Commit | Line | Data |
---|---|---|
35728b82 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
734efb46 | 2 | /* |
734efb46 | 3 | * This file contains the functions which manage clocksource drivers. |
4 | * | |
5 | * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com) | |
734efb46 | 6 | */ |
7 | ||
45bbfe64 JP |
8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
9 | ||
d369a5d8 | 10 | #include <linux/device.h> |
734efb46 | 11 | #include <linux/clocksource.h> |
734efb46 | 12 | #include <linux/init.h> |
13 | #include <linux/module.h> | |
dc29a365 | 14 | #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */ |
79bf2bb3 | 15 | #include <linux/tick.h> |
01548f4d | 16 | #include <linux/kthread.h> |
734efb46 | 17 | |
c1797baf | 18 | #include "tick-internal.h" |
3a978377 | 19 | #include "timekeeping_internal.h" |
03e13cf5 | 20 | |
7d2f944a TG |
21 | /** |
22 | * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks | |
23 | * @mult: pointer to mult variable | |
24 | * @shift: pointer to shift variable | |
25 | * @from: frequency to convert from | |
26 | * @to: frequency to convert to | |
5fdade95 | 27 | * @maxsec: guaranteed runtime conversion range in seconds |
7d2f944a TG |
28 | * |
29 | * The function evaluates the shift/mult pair for the scaled math | |
30 | * operations of clocksources and clockevents. | |
31 | * | |
32 | * @to and @from are frequency values in HZ. For clock sources @to is | |
33 | * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock | |
34 | * event @to is the counter frequency and @from is NSEC_PER_SEC. | |
35 | * | |
5fdade95 | 36 | * The @maxsec conversion range argument controls the time frame in |
7d2f944a TG |
37 | * seconds which must be covered by the runtime conversion with the |
38 | * calculated mult and shift factors. This guarantees that no 64bit | |
39 | * overflow happens when the input value of the conversion is | |
40 | * multiplied with the calculated mult factor. Larger ranges may | |
41 | * reduce the conversion accuracy by chosing smaller mult and shift | |
42 | * factors. | |
43 | */ | |
44 | void | |
5fdade95 | 45 | clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec) |
7d2f944a TG |
46 | { |
47 | u64 tmp; | |
48 | u32 sft, sftacc= 32; | |
49 | ||
50 | /* | |
51 | * Calculate the shift factor which is limiting the conversion | |
52 | * range: | |
53 | */ | |
5fdade95 | 54 | tmp = ((u64)maxsec * from) >> 32; |
7d2f944a TG |
55 | while (tmp) { |
56 | tmp >>=1; | |
57 | sftacc--; | |
58 | } | |
59 | ||
60 | /* | |
61 | * Find the conversion shift/mult pair which has the best | |
62 | * accuracy and fits the maxsec conversion range: | |
63 | */ | |
64 | for (sft = 32; sft > 0; sft--) { | |
65 | tmp = (u64) to << sft; | |
b5776c4a | 66 | tmp += from / 2; |
7d2f944a TG |
67 | do_div(tmp, from); |
68 | if ((tmp >> sftacc) == 0) | |
69 | break; | |
70 | } | |
71 | *mult = tmp; | |
72 | *shift = sft; | |
73 | } | |
5304121a | 74 | EXPORT_SYMBOL_GPL(clocks_calc_mult_shift); |
7d2f944a | 75 | |
734efb46 | 76 | /*[Clocksource internal variables]--------- |
77 | * curr_clocksource: | |
f1b82746 | 78 | * currently selected clocksource. |
39232ed5 BW |
79 | * suspend_clocksource: |
80 | * used to calculate the suspend time. | |
734efb46 | 81 | * clocksource_list: |
82 | * linked list with the registered clocksources | |
75c5158f MS |
83 | * clocksource_mutex: |
84 | * protects manipulations to curr_clocksource and the clocksource_list | |
734efb46 | 85 | * override_name: |
86 | * Name of the user-specified clocksource. | |
87 | */ | |
f1b82746 | 88 | static struct clocksource *curr_clocksource; |
39232ed5 | 89 | static struct clocksource *suspend_clocksource; |
734efb46 | 90 | static LIST_HEAD(clocksource_list); |
75c5158f | 91 | static DEFINE_MUTEX(clocksource_mutex); |
29b54078 | 92 | static char override_name[CS_NAME_LEN]; |
54a6bc0b | 93 | static int finished_booting; |
39232ed5 | 94 | static u64 suspend_start; |
734efb46 | 95 | |
5d8b34fd | 96 | #ifdef CONFIG_CLOCKSOURCE_WATCHDOG |
f79e0258 | 97 | static void clocksource_watchdog_work(struct work_struct *work); |
332962f2 | 98 | static void clocksource_select(void); |
f79e0258 | 99 | |
5d8b34fd TG |
100 | static LIST_HEAD(watchdog_list); |
101 | static struct clocksource *watchdog; | |
102 | static struct timer_list watchdog_timer; | |
f79e0258 | 103 | static DECLARE_WORK(watchdog_work, clocksource_watchdog_work); |
5d8b34fd | 104 | static DEFINE_SPINLOCK(watchdog_lock); |
fb63a0eb | 105 | static int watchdog_running; |
9fb60336 | 106 | static atomic_t watchdog_reset_pending; |
b52f52a0 | 107 | |
0f48b41f | 108 | static inline void clocksource_watchdog_lock(unsigned long *flags) |
2aae7bcf PZ |
109 | { |
110 | spin_lock_irqsave(&watchdog_lock, *flags); | |
111 | } | |
112 | ||
0f48b41f | 113 | static inline void clocksource_watchdog_unlock(unsigned long *flags) |
2aae7bcf PZ |
114 | { |
115 | spin_unlock_irqrestore(&watchdog_lock, *flags); | |
116 | } | |
117 | ||
e2c631ba PZ |
118 | static int clocksource_watchdog_kthread(void *data); |
119 | static void __clocksource_change_rating(struct clocksource *cs, int rating); | |
120 | ||
5d8b34fd | 121 | /* |
35c35d1a | 122 | * Interval: 0.5sec Threshold: 0.0625s |
5d8b34fd TG |
123 | */ |
124 | #define WATCHDOG_INTERVAL (HZ >> 1) | |
35c35d1a | 125 | #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4) |
5d8b34fd | 126 | |
e2c631ba PZ |
127 | static void clocksource_watchdog_work(struct work_struct *work) |
128 | { | |
129 | /* | |
130 | * We cannot directly run clocksource_watchdog_kthread() here, because | |
131 | * clocksource_select() calls timekeeping_notify() which uses | |
132 | * stop_machine(). One cannot use stop_machine() from a workqueue() due | |
133 | * lock inversions wrt CPU hotplug. | |
134 | * | |
135 | * Also, we only ever run this work once or twice during the lifetime | |
136 | * of the kernel, so there is no point in creating a more permanent | |
137 | * kthread for this. | |
138 | * | |
139 | * If kthread_run fails the next watchdog scan over the | |
140 | * watchdog_list will find the unstable clock again. | |
141 | */ | |
142 | kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog"); | |
