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 = { {
 139		.cpu_base = &migration_cpu_base,
 140		.seq      = SEQCNT_RAW_SPINLOCK_ZERO(migration_cpu_base.seq,
 141						     &migration_cpu_base.lock),
 142	}, },
 143};
 144
 145#define migration_base	migration_cpu_base.clock_base[0]
 146
 147static inline bool is_migration_base(struct hrtimer_clock_base *base)
 148{
 149	return base == &migration_base;
 150}
 151
 152/*
 153 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
 154 * means that all timers which are tied to this base via timer->base are
 155 * locked, and the base itself is locked too.
 156 *
 157 * So __run_timers/migrate_timers can safely modify all timers which could
 158 * be found on the lists/queues.
 159 *
 160 * When the timer's base is locked, and the timer removed from list, it is
 161 * possible to set timer->base = &migration_base and drop the lock: the timer
 162 * remains locked.
 163 */
 164static
 165struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
 166					     unsigned long *flags)
 
 167{
 168	struct hrtimer_clock_base *base;
 169
 170	for (;;) {
 171		base = READ_ONCE(timer->base);
 172		if (likely(base != &migration_base)) {
 173			raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 174			if (likely(base == timer->base))
 175				return base;
 176			/* The timer has migrated to another CPU: */
 177			raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
 178		}
 179		cpu_relax();
 180	}
 181}
 182
 183/*
 184 * We do not migrate the timer when it is expiring before the next
 185 * event on the target cpu. When high resolution is enabled, we cannot
 186 * reprogram the target cpu hardware and we would cause it to fire
 187 * late. To keep it simple, we handle the high resolution enabled and
 188 * disabled case similar.
 189 *
 190 * Called with cpu_base->lock of target cpu held.
 191 */
 192static int
 193hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
 194{
 195	ktime_t expires;
 196
 197	expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
 198	return expires < new_base->cpu_base->expires_next;
 199}
 200
 201static inline
 202struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
 203					 int pinned)
 204{
 205#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
 206	if (static_branch_likely(&timers_migration_enabled) && !pinned)
 207		return &per_cpu(hrtimer_bases, get_nohz_timer_target());
 208#endif
 209	return base;
 210}
 211
 212/*
 213 * We switch the timer base to a power-optimized selected CPU target,
 214 * if:
 215 *	- NO_HZ_COMMON is enabled
 216 *	- timer migration is enabled
 217 *	- the timer callback is not running
 218 *	- the timer is not the first expiring timer on the new target
 219 *
 220 * If one of the above requirements is not fulfilled we move the timer
 221 * to the current CPU or leave it on the previously assigned CPU if
 222 * the timer callback is currently running.
 223 */
 224static inline struct hrtimer_clock_base *
 225switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 226		    int pinned)
 227{
 228	struct hrtimer_cpu_base *new_cpu_base, *this_cpu_base;
 229	struct hrtimer_clock_base *new_base;
 230	int basenum = base->index;
 231
 232	this_cpu_base = this_cpu_ptr(&hrtimer_bases);
 233	new_cpu_base = get_target_base(this_cpu_base, pinned);
 234again:
 235	new_base = &new_cpu_base->clock_base[basenum];
 236
 237	if (base != new_base) {
 238		/*
 239		 * We are trying to move timer to new_base.
 240		 * However we can't change timer's base while it is running,
 241		 * so we keep it on the same CPU. No hassle vs. reprogramming
 242		 * the event source in the high resolution case. The softirq
 243		 * code will take care of this when the timer function has
 244		 * completed. There is no conflict as we hold the lock until
 245		 * the timer is enqueued.
 246		 */
 247		if (unlikely(hrtimer_callback_running(timer)))
 248			return base;
 249
 250		/* See the comment in lock_hrtimer_base() */
 251		WRITE_ONCE(timer->base, &migration_base);
 252		raw_spin_unlock(&base->cpu_base->lock);
 253		raw_spin_lock(&new_base->cpu_base->lock);
 254
 255		if (new_cpu_base != this_cpu_base &&
 256		    hrtimer_check_target(timer, new_base)) {
 257			raw_spin_unlock(&new_base->cpu_base->lock);
 258			raw_spin_lock(&base->cpu_base->lock);
 259			new_cpu_base = this_cpu_base;
 260			WRITE_ONCE(timer->base, base);
 261			goto again;
 262		}
 263		WRITE_ONCE(timer->base, new_base);
 264	} else {
 265		if (new_cpu_base != this_cpu_base &&
 266		    hrtimer_check_target(timer, new_base)) {
 267			new_cpu_base = this_cpu_base;
 268			goto again;
 269		}
 270	}
 271	return new_base;
 272}
 273
 274#else /* CONFIG_SMP */
 275
 276static inline bool is_migration_base(struct hrtimer_clock_base *base)
 277{
 278	return false;
 279}
 280
 281static inline struct hrtimer_clock_base *
 282lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 
 283{
 284	struct hrtimer_clock_base *base = timer->base;
 285
 286	raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 287
 288	return base;
 289}
 290
 291# define switch_hrtimer_base(t, b, p)	(b)
 292
 293#endif	/* !CONFIG_SMP */
 294
 295/*
 296 * Functions for the union type storage format of ktime_t which are
 297 * too large for inlining:
 298 */
 299#if BITS_PER_LONG < 64
 300/*
 301 * Divide a ktime value by a nanosecond value
 302 */
 303s64 __ktime_divns(const ktime_t kt, s64 div)
 304{
 305	int sft = 0;
 306	s64 dclc;
 307	u64 tmp;
 308
 309	dclc = ktime_to_ns(kt);
 310	tmp = dclc < 0 ? -dclc : dclc;
 311
 312	/* Make sure the divisor is less than 2^32: */
 313	while (div >> 32) {
 314		sft++;
 315		div >>= 1;
 316	}
 317	tmp >>= sft;
 318	do_div(tmp, (u32) div);
 319	return dclc < 0 ? -tmp : tmp;
 320}
 321EXPORT_SYMBOL_GPL(__ktime_divns);
 322#endif /* BITS_PER_LONG >= 64 */
 323
 324/*
 325 * Add two ktime values and do a safety check for overflow:
 326 */
 327ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
 328{
 329	ktime_t res = ktime_add_unsafe(lhs, rhs);
 330
 331	/*
 332	 * We use KTIME_SEC_MAX here, the maximum timeout which we can
 333	 * return to user space in a timespec:
 334	 */
 335	if (res < 0 || res < lhs || res < rhs)
 336		res = ktime_set(KTIME_SEC_MAX, 0);
 337
 338	return res;
 339}
 340
 341EXPORT_SYMBOL_GPL(ktime_add_safe);
 342
 343#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 344
 345static const struct debug_obj_descr hrtimer_debug_descr;
 346
 347static void *hrtimer_debug_hint(void *addr)
 348{
 349	return ((struct hrtimer *) addr)->function;
 350}
 351
 352/*
 353 * fixup_init is called when:
 354 * - an active object is initialized
 355 */
 356static bool hrtimer_fixup_init(void *addr, enum debug_obj_state state)
 357{
 358	struct hrtimer *timer = addr;
 359
 360	switch (state) {
 361	case ODEBUG_STATE_ACTIVE:
 362		hrtimer_cancel(timer);
 363		debug_object_init(timer, &hrtimer_debug_descr);
 364		return true;
 365	default:
 366		return false;
 367	}
 368}
 369
 370/*
 371 * fixup_activate is called when:
 372 * - an active object is activated
 373 * - an unknown non-static object is activated
 374 */
 375static bool hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
 376{
 377	switch (state) {
 378	case ODEBUG_STATE_ACTIVE:
 379		WARN_ON(1);
 380		fallthrough;
 381	default:
 382		return false;
 383	}
 384}
 385
 386/*
 387 * fixup_free is called when:
 388 * - an active object is freed
 389 */
 390static bool hrtimer_fixup_free(void *addr, enum debug_obj_state state)
 391{
 392	struct hrtimer *timer = addr;
 393
 394	switch (state) {
 395	case ODEBUG_STATE_ACTIVE:
 396		hrtimer_cancel(timer);
 397		debug_object_free(timer, &hrtimer_debug_descr);
 398		return true;
 399	default:
 400		return false;
 401	}
 402}
 403
 404static const struct debug_obj_descr hrtimer_debug_descr = {
 405	.name		= "hrtimer",
 406	.debug_hint	= hrtimer_debug_hint,
 407	.fixup_init	= hrtimer_fixup_init,
 408	.fixup_activate	= hrtimer_fixup_activate,
 409	.fixup_free	= hrtimer_fixup_free,
 410};
 411
 412static inline void debug_hrtimer_init(struct hrtimer *timer)
 413{
 414	debug_object_init(timer, &hrtimer_debug_descr);
 415}
 416
 417static inline void debug_hrtimer_activate(struct hrtimer *timer,
 418					  enum hrtimer_mode mode)
 419{
 420	debug_object_activate(timer, &hrtimer_debug_descr);
 421}
 422
 423static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
 424{
