~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/time/hrtimer.c

Version: ~ [ linux-6.0-rc6 ] ~ [ linux-5.19.10 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.69 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.144 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.214 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.259 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.294 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.329 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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  */
 68 DEFINE_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 
116 static 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  */
137 static 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 
147 static 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  */
164 static
165 struct 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  */
192 static int
193 hrtimer_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 
201 static inline
202 struct 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  */
224 static inline struct hrtimer_clock_base *
225 switch_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);
234 again:
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 
276 static inline bool is_migration_base(struct hrtimer_clock_base *base)
277 {
278         return false;
279 }
280 
281 static inline struct hrtimer_clock_base *
282 lock_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  */
303 s64 __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 }
321 EXPORT_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  */
327 ktime_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 
341 EXPORT_SYMBOL_GPL(ktime_add_safe);
342 
343 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
344 
345 static const struct debug_obj_descr hrtimer_debug_descr;
346 
347 static 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  */
356 static 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  */
375 static 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  */
390 static 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 
404 static 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 
412 static inline void debug_hrtimer_init(struct hrtimer *timer)
413 {
414         debug_object_init(timer, &hrtimer_debug_descr);
415 }
416 
417 static inline void debug_hrtimer_activate(struct hrtimer *timer,
418                                           enum hrtimer_mode mode)
419 {
420         debug_object_activate(timer, &hrtimer_debug_descr);
421 }
422 
423 static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
424 {
425         debug_object_deactivate(timer, &hrtimer_debug_descr);
426 }
427 
428 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
429                            enum hrtimer_mode mode);
430 
431 void 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 }
437 EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
438 
439 static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
440                                    clockid_t clock_id, enum hrtimer_mode mode);
441 
442 void 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 }
448 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper_on_stack);
449 
450 void destroy_hrtimer_on_stack(struct hrtimer *timer)
451 {
452         debug_object_free(timer, &hrtimer_debug_descr);
453 }
454 EXPORT_SYMBOL_GPL(destroy_hrtimer_on_stack);
455 
456 #else
457 
458 static inline void debug_hrtimer_init(struct hrtimer *timer) { }
459 static inline void debug_hrtimer_activate(struct hrtimer *timer,
460                                           enum hrtimer_mode mode) { }
461 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
462 #endif
463 
464 static inline void
465 debug_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 
472 static 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 
479 static inline void debug_deactivate(struct hrtimer *timer)
480 {
481         debug_hrtimer_deactivate(timer);
482         trace_hrtimer_cancel(timer);
483 }
484 
485 static 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 
502 static 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  */
568 static 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 
594 static 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 
625 static 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  */
644 static 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 
650 static inline int hrtimer_hres_active(void)
651 {
652         return __hrtimer_hres_active(this_cpu_ptr(&hrtimer_bases));
653 }
654 
655 /*
656  * Reprogram the event source with checking both queues for the
657  * next event
658  * Called with interrupts disabled and base->lock held
659  */
660 static void
661 hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
662 {
663         ktime_t expires_next;
664 
665         expires_next = hrtimer_update_next_event(cpu_base);
666 
667         if (skip_equal && expires_next == cpu_base->expires_next)
668                 return;
669 
670         cpu_base->expires_next = expires_next;
671 
672         /*
673          * If hres is not active, hardware does not have to be
674          * reprogrammed yet.
675          *
676          * If a hang was detected in the last timer interrupt then we
677          * leave the hang delay active in the hardware. We want the
678          * system to make progress. That also prevents the following
679          * scenario:
680          * T1 expires 50ms from now
681          * T2 expires 5s from now
682          *
683          * T1 is removed, so this code is called and would reprogram
684          * the hardware to 5s from now. Any hrtimer_start after that
685          * will not reprogram the hardware due to hang_detected being
686          * set. So we'd effectivly block all timers until the T2 event
687          * fires.
688          */
689         if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
690                 return;
691 
692         tick_program_event(cpu_base->expires_next, 1);
693 }
694 
695 /* High resolution timer related functions */
696 #ifdef CONFIG_HIGH_RES_TIMERS
697 
698 /*
699  * High resolution timer enabled ?
700  */
701 static bool hrtimer_hres_enabled __read_mostly  = true;
702 unsigned int hrtimer_resolution __read_mostly = LOW_RES_NSEC;
703 EXPORT_SYMBOL_GPL(hrtimer_resolution);
704 
705 /*
706  * Enable / Disable high resolution mode
707  */
708 static int __init setup_hrtimer_hres(char *str)
709 {
710         return (kstrtobool(str, &hrtimer_hres_enabled) == 0);
711 }
712 
713 __setup("highres=", setup_hrtimer_hres);
714 
715 /*
716  * hrtimer_high_res_enabled - query, if the highres mode is enabled
717  */
718 static inline int hrtimer_is_hres_enabled(void)
719 {
720         return hrtimer_hres_enabled;
721 }
722 
723 /*
724  * Retrigger next event is called after clock was set
725  *
726  * Called with interrupts disabled via on_each_cpu()
727  */
728 static void retrigger_next_event(void *arg)
729 {
730         struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
731 
732         if (!__hrtimer_hres_active(base))
733                 return;
734 
735         raw_spin_lock(&base->lock);
736         hrtimer_update_base(base);
737         hrtimer_force_reprogram(base, 0);
738         raw_spin_unlock(&base->lock);
739 }
740 
741 /*
742  * Switch to high resolution mode
743  */
744 static void hrtimer_switch_to_hres(void)
745 {
746         struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
747 
748         if (tick_init_highres()) {
749                 pr_warn("Could not switch to high resolution mode on CPU %u\n",
750                         base->cpu);
751                 return;
752         }
753         base->hres_active = 1;
754         hrtimer_resolution = HIGH_RES_NSEC;
755 
756         tick_setup_sched_timer();
757         /* "Retrigger" the interrupt to get things going */
758         retrigger_next_event(NULL);
759 }
760 
761 static void clock_was_set_work(struct work_struct *work)
762 {
763         clock_was_set();
764 }
765 
766 static DECLARE_WORK(hrtimer_work, clock_was_set_work);
767 
768 /*
769  * Called from timekeeping and resume code to reprogram the hrtimer
770  * interrupt device on all cpus.
771  */
772 void clock_was_set_delayed(void)
773 {
774         schedule_work(&hrtimer_work);
775 }
776 
777 #else
778 
779 static inline int hrtimer_is_hres_enabled(void) { return 0; }
780 static inline void hrtimer_switch_to_hres(void) { }
781 static inline void retrigger_next_event(void *arg) { }
782 
783 #endif /* CONFIG_HIGH_RES_TIMERS */
784 
785 /*
786  * When a timer is enqueued and expires earlier than the already enqueued
787  * timers, we have to check, whether it expires earlier than the timer for
788  * which the clock event device was armed.
789  *
790  * Called with interrupts disabled and base->cpu_base.lock held
791  */
792 static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
793 {
794         struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
795         struct hrtimer_clock_base *base = timer->base;
796         ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
797 
798         WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
799 
800         /*
801          * CLOCK_REALTIME timer might be requested with an absolute
802          * expiry time which is less than base->offset. Set it to 0.
803          */
804         if (expires < 0)
805                 expires = 0;
806 
807         if (timer->is_soft) {
808                 /*
809                  * soft hrtimer could be started on a remote CPU. In this
810                  * case softirq_expires_next needs to be updated on the
811                  * remote CPU. The soft hrtimer will not expire before the
812                  * first hard hrtimer on the remote CPU -
813                  * hrtimer_check_target() prevents this case.
