1 | // SPDX-License-Identifier: GPL-2.0+ |
2 | /* |
3 | * This file contains the functions which manage clocksource drivers. |
4 | * |
5 | * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com) |
6 | */ |
7 | |
8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
9 | |
10 | #include <linux/device.h> |
11 | #include <linux/clocksource.h> |
12 | #include <linux/init.h> |
13 | #include <linux/module.h> |
14 | #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */ |
15 | #include <linux/tick.h> |
16 | #include <linux/kthread.h> |
17 | #include <linux/prandom.h> |
18 | #include <linux/cpu.h> |
19 | |
20 | #include "tick-internal.h" |
21 | #include "timekeeping_internal.h" |
22 | |
23 | /** |
24 | * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks |
25 | * @mult: pointer to mult variable |
26 | * @shift: pointer to shift variable |
27 | * @from: frequency to convert from |
28 | * @to: frequency to convert to |
29 | * @maxsec: guaranteed runtime conversion range in seconds |
30 | * |
31 | * The function evaluates the shift/mult pair for the scaled math |
32 | * operations of clocksources and clockevents. |
33 | * |
34 | * @to and @from are frequency values in HZ. For clock sources @to is |
35 | * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock |
36 | * event @to is the counter frequency and @from is NSEC_PER_SEC. |
37 | * |
38 | * The @maxsec conversion range argument controls the time frame in |
39 | * seconds which must be covered by the runtime conversion with the |
40 | * calculated mult and shift factors. This guarantees that no 64bit |
41 | * overflow happens when the input value of the conversion is |
42 | * multiplied with the calculated mult factor. Larger ranges may |
43 | * reduce the conversion accuracy by choosing smaller mult and shift |
44 | * factors. |
45 | */ |
46 | void |
47 | clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec) |
48 | { |
49 | u64 tmp; |
50 | u32 sft, sftacc= 32; |
51 | |
52 | /* |
53 | * Calculate the shift factor which is limiting the conversion |
54 | * range: |
55 | */ |
56 | tmp = ((u64)maxsec * from) >> 32; |
57 | while (tmp) { |
58 | tmp >>=1; |
59 | sftacc--; |
60 | } |
61 | |
62 | /* |
63 | * Find the conversion shift/mult pair which has the best |
64 | * accuracy and fits the maxsec conversion range: |
65 | */ |
66 | for (sft = 32; sft > 0; sft--) { |
67 | tmp = (u64) to << sft; |
68 | tmp += from / 2; |
69 | do_div(tmp, from); |
70 | if ((tmp >> sftacc) == 0) |
71 | break; |
72 | } |
73 | *mult = tmp; |
74 | *shift = sft; |
75 | } |
76 | EXPORT_SYMBOL_GPL(clocks_calc_mult_shift); |
77 | |
78 | /*[Clocksource internal variables]--------- |
79 | * curr_clocksource: |
80 | * currently selected clocksource. |
81 | * suspend_clocksource: |
82 | * used to calculate the suspend time. |
83 | * clocksource_list: |
84 | * linked list with the registered clocksources |
85 | * clocksource_mutex: |
86 | * protects manipulations to curr_clocksource and the clocksource_list |
87 | * override_name: |
88 | * Name of the user-specified clocksource. |
89 | */ |
90 | static struct clocksource *curr_clocksource; |
91 | static struct clocksource *suspend_clocksource; |
92 | static LIST_HEAD(clocksource_list); |
93 | static DEFINE_MUTEX(clocksource_mutex); |
94 | static char override_name[CS_NAME_LEN]; |
95 | static int finished_booting; |
96 | static u64 suspend_start; |
97 | |
98 | /* |
99 | * Interval: 0.5sec. |
100 | */ |
101 | #define WATCHDOG_INTERVAL (HZ >> 1) |
102 | |
103 | /* |
104 | * Threshold: 0.0312s, when doubled: 0.0625s. |
105 | * Also a default for cs->uncertainty_margin when registering clocks. |
106 | */ |
107 | #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5) |
108 | |
109 | /* |
110 | * Maximum permissible delay between two readouts of the watchdog |
111 | * clocksource surrounding a read of the clocksource being validated. |
112 | * This delay could be due to SMIs, NMIs, or to VCPU preemptions. Used as |
113 | * a lower bound for cs->uncertainty_margin values when registering clocks. |
114 | * |
115 | * The default of 500 parts per million is based on NTP's limits. |
116 | * If a clocksource is good enough for NTP, it is good enough for us! |
117 | */ |
118 | #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US |
119 | #define MAX_SKEW_USEC CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US |
120 | #else |
121 | #define MAX_SKEW_USEC (125 * WATCHDOG_INTERVAL / HZ) |
122 | #endif |
123 | |
124 | #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC) |
125 | |
126 | #ifdef CONFIG_CLOCKSOURCE_WATCHDOG |
127 | static void clocksource_watchdog_work(struct work_struct *work); |
128 | static void clocksource_select(void); |
129 | |
130 | static LIST_HEAD(watchdog_list); |
131 | static struct clocksource *watchdog; |
132 | static struct timer_list watchdog_timer; |
133 | static DECLARE_WORK(watchdog_work, clocksource_watchdog_work); |
134 | static DEFINE_SPINLOCK(watchdog_lock); |
135 | static int watchdog_running; |
136 | static atomic_t watchdog_reset_pending; |
137 | |
138 | static inline void clocksource_watchdog_lock(unsigned long *flags) |
139 | { |
140 | spin_lock_irqsave(&watchdog_lock, *flags); |
141 | } |
142 | |
143 | static inline void clocksource_watchdog_unlock(unsigned long *flags) |
144 | { |
145 | spin_unlock_irqrestore(lock: &watchdog_lock, flags: *flags); |
146 | } |
147 | |
148 | static int clocksource_watchdog_kthread(void *data); |
149 | static void __clocksource_change_rating(struct clocksource *cs, int rating); |
150 | |
151 | static void clocksource_watchdog_work(struct work_struct *work) |
152 | { |
153 | /* |
154 | * We cannot directly run clocksource_watchdog_kthread() here, because |
155 | * clocksource_select() calls timekeeping_notify() which uses |
156 | * stop_machine(). One cannot use stop_machine() from a workqueue() due |
157 | * lock inversions wrt CPU hotplug. |
158 | * |
159 | * Also, we only ever run this work once or twice during the lifetime |
160 | * of the kernel, so there is no point in creating a more permanent |
161 | * kthread for this. |
162 | * |
163 | * If kthread_run fails the next watchdog scan over the |
164 | * watchdog_list will find the unstable clock again. |
165 | */ |
166 | kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog" ); |
167 | } |
168 | |
169 | static void __clocksource_unstable(struct clocksource *cs) |
170 | { |
171 | cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG); |
172 | cs->flags |= CLOCK_SOURCE_UNSTABLE; |
173 | |
174 | /* |
175 | * If the clocksource is registered clocksource_watchdog_kthread() will |
176 | * re-rate and re-select. |
177 | */ |
178 | if (list_empty(head: &cs->list)) { |
179 | cs->rating = 0; |
180 | return; |
181 | } |
182 | |
183 | if (cs->mark_unstable) |
184 | cs->mark_unstable(cs); |
