->total_sleep_time.tv_nsec
1462	};
1463
1464	set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1465}
1466EXPORT_SYMBOL_GPL(getboottime);
1467
 
1468/**
1469 * get_monotonic_boottime - Returns monotonic time since boot
1470 * @ts:		pointer to the timespec to be set
1471 *
1472 * Returns the monotonic time since boot in a timespec.
1473 *
1474 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1475 * includes the time spent in suspend.
1476 */
1477void get_monotonic_boottime(struct timespec *ts)
1478{
1479	struct timekeeper *tk = &timekeeper;
1480	struct timespec tomono, sleep;
1481	s64 nsec;
1482	unsigned int seq;
 
1483
1484	WARN_ON(timekeeping_suspended);
1485
1486	do {
1487		seq = read_seqcount_begin(&timekeeper_seq);
1488		ts->tv_sec = tk->xtime_sec;
1489		nsec = timekeeping_get_ns(tk);
1490		tomono = tk->wall_to_monotonic;
1491		sleep = tk->total_sleep_time;
1492
1493	} while (read_seqcount_retry(&timekeeper_seq, seq));
1494
1495	ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1496	ts->tv_nsec = 0;
1497	timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1498}
1499EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1500
1501/**
1502 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1503 *
1504 * Returns the monotonic time since boot in a ktime
1505 *
1506 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1507 * includes the time spent in suspend.
1508 */
1509ktime_t ktime_get_boottime(void)
1510{
1511	struct timespec ts;
1512
1513	get_monotonic_boottime(&ts);
1514	return timespec_to_ktime(ts);
1515}
1516EXPORT_SYMBOL_GPL(ktime_get_boottime);
1517
1518/**
1519 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1520 * @ts:		pointer to the timespec to be converted
1521 */
1522void monotonic_to_bootbased(struct timespec *ts)
1523{
1524	struct timekeeper *tk = &timekeeper;
1525
1526	*ts = timespec_add(*ts, tk->total_sleep_time);
1527}
1528EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1529
1530unsigned long get_seconds(void)
1531{
1532	struct timekeeper *tk = &timekeeper;
1533
1534	return tk->xtime_sec;
1535}
1536EXPORT_SYMBOL(get_seconds);
1537
1538struct timespec __current_kernel_time(void)
1539{
1540	struct timekeeper *tk = &timekeeper;
1541
1542	return tk_xtime(tk);
1543}
1544
1545struct timespec current_kernel_time(void)
1546{
1547	struct timekeeper *tk = &timekeeper;
1548	struct timespec now;
1549	unsigned long seq;
1550
1551	do {
1552		seq = read_seqcount_begin(&timekeeper_seq);
1553
1554		now = tk_xtime(tk);
1555	} while (read_seqcount_retry(&timekeeper_seq, seq));
1556
1557	return now;
1558}
1559EXPORT_SYMBOL(current_kernel_time);
1560
1561struct timespec get_monotonic_coarse(void)
1562{
1563	struct timekeeper *tk = &timekeeper;
1564	struct timespec now, mono;
1565	unsigned long seq;
1566
1567	do {
1568		seq = read_seqcount_begin(&timekeeper_seq);
1569
1570		now = tk_xtime(tk);
1571		mono = tk->wall_to_monotonic;
1572	} while (read_seqcount_retry(&timekeeper_seq, seq));
1573
1574	set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1575				now.tv_nsec + mono.tv_nsec);
1576	return now;
1577}
1578
1579/*
1580 * Must hold jiffies_lock
 
 
1581 */
1582void do_timer(unsigned long ticks)
1583{
1584	jiffies_64 += ticks;
 