814                  */
815                 struct hrtimer_cpu_base *timer_cpu_base = base->cpu_base;
816 
817                 if (timer_cpu_base->softirq_activated)
818                         return;
819 
820                 if (!ktime_before(expires, timer_cpu_base->softirq_expires_next))
821                         return;
822 
823                 timer_cpu_base->softirq_next_timer = timer;
824                 timer_cpu_base->softirq_expires_next = expires;
825 
826                 if (!ktime_before(expires, timer_cpu_base->expires_next) ||
827                     !reprogram)
828                         return;
829         }
830 
831         /*
832          * If the timer is not on the current cpu, we cannot reprogram
833          * the other cpus clock event device.
834          */
835         if (base->cpu_base != cpu_base)
836                 return;
837 
838         /*
839          * If the hrtimer interrupt is running, then it will
840          * reevaluate the clock bases and reprogram the clock event
841          * device. The callbacks are always executed in hard interrupt
842          * context so we don't need an extra check for a running
843          * callback.
844          */
845         if (cpu_base->in_hrtirq)
846                 return;
847 
848         if (expires >= cpu_base->expires_next)
849                 return;
850 
851         /* Update the pointer to the next expiring timer */
852         cpu_base->next_timer = timer;
853         cpu_base->expires_next = expires;
854 
855         /*
856          * If hres is not active, hardware does not have to be
857          * programmed yet.
858          *
859          * If a hang was detected in the last timer interrupt then we
860          * do not schedule a timer which is earlier than the expiry
861          * which we enforced in the hang detection. We want the system
862          * to make progress.
863          */
864         if (!__hrtimer_hres_active(cpu_base) || cpu_base->hang_detected)
865                 return;
866 
867         /*
868          * Program the timer hardware. We enforce the expiry for
869          * events which are already in the past.
870          */
871         tick_program_event(expires, 1);
872 }
873 
874 /*
875  * Clock realtime was set
876  *
877  * Change the offset of the realtime clock vs. the monotonic
878  * clock.
879  *
880  * We might have to reprogram the high resolution timer interrupt. On
881  * SMP we call the architecture specific code to retrigger _all_ high
882  * resolution timer interrupts. On UP we just disable interrupts and
883  * call the high resolution interrupt code.
884  */
885 void clock_was_set(void)
886 {
887 #ifdef CONFIG_HIGH_RES_TIMERS
888         /* Retrigger the CPU local events everywhere */
889         on_each_cpu(retrigger_next_event, NULL, 1);
890 #endif
891         timerfd_clock_was_set();
892 }
893 
894 /*
895  * During resume we might have to reprogram the high resolution timer
896  * interrupt on all online CPUs.  However, all other CPUs will be
897  * stopped with IRQs interrupts disabled so the clock_was_set() call
898  * must be deferred.
899  */
900 void hrtimers_resume(void)
901 {
902         lockdep_assert_irqs_disabled();
903         /* Retrigger on the local CPU */
904         retrigger_next_event(NULL);
905         /* And schedule a retrigger for all others */
906         clock_was_set_delayed();
907 }
908 
909 /*
910  * Counterpart to lock_hrtimer_base above:
911  */
912 static inline
913 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
914 {
915         raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
916 }
917 
918 /**
919  * hrtimer_forward - forward the timer expiry
920  * @timer:      hrtimer to forward
921  * @now:        forward past this time
922  * @interval:   the interval to forward
923  *
924  * Forward the timer expiry so it will expire in the future.
925  * Returns the number of overruns.
926  *
927  * Can be safely called from the callback function of @timer. If
928  * called from other contexts @timer must neither be enqueued nor
929  * running the callback and the caller needs to take care of
930  * serialization.
931  *
932  * Note: This only updates the timer expiry value and does not requeue
933  * the timer.
934  */
935 u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
936 {
937         u64 orun = 1;
938         ktime_t delta;
939 
940         delta = ktime_sub(now, hrtimer_get_expires(timer));
941 
942         if (delta < 0)
943                 return 0;
944 
945         if (WARN_ON(timer->state & HRTIMER_STATE_ENQUEUED))
946                 return 0;
947 
948         if (interval < hrtimer_resolution)
949                 interval = hrtimer_resolution;
950 
951         if (unlikely(delta >= interval)) {
952                 s64 incr = ktime_to_ns(interval);
953 
954                 orun = ktime_divns(delta, incr);
955                 hrtimer_add_expires_ns(timer, incr * orun);
956                 if (hrtimer_get_expires_tv64(timer) > now)
957                         return orun;
958                 /*
959                  * This (and the ktime_add() below) is the
960                  * correction for exact:
961                  */
962                 orun++;
963         }
964         hrtimer_add_expires(timer, interval);
965 
966         return orun;
967 }
968 EXPORT_SYMBOL_GPL(hrtimer_forward);
969 
970 /*
971  * enqueue_hrtimer - internal function to (re)start a timer
972  *
973  * The timer is inserted in expiry order. Insertion into the
974  * red black tree is O(log(n)). Must hold the base lock.
975  *
976  * Returns 1 when the new timer is the leftmost timer in the tree.
977  */
978 static int enqueue_hrtimer(struct hrtimer *timer,
979                            struct hrtimer_clock_base *base,
980                            enum hrtimer_mode mode)
981 {
982         debug_activate(timer, mode);
983 
984         base->cpu_base->active_bases |= 1 << base->index;
985 
986         /* Pairs with the lockless read in hrtimer_is_queued() */
987         WRITE_ONCE(timer->state, HRTIMER_STATE_ENQUEUED);
988 
989         return timerqueue_add(&base->active, &timer->node);
990 }
991 
992 /*
993  * __remove_hrtimer - internal function to remove a timer
994  *
995  * Caller must hold the base lock.
996  *
997  * High resolution timer mode reprograms the clock event device when the
998  * timer is the one which expires next. The caller can disable this by setting
999  * reprogram to zero. This is useful, when the context does a reprogramming
1000  * anyway (e.g. timer interrupt)
1001  */
1002 static void __remove_hrtimer(struct hrtimer *timer,
1003                              struct hrtimer_clock_base *base,
1004                              u8 newstate, int reprogram)
1005 {
1006         struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1007         u8 state = timer->state;
1008 
1009         /* Pairs with the lockless read in hrtimer_is_queued() */
1010         WRITE_ONCE(timer->state, newstate);
1011         if (!(state & HRTIMER_STATE_ENQUEUED))
1012                 return;
1013 
1014         if (!timerqueue_del(&base->active, &timer->node))
1015                 cpu_base->active_bases &= ~(1 << base->index);
1016 
1017         /*
1018          * Note: If reprogram is false we do not update
1019          * cpu_base->next_timer. This happens when we remove the first
1020          * timer on a remote cpu. No harm as we never dereference
1021          * cpu_base->next_timer. So the worst thing what can happen is
1022          * an superflous call to hrtimer_force_reprogram() on the
1023          * remote cpu later on if the same timer gets enqueued again.
1024          */
1025         if (reprogram && timer == cpu_base->next_timer)
1026                 hrtimer_force_reprogram(cpu_base, 1);
1027 }
1028 
1029 /*
1030  * remove hrtimer, called with base lock held
1031  */
1032 static inline int
1033 remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base, bool restart)
1034 {
1035         u8 state = timer->state;
1036 
1037         if (state & HRTIMER_STATE_ENQUEUED) {
1038                 int reprogram;
1039 
1040                 /*
1041                  * Remove the timer and force reprogramming when high
1042                  * resolution mode is active and the timer is on the current
1043                  * CPU. If we remove a timer on another CPU, reprogramming is
1044                  * skipped. The interrupt event on this CPU is fired and
1045                  * reprogramming happens in the interrupt handler. This is a
1046                  * rare case and less expensive than a smp call.