185 | |
186 | /* kick clocksource_watchdog_kthread() */ |
187 | if (finished_booting) |
188 | schedule_work(work: &watchdog_work); |
189 | } |
190 | |
191 | /** |
192 | * clocksource_mark_unstable - mark clocksource unstable via watchdog |
193 | * @cs: clocksource to be marked unstable |
194 | * |
195 | * This function is called by the x86 TSC code to mark clocksources as unstable; |
196 | * it defers demotion and re-selection to a kthread. |
197 | */ |
198 | void clocksource_mark_unstable(struct clocksource *cs) |
199 | { |
200 | unsigned long flags; |
201 | |
202 | spin_lock_irqsave(&watchdog_lock, flags); |
203 | if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) { |
204 | if (!list_empty(head: &cs->list) && list_empty(head: &cs->wd_list)) |
205 | list_add(new: &cs->wd_list, head: &watchdog_list); |
206 | __clocksource_unstable(cs); |
207 | } |
208 | spin_unlock_irqrestore(lock: &watchdog_lock, flags); |
209 | } |
210 | |
211 | ulong max_cswd_read_retries = 2; |
212 | module_param(max_cswd_read_retries, ulong, 0644); |
213 | EXPORT_SYMBOL_GPL(max_cswd_read_retries); |
214 | static int verify_n_cpus = 8; |
215 | module_param(verify_n_cpus, int, 0644); |
216 | |
217 | enum wd_read_status { |
218 | WD_READ_SUCCESS, |
219 | WD_READ_UNSTABLE, |
220 | WD_READ_SKIP |
221 | }; |
222 | |
223 | static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow) |
224 | { |
225 | unsigned int nretries; |
226 | u64 wd_end, wd_end2, wd_delta; |
227 | int64_t wd_delay, wd_seq_delay; |
228 | |
229 | for (nretries = 0; nretries <= max_cswd_read_retries; nretries++) { |
230 | local_irq_disable(); |
231 | *wdnow = watchdog->read(watchdog); |
232 | *csnow = cs->read(cs); |
233 | wd_end = watchdog->read(watchdog); |
234 | wd_end2 = watchdog->read(watchdog); |
235 | local_irq_enable(); |
236 | |
237 | wd_delta = clocksource_delta(now: wd_end, last: *wdnow, mask: watchdog->mask); |
238 | wd_delay = clocksource_cyc2ns(cycles: wd_delta, mult: watchdog->mult, |
239 | shift: watchdog->shift); |
240 | if (wd_delay <= WATCHDOG_MAX_SKEW) { |
241 | if (nretries > 1 || nretries >= max_cswd_read_retries) { |
242 | pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n" , |
243 | smp_processor_id(), watchdog->name, nretries); |
244 | } |
245 | return WD_READ_SUCCESS; |
246 | } |
247 | |
248 | /* |
249 | * Now compute delay in consecutive watchdog read to see if |
250 | * there is too much external interferences that cause |
251 | * significant delay in reading both clocksource and watchdog. |
252 | * |
253 | * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2, |
254 | * report system busy, reinit the watchdog and skip the current |
255 | * watchdog test. |
256 | */ |
257 | wd_delta = clocksource_delta(now: wd_end2, last: wd_end, mask: watchdog->mask); |
258 | wd_seq_delay = clocksource_cyc2ns(cycles: wd_delta, mult: watchdog->mult, shift: watchdog->shift); |
259 | if (wd_seq_delay > WATCHDOG_MAX_SKEW/2) |
260 | goto skip_test; |
261 | } |
262 | |
263 | pr_warn("timekeeping watchdog on CPU%d: wd-%s-wd excessive read-back delay of %lldns vs. limit of %ldns, wd-wd read-back delay only %lldns, attempt %d, marking %s unstable\n" , |
264 | smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name); |
265 | return WD_READ_UNSTABLE; |
266 | |
267 | skip_test: |
268 | pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n" , |
269 | smp_processor_id(), watchdog->name, wd_seq_delay); |
270 | pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n" , |
271 | cs->name, wd_delay); |
272 | return WD_READ_SKIP; |
273 | } |
274 | |
275 | static u64 csnow_mid; |
276 | static cpumask_t cpus_ahead; |
277 | static cpumask_t cpus_behind; |
278 | static cpumask_t cpus_chosen; |
279 | |
280 | static void clocksource_verify_choose_cpus(void) |
281 | { |
282 | int cpu, i, n = verify_n_cpus; |
283 | |
284 | if (n < 0) { |
285 | /* Check all of the CPUs. */ |
286 | cpumask_copy(dstp: &cpus_chosen, cpu_online_mask); |
287 | cpumask_clear_cpu(smp_processor_id(), dstp: &cpus_chosen); |
288 | return; |
289 | } |
290 | |
291 | /* If no checking desired, or no other CPU to check, leave. */ |
292 | cpumask_clear(dstp: &cpus_chosen); |
293 | if (n == 0 || num_online_cpus() <= 1) |
294 | return; |
295 | |
296 | /* Make sure to select at least one CPU other than the current CPU. */ |
297 | cpu = cpumask_first(cpu_online_mask); |
298 | if (cpu == smp_processor_id()) |
299 | cpu = cpumask_next(n: cpu, cpu_online_mask); |
300 | if (WARN_ON_ONCE(cpu >= nr_cpu_ids)) |
301 | return; |
302 | cpumask_set_cpu(cpu, dstp: &cpus_chosen); |
303 | |
304 | /* Force a sane value for the boot parameter. */ |
305 | if (n > nr_cpu_ids) |
306 | n = nr_cpu_ids; |
307 | |
308 | /* |
309 | * Randomly select the specified number of CPUs. If the same |
310 | * CPU is selected multiple times, that CPU is checked only once, |
311 | * and no replacement CPU is selected. This gracefully handles |
312 | * situations where verify_n_cpus is greater than the number of |
313 | * CPUs that are currently online. |
314 | */ |
315 | for (i = 1; i < n; i++) { |
316 | cpu = get_random_u32_below(ceil: nr_cpu_ids); |
317 | cpu = cpumask_next(n: cpu - 1, cpu_online_mask); |
318 | if (cpu >= nr_cpu_ids) |
319 | cpu = cpumask_first(cpu_online_mask); |
320 | if (!WARN_ON_ONCE(cpu >= nr_cpu_ids)) |
321 | cpumask_set_cpu(cpu, dstp: &cpus_chosen); |
322 | } |
323 | |
324 | /* Don't verify ourselves. */ |
325 | cpumask_clear_cpu(smp_processor_id(), dstp: &cpus_chosen); |
326 | } |
327 | |
328 | static void clocksource_verify_one_cpu(void *csin) |
329 | { |
330 | struct clocksource *cs = (struct clocksource *)csin; |
331 | |
332 | csnow_mid = cs->read(cs); |
333 | } |
334 | |
335 | void clocksource_verify_percpu(struct clocksource *cs) |
336 | { |
337 | int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX; |
338 | u64 csnow_begin, csnow_end; |
339 | int cpu, testcpu; |
340 | s64 delta; |
341 | |
342 | if (verify_n_cpus == 0) |
343 | return; |
344 | cpumask_clear(dstp: &cpus_ahead); |
345 | cpumask_clear(dstp: &cpus_behind); |
346 | cpus_read_lock(); |
347 | preempt_disable(); |
348 | clocksource_verify_choose_cpus(); |
349 | if (cpumask_empty(srcp: &cpus_chosen)) { |
350 | preempt_enable(); |
351 | cpus_read_unlock(); |
352 | pr_warn("Not enough CPUs to check clocksource '%s'.\n" , cs->name); |
353 | return; |
354 | } |
355 | testcpu = smp_processor_id(); |
356 | pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n" , cs->name, testcpu, cpumask_pr_args(&cpus_chosen)); |
357 | for_each_cpu(cpu, &cpus_chosen) { |
358 | if (cpu == testcpu) |
359 | continue; |
360 | csnow_begin = cs->read(cs); |
361 | smp_call_function_single(cpuid: cpu, func: clocksource_verify_one_cpu, info: cs, wait: 1); |
362 | csnow_end = cs->read(cs); |
363 | delta = (s64)((csnow_mid - csnow_begin) & cs->mask); |