143 | } | |
144 | ||
7285dd7f | 145 | static void __clocksource_unstable(struct clocksource *cs) |
5d8b34fd | 146 | { |
5d8b34fd | 147 | cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG); |
c55c87c8 | 148 | cs->flags |= CLOCK_SOURCE_UNSTABLE; |
12907fbb | 149 | |
cd2af07d | 150 | /* |
e2c631ba | 151 | * If the clocksource is registered clocksource_watchdog_kthread() will |
cd2af07d PZ |
152 | * re-rate and re-select. |
153 | */ | |
154 | if (list_empty(&cs->list)) { | |
155 | cs->rating = 0; | |
2aae7bcf | 156 | return; |
cd2af07d | 157 | } |
2aae7bcf | 158 | |
12907fbb TG |
159 | if (cs->mark_unstable) |
160 | cs->mark_unstable(cs); | |
161 | ||
e2c631ba | 162 | /* kick clocksource_watchdog_kthread() */ |
54a6bc0b TG |
163 | if (finished_booting) |
164 | schedule_work(&watchdog_work); | |
5d8b34fd TG |
165 | } |
166 | ||
7285dd7f TG |
167 | /** |
168 | * clocksource_mark_unstable - mark clocksource unstable via watchdog | |
169 | * @cs: clocksource to be marked unstable | |
170 | * | |
7dba33c6 | 171 | * This function is called by the x86 TSC code to mark clocksources as unstable; |
e2c631ba | 172 | * it defers demotion and re-selection to a kthread. |
7285dd7f TG |
173 | */ |
174 | void clocksource_mark_unstable(struct clocksource *cs) | |
175 | { | |
176 | unsigned long flags; | |
177 | ||
178 | spin_lock_irqsave(&watchdog_lock, flags); | |
179 | if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) { | |
2aae7bcf | 180 | if (!list_empty(&cs->list) && list_empty(&cs->wd_list)) |
7285dd7f TG |
181 | list_add(&cs->wd_list, &watchdog_list); |
182 | __clocksource_unstable(cs); | |
183 | } | |
184 | spin_unlock_irqrestore(&watchdog_lock, flags); | |
185 | } | |
186 | ||
e99e88a9 | 187 | static void clocksource_watchdog(struct timer_list *unused) |
5d8b34fd | 188 | { |
c55c87c8 | 189 | struct clocksource *cs; |
a5a1d1c2 | 190 | u64 csnow, wdnow, cslast, wdlast, delta; |
5d8b34fd | 191 | int64_t wd_nsec, cs_nsec; |
9fb60336 | 192 | int next_cpu, reset_pending; |
5d8b34fd TG |
193 | |
194 | spin_lock(&watchdog_lock); | |
fb63a0eb MS |
195 | if (!watchdog_running) |
196 | goto out; | |
5d8b34fd | 197 | |
9fb60336 TG |
198 | reset_pending = atomic_read(&watchdog_reset_pending); |
199 | ||
c55c87c8 MS |
200 | list_for_each_entry(cs, &watchdog_list, wd_list) { |
201 | ||
202 | /* Clocksource already marked unstable? */ | |
01548f4d | 203 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) { |
54a6bc0b TG |
204 | if (finished_booting) |
205 | schedule_work(&watchdog_work); | |
c55c87c8 | 206 | continue; |
01548f4d | 207 | } |
c55c87c8 | 208 | |
b5199515 | 209 | local_irq_disable(); |
8e19608e | 210 | csnow = cs->read(cs); |
b5199515 TG |
211 | wdnow = watchdog->read(watchdog); |
212 | local_irq_enable(); | |
b52f52a0 | 213 | |
8cf4e750 | 214 | /* Clocksource initialized ? */ |
9fb60336 TG |
215 | if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) || |
216 | atomic_read(&watchdog_reset_pending)) { | |
8cf4e750 | 217 | cs->flags |= CLOCK_SOURCE_WATCHDOG; |
b5199515 TG |
218 | cs->wd_last = wdnow; |
219 | cs->cs_last = csnow; | |
b52f52a0 TG |
220 | continue; |
221 | } | |
222 | ||
3a978377 TG |
223 | delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask); |
224 | wd_nsec = clocksource_cyc2ns(delta, watchdog->mult, | |
225 | watchdog->shift); | |
b5199515 | 226 | |
3a978377 TG |
227 | delta = clocksource_delta(csnow, cs->cs_last, cs->mask); |
228 | cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift); | |
0b046b21 JS |
229 | wdlast = cs->wd_last; /* save these in case we print them */ |
230 | cslast = cs->cs_last; | |
b5199515 TG |
231 | cs->cs_last = csnow; |
232 | cs->wd_last = wdnow; | |
233 | ||
9fb60336 TG |
234 | if (atomic_read(&watchdog_reset_pending)) |
235 | continue; | |
236 | ||
b5199515 | 237 | /* Check the deviation from the watchdog clocksource. */ |
79211c8e | 238 | if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) { |
390dd67c SI |
239 | pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n", |
240 | smp_processor_id(), cs->name); | |
45bbfe64 | 241 | pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n", |
0b046b21 | 242 | watchdog->name, wdnow, wdlast, watchdog->mask); |
45bbfe64 | 243 | pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n", |
0b046b21 JS |
244 | cs->name, csnow, cslast, cs->mask); |
245 | __clocksource_unstable(cs); | |
8cf4e750 MS |
246 | continue; |
247 | } | |
248 | ||
b421b22b PZ |
249 | if (cs == curr_clocksource && cs->tick_stable) |
250 | cs->tick_stable(cs); | |
251 | ||
8cf4e750 MS |
252 | if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && |
253 | (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) && | |
254 | (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) { | |
332962f2 | 255 | /* Mark it valid for high-res. */ |
8cf4e750 | 256 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; |
332962f2 TG |
257 | |
258 | /* | |
259 | * clocksource_done_booting() will sort it if | |
260 | * finished_booting is not set yet. | |
261 | */ | |
262 | if (!finished_booting) | |
263 | continue; | |
264 | ||
8cf4e750 | 265 | /* |
332962f2 TG |
266 | * If this is not the current clocksource let |
267 | * the watchdog thread reselect it. Due to the | |
268 | * change to high res this clocksource might | |
269 | * be preferred now. If it is the current | |
270 | * clocksource let the tick code know about | |
271 | * that change. | |
8cf4e750 | 272 | */ |
332962f2 TG |
273 | if (cs != curr_clocksource) { |
274 | cs->flags |= CLOCK_SOURCE_RESELECT; | |
275 | schedule_work(&watchdog_work); | |
276 | } else { | |
277 | tick_clock_notify(); | |
278 | } | |
5d8b34fd TG |
279 | } |
280 | } | |
281 | ||
9fb60336 TG |
282 | /* |
283 | * We only clear the watchdog_reset_pending, when we did a | |
284 | * full cycle through all clocksources. | |
285 | */ | |
286 | if (reset_pending) | |
287 | atomic_dec(&watchdog_reset_pending); | |
288 | ||
c55c87c8 MS |
289 | /* |
290 | * Cycle through CPUs to check if the CPUs stay synchronized | |
291 | * to each other. | |
292 | */ | |
293 | next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask); | |
294 | if (next_cpu >= nr_cpu_ids) | |
