1// SPDX-License-Identifier: GPL-2.0
   2/*
 
 
   3 *  Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
   4 *  Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
   5 *  Copyright(C) 2006-2007  Timesys Corp., Thomas Gleixner
   6 *
   7 *  High-resolution kernel timers
   8 *
   9 *  In contrast to the low-resolution timeout API, aka timer wheel,
  10 *  hrtimers provide finer resolution and accuracy depending on system
  11 *  configuration and capabilities.
 
 
 
 
 
 
  12 *
  13 *  Started by: Thomas Gleixner and Ingo Molnar
  14 *
  15 *  Credits:
  16 *	Based on the original timer wheel code
  17 *
  18 *	Help, testing, suggestions, bugfixes, improvements were
  19 *	provided by:
  20 *
  21 *	George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
  22 *	et. al.
 
 
  23 */
  24
  25#include <linux/cpu.h>
  26#include <linux/export.h>
  27#include <linux/percpu.h>
  28#include <linux/hrtimer.h>
  29#include <linux/notifier.h>
  30#include <linux/syscalls.h>
  31#include <linux/interrupt.h>
  32#include <linux/tick.h>
 
  33#include <linux/err.h>
  34#include <linux/debugobjects.h>
  35#include <linux/sched/signal.h>
  36#include <linux/sched/sysctl.h>
  37#include <linux/sched/rt.h>
  38#include <linux/sched/deadline.h>
  39#include <linux/sched/nohz.h>
  40#include <linux/sched/debug.h>
  41#include <linux/timer.h>
  42#include <linux/freezer.h>
  43#include <linux/compat.h>
  44
  45#include <linux/uaccess.h>
  46
  47#include <trace/events/timer.h>
  48
  49#include "tick-internal.h"
  50
  51/*
  52 * Masks for selecting the soft and hard context timers from
  53 * cpu_base->active
  54 */
  55#define MASK_SHIFT		(HRTIMER_BASE_MONOTONIC_SOFT)
  56#define HRTIMER_ACTIVE_HARD	((1U << MASK_SHIFT) - 1)
  57#define HRTIMER_ACTIVE_SOFT	(HRTIMER_ACTIVE_HARD << MASK_SHIFT)
  58#define HRTIMER_ACTIVE_ALL	(HRTIMER_ACTIVE_SOFT | HRTIMER_ACTIVE_HARD)
  59
  60/*
  61 * The timer bases:
  62 *
  63 * There are more clockids than hrtimer bases. Thus, we index
  64 * into the timer bases by the hrtimer_base_type enum. When trying
  65 * to reach a base using a clockid, hrtimer_clockid_to_base()
  66 * is used to convert from clockid to the proper hrtimer_base_type.
  67 */
  68DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
  69{
  70	.lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
  71	.clock_base =
  72	{
  73		{
  74			.index = HRTIMER_BASE_MONOTONIC,
  75			.clockid = CLOCK_MONOTONIC,
  76			.get_time = &ktime_get,
  77		},
  78		{
  79			.index = HRTIMER_BASE_REALTIME,
  80			.clockid = CLOCK_REALTIME,
  81			.get_time = &ktime_get_real,
  82		},
  83		{
  84			.index = HRTIMER_BASE_BOOTTIME,
  85			.clockid = CLOCK_BOOTTIME,
  86			.get_time = &ktime_get_boottime,
  87		},
  88		{
  89			.index = HRTIMER_BASE_TAI,
  90			.clockid = CLOCK_TAI,
  91			.get_time = &ktime_get_clocktai,
  92		},
  93		{
  94			.index = HRTIMER_BASE_MONOTONIC_SOFT,
  95			.clockid = CLOCK_MONOTONIC,
  96			.get_time = &ktime_get,
  97		},
  98		{
  99			.index = HRTIMER_BASE_REALTIME_SOFT,
 100			.clockid = CLOCK_REALTIME,
 101			.get_time = &ktime_get_real,
 102		},
 103		{
 104			.index = HRTIMER_BASE_BOOTTIME_SOFT,
 105			.clockid = CLOCK_BOOTTIME,
 106			.get_time = &ktime_get_boottime,
 107		},
 108		{
 109			.index = HRTIMER_BASE_TAI_SOFT,
 110			.clockid = CLOCK_TAI,
 111			.get_time = &ktime_get_clocktai,
 112		},
 113	}
 114};
 115
 116static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
 117	/* Make sure we catch unsupported clockids */
 118	[0 ... MAX_CLOCKS - 1]	= HRTIMER_MAX_CLOCK_BASES,
 119
 120	[CLOCK_REALTIME]	= HRTIMER_BASE_REALTIME,
 121	[CLOCK_MONOTONIC]	= HRTIMER_BASE_MONOTONIC,
 122	[CLOCK_BOOTTIME]	= HRTIMER_BASE_BOOTTIME,
 123	[CLOCK_TAI]		= HRTIMER_BASE_TAI,
 124};
 125
 126/*
 127 * Functions and macros which are different for UP/SMP systems are kept in a
 128 * single place
 129 */
 130#ifdef CONFIG_SMP
 131
 132/*
 133 * We require the migration_base for lock_hrtimer_base()/switch_hrtimer_base()
 134 * such that hrtimer_callback_running() can unconditionally dereference
 135 * timer->base->cpu_base
 136 */
 137static struct hrtimer_cpu_base migration_cpu_base = {
 138	.clock_base = { { .cpu_base = &migration_cpu_base, }, },
 139};
 140
 141#define migration_base	migration_cpu_base.clock_base[0]
 142
 143static inline bool is_migration_base(struct hrtimer_clock_base *base)
 144{
 145	return base == &migration_base;
 146}
 147
 148/*
 149 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
 150 * means that all timers which are tied to this base via timer->base are
 151 * locked, and the base itself is locked too.
 152 *
 153 * So __run_timers/migrate_timers can safely modify all timers which could
 154 * be found on the lists/queues.
 155 *
 156 * When the timer's base is locked, and the timer removed from list, it is
 157 * possible to set timer->base = &migration_base and drop the lock: the timer
 158 * remains locked.
 159 */
 160static
 161struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
 162					     unsigned long *flags)
 163{
 164	struct hrtimer_clock_base *base;
 165
 166	for (;;) {
 167		base = READ_ONCE(timer->base);
 168		if (likely(base != &migration_base)) {
 169			raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 170			if (likely(base == timer->base))
 171				return base;
 172			/* The timer has migrated to another CPU: */
 173			raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
 174		}
 175		cpu_relax();
 176	}
 177}
 178
 179/*
 180 * We do not migrate the timer when it is expiring before the next
 181 * event on the target cpu. When high resolution is enabled, we cannot
 182 * reprogram the target cpu hardware and we would cause it to fire
 183 * late. To keep it simple, we handle the high resolution enabled and
 184 * disabled case similar.
 185 *
 186 * Called with cpu_base->lock of target cpu held.
 187 */
 188static int
 189hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
 190{
 191	ktime_t expires;
 192
 193	expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
 194	return expires < new_base->cpu_base->expires_next;
 195}
 196
 197static inline
 198struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
 199					 int pinned)
 200{
 201#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
 202	if (static_branch_likely(&timers_migration_enabled) && !pinned)
 203		return &per_cpu(hrtimer_bases, get_nohz_timer_target());
 204#endif
 205	return base;
 206}
 207
 208/*
 209 * We switch the timer base to a power-optimized selected CPU target,
 210 * if:
 211 *	- NO_HZ_COMMON is enabled
 212 *	- timer migration is enabled
 213 *	- the timer callback is not running
 214 *	- the timer is not the first expiring timer on the new target
 215 *
 216 * If one of the above requirements is not fulfilled we move the timer
 217 * to the current CPU or leave it on the previously assigned CPU if
 218 * the timer callback is currently running.
 219 */
 220static inline struct hrtimer_clock_base *
 221switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 222		    int pinned)
 223{
 224	struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
 225	struct hrtimer_clock_base *new_base;
 226	int basenum = base->index;
 227
 228	this_cpu_base = this_cpu_ptr(&hrtimer_bases);
 229	new_cpu_base = get_target_base(this_cpu_base, pinned);
 230again:
 231	new_base = &new_cpu_base->clock_base[basenum];
 232
 233	if (base != new_base) {
 234		/*
 235		 * We are trying to move timer to new_base.
 236		 * However we can't change timer's base while it is running,
 237		 * so we keep it on the same CPU. No hassle vs. reprogramming
 238		 * the event source in the high resolution case. The softirq
 239		 * code will take care of this when the timer function has
 240		 * completed. There is no conflict as we hold the lock until
 241		 * the timer is enqueued.