 425	debug_object_deactivate(timer, &hrtimer_debug_descr);
 426}
 427
 428static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
 429			   enum hrtimer_mode mode);
 430
 431void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
 432			   enum hrtimer_mode mode)
 433{
 434	debug_object_init_on_stack(timer, &hrtimer_debug_descr);
 435	__hrtimer_init(timer, clock_id, mode);
 436}
 437EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
 438
 439static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
 440				   clockid_t clock_id, enum hrtimer_mode mode);
 441
 442void hrtimer_init_sleeper_on_stack(struct hrtimer_sleeper *sl,
 443				   clockid_t clock_id, enum hrtimer_mode mode)
 444{
 445	debug_object_init_on_stack(&sl->timer, &hrtimer_debug_descr);
 446	__hrtimer_init_sleeper(sl, clock_id, mode);
 447}
 448EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack);
 449
 450void destroy_hrtimer_on_stack(struct hrtimer *timer)
 451{
 452	debug_object_free(timer, &hrtimer_debug_descr);
 453}
 454EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
 455
 456#else
 457
 458static inline void debug_hrtimer_init(struct hrtimer *timer) { }
 459static inline void debug_hrtimer_activate(struct hrtimer *timer,
 460					  enum hrtimer_mode mode) { }
 461static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
 462#endif
 463
 464static inline void
 465debug_init(struct hrtimer *timer, clockid_t clockid,
 466	   enum hrtimer_mode mode)
 467{
 468	debug_hrtimer_init(timer);
 469	trace_hrtimer_init(timer, clockid, mode);
 470}
 471
 472static inline void debug_activate(struct hrtimer *timer,
 473				  enum hrtimer_mode mode)
 474{
 475	debug_hrtimer_activate(timer, mode);
 476	trace_hrtimer_start(timer, mode);
 477}
 478
 479static inline void debug_deactivate(struct hrtimer *timer)
 480{
 481	debug_hrtimer_deactivate(timer);
 482	trace_hrtimer_cancel(timer);
 483}
 484
 485static struct hrtimer_clock_base *
 486__next_base(struct hrtimer_cpu_base *cpu_base, unsigned int *active)
 487{
 488	unsigned int idx;
 489
 490	if (!*active)
 491		return NULL;
 492
 493	idx = __ffs(*active);
 494	*active &= ~(1U << idx);
 495
 496	return &cpu_base->clock_base[idx];
 497}
 498
 499#define for_each_active_base(base, cpu_base, active)	\
 500	while ((base = __next_base((cpu_base), &(active))))
 501
 502static ktime_t __hrtimer_next_event_base(struct hrtimer_cpu_base *cpu_base,
 503					 const struct hrtimer *exclude,
 504					 unsigned int active,
 505					 ktime_t expires_next)
 506{
 507	struct hrtimer_clock_base *base;
 508	ktime_t expires;
 509
 510	for_each_active_base(base, cpu_base, active) {
 511		struct timerqueue_node *next;
 512		struct hrtimer *timer;
 513
 514		next = timerqueue_getnext(&base->active);
 515		timer = container_of(next, struct hrtimer, node);
 516		if (timer == exclude) {
 517			/* Get to the next timer in the queue. */
 518			next = timerqueue_iterate_next(next);
 519			if (!next)
 520				continue;
 521
 522			timer = container_of(next, struct hrtimer, node);
 523		}
 524		expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 525		if (expires < expires_next) {
 526			expires_next = expires;
 527
 528			/* Skip cpu_base update if a timer is being excluded. */
 529			if (exclude)
 530				continue;
 531
 532			if (timer->is_soft)
 533				cpu_base->softirq_next_timer = timer;
 534			else
 535				cpu_base->next_timer = timer;
 536		}
 537	}
 538	/*
 539	 * clock_was_set() might have changed base->offset of any of
 540	 * the clock bases so the result might be negative. Fix it up
 541	 * to prevent a false positive in clockevents_program_event().
 542	 */
 543	if (expires_next < 0)
 544		expires_next = 0;
 545	return expires_next;
 546}
 547
 548/*
 549 * Recomputes cpu_base::*next_timer and returns the earliest expires_next
 550 * but does not set cpu_base::*expires_next, that is done by
 551 * hrtimer[_force]_reprogram and hrtimer_interrupt only. When updating
 552 * cpu_base::*expires_next right away, reprogramming logic would no longer
 553 * work.
 554 *
 555 * When a softirq is pending, we can ignore the HRTIMER_ACTIVE_SOFT bases,
 556 * those timers will get run whenever the softirq gets handled, at the end of
 557 * hrtimer_run_softirq(), hrtimer_update_softirq_timer() will re-add these bases.
 558 *
 559 * Therefore softirq values are those from the HRTIMER_ACTIVE_SOFT clock bases.
 560 * The !softirq values are the minima across HRTIMER_ACTIVE_ALL, unless an actual
 561 * softirq is pending, in which case they're the minima of HRTIMER_ACTIVE_HARD.
 562 *
 563 * @active_mask must be one of:
 564 *  - HRTIMER_ACTIVE_ALL,
 565 *  - HRTIMER_ACTIVE_SOFT, or
 566 *  - HRTIMER_ACTIVE_HARD.
 567 */
 568static ktime_t
 569__hrtimer_get_next_event(struct hrtimer_cpu_base *cpu_base, unsigned int active_mask)
 570{
 571	unsigned int active;
 572	struct hrtimer *next_timer = NULL;
 573	ktime_t expires_next = KTIME_MAX;
 574
 575	if (!cpu_base->softirq_activated && (active_mask & HRTIMER_ACTIVE_SOFT)) {
 576		active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
 577		cpu_base->softirq_next_timer = NULL;
 578		expires_next = __hrtimer_next_event_base(cpu_base, NULL,
 579							 active, KTIME_MAX);
 580
 581		next_timer = cpu_base->softirq_next_timer;
 582	}
 583
 584	if (active_mask & HRTIMER_ACTIVE_HARD) {
 585		active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
 586		cpu_base->next_timer = next_timer;
 587		expires_next = __hrtimer_next_event_base(cpu_base, NULL, active,
 588							 expires_next);
 589	}
 590
 591	return expires_next;
 592}
 593
 594static ktime_t hrtimer_update_next_event(struct hrtimer_cpu_base *cpu_base)
 595{
 596	ktime_t expires_next, soft = KTIME_MAX;
 597
 598	/*
 599	 * If the soft interrupt has already been activated, ignore the
 600	 * soft bases. They will be handled in the already raised soft
 601	 * interrupt.
 602	 */
 603	if (!cpu_base->softirq_activated) {
 604		soft = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
 605		/*
 606		 * Update the soft expiry time. clock_settime() might have
 607		 * affected it.
 608		 */
 609		cpu_base->softirq_expires_next = soft;
 610	}
 611
 612	expires_next = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_HARD);
 613	/*
 614	 * If a softirq timer is expiring first, update cpu_base->next_timer
 615	 * and program the hardware with the soft expiry time.
 616	 */
 617	if (expires_next > soft) {
 618		cpu_base->next_timer = cpu_base->softirq_next_timer;
 619		expires_next = soft;
 620	}
 621
 622	return expires_next;
 623}
 624
 625static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
 626{
 627	ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
 628	ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
 629	ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
 630
 631	ktime_t now = ktime_get_update_offsets_now(&base->clock_was_set_seq,
 632					    offs_real, offs_boot, offs_tai);
 633
 634	base->clock_base[HRTIMER_BASE_REALTIME_SOFT].offset = *offs_real;
 635	base->clock_base[HRTIMER_BASE_BOOTTIME_SOFT].offset = *offs_boot;
 636	base->clock_base[HRTIMER_BASE_TAI_SOFT].offset = *offs_tai;
 637
 638	return now;
 639}
 640
 641/*
 642 * Is the high resolution mode active ?
 643 */
 644static inline int __hrtimer_hres_active(struct hrtimer_cpu_base *cpu_base)
 645{
 646	return IS_ENABLED(CONFIG_HIGH_RES_TIMERS) ?
 647		cpu_base->hres_active : 0;
 648}
 649
 650static inline int hrtimer_hres_active(void)
 651{
 652	return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
 653}
 654
 655static void __hrtimer_reprogram(struct hrtimer_cpu_base *cpu_base,
 656				struct hrtimer *next_timer,
 657				ktime_t expires_next)
 658{
 659	cpu_base->expires_next = expires_next;
 660
 661	/*
 662	 * If hres is not active, hardware does not have to be
 663	 * reprogrammed yet.
 664	 *
 665	 * If a hang was detected in the last timer interrupt then we
 666	 * leave the hang delay active in the hardware. We want the
 667	 * system to make progress. That also prevents the following
 668	 * scenario:
 669	 * T1 expires 50ms from now
 670	 * T2 expires 5s from now
 671	 *
 672	 * T1 is removed, so this code is called and would reprogram
 673	 * the hardware to 5s from now. Any hrtimer_start after that
 674	 * will not reprogram the hardware due to hang_detected being
 675	 * set. So we'd effectively block all timers until the T2 event
 676	 * fires.