295 | next_cpu = cpumask_first(cpu_online_mask); | |
296 | watchdog_timer.expires += WATCHDOG_INTERVAL; | |
297 | add_timer_on(&watchdog_timer, next_cpu); | |
fb63a0eb | 298 | out: |
5d8b34fd TG |
299 | spin_unlock(&watchdog_lock); |
300 | } | |
0f8e8ef7 | 301 | |
fb63a0eb MS |
302 | static inline void clocksource_start_watchdog(void) |
303 | { | |
304 | if (watchdog_running || !watchdog || list_empty(&watchdog_list)) | |
305 | return; | |
e99e88a9 | 306 | timer_setup(&watchdog_timer, clocksource_watchdog, 0); |
fb63a0eb MS |
307 | watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL; |
308 | add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask)); | |
309 | watchdog_running = 1; | |
310 | } | |
311 | ||
312 | static inline void clocksource_stop_watchdog(void) | |
313 | { | |
314 | if (!watchdog_running || (watchdog && !list_empty(&watchdog_list))) | |
315 | return; | |
316 | del_timer(&watchdog_timer); | |
317 | watchdog_running = 0; | |
318 | } | |
319 | ||
0f8e8ef7 MS |
320 | static inline void clocksource_reset_watchdog(void) |
321 | { | |
322 | struct clocksource *cs; | |
323 | ||
324 | list_for_each_entry(cs, &watchdog_list, wd_list) | |
325 | cs->flags &= ~CLOCK_SOURCE_WATCHDOG; | |
326 | } | |
327 | ||
b52f52a0 TG |
328 | static void clocksource_resume_watchdog(void) |
329 | { | |
9fb60336 | 330 | atomic_inc(&watchdog_reset_pending); |
b52f52a0 TG |
331 | } |
332 | ||
fb63a0eb | 333 | static void clocksource_enqueue_watchdog(struct clocksource *cs) |
5d8b34fd | 334 | { |
5b9e886a PZ |
335 | INIT_LIST_HEAD(&cs->wd_list); |
336 | ||
5d8b34fd | 337 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { |
fb63a0eb | 338 | /* cs is a clocksource to be watched. */ |
5d8b34fd | 339 | list_add(&cs->wd_list, &watchdog_list); |
fb63a0eb | 340 | cs->flags &= ~CLOCK_SOURCE_WATCHDOG; |
948ac6d7 | 341 | } else { |
fb63a0eb | 342 | /* cs is a watchdog. */ |
948ac6d7 | 343 | if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) |
5d8b34fd | 344 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; |
bbf66d89 | 345 | } |
bbf66d89 VK |
346 | } |
347 | ||
348 | static void clocksource_select_watchdog(bool fallback) | |
349 | { | |
350 | struct clocksource *cs, *old_wd; | |
351 | unsigned long flags; | |
352 | ||
353 | spin_lock_irqsave(&watchdog_lock, flags); | |
354 | /* save current watchdog */ | |
355 | old_wd = watchdog; | |
356 | if (fallback) | |
357 | watchdog = NULL; | |
358 | ||
359 | list_for_each_entry(cs, &clocksource_list, list) { | |
360 | /* cs is a clocksource to be watched. */ | |
361 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) | |
362 | continue; | |
363 | ||
364 | /* Skip current if we were requested for a fallback. */ | |
365 | if (fallback && cs == old_wd) | |
366 | continue; | |
367 | ||
fb63a0eb | 368 | /* Pick the best watchdog. */ |
bbf66d89 | 369 | if (!watchdog || cs->rating > watchdog->rating) |
5d8b34fd | 370 | watchdog = cs; |
5d8b34fd | 371 | } |
bbf66d89 VK |
372 | /* If we failed to find a fallback restore the old one. */ |
373 | if (!watchdog) | |
374 | watchdog = old_wd; | |
375 | ||
376 | /* If we changed the watchdog we need to reset cycles. */ | |
377 | if (watchdog != old_wd) | |
378 | clocksource_reset_watchdog(); | |
379 | ||
fb63a0eb MS |
380 | /* Check if the watchdog timer needs to be started. */ |
381 | clocksource_start_watchdog(); | |
5d8b34fd TG |
382 | spin_unlock_irqrestore(&watchdog_lock, flags); |
383 | } | |
fb63a0eb MS |
384 | |
385 | static void clocksource_dequeue_watchdog(struct clocksource *cs) | |
386 | { | |
a89c7edb TG |
387 | if (cs != watchdog) { |
388 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { | |
389 | /* cs is a watched clocksource. */ | |
390 | list_del_init(&cs->wd_list); | |
391 | /* Check if the watchdog timer needs to be stopped. */ | |
392 | clocksource_stop_watchdog(); | |
fb63a0eb MS |
393 | } |
394 | } | |
fb63a0eb MS |
395 | } |
396 | ||
e2c631ba | 397 | static int __clocksource_watchdog_kthread(void) |
c55c87c8 MS |
398 | { |
399 | struct clocksource *cs, *tmp; | |
400 | unsigned long flags; | |
332962f2 | 401 | int select = 0; |
c55c87c8 MS |
402 | |
403 | spin_lock_irqsave(&watchdog_lock, flags); | |
332962f2 | 404 | list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) { |
c55c87c8 MS |
405 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) { |
406 | list_del_init(&cs->wd_list); | |
2aae7bcf | 407 | __clocksource_change_rating(cs, 0); |
332962f2 TG |
408 | select = 1; |
409 | } | |
410 | if (cs->flags & CLOCK_SOURCE_RESELECT) { | |
411 | cs->flags &= ~CLOCK_SOURCE_RESELECT; | |
412 | select = 1; | |
c55c87c8 | 413 | } |
332962f2 | 414 | } |
c55c87c8 MS |
415 | /* Check if the watchdog timer needs to be stopped. */ |
416 | clocksource_stop_watchdog(); | |
6ea41d25 TG |
417 | spin_unlock_irqrestore(&watchdog_lock, flags); |
418 | ||
332962f2 TG |
419 | return select; |
420 | } | |
421 | ||
e2c631ba | 422 | static int clocksource_watchdog_kthread(void *data) |
332962f2 TG |
423 | { |
424 | mutex_lock(&clocksource_mutex); | |
e2c631ba | 425 | if (__clocksource_watchdog_kthread()) |
332962f2 | 426 | clocksource_select(); |
d0981a1b | 427 | mutex_unlock(&clocksource_mutex); |
e2c631ba | 428 | return 0; |
c55c87c8 MS |
429 | } |
430 | ||
7eaeb343 TG |
431 | static bool clocksource_is_watchdog(struct clocksource *cs) |
432 | { | |
433 | return cs == watchdog; | |
434 | } | |
435 | ||
fb63a0eb MS |
436 | #else /* CONFIG_CLOCKSOURCE_WATCHDOG */ |
437 | ||
438 | static void clocksource_enqueue_watchdog(struct clocksource *cs) | |
5d8b34fd TG |
439 | { |
440 | if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) | |
441 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; | |
442 | } | |
b52f52a0 | 443 | |
bbf66d89 | 444 | static void clocksource_select_watchdog(bool fallback) { } |
fb63a0eb | 445 | static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { } |
b52f52a0 | 446 | static inline void clocksource_resume_watchdog(void) { } |
e2c631ba | 447 | static inline int __clocksource_watchdog_kthread(void) { return 0; } |
7eaeb343 | 448 | static bool clocksource_is_watchdog(struct clocksource *cs) { return false; } |