1047                  */
1048                 debug_deactivate(timer);
1049                 reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
1050 
1051                 if (!restart)
1052                         state = HRTIMER_STATE_INACTIVE;
1053 
1054                 __remove_hrtimer(timer, base, state, reprogram);
1055                 return 1;
1056         }
1057         return 0;
1058 }
1059 
1060 static inline ktime_t hrtimer_update_lowres(struct hrtimer *timer, ktime_t tim,
1061                                             const enum hrtimer_mode mode)
1062 {
1063 #ifdef CONFIG_TIME_LOW_RES
1064         /*
1065          * CONFIG_TIME_LOW_RES indicates that the system has no way to return
1066          * granular time values. For relative timers we add hrtimer_resolution
1067          * (i.e. one jiffie) to prevent short timeouts.
1068          */
1069         timer->is_rel = mode & HRTIMER_MODE_REL;
1070         if (timer->is_rel)
1071                 tim = ktime_add_safe(tim, hrtimer_resolution);
1072 #endif
1073         return tim;
1074 }
1075 
1076 static void
1077 hrtimer_update_softirq_timer(struct hrtimer_cpu_base *cpu_base, bool reprogram)
1078 {
1079         ktime_t expires;
1080 
1081         /*
1082          * Find the next SOFT expiration.
1083          */
1084         expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_SOFT);
1085 
1086         /*
1087          * reprogramming needs to be triggered, even if the next soft
1088          * hrtimer expires at the same time than the next hard
1089          * hrtimer. cpu_base->softirq_expires_next needs to be updated!
1090          */
1091         if (expires == KTIME_MAX)
1092                 return;
1093 
1094         /*
1095          * cpu_base->*next_timer is recomputed by __hrtimer_get_next_event()
1096          * cpu_base->*expires_next is only set by hrtimer_reprogram()
1097          */
1098         hrtimer_reprogram(cpu_base->softirq_next_timer, reprogram);
1099 }
1100 
1101 static int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1102                                     u64 delta_ns, const enum hrtimer_mode mode,
1103                                     struct hrtimer_clock_base *base)
1104 {
1105         struct hrtimer_clock_base *new_base;
1106 
1107         /* Remove an active timer from the queue: */
1108         remove_hrtimer(timer, base, true);
1109 
1110         if (mode & HRTIMER_MODE_REL)
1111                 tim = ktime_add_safe(tim, base->get_time());
1112 
1113         tim = hrtimer_update_lowres(timer, tim, mode);
1114 
1115         hrtimer_set_expires_range_ns(timer, tim, delta_ns);
1116 
1117         /* Switch the timer base, if necessary: */
1118         new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
1119 
1120         return enqueue_hrtimer(timer, new_base, mode);
1121 }
1122 
1123 /**
1124  * hrtimer_start_range_ns - (re)start an hrtimer
1125  * @timer:      the timer to be added
1126  * @tim:        expiry time
1127  * @delta_ns:   "slack" range for the timer
1128  * @mode:       timer mode: absolute (HRTIMER_MODE_ABS) or
1129  *              relative (HRTIMER_MODE_REL), and pinned (HRTIMER_MODE_PINNED);
1130  *              softirq based mode is considered for debug purpose only!
1131  */
1132 void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1133                             u64 delta_ns, const enum hrtimer_mode mode)
1134 {
1135         struct hrtimer_clock_base *base;
1136         unsigned long flags;
1137 
1138         /*
1139          * Check whether the HRTIMER_MODE_SOFT bit and hrtimer.is_soft
1140          * match on CONFIG_PREEMPT_RT = n. With PREEMPT_RT check the hard
1141          * expiry mode because unmarked timers are moved to softirq expiry.
1142          */
1143         if (!IS_ENABLED(CONFIG_PREEMPT_RT))
1144                 WARN_ON_ONCE(!(mode & HRTIMER_MODE_SOFT) ^ !timer->is_soft);
1145         else
1146                 WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard);
1147 
1148         base = lock_hrtimer_base(timer, &flags);
1149 
1150         if (__hrtimer_start_range_ns(timer, tim, delta_ns, mode, base))
1151                 hrtimer_reprogram(timer, true);
1152 
1153         unlock_hrtimer_base(timer, &flags);
1154 }
1155 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1156 
1157 /**
1158  * hrtimer_try_to_cancel - try to deactivate a timer
1159  * @timer:      hrtimer to stop
1160  *
1161  * Returns:
1162  *
1163  *  *  0 when the timer was not active
1164  *  *  1 when the timer was active
1165  *  * -1 when the timer is currently executing the callback function and
1166  *    cannot be stopped
1167  */
1168 int hrtimer_try_to_cancel(struct hrtimer *timer)
1169 {
1170         struct hrtimer_clock_base *base;
1171         unsigned long flags;
1172         int ret = -1;
1173 
1174         /*
1175          * Check lockless first. If the timer is not active (neither
1176          * enqueued nor running the callback, nothing to do here.  The
1177          * base lock does not serialize against a concurrent enqueue,
1178          * so we can avoid taking it.
1179          */
1180         if (!hrtimer_active(timer))
1181                 return 0;
1182 
1183         base = lock_hrtimer_base(timer, &flags);
1184 
1185         if (!hrtimer_callback_running(timer))
1186                 ret = remove_hrtimer(timer, base, false);
1187 
1188         unlock_hrtimer_base(timer, &flags);
1189 
1190         return ret;
1191 
1192 }
1193 EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1194 
1195 #ifdef CONFIG_PREEMPT_RT
1196 static void hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base)
1197 {
1198         spin_lock_init(&base->softirq_expiry_lock);
1199 }
1200 
1201 static void hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base)
1202 {
1203         spin_lock(&base->softirq_expiry_lock);
1204 }
1205 
1206 static void hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base)
1207 {
1208         spin_unlock(&base->softirq_expiry_lock);
1209 }
1210 
1211 /*
1212  * The counterpart to hrtimer_cancel_wait_running().
1213  *
1214  * If there is a waiter for cpu_base->expiry_lock, then it was waiting for
1215  * the timer callback to finish. Drop expiry_lock and reaquire it. That
1216  * allows the waiter to acquire the lock and make progress.
1217  */
1218 static void hrtimer_sync_wait_running(struct hrtimer_cpu_base *cpu_base,
1219                                       unsigned long flags)
1220 {
1221         if (atomic_read(&cpu_base->timer_waiters)) {
1222                 raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1223                 spin_unlock(&cpu_base->softirq_expiry_lock);
1224                 spin_lock(&cpu_base->softirq_expiry_lock);
1225                 raw_spin_lock_irq(&cpu_base->lock);
1226         }
1227 }
1228 
1229 /*
1230  * This function is called on PREEMPT_RT kernels when the fast path
1231  * deletion of a timer failed because the timer callback function was
1232  * running.
1233  *
1234  * This prevents priority inversion: if the soft irq thread is preempted
1235  * in the middle of a timer callback, then calling del_timer_sync() can
1236  * lead to two issues:
1237  *
1238  *  - If the caller is on a remote CPU then it has to spin wait for the timer
1239  *    handler to complete. This can result in unbound priority inversion.