 677	 */
 678	if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
 679		return;
 680
 681	tick_program_event(expires_next, 1);
 682}
 683
 684/*
 685 * Reprogram the event source with checking both queues for the
 686 * next event
 687 * Called with interrupts disabled and base->lock held
 688 */
 689static void
 690hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
 691{
 692	ktime_t expires_next;
 693
 694	expires_next = hrtimer_update_next_event(cpu_base);
 695
 696	if (skip_equal && expires_next == cpu_base->expires_next)
 697		return;
 698
 699	__hrtimer_reprogram(cpu_base, cpu_base->next_timer, expires_next);
 700}
 701
 702/* High resolution timer related functions */
 703#ifdef CONFIG_HIGH_RES_TIMERS
 704
 705/*
 706 * High resolution timer enabled ?
 707 */
 708static bool hrtimer_hres_enabled __read_mostly  = true;
 709unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
 710EXPORT_SYMBOL_GPL(hrtimer_resolution);
 711
 712/*
 713 * Enable / Disable high resolution mode
 714 */
 715static int __init setup_hrtimer_hres(char *str)
 716{
 717	return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
 718}
 719
 720__setup("highres=", setup_hrtimer_hres);
 721
 722/*
 723 * hrtimer_high_res_enabled - query, if the highres mode is enabled
 724 */
 725static inline int hrtimer_is_hres_enabled(void)
 726{
 727	return hrtimer_hres_enabled;
 728}
 729
 730static void retrigger_next_event(void *arg);
 731
 732/*
 733 * Switch to high resolution mode
 734 */
 735static void hrtimer_switch_to_hres(void)
 736{
 737	struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
 738
 739	if (tick_init_highres()) {
 740		pr_warn("Could not switch to high resolution mode on CPU %u\n",
 741			base->cpu);
 742		return;
 743	}
 744	base->hres_active = 1;
 745	hrtimer_resolution = HIGH_RES_NSEC;
 746
 747	tick_setup_sched_timer();
 748	/* "Retrigger" the interrupt to get things going */
 749	retrigger_next_event(NULL);
 750}
 751
 752#else
 753
 754static inline int hrtimer_is_hres_enabled(void) { return 0; }
 755static inline void hrtimer_switch_to_hres(void) { }
 756
 757#endif /* CONFIG_HIGH_RES_TIMERS */
 758/*
 759 * Retrigger next event is called after clock was set with interrupts
 760 * disabled through an SMP function call or directly from low level
 761 * resume code.
 762 *
 763 * This is only invoked when:
 764 *	- CONFIG_HIGH_RES_TIMERS is enabled.
 765 *	- CONFIG_NOHZ_COMMON is enabled
 766 *
 767 * For the other cases this function is empty and because the call sites
 768 * are optimized out it vanishes as well, i.e. no need for lots of
 769 * #ifdeffery.
 770 */
 771static void retrigger_next_event(void *arg)
 772{
 773	struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
 774
 775	/*
 776	 * When high resolution mode or nohz is active, then the offsets of
 777	 * CLOCK_REALTIME/TAI/BOOTTIME have to be updated. Otherwise the
 778	 * next tick will take care of that.
 779	 *
 780	 * If high resolution mode is active then the next expiring timer
 781	 * must be reevaluated and the clock event device reprogrammed if
 782	 * necessary.
 783	 *
 784	 * In the NOHZ case the update of the offset and the reevaluation
 785	 * of the next expiring timer is enough. The return from the SMP
 786	 * function call will take care of the reprogramming in case the
 787	 * CPU was in a NOHZ idle sleep.
 788	 */
 789	if (!__hrtimer_hres_active(base) && !tick_nohz_active)
 790		return;
 791
 792	raw_spin_lock(&base->lock);
 793	hrtimer_update_base(base);
 794	if (__hrtimer_hres_active(base))
 795		hrtimer_force_reprogram(base, 0);
 796	else
 797		hrtimer_update_next_event(base);
 798	raw_spin_unlock(&base->lock);
 799}
 800
 801/*
 802 * When a timer is enqueued and expires earlier than the already enqueued
 803 * timers, we have to check, whether it expires earlier than the timer for
 804 * which the clock event device was armed.
 805 *
 806 * Called with interrupts disabled and base->cpu_base.lock held
 807 */
 808static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
 809{
 810	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
 811	struct hrtimer_clock_base *base = timer->base;
 812	ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 813
 814	WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
 815
 816	/*
 817	 * CLOCK_REALTIME timer might be requested with an absolute
 818	 * expiry time which is less than base->offset. Set it to 0.
 819	 */
 820	if (expires < 0)
 821		expires = 0;
 822
 823	if (timer->is_soft) {
 824		/*
 825		 * soft hrtimer could be started on a remote CPU. In this
 826		 * case softirq_expires_next needs to be updated on the
 827		 * remote CPU. The soft hrtimer will not expire before the
 828		 * first hard hrtimer on the remote CPU -
 829		 * hrtimer_check_target() prevents this case.
 830		 */
 831		struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
 832
 833		if (timer_cpu_base->softirq_activated)
 834			return;
 835
 836		if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
 837			return;
 838
 839		timer_cpu_base->softirq_next_timer = timer;
 840		timer_cpu_base->softirq_expires_next = expires;
 841
 842		if (!ktime_before(expires, timer_cpu_base->expires_next) ||
 843		    !reprogram)
 844			return;
 845	}
 846
 847	/*
 848	 * If the timer is not on the current cpu, we cannot reprogram
 849	 * the other cpus clock event device.
 850	 */
 851	if (base->cpu_base != cpu_base)
 852		return;
 853
 854	if (expires >= cpu_base->expires_next)
 855		return;
 856
 857	/*
 858	 * If the hrtimer interrupt is running, then it will reevaluate the
 859	 * clock bases and reprogram the clock event device.
 860	 */
 861	if (cpu_base->in_hrtirq)
 862		return;
 863
 864	cpu_base->next_timer = timer;
 865
 866	__hrtimer_reprogram(cpu_base, timer, expires);
 867}
 868
 869static bool update_needs_ipi(struct hrtimer_cpu_base *cpu_base,
 870			     unsigned int active)
 871{
 872	struct hrtimer_clock_base *base;
 873	unsigned int seq;
 874	ktime_t expires;
 875
 876	/*
 877	 * Update the base offsets unconditionally so the following
 878	 * checks whether the SMP function call is required works.
 879	 *
 880	 * The update is safe even when the remote CPU is in the hrtimer
 881	 * interrupt or the hrtimer soft interrupt and expiring affected
 882	 * bases. Either it will see the update before handling a base or
 883	 * it will see it when it finishes the processing and reevaluates
 884	 * the next expiring timer.
 885	 */
 886	seq = cpu_base->clock_was_set_seq;
 887	hrtimer_update_base(cpu_base);
 888
 889	/*
 890	 * If the sequence did not change over the update then the
 891	 * remote CPU already handled it.
 892	 */
 893	if (seq == cpu_base->clock_was_set_seq)
 894		return false;
 895
 896	/*
 897	 * If the remote CPU is currently handling an hrtimer interrupt, it
 898	 * will reevaluate the first expiring timer of all clock bases
 899	 * before reprogramming. Nothing to do here.
 900	 */
 901	if (cpu_base->in_hrtirq)
 902		return false;
 903
 904	/*
 905	 * Walk the affected clock bases and check whether the first expiring
 906	 * timer in a clock base is moving ahead of the first expiring timer of
 907	 * @cpu_base. If so, the IPI must be invoked because per CPU clock
 908	 * event devices cannot be remotely reprogrammed.
 909	 */
 910	active &= cpu_base->active_bases;
 911
 912	for_each_active_base(base, cpu_base, active) {
 913		struct timerqueue_node *next;
 914
 915		next = timerqueue_getnext(&base->active);
 916		expires = ktime_sub(next->expires, base->offset);
 917		if (expires < cpu_base->expires_next)
 918			return true;
 919
 920		/* Extra check for softirq clock bases */
 921		if (base->clockid < HRTIMER_BASE_MONOTONIC_SOFT)
 922			continue;
 923		if (cpu_base->softirq_activated)
 924			continue;
 925		if (expires < cpu_base->softirq_expires_next)
 926			return true;
 927	}
 928	return false;
 929}
 930
 931/*
 932 * Clock was set. This might affect CLOCK_REALTIME, CLOCK_TAI and
 933 * CLOCK_BOOTTIME (for late sleep time injection).
 934 *
 935 * This requires to update the offsets for these clocks
 936 * vs. CLOCK_MONOTONIC. When high resolution timers are enabled, then this
 937 * also requires to eventually reprogram the per CPU clock event devices
 938 * when the change moves an affected timer ahead of the first expiring
 939 * timer on that CPU. Obviously remote per CPU clock event devices cannot
 940 * be reprogrammed. The other reason why an IPI has to be sent is when the
 941 * system is in !HIGH_RES and NOHZ mode. The NOHZ mode updates the offsets
 942 * in the tick, which obviously might be stopped, so this has to bring out
 943 * the remote CPU which might sleep in idle to get this sorted.