397bbf6d | 449 | void clocksource_mark_unstable(struct clocksource *cs) { } |
fb63a0eb | 450 | |
db6f9e55 MM |
451 | static inline void clocksource_watchdog_lock(unsigned long *flags) { } |
452 | static inline void clocksource_watchdog_unlock(unsigned long *flags) { } | |
2aae7bcf | 453 | |
fb63a0eb | 454 | #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */ |
5d8b34fd | 455 | |
39232ed5 BW |
456 | static bool clocksource_is_suspend(struct clocksource *cs) |
457 | { | |
458 | return cs == suspend_clocksource; | |
459 | } | |
460 | ||
461 | static void __clocksource_suspend_select(struct clocksource *cs) | |
462 | { | |
463 | /* | |
464 | * Skip the clocksource which will be stopped in suspend state. | |
465 | */ | |
466 | if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP)) | |
467 | return; | |
468 | ||
469 | /* | |
470 | * The nonstop clocksource can be selected as the suspend clocksource to | |
471 | * calculate the suspend time, so it should not supply suspend/resume | |
472 | * interfaces to suspend the nonstop clocksource when system suspends. | |
473 | */ | |
474 | if (cs->suspend || cs->resume) { | |
475 | pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n", | |
476 | cs->name); | |
477 | } | |
478 | ||
479 | /* Pick the best rating. */ | |
480 | if (!suspend_clocksource || cs->rating > suspend_clocksource->rating) | |
481 | suspend_clocksource = cs; | |
482 | } | |
483 | ||
484 | /** | |
485 | * clocksource_suspend_select - Select the best clocksource for suspend timing | |
486 | * @fallback: if select a fallback clocksource | |
487 | */ | |
488 | static void clocksource_suspend_select(bool fallback) | |
489 | { | |
490 | struct clocksource *cs, *old_suspend; | |
491 | ||
492 | old_suspend = suspend_clocksource; | |
493 | if (fallback) | |
494 | suspend_clocksource = NULL; | |
495 | ||
496 | list_for_each_entry(cs, &clocksource_list, list) { | |
497 | /* Skip current if we were requested for a fallback. */ | |
498 | if (fallback && cs == old_suspend) | |
499 | continue; | |
500 | ||
501 | __clocksource_suspend_select(cs); | |
502 | } | |
503 | } | |
504 | ||
505 | /** | |
506 | * clocksource_start_suspend_timing - Start measuring the suspend timing | |
507 | * @cs: current clocksource from timekeeping | |
508 | * @start_cycles: current cycles from timekeeping | |
509 | * | |
510 | * This function will save the start cycle values of suspend timer to calculate | |
511 | * the suspend time when resuming system. | |
512 | * | |
513 | * This function is called late in the suspend process from timekeeping_suspend(), | |
514 | * that means processes are freezed, non-boot cpus and interrupts are disabled | |
515 | * now. It is therefore possible to start the suspend timer without taking the | |
516 | * clocksource mutex. | |
517 | */ | |
518 | void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles) | |
519 | { | |
520 | if (!suspend_clocksource) | |
521 | return; | |
522 | ||
523 | /* | |
524 | * If current clocksource is the suspend timer, we should use the | |
525 | * tkr_mono.cycle_last value as suspend_start to avoid same reading | |
526 | * from suspend timer. | |
527 | */ | |
528 | if (clocksource_is_suspend(cs)) { | |
529 | suspend_start = start_cycles; | |
530 | return; | |
531 | } | |
532 | ||
533 | if (suspend_clocksource->enable && | |
534 | suspend_clocksource->enable(suspend_clocksource)) { | |
535 | pr_warn_once("Failed to enable the non-suspend-able clocksource.\n"); | |
536 | return; | |
537 | } | |
538 | ||
539 | suspend_start = suspend_clocksource->read(suspend_clocksource); | |
540 | } | |
541 | ||
542 | /** | |
543 | * clocksource_stop_suspend_timing - Stop measuring the suspend timing | |
544 | * @cs: current clocksource from timekeeping | |
545 | * @cycle_now: current cycles from timekeeping | |
546 | * | |
547 | * This function will calculate the suspend time from suspend timer. | |
548 | * | |
549 | * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource. | |
550 | * | |
551 | * This function is called early in the resume process from timekeeping_resume(), | |
552 | * that means there is only one cpu, no processes are running and the interrupts | |
553 | * are disabled. It is therefore possible to stop the suspend timer without | |
554 | * taking the clocksource mutex. | |
555 | */ | |
556 | u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now) | |
557 | { | |
558 | u64 now, delta, nsec = 0; | |
559 | ||
560 | if (!suspend_clocksource) | |
561 | return 0; | |
562 | ||
563 | /* | |
564 | * If current clocksource is the suspend timer, we should use the | |
565 | * tkr_mono.cycle_last value from timekeeping as current cycle to | |
566 | * avoid same reading from suspend timer. | |
567 | */ | |
568 | if (clocksource_is_suspend(cs)) | |
569 | now = cycle_now; | |
570 | else | |
571 | now = suspend_clocksource->read(suspend_clocksource); | |
572 | ||
573 | if (now > suspend_start) { | |
574 | delta = clocksource_delta(now, suspend_start, | |
575 | suspend_clocksource->mask); | |
576 | nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult, | |
577 | suspend_clocksource->shift); | |
578 | } | |
579 | ||
580 | /* | |
581 | * Disable the suspend timer to save power if current clocksource is | |
582 | * not the suspend timer. | |
583 | */ | |
584 | if (!clocksource_is_suspend(cs) && suspend_clocksource->disable) | |
585 | suspend_clocksource->disable(suspend_clocksource); | |
586 | ||
587 | return nsec; | |
588 | } | |
589 | ||
c54a42b1 MD |
590 | /** |
591 | * clocksource_suspend - suspend the clocksource(s) | |
592 | */ | |
593 | void clocksource_suspend(void) | |
594 | { | |
595 | struct clocksource *cs; | |
596 | ||
597 | list_for_each_entry_reverse(cs, &clocksource_list, list) | |
598 | if (cs->suspend) | |
599 | cs->suspend(cs); | |
600 | } | |
601 | ||
b52f52a0 TG |
602 | /** |
603 | * clocksource_resume - resume the clocksource(s) | |
604 | */ | |
605 | void clocksource_resume(void) | |
606 | { | |
2e197586 | 607 | struct clocksource *cs; |
b52f52a0 | 608 | |
75c5158f | 609 | list_for_each_entry(cs, &clocksource_list, list) |
b52f52a0 | 610 | if (cs->resume) |
17622339 | 611 | cs->resume(cs); |
b52f52a0 TG |
612 | |
613 | clocksource_resume_watchdog(); | |