1240  *
1241  *  - If the caller originates from the task which preempted the timer
1242  *    handler on the same CPU, then spin waiting for the timer handler to
1243  *    complete is never going to end.
1244  */
1245 void hrtimer_cancel_wait_running(const struct hrtimer *timer)
1246 {
1247         /* Lockless read. Prevent the compiler from reloading it below */
1248         struct hrtimer_clock_base *base = READ_ONCE(timer->base);
1249 
1250         /*
1251          * Just relax if the timer expires in hard interrupt context or if
1252          * it is currently on the migration base.
1253          */
1254         if (!timer->is_soft || is_migration_base(base)) {
1255                 cpu_relax();
1256                 return;
1257         }
1258 
1259         /*
1260          * Mark the base as contended and grab the expiry lock, which is
1261          * held by the softirq across the timer callback. Drop the lock
1262          * immediately so the softirq can expire the next timer. In theory
1263          * the timer could already be running again, but that's more than
1264          * unlikely and just causes another wait loop.
1265          */
1266         atomic_inc(&base->cpu_base->timer_waiters);
1267         spin_lock_bh(&base->cpu_base->softirq_expiry_lock);
1268         atomic_dec(&base->cpu_base->timer_waiters);
1269         spin_unlock_bh(&base->cpu_base->softirq_expiry_lock);
1270 }
1271 #else
1272 static inline void
1273 hrtimer_cpu_base_init_expiry_lock(struct hrtimer_cpu_base *base) { }
1274 static inline void
1275 hrtimer_cpu_base_lock_expiry(struct hrtimer_cpu_base *base) { }
1276 static inline void
1277 hrtimer_cpu_base_unlock_expiry(struct hrtimer_cpu_base *base) { }
1278 static inline void hrtimer_sync_wait_running(struct hrtimer_cpu_base *base,
1279                                              unsigned long flags) { }
1280 #endif
1281 
1282 /**
1283  * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1284  * @timer:      the timer to be cancelled
1285  *
1286  * Returns:
1287  *  0 when the timer was not active
1288  *  1 when the timer was active
1289  */
1290 int hrtimer_cancel(struct hrtimer *timer)
1291 {
1292         int ret;
1293 
1294         do {
1295                 ret = hrtimer_try_to_cancel(timer);
1296 
1297                 if (ret < 0)
1298                         hrtimer_cancel_wait_running(timer);
1299         } while (ret < 0);
1300         return ret;
1301 }
1302 EXPORT_SYMBOL_GPL(hrtimer_cancel);
1303 
1304 /**
1305  * __hrtimer_get_remaining - get remaining time for the timer
1306  * @timer:      the timer to read
1307  * @adjust:     adjust relative timers when CONFIG_TIME_LOW_RES=y
1308  */
1309 ktime_t __hrtimer_get_remaining(const struct hrtimer *timer, bool adjust)
1310 {
1311         unsigned long flags;
1312         ktime_t rem;
1313 
1314         lock_hrtimer_base(timer, &flags);
1315         if (IS_ENABLED(CONFIG_TIME_LOW_RES) && adjust)
1316                 rem = hrtimer_expires_remaining_adjusted(timer);
1317         else
1318                 rem = hrtimer_expires_remaining(timer);
1319         unlock_hrtimer_base(timer, &flags);
1320 
1321         return rem;
1322 }
1323 EXPORT_SYMBOL_GPL(__hrtimer_get_remaining);
1324 
1325 #ifdef CONFIG_NO_HZ_COMMON
1326 /**
1327  * hrtimer_get_next_event - get the time until next expiry event
1328  *
1329  * Returns the next expiry time or KTIME_MAX if no timer is pending.
1330  */
1331 u64 hrtimer_get_next_event(void)
1332 {
1333         struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1334         u64 expires = KTIME_MAX;
1335         unsigned long flags;
1336 
1337         raw_spin_lock_irqsave(&cpu_base->lock, flags);
1338 
1339         if (!__hrtimer_hres_active(cpu_base))
1340                 expires = __hrtimer_get_next_event(cpu_base, HRTIMER_ACTIVE_ALL);
1341 
1342         raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1343 
1344         return expires;
1345 }
1346 
1347 /**
1348  * hrtimer_next_event_without - time until next expiry event w/o one timer
1349  * @exclude:    timer to exclude
1350  *
1351  * Returns the next expiry time over all timers except for the @exclude one or
1352  * KTIME_MAX if none of them is pending.
1353  */
1354 u64 hrtimer_next_event_without(const struct hrtimer *exclude)
1355 {
1356         struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1357         u64 expires = KTIME_MAX;
1358         unsigned long flags;
1359 
1360         raw_spin_lock_irqsave(&cpu_base->lock, flags);
1361 
1362         if (__hrtimer_hres_active(cpu_base)) {
1363                 unsigned int active;
1364 
1365                 if (!cpu_base->softirq_activated) {
1366                         active = cpu_base->active_bases & HRTIMER_ACTIVE_SOFT;
1367                         expires = __hrtimer_next_event_base(cpu_base, exclude,
1368                                                             active, KTIME_MAX);
1369                 }
1370                 active = cpu_base->active_bases & HRTIMER_ACTIVE_HARD;
1371                 expires = __hrtimer_next_event_base(cpu_base, exclude, active,
1372                                                     expires);
1373         }
1374 
1375         raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1376 
1377         return expires;
1378 }
1379 #endif
1380 
1381 static inline int hrtimer_clockid_to_base(clockid_t clock_id)
1382 {
1383         if (likely(clock_id < MAX_CLOCKS)) {
1384                 int base = hrtimer_clock_to_base_table[clock_id];
1385 
1386                 if (likely(base != HRTIMER_MAX_CLOCK_BASES))
1387                         return base;
1388         }
1389         WARN(1, "Invalid clockid %d. Using MONOTONIC\n", clock_id);
1390         return HRTIMER_BASE_MONOTONIC;
1391 }
1392 
1393 static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1394                            enum hrtimer_mode mode)
1395 {
1396         bool softtimer = !!(mode & HRTIMER_MODE_SOFT);
1397         struct hrtimer_cpu_base *cpu_base;
1398         int base;
1399 
1400         /*
1401          * On PREEMPT_RT enabled kernels hrtimers which are not explicitely
1402          * marked for hard interrupt expiry mode are moved into soft
1403          * interrupt context for latency reasons and because the callbacks
1404          * can invoke functions which might sleep on RT, e.g. spin_lock().
1405          */
1406         if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(mode & HRTIMER_MODE_HARD))
1407                 softtimer = true;
1408 
1409         memset(timer, 0, sizeof(struct hrtimer));
1410 
1411         cpu_base = raw_cpu_ptr(&hrtimer_bases);
1412 
1413         /*
1414          * POSIX magic: Relative CLOCK_REALTIME timers are not affected by
1415          * clock modifications, so they needs to become CLOCK_MONOTONIC to
1416          * ensure POSIX compliance.
