| 1 | |
| 2 | * linux/kernel/time/clocksource.c
|
| 3 | *
|
| 4 | * This file contains the functions which manage clocksource drivers.
|
| 5 | *
|
| 6 | * Copyright (C) 2004, 2005 IBM, John Stultz ([email protected])
|
| 7 | *
|
| 8 | * This program is free software; you can redistribute it and/or modify
|
| 9 | * it under the terms of the GNU General Public License as published by
|
| 10 | * the Free Software Foundation; either version 2 of the License, or
|
| 11 | * (at your option) any later version.
|
| 12 | *
|
| 13 | * This program is distributed in the hope that it will be useful,
|
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
| 16 | * GNU General Public License for more details.
|
| 17 | *
|
| 18 | * You should have received a copy of the GNU General Public License
|
| 19 | * along with this program; if not, write to the Free Software
|
| 20 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
| 21 | *
|
| 22 | * TODO WishList:
|
| 23 | * o Allow clocksource drivers to be unregistered
|
| 24 |
|
| 25 |
|
| 26 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
|
| 27 |
|
| 28 | #include <linux/device.h>
|
| 29 | #include <linux/clocksource.h>
|
| 30 | #include <linux/init.h>
|
| 31 | #include <linux/module.h>
|
| 32 | #include <linux/sched.h>
|
| 33 | #include <linux/tick.h>
|
| 34 | #include <linux/kthread.h>
|
| 35 |
|
| 36 | #include "tick-internal.h"
|
| 37 | #include "timekeeping_internal.h"
|
| 38 |
|
| 39 | |
| 40 | * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks
|
| 41 | * @mult: pointer to mult variable
|
| 42 | * @shift: pointer to shift variable
|
| 43 | * @from: frequency to convert from
|
| 44 | * @to: frequency to convert to
|
| 45 | * @maxsec: guaranteed runtime conversion range in seconds
|
| 46 | *
|
| 47 | * The function evaluates the shift/mult pair for the scaled math
|
| 48 | * operations of clocksources and clockevents.
|
| 49 | *
|
| 50 | * @to and @from are frequency values in HZ. For clock sources @to is
|
| 51 | * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock
|
| 52 | * event @to is the counter frequency and @from is NSEC_PER_SEC.
|
| 53 | *
|
| 54 | * The @maxsec conversion range argument controls the time frame in
|
| 55 | * seconds which must be covered by the runtime conversion with the
|
| 56 | * calculated mult and shift factors. This guarantees that no 64bit
|
| 57 | * overflow happens when the input value of the conversion is
|
| 58 | * multiplied with the calculated mult factor. Larger ranges may
|
| 59 | * reduce the conversion accuracy by chosing smaller mult and shift
|
| 60 | * factors.
|
| 61 |
|
| 62 | void
|
| 63 | clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec)
|
| 64 | {
|
| 65 | u64 tmp;
|
| 66 | u32 sft, sftacc= 32;
|
| 67 |
|
| 68 | |
| 69 | * Calculate the shift factor which is limiting the conversion
|
| 70 | * range:
|
| 71 |
|
| 72 | tmp = ((u64)maxsec * from) >> 32;
|
| 73 | while (tmp) {
|
| 74 | tmp >>=1;
|
| 75 | sftacc--;
|
| 76 | }
|
| 77 |
|
| 78 | |
| 79 | * Find the conversion shift/mult pair which has the best
|
| 80 | * accuracy and fits the maxsec conversion range:
|
| 81 |
|
| 82 | for (sft = 32; sft > 0; sft--) {
|
| 83 | tmp = (u64) to << sft;
|
| 84 | tmp += from / 2;
|
| 85 | do_div(tmp, from);
|
| 86 | if ((tmp >> sftacc) == 0)
|
| 87 | break;
|
| 88 | }
|
| 89 | *mult = tmp;
|
| 90 | *shift = sft;
|
| 91 | }
|
| 92 | EXPORT_SYMBOL_GPL(clocks_calc_mult_shift);
|
| 93 |
|
| 94 | |
| 95 | * curr_clocksource:
|
| 96 | * currently selected clocksource.