b52f52a0 TG |
614 | } |
615 | ||
7c3078b6 JW |
616 | /** |
617 | * clocksource_touch_watchdog - Update watchdog | |
618 | * | |
619 | * Update the watchdog after exception contexts such as kgdb so as not | |
7b7422a5 TG |
620 | * to incorrectly trip the watchdog. This might fail when the kernel |
621 | * was stopped in code which holds watchdog_lock. | |
7c3078b6 JW |
622 | */ |
623 | void clocksource_touch_watchdog(void) | |
624 | { | |
625 | clocksource_resume_watchdog(); | |
626 | } | |
627 | ||
d65670a7 JS |
628 | /** |
629 | * clocksource_max_adjustment- Returns max adjustment amount | |
630 | * @cs: Pointer to clocksource | |
631 | * | |
632 | */ | |
633 | static u32 clocksource_max_adjustment(struct clocksource *cs) | |
634 | { | |
635 | u64 ret; | |
636 | /* | |
88b28adf | 637 | * We won't try to correct for more than 11% adjustments (110,000 ppm), |
d65670a7 JS |
638 | */ |
639 | ret = (u64)cs->mult * 11; | |
640 | do_div(ret,100); | |
641 | return (u32)ret; | |
642 | } | |
643 | ||
98962465 | 644 | /** |
87d8b9eb SB |
645 | * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted |
646 | * @mult: cycle to nanosecond multiplier | |
647 | * @shift: cycle to nanosecond divisor (power of two) | |
648 | * @maxadj: maximum adjustment value to mult (~11%) | |
649 | * @mask: bitmask for two's complement subtraction of non 64 bit counters | |
fb82fe2f JS |
650 | * @max_cyc: maximum cycle value before potential overflow (does not include |
651 | * any safety margin) | |
362fde04 | 652 | * |
8e56f33f JS |
653 | * NOTE: This function includes a safety margin of 50%, in other words, we |
654 | * return half the number of nanoseconds the hardware counter can technically | |
655 | * cover. This is done so that we can potentially detect problems caused by | |
656 | * delayed timers or bad hardware, which might result in time intervals that | |
571af55a | 657 | * are larger than what the math used can handle without overflows. |
98962465 | 658 | */ |
fb82fe2f | 659 | u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc) |
98962465 JH |
660 | { |
661 | u64 max_nsecs, max_cycles; | |
662 | ||
663 | /* | |
664 | * Calculate the maximum number of cycles that we can pass to the | |
6086e346 | 665 | * cyc2ns() function without overflowing a 64-bit result. |
98962465 | 666 | */ |
6086e346 JS |
667 | max_cycles = ULLONG_MAX; |
668 | do_div(max_cycles, mult+maxadj); | |
98962465 JH |
669 | |
670 | /* | |
671 | * The actual maximum number of cycles we can defer the clocksource is | |
87d8b9eb | 672 | * determined by the minimum of max_cycles and mask. |
d65670a7 JS |
673 | * Note: Here we subtract the maxadj to make sure we don't sleep for |
674 | * too long if there's a large negative adjustment. | |
98962465 | 675 | */ |
87d8b9eb SB |
676 | max_cycles = min(max_cycles, mask); |
677 | max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift); | |
678 | ||
fb82fe2f JS |
679 | /* return the max_cycles value as well if requested */ |
680 | if (max_cyc) | |
681 | *max_cyc = max_cycles; | |
682 | ||
362fde04 JS |
683 | /* Return 50% of the actual maximum, so we can detect bad values */ |
684 | max_nsecs >>= 1; | |
685 | ||
87d8b9eb SB |
686 | return max_nsecs; |
687 | } | |
688 | ||
689 | /** | |
fb82fe2f JS |
690 | * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles |
691 | * @cs: Pointer to clocksource to be updated | |
87d8b9eb SB |
692 | * |
693 | */ | |
fb82fe2f | 694 | static inline void clocksource_update_max_deferment(struct clocksource *cs) |
87d8b9eb | 695 | { |
fb82fe2f JS |
696 | cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift, |
697 | cs->maxadj, cs->mask, | |
698 | &cs->max_cycles); | |
98962465 JH |
699 | } |
700 | ||
592913ec | 701 | #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET |
734efb46 | 702 | |
f5a2e343 | 703 | static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur) |
5d33b883 TG |
704 | { |
705 | struct clocksource *cs; | |
706 | ||
707 | if (!finished_booting || list_empty(&clocksource_list)) | |
708 | return NULL; | |
709 | ||
710 | /* | |
711 | * We pick the clocksource with the highest rating. If oneshot | |
712 | * mode is active, we pick the highres valid clocksource with | |
713 | * the best rating. | |
714 | */ | |
715 | list_for_each_entry(cs, &clocksource_list, list) { | |
f5a2e343 TG |
716 | if (skipcur && cs == curr_clocksource) |
717 | continue; | |
5d33b883 TG |
718 | if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES)) |
719 | continue; | |
720 | return cs; | |
721 | } | |
722 | return NULL; | |
723 | } | |
724 | ||
f5a2e343 | 725 | static void __clocksource_select(bool skipcur) |
734efb46 | 726 | { |
5d33b883 | 727 | bool oneshot = tick_oneshot_mode_active(); |
f1b82746 | 728 | struct clocksource *best, *cs; |
5d8b34fd | 729 | |
5d33b883 | 730 | /* Find the best suitable clocksource */ |
f5a2e343 | 731 | best = clocksource_find_best(oneshot, skipcur); |
5d33b883 | 732 | if (!best) |
f1b82746 | 733 | return; |
5d33b883 | 734 | |
7f852afe BW |
735 | if (!strlen(override_name)) |
736 | goto found; | |
737 | ||
f1b82746 MS |
738 | /* Check for the override clocksource. */ |
739 | list_for_each_entry(cs, &clocksource_list, list) { | |
f5a2e343 TG |
740 | if (skipcur && cs == curr_clocksource) |
741 | continue; | |
f1b82746 MS |
742 | if (strcmp(cs->name, override_name) != 0) |
743 | continue; | |
744 | /* | |
745 | * Check to make sure we don't switch to a non-highres | |
746 | * capable clocksource if the tick code is in oneshot | |
747 | * mode (highres or nohz) | |
748 | */ | |
5d33b883 | 749 | if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) { |
f1b82746 | 750 | /* Override clocksource cannot be used. */ |
36374583 KW |
751 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) { |
752 | pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n", | |
753 | cs->name); | |
754 | override_name[0] = 0; | |
755 | } else { | |
756 | /* | |
757 | * The override cannot be currently verified. | |
758 | * Deferring to let the watchdog check. | |
759 | */ | |
760 | pr_info("Override clocksource %s is not currently HRT compatible - deferring\n", | |
761 | cs->name); | |
762 | } | |
f1b82746 MS |