1417          */
1418         if (clock_id == CLOCK_REALTIME && mode & HRTIMER_MODE_REL)
1419                 clock_id = CLOCK_MONOTONIC;
1420 
1421         base = softtimer ? HRTIMER_MAX_CLOCK_BASES / 2 : 0;
1422         base += hrtimer_clockid_to_base(clock_id);
1423         timer->is_soft = softtimer;
1424         timer->is_hard = !!(mode & HRTIMER_MODE_HARD);
1425         timer->base = &cpu_base->clock_base[base];
1426         timerqueue_init(&timer->node);
1427 }
1428 
1429 /**
1430  * hrtimer_init - initialize a timer to the given clock
1431  * @timer:      the timer to be initialized
1432  * @clock_id:   the clock to be used
1433  * @mode:       The modes which are relevant for intitialization:
1434  *              HRTIMER_MODE_ABS, HRTIMER_MODE_REL, HRTIMER_MODE_ABS_SOFT,
1435  *              HRTIMER_MODE_REL_SOFT
1436  *
1437  *              The PINNED variants of the above can be handed in,
1438  *              but the PINNED bit is ignored as pinning happens
1439  *              when the hrtimer is started
1440  */
1441 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1442                   enum hrtimer_mode mode)
1443 {
1444         debug_init(timer, clock_id, mode);
1445         __hrtimer_init(timer, clock_id, mode);
1446 }
1447 EXPORT_SYMBOL_GPL(hrtimer_init);
1448 
1449 /*
1450  * A timer is active, when it is enqueued into the rbtree or the
1451  * callback function is running or it's in the state of being migrated
1452  * to another cpu.
1453  *
1454  * It is important for this function to not return a false negative.
1455  */
1456 bool hrtimer_active(const struct hrtimer *timer)
1457 {
1458         struct hrtimer_clock_base *base;
1459         unsigned int seq;
1460 
1461         do {
1462                 base = READ_ONCE(timer->base);
1463                 seq = raw_read_seqcount_begin(&base->seq);
1464 
1465                 if (timer->state != HRTIMER_STATE_INACTIVE ||
1466                     base->running == timer)
1467                         return true;
1468 
1469         } while (read_seqcount_retry(&base->seq, seq) ||
1470                  base != READ_ONCE(timer->base));
1471 
1472         return false;
1473 }
1474 EXPORT_SYMBOL_GPL(hrtimer_active);
1475 
1476 /*
1477  * The write_seqcount_barrier()s in __run_hrtimer() split the thing into 3
1478  * distinct sections:
1479  *
1480  *  - queued:   the timer is queued
1481  *  - callback: the timer is being ran
1482  *  - post:     the timer is inactive or (re)queued
1483  *
1484  * On the read side we ensure we observe timer->state and cpu_base->running
1485  * from the same section, if anything changed while we looked at it, we retry.
1486  * This includes timer->base changing because sequence numbers alone are
1487  * insufficient for that.
1488  *
1489  * The sequence numbers are required because otherwise we could still observe
1490  * a false negative if the read side got smeared over multiple consequtive
1491  * __run_hrtimer() invocations.
1492  */
1493 
1494 static void __run_hrtimer(struct hrtimer_cpu_base *cpu_base,
1495                           struct hrtimer_clock_base *base,
1496                           struct hrtimer *timer, ktime_t *now,
1497                           unsigned long flags) __must_hold(&cpu_base->lock)
1498 {
1499         enum hrtimer_restart (*fn)(struct hrtimer *);
1500         bool expires_in_hardirq;
1501         int restart;
1502 
1503         lockdep_assert_held(&cpu_base->lock);
1504 
1505         debug_deactivate(timer);
1506         base->running = timer;
1507 
1508         /*
1509          * Separate the ->running assignment from the ->state assignment.
1510          *
1511          * As with a regular write barrier, this ensures the read side in
1512          * hrtimer_active() cannot observe base->running == NULL &&
1513          * timer->state == INACTIVE.
1514          */
1515         raw_write_seqcount_barrier(&base->seq);
1516 
1517         __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 0);
1518         fn = timer->function;
1519 
1520         /*
1521          * Clear the 'is relative' flag for the TIME_LOW_RES case. If the
1522          * timer is restarted with a period then it becomes an absolute
1523          * timer. If its not restarted it does not matter.
1524          */
1525         if (IS_ENABLED(CONFIG_TIME_LOW_RES))
1526                 timer->is_rel = false;
1527 
1528         /*
1529          * The timer is marked as running in the CPU base, so it is
1530          * protected against migration to a different CPU even if the lock
1531          * is dropped.
1532          */
1533         raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1534         trace_hrtimer_expire_entry(timer, now);
1535         expires_in_hardirq = lockdep_hrtimer_enter(timer);
1536 
1537         restart = fn(timer);
1538 
1539         lockdep_hrtimer_exit(expires_in_hardirq);
1540         trace_hrtimer_expire_exit(timer);
1541         raw_spin_lock_irq(&cpu_base->lock);
1542 
1543         /*
1544          * Note: We clear the running state after enqueue_hrtimer and
1545          * we do not reprogram the event hardware. Happens either in
1546          * hrtimer_start_range_ns() or in hrtimer_interrupt()
1547          *
1548          * Note: Because we dropped the cpu_base->lock above,
1549          * hrtimer_start_range_ns() can have popped in and enqueued the timer
1550          * for us already.
1551          */
1552         if (restart != HRTIMER_NORESTART &&
1553             !(timer->state & HRTIMER_STATE_ENQUEUED))
1554                 enqueue_hrtimer(timer, base, HRTIMER_MODE_ABS);
1555 
1556         /*
1557          * Separate the ->running assignment from the ->state assignment.
1558          *
1559          * As with a regular write barrier, this ensures the read side in
1560          * hrtimer_active() cannot observe base->running.timer == NULL &&
1561          * timer->state == INACTIVE.
1562          */
1563         raw_write_seqcount_barrier(&base->seq);
1564 
1565         WARN_ON_ONCE(base->running != timer);
1566         base->running = NULL;
1567 }
1568 
1569 static void __hrtimer_run_queues(struct hrtimer_cpu_base *cpu_base, ktime_t now,
1570                                  unsigned long flags, unsigned int active_mask)
1571 {
1572         struct hrtimer_clock_base *base;
1573         unsigned int active = cpu_base->active_bases & active_mask;
1574 
1575         for_each_active_base(base, cpu_base, active) {
1576                 struct timerqueue_node *node;
1577                 ktime_t basenow;
1578 
1579                 basenow = ktime_add(now, base->offset);
1580 
1581                 while ((node = timerqueue_getnext(&base->active))) {
1582                         struct hrtimer *timer;
1583 
1584                         timer = container_of(node, struct hrtimer, node);
1585 
1586                         /*
1587                          * The immediate goal for using the softexpires is
1588                          * minimizing wakeups, not running timers at the
1589                          * earliest interrupt after their soft expiration.
1590                          * This allows us to avoid using a Priority Search
1591                          * Tree, which can answer a stabbing querry for
1592                          * overlapping intervals and instead use the simple
1593                          * BST we already have.
1594                          * We don't add extra wakeups by delaying timers that
1595                          * are right-of a not yet expired timer, because that
1596                          * timer will have to trigger a wakeup anyway.