|
| 97 | * clocksource_list:
|
| 98 | * linked list with the registered clocksources
|
| 99 | * clocksource_mutex:
|
| 100 | * protects manipulations to curr_clocksource and the clocksource_list
|
| 101 | * override_name:
|
| 102 | * Name of the user-specified clocksource.
|
| 103 |
|
| 104 | static struct clocksource *curr_clocksource;
|
| 105 | static LIST_HEAD(clocksource_list);
|
| 106 | static DEFINE_MUTEX(clocksource_mutex);
|
| 107 | static char override_name[CS_NAME_LEN];
|
| 108 | static int finished_booting;
|
| 109 |
|
| 110 | #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
|
| 111 | static void clocksource_watchdog_work(struct work_struct *work);
|
| 112 | static void clocksource_select(void);
|
| 113 |
|
| 114 | static LIST_HEAD(watchdog_list);
|
| 115 | static struct clocksource *watchdog;
|
| 116 | static struct timer_list watchdog_timer;
|
| 117 | static DECLARE_WORK(watchdog_work, clocksource_watchdog_work);
|
| 118 | static DEFINE_SPINLOCK(watchdog_lock);
|
| 119 | static int watchdog_running;
|
| 120 | static atomic_t watchdog_reset_pending;
|
| 121 |
|
| 122 | static void inline clocksource_watchdog_lock(unsigned long *flags)
|
| 123 | {
|
| 124 | spin_lock_irqsave(&watchdog_lock, *flags);
|
| 125 | }
|
| 126 |
|
| 127 | static void inline clocksource_watchdog_unlock(unsigned long *flags)
|
| 128 | {
|
| 129 | spin_unlock_irqrestore(&watchdog_lock, *flags);
|
| 130 | }
|
| 131 |
|
| 132 | |
| 133 | * Interval: 0.5sec Threshold: 0.0625s
|
| 134 |
|
| 135 | #define WATCHDOG_INTERVAL (HZ >> 1)
|
| 136 | #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
|
| 137 |
|
| 138 | static void __clocksource_unstable(struct clocksource *cs)
|
| 139 | {
|
| 140 | cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
|
| 141 | cs->flags |= CLOCK_SOURCE_UNSTABLE;
|
| 142 |
|
| 143 | |
| 144 | * If the clocksource is registered clocksource_watchdog_work() will
|
| 145 | * re-rate and re-select.
|
| 146 |
|
| 147 | if (list_empty(&cs->list)) {
|
| 148 | cs->rating = 0;
|
| 149 | return;
|
| 150 | }
|
| 151 |
|
| 152 | if (cs->mark_unstable)
|
| 153 | cs->mark_unstable(cs);
|
| 154 |
|
| 155 |
|
| 156 | if (finished_booting)
|
| 157 | schedule_work(&watchdog_work);
|
| 158 | }
|
| 159 |
|
| 160 | |
| 161 | * clocksource_mark_unstable - mark clocksource unstable via watchdog
|
| 162 | * @cs: clocksource to be marked unstable
|
| 163 | *
|
| 164 | * This function is called by the x86 TSC code to mark clocksources as unstable;
|
| 165 | * it defers demotion and re-selection to a work.