763 | } else |
764 | /* Override clocksource can be used. */ | |
765 | best = cs; | |
766 | break; | |
767 | } | |
ba919d1c | 768 | |
7f852afe | 769 | found: |
ba919d1c TG |
770 | if (curr_clocksource != best && !timekeeping_notify(best)) { |
771 | pr_info("Switched to clocksource %s\n", best->name); | |
75c5158f | 772 | curr_clocksource = best; |
75c5158f | 773 | } |
f1b82746 | 774 | } |
734efb46 | 775 | |
f5a2e343 TG |
776 | /** |
777 | * clocksource_select - Select the best clocksource available | |
778 | * | |
779 | * Private function. Must hold clocksource_mutex when called. | |
780 | * | |
781 | * Select the clocksource with the best rating, or the clocksource, | |
782 | * which is selected by userspace override. | |
783 | */ | |
784 | static void clocksource_select(void) | |
785 | { | |
cfed432d | 786 | __clocksource_select(false); |
f5a2e343 TG |
787 | } |
788 | ||
7eaeb343 TG |
789 | static void clocksource_select_fallback(void) |
790 | { | |
cfed432d | 791 | __clocksource_select(true); |
7eaeb343 TG |
792 | } |
793 | ||
592913ec | 794 | #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */ |
54a6bc0b | 795 | static inline void clocksource_select(void) { } |
1eaff672 | 796 | static inline void clocksource_select_fallback(void) { } |
54a6bc0b TG |
797 | |
798 | #endif | |
799 | ||
75c5158f MS |
800 | /* |
801 | * clocksource_done_booting - Called near the end of core bootup | |
802 | * | |
803 | * Hack to avoid lots of clocksource churn at boot time. | |
804 | * We use fs_initcall because we want this to start before | |
805 | * device_initcall but after subsys_initcall. | |
806 | */ | |
807 | static int __init clocksource_done_booting(void) | |
808 | { | |
ad6759fb | 809 | mutex_lock(&clocksource_mutex); |
810 | curr_clocksource = clocksource_default_clock(); | |
75c5158f | 811 | finished_booting = 1; |
54a6bc0b TG |
812 | /* |
813 | * Run the watchdog first to eliminate unstable clock sources | |
814 | */ | |
e2c631ba | 815 | __clocksource_watchdog_kthread(); |
75c5158f | 816 | clocksource_select(); |
e6c73305 | 817 | mutex_unlock(&clocksource_mutex); |
75c5158f MS |
818 | return 0; |
819 | } | |
820 | fs_initcall(clocksource_done_booting); | |
821 | ||
92c7e002 TG |
822 | /* |
823 | * Enqueue the clocksource sorted by rating | |
734efb46 | 824 | */ |
f1b82746 | 825 | static void clocksource_enqueue(struct clocksource *cs) |
734efb46 | 826 | { |
f1b82746 MS |
827 | struct list_head *entry = &clocksource_list; |
828 | struct clocksource *tmp; | |
92c7e002 | 829 | |
0fb71d34 | 830 | list_for_each_entry(tmp, &clocksource_list, list) { |
92c7e002 | 831 | /* Keep track of the place, where to insert */ |
0fb71d34 MH |
832 | if (tmp->rating < cs->rating) |
833 | break; | |
834 | entry = &tmp->list; | |
835 | } | |
f1b82746 | 836 | list_add(&cs->list, entry); |
734efb46 | 837 | } |
838 | ||
d7e81c26 | 839 | /** |
fba9e072 | 840 | * __clocksource_update_freq_scale - Used update clocksource with new freq |
b1b73d09 | 841 | * @cs: clocksource to be registered |
d7e81c26 JS |
842 | * @scale: Scale factor multiplied against freq to get clocksource hz |
843 | * @freq: clocksource frequency (cycles per second) divided by scale | |
844 | * | |
852db46d | 845 | * This should only be called from the clocksource->enable() method. |
d7e81c26 JS |
846 | * |
847 | * This *SHOULD NOT* be called directly! Please use the | |
fba9e072 JS |
848 | * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper |
849 | * functions. | |
d7e81c26 | 850 | */ |
fba9e072 | 851 | void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq) |
d7e81c26 | 852 | { |
c0e299b1 | 853 | u64 sec; |
f8935983 | 854 | |
d7e81c26 | 855 | /* |
f8935983 JS |
856 | * Default clocksources are *special* and self-define their mult/shift. |
857 | * But, you're not special, so you should specify a freq value. | |
d7e81c26 | 858 | */ |
f8935983 JS |
859 | if (freq) { |
860 | /* | |
861 | * Calc the maximum number of seconds which we can run before | |
862 | * wrapping around. For clocksources which have a mask > 32-bit | |
863 | * we need to limit the max sleep time to have a good | |
864 | * conversion precision. 10 minutes is still a reasonable | |
865 | * amount. That results in a shift value of 24 for a | |
866 | * clocksource with mask >= 40-bit and f >= 4GHz. That maps to | |
867 | * ~ 0.06ppm granularity for NTP. | |
868 | */ | |
869 | sec = cs->mask; | |
870 | do_div(sec, freq); | |
871 | do_div(sec, scale); | |
872 | if (!sec) | |
873 | sec = 1; | |
874 | else if (sec > 600 && cs->mask > UINT_MAX) | |
875 | sec = 600; | |
876 | ||
877 | clocks_calc_mult_shift(&cs->mult, &cs->shift, freq, | |
878 | NSEC_PER_SEC / scale, sec * scale); | |
879 | } | |
d65670a7 | 880 | /* |
362fde04 JS |
881 | * Ensure clocksources that have large 'mult' values don't overflow |
882 | * when adjusted. | |
d65670a7 JS |
883 | */ |
884 | cs->maxadj = clocksource_max_adjustment(cs); | |
f8935983 JS |
885 | while (freq && ((cs->mult + cs->maxadj < cs->mult) |
886 | || (cs->mult - cs->maxadj > cs->mult))) { | |
d65670a7 JS |
887 | cs->mult >>= 1; |
888 | cs->shift--; | |
889 | cs->maxadj = clocksource_max_adjustment(cs); | |
890 | } | |
891 | ||
f8935983 JS |
892 | /* |
893 | * Only warn for *special* clocksources that self-define | |
894 | * their mult/shift values and don't specify a freq. | |
895 | */ | |
896 | WARN_ONCE(cs->mult + cs->maxadj < cs->mult, | |
897 | "timekeeping: Clocksource %s might overflow on 11%% adjustment\n", | |
898 | cs->name); | |
899 | ||
fb82fe2f | 900 | clocksource_update_max_deferment(cs); |
8cc8c525 | 901 | |
45bbfe64 JP |
902 | pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n", |
903 | cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns); | |
852db46d | 904 | } |
fba9e072 | 905 | EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale); |
852db46d JS |
906 | |
907 | /** | |
908 | * __clocksource_register_scale - Used to install new clocksources | |
b1b73d09 | 909 | * @cs: clocksource to be registered |
852db46d JS |
910 | * @scale: Scale factor multiplied against freq to get clocksource hz |
911 | * @freq: clocksource frequency (cycles per second) divided by scale | |
912 | * | |