1597                          */
1598                         if (basenow < hrtimer_get_softexpires_tv64(timer))
1599                                 break;
1600 
1601                         __run_hrtimer(cpu_base, base, timer, &basenow, flags);
1602                         if (active_mask == HRTIMER_ACTIVE_SOFT)
1603                                 hrtimer_sync_wait_running(cpu_base, flags);
1604                 }
1605         }
1606 }
1607 
1608 static __latent_entropy void hrtimer_run_softirq(struct softirq_action *h)
1609 {
1610         struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1611         unsigned long flags;
1612         ktime_t now;
1613 
1614         hrtimer_cpu_base_lock_expiry(cpu_base);
1615         raw_spin_lock_irqsave(&cpu_base->lock, flags);
1616 
1617         now = hrtimer_update_base(cpu_base);
1618         __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_SOFT);
1619 
1620         cpu_base->softirq_activated = 0;
1621         hrtimer_update_softirq_timer(cpu_base, true);
1622 
1623         raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1624         hrtimer_cpu_base_unlock_expiry(cpu_base);
1625 }
1626 
1627 #ifdef CONFIG_HIGH_RES_TIMERS
1628 
1629 /*
1630  * High resolution timer interrupt
1631  * Called with interrupts disabled
1632  */
1633 void hrtimer_interrupt(struct clock_event_device *dev)
1634 {
1635         struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1636         ktime_t expires_next, now, entry_time, delta;
1637         unsigned long flags;
1638         int retries = 0;
1639 
1640         BUG_ON(!cpu_base->hres_active);
1641         cpu_base->nr_events++;
1642         dev->next_event = KTIME_MAX;
1643 
1644         raw_spin_lock_irqsave(&cpu_base->lock, flags);
1645         entry_time = now = hrtimer_update_base(cpu_base);
1646 retry:
1647         cpu_base->in_hrtirq = 1;
1648         /*
1649          * We set expires_next to KTIME_MAX here with cpu_base->lock
1650          * held to prevent that a timer is enqueued in our queue via
1651          * the migration code. This does not affect enqueueing of
1652          * timers which run their callback and need to be requeued on
1653          * this CPU.
1654          */
1655         cpu_base->expires_next = KTIME_MAX;
1656 
1657         if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1658                 cpu_base->softirq_expires_next = KTIME_MAX;
1659                 cpu_base->softirq_activated = 1;
1660                 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1661         }
1662 
1663         __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1664 
1665         /* Reevaluate the clock bases for the [soft] next expiry */
1666         expires_next = hrtimer_update_next_event(cpu_base);
1667         /*
1668          * Store the new expiry value so the migration code can verify
1669          * against it.
1670          */
1671         cpu_base->expires_next = expires_next;
1672         cpu_base->in_hrtirq = 0;
1673         raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1674 
1675         /* Reprogramming necessary ? */
1676         if (!tick_program_event(expires_next, 0)) {
1677                 cpu_base->hang_detected = 0;
1678                 return;
1679         }
1680 
1681         /*
1682          * The next timer was already expired due to:
1683          * - tracing
1684          * - long lasting callbacks
1685          * - being scheduled away when running in a VM
1686          *
1687          * We need to prevent that we loop forever in the hrtimer
1688          * interrupt routine. We give it 3 attempts to avoid
1689          * overreacting on some spurious event.
1690          *
1691          * Acquire base lock for updating the offsets and retrieving
1692          * the current time.
1693          */
1694         raw_spin_lock_irqsave(&cpu_base->lock, flags);
1695         now = hrtimer_update_base(cpu_base);
1696         cpu_base->nr_retries++;
1697         if (++retries < 3)
1698                 goto retry;
1699         /*
1700          * Give the system a chance to do something else than looping
1701          * here. We stored the entry time, so we know exactly how long
1702          * we spent here. We schedule the next event this amount of
1703          * time away.
1704          */
1705         cpu_base->nr_hangs++;
1706         cpu_base->hang_detected = 1;
1707         raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1708 
1709         delta = ktime_sub(now, entry_time);
1710         if ((unsigned int)delta > cpu_base->max_hang_time)
1711                 cpu_base->max_hang_time = (unsigned int) delta;
1712         /*
1713          * Limit it to a sensible value as we enforce a longer
1714          * delay. Give the CPU at least 100ms to catch up.
1715          */
1716         if (delta > 100 * NSEC_PER_MSEC)
1717                 expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1718         else
1719                 expires_next = ktime_add(now, delta);
1720         tick_program_event(expires_next, 1);
1721         pr_warn_once("hrtimer: interrupt took %llu ns\n", ktime_to_ns(delta));
1722 }
1723 
1724 /* called with interrupts disabled */
1725 static inline void __hrtimer_peek_ahead_timers(void)
1726 {
1727         struct tick_device *td;
1728 
1729         if (!hrtimer_hres_active())
1730                 return;
1731 
1732         td = this_cpu_ptr(&tick_cpu_device);
1733         if (td && td->evtdev)
1734                 hrtimer_interrupt(td->evtdev);
1735 }
1736 
1737 #else /* CONFIG_HIGH_RES_TIMERS */
1738 
1739 static inline void __hrtimer_peek_ahead_timers(void) { }
1740 
1741 #endif  /* !CONFIG_HIGH_RES_TIMERS */
1742 
1743 /*
1744  * Called from run_local_timers in hardirq context every jiffy
1745  */
1746 void hrtimer_run_queues(void)
1747 {
1748         struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
1749         unsigned long flags;
1750         ktime_t now;
1751 
1752         if (__hrtimer_hres_active(cpu_base))
1753                 return;
1754 
1755         /*
1756          * This _is_ ugly: We have to check periodically, whether we
1757          * can switch to highres and / or nohz mode. The clocksource
1758          * switch happens with xtime_lock held. Notification from
1759          * there only sets the check bit in the tick_oneshot code,
1760          * otherwise we might deadlock vs. xtime_lock.
1761          */
1762         if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) {
1763                 hrtimer_switch_to_hres();
1764                 return;
1765         }
1766 
1767         raw_spin_lock_irqsave(&cpu_base->lock, flags);
1768         now = hrtimer_update_base(cpu_base);
1769 
1770         if (!ktime_before(now, cpu_base->softirq_expires_next)) {
1771                 cpu_base->softirq_expires_next = KTIME_MAX;
1772                 cpu_base->softirq_activated = 1;
1773                 raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1774         }
1775 
1776         __hrtimer_run_queues(cpu_base, now, flags, HRTIMER_ACTIVE_HARD);
1777         raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1778 }
1779 
1780 /*
1781  * Sleep related functions:
1782  */
1783 static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1784 {
1785         struct hrtimer_sleeper *t =
1786                 container_of(timer, struct hrtimer_sleeper, timer);
1787         struct task_struct *task = t->task;
1788 
1789         t->task = NULL;
1790         if (task)
1791                 wake_up_process(task);
1792 
1793         return HRTIMER_NORESTART;
1794 }
1795 
1796 /**
1797  * hrtimer_sleeper_start_expires - Start a hrtimer sleeper timer
1798  * @sl:         sleeper to be started
1799  * @mode:       timer mode abs/rel
1800  *
1801  * Wrapper around hrtimer_start_expires() for hrtimer_sleeper based timers
1802  * to allow PREEMPT_RT to tweak the delivery mode (soft/hardirq context)
1803  */
1804 void hrtimer_sleeper_start_expires(struct hrtimer_sleeper *sl,
1805                                    enum hrtimer_mode mode)
1806 {
1807         /*
1808          * Make the enqueue delivery mode check work on RT. If the sleeper
1809          * was initialized for hard interrupt delivery, force the mode bit.
1810          * This is a special case for hrtimer_sleepers because
1811          * hrtimer_init_sleeper() determines the delivery mode on RT so the
1812          * fiddling with this decision is avoided at the call sites.
1813          */
1814         if (IS_ENABLED(CONFIG_PREEMPT_RT) && sl->timer.is_hard)
1815                 mode |= HRTIMER_MODE_HARD;
1816 
1817         hrtimer_start_expires(&sl->timer, mode);
1818 }
1819 EXPORT_SYMBOL_GPL(hrtimer_sleeper_start_expires);
1820 
1821 static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
1822                                    clockid_t clock_id, enum hrtimer_mode mode)
1823 {
1824         /*
1825          * On PREEMPT_RT enabled kernels hrtimers which are not explicitely
1826          * marked for hard interrupt expiry mode are moved into soft
1827          * interrupt context either for latency reasons or because the
1828          * hrtimer callback takes regular spinlocks or invokes other
1829          * functions which are not suitable for hard interrupt context on
1830          * PREEMPT_RT.