 242		 */
 243		if (unlikely(hrtimer_callback_running(timer)))
 244			return base;
 245
 246		/* See the comment in lock_hrtimer_base() */
 247		WRITE_ONCE(timer->base, &migration_base);
 248		raw_spin_unlock(&base->cpu_base->lock);
 249		raw_spin_lock(&new_base->cpu_base->lock);
 250
 251		if (new_cpu_base != this_cpu_base &&
 252		    hrtimer_check_target(timer, new_base)) {
 253			raw_spin_unlock(&new_base->cpu_base->lock);
 254			raw_spin_lock(&base->cpu_base->lock);
 255			new_cpu_base = this_cpu_base;
 256			WRITE_ONCE(timer->base, base);
 257			goto again;
 258		}
 259		WRITE_ONCE(timer->base, new_base);
 260	} else {
 261		if (new_cpu_base != this_cpu_base &&
 262		    hrtimer_check_target(timer, new_base)) {
 263			new_cpu_base = this_cpu_base;
 264			goto again;
 265		}
 266	}
 267	return new_base;
 268}
 269
 270#else /* CONFIG_SMP */
 271
 272static inline bool is_migration_base(struct hrtimer_clock_base *base)
 273{
 274	return false;
 275}
 276
 277static inline struct hrtimer_clock_base *
 278lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 279{
 280	struct hrtimer_clock_base *base = timer->base;
 281
 282	raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 283
 284	return base;
 285}
 286
 287# define switch_hrtimer_base(t, b, p)	(b)
 288
 289#endif	/* !CONFIG_SMP */
 290
 291/*
 292 * Functions for the union type storage format of ktime_t which are
 293 * too large for inlining:
 294 */
 295#if BITS_PER_LONG < 64
 296/*
 297 * Divide a ktime value by a nanosecond value
 298 */
 299s64 __ktime_divns(const ktime_t kt, s64 div)
 300{
 301	int sft = 0;
 302	s64 dclc;
 303	u64 tmp;
 304
 305	dclc = ktime_to_ns(kt);
 306	tmp = dclc < 0 ? -dclc : dclc;
 307
 308	/* Make sure the divisor is less than 2^32: */
 309	while (div >> 32) {
 310		sft++;
 311		div >>= 1;
 312	}
 313	tmp >>= sft;
 314	do_div(tmp, (unsigned long) div);
 315	return dclc < 0 ? -tmp : tmp;
 316}
 317EXPORT_SYMBOL_GPL(__ktime_divns);
 318#endif /* BITS_PER_LONG >= 64 */
 319
 320/*
 321 * Add two ktime values and do a safety check for overflow:
 322 */
 323ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
 324{
 325	ktime_t res = ktime_add_unsafe(lhs, rhs);
 326
 327	/*
 328	 * We use KTIME_SEC_MAX here, the maximum timeout which we can
 329	 * return to user space in a timespec:
 330	 */
 331	if (res < 0 || res < lhs || res < rhs)
 332		res = ktime_set(KTIME_SEC_MAX, 0);
 333
 334	return res;
 335}
 336
 337EXPORT_SYMBOL_GPL(ktime_add_safe);
 338
 339#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 340
 341static struct debug_obj_descr hrtimer_debug_descr;
 342
 343static void *hrtimer_debug_hint(void *addr)
 344{
 345	return ((struct hrtimer *) addr)->function;
 346}
 347
 348/*
 349 * fixup_init is called when:
 350 * - an active object is initialized
 351 */
 352static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
 353{
 354	struct hrtimer *timer = addr;
 355
 356	switch (state) {
 357	case ODEBUG_STATE_ACTIVE:
 358		hrtimer_cancel(timer);
 359		debug_object_init(timer, &hrtimer_debug_descr);
 360		return true;
 361	default:
 362		return false;
 363	}
 364}
 365
 366/*
 367 * fixup_activate is called when:
 368 * - an active object is activated
 369 * - an unknown non-static object is activated
 370 */
 371static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
 372{
 373	switch (state) {
 374	case ODEBUG_STATE_ACTIVE:
 375		WARN_ON(1);
 376		/* fall through */
 377	default:
 378		return false;
 379	}
 380}
 381
 382/*
 383 * fixup_free is called when:
 384 * - an active object is freed
 385 */
 386static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
 387{
 388	struct hrtimer *timer = addr;
 389
 390	switch (state) {
 391	case ODEBUG_STATE_ACTIVE:
 392		hrtimer_cancel(timer);
 393		debug_object_free(timer, &hrtimer_debug_descr);
 394		return true;
 395	default:
 396		return false;
 397	}
 398}
 399
 400static struct debug_obj_descr hrtimer_debug_descr = {
 401	.name		= "hrtimer",
 402	.debug_hint	= hrtimer_debug_hint,
 403	.fixup_init	= hrtimer_fixup_init,
 404	.fixup_activate	= hrtimer_fixup_activate,
 405	.fixup_free	= hrtimer_fixup_free,
 406};
 407
 408static inline void debug_hrtimer_init(struct hrtimer *timer)
 409{
 410	debug_object_init(timer, &hrtimer_debug_descr);
 411}
 412
 413static inline void debug_hrtimer_activate(struct hrtimer *timer,
 414					  enum hrtimer_mode mode)
 415{
 416	debug_object_activate(timer, &hrtimer_debug_descr);
 417}
 418
 419static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
 420{
 421	debug_object_deactivate(timer, &hrtimer_debug_descr);
 422}
 423
 424static inline void debug_hrtimer_free(struct hrtimer *timer)
 425{
 426	debug_object_free(timer, &hrtimer_debug_descr);
 427}
 428
 429static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
 430			   enum hrtimer_mode mode);
 431
 432void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
 433			   enum hrtimer_mode mode)
 434{
 435	debug_object_init_on_stack(timer, &hrtimer_debug_descr);
 436	__hrtimer_init(timer, clock_id, mode);
 437}
 438EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
 439
 440static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
 441				   clockid_t clock_id, enum hrtimer_mode mode);
 442
 443void hrtimer_init_sleeper_on_stack(struct hrtimer_sleeper *sl,
 444				   clockid_t clock_id, enum hrtimer_mode mode)
 445{
 446	debug_object_init_on_stack(&sl->timer, &hrtimer_debug_descr);
 447	__hrtimer_init_sleeper(sl, clock_id, mode);
 448}
 449EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack);
 450
 451void destroy_hrtimer_on_stack(struct hrtimer *timer)
 452{
 453	debug_object_free(timer, &hrtimer_debug_descr);
 454}
 455EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
 456
 457#else
 458
 459static inline void debug_hrtimer_init(struct hrtimer *timer) { }
 460static inline void debug_hrtimer_activate(struct hrtimer *timer,
 461					  enum hrtimer_mode mode) { }
 462static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
 463#endif
 464
 465static inline void
 466debug_init(struct hrtimer *timer, clockid_t clockid,
 467	   enum hrtimer_mode mode)
 468{
 469	debug_hrtimer_init(timer);
 470	trace_hrtimer_init(timer, clockid, mode);
 471}
 472
 473static inline void debug_activate(struct hrtimer *timer,
 474				  enum hrtimer_mode mode)
 475{
 476	debug_hrtimer_activate(timer, mode);
 477	trace_hrtimer_start(timer, mode);
 478}
 479
 480static inline void debug_deactivate(struct hrtimer *timer)
 481{
 482	debug_hrtimer_deactivate(timer);
 483	trace_hrtimer_cancel(timer);
 484}
 485
 486static struct hrtimer_clock_base *
 487__next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
 488{
 489	unsigned int idx;
 490
 491	if (!*active)
 492		return NULL;
 493
 494	idx = __ffs(*active);
 495	*active &= ~(1U << idx);
 496
 497	return &cpu_base->clock_base[idx];
 498}
 499
 500#define for_each_active_base(base, cpu_base, active)	\
 501	while ((base = __next_base((cpu_base), &(active))))
 502
 503static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
 504					 const struct hrtimer *exclude,
 505					 unsigned int active,
 506					 ktime_t expires_next)
 507{
 508	struct hrtimer_clock_base *base;
 509	ktime_t expires;
 510
 511	for_each_active_base(base, cpu_base, active) {
 512		struct timerqueue_node *next;
 513		struct hrtimer *timer;
 514
 515		next = timerqueue_getnext(&base->active);
 516		timer = container_of(next, struct hrtimer, node);
 517		if (timer == exclude) {
 518			/* Get to the next timer in the queue. */
 519			next = timerqueue_iterate_next(next);
 520			if (!next)
 521				continue;
 522
 523			timer = container_of(next, struct hrtimer, node);
 524		}
 525		expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 526		if (expires < expires_next) {
 527			expires_next = expires;
 528
 529			/* Skip cpu_base update if a timer is being excluded. */
 530			if (exclude)
 531				continue;
 532
 533			if (timer->is_soft)
 534				cpu_base->softirq_next_timer = timer;
 535			else
 536				cpu_base->next_timer = timer;
 537		}
 538	}
 539	/*
 540	 * clock_was_set() might have changed base->offset of any of
 541	 * the clock bases so the result might be negative. Fix it up
 542	 * to prevent a false positive in clockevents_program_event().
