Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/kernel/time/clocksource.c

  1 /*
  2  * linux/kernel/time/clocksource.c
  3  *
  4  * This file contains the functions which manage clocksource drivers.
  5  *
  6  * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
  7  *
  8  * This program is free software; you can redistribute it and/or modify
  9  * it under the terms of the GNU General Public License as published by
 10  * the Free Software Foundation; either version 2 of the License, or
 11  * (at your option) any later version.
 12  *
 13  * This program is distributed in the hope that it will be useful,
 14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 16  * GNU General Public License for more details.
 17  *
 18  * You should have received a copy of the GNU General Public License
 19  * along with this program; if not, write to the Free Software
 20  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 21  *
 22  * TODO WishList:
 23  *   o Allow clocksource drivers to be unregistered
 24  */
 25 
 26 #include <linux/device.h>
 27 #include <linux/clocksource.h>
 28 #include <linux/init.h>
 29 #include <linux/module.h>
 30 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
 31 #include <linux/tick.h>
 32 #include <linux/kthread.h>
 33 
 34 #include "tick-internal.h"
 35 #include "timekeeping_internal.h"
 36 
 37 void timecounter_init(struct timecounter *tc,
 38                       const struct cyclecounter *cc,
 39                       u64 start_tstamp)
 40 {
 41         tc->cc = cc;
 42         tc->cycle_last = cc->read(cc);
 43         tc->nsec = start_tstamp;
 44 }
 45 EXPORT_SYMBOL_GPL(timecounter_init);
 46 
 47 /**
 48  * timecounter_read_delta - get nanoseconds since last call of this function
 49  * @tc:         Pointer to time counter
 50  *
 51  * When the underlying cycle counter runs over, this will be handled
 52  * correctly as long as it does not run over more than once between
 53  * calls.
 54  *
 55  * The first call to this function for a new time counter initializes
 56  * the time tracking and returns an undefined result.
 57  */
 58 static u64 timecounter_read_delta(struct timecounter *tc)
 59 {
 60         cycle_t cycle_now, cycle_delta;
 61         u64 ns_offset;
 62 
 63         /* read cycle counter: */
 64         cycle_now = tc->cc->read(tc->cc);
 65 
 66         /* calculate the delta since the last timecounter_read_delta(): */
 67         cycle_delta = (cycle_now - tc->cycle_last) & tc->cc->mask;
 68 
 69         /* convert to nanoseconds: */
 70         ns_offset = cyclecounter_cyc2ns(tc->cc, cycle_delta);
 71 
 72         /* update time stamp of timecounter_read_delta() call: */
 73         tc->cycle_last = cycle_now;
 74 
 75         return ns_offset;
 76 }
 77 
 78 u64 timecounter_read(struct timecounter *tc)
 79 {
 80         u64 nsec;
 81 
 82         /* increment time by nanoseconds since last call */
 83         nsec = timecounter_read_delta(tc);
 84         nsec += tc->nsec;
 85         tc->nsec = nsec;
 86 
 87         return nsec;
 88 }
 89 EXPORT_SYMBOL_GPL(timecounter_read);
 90 
 91 u64 timecounter_cyc2time(struct timecounter *tc,
 92                          cycle_t cycle_tstamp)
 93 {
 94         u64 cycle_delta = (cycle_tstamp - tc->cycle_last) & tc->cc->mask;
 95         u64 nsec;
 96 
 97         /*
 98          * Instead of always treating cycle_tstamp as more recent
 99          * than tc->cycle_last, detect when it is too far in the
100          * future and treat it as old time stamp instead.
101          */
102         if (cycle_delta > tc->cc->mask / 2) {
103                 cycle_delta = (tc->cycle_last - cycle_tstamp) & tc->cc->mask;
104                 nsec = tc->nsec - cyclecounter_cyc2ns(tc->cc, cycle_delta);
105         } else {
106                 nsec = cyclecounter_cyc2ns(tc->cc, cycle_delta) + tc->nsec;
107         }
108 
109         return nsec;
110 }
111 EXPORT_SYMBOL_GPL(timecounter_cyc2time);
112 
113 /**
114  * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
115  * @mult:       pointer to mult variable
116  * @shift:      pointer to shift variable
117  * @from:       frequency to convert from
118  * @to:         frequency to convert to
119  * @maxsec:     guaranteed runtime conversion range in seconds
120  *
121  * The function evaluates the shift/mult pair for the scaled math
122  * operations of clocksources and clockevents.
123  *
124  * @to and @from are frequency values in HZ. For clock sources @to is
125  * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
126  * event @to is the counter frequency and @from is NSEC_PER_SEC.