1831          *
1832          * The hrtimer_sleeper callback is RT compatible in hard interrupt
1833          * context, but there is a latency concern: Untrusted userspace can
1834          * spawn many threads which arm timers for the same expiry time on
1835          * the same CPU. That causes a latency spike due to the wakeup of
1836          * a gazillion threads.
1837          *
1838          * OTOH, priviledged real-time user space applications rely on the
1839          * low latency of hard interrupt wakeups. If the current task is in
1840          * a real-time scheduling class, mark the mode for hard interrupt
1841          * expiry.
1842          */
1843         if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1844                 if (task_is_realtime(current) && !(mode & HRTIMER_MODE_SOFT))
1845                         mode |= HRTIMER_MODE_HARD;
1846         }
1847 
1848         __hrtimer_init(&sl->timer, clock_id, mode);
1849         sl->timer.function = hrtimer_wakeup;
1850         sl->task = current;
1851 }
1852 
1853 /**
1854  * hrtimer_init_sleeper - initialize sleeper to the given clock
1855  * @sl:         sleeper to be initialized
1856  * @clock_id:   the clock to be used
1857  * @mode:       timer mode abs/rel
1858  */
1859 void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, clockid_t clock_id,
1860                           enum hrtimer_mode mode)
1861 {
1862         debug_init(&sl->timer, clock_id, mode);
1863         __hrtimer_init_sleeper(sl, clock_id, mode);
1864 
1865 }
1866 EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1867 
1868 int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
1869 {
1870         switch(restart->nanosleep.type) {
1871 #ifdef CONFIG_COMPAT_32BIT_TIME
1872         case TT_COMPAT:
1873                 if (put_old_timespec32(ts, restart->nanosleep.compat_rmtp))
1874                         return -EFAULT;
1875                 break;
1876 #endif
1877         case TT_NATIVE:
1878                 if (put_timespec64(ts, restart->nanosleep.rmtp))
1879                         return -EFAULT;
1880                 break;
1881         default:
1882                 BUG();
1883         }
1884         return -ERESTART_RESTARTBLOCK;
1885 }
1886 
1887 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1888 {
1889         struct restart_block *restart;
1890 
1891         do {
1892                 set_current_state(TASK_INTERRUPTIBLE);
1893                 hrtimer_sleeper_start_expires(t, mode);
1894 
1895                 if (likely(t->task))
1896                         freezable_schedule();
1897 
1898                 hrtimer_cancel(&t->timer);
1899                 mode = HRTIMER_MODE_ABS;
1900 
1901         } while (t->task && !signal_pending(current));
1902 
1903         __set_current_state(TASK_RUNNING);
1904 
1905         if (!t->task)
1906                 return 0;
1907 
1908         restart = &current->restart_block;
1909         if (restart->nanosleep.type != TT_NONE) {
1910                 ktime_t rem = hrtimer_expires_remaining(&t->timer);
1911                 struct timespec64 rmt;
1912 
1913                 if (rem <= 0)
1914                         return 0;
1915                 rmt = ktime_to_timespec64(rem);
1916 
1917                 return nanosleep_copyout(restart, &rmt);
1918         }
1919         return -ERESTART_RESTARTBLOCK;
1920 }
1921 
1922 static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1923 {
1924         struct hrtimer_sleeper t;
1925         int ret;
1926 
1927         hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
1928                                       HRTIMER_MODE_ABS);
1929         hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1930         ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
1931         destroy_hrtimer_on_stack(&t.timer);
1932         return ret;
1933 }
1934 
1935 long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
1936                        const clockid_t clockid)
1937 {
1938         struct restart_block *restart;
1939         struct hrtimer_sleeper t;
1940         int ret = 0;
1941         u64 slack;
1942 
1943         slack = current->timer_slack_ns;
1944         if (dl_task(current) || rt_task(current))
1945                 slack = 0;
1946 
1947         hrtimer_init_sleeper_on_stack(&t, clockid, mode);
1948         hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
1949         ret = do_nanosleep(&t, mode);
1950         if (ret != -ERESTART_RESTARTBLOCK)
1951                 goto out;
1952 
1953         /* Absolute timers do not update the rmtp value and restart: */
1954         if (mode == HRTIMER_MODE_ABS) {
1955                 ret = -ERESTARTNOHAND;
1956                 goto out;
1957         }
1958 
1959         restart = &current->restart_block;
1960         restart->nanosleep.clockid = t.timer.base->clockid;
1961         restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1962         set_restart_fn(restart, hrtimer_nanosleep_restart);
1963 out:
1964         destroy_hrtimer_on_stack(&t.timer);
1965         return ret;
1966 }
1967 
1968 #ifdef CONFIG_64BIT
1969 
1970 SYSCALL_DEFINE2(nanosleep, struct __kernel_timespec __user *, rqtp,
1971                 struct __kernel_timespec __user *, rmtp)
1972 {
1973         struct timespec64 tu;
1974 
1975         if (get_timespec64(&tu, rqtp))
1976                 return -EFAULT;
1977 
1978         if (!timespec64_valid(&tu))
1979                 return -EINVAL;
1980 
1981         current->restart_block.nanosleep.type = rmtp ? TT_NATIVE : TT_NONE;
1982         current->restart_block.nanosleep.rmtp = rmtp;
1983         return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
1984                                  CLOCK_MONOTONIC);
1985 }
1986 
1987 #endif
1988 
1989 #ifdef CONFIG_COMPAT_32BIT_TIME
1990 
1991 SYSCALL_DEFINE2(nanosleep_time32, struct old_timespec32 __user *, rqtp,
1992                        struct old_timespec32 __user *, rmtp)
1993 {
1994         struct timespec64 tu;
1995 
1996         if (get_old_timespec32(&tu, rqtp))
1997                 return -EFAULT;
1998 
1999         if (!timespec64_valid(&tu))
2000                 return -EINVAL;
2001 
2002         current->restart_block.nanosleep.type = rmtp ? TT_COMPAT : TT_NONE;
2003         current->restart_block.nanosleep.compat_rmtp = rmtp;
2004         return hrtimer_nanosleep(timespec64_to_ktime(tu), HRTIMER_MODE_REL,
2005                                  CLOCK_MONOTONIC);
2006 }
2007 #endif
2008 
2009 /*
2010  * Functions related to boot-time initialization:
2011  */
2012 int hrtimers_prepare_cpu(unsigned int cpu)
2013 {
2014         struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
2015         int i;
2016 
2017         for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2018                 struct hrtimer_clock_base *clock_b = &cpu_base->clock_base[i];
2019 
2020                 clock_b->cpu_base = cpu_base;
2021                 seqcount_raw_spinlock_init(&clock_b->seq, &cpu_base->lock);
2022                 timerqueue_init_head(&clock_b->active);
2023         }
2024 
2025         cpu_base->cpu = cpu;
2026         cpu_base->active_bases = 0;
2027         cpu_base->hres_active = 0;
2028         cpu_base->hang_detected = 0;
2029         cpu_base->next_timer = NULL;
2030         cpu_base->softirq_next_timer = NULL;
2031         cpu_base->expires_next = KTIME_MAX;
2032         cpu_base->softirq_expires_next = KTIME_MAX;
2033         hrtimer_cpu_base_init_expiry_lock(cpu_base);
2034         return 0;
2035 }
2036 
2037 #ifdef CONFIG_HOTPLUG_CPU
2038 
2039 static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
2040                                 struct hrtimer_clock_base *new_base)
2041 {
2042         struct hrtimer *timer;
2043         struct timerqueue_node *node;
2044 
2045         while ((node = timerqueue_getnext(&old_base->active))) {
2046                 timer = container_of(node, struct hrtimer, node);
2047                 BUG_ON(hrtimer_callback_running(timer));
2048                 debug_deactivate(timer);
2049 
2050                 /*
2051                  * Mark it as ENQUEUED not INACTIVE otherwise the
2052                  * timer could be seen as !active and just vanish away
2053                  * under us on another CPU
2054                  */
2055                 __remove_hrtimer(timer, old_base, HRTIMER_STATE_ENQUEUED, 0);
2056                 timer->base = new_base;
2057                 /*
2058                  * Enqueue the timers on the new cpu. This does not
2059                  * reprogram the event device in case the timer
2060                  * expires before the earliest on this CPU, but we run
2061                  * hrtimer_interrupt after we migrated everything to
2062                  * sort out already expired timers and reprogram the
2063                  * event device.