 944 */
 945void clock_was_set(unsigned int bases)
 946{
 947	struct hrtimer_cpu_base *cpu_base = raw_cpu_ptr(&hrtimer_bases);
 948	cpumask_var_t mask;
 949	int cpu;
 950
 951	if (!__hrtimer_hres_active(cpu_base) && !tick_nohz_active)
 952		goto out_timerfd;
 953
 954	if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
 955		on_each_cpu(retrigger_next_event, NULL, 1);
 956		goto out_timerfd;
 957	}
 958
 959	/* Avoid interrupting CPUs if possible */
 960	cpus_read_lock();
 961	for_each_online_cpu(cpu) {
 962		unsigned long flags;
 963
 964		cpu_base = &per_cpu(hrtimer_bases, cpu);
 965		raw_spin_lock_irqsave(&cpu_base->lock, flags);
 966
 967		if (update_needs_ipi(cpu_base, bases))
 968			cpumask_set_cpu(cpu, mask);
 969
 970		raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
 971	}
 972
 973	preempt_disable();
 974	smp_call_function_many(mask, retrigger_next_event, NULL, 1);
 975	preempt_enable();
 976	cpus_read_unlock();
 977	free_cpumask_var(mask);
 978
 979out_timerfd:
 980	timerfd_clock_was_set();
 981}
 982
 983static void clock_was_set_work(struct work_struct *work)
 984{
 985	clock_was_set(CLOCK_SET_WALL);
 986}
 987
 988static DECLARE_WORK(hrtimer_work, clock_was_set_work);
 989
 990/*
 991 * Called from timekeeping code to reprogram the hrtimer interrupt device
 992 * on all cpus and to notify timerfd.
 993 */
 994void clock_was_set_delayed(void)
 995{
 996	schedule_work(&hrtimer_work);
 997}
 998
 999/*
1000 * Called during resume either directly from via timekeeping_resume()
1001 * or in the case of s2idle from tick_unfreeze() to ensure that the
1002 * hrtimers are up to date.
1003 */
1004void hrtimers_resume_local(void)
1005{
1006	lockdep_assert_irqs_disabled();
1007	/* Retrigger on the local CPU */
1008	retrigger_next_event(NULL);
1009}
1010
1011/*
1012 * Counterpart to lock_hrtimer_base above:
1013 */
1014static inline
1015void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 
1016{
1017	raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
1018}
1019
1020/**
1021 * hrtimer_forward - forward the timer expiry
1022 * @timer:	hrtimer to forward
1023 * @now:	forward past this time
1024 * @interval:	the interval to forward
1025 *
1026 * Forward the timer expiry so it will expire in the future.
1027 * Returns the number of overruns.
1028 *
1029 * Can be safely called from the callback function of @timer. If
1030 * called from other contexts @timer must neither be enqueued nor
1031 * running the callback and the caller needs to take care of
1032 * serialization.
1033 *
1034 * Note: This only updates the timer expiry value and does not requeue
1035 * the timer.
1036 */
1037u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
1038{
1039	u64 orun = 1;
1040	ktime_t delta;
1041
1042	delta = ktime_sub(now, hrtimer_get_expires(timer));
1043
1044	if (delta < 0)
1045		return 0;
1046
1047	if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
1048		return 0;
1049
1050	if (interval < hrtimer_resolution)
1051		interval = hrtimer_resolution;
1052
1053	if (unlikely(delta >= interval)) {
1054		s64 incr = ktime_to_ns(interval);
1055
1056		orun = ktime_divns(delta, incr);
1057		hrtimer_add_expires_ns(timer, incr * orun);
1058		if (hrtimer_get_expires_tv64(timer) > now)
1059			return orun;
1060		/*
1061		 * This (and the ktime_add() below) is the
1062		 * correction for exact:
1063		 */
1064		orun++;
1065	}
1066	hrtimer_add_expires(timer, interval);
1067
1068	return orun;
1069}
1070EXPORT_SYMBOL_GPL(hrtimer_forward);
1071
1072/*
1073 * enqueue_hrtimer - internal function to (re)start a timer
1074 *
1075 * The timer is inserted in expiry order. Insertion into the
1076 * red black tree is O(log(n)). Must hold the base lock.
1077 *
1078 * Returns 1 when the new timer is the leftmost timer in the tree.
1079 */
1080static int enqueue_hrtimer(struct hrtimer *timer,
1081			   struct hrtimer_clock_base *base,
1082			   enum hrtimer_mode mode)
1083{
1084	debug_activate(timer, mode);
 
1085
1086	base->cpu_base->active_bases |= 1 << base->index;
1087
1088	/* Pairs with the lockless read in hrtimer_is_queued() */
1089	WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
1090
1091	return timerqueue_add(&base->active, &timer->node);
1092}
1093
1094/*
1095 * __remove_hrtimer - internal function to remove a timer
1096 *
1097 * Caller must hold the base lock.
1098 *
1099 * High resolution timer mode reprograms the clock event device when the
1100 * timer is the one which expires next. The caller can disable this by setting
1101 * reprogram to zero. This is useful, when the context does a reprogramming
1102 * anyway (e.g. timer interrupt)
1103 */
1104static void __remove_hrtimer(struct hrtimer *timer,
1105			     struct hrtimer_clock_base *base,
1106			     u8 newstate, int reprogram)
1107{
1108	struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1109	u8 state = timer->state;
1110
1111	/* Pairs with the lockless read in hrtimer_is_queued() */
1112	WRITE_ONCE(timer->state, newstate);
1113	if (!(state & HRTIMER_STATE_ENQUEUED))
1114		return;
1115
1116	if (!timerqueue_del(&base->active, &timer->node))
1117		cpu_base->active_bases &= ~(1 << base->index);
1118
1119	/*
1120	 * Note: If reprogram is false we do not update
1121	 * cpu_base->next_timer. This happens when we remove the first
1122	 * timer on a remote cpu. No harm as we never dereference
1123	 * cpu_base->next_timer. So the worst thing what can happen is
1124	 * an superfluous call to hrtimer_force_reprogram() on the
1125	 * remote cpu later on if the same timer gets enqueued again.
1126	 */
1127	if (reprogram && timer == cpu_base->next_timer)
1128		hrtimer_force_reprogram(cpu_base, 1);
1129}
1130
1131/*
1132 * remove hrtimer, called with base lock held
1133 */
1134static inline int
1135remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base,
1136	       bool restart, bool keep_local)
1137{
1138	u8 state = timer->state;
1139
1140	if (state & HRTIMER_STATE_ENQUEUED) {
1141		bool reprogram;
1142
1143		/*
1144		 * Remove the timer and force reprogramming when high
1145		 * resolution mode is active and the timer is on the current
1146		 * CPU. If we remove a timer on another CPU, reprogramming is
1147		 * skipped. The interrupt event on this CPU is fired and
1148		 * reprogramming happens in the interrupt handler. This is a
1149		 * rare case and less expensive than a smp call.
1150		 */
1151		debug_deactivate(timer);
1152		reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1153
1154		/*
1155		 * If the timer is not restarted then reprogramming is
1156		 * required if the timer is local. If it is local and about
1157		 * to be restarted, avoid programming it twice (on removal
1158		 * and a moment later when it's requeued).
1159		 */
1160		if (!restart)
1161			state = HRTIMER_STATE_INACTIVE;
1162		else
1163			reprogram &= !keep_local;
1164
1165		__remove_hrtimer(timer, base, state, reprogram);
1166		return 1;
1167	}
1168	return 0;
1169}
1170
1171static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1172					    const enum hrtimer_mode mode)
1173{
1174#ifdef CONFIG_TIME_LOW_RES
1175	/*
1176	 * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1177	 * granular time values. For relative timers we add hrtimer_resolution
1178	 * (i.e. one jiffie) to prevent short timeouts.
1179	 */
1180	timer->is_rel = mode & HRTIMER_MODE_REL;
1181	if (timer->is_rel)
1182		tim = ktime_add_safe(tim, hrtimer_resolution);
1183#endif
1184	return tim;
1185}
1186
1187static void
1188hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1189{
1190	ktime_t expires;
1191
1192	/*
1193	 * Find the next SOFT expiration.
1194	 */
1195	expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1196
1197	/*
1198	 * reprogramming needs to be triggered, even if the next soft
1199	 * hrtimer expires at the same time than the next hard
1200	 * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1201	 */
1202	if (expires == KTIME_MAX)
1203		return;
1204
1205	/*
1206	 * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1207	 * cpu_base->*expires_next is only set by hrtimer_reprogram()
1208	 */
1209	hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1210}
1211
1212static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1213				    u64 delta_ns, const enum hrtimer_mode mode,
1214				    struct hrtimer_clock_base *base)
1215{
1216	struct hrtimer_clock_base *new_base;
1217	bool force_local, first;
1218
1219	/*
1220	 * If the timer is on the local cpu base and is the first expiring
1221	 * timer then this might end up reprogramming the hardware twice
1222	 * (on removal and on enqueue). To avoid that by prevent the
1223	 * reprogram on removal, keep the timer local to the current CPU
1224	 * and enforce reprogramming after it is queued no matter whether
1225	 * it is the new first expiring timer again or not.
1226	 */
1227	force_local = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1228	force_local &= base->cpu_base->next_timer == timer;
1229
1230	/*
1231	 * Remove an active timer from the queue. In case it is not queued
1232	 * on the current CPU, make sure that remove_hrtimer() updates the
1233	 * remote data correctly.
1234	 *
1235	 * If it's on the current CPU and the first expiring timer, then
1236	 * skip reprogramming, keep the timer local and enforce
1237	 * reprogramming later if it was the first expiring timer.  This
1238	 * avoids programming the underlying clock event twice (once at
1239	 * removal and once after enqueue).