|
| 166 |
|
| 167 | void clocksource_mark_unstable(struct clocksource *cs)
|
| 168 | {
|
| 169 | unsigned long flags;
|
| 170 |
|
| 171 | spin_lock_irqsave(&watchdog_lock, flags);
|
| 172 | if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) {
|
| 173 | if (!list_empty(&cs->list) && list_empty(&cs->wd_list))
|
| 174 | list_add(&cs->wd_list, &watchdog_list);
|
| 175 | __clocksource_unstable(cs);
|
| 176 | }
|
| 177 | spin_unlock_irqrestore(&watchdog_lock, flags);
|
| 178 | }
|
| 179 |
|
| 180 | static void clocksource_watchdog(struct timer_list *unused)
|
| 181 | {
|
| 182 | struct clocksource *cs;
|
| 183 | u64 csnow, wdnow, cslast, wdlast, delta;
|
| 184 | int64_t wd_nsec, cs_nsec;
|
| 185 | int next_cpu, reset_pending;
|
| 186 |
|
| 187 | spin_lock(&watchdog_lock);
|
| 188 | if (!watchdog_running)
|
| 189 | goto out;
|
| 190 |
|
| 191 | reset_pending = atomic_read(&watchdog_reset_pending);
|
| 192 |
|
| 193 | list_for_each_entry(cs, &watchdog_list, wd_list) {
|
| 194 |
|
| 195 |
|
| 196 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
|
| 197 | if (finished_booting)
|
| 198 | schedule_work(&watchdog_work);
|
| 199 | continue;
|
| 200 | }
|
| 201 |
|
| 202 | local_irq_disable();
|
| 203 | csnow = cs->read(cs);
|
| 204 | wdnow = watchdog->read(watchdog);
|
| 205 | local_irq_enable();
|
| 206 |
|
| 207 |
|
| 208 | if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) ||
|
| 209 | atomic_read(&watchdog_reset_pending)) {
|
| 210 | cs->flags |= CLOCK_SOURCE_WATCHDOG;
|
| 211 | cs->wd_last = wdnow;
|
| 212 | cs->cs_last = csnow;
|
| 213 | continue;
|
| 214 | }
|
| 215 |
|
| 216 | delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask);
|
| 217 | wd_nsec = clocksource_cyc2ns(delta, watchdog->mult,
|
| 218 | watchdog->shift);
|
| 219 |
|
| 220 | delta = clocksource_delta(csnow, cs->cs_last, cs->mask);
|
| 221 | cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift);
|
| 222 | wdlast = cs->wd_last;
|
| 223 | cslast = cs->cs_last;
|
| 224 | cs->cs_last = csnow;
|
| 225 | cs->wd_last = wdnow;
|
| 226 |
|
| 227 | if (atomic_read(&watchdog_reset_pending))
|
| 228 | continue;
|
| 229 |
|
| 230 |
|
| 231 | if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) {
|
| 232 | pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n",
|
| 233 | smp_processor_id(), cs->name);
|
| 234 | pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n",
|
| 235 | watchdog->name, wdnow, wdlast, watchdog->mask);
|
| 236 | pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n",
|
| 237 | cs->name, csnow, cslast, cs->mask);
|
| 238 | __clocksource_unstable(cs);
|
| 239 | continue;
|
| 240 | }
|
| 241 |
|
| 242 | if (cs == curr_clocksource && cs->tick_stable)
|
| 243 | cs->tick_stable(cs);
|
| 244 |
|
| 245 | if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) &&
|
| 246 | (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
|
| 247 | (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
|
| 248 |
|
| 249 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
|
| 250 |
|
| 251 | |
| 252 | * clocksource_done_booting() will sort it if
|
| 253 | * finished_booting is not set yet.
|
| 254 |
|
| 255 | if (!finished_booting)
|
| 256 | continue;
|
| 257 |
|
| 258 | |
| 259 | * If this is not the current clocksource let
|
| 260 | * the watchdog thread reselect it. Due to the
|
| 261 | * change to high res this clocksource might
|
| 262 | * be preferred now. If it is the current
|
| 263 | * clocksource let the tick code know about
|
| 264 | * that change.
|
| 265 |
|
| 266 | if (cs != curr_clocksource) {
|
| 267 | cs->flags |= CLOCK_SOURCE_RESELECT;
|
| 268 | schedule_work(&watchdog_work);
|
| 269 | } else {
|
| 270 | tick_clock_notify();
|
| 271 | }
|
| 272 | }
|
| 273 | }
|
| 274 |
|
| 275 | |
| 276 | * We only clear the watchdog_reset_pending, when we did a
|
| 277 | * full cycle through all clocksources.
|
| 278 |
|
| 279 | if (reset_pending)
|
| 280 | atomic_dec(&watchdog_reset_pending);
|
| 281 |
|
| 282 | |
| 283 | * Cycle through CPUs to check if the CPUs stay synchronized
|
| 284 | * to each other.
|
| 285 |
|
| 286 | next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
|
| 287 | if (next_cpu >= nr_cpu_ids)
|
| 288 | next_cpu = cpumask_first(cpu_online_mask);
|
| 289 | watchdog_timer.expires += WATCHDOG_INTERVAL;
|
| 290 | add_timer_on(&watchdog_timer, next_cpu);
|
| 291 | out:
|
| 292 | spin_unlock(&watchdog_lock);
|
| 293 | }
|
| 294 |
|
| 295 | static inline void clocksource_start_watchdog(void)
|
| 296 | {
|
| 297 | if (watchdog_running || !watchdog || list_empty(&watchdog_list))
|
| 298 | return;
|
| 299 | timer_setup(&watchdog_timer, clocksource_watchdog, 0);
|
| 300 | watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
|
| 301 | add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
|
| 302 | watchdog_running = 1;
|
| 303 | }
|
| 304 |
|
| 305 | static inline void clocksource_stop_watchdog(void)
|
| 306 | {
|
| 307 | if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
|
| 308 | return;
|
| 309 | del_timer(&watchdog_timer);
|
| 310 | watchdog_running = 0;
|
| 311 | }
|
| 312 |
|
| 313 | static inline void clocksource_reset_watchdog(void)
|
| 314 | {
|
| 315 | struct clocksource *cs;
|
| 316 |
|
| 317 | list_for_each_entry(cs, &watchdog_list, wd_list)
|
| 318 | cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
|
| 319 | }
|
| 320 |
|
| 321 | static void clocksource_resume_watchdog(void)
|
| 322 | {
|
| 323 | atomic_inc(&watchdog_reset_pending);
|
| 324 | }
|
| 325 |
|
| 326 | static void clocksource_enqueue_watchdog(struct clocksource *cs)
|
| 327 | {
|
| 328 | INIT_LIST_HEAD(&cs->wd_list);
|
| 329 |
|
| 330 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
|
| 331 |
|
| 332 | list_add(&cs->wd_list, &watchdog_list);
|
| 333 | cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
|
| 334 | } else {
|
| 335 |
|
| 336 | if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
|
| 337 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
|
| 338 | }
|
| 339 | }
|
| 340 |
|
| 341 | static void clocksource_select_watchdog(bool fallback)
|
| 342 | {
|
| 343 | struct clocksource *cs, *old_wd;
|
| 344 | unsigned long flags;
|
| 345 |
|
| 346 | spin_lock_irqsave(&watchdog_lock, flags);
|
| 347 |
|
| 348 | old_wd = watchdog;
|
| 349 | if (fallback)
|
| 350 | watchdog = NULL;
|
| 351 |
|
| 352 | list_for_each_entry(cs, &clocksource_list, list) {
|
| 353 |
|
| 354 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY)
|
| 355 | continue;
|
| 356 |
|
| 357 |
|
| 358 | if (fallback && cs == old_wd)
|
| 359 | continue;
|
| 360 |
|
| 361 |
|
| 362 | if (!watchdog || cs->rating > watchdog->rating)
|
| 363 | watchdog = cs;
|
| 364 | }
|
| 365 |
|
| 366 | if (!watchdog)
|
| 367 | watchdog = old_wd;
|
| 368 |
|
| 369 |
|
| 370 | if (watchdog != old_wd)
|
| 371 | clocksource_reset_watchdog();
|
| 372 |
|
| 373 |
|
| 374 | clocksource_start_watchdog();
|
| 375 | spin_unlock_irqrestore(&watchdog_lock, flags);
|
| 376 | }
|
| 377 |
|
| 378 | static void clocksource_dequeue_watchdog(struct clocksource *cs)
|
| 379 | {
|
| 380 | if (cs != watchdog) {
|
| 381 | if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
|
| 382 |
|
| 383 | list_del_init(&cs->wd_list);
|
| 384 |
|
| 385 | clocksource_stop_watchdog();
|
| 386 | }
|
| 387 | }
|
| 388 | }
|
| 389 |
|
| 390 | static void __clocksource_change_rating(struct clocksource *cs, int rating);
|
| 391 |
|
| 392 | static int __clocksource_watchdog_work(void)
|
| 393 | {
|
| 394 | struct clocksource *cs, *tmp;
|
| 395 | unsigned long flags;
|
| 396 | int select = 0;
|
| 397 |
|
| 398 | spin_lock_irqsave(&watchdog_lock, flags);
|
| 399 | list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
|
| 400 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
|
| 401 | list_del_init(&cs->wd_list);
|
| 402 | __clocksource_change_rating(cs, 0);
|
| 403 | select = 1;
|
| 404 | }
|
| 405 | if (cs->flags & CLOCK_SOURCE_RESELECT) {
|
| 406 | cs->flags &= ~CLOCK_SOURCE_RESELECT;
|
| 407 | select = 1;
|
| 408 | }
|
| 409 | }
|
| 410 |
|
| 411 | clocksource_stop_watchdog();
|
| 412 | spin_unlock_irqrestore(&watchdog_lock, flags);
|
| 413 |
|
| 414 | return select;
|
| 415 | }
|
| 416 |
|
| 417 | static void clocksource_watchdog_work(struct work_struct *work)
|
| 418 | {
|
| 419 | mutex_lock(&clocksource_mutex);
|
| 420 | if (__clocksource_watchdog_work())
|
| 421 | clocksource_select();
|
| 422 | mutex_unlock(&clocksource_mutex);
|
| 423 | }
|
| 424 |
|
| 425 | static bool clocksource_is_watchdog(struct clocksource *cs)
|
| 426 | {
|
| 427 | return cs == watchdog;
|
| 428 | }
|
| 429 |
|
| 430 | #else
|
| 431 |
|
| 432 | static void clocksource_enqueue_watchdog(struct clocksource *cs)
|
| 433 | {
|
| 434 | if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
|
| 435 | cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
|
| 436 | }
|
| 437 |
|
| 438 | static void clocksource_select_watchdog(bool fallback) { }
|
| 439 | static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { }
|
| 440 | static inline void clocksource_resume_watchdog(void) { }
|
| 441 | static inline int __clocksource_watchdog_work(void) { return 0; }
|
| 442 | static bool clocksource_is_watchdog(struct clocksource *cs) { return false; }
|
| 443 | void clocksource_mark_unstable(struct clocksource *cs) { }
|
| 444 |
|
| 445 | static inline void clocksource_watchdog_lock(unsigned long *flags) { }
|
| 446 | static inline void clocksource_watchdog_unlock(unsigned long *flags) { }
|
| 447 |
|
| 448 | #endif
|
| 449 |
|
| 450 | |
| 451 | * clocksource_suspend - suspend the clocksource(s)
|
| 452 |
|
| 453 | void clocksource_suspend(void)
|
| 454 | {
|
| 455 | struct clocksource *cs;
|
| 456 |
|
| 457 | list_for_each_entry_reverse(cs, &clocksource_list, list)
|
| 458 | if (cs->suspend)
|
| 459 | cs->suspend(cs);
|
| 460 | }
|
| 461 |
|
| 462 | |
| 463 | * clocksource_resume - resume the clocksource(s)
|
| 464 |
|
| 465 | void clocksource_resume(void)
|
| 466 | {
|
| 467 | struct clocksource *cs;
|
| 468 |
|
| 469 | list_for_each_entry(cs, &clocksource_list, list)
|
| 470 | if (cs->resume)
|
| 471 | cs->resume(cs);
|
| 472 |
|
| 473 | clocksource_resume_watchdog();
|
| 474 | }
|
| 475 |
|
| 476 | |
| 477 | * clocksource_touch_watchdog - Update watchdog
|
| 478 | *
|
| 479 | * Update the watchdog after exception contexts such as kgdb so as not
|
| 480 | * to incorrectly trip the watchdog. This might fail when the kernel
|
| 481 | * was stopped in code which holds watchdog_lock.