1585	calc_global_load(ticks);
1586}
1587
1588/**
1589 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1590 *    and sleep offsets.
1591 * @xtim:	pointer to timespec to be set with xtime
1592 * @wtom:	pointer to timespec to be set with wall_to_monotonic
1593 * @sleep:	pointer to timespec to be set with time in suspend
1594 */
1595void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1596				struct timespec *wtom, struct timespec *sleep)
1597{
1598	struct timekeeper *tk = &timekeeper;
1599	unsigned long seq;
1600
1601	do {
1602		seq = read_seqcount_begin(&timekeeper_seq);
1603		*xtim = tk_xtime(tk);
1604		*wtom = tk->wall_to_monotonic;
1605		*sleep = tk->total_sleep_time;
1606	} while (read_seqcount_retry(&timekeeper_seq, seq));
1607}
1608
1609#ifdef CONFIG_HIGH_RES_TIMERS
1610/**
1611 * ktime_get_update_offsets - hrtimer helper
1612 * @offs_real:	pointer to storage for monotonic -> realtime offset
1613 * @offs_boot:	pointer to storage for monotonic -> boottime offset
1614 * @offs_tai:	pointer to storage for monotonic -> clock tai offset
1615 *
1616 * Returns current monotonic time and updates the offsets
1617 * Called from hrtimer_interrupt() or retrigger_next_event()
1618 */
1619ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot,
1620							ktime_t *offs_tai)
1621{
1622	struct timekeeper *tk = &timekeeper;
1623	ktime_t now;
1624	unsigned int seq;
1625	u64 secs, nsecs;
1626
1627	do {
1628		seq = read_seqcount_begin(&timekeeper_seq);
1629
1630		secs = tk->xtime_sec;
1631		nsecs = timekeeping_get_ns(tk);
1632
1633		*offs_real = tk->offs_real;
1634		*offs_boot = tk->offs_boot;
1635		*offs_tai = tk->offs_tai;
1636	} while (read_seqcount_retry(&timekeeper_seq, seq));
1637
1638	now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1639	now = ktime_sub(now, *offs_real);
1640	return now;
1641}
1642#endif
1643
1644/**
1645 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1646 */
1647ktime_t ktime_get_monotonic_offset(void)
1648{
1649	struct timekeeper *tk = &timekeeper;
1650	unsigned long seq;
1651	struct timespec wtom;
1652
1653	do {
1654		seq = read_seqcount_begin(&timekeeper_seq);
1655		wtom = tk->wall_to_monotonic;
1656	} while (read_seqcount_retry(&timekeeper_seq, seq));
1657
1658	return timespec_to_ktime(wtom);
1659}
1660EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1661
1662/**
1663 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1664 */
1665int do_adjtimex(struct timex *txc)
1666{
1667	struct timekeeper *tk = &timekeeper;
1668	unsigned long flags;
1669	struct timespec ts;
1670	s32 orig_tai, tai;
1671	int ret;
1672
1673	/* Validate the data before disabling interrupts */
1674	ret = ntp_validate_timex(txc);
1675	if (ret)
1676		return ret;
1677
1678	if (txc->modes & ADJ_SETOFFSET) {
1679		struct timespec delta;
1680		delta.tv_sec  = txc->time.tv_sec;
1681		delta.tv_nsec = txc->time.tv_usec;
1682		if (!(txc->modes & ADJ_NANO))
1683			delta.tv_nsec *= 1000;
1684		ret = timekeeping_inject_offset(&delta);
1685		if (ret)
1686			return ret;
1687	}
1688
1689	getnstimeofday(&ts);
1690
1691	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1692	write_seqcount_begin(&timekeeper_seq);
1693
1694	orig_tai = tai = tk->tai_offset;
1695	ret = __do_adjtimex(txc, &ts, &tai);
1696
1697	if (tai != orig_tai) {
1698		__timekeeping_set_tai_offset(tk, tai);
1699		timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1700	}
1701	write_seqcount_end(&timekeeper_seq);
1702	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1703
1704	if (tai != orig_tai)
1705		clock_was_set();
1706
1707	ntp_notify_cmos_timer();
1708
1709	return ret;
1710}
1711
1712#ifdef CONFIG_NTP_PPS
1713/**
1714 * hardpps() - Accessor function to NTP __hardpps function
1715 */
1716void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1717{
1718	unsigned long flags;
1719
1720	raw_spin_lock_irqsave(&timekeeper_lock, flags);
1721	write_seqcount_begin(&timekeeper_seq);
1722
1723	__hardpps(phase_ts, raw_ts);
1724
1725	write_seqcount_end(&timekeeper_seq);
1726	raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1727}
1728EXPORT_SYMBOL(hardpps);
1729#endif
1730
1731/**
1732 * xtime_update() - advances the timekeeping infrastructure
1733 * @ticks:	number of ticks, that have elapsed since the last call.
1734 *
1735 * Must be called with interrupts disabled.
1736 */
1737void xtime_update(unsigned long ticks)
1738{
1739	write_seqlock(&jiffies_lock);
1740	do_timer(ticks);
1741	write_sequnlock(&jiffies_lock);
1742	update_wall_time();
1743}