1240	 */
1241	remove_hrtimer(timer, base, true, force_local);
1242
1243	if (mode & HRTIMER_MODE_REL)
1244		tim = ktime_add_safe(tim, base->get_time());
1245
1246	tim = hrtimer_update_lowres(timer, tim, mode);
1247
1248	hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1249
1250	/* Switch the timer base, if necessary: */
1251	if (!force_local) {
1252		new_base = switch_hrtimer_base(timer, base,
1253					       mode & HRTIMER_MODE_PINNED);
1254	} else {
1255		new_base = base;
1256	}
1257
1258	first = enqueue_hrtimer(timer, new_base, mode);
1259	if (!force_local)
1260		return first;
1261
1262	/*
1263	 * Timer was forced to stay on the current CPU to avoid
1264	 * reprogramming on removal and enqueue. Force reprogram the
1265	 * hardware by evaluating the new first expiring timer.
1266	 */
1267	hrtimer_force_reprogram(new_base->cpu_base, 1);
1268	return 0;
1269}
1270
1271/**
1272 * hrtimer_start_range_ns - (re)start an hrtimer
1273 * @timer:	the timer to be added
1274 * @tim:	expiry time
1275 * @delta_ns:	"slack" range for the timer
1276 * @mode:	timer mode: absolute (HRTIMER_MODE_ABS) or
1277 *		relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1278 *		softirq based mode is considered for debug purpose only!
1279 */
1280void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1281			    u64 delta_ns, const enum hrtimer_mode mode)
1282{
1283	struct hrtimer_clock_base *base;
1284	unsigned long flags;
1285
1286	/*
1287	 * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1288	 * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1289	 * expiry mode because unmarked timers are moved to softirq expiry.
1290	 */
1291	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1292		WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1293	else
1294		WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1295
1296	base = lock_hrtimer_base(timer, &flags);
1297
1298	if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1299		hrtimer_reprogram(timer, true);
1300
1301	unlock_hrtimer_base(timer, &flags);
1302}
1303EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1304
1305/**
1306 * hrtimer_try_to_cancel - try to deactivate a timer
1307 * @timer:	hrtimer to stop
1308 *
1309 * Returns:
1310 *
1311 *  *  0 when the timer was not active
1312 *  *  1 when the timer was active
1313 *  * -1 when the timer is currently executing the callback function and
1314 *    cannot be stopped
1315 */
1316int hrtimer_try_to_cancel(struct hrtimer *timer)
1317{
1318	struct hrtimer_clock_base *base;
1319	unsigned long flags;
1320	int ret = -1;
1321
1322	/*
1323	 * Check lockless first. If the timer is not active (neither
1324	 * enqueued nor running the callback, nothing to do here.  The
1325	 * base lock does not serialize against a concurrent enqueue,
1326	 * so we can avoid taking it.
1327	 */
1328	if (!hrtimer_active(timer))
1329		return 0;
1330
1331	base = lock_hrtimer_base(timer, &flags);
1332
1333	if (!hrtimer_callback_running(timer))
1334		ret = remove_hrtimer(timer, base, false, false);
1335
1336	unlock_hrtimer_base(timer, &flags);
1337
1338	return ret;
1339
1340}
1341EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1342
1343#ifdef CONFIG_PREEMPT_RT
1344static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1345{
1346	spin_lock_init(&base->softirq_expiry_lock);
1347}
1348
1349static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1350{
1351	spin_lock(&base->softirq_expiry_lock);
1352}
1353
1354static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1355{
1356	spin_unlock(&base->softirq_expiry_lock);
1357}
1358
1359/*
1360 * The counterpart to hrtimer_cancel_wait_running().
1361 *
1362 * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1363 * the timer callback to finish. Drop expiry_lock and reacquire it. That
1364 * allows the waiter to acquire the lock and make progress.
1365 */
1366static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1367				      unsigned long flags)
1368{
1369	if (atomic_read(&cpu_base->timer_waiters)) {
1370		raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1371		spin_unlock(&cpu_base->softirq_expiry_lock);
1372		spin_lock(&cpu_base->softirq_expiry_lock);
1373		raw_spin_lock_irq(&cpu_base->lock);
1374	}
1375}
1376
1377/*
1378 * This function is called on PREEMPT_RT kernels when the fast path
1379 * deletion of a timer failed because the timer callback function was
1380 * running.
1381 *
1382 * This prevents priority inversion: if the soft irq thread is preempted
1383 * in the middle of a timer callback, then calling del_timer_sync() can
1384 * lead to two issues:
1385 *
1386 *  - If the caller is on a remote CPU then it has to spin wait for the timer
1387 *    handler to complete. This can result in unbound priority inversion.
1388 *
1389 *  - If the caller originates from the task which preempted the timer
1390 *    handler on the same CPU, then spin waiting for the timer handler to
1391 *    complete is never going to end.
1392 */
1393void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1394{
1395	/* Lockless read. Prevent the compiler from reloading it below */
1396	struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1397
1398	/*
1399	 * Just relax if the timer expires in hard interrupt context or if
1400	 * it is currently on the migration base.
1401	 */
1402	if (!timer->is_soft || is_migration_base(base)) {
1403		cpu_relax();
1404		return;
1405	}
1406
1407	/*
1408	 * Mark the base as contended and grab the expiry lock, which is
1409	 * held by the softirq across the timer callback. Drop the lock
1410	 * immediately so the softirq can expire the next timer. In theory
1411	 * the timer could already be running again, but that's more than
1412	 * unlikely and just causes another wait loop.
1413	 */
1414	atomic_inc(&base->cpu_base->timer_waiters);
1415	spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1416	atomic_dec(&base->cpu_base->timer_waiters);
1417	spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1418}
1419#else
1420static inline void
1421hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1422static inline void
1423hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1424static inline void
1425hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1426static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1427					     unsigned long flags) { }
1428#endif
1429
1430/**
1431 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1432 * @timer:	the timer to be cancelled
1433 *
1434 * Returns:
1435 *  0 when the timer was not active
1436 *  1 when the timer was active
1437 */
1438int hrtimer_cancel(struct hrtimer *timer)
1439{
1440	int ret;
1441
1442	do {
1443		ret = hrtimer_try_to_cancel(timer);
1444
1445		if (ret < 0)
1446			hrtimer_cancel_wait_running(timer);
1447	} while (ret < 0);
1448	return ret;
1449}
1450EXPORT_SYMBOL_GPL(hrtimer_cancel);
1451
1452/**
1453 * __hrtimer_get_remaining - get remaining time for the timer
1454 * @timer:	the timer to read
1455 * @adjust:	adjust relative timers when CONFIG_TIME_LOW_RES=y
1456 */
1457ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1458{
1459	unsigned long flags;
1460	ktime_t rem;
1461
1462	lock_hrtimer_base(timer, &flags);
1463	if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1464		rem = hrtimer_expires_remaining_adjusted(timer);
1465	else
1466		rem = hrtimer_expires_remaining(timer);
1467	unlock_hrtimer_base(timer, &flags);
1468
1469	return rem;
1470}
1471EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1472
1473#ifdef CONFIG_NO_HZ_COMMON
1474/**
1475 * hrtimer_get_next_event - get the time until next expiry event
1476 *
1477 * Returns the next expiry time or KTIME_MAX if no timer is pending.
1478 */
1479u64 hrtimer_get_next_event(void)
1480{
1481	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1482	u64 expires = KTIME_MAX;
1483	unsigned long flags;
1484
1485	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1486
1487	if (!__hrtimer_hres_active(cpu_base))
1488		expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1489
1490	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1491
1492	return expires;
1493}
1494
1495/**
1496 * hrtimer_next_event_without - time until next expiry event w/o one timer
1497 * @exclude:	timer to exclude
1498 *
1499 * Returns the next expiry time over all timers except for the @exclude one or
1500 * KTIME_MAX if none of them is pending.
1501 */
1502u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1503{
1504	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1505	u64 expires = KTIME_MAX;
1506	unsigned long flags;
1507
1508	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1509
1510	if (__hrtimer_hres_active(cpu_base)) {
1511		unsigned int active;
1512
1513		if (!cpu_base->softirq_activated) {
1514			active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1515			expires = __hrtimer_next_event_base(cpu_base, exclude,
1516							    active, KTIME_MAX);
1517		}
1518		active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1519		expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1520						    expires);
1521	}
1522
1523	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1524
1525	return expires;
1526}
1527#endif
1528
1529static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1530{
1531	if (likely(clock_id < MAX_CLOCKS)) {
1532		int base = hrtimer_clock_to_base_table[clock_id];
1533
1534		if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1535			return base;
1536	}
1537	WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1538	return HRTIMER_BASE_MONOTONIC;
1539}
1540
1541static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1542			   enum hrtimer_mode mode)
1543{
1544	bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1545	struct hrtimer_cpu_base *cpu_base;
1546	int base;
1547
1548	/*
1549	 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1550	 * marked for hard interrupt expiry mode are moved into soft
1551	 * interrupt context for latency reasons and because the callbacks
1552	 * can invoke functions which might sleep on RT, e.g. spin_lock().