364 | if (delta < 0) |
365 | cpumask_set_cpu(cpu, dstp: &cpus_behind); |
366 | delta = (csnow_end - csnow_mid) & cs->mask; |
367 | if (delta < 0) |
368 | cpumask_set_cpu(cpu, dstp: &cpus_ahead); |
369 | delta = clocksource_delta(now: csnow_end, last: csnow_begin, mask: cs->mask); |
370 | cs_nsec = clocksource_cyc2ns(cycles: delta, mult: cs->mult, shift: cs->shift); |
371 | if (cs_nsec > cs_nsec_max) |
372 | cs_nsec_max = cs_nsec; |
373 | if (cs_nsec < cs_nsec_min) |
374 | cs_nsec_min = cs_nsec; |
375 | } |
376 | preempt_enable(); |
377 | cpus_read_unlock(); |
378 | if (!cpumask_empty(srcp: &cpus_ahead)) |
379 | pr_warn(" CPUs %*pbl ahead of CPU %d for clocksource %s.\n" , |
380 | cpumask_pr_args(&cpus_ahead), testcpu, cs->name); |
381 | if (!cpumask_empty(srcp: &cpus_behind)) |
382 | pr_warn(" CPUs %*pbl behind CPU %d for clocksource %s.\n" , |
383 | cpumask_pr_args(&cpus_behind), testcpu, cs->name); |
384 | if (!cpumask_empty(srcp: &cpus_ahead) || !cpumask_empty(srcp: &cpus_behind)) |
385 | pr_warn(" CPU %d check durations %lldns - %lldns for clocksource %s.\n" , |
386 | testcpu, cs_nsec_min, cs_nsec_max, cs->name); |
387 | } |
388 | EXPORT_SYMBOL_GPL(clocksource_verify_percpu); |
389 | |
390 | static inline void clocksource_reset_watchdog(void) |
391 | { |
392 | struct clocksource *cs; |
393 | |
394 | list_for_each_entry(cs, &watchdog_list, wd_list) |
395 | cs->flags &= ~CLOCK_SOURCE_WATCHDOG; |
396 | } |
397 | |
398 | |
399 | static void clocksource_watchdog(struct timer_list *unused) |
400 | { |
401 | u64 csnow, wdnow, cslast, wdlast, delta; |
402 | int next_cpu, reset_pending; |
403 | int64_t wd_nsec, cs_nsec; |
404 | struct clocksource *cs; |
405 | enum wd_read_status read_ret; |
406 | unsigned long = 0; |
407 | u32 md; |
408 | |
409 | spin_lock(lock: &watchdog_lock); |
410 | if (!watchdog_running) |
411 | goto out; |
412 | |
413 | reset_pending = atomic_read(v: &watchdog_reset_pending); |
414 | |
415 | list_for_each_entry(cs, &watchdog_list, wd_list) { |
416 | |
417 | /* Clocksource already marked unstable? */ |
418 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) { |
419 | if (finished_booting) |
420 | schedule_work(work: &watchdog_work); |
421 | continue; |
422 | } |
423 | |
424 | read_ret = cs_watchdog_read(cs, csnow: &csnow, wdnow: &wdnow); |
425 | |
426 | if (read_ret == WD_READ_UNSTABLE) { |
427 | /* Clock readout unreliable, so give it up. */ |
428 | __clocksource_unstable(cs); |
429 | continue; |
430 | } |
431 | |
432 | /* |
433 | * When WD_READ_SKIP is returned, it means the system is likely |
434 | * under very heavy load, where the latency of reading |
435 | * watchdog/clocksource is very big, and affect the accuracy of |
436 | * watchdog check. So give system some space and suspend the |
437 | * watchdog check for 5 minutes. |
438 | */ |
439 | if (read_ret == WD_READ_SKIP) { |
440 | /* |
441 | * As the watchdog timer will be suspended, and |
442 | * cs->last could keep unchanged for 5 minutes, reset |
443 | * the counters. |
444 | */ |
445 | clocksource_reset_watchdog(); |
446 | extra_wait = HZ * 300; |
447 | break; |
448 | } |
449 | |
450 | /* Clocksource initialized ? */ |
451 | if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) || |
452 | atomic_read(v: &watchdog_reset_pending)) { |
453 | cs->flags |= CLOCK_SOURCE_WATCHDOG; |
454 | cs->wd_last = wdnow; |
455 | cs->cs_last = csnow; |
456 | continue; |
457 | } |
458 | |
459 | delta = clocksource_delta(now: wdnow, last: cs->wd_last, mask: watchdog->mask); |
460 | wd_nsec = clocksource_cyc2ns(cycles: delta, mult: watchdog->mult, |
461 | shift: watchdog->shift); |
462 | |
463 | delta = clocksource_delta(now: csnow, last: cs->cs_last, mask: cs->mask); |
464 | cs_nsec = clocksource_cyc2ns(cycles: delta, mult: cs->mult, shift: cs->shift); |
465 | wdlast = cs->wd_last; /* save these in case we print them */ |
466 | cslast = cs->cs_last; |
467 | cs->cs_last = csnow; |
468 | cs->wd_last = wdnow; |
469 | |
470 | if (atomic_read(v: &watchdog_reset_pending)) |
471 | continue; |
472 | |
473 | /* Check the deviation from the watchdog clocksource. */ |
474 | md = cs->uncertainty_margin + watchdog->uncertainty_margin; |
475 | if (abs(cs_nsec - wd_nsec) > md) { |
476 | s64 cs_wd_msec; |
477 | s64 wd_msec; |
478 | u32 wd_rem; |
479 | |
480 | pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n" , |
481 | smp_processor_id(), cs->name); |
482 | pr_warn(" '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n" , |
483 | watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask); |
484 | pr_warn(" '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n" , |
485 | cs->name, cs_nsec, csnow, cslast, cs->mask); |
486 | cs_wd_msec = div_s64_rem(dividend: cs_nsec - wd_nsec, divisor: 1000 * 1000, remainder: &wd_rem); |
487 | wd_msec = div_s64_rem(dividend: wd_nsec, divisor: 1000 * 1000, remainder: &wd_rem); |
488 | pr_warn(" Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n" , |
489 | cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec); |
490 | if (curr_clocksource == cs) |
491 | pr_warn(" '%s' is current clocksource.\n" , cs->name); |
492 | else if (curr_clocksource) |
493 | pr_warn(" '%s' (not '%s') is current clocksource.\n" , curr_clocksource->name, cs->name); |
494 | else |
495 | pr_warn(" No current clocksource.\n" ); |
496 | __clocksource_unstable(cs); |
497 | continue; |
498 | } |
499 | |
500 | if (cs == curr_clocksource && cs->tick_stable) |
501 | cs->tick_stable(cs); |
502 | |
503 | if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && |
504 | (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) && |
505 | (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) { |
506 | /* Mark it valid for high-res. */ |
507 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; |
508 | |
509 | /* |
510 | * clocksource_done_booting() will sort it if |
511 | * finished_booting is not set yet. |
512 | */ |
513 | if (!finished_booting) |
514 | continue; |
515 | |
516 | /* |
517 | * If this is not the current clocksource let |
518 | * the watchdog thread reselect it. Due to the |
519 | * change to high res this clocksource might |
520 | * be preferred now. If it is the current |
521 | * clocksource let the tick code know about |
522 | * that change. |
523 | */ |
524 | if (cs != curr_clocksource) { |
525 | cs->flags |= CLOCK_SOURCE_RESELECT; |
526 | schedule_work(work: &watchdog_work); |
527 | } else { |
528 | tick_clock_notify(); |
529 | } |
530 | } |
531 | } |
532 | |
533 | /* |
534 | * We only clear the watchdog_reset_pending, when we did a |
535 | * full cycle through all clocksources. |
536 | */ |
537 | if (reset_pending) |
538 | atomic_dec(v: &watchdog_reset_pending); |
539 | |
540 | /* |
541 | * Cycle through CPUs to check if the CPUs stay synchronized |
542 | * to each other. |