127  *
128  * The @maxsec conversion range argument controls the time frame in
129  * seconds which must be covered by the runtime conversion with the
130  * calculated mult and shift factors. This guarantees that no 64bit
131  * overflow happens when the input value of the conversion is
132  * multiplied with the calculated mult factor. Larger ranges may
133  * reduce the conversion accuracy by chosing smaller mult and shift
134  * factors.
135  */
136 void
137 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
138 {
139         u64 tmp;
140         u32 sft, sftacc= 32;
141 
142         /*
143          * Calculate the shift factor which is limiting the conversion
144          * range:
145          */
146         tmp = ((u64)maxsec * from) >> 32;
147         while (tmp) {
148                 tmp >>=1;
149                 sftacc--;
150         }
151 
152         /*
153          * Find the conversion shift/mult pair which has the best
154          * accuracy and fits the maxsec conversion range:
155          */
156         for (sft = 32; sft > 0; sft--) {
157                 tmp = (u64) to << sft;
158                 tmp += from / 2;
159                 do_div(tmp, from);
160                 if ((tmp >> sftacc) == 0)
161                         break;
162         }
163         *mult = tmp;
164         *shift = sft;
165 }
166 
167 /*[Clocksource internal variables]---------
168  * curr_clocksource:
169  *      currently selected clocksource.
170  * clocksource_list:
171  *      linked list with the registered clocksources
172  * clocksource_mutex:
173  *      protects manipulations to curr_clocksource and the clocksource_list
174  * override_name:
175  *      Name of the user-specified clocksource.
176  */
177 static struct clocksource *curr_clocksource;
178 static LIST_HEAD(clocksource_list);
179 static DEFINE_MUTEX(clocksource_mutex);
180 static char override_name[CS_NAME_LEN];
181 static int finished_booting;
182 
183 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
184 static void clocksource_watchdog_work(struct work_struct *work);
185 static void clocksource_select(void);
186 
187 static LIST_HEAD(watchdog_list);
188 static struct clocksource *watchdog;
189 static struct timer_list watchdog_timer;
190 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
191 static DEFINE_SPINLOCK(watchdog_lock);
192 static int watchdog_running;
193 static atomic_t watchdog_reset_pending;
194 
195 static int clocksource_watchdog_kthread(void *data);
196 static void __clocksource_change_rating(struct clocksource *cs, int rating);
197 
198 /*
199  * Interval: 0.5sec Threshold: 0.0625s
200  */
201 #define WATCHDOG_INTERVAL (HZ >> 1)
202 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
203 
204 static void clocksource_watchdog_work(struct work_struct *work)
205 {
206         /*
207          * If kthread_run fails the next watchdog scan over the
208          * watchdog_list will find the unstable clock again.
209          */
210         kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog");
211 }
212 
213 static void __clocksource_unstable(struct clocksource *cs)
214 {
215         cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
216         cs->flags |= CLOCK_SOURCE_UNSTABLE;
217         if (finished_booting)
218                 schedule_work(&watchdog_work);
219 }
220 
221 static void clocksource_unstable(struct clocksource *cs, int64_t delta)
222 {
223         printk(KERN_WARNING "Clocksource %s unstable (delta = %Ld ns)\n",
224                cs->name, delta);
225         __clocksource_unstable(cs);
226 }
227 
228 /**
229  * clocksource_mark_unstable - mark clocksource unstable via watchdog
230  * @cs:         clocksource to be marked unstable
231  *
232  * This function is called instead of clocksource_change_rating from
233  * cpu hotplug code to avoid a deadlock between the clocksource mutex
234  * and the cpu hotplug mutex. It defers the update of the clocksource
235  * to the watchdog thread.
236  */
237 void clocksource_mark_unstable(struct clocksource *cs)
238 {
239         unsigned long flags;
240 
241         spin_lock_irqsave(&watchdog_lock, flags);
242         if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
243                 if (list_empty(&cs->wd_list))
244                         list_add(&cs->wd_list, &watchdog_list);
245                 __clocksource_unstable(cs);
246         }
247         spin_unlock_irqrestore(&watchdog_lock, flags);
248 }
249 
250 static void clocksource_watchdog(unsigned long data)
251 {
252         struct clocksource *cs;
253         cycle_t csnow, wdnow, delta;
254         int64_t wd_nsec, cs_nsec;
255         int next_cpu, reset_pending;
256 
257         spin_lock(&watchdog_lock);
258         if (!watchdog_running)
259                 goto out;
260 
261         reset_pending = atomic_read(&watchdog_reset_pending);
262 
263         list_for_each_entry(cs, &watchdog_list, wd_list) {
264 
265                 /* Clocksource already marked unstable? */
266                 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
267                         if (finished_booting)
268                                 schedule_work(&watchdog_work);
269                         continue;
270                 }
271 
272                 local_irq_disable();
273                 csnow = cs->read(cs);
274                 wdnow = watchdog->read(watchdog);
275                 local_irq_enable();
276 
277                 /* Clocksource initialized ? */
278                 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
279                     atomic_read(&watchdog_reset_pending)) {
280                         cs->flags |= CLOCK_SOURCE_WATCHDOG;
281                         cs->wd_last = wdnow;
282                         cs->cs_last = csnow;
283                         continue;
284                 }
285 
286                 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
287                 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
288                                              watchdog->shift);
289 
290                 delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
291                 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
292                 cs->cs_last = csnow;
293                 cs->wd_last = wdnow;
294 
295                 if (atomic_read(&watchdog_reset_pending))
296                         continue;
297 
298                 /* Check the deviation from the watchdog clocksource. */
299                 if ((abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD)) {
300                         clocksource_unstable(cs, cs_nsec - wd_nsec);
301                         continue;
302                 }
303 
304                 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
305                     (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
306                     (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
307                         /* Mark it valid for high-res. */
308                         cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
309 
310                         /*
311                          * clocksource_done_booting() will sort it if
312                          * finished_booting is not set yet.
313                          */
314                         if (!finished_booting)
315                                 continue;
316 
317                         /*
318                          * If this is not the current clocksource let
319                          * the watchdog thread reselect it. Due to the
320                          * change to high res this clocksource might
321                          * be preferred now. If it is the current
322                          * clocksource let the tick code know about
323                          * that change.
324                          */
325                         if (cs != curr_clocksource) {
326                                 cs->flags |= CLOCK_SOURCE_RESELECT;
327                                 schedule_work(&watchdog_work);
328                         } else {
329                                 tick_clock_notify();
330                         }
331                 }
332         }
333 
334         /*
335          * We only clear the watchdog_reset_pending, when we did a
336          * full cycle through all clocksources.
337          */
338         if (reset_pending)
339                 atomic_dec(&watchdog_reset_pending);
340 
341         /*
342          * Cycle through CPUs to check if the CPUs stay synchronized
343          * to each other.
344          */
345         next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
346         if (next_cpu >= nr_cpu_ids)
347                 next_cpu = cpumask_first(cpu_online_mask);
348         watchdog_timer.expires += WATCHDOG_INTERVAL;
349         add_timer_on(&watchdog_timer, next_cpu);
350 out:
351         spin_unlock(&watchdog_lock);
352 }
353 
354 static inline void clocksource_start_watchdog(void)
355 {
356         if (watchdog_running || !watchdog || list_empty(&watchdog_list))
357                 return;
358         init_timer(&watchdog_timer);
359         watchdog_timer.function = clocksource_watchdog;
360         watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
361         add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
362         watchdog_running = 1;
363 }
364 
365 static inline void clocksource_stop_watchdog(void)
366 {
367         if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
368                 return;
369         del_timer(&watchdog_timer);
370         watchdog_running = 0;
371 }
372 
373 static inline void clocksource_reset_watchdog(void)
374 {
375         struct clocksource *cs;
376 
377         list_for_each_entry(cs, &watchdog_list, wd_list)
378                 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
379 }
380 
381 static void clocksource_resume_watchdog(void)
382 {
383         atomic_inc(&watchdog_reset_pending);
384 }
385 
386 static void clocksource_enqueue_watchdog(struct clocksource *cs)
387 {
388         unsigned long flags;
389 
390         spin_lock_irqsave(&watchdog_lock, flags);
391         if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
392                 /* cs is a clocksource to be watched. */
393                 list_add(&cs->wd_list, &watchdog_list);
394                 cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
395         } else {
396                 /* cs is a watchdog. */
397                 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
398                         cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
399                 /* Pick the best watchdog. */
400                 if (!watchdog || cs->rating > watchdog->rating) {
401                         watchdog = cs;
402                         /* Reset watchdog cycles */
403                         clocksource_reset_watchdog();
404                 }
405         }
406         /* Check if the watchdog timer needs to be started. */
407         clocksource_start_watchdog();
408         spin_unlock_irqrestore(&watchdog_lock, flags);
409 }
410 
411 static void clocksource_dequeue_watchdog(struct clocksource *cs)
412 {
413         unsigned long flags;
414 
415         spin_lock_irqsave(&watchdog_lock, flags);
416         if (cs != watchdog) {
417                 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
418                         /* cs is a watched clocksource. */
419                         list_del_init(&cs->wd_list);
420                         /* Check if the watchdog timer needs to be stopped. */
421                         clocksource_stop_watchdog();
422                 }
423         }
424         spin_unlock_irqrestore(&watchdog_lock, flags);
425 }
426 
427 static int __clocksource_watchdog_kthread(void)
428 {
429         struct clocksource *cs, *tmp;
430         unsigned long flags;
431         LIST_HEAD(unstable);
432         int select = 0;
433 
434         spin_lock_irqsave(&watchdog_lock, flags);
435         list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
436                 if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
437                         list_del_init(&cs->wd_list);
438                         list_add(&cs->wd_list, &unstable);
439                         select = 1;
440                 }
441                 if (cs->flags & CLOCK_SOURCE_RESELECT) {
442                         cs->flags &= ~CLOCK_SOURCE_RESELECT;
443                         select = 1;
444                 }
445         }
446         /* Check if the watchdog timer needs to be stopped. */
447         clocksource_stop_watchdog();
448         spin_unlock_irqrestore(&watchdog_lock, flags);
449 
450         /* Needs to be done outside of watchdog lock */
451         list_for_each_entry_safe(cs, tmp, &unstable, wd_list) {
452                 list_del_init(&cs->wd_list);
453                 __clocksource_change_rating(cs, 0);
454         }
455         return select;
456 }
457 
458 static int clocksource_watchdog_kthread(void *data)
459 {
460         mutex_lock(&clocksource_mutex);
461         if (__clocksource_watchdog_kthread())
462                 clocksource_select();
463         mutex_unlock(&clocksource_mutex);
464         return 0;
465 }
466 
467 static bool clocksource_is_watchdog(struct clocksource *cs)
468 {
469         return cs == watchdog;
470 }
471 
472 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */
473 
474 static void clocksource_enqueue_watchdog(struct clocksource *cs)
475 {
476         if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
477                 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
478 }
479 
480 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
481 static inline void clocksource_resume_watchdog(void) { }
482 static inline int __clocksource_watchdog_kthread(void) { return 0; }
483 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
484 void clocksource_mark_unstable(struct clocksource *cs) { }
485 
486 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */
487 
488 /**
489  * clocksource_suspend - suspend the clocksource(s)
490  */
491 void clocksource_suspend(void)
492 {
493         struct clocksource *cs;
494 
495         list_for_each_entry_reverse(cs, &clocksource_list, list)
496                 if (cs->suspend)
497                         cs->suspend(cs);
498 }
499 
500 /**
501  * clocksource_resume - resume the clocksource(s)
502  */
503 void clocksource_resume(void)
504 {
505         struct clocksource *cs;
506 
507         list_for_each_entry(cs, &clocksource_list, list)
508                 if (cs->resume)
509                         cs->resume(cs);
510 
511         clocksource_resume_watchdog();
512 }
513 
514 /**
515  * clocksource_touch_watchdog - Update watchdog
516  *
517  * Update the watchdog after exception contexts such as kgdb so as not
518  * to incorrectly trip the watchdog. This might fail when the kernel
519  * was stopped in code which holds watchdog_lock.
520  */
521 void clocksource_touch_watchdog(void)
522 {
523         clocksource_resume_watchdog();
524 }
525 
526 /**
527  * clocksource_max_adjustment- Returns max adjustment amount
528  * @cs:         Pointer to clocksource
529  *
530  */
531 static u32 clocksource_max_adjustment(struct clocksource *cs)
532 {
533         u64 ret;
534         /*
535          * We won't try to correct for more than 11% adjustments (110,000 ppm),
536          */
537         ret = (u64)cs->mult * 11;
538         do_div(ret,100);
539         return (u32)ret;
540 }
541 
542 /**
543  * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
544  * @mult:       cycle to nanosecond multiplier
545  * @shift:      cycle to nanosecond divisor (power of two)
546  * @maxadj:     maximum adjustment value to mult (~11%)
547  * @mask:       bitmask for two's complement subtraction of non 64 bit counters
548  */
549 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask)
550 {
551         u64 max_nsecs, max_cycles;
552 
553         /*
554          * Calculate the maximum number of cycles that we can pass to the
555          * cyc2ns function without overflowing a 64-bit signed result. The
556          * maximum number of cycles is equal to ULLONG_MAX/(mult+maxadj)
557          * which is equivalent to the below.
558          * max_cycles < (2^63)/(mult + maxadj)
559          * max_cycles < 2^(log2((2^63)/(mult + maxadj)))
560          * max_cycles < 2^(log2(2^63) - log2(mult + maxadj))
561          * max_cycles < 2^(63 - log2(mult + maxadj))
562          * max_cycles < 1 << (63 - log2(mult + maxadj))
563          * Please note that we add 1 to the result of the log2 to account for
564          * any rounding errors, ensure the above inequality is satisfied and
565          * no overflow will occur.
566          */
567         max_cycles = 1ULL << (63 - (ilog2(mult + maxadj) + 1));
568 
569         /*
570          * The actual maximum number of cycles we can defer the clocksource is
571          * determined by the minimum of max_cycles and mask.
572          * Note: Here we subtract the maxadj to make sure we don't sleep for
573          * too long if there's a large negative adjustment.
574          */
575         max_cycles = min(max_cycles, mask);
576         max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
577 
578         return max_nsecs;
579 }
580 
581 /**
582  * clocksource_max_deferment - Returns max time the clocksource can be deferred
583  * @cs:         Pointer to clocksource
584  *
585  */
586 static u64 clocksource_max_deferment(struct clocksource *cs)
587 {
588         u64 max_nsecs;
589 
590         max_nsecs = clocks_calc_max_nsecs(cs->mult, cs->shift, cs->maxadj,
591                                           cs->mask);
592         /*
593          * To ensure that the clocksource does not wrap whilst we are idle,
594          * limit the time the clocksource can be deferred by 12.5%. Please
595          * note a margin of 12.5% is used because this can be computed with
596          * a shift, versus say 10% which would require division.
597          */
598         return max_nsecs - (max_nsecs >> 3);
599 }
600 
601 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
602 
603 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
604 {
605         struct clocksource *cs;
606 
607         if (!finished_booting || list_empty(&clocksource_list))
608                 return NULL;
609 
610         /*
611          * We pick the clocksource with the highest rating. If oneshot
612          * mode is active, we pick the highres valid clocksource with
613          * the best rating.
614          */
615         list_for_each_entry(cs, &clocksource_list, list) {
616                 if (skipcur && cs == curr_clocksource)
617                         continue;
618                 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
619                         continue;
620                 return cs;
621         }
622         return NULL;
623 }
624 
625 static void __clocksource_select(bool skipcur)
626 {
627         bool oneshot = tick_oneshot_mode_active();
628         struct clocksource *best, *cs;
629 
630         /* Find the best suitable clocksource */
631         best = clocksource_find_best(oneshot, skipcur);
632         if (!best)
633                 return;
634 
635         /* Check for the override clocksource. */
636         list_for_each_entry(cs, &clocksource_list, list) {
637                 if (skipcur && cs == curr_clocksource)
638                         continue;
639                 if (strcmp(cs->name, override_name) != 0)
640                         continue;
641                 /*
642                  * Check to make sure we don't switch to a non-highres
643                  * capable clocksource if the tick code is in oneshot
644                  * mode (highres or nohz)
645                  */
646                 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
647                         /* Override clocksource cannot be used. */
648                         printk(KERN_WARNING "Override clocksource %s is not "
649                                "HRT compatible. Cannot switch while in "
650                                "HRT/NOHZ mode\n", cs->name);
651                         override_name[0] = 0;
652                 } else
653                         /* Override clocksource can be used. */
654                         best = cs;
655                 break;
656         }
657 
658         if (curr_clocksource != best && !timekeeping_notify(best)) {
659                 pr_info("Switched to clocksource %s\n", best->name);
660                 curr_clocksource = best;
661         }
662 }
663 
664 /**
665  * clocksource_select - Select the best clocksource available
666  *
667  * Private function. Must hold clocksource_mutex when called.
668  *
669  * Select the clocksource with the best rating, or the clocksource,
670  * which is selected by userspace override.
671  */
672 static void clocksource_select(void)
673 {
674         return __clocksource_select(false);
675 }
676 
677 static void clocksource_select_fallback(void)
678 {
679         return __clocksource_select(true);
680 }
681 
682 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */
683 
684 static inline void clocksource_select(void) { }
685 static inline void clocksource_select_fallback(void) { }
686 
687 #endif
688 
689 /*
690  * clocksource_done_booting - Called near the end of core bootup
691  *
692  * Hack to avoid lots of clocksource churn at boot time.
693  * We use fs_initcall because we want this to start before
694  * device_initcall but after subsys_initcall.
695  */
696 static int __init clocksource_done_booting(void)
697 {
698         mutex_lock(&clocksource_mutex);
699         curr_clocksource = clocksource_default_clock();
700         finished_booting = 1;
701         /*
702          * Run the watchdog first to eliminate unstable clock sources
703          */
704         __clocksource_watchdog_kthread();
705         clocksource_select();
706         mutex_unlock(&clocksource_mutex);
707         return 0;
708 }
709 fs_initcall(clocksource_done_booting);
710 
711 /*
712  * Enqueue the clocksource sorted by rating
713  */
714 static void clocksource_enqueue(struct clocksource *cs)
715 {
716         struct list_head *entry = &clocksource_list;
717         struct clocksource *tmp;
718 
719         list_for_each_entry(tmp, &clocksource_list, list)
720                 /* Keep track of the place, where to insert */
721                 if (tmp->rating >= cs->rating)
722                         entry = &tmp->list;
723         list_add(&cs->list, entry);
724 }
725 
726 /**
727  * __clocksource_updatefreq_scale - Used update clocksource with new freq
728  * @cs:         clocksource to be registered
729  * @scale:      Scale factor multiplied against freq to get clocksource hz
730  * @freq:       clocksource frequency (cycles per second) divided by scale
731  *
732  * This should only be called from the clocksource->enable() method.
733  *
734  * This *SHOULD NOT* be called directly! Please use the
735  * clocksource_updatefreq_hz() or clocksource_updatefreq_khz helper functions.
736  */
737 void __clocksource_updatefreq_scale(struct clocksource *cs, u32 scale, u32 freq)
738 {
739         u64 sec;
740         /*
741          * Calc the maximum number of seconds which we can run before
742          * wrapping around. For clocksources which have a mask > 32bit
743          * we need to limit the max sleep time to have a good
744          * conversion precision. 10 minutes is still a reasonable
745          * amount. That results in a shift value of 24 for a
746          * clocksource with mask >= 40bit and f >= 4GHz. That maps to
747          * ~ 0.06ppm granularity for NTP. We apply the same 12.5%
748          * margin as we do in clocksource_max_deferment()
749          */
750         sec = (cs->mask - (cs->mask >> 3));
751         do_div(sec, freq);
752         do_div(sec, scale);
753         if (!sec)
754                 sec = 1;
755         else if (sec > 600 && cs->mask > UINT_MAX)
756                 sec = 600;
757 
758         clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
759                                NSEC_PER_SEC / scale, sec * scale);
760 
761         /*
762          * for clocksources that have large mults, to avoid overflow.
763          * Since mult may be adjusted by ntp, add an safety extra margin
764          *
765          */
766         cs->maxadj = clocksource_max_adjustment(cs);
767         while ((cs->mult + cs->maxadj < cs->mult)
768                 || (cs->mult - cs->maxadj > cs->mult)) {
769                 cs->mult >>= 1;
770                 cs->shift--;
771                 cs->maxadj = clocksource_max_adjustment(cs);
772         }
773 
774         cs->max_idle_ns = clocksource_max_deferment(cs);
775 }
776 EXPORT_SYMBOL_GPL(__clocksource_updatefreq_scale);
777 
778 /**
779  * __clocksource_register_scale - Used to install new clocksources
780  * @cs:         clocksource to be registered
781  * @scale:      Scale factor multiplied against freq to get clocksource hz
782  * @freq:       clocksource frequency (cycles per second) divided by scale
783  *
784  * Returns -EBUSY if registration fails, zero otherwise.
785  *
786  * This *SHOULD NOT* be called directly! Please use the
787  * clocksource_register_hz() or clocksource_register_khz helper functions.
788  */
789 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
790 {
791 
792         /* Initialize mult/shift and max_idle_ns */
793         __clocksource_updatefreq_scale(cs, scale, freq);
794 
795         /* Add clocksource to the clcoksource list */
796         mutex_lock(&clocksource_mutex);
797         clocksource_enqueue(cs);
798         clocksource_enqueue_watchdog(cs);
799         clocksource_select();
800         mutex_unlock(&clocksource_mutex);
801         return 0;
802 }
803 EXPORT_SYMBOL_GPL(__clocksource_register_scale);
804 
805 
806 /**
807  * clocksource_register - Used to install new clocksources
808  * @cs:         clocksource to be registered
809  *
810  * Returns -EBUSY if registration fails, zero otherwise.
811  */
812 int clocksource_register(struct clocksource *cs)
813 {
814         /* calculate max adjustment for given mult/shift */
815         cs->maxadj = clocksource_max_adjustment(cs);
816         WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
817                 "Clocksource %s might overflow on 11%% adjustment\n",
818                 cs->name);
819 
820         /* calculate max idle time permitted for this clocksource */
821         cs->max_idle_ns = clocksource_max_deferment(cs);
822 
823         mutex_lock(&clocksource_mutex);
824         clocksource_enqueue(cs);
825         clocksource_enqueue_watchdog(cs);
826         clocksource_select();
827         mutex_unlock(&clocksource_mutex);
828         return 0;
829 }
830 EXPORT_SYMBOL(clocksource_register);
831 
832 static void __clocksource_change_rating(struct clocksource *cs, int rating)
833 {
834         list_del(&cs->list);
835         cs->rating = rating;
836         clocksource_enqueue(cs);
837 }
838 
839 /**
840  * clocksource_change_rating - Change the rating of a registered clocksource
841  * @cs:         clocksource to be changed
842  * @rating:     new rating
843  */
844 void clocksource_change_rating(struct clocksource *cs, int rating)
845 {
846         mutex_lock(&clocksource_mutex);
847         __clocksource_change_rating(cs, rating);
848         clocksource_select();
849         mutex_unlock(&clocksource_mutex);
850 }
851 EXPORT_SYMBOL(clocksource_change_rating);
852 
853 /*
854  * Unbind clocksource @cs. Called with clocksource_mutex held
855  */
856 static int clocksource_unbind(struct clocksource *cs)
857 {
858         /*
859          * I really can't convince myself to support this on hardware
860          * designed by lobotomized monkeys.
861          */
862         if (clocksource_is_watchdog(cs))
863                 return -EBUSY;
864 
865         if (cs == curr_clocksource) {
866                 /* Select and try to install a replacement clock source */
867                 clocksource_select_fallback();
868                 if (curr_clocksource == cs)
869                         return -EBUSY;
870         }
871         clocksource_dequeue_watchdog(cs);
872         list_del_init(&cs->list);
873         return 0;
874 }
875 
876 /**
877  * clocksource_unregister - remove a registered clocksource
878  * @cs: clocksource to be unregistered
879  */
880 int clocksource_unregister(struct clocksource *cs)
881 {
882         int ret = 0;
883 
884         mutex_lock(&clocksource_mutex);
885         if (!list_empty(&cs->list))
886                 ret = clocksource_unbind(cs);
887         mutex_unlock(&clocksource_mutex);
888         return ret;
889 }
890 EXPORT_SYMBOL(clocksource_unregister);
891 
892 #ifdef CONFIG_SYSFS
893 /**
894  * sysfs_show_current_clocksources - sysfs interface for current clocksource
895  * @dev:        unused
896  * @attr:       unused
897  * @buf:        char buffer to be filled with clocksource list
898  *
899  * Provides sysfs interface for listing current clocksource.
900  */
901 static ssize_t
902 sysfs_show_current_clocksources(struct device *dev,
903                                 struct device_attribute *attr, char *buf)
904 {
905         ssize_t count = 0;
906 
907         mutex_lock(&clocksource_mutex);
908         count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
909         mutex_unlock(&clocksource_mutex);
910 
911         return count;
912 }
913 
914 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
915 {
916         size_t ret = cnt;
917 
918         /* strings from sysfs write are not 0 terminated! */
919         if (!cnt || cnt >= CS_NAME_LEN)
920                 return -EINVAL;
921 
922         /* strip of \n: */
923         if (buf[cnt-1] == '\n')
924                 cnt--;
925         if (cnt > 0)
926                 memcpy(dst, buf, cnt);
927         dst[cnt] = 0;
928         return ret;
929 }
930 
931 /**
932  * sysfs_override_clocksource - interface for manually overriding clocksource
933  * @dev:        unused
934  * @attr:       unused
935  * @buf:        name of override clocksource
936  * @count:      length of buffer
937  *
938  * Takes input from sysfs interface for manually overriding the default
939  * clocksource selection.
940  */
941 static ssize_t sysfs_override_clocksource(struct device *dev,
942                                           struct device_attribute *attr,
943                                           const char *buf, size_t count)
944 {
945         ssize_t ret;
946 
947         mutex_lock(&clocksource_mutex);
948 
949         ret = sysfs_get_uname(buf, override_name, count);
950         if (ret >= 0)
951                 clocksource_select();
952 
953         mutex_unlock(&clocksource_mutex);
954 
955         return ret;
956 }
957 
958 /**
959  * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource
960  * @dev:        unused
961  * @attr:       unused
962  * @buf:        unused
963  * @count:      length of buffer
964  *
965  * Takes input from sysfs interface for manually unbinding a clocksource.
966  */
967 static ssize_t sysfs_unbind_clocksource(struct device *dev,
968                                         struct device_attribute *attr,
969                                         const char *buf, size_t count)
970 {
971         struct clocksource *cs;
972         char name[CS_NAME_LEN];
973         ssize_t ret;
974 
975         ret = sysfs_get_uname(buf, name, count);
976         if (ret < 0)
977                 return ret;
978 
979         ret = -ENODEV;
980         mutex_lock(&clocksource_mutex);
981         list_for_each_entry(cs, &clocksource_list, list) {
982                 if (strcmp(cs->name, name))
983                         continue;
984                 ret = clocksource_unbind(cs);
985                 break;
986         }
987         mutex_unlock(&clocksource_mutex);
988 
989         return ret ? ret : count;
990 }
991 
992 /**
993  * sysfs_show_available_clocksources - sysfs interface for listing clocksource
994  * @dev:        unused
995  * @attr:       unused
996  * @buf:        char buffer to be filled with clocksource list
997  *
998  * Provides sysfs interface for listing registered clocksources
999  */
1000 static ssize_t
1001 sysfs_show_available_clocksources(struct device *dev,
1002                                   struct device_attribute *attr,
1003                                   char *buf)
1004 {
1005         struct clocksource *src;
1006         ssize_t count = 0;
1007 
1008         mutex_lock(&clocksource_mutex);
1009         list_for_each_entry(src, &clocksource_list, list) {
1010                 /*
1011                  * Don't show non-HRES clocksource if the tick code is
1012                  * in one shot mode (highres=on or nohz=on)
1013                  */
1014                 if (!tick_oneshot_mode_active() ||
1015                     (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
1016                         count += snprintf(buf + count,
1017                                   max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
1018                                   "%s ", src->name);
1019         }
1020         mutex_unlock(&clocksource_mutex);
1021 
1022         count += snprintf(buf + count,
1023                           max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
1024 
1025         return count;
1026 }
1027 
1028 /*
1029  * Sysfs setup bits:
1030  */
1031 static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
1032                    sysfs_override_clocksource);
1033 
1034 static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource);
1035 
1036 static DEVICE_ATTR(available_clocksource, 0444,
1037                    sysfs_show_available_clocksources, NULL);
1038 
1039 static struct bus_type clocksource_subsys = {
1040         .name = "clocksource",
1041         .dev_name = "clocksource",
1042 };
1043 
1044 static struct device device_clocksource = {
1045         .id     = 0,
1046         .bus    = &clocksource_subsys,
1047 };
1048 
1049 static int __init init_clocksource_sysfs(void)
1050 {
1051         int error = subsys_system_register(&clocksource_subsys, NULL);
1052 
1053         if (!error)
1054                 error = device_register(&device_clocksource);
1055         if (!error)
1056                 error = device_create_file(
1057                                 &device_clocksource,
1058                                 &dev_attr_current_clocksource);
1059         if (!error)
1060                 error = device_create_file(&device_clocksource,
1061                                            &dev_attr_unbind_clocksource);
1062         if (!error)
1063                 error = device_create_file(
1064                                 &device_clocksource,
1065                                 &dev_attr_available_clocksource);
1066         return error;
1067 }
1068 
1069 device_initcall(init_clocksource_sysfs);
1070 #endif /* CONFIG_SYSFS */
1071 
1072 /**
1073  * boot_override_clocksource - boot clock override
1074  * @str:        override name
1075  *
1076  * Takes a clocksource= boot argument and uses it
1077  * as the clocksource override name.
1078  */
1079 static int __init boot_override_clocksource(char* str)
1080 {
1081         mutex_lock(&clocksource_mutex);
1082         if (str)
1083                 strlcpy(override_name, str, sizeof(override_name));
1084         mutex_unlock(&clocksource_mutex);
1085         return 1;
1086 }
1087 
1088 __setup("clocksource=", boot_override_clocksource);
1089 
1090 /**
1091  * boot_override_clock - Compatibility layer for deprecated boot option
1092  * @str:        override name
1093  *
1094  * DEPRECATED! Takes a clock= boot argument and uses it
1095  * as the clocksource override name
1096  */
1097 static int __init boot_override_clock(char* str)
1098 {
1099         if (!strcmp(str, "pmtmr")) {
1100                 printk("Warning: clock=pmtmr is deprecated. "
1101                         "Use clocksource=acpi_pm.\n");
1102                 return boot_override_clocksource("acpi_pm");
1103         }
1104         printk("Warning! clock= boot option is deprecated. "
1105                 "Use clocksource=xyz\n");
1106         return boot_override_clocksource(str);
1107 }
1108 
1109 __setup("clock=", boot_override_clock);
1110 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us