1553	 */
1554	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1555		softtimer = true;
1556
1557	memset(timer, 0, sizeof(struct hrtimer));
1558
1559	cpu_base = raw_cpu_ptr(&hrtimer_bases);
1560
1561	/*
1562	 * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1563	 * clock modifications, so they needs to become CLOCK_MONOTONIC to
1564	 * ensure POSIX compliance.
1565	 */
1566	if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1567		clock_id = CLOCK_MONOTONIC;
1568
1569	base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1570	base += hrtimer_clockid_to_base(clock_id);
1571	timer->is_soft = softtimer;
1572	timer->is_hard = !!(mode & HRTIMER_MODE_HARD);
1573	timer->base = &cpu_base->clock_base[base];
1574	timerqueue_init(&timer->node);
1575}
1576
1577/**
1578 * hrtimer_init - initialize a timer to the given clock
1579 * @timer:	the timer to be initialized
1580 * @clock_id:	the clock to be used
1581 * @mode:       The modes which are relevant for initialization:
1582 *              HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1583 *              HRTIMER_MODE_REL_SOFT
1584 *
1585 *              The PINNED variants of the above can be handed in,
1586 *              but the PINNED bit is ignored as pinning happens
1587 *              when the hrtimer is started
1588 */
1589void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1590		  enum hrtimer_mode mode)
1591{
1592	debug_init(timer, clock_id, mode);
1593	__hrtimer_init(timer, clock_id, mode);
1594}
1595EXPORT_SYMBOL_GPL(hrtimer_init);
1596
1597/*
1598 * A timer is active, when it is enqueued into the rbtree or the
1599 * callback function is running or it's in the state of being migrated
1600 * to another cpu.
1601 *
1602 * It is important for this function to not return a false negative.
1603 */
1604bool hrtimer_active(const struct hrtimer *timer)
1605{
1606	struct hrtimer_clock_base *base;
1607	unsigned int seq;
1608
1609	do {
1610		base = READ_ONCE(timer->base);
1611		seq = raw_read_seqcount_begin(&base->seq);
1612
1613		if (timer->state != HRTIMER_STATE_INACTIVE ||
1614		    base->running == timer)
1615			return true;
1616
1617	} while (read_seqcount_retry(&base->seq, seq) ||
1618		 base != READ_ONCE(timer->base));
1619
1620	return false;
1621}
1622EXPORT_SYMBOL_GPL(hrtimer_active);
1623
1624/*
1625 * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1626 * distinct sections:
1627 *
1628 *  - queued:	the timer is queued
1629 *  - callback:	the timer is being ran
1630 *  - post:	the timer is inactive or (re)queued
1631 *
1632 * On the read side we ensure we observe timer->state and cpu_base->running
1633 * from the same section, if anything changed while we looked at it, we retry.
1634 * This includes timer->base changing because sequence numbers alone are
1635 * insufficient for that.
1636 *
1637 * The sequence numbers are required because otherwise we could still observe
1638 * a false negative if the read side got smeared over multiple consecutive
1639 * __run_hrtimer() invocations.
1640 */
1641
1642static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1643			  struct hrtimer_clock_base *base,
1644			  struct hrtimer *timer, ktime_t *now,
1645			  unsigned long flags) __must_hold(&cpu_base->lock)
1646{
1647	enum hrtimer_restart (*fn)(struct hrtimer *);
1648	bool expires_in_hardirq;
1649	int restart;
1650
1651	lockdep_assert_held(&cpu_base->lock);
1652
1653	debug_deactivate(timer);
1654	base->running = timer;
1655
1656	/*
1657	 * Separate the ->running assignment from the ->state assignment.
1658	 *
1659	 * As with a regular write barrier, this ensures the read side in
1660	 * hrtimer_active() cannot observe base->running == NULL &&
1661	 * timer->state == INACTIVE.
1662	 */
1663	raw_write_seqcount_barrier(&base->seq);
1664
1665	__remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1666	fn = timer->function;
1667
1668	/*
1669	 * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1670	 * timer is restarted with a period then it becomes an absolute
1671	 * timer. If its not restarted it does not matter.
1672	 */
1673	if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1674		timer->is_rel = false;
1675
1676	/*
1677	 * The timer is marked as running in the CPU base, so it is
1678	 * protected against migration to a different CPU even if the lock
1679	 * is dropped.
1680	 */
1681	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1682	trace_hrtimer_expire_entry(timer, now);
1683	expires_in_hardirq = lockdep_hrtimer_enter(timer);
1684
1685	restart = fn(timer);
1686
1687	lockdep_hrtimer_exit(expires_in_hardirq);
1688	trace_hrtimer_expire_exit(timer);
1689	raw_spin_lock_irq(&cpu_base->lock);
1690
1691	/*
1692	 * Note: We clear the running state after enqueue_hrtimer and
1693	 * we do not reprogram the event hardware. Happens either in
1694	 * hrtimer_start_range_ns() or in hrtimer_interrupt()
1695	 *
1696	 * Note: Because we dropped the cpu_base->lock above,
1697	 * hrtimer_start_range_ns() can have popped in and enqueued the timer
1698	 * for us already.
1699	 */
1700	if (restart != HRTIMER_NORESTART &&
1701	    !(timer->state & HRTIMER_STATE_ENQUEUED))
1702		enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1703
1704	/*
1705	 * Separate the ->running assignment from the ->state assignment.
1706	 *
1707	 * As with a regular write barrier, this ensures the read side in
1708	 * hrtimer_active() cannot observe base->running.timer == NULL &&
1709	 * timer->state == INACTIVE.
1710	 */
1711	raw_write_seqcount_barrier(&base->seq);
1712
1713	WARN_ON_ONCE(base->running != timer);
1714	base->running = NULL;
1715}
1716
1717static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1718				 unsigned long flags, unsigned int active_mask)
1719{
1720	struct hrtimer_clock_base *base;
1721	unsigned int active = cpu_base->active_bases & active_mask;
1722
1723	for_each_active_base(base, cpu_base, active) {
1724		struct timerqueue_node *node;
1725		ktime_t basenow;
1726
1727		basenow = ktime_add(now, base->offset);
1728
1729		while ((node = timerqueue_getnext(&base->active))) {
1730			struct hrtimer *timer;
1731
1732			timer = container_of(node, struct hrtimer, node);
1733
1734			/*
1735			 * The immediate goal for using the softexpires is
1736			 * minimizing wakeups, not running timers at the
1737			 * earliest interrupt after their soft expiration.
1738			 * This allows us to avoid using a Priority Search
1739			 * Tree, which can answer a stabbing query for
1740			 * overlapping intervals and instead use the simple
1741			 * BST we already have.
1742			 * We don't add extra wakeups by delaying timers that
1743			 * are right-of a not yet expired timer, because that
1744			 * timer will have to trigger a wakeup anyway.
1745			 */
1746			if (basenow < hrtimer_get_softexpires_tv64(timer))
1747				break;
1748
1749			__run_hrtimer(cpu_base, base, timer, &basenow, flags);
1750			if (active_mask == HRTIMER_ACTIVE_SOFT)
1751				hrtimer_sync_wait_running(cpu_base, flags);
1752		}
1753	}
1754}
1755
1756static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1757{
1758	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1759	unsigned long flags;
1760	ktime_t now;
1761
1762	hrtimer_cpu_base_lock_expiry(cpu_base);
1763	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1764
1765	now = hrtimer_update_base(cpu_base);
1766	__hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1767
1768	cpu_base->softirq_activated = 0;
1769	hrtimer_update_softirq_timer(cpu_base, true);
1770
1771	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1772	hrtimer_cpu_base_unlock_expiry(cpu_base);
1773}
1774
1775#ifdef CONFIG_HIGH_RES_TIMERS
1776
1777/*
1778 * High resolution timer interrupt
1779 * Called with interrupts disabled
1780 */
1781void hrtimer_interrupt(struct clock_event_device *dev)
1782{
1783	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1784	ktime_t expires_next, now, entry_time, delta;
1785	unsigned long flags;
1786	int retries = 0;
1787
1788	BUG_ON(!cpu_base->hres_active);
1789	cpu_base->nr_events++;
1790	dev->next_event = KTIME_MAX;
1791
1792	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1793	entry_time = now = hrtimer_update_base(cpu_base);
1794retry:
1795	cpu_base->in_hrtirq = 1;
1796	/*
1797	 * We set expires_next to KTIME_MAX here with cpu_base->lock
1798	 * held to prevent that a timer is enqueued in our queue via
1799	 * the migration code. This does not affect enqueueing of
1800	 * timers which run their callback and need to be requeued on
1801	 * this CPU.
1802	 */
1803	cpu_base->expires_next = KTIME_MAX;
1804
1805	if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1806		cpu_base->softirq_expires_next = KTIME_MAX;
1807		cpu_base->softirq_activated = 1;
1808		raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1809	}
1810
1811	__hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1812
1813	/* Reevaluate the clock bases for the [soft] next expiry */
1814	expires_next = hrtimer_update_next_event(cpu_base);
1815	/*
1816	 * Store the new expiry value so the migration code can verify
1817	 * against it.
1818	 */
1819	cpu_base->expires_next = expires_next;
1820	cpu_base->in_hrtirq = 0;
1821	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1822
1823	/* Reprogramming necessary ? */
1824	if (!tick_program_event(expires_next, 0)) {
1825		cpu_base->hang_detected = 0;
1826		return;
1827	}
1828
1829	/*
1830	 * The next timer was already expired due to:
1831	 * - tracing
1832	 * - long lasting callbacks
1833	 * - being scheduled away when running in a VM
1834	 *
1835	 * We need to prevent that we loop forever in the hrtimer
1836	 * interrupt routine. We give it 3 attempts to avoid
1837	 * overreacting on some spurious event.
1838	 *
1839	 * Acquire base lock for updating the offsets and retrieving
1840	 * the current time.
1841	 */
1842	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1843	now = hrtimer_update_base(cpu_base);
1844	cpu_base->nr_retries++;
1845	if (++retries < 3)
1846		goto retry;
1847	/*
1848	 * Give the system a chance to do something else than looping
1849	 * here. We stored the entry time, so we know exactly how long
1850	 * we spent here. We schedule the next event this amount of
1851	 * time away.
1852	 */
1853	cpu_base->nr_hangs++;
1854	cpu_base->hang_detected = 1;
1855	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1856
1857	delta = ktime_sub(now, entry_time);
1858	if ((unsigned int)delta > cpu_base->max_hang_time)
1859		cpu_base->max_hang_time = (unsigned int) delta;
1860	/*
1861	 * Limit it to a sensible value as we enforce a longer
1862	 * delay. Give the CPU at least 100ms to catch up.
1863	 */
1864	if (delta > 100 * NSEC_PER_MSEC)
1865		expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1866	else
1867		expires_next = ktime_add(now, delta);
1868	tick_program_event(expires_next, 1);
1869	pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1870}
1871
1872/* called with interrupts disabled */
1873static inline void __hrtimer_peek_ahead_timers(void)
1874{
1875	struct tick_device *td;
1876
1877	if (!hrtimer_hres_active())
1878		return;
1879
1880	td = this_cpu_ptr(&tick_cpu_device);
1881	if (td && td->evtdev)
1882		hrtimer_interrupt(td->evtdev);
1883}
1884
1885#else /* CONFIG_HIGH_RES_TIMERS */
1886
1887static inline void __hrtimer_peek_ahead_timers(void) { }
1888
1889#endif	/* !CONFIG_HIGH_RES_TIMERS */
1890
1891/*
1892 * Called from run_local_timers in hardirq context every jiffy
1893 */
1894void hrtimer_run_queues(void)
1895{
1896	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1897	unsigned long flags;
1898	ktime_t now;
1899
1900	if (__hrtimer_hres_active(cpu_base))
1901		return;
1902
1903	/*
1904	 * This _is_ ugly: We have to check periodically, whether we
1905	 * can switch to highres and / or nohz mode. The clocksource
1906	 * switch happens with xtime_lock held. Notification from
1907	 * there only sets the check bit in the tick_oneshot code,
1908	 * otherwise we might deadlock vs. xtime_lock.
1909	 */
1910	if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1911		hrtimer_switch_to_hres();
1912		return;
1913	}
1914
1915	raw_spin_lock_irqsave(&cpu_base->lock, flags);
1916	now = hrtimer_update_base(cpu_base);
1917
1918	if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1919		cpu_base->softirq_expires_next = KTIME_MAX;
1920		cpu_base->softirq_activated = 1;
1921		raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1922	}
1923
1924	__hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1925	raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1926}
1927
1928/*
1929 * Sleep related functions:
1930 */
1931static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1932{
1933	struct hrtimer_sleeper *t =
1934		container_of(timer, struct hrtimer_sleeper, timer);
1935	struct task_struct *task = t->task;
1936
1937	t->task = NULL;
1938	if (task)
1939		wake_up_process(task);
1940
1941	return HRTIMER_NORESTART;
1942}
1943
1944/**
1945 * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
1946 * @sl:		sleeper to be started
1947 * @mode:	timer mode abs/rel
1948 *
1949 * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
1950 * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
1951 */
1952void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
1953				   enum hrtimer_mode mode)
1954{
1955	/*
1956	 * Make the enqueue delivery mode check work on RT. If the sleeper
1957	 * was initialized for hard interrupt delivery, force the mode bit.
1958	 * This is a special case for hrtimer_sleepers because
1959	 * hrtimer_init_sleeper() determines the delivery mode on RT so the
1960	 * fiddling with this decision is avoided at the call sites.
1961	 */
1962	if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
1963		mode |= HRTIMER_MODE_HARD;
1964
1965	hrtimer_start_expires(&sl->timer, mode);
1966}
1967EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
1968
1969static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
1970				   clockid_t clock_id, enum hrtimer_mode mode)
1971{
1972	/*
1973	 * On PREEMPT_RT enabled kernels hrtimers which are not explicitly
1974	 * marked for hard interrupt expiry mode are moved into soft
1975	 * interrupt context either for latency reasons or because the
1976	 * hrtimer callback takes regular spinlocks or invokes other
1977	 * functions which are not suitable for hard interrupt context on
1978	 * PREEMPT_RT.
1979	 *
1980	 * The hrtimer_sleeper callback is RT compatible in hard interrupt
1981	 * context, but there is a latency concern: Untrusted userspace can
1982	 * spawn many threads which arm timers for the same expiry time on
1983	 * the same CPU. That causes a latency spike due to the wakeup of
1984	 * a gazillion threads.
1985	 *
1986	 * OTOH, privileged real-time user space applications rely on the
1987	 * low latency of hard interrupt wakeups. If the current task is in
1988	 * a real-time scheduling class, mark the mode for hard interrupt
1989	 * expiry.
1990	 */
1991	if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1992		if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT))
1993			mode |= HRTIMER_MODE_HARD;
1994	}
1995
1996	__hrtimer_init(&sl->timer, clock_id, mode);
1997	sl->timer.function = hrtimer_wakeup;
1998	sl->task = current;
1999}
2000
2001/**
2002 * hrtimer_init_sleeper - initialize sleeper to the given clock
2003 * @sl:		sleeper to be initialized
2004 * @clock_id:	the clock to be used
2005 * @mode:	timer mode abs/rel
2006 */
2007void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id,
2008			  enum hrtimer_mode mode)
2009{
2010	debug_init(&sl->timer, clock_id, mode);
2011	__hrtimer_init_sleeper(sl, clock_id, mode);
2012
2013}
2014EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
2015
2016int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
2017{
2018	switch(restart->nanosleep.type) {
2019#ifdef CONFIG_COMPAT_32BIT_TIME
2020	case TT_COMPAT:
2021		if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
2022			return -EFAULT;
2023		break;
2024#endif
2025	case TT_NATIVE:
2026		if (put_timespec64(ts, restart->nanosleep.rmtp))
2027			return -EFAULT;
2028		break;
2029	default:
2030		BUG();
2031	}
2032	return -ERESTART_RESTARTBLOCK;
2033}
2034
2035static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
2036{
2037	struct restart_block *restart;
2038
2039	do {
2040		set_current_state(TASK_INTERRUPTIBLE|TASK_FREEZABLE);
2041		hrtimer_sleeper_start_expires(t, mode);
2042
2043		if (likely(t->task))
2044			schedule();
2045
2046		hrtimer_cancel(&t->timer);
2047		mode = HRTIMER_MODE_ABS;
2048
2049	} while (t->task && !signal_pending(current));
2050
2051	__set_current_state(TASK_RUNNING);
2052
2053	if (!t->task)
2054		return 0;
2055
2056	restart = &current->restart_block;
2057	if (restart->nanosleep.type != TT_NONE) {
2058		ktime_t rem = hrtimer_expires_remaining(&t->timer);
2059		struct timespec64 rmt;
2060
2061		if (rem <= 0)
2062			return 0;
2063		rmt = ktime_to_timespec64(rem);
2064
2065		return nanosleep_copyout(restart, &rmt);
2066	}
2067	return -ERESTART_RESTARTBLOCK;
2068}
2069
2070static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
2071{
2072	struct hrtimer_sleeper t;
2073	int ret;
2074
2075	hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
2076				      HRTIMER_MODE_ABS);
2077	hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
2078	ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
2079	destroy_hrtimer_on_stack(&t.timer);
2080	return ret;
2081}
2082
2083long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
2084		       const clockid_t clockid)
2085{
2086	struct restart_block *restart;
2087	struct hrtimer_sleeper t;
2088	int ret = 0;
2089	u64 slack;
2090
2091	slack = current->timer_slack_ns;
2092	if (dl_task(current) || rt_task(current))
2093		slack = 0;
2094
2095	hrtimer_init_sleeper_on_stack(&t, clockid, mode);
2096	hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
2097	ret = do_nanosleep(&t, mode);
2098	if (ret != -ERESTART_RESTARTBLOCK)
2099		goto out;
2100
2101	/* Absolute timers do not update the rmtp value and restart: */
2102	if (mode == HRTIMER_MODE_ABS) {
2103		ret = -ERESTARTNOHAND;
2104		goto out;
2105	}
2106