2064                  */
2065                 enqueue_hrtimer(timer, new_base, HRTIMER_MODE_ABS);
2066         }
2067 }
2068 
2069 int hrtimers_dead_cpu(unsigned int scpu)
2070 {
2071         struct hrtimer_cpu_base *old_base, *new_base;
2072         int i;
2073 
2074         BUG_ON(cpu_online(scpu));
2075         tick_cancel_sched_timer(scpu);
2076 
2077         /*
2078          * this BH disable ensures that raise_softirq_irqoff() does
2079          * not wakeup ksoftirqd (and acquire the pi-lock) while
2080          * holding the cpu_base lock
2081          */
2082         local_bh_disable();
2083         local_irq_disable();
2084         old_base = &per_cpu(hrtimer_bases, scpu);
2085         new_base = this_cpu_ptr(&hrtimer_bases);
2086         /*
2087          * The caller is globally serialized and nobody else
2088          * takes two locks at once, deadlock is not possible.
2089          */
2090         raw_spin_lock(&new_base->lock);
2091         raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
2092 
2093         for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
2094                 migrate_hrtimer_list(&old_base->clock_base[i],
2095                                      &new_base->clock_base[i]);
2096         }
2097 
2098         /*
2099          * The migration might have changed the first expiring softirq
2100          * timer on this CPU. Update it.
2101          */
2102         hrtimer_update_softirq_timer(new_base, false);
2103 
2104         raw_spin_unlock(&old_base->lock);
2105         raw_spin_unlock(&new_base->lock);
2106 
2107         /* Check, if we got expired work to do */
2108         __hrtimer_peek_ahead_timers();
2109         local_irq_enable();
2110         local_bh_enable();
2111         return 0;
2112 }
2113 
2114 #endif /* CONFIG_HOTPLUG_CPU */
2115 
2116 void __init hrtimers_init(void)
2117 {
2118         hrtimers_prepare_cpu(smp_processor_id());
2119         open_softirq(HRTIMER_SOFTIRQ, hrtimer_run_softirq);
2120 }
2121 
2122 /**
2123  * schedule_hrtimeout_range_clock - sleep until timeout
2124  * @expires:    timeout value (ktime_t)
2125  * @delta:      slack in expires timeout (ktime_t)
2126  * @mode:       timer mode
2127  * @clock_id:   timer clock to be used
2128  */
2129 int __sched
2130 schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
2131                                const enum hrtimer_mode mode, clockid_t clock_id)
2132 {
2133         struct hrtimer_sleeper t;
2134 
2135         /*
2136          * Optimize when a zero timeout value is given. It does not
2137          * matter whether this is an absolute or a relative time.
2138          */
2139         if (expires && *expires == 0) {
2140                 __set_current_state(TASK_RUNNING);
2141                 return 0;
2142         }
2143 
2144         /*
2145          * A NULL parameter means "infinite"
2146          */
2147         if (!expires) {
2148                 schedule();
2149                 return -EINTR;
2150         }
2151 
2152         hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
2153         hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
2154         hrtimer_sleeper_start_expires(&t, mode);
2155 
2156         if (likely(t.task))
2157                 schedule();
2158 
2159         hrtimer_cancel(&t.timer);
2160         destroy_hrtimer_on_stack(&t.timer);
2161 
2162         __set_current_state(TASK_RUNNING);
2163 
2164         return !t.task ? 0 : -EINTR;
2165 }
2166 
2167 /**
2168  * schedule_hrtimeout_range - sleep until timeout
2169  * @expires:    timeout value (ktime_t)
2170  * @delta:      slack in expires timeout (ktime_t)
2171  * @mode:       timer mode
2172  *
2173  * Make the current task sleep until the given expiry time has
2174  * elapsed. The routine will return immediately unless
2175  * the current task state has been set (see set_current_state()).
2176  *
2177  * The @delta argument gives the kernel the freedom to schedule the
2178  * actual wakeup to a time that is both power and performance friendly.
2179  * The kernel give the normal best effort behavior for "@expires+@delta",
2180  * but may decide to fire the timer earlier, but no earlier than @expires.
2181  *
2182  * You can set the task state as follows -
2183  *
2184  * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2185  * pass before the routine returns unless the current task is explicitly
2186  * woken up, (e.g. by wake_up_process()).
2187  *
2188  * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2189  * delivered to the current task or the current task is explicitly woken
2190  * up.
2191  *
2192  * The current task state is guaranteed to be TASK_RUNNING when this
2193  * routine returns.
2194  *
2195  * Returns 0 when the timer has expired. If the task was woken before the
2196  * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2197  * by an explicit wakeup, it returns -EINTR.
2198  */
2199 int __sched schedule_hrtimeout_range(ktime_t *expires, u64 delta,
2200                                      const enum hrtimer_mode mode)
2201 {
2202         return schedule_hrtimeout_range_clock(expires, delta, mode,
2203                                               CLOCK_MONOTONIC);
2204 }
2205 EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
2206 
2207 /**
2208  * schedule_hrtimeout - sleep until timeout
2209  * @expires:    timeout value (ktime_t)
2210  * @mode:       timer mode
2211  *
2212  * Make the current task sleep until the given expiry time has
2213  * elapsed. The routine will return immediately unless
2214  * the current task state has been set (see set_current_state()).
2215  *
2216  * You can set the task state as follows -
2217  *
2218  * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
2219  * pass before the routine returns unless the current task is explicitly
2220  * woken up, (e.g. by wake_up_process()).
2221  *
2222  * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
2223  * delivered to the current task or the current task is explicitly woken
2224  * up.
2225  *
2226  * The current task state is guaranteed to be TASK_RUNNING when this
2227  * routine returns.
2228  *
2229  * Returns 0 when the timer has expired. If the task was woken before the
2230  * timer expired by a signal (only possible in state TASK_INTERRUPTIBLE) or
2231  * by an explicit wakeup, it returns -EINTR.
2232  */
2233 int __sched schedule_hrtimeout(ktime_t *expires,
2234                                const enum hrtimer_mode mode)
2235 {
2236         return schedule_hrtimeout_range(expires, 0, mode);
2237 }
2238 EXPORT_SYMBOL_GPL(schedule_hrtimeout);
2239 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp