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

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

Version: ~ [ linux-5.17-rc1 ] ~ [ linux-5.16.2 ] ~ [ linux-5.15.16 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.93 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.173 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.225 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.262 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.297 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.299 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ 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