543 | */ |
544 | next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask); |
545 | if (next_cpu >= nr_cpu_ids) |
546 | next_cpu = cpumask_first(cpu_online_mask); |
547 | |
548 | /* |
549 | * Arm timer if not already pending: could race with concurrent |
550 | * pair clocksource_stop_watchdog() clocksource_start_watchdog(). |
551 | */ |
552 | if (!timer_pending(timer: &watchdog_timer)) { |
553 | watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait; |
554 | add_timer_on(timer: &watchdog_timer, cpu: next_cpu); |
555 | } |
556 | out: |
557 | spin_unlock(lock: &watchdog_lock); |
558 | } |
559 | |
560 | static inline void clocksource_start_watchdog(void) |
561 | { |
562 | if (watchdog_running || !watchdog || list_empty(head: &watchdog_list)) |
563 | return; |
564 | timer_setup(&watchdog_timer, clocksource_watchdog, 0); |
565 | watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL; |
566 | add_timer_on(timer: &watchdog_timer, cpu: cpumask_first(cpu_online_mask)); |
567 | watchdog_running = 1; |
568 | } |
569 | |
570 | static inline void clocksource_stop_watchdog(void) |
571 | { |
572 | if (!watchdog_running || (watchdog && !list_empty(head: &watchdog_list))) |
573 | return; |
574 | del_timer(timer: &watchdog_timer); |
575 | watchdog_running = 0; |
576 | } |
577 | |
578 | static void clocksource_resume_watchdog(void) |
579 | { |
580 | atomic_inc(v: &watchdog_reset_pending); |
581 | } |
582 | |
583 | static void clocksource_enqueue_watchdog(struct clocksource *cs) |
584 | { |
585 | INIT_LIST_HEAD(list: &cs->wd_list); |
586 | |
587 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { |
588 | /* cs is a clocksource to be watched. */ |
589 | list_add(new: &cs->wd_list, head: &watchdog_list); |
590 | cs->flags &= ~CLOCK_SOURCE_WATCHDOG; |
591 | } else { |
592 | /* cs is a watchdog. */ |
593 | if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) |
594 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; |
595 | } |
596 | } |
597 | |
598 | static void clocksource_select_watchdog(bool fallback) |
599 | { |
600 | struct clocksource *cs, *old_wd; |
601 | unsigned long flags; |
602 | |
603 | spin_lock_irqsave(&watchdog_lock, flags); |
604 | /* save current watchdog */ |
605 | old_wd = watchdog; |
606 | if (fallback) |
607 | watchdog = NULL; |
608 | |
609 | list_for_each_entry(cs, &clocksource_list, list) { |
610 | /* cs is a clocksource to be watched. */ |
611 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) |
612 | continue; |
613 | |
614 | /* Skip current if we were requested for a fallback. */ |
615 | if (fallback && cs == old_wd) |
616 | continue; |
617 | |
618 | /* Pick the best watchdog. */ |
619 | if (!watchdog || cs->rating > watchdog->rating) |
620 | watchdog = cs; |
621 | } |
622 | /* If we failed to find a fallback restore the old one. */ |
623 | if (!watchdog) |
624 | watchdog = old_wd; |
625 | |
626 | /* If we changed the watchdog we need to reset cycles. */ |
627 | if (watchdog != old_wd) |
628 | clocksource_reset_watchdog(); |
629 | |
630 | /* Check if the watchdog timer needs to be started. */ |
631 | clocksource_start_watchdog(); |
632 | spin_unlock_irqrestore(lock: &watchdog_lock, flags); |
633 | } |
634 | |
635 | static void clocksource_dequeue_watchdog(struct clocksource *cs) |
636 | { |
637 | if (cs != watchdog) { |
638 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { |
639 | /* cs is a watched clocksource. */ |
640 | list_del_init(entry: &cs->wd_list); |
641 | /* Check if the watchdog timer needs to be stopped. */ |
642 | clocksource_stop_watchdog(); |
643 | } |
644 | } |
645 | } |
646 | |
647 | static int __clocksource_watchdog_kthread(void) |
648 | { |
649 | struct clocksource *cs, *tmp; |
650 | unsigned long flags; |
651 | int select = 0; |
652 | |
653 | /* Do any required per-CPU skew verification. */ |
654 | if (curr_clocksource && |
655 | curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE && |
656 | curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU) |
657 | clocksource_verify_percpu(curr_clocksource); |
658 | |
659 | spin_lock_irqsave(&watchdog_lock, flags); |
660 | list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) { |
661 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) { |
662 | list_del_init(entry: &cs->wd_list); |
663 | __clocksource_change_rating(cs, rating: 0); |
664 | select = 1; |
665 | } |
666 | if (cs->flags & CLOCK_SOURCE_RESELECT) { |
667 | cs->flags &= ~CLOCK_SOURCE_RESELECT; |
668 | select = 1; |
669 | } |
670 | } |
671 | /* Check if the watchdog timer needs to be stopped. */ |
672 | clocksource_stop_watchdog(); |
673 | spin_unlock_irqrestore(lock: &watchdog_lock, flags); |
674 | |
675 | return select; |
676 | } |
677 | |
678 | static int clocksource_watchdog_kthread(void *data) |
679 | { |
680 | mutex_lock(&clocksource_mutex); |
681 | if (__clocksource_watchdog_kthread()) |
682 | clocksource_select(); |
683 | mutex_unlock(lock: &clocksource_mutex); |
684 | return 0; |
685 | } |
686 | |
687 | static bool clocksource_is_watchdog(struct clocksource *cs) |
688 | { |
689 | return cs == watchdog; |
690 | } |
691 | |
692 | #else /* CONFIG_CLOCKSOURCE_WATCHDOG */ |
693 | |
694 | static void clocksource_enqueue_watchdog(struct clocksource *cs) |
695 | { |
696 | if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) |
697 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; |
698 | } |
699 | |
700 | static void clocksource_select_watchdog(bool fallback) { } |
701 | static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { } |
702 | static inline void clocksource_resume_watchdog(void) { } |
703 | static inline int __clocksource_watchdog_kthread(void) { return 0; } |
704 | static bool clocksource_is_watchdog(struct clocksource *cs) { return false; } |
705 | void clocksource_mark_unstable(struct clocksource *cs) { } |
706 | |
707 | static inline void clocksource_watchdog_lock(unsigned long *flags) { } |
708 | static inline void clocksource_watchdog_unlock(unsigned long *flags) { } |
709 | |
710 | #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */ |
711 | |
712 | static bool clocksource_is_suspend(struct clocksource *cs) |
713 | { |
714 | return cs == suspend_clocksource; |
715 | } |
716 | |
717 | static void __clocksource_suspend_select(struct clocksource *cs) |
718 | { |
719 | /* |
720 | * Skip the clocksource which will be stopped in suspend state. |
721 | */ |
722 | if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP)) |
723 | return; |
724 | |
725 | /* |
726 | * The nonstop clocksource can be selected as the suspend clocksource to |
727 | * calculate the suspend time, so it should not supply suspend/resume |
728 | * interfaces to suspend the nonstop clocksource when system suspends. |
729 | */ |
730 | if (cs->suspend || cs->resume) { |