 543	 */
 544	if (expires_next < 0)
 545		expires_next = 0;
 546	return expires_next;
 547}
 548
 549/*
 550 * Recomputes cpu_base::*next_timer and returns the earliest expires_next but
 551 * does not set cpu_base::*expires_next, that is done by hrtimer_reprogram.
 552 *
 553 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
 554 * those timers will get run whenever the softirq gets handled, at the end of
 555 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
 556 *
 557 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
 558 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
 559 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
 560 *
 561 * @active_mask must be one of:
 562 *  - HRTIMER_ACTIVE_ALL,
 563 *  - HRTIMER_ACTIVE_SOFT, or
 564 *  - HRTIMER_ACTIVE_HARD.
 565 */
 566static ktime_t
 567__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
 568{
 569	unsigned int active;
 570	struct hrtimer *next_timer = NULL;
 571	ktime_t expires_next = KTIME_MAX;
 572
 573	if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
 574		active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
 575		cpu_base->softirq_next_timer = NULL;
 576		expires_next = __hrtimer_next_event_base(cpu_base, NULL,
 577							 active, KTIME_MAX);
 578
 579		next_timer = cpu_base->softirq_next_timer;
 580	}
 581
 582	if (active_mask & HRTIMER_ACTIVE_HARD) {
 583		active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
 584		cpu_base->next_timer = next_timer;
 585		expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
 586							 expires_next);
 587	}
 588
 589	return expires_next;
 590}
 591
 592static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
 593{
 594	ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
 595	ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
 596	ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
 597
 598	ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
 599					    offs_real, offs_boot, offs_tai);
 600
 601	base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
 602	base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
 603	base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
 604
 605	return now;
 606}
 607
 608/*
 609 * Is the high resolution mode active ?
 610 */
 611static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
 612{
 613	return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
 614		cpu_base->hres_active : 0;
 615}
 616
 617static inline int hrtimer_hres_active(void)
 618{
 619	return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
 620}
 621
 622/*
 623 * Reprogram the event source with checking both queues for the
 624 * next event
 625 * Called with interrupts disabled and base->lock held
 626 */
 627static void
 628hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
 629{
 630	ktime_t expires_next;
 631
 632	/*
 633	 * Find the current next expiration time.
 634	 */
 635	expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
 636
 637	if (cpu_base->next_timer && cpu_base->next_timer->is_soft) {
 638		/*
 639		 * When the softirq is activated, hrtimer has to be
 640		 * programmed with the first hard hrtimer because soft
 641		 * timer interrupt could occur too late.
 642		 */
 643		if (cpu_base->softirq_activated)
 644			expires_next = __hrtimer_get_next_event(cpu_base,
 645								HRTIMER_ACTIVE_HARD);
 646		else
 647			cpu_base->softirq_expires_next = expires_next;
 648	}
 649
 650	if (skip_equal && expires_next == cpu_base->expires_next)
 651		return;
 652
 653	cpu_base->expires_next = expires_next;
 654
 655	/*
 656	 * If hres is not active, hardware does not have to be
 657	 * reprogrammed yet.
 658	 *
 659	 * If a hang was detected in the last timer interrupt then we
 660	 * leave the hang delay active in the hardware. We want the
 661	 * system to make progress. That also prevents the following
 662	 * scenario:
 663	 * T1 expires 50ms from now
 664	 * T2 expires 5s from now
 665	 *
 666	 * T1 is removed, so this code is called and would reprogram
 667	 * the hardware to 5s from now. Any hrtimer_start after that
 668	 * will not reprogram the hardware due to hang_detected being
 669	 * set. So we'd effectivly block all timers until the T2 event
 670	 * fires.
 671	 */
 672	if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
 673		return;
 674
 675	tick_program_event(cpu_base->expires_next, 1);
 676}
 677
 678/* High resolution timer related functions */
 679#ifdef CONFIG_HIGH_RES_TIMERS
 680
 681/*
 682 * High resolution timer enabled ?
 683 */
 684static bool hrtimer_hres_enabled __read_mostly  = true;
 685unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
 686EXPORT_SYMBOL_GPL(hrtimer_resolution);
 687
 688/*
 689 * Enable / Disable high resolution mode
 690 */
 691static int __init setup_hrtimer_hres(char *str)
 692{
 693	return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
 694}
 695
 696__setup("highres=", setup_hrtimer_hres);
 697
 698/*
 699 * hrtimer_high_res_enabled - query, if the highres mode is enabled
 700 */
 701static inline int hrtimer_is_hres_enabled(void)
 702{
 703	return hrtimer_hres_enabled;
 704}
 705
 706/*
 707 * Retrigger next event is called after clock was set
 708 *
 709 * Called with interrupts disabled via on_each_cpu()
 710 */
 711static void retrigger_next_event(void *arg)
 712{
 713	struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
 714
 715	if (!__hrtimer_hres_active(base))
 716		return;
 717
 718	raw_spin_lock(&base->lock);
 719	hrtimer_update_base(base);
 720	hrtimer_force_reprogram(base, 0);
 721	raw_spin_unlock(&base->lock);
 722}
 723
 724/*
 725 * Switch to high resolution mode
 726 */
 727static void hrtimer_switch_to_hres(void)
 728{
 729	struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
 730
 731	if (tick_init_highres()) {
 732		pr_warn("Could not switch to high resolution mode on CPU %u\n",
 733			base->cpu);
 734		return;
 735	}
 736	base->hres_active = 1;
 737	hrtimer_resolution = HIGH_RES_NSEC;
 738
 739	tick_setup_sched_timer();
 740	/* "Retrigger" the interrupt to get things going */
 741	retrigger_next_event(NULL);
 742}
 743
 744static void clock_was_set_work(struct work_struct *work)
 745{
 746	clock_was_set();
 747}
 748
 749static DECLARE_WORK(hrtimer_work, clock_was_set_work);
 750
 751/*
 752 * Called from timekeeping and resume code to reprogram the hrtimer
 753 * interrupt device on all cpus.
 754 */
 755void clock_was_set_delayed(void)
 756{
 757	schedule_work(&hrtimer_work);
 758}
 759
 760#else
 761
 762static inline int hrtimer_is_hres_enabled(void) { return 0; }
 763static inline void hrtimer_switch_to_hres(void) { }
 764static inline void retrigger_next_event(void *arg) { }
 765
 766#endif /* CONFIG_HIGH_RES_TIMERS */
 767
 768/*
 769 * When a timer is enqueued and expires earlier than the already enqueued
 770 * timers, we have to check, whether it expires earlier than the timer for
 771 * which the clock event device was armed.
 772 *
 773 * Called with interrupts disabled and base->cpu_base.lock held
 774 */
 775static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
 776{
 777	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
 778	struct hrtimer_clock_base *base = timer->base;
 779	ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 780
 781	WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
 782
 783	/*
 784	 * CLOCK_REALTIME timer might be requested with an absolute
 785	 * expiry time which is less than base->offset. Set it to 0.
 786	 */
 787	if (expires < 0)
 788		expires = 0;
 789
 790	if (timer->is_soft) {
 791		/*
 792		 * soft hrtimer could be started on a remote CPU. In this
 793		 * case softirq_expires_next needs to be updated on the
 794		 * remote CPU. The soft hrtimer will not expire before the
 795		 * first hard hrtimer on the remote CPU -
 796		 * hrtimer_check_target() prevents this case.
 797		 */
 798		struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
 799
 800		if (timer_cpu_base->softirq_activated)
 801			return;
 802
 803		if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
 804			return;
 805
 806		timer_cpu_base->softirq_next_timer = timer;
 807		timer_cpu_base->softirq_expires_next = expires;
 808
 809		if (!ktime_before(expires, timer_cpu_base->expires_next) ||
 810		    !reprogram)
 811			return;
 812	}
 813
 814	/*
 815	 * If the timer is not on the current cpu, we cannot reprogram
 816	 * the other cpus clock event device.
 817	 */
 818	if (base->cpu_base != cpu_base)
 819		return;
 820
 821	/*
 822	 * If the hrtimer interrupt is running, then it will
 823	 * reevaluate the clock bases and reprogram the clock event
 824	 * device. The callbacks are always executed in hard interrupt
 825	 * context so we don't need an extra check for a running
 826	 * callback.
