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