731 | pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n" , |
732 | cs->name); |
733 | } |
734 | |
735 | /* Pick the best rating. */ |
736 | if (!suspend_clocksource || cs->rating > suspend_clocksource->rating) |
737 | suspend_clocksource = cs; |
738 | } |
739 | |
740 | /** |
741 | * clocksource_suspend_select - Select the best clocksource for suspend timing |
742 | * @fallback: if select a fallback clocksource |
743 | */ |
744 | static void clocksource_suspend_select(bool fallback) |
745 | { |
746 | struct clocksource *cs, *old_suspend; |
747 | |
748 | old_suspend = suspend_clocksource; |
749 | if (fallback) |
750 | suspend_clocksource = NULL; |
751 | |
752 | list_for_each_entry(cs, &clocksource_list, list) { |
753 | /* Skip current if we were requested for a fallback. */ |
754 | if (fallback && cs == old_suspend) |
755 | continue; |
756 | |
757 | __clocksource_suspend_select(cs); |
758 | } |
759 | } |
760 | |
761 | /** |
762 | * clocksource_start_suspend_timing - Start measuring the suspend timing |
763 | * @cs: current clocksource from timekeeping |
764 | * @start_cycles: current cycles from timekeeping |
765 | * |
766 | * This function will save the start cycle values of suspend timer to calculate |
767 | * the suspend time when resuming system. |
768 | * |
769 | * This function is called late in the suspend process from timekeeping_suspend(), |
770 | * that means processes are frozen, non-boot cpus and interrupts are disabled |
771 | * now. It is therefore possible to start the suspend timer without taking the |
772 | * clocksource mutex. |
773 | */ |
774 | void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles) |
775 | { |
776 | if (!suspend_clocksource) |
777 | return; |
778 | |
779 | /* |
780 | * If current clocksource is the suspend timer, we should use the |
781 | * tkr_mono.cycle_last value as suspend_start to avoid same reading |
782 | * from suspend timer. |
783 | */ |
784 | if (clocksource_is_suspend(cs)) { |
785 | suspend_start = start_cycles; |
786 | return; |
787 | } |
788 | |
789 | if (suspend_clocksource->enable && |
790 | suspend_clocksource->enable(suspend_clocksource)) { |
791 | pr_warn_once("Failed to enable the non-suspend-able clocksource.\n" ); |
792 | return; |
793 | } |
794 | |
795 | suspend_start = suspend_clocksource->read(suspend_clocksource); |
796 | } |
797 | |
798 | /** |
799 | * clocksource_stop_suspend_timing - Stop measuring the suspend timing |
800 | * @cs: current clocksource from timekeeping |
801 | * @cycle_now: current cycles from timekeeping |
802 | * |
803 | * This function will calculate the suspend time from suspend timer. |
804 | * |
805 | * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource. |
806 | * |
807 | * This function is called early in the resume process from timekeeping_resume(), |
808 | * that means there is only one cpu, no processes are running and the interrupts |
809 | * are disabled. It is therefore possible to stop the suspend timer without |
810 | * taking the clocksource mutex. |
811 | */ |
812 | u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now) |
813 | { |
814 | u64 now, delta, nsec = 0; |
815 | |
816 | if (!suspend_clocksource) |
817 | return 0; |
818 | |
819 | /* |
820 | * If current clocksource is the suspend timer, we should use the |
821 | * tkr_mono.cycle_last value from timekeeping as current cycle to |
822 | * avoid same reading from suspend timer. |
823 | */ |
824 | if (clocksource_is_suspend(cs)) |
825 | now = cycle_now; |
826 | else |
827 | now = suspend_clocksource->read(suspend_clocksource); |
828 | |
829 | if (now > suspend_start) { |
830 | delta = clocksource_delta(now, last: suspend_start, |
831 | mask: suspend_clocksource->mask); |
832 | nsec = mul_u64_u32_shr(a: delta, mul: suspend_clocksource->mult, |
833 | shift: suspend_clocksource->shift); |
834 | } |
835 | |
836 | /* |
837 | * Disable the suspend timer to save power if current clocksource is |
838 | * not the suspend timer. |
839 | */ |
840 | if (!clocksource_is_suspend(cs) && suspend_clocksource->disable) |
841 | suspend_clocksource->disable(suspend_clocksource); |
842 | |
843 | return nsec; |
844 | } |
845 | |
846 | /** |
847 | * clocksource_suspend - suspend the clocksource(s) |
848 | */ |
849 | void clocksource_suspend(void) |
850 | { |
851 | struct clocksource *cs; |
852 | |
853 | list_for_each_entry_reverse(cs, &clocksource_list, list) |
854 | if (cs->suspend) |
855 | cs->suspend(cs); |
856 | } |
857 | |
858 | /** |
859 | * clocksource_resume - resume the clocksource(s) |
860 | */ |
861 | void clocksource_resume(void) |
862 | { |
863 | struct clocksource *cs; |
864 | |
865 | list_for_each_entry(cs, &clocksource_list, list) |
866 | if (cs->resume) |
867 | cs->resume(cs); |
868 | |
869 | clocksource_resume_watchdog(); |
870 | } |
871 | |
872 | /** |
873 | * clocksource_touch_watchdog - Update watchdog |
874 | * |
875 | * Update the watchdog after exception contexts such as kgdb so as not |
876 | * to incorrectly trip the watchdog. This might fail when the kernel |
877 | * was stopped in code which holds watchdog_lock. |
878 | */ |
879 | void clocksource_touch_watchdog(void) |
880 | { |
881 | clocksource_resume_watchdog(); |
882 | } |
883 | |
884 | /** |
885 | * clocksource_max_adjustment- Returns max adjustment amount |
886 | * @cs: Pointer to clocksource |
887 | * |
888 | */ |
889 | static u32 clocksource_max_adjustment(struct clocksource *cs) |
890 | { |
891 | u64 ret; |
892 | /* |
893 | * We won't try to correct for more than 11% adjustments (110,000 ppm), |
894 | */ |
895 | ret = (u64)cs->mult * 11; |
896 | do_div(ret,100); |
897 | return (u32)ret; |
898 | } |
899 | |
900 | /** |
901 | * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted |
902 | * @mult: cycle to nanosecond multiplier |
903 | * @shift: cycle to nanosecond divisor (power of two) |
904 | * @maxadj: maximum adjustment value to mult (~11%) |
905 | * @mask: bitmask for two's complement subtraction of non 64 bit counters |
906 | * @max_cyc: maximum cycle value before potential overflow (does not include |
907 | * any safety margin) |
908 | * |
909 | * NOTE: This function includes a safety margin of 50%, in other words, we |
910 | * return half the number of nanoseconds the hardware counter can technically |
911 | * cover. This is done so that we can potentially detect problems caused by |
912 | * delayed timers or bad hardware, which might result in time intervals that |
913 | * are larger than what the math used can handle without overflows. |
914 | */ |
915 | u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc) |
916 | { |
917 | u64 max_nsecs, max_cycles; |
918 | |
919 | /* |
920 | * Calculate the maximum number of cycles that we can pass to the |
921 | * cyc2ns() function without overflowing a 64-bit result. |
922 | */ |
923 | max_cycles = ULLONG_MAX; |
924 | do_div(max_cycles, mult+maxadj); |
925 | |
926 | /* |
927 | * The actual maximum number of cycles we can defer the clocksource is |
928 | * determined by the minimum of max_cycles and mask. |
929 | * Note: Here we subtract the maxadj to make sure we don't sleep for |
930 | * too long if there's a large negative adjustment. |
931 | */ |
932 | max_cycles = min(max_cycles, mask); |
933 | max_nsecs = clocksource_cyc2ns(cycles: max_cycles, mult: mult - maxadj, shift); |
934 | |
935 | /* return the max_cycles value as well if requested */ |
936 | if (max_cyc) |
937 | *max_cyc = max_cycles; |
938 | |
939 | /* Return 50% of the actual maximum, so we can detect bad values */ |
940 | max_nsecs >>= 1; |
941 | |
942 | return max_nsecs; |
943 | } |
944 | |
945 | /** |
946 | * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles |
947 | * @cs: Pointer to clocksource to be updated |
948 | * |
949 | */ |
950 | static inline void clocksource_update_max_deferment(struct clocksource *cs) |
951 | { |
952 | cs->max_idle_ns = clocks_calc_max_nsecs(mult: cs->mult, shift: cs->shift, |
953 | maxadj: cs->maxadj, mask: cs->mask, |
954 | max_cyc: &cs->max_cycles); |
955 | } |
956 | |
957 | static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur) |
958 | { |
959 | struct clocksource *cs; |
960 | |
961 | if (!finished_booting || list_empty(head: &clocksource_list)) |
962 | return NULL; |
963 | |
964 | /* |
965 | * We pick the clocksource with the highest rating. If oneshot |
966 | * mode is active, we pick the highres valid clocksource with |
967 | * the best rating. |
968 | */ |
969 | list_for_each_entry(cs, &clocksource_list, list) { |
970 | if (skipcur && cs == curr_clocksource) |
971 | continue; |
972 | if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES)) |
973 | continue; |
974 | return cs; |
975 | } |
976 | return NULL; |
977 | } |
978 | |
979 | static void __clocksource_select(bool skipcur) |
980 | { |
981 | bool oneshot = tick_oneshot_mode_active(); |
982 | struct clocksource *best, *cs; |
983 | |
984 | /* Find the best suitable clocksource */ |
985 | best = clocksource_find_best(oneshot, skipcur); |
986 | if (!best) |
987 | return; |
988 | |
989 | if (!strlen(override_name)) |
990 | goto found; |
991 | |
992 | /* Check for the override clocksource. */ |
993 | list_for_each_entry(cs, &clocksource_list, list) { |
994 | if (skipcur && cs == curr_clocksource) |
995 | continue; |
996 | if (strcmp(cs->name, override_name) != 0) |
997 | continue; |
998 | /* |
999 | * Check to make sure we don't switch to a non-highres |
1000 | * capable clocksource if the tick code is in oneshot |
1001 | * mode (highres or nohz) |
1002 | */ |
1003 | if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) { |
1004 | /* Override clocksource cannot be used. */ |
1005 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) { |
1006 | pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n" , |
1007 | cs->name); |
1008 | override_name[0] = 0; |
1009 | } else { |
1010 | /* |
1011 | * The override cannot be currently verified. |
1012 | * Deferring to let the watchdog check. |
1013 | */ |
1014 | pr_info("Override clocksource %s is not currently HRT compatible - deferring\n" , |
1015 | cs->name); |
1016 | } |
1017 | } else |
1018 | /* Override clocksource can be used. */ |
1019 | best = cs; |
1020 | break; |
1021 | } |
1022 | |
1023 | found: |
1024 | if (curr_clocksource != best && !timekeeping_notify(clock: best)) { |
1025 | pr_info("Switched to clocksource %s\n" , best->name); |
1026 | curr_clocksource = best; |
1027 | } |
1028 | } |
1029 | |
1030 | /** |
1031 | * clocksource_select - Select the best clocksource available |
1032 | * |
1033 | * Private function. Must hold clocksource_mutex when called. |
1034 | * |
1035 | * Select the clocksource with the best rating, or the clocksource, |
1036 | * which is selected by userspace override. |
1037 | */ |
1038 | static void clocksource_select(void) |
1039 | { |
1040 | __clocksource_select(skipcur: false); |
1041 | } |
1042 | |
1043 | static void clocksource_select_fallback(void) |
1044 | { |
1045 | __clocksource_select(skipcur: true); |
1046 | } |
1047 | |
1048 | /* |
1049 | * clocksource_done_booting - Called near the end of core bootup |
1050 | * |
1051 | * Hack to avoid lots of clocksource churn at boot time. |
1052 | * We use fs_initcall because we want this to start before |
1053 | * device_initcall but after subsys_initcall. |
1054 | */ |
1055 | static int __init clocksource_done_booting(void) |
1056 | { |
1057 | mutex_lock(&clocksource_mutex); |
1058 | curr_clocksource = clocksource_default_clock(); |
1059 | finished_booting = 1; |
1060 | /* |
1061 | * Run the watchdog first to eliminate unstable clock sources |
1062 | */ |
1063 | __clocksource_watchdog_kthread(); |
1064 | clocksource_select(); |
1065 | mutex_unlock(lock: &clocksource_mutex); |
1066 | return 0; |
1067 | } |
1068 | fs_initcall(clocksource_done_booting); |
1069 | |
1070 | /* |
1071 | * Enqueue the clocksource sorted by rating |
1072 | */ |
1073 | static void clocksource_enqueue(struct clocksource *cs) |
1074 | { |
1075 | struct list_head *entry = &clocksource_list; |
1076 | struct clocksource *tmp; |
1077 | |
1078 | list_for_each_entry(tmp, &clocksource_list, list) { |
1079 | /* Keep track of the place, where to insert */ |
1080 | if (tmp->rating < cs->rating) |
1081 | break; |
1082 | entry = &tmp->list; |
1083 | } |
1084 | list_add(new: &cs->list, head: entry); |
1085 | } |
1086 | |
1087 | /** |
1088 | * __clocksource_update_freq_scale - Used update clocksource with new freq |
1089 | * @cs: clocksource to be registered |
1090 | * @scale: Scale factor multiplied against freq to get clocksource hz |
1091 | * @freq: clocksource frequency (cycles per second) divided by scale |
1092 | * |
1093 | * This should only be called from the clocksource->enable() method. |
1094 | * |
1095 | * This *SHOULD NOT* be called directly! Please use the |
1096 | * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper |
1097 | * functions. |
1098 | */ |
1099 | void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq) |
1100 | { |
1101 | u64 sec; |
1102 | |
1103 | /* |
1104 | * Default clocksources are *special* and self-define their mult/shift. |
1105 | * But, you're not special, so you should specify a freq value. |
1106 | */ |
1107 | if (freq) { |
1108 | /* |
1109 | * Calc the maximum number of seconds which we can run before |
1110 | * wrapping around. For clocksources which have a mask > 32-bit |
1111 | * we need to limit the max sleep time to have a good |
1112 | * conversion precision. 10 minutes is still a reasonable |
1113 | * amount. That results in a shift value of 24 for a |
1114 | * clocksource with mask >= 40-bit and f >= 4GHz. That maps to |
1115 | * ~ 0.06ppm granularity for NTP. |
1116 | */ |
1117 | sec = cs->mask; |
1118 | do_div(sec, freq); |
1119 | do_div(sec, scale); |
1120 | if (!sec) |
1121 | sec = 1; |
1122 | else if (sec > 600 && cs->mask > UINT_MAX) |
1123 | sec = 600; |
1124 | |
1125 | clocks_calc_mult_shift(&cs->mult, &cs->shift, freq, |
1126 | NSEC_PER_SEC / scale, sec * scale); |
1127 | } |
1128 | |
1129 | /* |
1130 | * If the uncertainty margin is not specified, calculate it. |
1131 | * If both scale and freq are non-zero, calculate the clock |
1132 | * period, but bound below at 2*WATCHDOG_MAX_SKEW. However, |
1133 | * if either of scale or freq is zero, be very conservative and |
1134 | * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the |
1135 | * uncertainty margin. Allow stupidly small uncertainty margins |
1136 | * to be specified by the caller for testing purposes, but warn |
1137 | * to discourage production use of this capability. |
1138 | */ |
1139 | if (scale && freq && !cs->uncertainty_margin) { |
1140 | cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq); |
1141 | if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW) |
1142 | cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW; |
1143 | } else if (!cs->uncertainty_margin) { |
1144 | cs->uncertainty_margin = WATCHDOG_THRESHOLD; |
1145 | } |
1146 | WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW); |
1147 | |
1148 | /* |
1149 | * Ensure clocksources that have large 'mult' values don't overflow |
1150 | * when adjusted. |
1151 | */ |
1152 | cs->maxadj = clocksource_max_adjustment(cs); |
1153 | while (freq && ((cs->mult + cs->maxadj < cs->mult) |
1154 | || (cs->mult - cs->maxadj > cs->mult))) { |
1155 | cs->mult >>= 1; |
1156 | cs->shift--; |
1157 | cs->maxadj = clocksource_max_adjustment(cs); |
1158 | } |
1159 | |
1160 | /* |
1161 | * Only warn for *special* clocksources that self-define |
1162 | * their mult/shift values and don't specify a freq. |
1163 | */ |
1164 | WARN_ONCE(cs->mult + cs->maxadj < cs->mult, |
1165 | "timekeeping: Clocksource %s might overflow on 11%% adjustment\n" , |
1166 | cs->name); |
1167 | |
1168 | clocksource_update_max_deferment(cs); |
1169 | |
1170 | pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n" , |
1171 | cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns); |
1172 | } |
1173 | EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale); |
1174 | |
1175 | /** |
1176 | * __clocksource_register_scale - Used to install new clocksources |
1177 | * @cs: clocksource to be registered |
1178 | * @scale: Scale factor multiplied against freq to get clocksource hz |
1179 | * @freq: clocksource frequency (cycles per second) divided by scale |
1180 | * |
1181 | * Returns -EBUSY if registration fails, zero otherwise. |
1182 | * |
1183 | * This *SHOULD NOT* be called directly! Please use the |
1184 | * clocksource_register_hz() or clocksource_register_khz helper functions. |
1185 | */ |
1186 | int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq) |
1187 | { |
1188 | unsigned long flags; |
1189 | |
1190 | clocksource_arch_init(cs); |
1191 | |
1192 | if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX)) |
1193 | cs->id = CSID_GENERIC; |
1194 | if (cs->vdso_clock_mode < 0 || |
1195 | cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) { |
1196 | pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n" , |
1197 | cs->name, cs->vdso_clock_mode); |
1198 | cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE; |
1199 | } |
1200 | |
1201 | /* Initialize mult/shift and max_idle_ns */ |
1202 | __clocksource_update_freq_scale(cs, scale, freq); |
1203 | |
1204 | /* Add clocksource to the clocksource list */ |
1205 | mutex_lock(&clocksource_mutex); |
1206 | |
1207 | clocksource_watchdog_lock(flags: &flags); |
1208 | clocksource_enqueue(cs); |
1209 | clocksource_enqueue_watchdog(cs); |
1210 | clocksource_watchdog_unlock(flags: &flags); |
1211 | |
1212 | clocksource_select(); |
1213 | clocksource_select_watchdog(fallback: false); |
1214 | __clocksource_suspend_select(cs); |
1215 | mutex_unlock(lock: &clocksource_mutex); |
1216 | return 0; |
1217 | } |
1218 | EXPORT_SYMBOL_GPL(__clocksource_register_scale); |
1219 | |
1220 | static void __clocksource_change_rating(struct clocksource *cs, int rating) |
1221 | { |
1222 | list_del(entry: &cs->list); |
1223 | cs->rating = rating; |
1224 | clocksource_enqueue(cs); |
1225 | } |
1226 | |
1227 | /** |
1228 | * clocksource_change_rating - Change the rating of a registered clocksource |
1229 | * @cs: clocksource to be changed |
1230 | * @rating: new rating |
1231 | */ |
1232 | void clocksource_change_rating(struct clocksource *cs, int rating) |
1233 | { |
1234 | unsigned long flags; |
1235 | |
1236 | mutex_lock(&clocksource_mutex); |
1237 | clocksource_watchdog_lock(flags: &flags); |
1238 | __clocksource_change_rating(cs, rating); |
1239 | clocksource_watchdog_unlock(flags: &flags); |
1240 | |
1241 | clocksource_select(); |
1242 | clocksource_select_watchdog(fallback: false); |
1243 | clocksource_suspend_select(fallback: false); |
1244 | mutex_unlock(lock: &clocksource_mutex); |
1245 | } |
1246 | EXPORT_SYMBOL(clocksource_change_rating); |
1247 | |
1248 | /* |
1249 | * Unbind clocksource @cs. Called with clocksource_mutex held |
1250 | */ |
1251 | static int clocksource_unbind(struct clocksource *cs) |
1252 | { |
1253 | unsigned long flags; |
1254 | |
1255 | if (clocksource_is_watchdog(cs)) { |
1256 | /* Select and try to install a replacement watchdog. */ |
1257 | clocksource_select_watchdog(fallback: true); |
1258 | if (clocksource_is_watchdog(cs)) |
1259 | return -EBUSY; |
1260 | } |
1261 | |
1262 | if (cs == curr_clocksource) { |
1263 | /* Select and try to install a replacement clock source */ |
1264 | clocksource_select_fallback(); |
1265 | if (curr_clocksource == cs) |
1266 | return -EBUSY; |
1267 | } |
1268 | |
1269 | if (clocksource_is_suspend(cs)) { |
1270 | /* |
1271 | * Select and try to install a replacement suspend clocksource. |
1272 | * If no replacement suspend clocksource, we will just let the |
1273 | * clocksource go and have no suspend clocksource. |
1274 | */ |
1275 | clocksource_suspend_select(fallback: true); |
1276 | } |
1277 | |
1278 | clocksource_watchdog_lock(flags: &flags); |
1279 | clocksource_dequeue_watchdog(cs); |
1280 | list_del_init(entry: &cs->list); |
1281 | clocksource_watchdog_unlock(flags: &flags); |
1282 | |
1283 | return 0; |
1284 | } |
1285 | |
1286 | /** |
1287 | * clocksource_unregister - remove a registered clocksource |
1288 | * @cs: clocksource to be unregistered |
1289 | */ |
1290 | int clocksource_unregister(struct clocksource *cs) |
1291 | { |
1292 | int ret = 0; |
1293 | |
1294 | mutex_lock(&clocksource_mutex); |
1295 | if (!list_empty(head: &cs->list)) |
1296 | ret = clocksource_unbind(cs); |
1297 | mutex_unlock(lock: &clocksource_mutex); |