 827	 */
 828	if (cpu_base->in_hrtirq)
 829		return;
 830
 831	if (expires >= cpu_base->expires_next)
 832		return;
 833
 834	/* Update the pointer to the next expiring timer */
 835	cpu_base->next_timer = timer;
 836	cpu_base->expires_next = expires;
 837
 838	/*
 839	 * If hres is not active, hardware does not have to be
 840	 * programmed yet.
 841	 *
 842	 * If a hang was detected in the last timer interrupt then we
 843	 * do not schedule a timer which is earlier than the expiry
 844	 * which we enforced in the hang detection. We want the system
 845	 * to make progress.
 846	 */
 847	if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
 848		return;
 849
 850	/*
 851	 * Program the timer hardware. We enforce the expiry for
 852	 * events which are already in the past.
 853	 */
 854	tick_program_event(expires, 1);
 855}
 856
 857/*
 858 * Clock realtime was set
 859 *
 860 * Change the offset of the realtime clock vs. the monotonic
 861 * clock.
 862 *
 863 * We might have to reprogram the high resolution timer interrupt. On
 864 * SMP we call the architecture specific code to retrigger _all_ high
 865 * resolution timer interrupts. On UP we just disable interrupts and
 866 * call the high resolution interrupt code.
 867 */
 868void clock_was_set(void)
 869{
 870#ifdef CONFIG_HIGH_RES_TIMERS
 871	/* Retrigger the CPU local events everywhere */
 872	on_each_cpu(retrigger_next_event, NULL, 1);
 873#endif
 874	timerfd_clock_was_set();
 875}
 876
 877/*
 878 * During resume we might have to reprogram the high resolution timer
 879 * interrupt on all online CPUs.  However, all other CPUs will be
 880 * stopped with IRQs interrupts disabled so the clock_was_set() call
 881 * must be deferred.
 882 */
 883void hrtimers_resume(void)
 884{
 885	lockdep_assert_irqs_disabled();
 886	/* Retrigger on the local CPU */
 887	retrigger_next_event(NULL);
 888	/* And schedule a retrigger for all others */
 889	clock_was_set_delayed();
 890}
 891
 892/*
 893 * Counterpart to lock_hrtimer_base above:
 894 */
 895static inline
 896void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 897{
 898	raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
 899}
 900
 901/**
 902 * hrtimer_forward - forward the timer expiry
 903 * @timer:	hrtimer to forward
 904 * @now:	forward past this time
 905 * @interval:	the interval to forward
 906 *
 907 * Forward the timer expiry so it will expire in the future.
 908 * Returns the number of overruns.
 909 *
 910 * Can be safely called from the callback function of @timer. If
 911 * called from other contexts @timer must neither be enqueued nor
 912 * running the callback and the caller needs to take care of
 913 * serialization.
 914 *
 915 * Note: This only updates the timer expiry value and does not requeue
 916 * the timer.
 917 */
 918u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
 919{
 920	u64 orun = 1;
 921	ktime_t delta;
 922
 923	delta = ktime_sub(now, hrtimer_get_expires(timer));
 924
 925	if (delta < 0)
 926		return 0;
 927
 928	if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
 929		return 0;
 930
 931	if (interval < hrtimer_resolution)
 932		interval = hrtimer_resolution;
 933
 934	if (unlikely(delta >= interval)) {
 935		s64 incr = ktime_to_ns(interval);
 936
 937		orun = ktime_divns(delta, incr);
 938		hrtimer_add_expires_ns(timer, incr * orun);
 939		if (hrtimer_get_expires_tv64(timer) > now)
 940			return orun;
 941		/*
 942		 * This (and the ktime_add() below) is the
 943		 * correction for exact:
 944		 */
 945		orun++;
 946	}
 947	hrtimer_add_expires(timer, interval);
 948
 949	return orun;
 950}
 951EXPORT_SYMBOL_GPL(hrtimer_forward);
 952
 953/*
 954 * enqueue_hrtimer - internal function to (re)start a timer
 955 *
 956 * The timer is inserted in expiry order. Insertion into the
 957 * red black tree is O(log(n)). Must hold the base lock.
 958 *
 959 * Returns 1 when the new timer is the leftmost timer in the tree.
 960 */
 961static int enqueue_hrtimer(struct hrtimer *timer,
 962			   struct hrtimer_clock_base *base,
 963			   enum hrtimer_mode mode)
 964{
 965	debug_activate(timer, mode);
 966
 967	base->cpu_base->active_bases |= 1 << base->index;
 968
 969	timer->state = HRTIMER_STATE_ENQUEUED;
 970
 971	return timerqueue_add(&base->active, &timer->node);
 972}
 973
 974/*
 975 * __remove_hrtimer - internal function to remove a timer
 976 *
 977 * Caller must hold the base lock.
 978 *
 979 * High resolution timer mode reprograms the clock event device when the
 980 * timer is the one which expires next. The caller can disable this by setting
 981 * reprogram to zero. This is useful, when the context does a reprogramming
 982 * anyway (e.g. timer interrupt)
 983 */
 984static void __remove_hrtimer(struct hrtimer *timer,
 985			     struct hrtimer_clock_base *base,
 986			     u8 newstate, int reprogram)
 987{
 988	struct hrtimer_cpu_base *cpu_base = base->cpu_base;
 989	u8 state = timer->state;
 990
 991	timer->state = newstate;
 992	if (!(state & HRTIMER_STATE_ENQUEUED))
 993		return;
 994
 995	if (!timerqueue_del(&base->active, &timer->node))
 996		cpu_base->active_bases &= ~(1 << base->index);
 997
 998	/*
 999	 * Note: If reprogram is false we do not update
1000	 * cpu_base->next_timer. This happens when we remove the first
1001	 * timer on a remote cpu. No harm as we never dereference
1002	 * cpu_base->next_timer. So the worst thing what can happen is
1003	 * an superflous call to hrtimer_force_reprogram() on the
1004	 * remote cpu later on if the same timer gets enqueued again.
1005	 */
1006	if (reprogram && timer == cpu_base->next_timer)
1007		hrtimer_force_reprogram(cpu_base, 1);
1008}
1009
1010/*
1011 * remove hrtimer, called with base lock held
1012 */
1013static inline int
1014remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
1015{
1016	if (hrtimer_is_queued(timer)) {
1017		u8 state = timer->state;
1018		int reprogram;
1019
1020		/*
1021		 * Remove the timer and force reprogramming when high
1022		 * resolution mode is active and the timer is on the current
1023		 * CPU. If we remove a timer on another CPU, reprogramming is
1024		 * skipped. The interrupt event on this CPU is fired and
1025		 * reprogramming happens in the interrupt handler. This is a
1026		 * rare case and less expensive than a smp call.
1027		 */
1028		debug_deactivate(timer);
1029		reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1030
1031		if (!restart)
1032			state = HRTIMER_STATE_INACTIVE;
1033
1034		__remove_hrtimer(timer, base, state, reprogram);
1035		return 1;
1036	}
1037	return 0;
1038}
1039
1040static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1041					    const enum hrtimer_mode mode)
1042{
1043#ifdef CONFIG_TIME_LOW_RES
1044	/*
1045	 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1046	 * granular time values. For relative timers we add hrtimer_resolution
1047	 * (i.e. one jiffie) to prevent short timeouts.
1048	 */
1049	timer->is_rel = mode & HRTIMER_MODE_REL;
1050	if (timer->is_rel)
1051		tim = ktime_add_safe(tim, hrtimer_resolution);
1052#endif
1053	return tim;
1054}
1055
1056static void
1057hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1058{
1059	ktime_t expires;
1060
1061	/*
1062	 * Find the next SOFT expiration.
1063	 */
1064	expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1065
1066	/*
1067	 * reprogramming needs to be triggered, even if the next soft
1068	 * hrtimer expires at the same time than the next hard
1069	 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1070	 */
1071	if (expires == KTIME_MAX)
1072		return;
1073
1074	/*
1075	 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1076	 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1077	 */
1078	hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1079}
1080
1081static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1082				    u64 delta_ns, const enum hrtimer_mode mode,
1083				    struct hrtimer_clock_base *base)
1084{
1085	struct hrtimer_clock_base *new_base;
1086
1087	/* Remove an active timer from the queue: */
1088	remove_hrtimer(timer, base, true);
1089
1090	if (mode & HRTIMER_MODE_REL)
1091		tim = ktime_add_safe(tim, base->get_time());
1092
1093	tim = hrtimer_update_lowres(timer, tim, mode);
1094
1095	hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1096
1097	/* Switch the timer base, if necessary: */
1098	new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1099
1100	return enqueue_hrtimer(timer, new_base, mode);
1101}
1102
1103/**
1104 * hrtimer_start_range_ns - (re)start an hrtimer
1105 * @timer:	the timer to be added
1106 * @tim:	expiry time
1107 * @delta_ns:	"slack" range for the timer
1108 * @mode:	timer mode: absolute (HRTIMER_MODE_ABS) or
1109 *		relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1110 *		softirq based mode is considered for debug purpose only!