913 | * Returns -EBUSY if registration fails, zero otherwise. | |
914 | * | |
915 | * This *SHOULD NOT* be called directly! Please use the | |
916 | * clocksource_register_hz() or clocksource_register_khz helper functions. | |
917 | */ | |
918 | int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq) | |
919 | { | |
2aae7bcf | 920 | unsigned long flags; |
852db46d | 921 | |
d67f34c1 TG |
922 | clocksource_arch_init(cs); |
923 | ||
b595076a | 924 | /* Initialize mult/shift and max_idle_ns */ |
fba9e072 | 925 | __clocksource_update_freq_scale(cs, scale, freq); |
d7e81c26 | 926 | |
be278e98 | 927 | /* Add clocksource to the clocksource list */ |
d7e81c26 | 928 | mutex_lock(&clocksource_mutex); |
2aae7bcf PZ |
929 | |
930 | clocksource_watchdog_lock(&flags); | |
d7e81c26 | 931 | clocksource_enqueue(cs); |
d7e81c26 | 932 | clocksource_enqueue_watchdog(cs); |
2aae7bcf PZ |
933 | clocksource_watchdog_unlock(&flags); |
934 | ||
e05b2efb | 935 | clocksource_select(); |
bbf66d89 | 936 | clocksource_select_watchdog(false); |
39232ed5 | 937 | __clocksource_suspend_select(cs); |
d7e81c26 JS |
938 | mutex_unlock(&clocksource_mutex); |
939 | return 0; | |
940 | } | |
941 | EXPORT_SYMBOL_GPL(__clocksource_register_scale); | |
942 | ||
d0981a1b TG |
943 | static void __clocksource_change_rating(struct clocksource *cs, int rating) |
944 | { | |
945 | list_del(&cs->list); | |
946 | cs->rating = rating; | |
947 | clocksource_enqueue(cs); | |
d0981a1b TG |
948 | } |
949 | ||
734efb46 | 950 | /** |
92c7e002 | 951 | * clocksource_change_rating - Change the rating of a registered clocksource |
b1b73d09 KK |
952 | * @cs: clocksource to be changed |
953 | * @rating: new rating | |
734efb46 | 954 | */ |
92c7e002 | 955 | void clocksource_change_rating(struct clocksource *cs, int rating) |
734efb46 | 956 | { |
2aae7bcf PZ |
957 | unsigned long flags; |
958 | ||
75c5158f | 959 | mutex_lock(&clocksource_mutex); |
2aae7bcf | 960 | clocksource_watchdog_lock(&flags); |
d0981a1b | 961 | __clocksource_change_rating(cs, rating); |
2aae7bcf PZ |
962 | clocksource_watchdog_unlock(&flags); |
963 | ||
332962f2 | 964 | clocksource_select(); |
bbf66d89 | 965 | clocksource_select_watchdog(false); |
39232ed5 | 966 | clocksource_suspend_select(false); |
75c5158f | 967 | mutex_unlock(&clocksource_mutex); |
734efb46 | 968 | } |
fb63a0eb | 969 | EXPORT_SYMBOL(clocksource_change_rating); |
734efb46 | 970 | |
7eaeb343 TG |
971 | /* |
972 | * Unbind clocksource @cs. Called with clocksource_mutex held | |
973 | */ | |
974 | static int clocksource_unbind(struct clocksource *cs) | |
975 | { | |
2aae7bcf PZ |
976 | unsigned long flags; |
977 | ||
bbf66d89 VK |
978 | if (clocksource_is_watchdog(cs)) { |
979 | /* Select and try to install a replacement watchdog. */ | |
980 | clocksource_select_watchdog(true); | |
981 | if (clocksource_is_watchdog(cs)) | |
982 | return -EBUSY; | |
983 | } | |
7eaeb343 TG |
984 | |
985 | if (cs == curr_clocksource) { | |
986 | /* Select and try to install a replacement clock source */ | |
987 | clocksource_select_fallback(); | |
988 | if (curr_clocksource == cs) | |
989 | return -EBUSY; | |
990 | } | |
2aae7bcf | 991 | |
39232ed5 BW |
992 | if (clocksource_is_suspend(cs)) { |
993 | /* | |
994 | * Select and try to install a replacement suspend clocksource. | |
995 | * If no replacement suspend clocksource, we will just let the | |
996 | * clocksource go and have no suspend clocksource. | |
997 | */ | |
998 | clocksource_suspend_select(true); | |
999 | } | |
1000 | ||
2aae7bcf | 1001 | clocksource_watchdog_lock(&flags); |
7eaeb343 TG |
1002 | clocksource_dequeue_watchdog(cs); |
1003 | list_del_init(&cs->list); | |
2aae7bcf PZ |
1004 | clocksource_watchdog_unlock(&flags); |
1005 | ||
7eaeb343 TG |
1006 | return 0; |
1007 | } | |
1008 | ||
4713e22c TG |
1009 | /** |
1010 | * clocksource_unregister - remove a registered clocksource | |
b1b73d09 | 1011 | * @cs: clocksource to be unregistered |
4713e22c | 1012 | */ |
a89c7edb | 1013 | int clocksource_unregister(struct clocksource *cs) |
4713e22c | 1014 | { |
a89c7edb TG |
1015 | int ret = 0; |
1016 | ||
75c5158f | 1017 | mutex_lock(&clocksource_mutex); |
a89c7edb TG |
1018 | if (!list_empty(&cs->list)) |
1019 | ret = clocksource_unbind(cs); | |
75c5158f | 1020 | mutex_unlock(&clocksource_mutex); |
a89c7edb | 1021 | return ret; |
4713e22c | 1022 | } |
fb63a0eb | 1023 | EXPORT_SYMBOL(clocksource_unregister); |
4713e22c | 1024 | |
2b013700 | 1025 | #ifdef CONFIG_SYSFS |
734efb46 | 1026 | /** |
e87821d1 | 1027 | * current_clocksource_show - sysfs interface for current clocksource |
734efb46 | 1028 | * @dev: unused |
b1b73d09 | 1029 | * @attr: unused |
734efb46 | 1030 | * @buf: char buffer to be filled with clocksource list |
1031 | * | |
1032 | * Provides sysfs interface for listing current clocksource. | |
1033 | */ | |
e87821d1 BW |
1034 | static ssize_t current_clocksource_show(struct device *dev, |
1035 | struct device_attribute *attr, | |
1036 | char *buf) | |
734efb46 | 1037 | { |
5e2cb101 | 1038 | ssize_t count = 0; |
734efb46 | 1039 | |
75c5158f | 1040 | mutex_lock(&clocksource_mutex); |
5e2cb101 | 1041 | count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name); |
75c5158f | 1042 | mutex_unlock(&clocksource_mutex); |
734efb46 | 1043 | |
5e2cb101 | 1044 | return count; |
734efb46 | 1045 | } |
1046 | ||
891292a7 | 1047 | ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt) |
29b54078 TG |
1048 | { |
1049 | size_t ret = cnt; | |
1050 | ||
1051 | /* strings from sysfs write are not 0 terminated! */ | |
1052 | if (!cnt || cnt >= CS_NAME_LEN) | |
1053 | return -EINVAL; | |
1054 | ||
1055 | /* strip of \n: */ | |
1056 | if (buf[cnt-1] == '\n') | |
1057 | cnt--; | |
1058 | if (cnt > 0) | |
1059 | memcpy(dst, buf, cnt); | |
1060 | dst[cnt] = 0; | |
1061 | return ret; | |
1062 | } | |
1063 | ||
734efb46 | 1064 | /** |
e87821d1 | 1065 | * current_clocksource_store - interface for manually overriding clocksource |
734efb46 | 1066 | * @dev: unused |
b1b73d09 | 1067 | * @attr: unused |
734efb46 | 1068 | * @buf: name of override clocksource |
1069 | * @count: length of buffer | |
1070 | * | |
1071 | * Takes input from sysfs interface for manually overriding the default | |