1298 | return ret; |
1299 | } |
1300 | EXPORT_SYMBOL(clocksource_unregister); |
1301 | |
1302 | #ifdef CONFIG_SYSFS |
1303 | /** |
1304 | * current_clocksource_show - sysfs interface for current clocksource |
1305 | * @dev: unused |
1306 | * @attr: unused |
1307 | * @buf: char buffer to be filled with clocksource list |
1308 | * |
1309 | * Provides sysfs interface for listing current clocksource. |
1310 | */ |
1311 | static ssize_t current_clocksource_show(struct device *dev, |
1312 | struct device_attribute *attr, |
1313 | char *buf) |
1314 | { |
1315 | ssize_t count = 0; |
1316 | |
1317 | mutex_lock(&clocksource_mutex); |
1318 | count = snprintf(buf, PAGE_SIZE, fmt: "%s\n" , curr_clocksource->name); |
1319 | mutex_unlock(lock: &clocksource_mutex); |
1320 | |
1321 | return count; |
1322 | } |
1323 | |
1324 | ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt) |
1325 | { |
1326 | size_t ret = cnt; |
1327 | |
1328 | /* strings from sysfs write are not 0 terminated! */ |
1329 | if (!cnt || cnt >= CS_NAME_LEN) |
1330 | return -EINVAL; |
1331 | |
1332 | /* strip of \n: */ |
1333 | if (buf[cnt-1] == '\n') |
1334 | cnt--; |
1335 | if (cnt > 0) |
1336 | memcpy(dst, buf, cnt); |
1337 | dst[cnt] = 0; |
1338 | return ret; |
1339 | } |
1340 | |
1341 | /** |
1342 | * current_clocksource_store - interface for manually overriding clocksource |
1343 | * @dev: unused |
1344 | * @attr: unused |
1345 | * @buf: name of override clocksource |
1346 | * @count: length of buffer |
1347 | * |
1348 | * Takes input from sysfs interface for manually overriding the default |
1349 | * clocksource selection. |
1350 | */ |
1351 | static ssize_t current_clocksource_store(struct device *dev, |
1352 | struct device_attribute *attr, |
1353 | const char *buf, size_t count) |
1354 | { |
1355 | ssize_t ret; |
1356 | |
1357 | mutex_lock(&clocksource_mutex); |
1358 | |
1359 | ret = sysfs_get_uname(buf, dst: override_name, cnt: count); |
1360 | if (ret >= 0) |
1361 | clocksource_select(); |
1362 | |
1363 | mutex_unlock(lock: &clocksource_mutex); |
1364 | |
1365 | return ret; |
1366 | } |
1367 | static DEVICE_ATTR_RW(current_clocksource); |
1368 | |
1369 | /** |
1370 | * unbind_clocksource_store - interface for manually unbinding clocksource |
1371 | * @dev: unused |
1372 | * @attr: unused |
1373 | * @buf: unused |
1374 | * @count: length of buffer |
1375 | * |
1376 | * Takes input from sysfs interface for manually unbinding a clocksource. |
1377 | */ |
1378 | static ssize_t unbind_clocksource_store(struct device *dev, |
1379 | struct device_attribute *attr, |
1380 | const char *buf, size_t count) |
1381 | { |
1382 | struct clocksource *cs; |
1383 | char name[CS_NAME_LEN]; |
1384 | ssize_t ret; |
1385 | |
1386 | ret = sysfs_get_uname(buf, dst: name, cnt: count); |
1387 | if (ret < 0) |
1388 | return ret; |
1389 | |
1390 | ret = -ENODEV; |
1391 | mutex_lock(&clocksource_mutex); |
1392 | list_for_each_entry(cs, &clocksource_list, list) { |
1393 | if (strcmp(cs->name, name)) |
1394 | continue; |
1395 | ret = clocksource_unbind(cs); |
1396 | break; |
1397 | } |
1398 | mutex_unlock(lock: &clocksource_mutex); |
1399 | |
1400 | return ret ? ret : count; |
1401 | } |
1402 | static DEVICE_ATTR_WO(unbind_clocksource); |
1403 | |
1404 | /** |
1405 | * available_clocksource_show - sysfs interface for listing clocksource |
1406 | * @dev: unused |
1407 | * @attr: unused |
1408 | * @buf: char buffer to be filled with clocksource list |
1409 | * |
1410 | * Provides sysfs interface for listing registered clocksources |
1411 | */ |
1412 | static ssize_t available_clocksource_show(struct device *dev, |
1413 | struct device_attribute *attr, |
1414 | char *buf) |
1415 | { |
1416 | struct clocksource *src; |
1417 | ssize_t count = 0; |
1418 | |
1419 | mutex_lock(&clocksource_mutex); |
1420 | list_for_each_entry(src, &clocksource_list, list) { |
1421 | /* |
1422 | * Don't show non-HRES clocksource if the tick code is |
1423 | * in one shot mode (highres=on or nohz=on) |
1424 | */ |
1425 | if (!tick_oneshot_mode_active() || |
1426 | (src->flags & CLOCK_SOURCE_VALID_FOR_HRES)) |
1427 | count += snprintf(buf: buf + count, |
1428 | max((ssize_t)PAGE_SIZE - count, (ssize_t)0), |
1429 | fmt: "%s " , src->name); |
1430 | } |
1431 | mutex_unlock(lock: &clocksource_mutex); |
1432 | |
1433 | count += snprintf(buf: buf + count, |
1434 | max((ssize_t)PAGE_SIZE - count, (ssize_t)0), fmt: "\n" ); |
1435 | |
1436 | return count; |
1437 | } |
1438 | static DEVICE_ATTR_RO(available_clocksource); |
1439 | |
1440 | static struct attribute *clocksource_attrs[] = { |
1441 | &dev_attr_current_clocksource.attr, |
1442 | &dev_attr_unbind_clocksource.attr, |
1443 | &dev_attr_available_clocksource.attr, |
1444 | NULL |
1445 | }; |
1446 | ATTRIBUTE_GROUPS(clocksource); |
1447 | |
1448 | static struct bus_type clocksource_subsys = { |
1449 | .name = "clocksource" , |
1450 | .dev_name = "clocksource" , |
1451 | }; |
1452 | |
1453 | static struct device device_clocksource = { |
1454 | .id = 0, |
1455 | .bus = &clocksource_subsys, |
1456 | .groups = clocksource_groups, |
1457 | }; |
1458 | |
1459 | static int __init init_clocksource_sysfs(void) |
1460 | { |
1461 | int error = subsys_system_register(subsys: &clocksource_subsys, NULL); |
1462 | |
1463 | if (!error) |
1464 | error = device_register(dev: &device_clocksource); |
1465 | |
1466 | return error; |
1467 | } |
1468 | |
1469 | device_initcall(init_clocksource_sysfs); |
1470 | #endif /* CONFIG_SYSFS */ |
1471 | |
1472 | /** |
1473 | * boot_override_clocksource - boot clock override |
1474 | * @str: override name |
1475 | * |
1476 | * Takes a clocksource= boot argument and uses it |
1477 | * as the clocksource override name. |
1478 | */ |
1479 | static int __init boot_override_clocksource(char* str) |
1480 | { |
1481 | mutex_lock(&clocksource_mutex); |
1482 | if (str) |
1483 | strscpy(p: override_name, q: str, size: sizeof(override_name)); |
1484 | mutex_unlock(lock: &clocksource_mutex); |
1485 | return 1; |
1486 | } |
1487 | |
1488 | __setup("clocksource=" , boot_override_clocksource); |
1489 | |
1490 | /** |
1491 | * boot_override_clock - Compatibility layer for deprecated boot option |
1492 | * @str: override name |
1493 | * |
1494 | * DEPRECATED! Takes a clock= boot argument and uses it |
1495 | * as the clocksource override name |
1496 | */ |
1497 | static int __init boot_override_clock(char* str) |
1498 | { |
1499 | if (!strcmp(str, "pmtmr" )) { |
1500 | pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n" ); |
1501 | return boot_override_clocksource(str: "acpi_pm" ); |
1502 | } |
1503 | pr_warn("clock= boot option is deprecated - use clocksource=xyz\n" ); |
1504 | return boot_override_clocksource(str); |
1505 | } |
1506 | |
1507 | __setup("clock=" , boot_override_clock); |
1508 | |