1111 */
1112void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1113			    u64 delta_ns, const enum hrtimer_mode mode)
1114{
1115	struct hrtimer_clock_base *base;
1116	unsigned long flags;
1117
1118	/*
1119	 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1120	 * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1121	 * expiry mode because unmarked timers are moved to softirq expiry.
1122	 */
1123	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1124		WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1125	else
1126		WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1127
1128	base = lock_hrtimer_base(timer, &flags);
1129
1130	if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1131		hrtimer_reprogram(timer, true);
1132
1133	unlock_hrtimer_base(timer, &flags);
1134}
1135EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1136
1137/**
1138 * hrtimer_try_to_cancel - try to deactivate a timer
1139 * @timer:	hrtimer to stop
1140 *
1141 * Returns:
1142 *
1143 *  *  0 when the timer was not active
1144 *  *  1 when the timer was active
1145 *  * -1 when the timer is currently executing the callback function and
1146 *    cannot be stopped
1147 */
1148int hrtimer_try_to_cancel(struct hrtimer *timer)
1149{
1150	struct hrtimer_clock_base *base;
1151	unsigned long flags;
1152	int ret = -1;
1153
1154	/*
1155	 * Check lockless first. If the timer is not active (neither
1156	 * enqueued nor running the callback, nothing to do here.  The
1157	 * base lock does not serialize against a concurrent enqueue,
1158	 * so we can avoid taking it.
1159	 */
1160	if (!hrtimer_active(timer))
1161		return 0;
1162
1163	base = lock_hrtimer_base(timer, &flags);
1164
1165	if (!hrtimer_callback_running(timer))
1166		ret = remove_hrtimer(timer, base, false);
1167
1168	unlock_hrtimer_base(timer, &flags);
1169
1170	return ret;
1171
1172}
1173EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1174
1175#ifdef CONFIG_PREEMPT_RT
1176static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1177{
1178	spin_lock_init(&base->softirq_expiry_lock);
1179}
1180
1181static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1182{
1183	spin_lock(&base->softirq_expiry_lock);
1184}
1185
1186static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1187{
1188	spin_unlock(&base->softirq_expiry_lock);
1189}
1190
1191/*
1192 * The counterpart to hrtimer_cancel_wait_running().
1193 *
1194 * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1195 * the timer callback to finish. Drop expiry_lock and reaquire it. That
1196 * allows the waiter to acquire the lock and make progress.
1197 */
1198static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1199				      unsigned long flags)
1200{
1201	if (atomic_read(&cpu_base->timer_waiters)) {
1202		raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1203		spin_unlock(&cpu_base->softirq_expiry_lock);
1204		spin_lock(&cpu_base->softirq_expiry_lock);
1205		raw_spin_lock_irq(&cpu_base->lock);
1206	}
1207}
1208
1209/*
1210 * This function is called on PREEMPT_RT kernels when the fast path
1211 * deletion of a timer failed because the timer callback function was
1212 * running.
1213 *
1214 * This prevents priority inversion: if the soft irq thread is preempted
1215 * in the middle of a timer callback, then calling del_timer_sync() can
1216 * lead to two issues:
1217 *
1218 *  - If the caller is on a remote CPU then it has to spin wait for the timer
1219 *    handler to complete. This can result in unbound priority inversion.
1220 *
1221 *  - If the caller originates from the task which preempted the timer
1222 *    handler on the same CPU, then spin waiting for the timer handler to
1223 *    complete is never going to end.
1224 */
1225void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1226{
1227	/* Lockless read. Prevent the compiler from reloading it below */
1228	struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1229
1230	/*
1231	 * Just relax if the timer expires in hard interrupt context or if
1232	 * it is currently on the migration base.
1233	 */
1234	if (!timer->is_soft || is_migration_base(base)) {
1235		cpu_relax();
1236		return;
1237	}
1238
1239	/*
1240	 * Mark the base as contended and grab the expiry lock, which is
1241	 * held by the softirq across the timer callback. Drop the lock
1242	 * immediately so the softirq can expire the next timer. In theory
1243	 * the timer could already be running again, but that's more than
1244	 * unlikely and just causes another wait loop.
1245	 */
1246	atomic_inc(&base->cpu_base->timer_waiters);
1247	spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1248	atomic_dec(&base->cpu_base->timer_waiters);
1249	spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1250}
1251#else
1252static inline void
1253hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1254static inline void
1255hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1256static inline void
1257hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1258static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1259					     unsigned long flags) { }
1260#endif
1261
1262/**
1263 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1264 * @timer:	the timer to be cancelled
1265 *
1266 * Returns:
1267 *  0 when the timer was not active
1268 *  1 when the timer was active
1269 */
1270int hrtimer_cancel(struct hrtimer *timer)
1271{
1272	int ret;
1273
1274	do {
1275		ret = hrtimer_try_to_cancel(timer);
1276
1277		if (ret < 0)
1278			hrtimer_cancel_wait_running(timer);
1279	} while (ret < 0);
1280	return ret;
1281}
1282EXPORT_SYMBOL_GPL(hrtimer_cancel);
1283
1284/**
1285 * hrtimer_get_remaining - get remaining time for the timer
1286 * @timer:	the timer to read
1287 * @adjust:	adjust relative timers when CONFIG_TIME_LOW_RES=y
1288 */
1289ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1290{
1291	unsigned long flags;
1292	ktime_t rem;
1293
1294	lock_hrtimer_base(timer, &flags);
1295	if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1296		rem = hrtimer_expires_remaining_adjusted(timer);
1297	else
1298		rem = hrtimer_expires_remaining(timer);
1299	unlock_hrtimer_base(timer, &flags);
1300
1301	return rem;
1302}
1303EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1304
1305#ifdef CONFIG_NO_HZ_COMMON
1306/**
1307 * hrtimer_get_next_event - get the time until next expiry event
1308 *
1309 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1310 */
1311u64 hrtimer_get_next_event(void)
1312{
1313	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1314	u64 expires = KTIME_MAX;
1315	unsigned long flags;
1316
1317	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1318
1319	if (!__hrtimer_hres_active(cpu_base))
1320		expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1321
1322	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1323
1324	return expires;
1325}
1326
1327/**
1328 * hrtimer_next_event_without - time until next expiry event w/o one timer
1329 * @exclude:	timer to exclude
1330 *
1331 * Returns the next expiry time over all timers except for the @exclude one or
1332 * KTIME_MAX if none of them is pending.
1333 */
1334u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1335{
1336	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1337	u64 expires = KTIME_MAX;
1338	unsigned long flags;
1339
1340	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1341
1342	if (__hrtimer_hres_active(cpu_base)) {
1343		unsigned int active;
1344
1345		if (!cpu_base->softirq_activated) {
1346			active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1347			expires = __hrtimer_next_event_base(cpu_base, exclude,
1348							    active, KTIME_MAX);
1349		}
1350		active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1351		expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1352						    expires);
1353	}
1354
1355	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1356
1357	return expires;
1358}
1359#endif
1360
1361static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1362{
1363	if (likely(clock_id < MAX_CLOCKS)) {
1364		int base = hrtimer_clock_to_base_table[clock_id];
1365
1366		if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1367			return base;
1368	}
1369	WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1370	return HRTIMER_BASE_MONOTONIC;
1371}
1372
1373static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1374			   enum hrtimer_mode mode)
1375{
1376	bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
 
1377	struct hrtimer_cpu_base *cpu_base;
1378	int base;
1379
1380	/*
1381	 * On PREEMPT_RT enabled kernels hrtimers which are not explicitely
1382	 * marked for hard interrupt expiry mode are moved into soft
1383	 * interrupt context for latency reasons and because the callbacks
1384	 * can invoke functions which might sleep on RT, e.g. spin_lock().
1385	 */
1386	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1387		softtimer = true;
1388
1389	memset(timer, 0, sizeof(struct hrtimer));
1390
1391	cpu_base = raw_cpu_ptr(&hrtimer_bases);
1392
1393	/*
1394	 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1395	 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1396	 * ensure POSIX compliance.