b71a8eb0 | 1072 | * clocksource selection. |
734efb46 | 1073 | */ |
e87821d1 BW |
1074 | static ssize_t current_clocksource_store(struct device *dev, |
1075 | struct device_attribute *attr, | |
1076 | const char *buf, size_t count) | |
734efb46 | 1077 | { |
233bcb41 | 1078 | ssize_t ret; |
734efb46 | 1079 | |
75c5158f | 1080 | mutex_lock(&clocksource_mutex); |
734efb46 | 1081 | |
03e13cf5 | 1082 | ret = sysfs_get_uname(buf, override_name, count); |
29b54078 TG |
1083 | if (ret >= 0) |
1084 | clocksource_select(); | |
734efb46 | 1085 | |
75c5158f | 1086 | mutex_unlock(&clocksource_mutex); |
734efb46 | 1087 | |
1088 | return ret; | |
1089 | } | |
e87821d1 | 1090 | static DEVICE_ATTR_RW(current_clocksource); |
734efb46 | 1091 | |
7eaeb343 | 1092 | /** |
e87821d1 | 1093 | * unbind_clocksource_store - interface for manually unbinding clocksource |
7eaeb343 TG |
1094 | * @dev: unused |
1095 | * @attr: unused | |
1096 | * @buf: unused | |
1097 | * @count: length of buffer | |
1098 | * | |
1099 | * Takes input from sysfs interface for manually unbinding a clocksource. | |
1100 | */ | |
e87821d1 | 1101 | static ssize_t unbind_clocksource_store(struct device *dev, |
7eaeb343 TG |
1102 | struct device_attribute *attr, |
1103 | const char *buf, size_t count) | |
1104 | { | |
1105 | struct clocksource *cs; | |
1106 | char name[CS_NAME_LEN]; | |
233bcb41 | 1107 | ssize_t ret; |
7eaeb343 | 1108 | |
03e13cf5 | 1109 | ret = sysfs_get_uname(buf, name, count); |
7eaeb343 TG |
1110 | if (ret < 0) |
1111 | return ret; | |
1112 | ||
1113 | ret = -ENODEV; | |
1114 | mutex_lock(&clocksource_mutex); | |
1115 | list_for_each_entry(cs, &clocksource_list, list) { | |
1116 | if (strcmp(cs->name, name)) | |
1117 | continue; | |
1118 | ret = clocksource_unbind(cs); | |
1119 | break; | |
1120 | } | |
1121 | mutex_unlock(&clocksource_mutex); | |
1122 | ||
1123 | return ret ? ret : count; | |
1124 | } | |
e87821d1 | 1125 | static DEVICE_ATTR_WO(unbind_clocksource); |
7eaeb343 | 1126 | |
734efb46 | 1127 | /** |
e87821d1 | 1128 | * available_clocksource_show - sysfs interface for listing clocksource |
734efb46 | 1129 | * @dev: unused |
b1b73d09 | 1130 | * @attr: unused |
734efb46 | 1131 | * @buf: char buffer to be filled with clocksource list |
1132 | * | |
1133 | * Provides sysfs interface for listing registered clocksources | |
1134 | */ | |
e87821d1 BW |
1135 | static ssize_t available_clocksource_show(struct device *dev, |
1136 | struct device_attribute *attr, | |
1137 | char *buf) | |
734efb46 | 1138 | { |
2e197586 | 1139 | struct clocksource *src; |
5e2cb101 | 1140 | ssize_t count = 0; |
734efb46 | 1141 | |
75c5158f | 1142 | mutex_lock(&clocksource_mutex); |
2e197586 | 1143 | list_for_each_entry(src, &clocksource_list, list) { |
cd6d95d8 TG |
1144 | /* |
1145 | * Don't show non-HRES clocksource if the tick code is | |
1146 | * in one shot mode (highres=on or nohz=on) | |
1147 | */ | |
1148 | if (!tick_oneshot_mode_active() || | |
1149 | (src->flags & CLOCK_SOURCE_VALID_FOR_HRES)) | |
3f68535a | 1150 | count += snprintf(buf + count, |
5e2cb101 MX |
1151 | max((ssize_t)PAGE_SIZE - count, (ssize_t)0), |
1152 | "%s ", src->name); | |
734efb46 | 1153 | } |
75c5158f | 1154 | mutex_unlock(&clocksource_mutex); |
734efb46 | 1155 | |
5e2cb101 MX |
1156 | count += snprintf(buf + count, |
1157 | max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n"); | |
734efb46 | 1158 | |
5e2cb101 | 1159 | return count; |
734efb46 | 1160 | } |
e87821d1 | 1161 | static DEVICE_ATTR_RO(available_clocksource); |
734efb46 | 1162 | |
27263e8d BW |
1163 | static struct attribute *clocksource_attrs[] = { |
1164 | &dev_attr_current_clocksource.attr, | |
1165 | &dev_attr_unbind_clocksource.attr, | |
1166 | &dev_attr_available_clocksource.attr, | |
1167 | NULL | |
1168 | }; | |
1169 | ATTRIBUTE_GROUPS(clocksource); | |
1170 | ||
d369a5d8 | 1171 | static struct bus_type clocksource_subsys = { |
af5ca3f4 | 1172 | .name = "clocksource", |
d369a5d8 | 1173 | .dev_name = "clocksource", |
734efb46 | 1174 | }; |
1175 | ||
d369a5d8 | 1176 | static struct device device_clocksource = { |
734efb46 | 1177 | .id = 0, |
d369a5d8 | 1178 | .bus = &clocksource_subsys, |
27263e8d | 1179 | .groups = clocksource_groups, |
734efb46 | 1180 | }; |
1181 | ||
ad596171 | 1182 | static int __init init_clocksource_sysfs(void) |
734efb46 | 1183 | { |
d369a5d8 | 1184 | int error = subsys_system_register(&clocksource_subsys, NULL); |
734efb46 | 1185 | |
1186 | if (!error) | |
d369a5d8 | 1187 | error = device_register(&device_clocksource); |
27263e8d | 1188 | |
734efb46 | 1189 | return error; |
1190 | } | |
1191 | ||
1192 | device_initcall(init_clocksource_sysfs); | |
2b013700 | 1193 | #endif /* CONFIG_SYSFS */ |
734efb46 | 1194 | |
1195 | /** | |
1196 | * boot_override_clocksource - boot clock override | |
1197 | * @str: override name | |
1198 | * | |
1199 | * Takes a clocksource= boot argument and uses it | |
1200 | * as the clocksource override name. | |
1201 | */ | |
1202 | static int __init boot_override_clocksource(char* str) | |
1203 | { | |
75c5158f | 1204 | mutex_lock(&clocksource_mutex); |
734efb46 | 1205 | if (str) |
1206 | strlcpy(override_name, str, sizeof(override_name)); | |
75c5158f | 1207 | mutex_unlock(&clocksource_mutex); |
734efb46 | 1208 | return 1; |
1209 | } | |
1210 | ||
1211 | __setup("clocksource=", boot_override_clocksource); | |
1212 | ||
1213 | /** | |
1214 | * boot_override_clock - Compatibility layer for deprecated boot option | |
1215 | * @str: override name | |
1216 | * | |
1217 | * DEPRECATED! Takes a clock= boot argument and uses it | |
1218 | * as the clocksource override name | |
1219 | */ | |
1220 | static int __init boot_override_clock(char* str) | |
1221 | { | |
5d0cf410 | 1222 | if (!strcmp(str, "pmtmr")) { |
45bbfe64 | 1223 | pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n"); |
5d0cf410 | 1224 | return boot_override_clocksource("acpi_pm"); |
1225 | } | |
45bbfe64 | 1226 | pr_warn("clock= boot option is deprecated - use clocksource=xyz\n"); |
734efb46 | 1227 | return boot_override_clocksource(str); |
1228 | } | |
1229 | ||
1230 | __setup("clock=", boot_override_clock); |