2107	restart = &current->restart_block;
2108	restart->nanosleep.clockid = t.timer.base->clockid;
2109	restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
2110	set_restart_fn(restart, hrtimer_nanosleep_restart);
2111out:
2112	destroy_hrtimer_on_stack(&t.timer);
2113	return ret;
2114}
2115
2116#ifdef CONFIG_64BIT
2117
2118SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
2119		struct __kernel_timespec __user *, rmtp)
2120{
2121	struct timespec64 tu;
2122
2123	if (get_timespec64(&tu, rqtp))
2124		return -EFAULT;
2125
2126	if (!timespec64_valid(&tu))
2127		return -EINVAL;
2128
 
2129	current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
2130	current->restart_block.nanosleep.rmtp = rmtp;
2131	return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2132				 CLOCK_MONOTONIC);
2133}
2134
2135#endif
2136
2137#ifdef CONFIG_COMPAT_32BIT_TIME
2138
2139SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
2140		       struct old_timespec32 __user *, rmtp)
2141{
2142	struct timespec64 tu;
2143
2144	if (get_old_timespec32(&tu, rqtp))
2145		return -EFAULT;
2146
2147	if (!timespec64_valid(&tu))
2148		return -EINVAL;
2149
 
2150	current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
2151	current->restart_block.nanosleep.compat_rmtp = rmtp;
2152	return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2153				 CLOCK_MONOTONIC);
2154}
2155#endif
2156
2157/*
2158 * Functions related to boot-time initialization:
2159 */
2160int hrtimers_prepare_cpu(unsigned int cpu)
2161{
2162	struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
2163	int i;
2164
2165	for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2166		struct hrtimer_clock_base *clock_b = &cpu_base->clock_base[i];
2167
2168		clock_b->cpu_base = cpu_base;
2169		seqcount_raw_spinlock_init(&clock_b->seq, &cpu_base->lock);
2170		timerqueue_init_head(&clock_b->active);
2171	}
2172
2173	cpu_base->cpu = cpu;
2174	cpu_base->active_bases = 0;
2175	cpu_base->hres_active = 0;
2176	cpu_base->hang_detected = 0;
2177	cpu_base->next_timer = NULL;
2178	cpu_base->softirq_next_timer = NULL;
2179	cpu_base->expires_next = KTIME_MAX;
2180	cpu_base->softirq_expires_next = KTIME_MAX;
 
2181	hrtimer_cpu_base_init_expiry_lock(cpu_base);
2182	return 0;
2183}
2184
2185#ifdef CONFIG_HOTPLUG_CPU
2186
2187static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2188				struct hrtimer_clock_base *new_base)
2189{
2190	struct hrtimer *timer;
2191	struct timerqueue_node *node;
2192
2193	while ((node = timerqueue_getnext(&old_base->active))) {
2194		timer = container_of(node, struct hrtimer, node);
2195		BUG_ON(hrtimer_callback_running(timer));
2196		debug_deactivate(timer);
2197
2198		/*
2199		 * Mark it as ENQUEUED not INACTIVE otherwise the
2200		 * timer could be seen as !active and just vanish away
2201		 * under us on another CPU
2202		 */
2203		__remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2204		timer->base = new_base;
2205		/*
2206		 * Enqueue the timers on the new cpu. This does not
2207		 * reprogram the event device in case the timer
2208		 * expires before the earliest on this CPU, but we run
2209		 * hrtimer_interrupt after we migrated everything to
2210		 * sort out already expired timers and reprogram the
2211		 * event device.
2212		 */
2213		enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2214	}
2215}
2216
2217int hrtimers_dead_cpu(unsigned int scpu)
2218{
2219	struct hrtimer_cpu_base *old_base, *new_base;
2220	int i;
2221
2222	BUG_ON(cpu_online(scpu));
2223	tick_cancel_sched_timer(scpu);
 
 
2224
2225	/*
2226	 * this BH disable ensures that raise_softirq_irqoff() does
2227	 * not wakeup ksoftirqd (and acquire the pi-lock) while
2228	 * holding the cpu_base lock
2229	 */
2230	local_bh_disable();
2231	local_irq_disable();
2232	old_base = &per_cpu(hrtimer_bases, scpu);
2233	new_base = this_cpu_ptr(&hrtimer_bases);
2234	/*
2235	 * The caller is globally serialized and nobody else
2236	 * takes two locks at once, deadlock is not possible.
2237	 */
2238	raw_spin_lock(&new_base->lock);
2239	raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2240
2241	for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2242		migrate_hrtimer_list(&old_base->clock_base[i],
2243				     &new_base->clock_base[i]);
2244	}
2245
2246	/*
2247	 * The migration might have changed the first expiring softirq
2248	 * timer on this CPU. Update it.
2249	 */
2250	hrtimer_update_softirq_timer(new_base, false);
 
 
2251
2252	raw_spin_unlock(&old_base->lock);
2253	raw_spin_unlock(&new_base->lock);
 
 
2254
2255	/* Check, if we got expired work to do */
2256	__hrtimer_peek_ahead_timers();
2257	local_irq_enable();
2258	local_bh_enable();
2259	return 0;
2260}
2261
2262#endif /* CONFIG_HOTPLUG_CPU */
2263
2264void __init hrtimers_init(void)
2265{
2266	hrtimers_prepare_cpu(smp_processor_id());
2267	open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2268}
2269
2270/**
2271 * schedule_hrtimeout_range_clock - sleep until timeout
2272 * @expires:	timeout value (ktime_t)
2273 * @delta:	slack in expires timeout (ktime_t)
2274 * @mode:	timer mode
2275 * @clock_id:	timer clock to be used
2276 */
2277int __sched
2278schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
2279			       const enum hrtimer_mode mode, clockid_t clock_id)
2280{
2281	struct hrtimer_sleeper t;
2282
2283	/*
2284	 * Optimize when a zero timeout value is given. It does not
2285	 * matter whether this is an absolute or a relative time.
2286	 */
2287	if (expires && *expires == 0) {
2288		__set_current_state(TASK_RUNNING);
2289		return 0;
2290	}
2291
2292	/*
2293	 * A NULL parameter means "infinite"
2294	 */
2295	if (!expires) {
2296		schedule();
2297		return -EINTR;
2298	}
2299
 
 
 
 
 
 
 
2300	hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
2301	hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2302	hrtimer_sleeper_start_expires(&t, mode);
2303
2304	if (likely(t.task))
2305		schedule();
2306
2307	hrtimer_cancel(&t.timer);
2308	destroy_hrtimer_on_stack(&t.timer);
2309
2310	__set_current_state(TASK_RUNNING);
2311
2312	return !t.task ? 0 : -EINTR;
2313}
2314EXPORT_SYMBOL_GPL(schedule_hrtimeout_range_clock);
2315
2316/**
2317 * schedule_hrtimeout_range - sleep until timeout
2318 * @expires:	timeout value (ktime_t)
2319 * @delta:	slack in expires timeout (ktime_t)
2320 * @mode:	timer mode
2321 *
2322 * Make the current task sleep until the given expiry time has
2323 * elapsed. The routine will return immediately unless
2324 * the current task state has been set (see set_current_state()).
2325 *
2326 * The @delta argument gives the kernel the freedom to schedule the
2327 * actual wakeup to a time that is both power and performance friendly.
 
2328 * The kernel give the normal best effort behavior for "@expires+@delta",
2329 * but may decide to fire the timer earlier, but no earlier than @expires.
2330 *
2331 * You can set the task state as follows -
2332 *
2333 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2334 * pass before the routine returns unless the current task is explicitly
2335 * woken up, (e.g. by wake_up_process()).
2336 *
2337 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2338 * delivered to the current task or the current task is explicitly woken
2339 * up.
2340 *
2341 * The current task state is guaranteed to be TASK_RUNNING when this
2342 * routine returns.
2343 *
2344 * Returns 0 when the timer has expired. If the task was woken before the
2345 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2346 * by an explicit wakeup, it returns -EINTR.
2347 */
2348int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2349				     const enum hrtimer_mode mode)
2350{
2351	return schedule_hrtimeout_range_clock(expires, delta, mode,
2352					      CLOCK_MONOTONIC);
2353}
2354EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2355
2356/**
2357 * schedule_hrtimeout - sleep until timeout
2358 * @expires:	timeout value (ktime_t)
2359 * @mode:	timer mode
2360 *
2361 * Make the current task sleep until the given expiry time has
2362 * elapsed. The routine will return immediately unless
2363 * the current task state has been set (see set_current_state()).
2364 *
2365 * You can set the task state as follows -
2366 *
2367 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2368 * pass before the routine returns unless the current task is explicitly
2369 * woken up, (e.g. by wake_up_process()).
2370 *
2371 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2372 * delivered to the current task or the current task is explicitly woken
2373 * up.
2374 *
2375 * The current task state is guaranteed to be TASK_RUNNING when this
2376 * routine returns.
2377 *
2378 * Returns 0 when the timer has expired. If the task was woken before the
2379 * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2380 * by an explicit wakeup, it returns -EINTR.
2381 */
2382int __sched schedule_hrtimeout(ktime_t *expires,
2383			       const enum hrtimer_mode mode)
2384{
2385	return schedule_hrtimeout_range(expires, 0, mode);
2386}
2387EXPORT_SYMBOL_GPL(schedule_hrtimeout);