1397	 */
1398	if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1399		clock_id = CLOCK_MONOTONIC;
1400
1401	base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1402	base += hrtimer_clockid_to_base(clock_id);
1403	timer->is_soft = softtimer;
1404	timer->is_hard = !softtimer;
1405	timer->base = &cpu_base->clock_base[base];
1406	timerqueue_init(&timer->node);
1407}
1408
1409/**
1410 * hrtimer_init - initialize a timer to the given clock
1411 * @timer:	the timer to be initialized
1412 * @clock_id:	the clock to be used
1413 * @mode:       The modes which are relevant for intitialization:
1414 *              HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1415 *              HRTIMER_MODE_REL_SOFT
1416 *
1417 *              The PINNED variants of the above can be handed in,
1418 *              but the PINNED bit is ignored as pinning happens
1419 *              when the hrtimer is started
1420 */
1421void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1422		  enum hrtimer_mode mode)
1423{
1424	debug_init(timer, clock_id, mode);
1425	__hrtimer_init(timer, clock_id, mode);
1426}
1427EXPORT_SYMBOL_GPL(hrtimer_init);
1428
1429/*
1430 * A timer is active, when it is enqueued into the rbtree or the
1431 * callback function is running or it's in the state of being migrated
1432 * to another cpu.
1433 *
1434 * It is important for this function to not return a false negative.
1435 */
1436bool hrtimer_active(const struct hrtimer *timer)
1437{
1438	struct hrtimer_clock_base *base;
1439	unsigned int seq;
1440
1441	do {
1442		base = READ_ONCE(timer->base);
1443		seq = raw_read_seqcount_begin(&base->seq);
1444
1445		if (timer->state != HRTIMER_STATE_INACTIVE ||
1446		    base->running == timer)
1447			return true;
1448
1449	} while (read_seqcount_retry(&base->seq, seq) ||
1450		 base != READ_ONCE(timer->base));
1451
1452	return false;
1453}
1454EXPORT_SYMBOL_GPL(hrtimer_active);
1455
1456/*
1457 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1458 * distinct sections:
1459 *
1460 *  - queued:	the timer is queued
1461 *  - callback:	the timer is being ran
1462 *  - post:	the timer is inactive or (re)queued
1463 *
1464 * On the read side we ensure we observe timer->state and cpu_base->running
1465 * from the same section, if anything changed while we looked at it, we retry.
1466 * This includes timer->base changing because sequence numbers alone are
1467 * insufficient for that.
1468 *
1469 * The sequence numbers are required because otherwise we could still observe
1470 * a false negative if the read side got smeared over multiple consequtive
1471 * __run_hrtimer() invocations.
1472 */
1473
1474static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1475			  struct hrtimer_clock_base *base,
1476			  struct hrtimer *timer, ktime_t *now,
1477			  unsigned long flags)
1478{
1479	enum hrtimer_restart (*fn)(struct hrtimer *);
1480	int restart;
1481
1482	lockdep_assert_held(&cpu_base->lock);
1483
1484	debug_deactivate(timer);
1485	base->running = timer;
1486
1487	/*
1488	 * Separate the ->running assignment from the ->state assignment.
1489	 *
1490	 * As with a regular write barrier, this ensures the read side in
1491	 * hrtimer_active() cannot observe base->running == NULL &&
1492	 * timer->state == INACTIVE.
1493	 */
1494	raw_write_seqcount_barrier(&base->seq);
1495
1496	__remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1497	fn = timer->function;
1498
1499	/*
1500	 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1501	 * timer is restarted with a period then it becomes an absolute
1502	 * timer. If its not restarted it does not matter.
1503	 */
1504	if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1505		timer->is_rel = false;
1506
1507	/*
1508	 * The timer is marked as running in the CPU base, so it is
1509	 * protected against migration to a different CPU even if the lock
1510	 * is dropped.
1511	 */
1512	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1513	trace_hrtimer_expire_entry(timer, now);
1514	restart = fn(timer);
1515	trace_hrtimer_expire_exit(timer);
1516	raw_spin_lock_irq(&cpu_base->lock);
1517
1518	/*
1519	 * Note: We clear the running state after enqueue_hrtimer and
1520	 * we do not reprogram the event hardware. Happens either in
1521	 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1522	 *
1523	 * Note: Because we dropped the cpu_base->lock above,
1524	 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1525	 * for us already.
1526	 */
1527	if (restart != HRTIMER_NORESTART &&
1528	    !(timer->state & HRTIMER_STATE_ENQUEUED))
1529		enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1530
1531	/*
1532	 * Separate the ->running assignment from the ->state assignment.
1533	 *
1534	 * As with a regular write barrier, this ensures the read side in
1535	 * hrtimer_active() cannot observe base->running.timer == NULL &&
1536	 * timer->state == INACTIVE.
1537	 */
1538	raw_write_seqcount_barrier(&base->seq);
1539
1540	WARN_ON_ONCE(base->running != timer);
1541	base->running = NULL;
1542}
1543
1544static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1545				 unsigned long flags, unsigned int active_mask)
1546{
1547	struct hrtimer_clock_base *base;
1548	unsigned int active = cpu_base->active_bases & active_mask;
1549
1550	for_each_active_base(base, cpu_base, active) {
1551		struct timerqueue_node *node;
1552		ktime_t basenow;
1553
1554		basenow = ktime_add(now, base->offset);
1555
1556		while ((node = timerqueue_getnext(&base->active))) {
1557			struct hrtimer *timer;
1558
1559			timer = container_of(node, struct hrtimer, node);
1560
1561			/*
1562			 * The immediate goal for using the softexpires is
1563			 * minimizing wakeups, not running timers at the
1564			 * earliest interrupt after their soft expiration.
1565			 * This allows us to avoid using a Priority Search
1566			 * Tree, which can answer a stabbing querry for
1567			 * overlapping intervals and instead use the simple
1568			 * BST we already have.
1569			 * We don't add extra wakeups by delaying timers that
1570			 * are right-of a not yet expired timer, because that
1571			 * timer will have to trigger a wakeup anyway.
1572			 */
1573			if (basenow < hrtimer_get_softexpires_tv64(timer))
1574				break;
1575
1576			__run_hrtimer(cpu_base, base, timer, &basenow, flags);
1577			if (active_mask == HRTIMER_ACTIVE_SOFT)
1578				hrtimer_sync_wait_running(cpu_base, flags);
1579		}
1580	}
1581}
1582
1583static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1584{
1585	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1586	unsigned long flags;
1587	ktime_t now;
1588
1589	hrtimer_cpu_base_lock_expiry(cpu_base);
1590	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1591
1592	now = hrtimer_update_base(cpu_base);
1593	__hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1594
1595	cpu_base->softirq_activated = 0;
1596	hrtimer_update_softirq_timer(cpu_base, true);
1597
1598	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1599	hrtimer_cpu_base_unlock_expiry(cpu_base);
1600}
1601
1602#ifdef CONFIG_HIGH_RES_TIMERS
1603
1604/*
1605 * High resolution timer interrupt
1606 * Called with interrupts disabled
1607 */
1608void hrtimer_interrupt(struct clock_event_device *dev)
1609{
1610	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1611	ktime_t expires_next, now, entry_time, delta;
1612	unsigned long flags;
1613	int retries = 0;
1614
1615	BUG_ON(!cpu_base->hres_active);
1616	cpu_base->nr_events++;
1617	dev->next_event = KTIME_MAX;
1618
1619	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1620	entry_time = now = hrtimer_update_base(cpu_base);
1621retry:
1622	cpu_base->in_hrtirq = 1;
1623	/*
1624	 * We set expires_next to KTIME_MAX here with cpu_base->lock
1625	 * held to prevent that a timer is enqueued in our queue via
1626	 * the migration code. This does not affect enqueueing of
1627	 * timers which run their callback and need to be requeued on
1628	 * this CPU.
1629	 */
1630	cpu_base->expires_next = KTIME_MAX;
1631
1632	if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1633		cpu_base->softirq_expires_next = KTIME_MAX;
1634		cpu_base->softirq_activated = 1;
1635		raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1636	}
1637
1638	__hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1639
1640	/* Reevaluate the clock bases for the next expiry */
1641	expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1642	/*
1643	 * Store the new expiry value so the migration code can verify
1644	 * against it.
1645	 */
1646	cpu_base->expires_next = expires_next;
1647	cpu_base->in_hrtirq = 0;
1648	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1649
1650	/* Reprogramming necessary ? */
1651	if (!tick_program_event(expires_next, 0)) {
1652		cpu_base->hang_detected = 0;
1653		return;
1654	}
1655
1656	/*
1657	 * The next timer was already expired due to:
1658	 * - tracing
1659	 * - long lasting callbacks
1660	 * - being scheduled away when running in a VM
1661	 *
1662	 * We need to prevent that we loop forever in the hrtimer
1663	 * interrupt routine. We give it 3 attempts to avoid
1664	 * overreacting on some spurious event.
1665	 *
1666	 * Acquire base lock for updating the offsets and retrieving
1667	 * the current time.
1668	 */
1669	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1670	now = hrtimer_update_base(cpu_base);
1671	cpu_base->nr_retries++;
1672	if (++retries < 3)
1673		goto retry;
1674	/*
1675	 * Give the system a chance to do something else than looping
1676	 * here. We stored the entry time, so we know exactly how long
1677	 * we spent here. We schedule the next event this amount of
1678	 * time away.
1679	 */
1680	cpu_base->nr_hangs++;
1681	cpu_base->hang_detected = 1;
1682	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1683
1684	delta = ktime_sub(now, entry_time);
1685	if ((unsigned int)delta > cpu_base->max_hang_time)
1686		cpu_base->max_hang_time = (unsigned int) delta;
1687	/*
1688	 * Limit it to a sensible value as we enforce a longer
1689	 * delay. Give the CPU at least 100ms to catch up.
1690	 */
1691	if (delta > 100 * NSEC_PER_MSEC)
1692		expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1693	else
1694		expires_next = ktime_add(now, delta);
1695	tick_program_event(expires_next, 1);
1696	pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
 
1697}
1698
1699/* called with interrupts disabled */
1700static inline void __hrtimer_peek_ahead_timers(void)
1701{
1702	struct tick_device *td;
1703
1704	if (!hrtimer_hres_active())
1705		return;
1706
1707	td = this_cpu_ptr(&tick_cpu_device);
1708	if (td && td->evtdev)
1709		hrtimer_interrupt(td->evtdev);
1710}
1711
1712#else /* CONFIG_HIGH_RES_TIMERS */
1713
1714static inline void __hrtimer_peek_ahead_timers(void) { }
1715
1716#endif	/* !CONFIG_HIGH_RES_TIMERS */
1717
1718/*
1719 * Called from run_local_timers in hardirq context every jiffy
1720 */
1721void hrtimer_run_queues(void)
1722{
1723	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1724	unsigned long flags;
1725	ktime_t now;
1726
1727	if (__hrtimer_hres_active(cpu_base))
1728		return;
1729
1730	/*
1731	 * This _is_ ugly: We have to check periodically, whether we
1732	 * can switch to highres and / or nohz mode. The clocksource
1733	 * switch happens with xtime_lock held. Notification from
1734	 * there only sets the check bit in the tick_oneshot code,
1735	 * otherwise we might deadlock vs. xtime_lock.
1736	 */
1737	if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1738		hrtimer_switch_to_hres();
1739		return;
1740	}
1741
1742	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1743	now = hrtimer_update_base(cpu_base);
1744
1745	if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1746		cpu_base->softirq_expires_next = KTIME_MAX;
1747		cpu_base->softirq_activated = 1;
1748		raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1749	}
1750
1751	__hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1752	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1753}
1754
1755/*
1756 * Sleep related functions:
1757 */
1758static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1759{
1760	struct hrtimer_sleeper *t =
1761		container_of(timer, struct hrtimer_sleeper, timer);
1762	struct task_struct *task = t->task;
1763
1764	t->task = NULL;
1765	if (task)
1766		wake_up_process(task);
1767
1768	return HRTIMER_NORESTART;
1769}
1770
1771/**
1772 * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
1773 * @sl:		sleeper to be started
1774 * @mode:	timer mode abs/rel
1775 *
1776 * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
1777 * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
1778 */
1779void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
1780				   enum hrtimer_mode mode)
1781{
1782	/*
1783	 * Make the enqueue delivery mode check work on RT. If the sleeper
1784	 * was initialized for hard interrupt delivery, force the mode bit.
1785	 * This is a special case for hrtimer_sleepers because
1786	 * hrtimer_init_sleeper() determines the delivery mode on RT so the
1787	 * fiddling with this decision is avoided at the call sites.
1788	 */
1789	if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
1790		mode |= HRTIMER_MODE_HARD;
1791
1792	hrtimer_start_expires(&sl->timer, mode);
1793}
1794EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
1795
1796static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
1797				   clockid_t clock_id, enum hrtimer_mode mode)
1798{
1799	/*
1800	 * On PREEMPT_RT enabled kernels hrtimers which are not explicitely
1801	 * marked for hard interrupt expiry mode are moved into soft
1802	 * interrupt context either for latency reasons or because the
1803	 * hrtimer callback takes regular spinlocks or invokes other
1804	 * functions which are not suitable for hard interrupt context on
1805	 * PREEMPT_RT.
1806	 *
1807	 * The hrtimer_sleeper callback is RT compatible in hard interrupt
1808	 * context, but there is a latency concern: Untrusted userspace can
1809	 * spawn many threads which arm timers for the same expiry time on
1810	 * the same CPU. That causes a latency spike due to the wakeup of
1811	 * a gazillion threads.
1812	 *
1813	 * OTOH, priviledged real-time user space applications rely on the
1814	 * low latency of hard interrupt wakeups. If the current task is in
1815	 * a real-time scheduling class, mark the mode for hard interrupt
1816	 * expiry.
1817	 */
1818	if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1819		if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT))
1820			mode |= HRTIMER_MODE_HARD;
1821	}
1822
1823	__hrtimer_init(&sl->timer, clock_id, mode);
1824	sl->timer.function = hrtimer_wakeup;
1825	sl->task = current;
1826}
1827
1828/**
1829 * hrtimer_init_sleeper - initialize sleeper to the given clock
1830 * @sl:		sleeper to be initialized
1831 * @clock_id:	the clock to be used
1832 * @mode:	timer mode abs/rel
1833 */
1834void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id,
1835			  enum hrtimer_mode mode)
1836{
1837	debug_init(&sl->timer, clock_id, mode);
1838	__hrtimer_init_sleeper(sl, clock_id, mode);
1839
1840}
1841EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1842
1843int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1844{
1845	switch(restart->nanosleep.type) {
1846#ifdef CONFIG_COMPAT_32BIT_TIME
1847	case TT_COMPAT:
1848		if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
1849			return -EFAULT;
1850		break;
1851#endif
1852	case TT_NATIVE:
1853		if (put_timespec64(ts, restart->nanosleep.rmtp))
1854			return -EFAULT;
1855		break;
1856	default:
1857		BUG();
1858	}
1859	return -ERESTART_RESTARTBLOCK;
1860}
1861
1862static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1863{
1864	struct restart_block *restart;
1865
 
 
1866	do {
1867		set_current_state(TASK_INTERRUPTIBLE);
1868		hrtimer_sleeper_start_expires(t, mode);
1869
1870		if (likely(t->task))
1871			freezable_schedule();
1872
1873		hrtimer_cancel(&t->timer);
1874		mode = HRTIMER_MODE_ABS;
1875
1876	} while (t->task && !signal_pending(current));
1877
1878	__set_current_state(TASK_RUNNING);
1879
1880	if (!t->task)
1881		return 0;
1882
1883	restart = &current->restart_block;
1884	if (restart->nanosleep.type != TT_NONE) {
1885		ktime_t rem = hrtimer_expires_remaining(&t->timer);
1886		struct timespec64 rmt;
1887
1888		if (rem <= 0)
1889			return 0;
1890		rmt = ktime_to_timespec64(rem);
1891
1892		return nanosleep_copyout(restart, &rmt);
1893	}
1894	return -ERESTART_RESTARTBLOCK;
1895}
1896
1897static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1898{
1899	struct hrtimer_sleeper t;
1900	int ret;
1901
1902	hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
1903				      HRTIMER_MODE_ABS);
1904	hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
 
1905	ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1906	destroy_hrtimer_on_stack(&t.timer);
1907	return ret;
1908}
1909
1910long hrtimer_nanosleep(const struct timespec64 *rqtp,
1911		       const enum hrtimer_mode mode, const clockid_t clockid)
1912{
1913	struct restart_block *restart;
1914	struct hrtimer_sleeper t;
1915	int ret = 0;
1916	u64 slack;
1917
1918	slack = current->timer_slack_ns;
1919	if (dl_task(current) || rt_task(current))
1920		slack = 0;
1921
1922	hrtimer_init_sleeper_on_stack(&t, clockid, mode);
1923	hrtimer_set_expires_range_ns(&t.timer, timespec64_to_ktime(*rqtp), slack);
1924	ret = do_nanosleep(&t, mode);
1925	if (ret != -ERESTART_RESTARTBLOCK)
1926		goto out;
1927
1928	/* Absolute timers do not update the rmtp value and restart: */
1929	if (mode == HRTIMER_MODE_ABS) {
1930		ret = -ERESTARTNOHAND;
1931		goto out;
1932	}
1933
1934	restart = &current->restart_block;
1935	restart->fn = hrtimer_nanosleep_restart;
1936	restart->nanosleep.clockid = t.timer.base->clockid;
1937	restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1938out:
1939	destroy_hrtimer_on_stack(&t.timer);
1940	return ret;
1941}
1942
1943#if !defined(CONFIG_64BIT_TIME) || defined(CONFIG_64BIT)
1944
1945SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
1946		struct __kernel_timespec __user *, rmtp)
1947{
1948	struct timespec64 tu;
1949
1950	if (get_timespec64(&tu, rqtp))
1951		return -EFAULT;
1952
1953	if (!timespec64_valid(&tu))
1954		return -EINVAL;
1955
1956	current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1957	current->restart_block.nanosleep.rmtp = rmtp;
1958	return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1959}
1960
1961#endif
1962
1963#ifdef CONFIG_COMPAT_32BIT_TIME
1964
1965SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
1966		       struct old_timespec32 __user *, rmtp)
1967{
1968	struct timespec64 tu;
1969
1970	if (get_old_timespec32(&tu, rqtp))
1971		return -EFAULT;
1972
1973	if (!timespec64_valid(&tu))
1974		return -EINVAL;
1975
1976	current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
1977	current->restart_block.nanosleep.compat_rmtp = rmtp;
1978	return hrtimer_nanosleep(&tu, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1979}
1980#endif
1981
1982/*
1983 * Functions related to boot-time initialization:
1984 */
1985int hrtimers_prepare_cpu(unsigned int cpu)
1986{
1987	struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1988	int i;
1989
1990	for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1991		cpu_base->clock_base[i].cpu_base = cpu_base;
1992		timerqueue_init_head(&cpu_base->clock_base[i].active);
1993	}
1994
1995	cpu_base->cpu = cpu;
1996	cpu_base->active_bases = 0;
1997	cpu_base->hres_active = 0;
1998	cpu_base->hang_detected = 0;
1999	cpu_base->next_timer = NULL;
2000	cpu_base->softirq_next_timer = NULL;
2001	cpu_base->expires_next = KTIME_MAX;
2002	cpu_base->softirq_expires_next = KTIME_MAX;
2003	hrtimer_cpu_base_init_expiry_lock(cpu_base);
2004	return 0;
2005}
2006
2007#ifdef CONFIG_HOTPLUG_CPU
2008
2009static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2010				struct hrtimer_clock_base *new_base)
2011{
2012	struct hrtimer *timer;
2013	struct timerqueue_node *node;
2014
2015	while ((node = timerqueue_getnext(&old_base->active))) {
2016		timer = container_of(node, struct hrtimer, node);
2017		BUG_ON(hrtimer_callback_running(timer));
2018		debug_deactivate(timer);
2019
2020		/*
2021		 * Mark it as ENQUEUED not INACTIVE otherwise the
2022		 * timer could be seen as !active and just vanish away
2023		 * under us on another CPU
2024		 */
2025		__remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2026		timer->base = new_base;
2027		/*
2028		 * Enqueue the timers on the new cpu. This does not
2029		 * reprogram the event device in case the timer
2030		 * expires before the earliest on this CPU, but we run
2031		 * hrtimer_interrupt after we migrated everything to
2032		 * sort out already expired timers and reprogram the
2033		 * event device.
2034		 */
2035		enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2036	}
2037}
2038
2039int hrtimers_dead_cpu(unsigned int scpu)
2040{
2041	struct hrtimer_cpu_base *old_base, *new_base;
2042	int i;
2043
2044	BUG_ON(cpu_online(scpu));
2045	tick_cancel_sched_timer(scpu);
2046
2047	/*
2048	 * this BH disable ensures that raise_softirq_irqoff() does
2049	 * not wakeup ksoftirqd (and acquire the pi-lock) while
2050	 * holding the cpu_base lock
2051	 */
2052	local_bh_disable();
2053	local_irq_disable();
2054	old_base = &per_cpu(hrtimer_bases, scpu);
2055	new_base = this_cpu_ptr(&hrtimer_bases);
2056	/*
2057	 * The caller is globally serialized and nobody else
2058	 * takes two locks at once, deadlock is not possible.
2059	 */
2060	raw_spin_lock(&new_base->lock);
2061	raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2062
2063	for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2064		migrate_hrtimer_list(&old_base->clock_base[i],
2065				     &new_base->clock_base[i]);
2066	}
2067
2068	/*
2069	 * The migration might have changed the first expiring softirq
2070	 * timer on this CPU. Update it.
2071	 */
2072	hrtimer_update_softirq_timer(new_base, false);
2073
2074	raw_spin_unlock(&old_base->lock);
2075	raw_spin_unlock(&new_base->lock);
2076
2077	/* Check, if we got expired work to do */
2078	__hrtimer_peek_ahead_timers();
2079	local_irq_enable();
2080	local_bh_enable();
2081	return 0;
2082}
2083
2084#endif /* CONFIG_HOTPLUG_CPU */
2085
2086void __init hrtimers_init(void)
2087{
2088	hrtimers_prepare_cpu(smp_processor_id());
2089	open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2090}
2091
2092/**
2093 * schedule_hrtimeout_range_clock - sleep until timeout
2094 * @expires:	timeout value (ktime_t)
2095 * @delta:	slack in expires timeout (ktime_t)
2096 * @mode:	timer mode
2097 * @clock_id:	timer clock to be used
2098 */
2099int __sched
2100schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
2101			       const enum hrtimer_mode mode, clockid_t clock_id)
2102{
2103	struct hrtimer_sleeper t;
2104
2105	/*
2106	 * Optimize when a zero timeout value is given. It does not
2107	 * matter whether this is an absolute or a relative time.
2108	 */
2109	if (expires && *expires == 0) {
2110		__set_current_state(TASK_RUNNING);
2111		return 0;
2112	}
2113
2114	/*
2115	 * A NULL parameter means "infinite"
2116	 */
2117	if (!expires) {
2118		schedule();
2119		return -EINTR;
2120	}
2121
2122	hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
2123	hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2124	hrtimer_sleeper_start_expires(&t, mode);
 
 
 
2125
2126	if (likely(t.task))
2127		schedule();
2128
2129	hrtimer_cancel(&t.timer);
2130	destroy_hrtimer_on_stack(&t.timer);
2131
2132	__set_current_state(TASK_RUNNING);
2133
2134	return !t.task ? 0 : -EINTR;
2135}
2136
2137/**
2138 * schedule_hrtimeout_range - sleep until timeout
2139 * @expires:	timeout value (ktime_t)
2140 * @delta:	slack in expires timeout (ktime_t)
2141 * @mode:	timer mode
2142 *
2143 * Make the current task sleep until the given expiry time has
2144 * elapsed. The routine will return immediately unless
2145 * the current task state has been set (see set_current_state()).
2146 *
2147 * The @delta argument gives the kernel the freedom to schedule the
2148 * actual wakeup to a time that is both power and performance friendly.
2149 * The kernel give the normal best effort behavior for "@expires+@delta",
2150 * but may decide to fire the timer earlier, but no earlier than @expires.
2151 *
2152 * You can set the task state as follows -
2153 *
2154 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2155 * pass before the routine returns unless the current task is explicitly
2156 * woken up, (e.g. by wake_up_process()).
2157 *
2158 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2159 * delivered to the current task or the current task is explicitly woken
2160 * up.
2161 *
2162 * The current task state is guaranteed to be TASK_RUNNING when this
2163 * routine returns.
2164 *
2165 * Returns 0 when the timer has expired. If the task was woken before the
2166 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2167 * by an explicit wakeup, it returns -EINTR.
2168 */
2169int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2170				     const enum hrtimer_mode mode)
2171{
2172	return schedule_hrtimeout_range_clock(expires, delta, mode,
2173					      CLOCK_MONOTONIC);
2174}
2175EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2176
2177/**
2178 * schedule_hrtimeout - sleep until timeout
2179 * @expires:	timeout value (ktime_t)
2180 * @mode:	timer mode
2181 *
2182 * Make the current task sleep until the given expiry time has
2183 * elapsed. The routine will return immediately unless
2184 * the current task state has been set (see set_current_state()).
2185 *
2186 * You can set the task state as follows -
2187 *
2188 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2189 * pass before the routine returns unless the current task is explicitly
2190 * woken up, (e.g. by wake_up_process()).
2191 *
2192 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2193 * delivered to the current task or the current task is explicitly woken
2194 * up.
2195 *
2196 * The current task state is guaranteed to be TASK_RUNNING when this
2197 * routine returns.
2198 *
2199 * Returns 0 when the timer has expired. If the task was woken before the
2200 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2201 * by an explicit wakeup, it returns -EINTR.
2202 */
2203int __sched schedule_hrtimeout(ktime_t *expires,
2204			       const enum hrtimer_mode mode)
2205{
2206	return schedule_hrtimeout_range(expires, 0, mode);
2207}
2208EXPORT_SYMBOL_GPL(schedule_hrtimeout);