|
| 482 |
|
| 483 | void clocksource_touch_watchdog(void)
|
| 484 | {
|
| 485 | clocksource_resume_watchdog();
|
| 486 | }
|
| 487 |
|
| 488 | |
| 489 | * clocksource_max_adjustment- Returns max adjustment amount
|
| 490 | * @cs: Pointer to clocksource
|
| 491 | *
|
| 492 |
|
| 493 | static u32 clocksource_max_adjustment(struct clocksource *cs)
|
| 494 | {
|
| 495 | u64 ret;
|
| 496 | |
| 497 | * We won't try to correct for more than 11% adjustments (110,000 ppm),
|
| 498 |
|
| 499 | ret = (u64)cs->mult * 11;
|
| 500 | do_div(ret,100);
|
| 501 | return (u32)ret;
|
| 502 | }
|
| 503 |
|
| 504 | |
| 505 | * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted
|
| 506 | * @mult: cycle to nanosecond multiplier
|
| 507 | * @shift: cycle to nanosecond divisor (power of two)
|
| 508 | * @maxadj: maximum adjustment value to mult (~11%)
|
| 509 | * @mask: bitmask for two's complement subtraction of non 64 bit counters
|
| 510 | * @max_cyc: maximum cycle value before potential overflow (does not include
|
| 511 | * any safety margin)
|
| 512 | *
|
| 513 | * NOTE: This function includes a safety margin of 50%, in other words, we
|
| 514 | * return half the number of nanoseconds the hardware counter can technically
|
| 515 | * cover. This is done so that we can potentially detect problems caused by
|
| 516 | * delayed timers or bad hardware, which might result in time intervals that
|
| 517 | * are larger than what the math used can handle without overflows.
|
| 518 |
|
| 519 | u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc)
|
| 520 | {
|
| 521 | u64 max_nsecs, max_cycles;
|
| 522 |
|
| 523 | |
| 524 | * Calculate the maximum number of cycles that we can pass to the
|
| 525 | * cyc2ns() function without overflowing a 64-bit result.
|
| 526 |
|
| 527 | max_cycles = ULLONG_MAX;
|
| 528 | do_div(max_cycles, mult+maxadj);
|
| 529 |
|
| 530 | |
| 531 | * The actual maximum number of cycles we can defer the clocksource is
|
| 532 | * determined by the minimum of max_cycles and mask.
|
| 533 | * Note: Here we subtract the maxadj to make sure we don't sleep for
|
| 534 | * too long if there's a large negative adjustment.
|
| 535 |
|
| 536 | max_cycles = min(max_cycles, mask);
|
| 537 | max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift);
|
| 538 |
|
| 539 |
|
| 540 | if (max_cyc)
|
| 541 | *max_cyc = max_cycles;
|
| 542 |
|
| 543 |
|
| 544 | max_nsecs >>= 1;
|
| 545 |
|
| 546 | return max_nsecs;
|
| 547 | }
|
| 548 |
|
| 549 | |
| 550 | * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles
|
| 551 | * @cs: Pointer to clocksource to be updated
|
| 552 | *
|
| 553 |
|
| 554 | static inline void clocksource_update_max_deferment(struct clocksource *cs)
|
| 555 | {
|
| 556 | cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift,
|
| 557 | cs->maxadj, cs->mask,
|
| 558 | &cs->max_cycles);
|
| 559 | }
|
| 560 |
|
| 561 | #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET
|
| 562 |
|
| 563 | static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur)
|
| 564 | {
|
| 565 | struct clocksource *cs;
|
| 566 |
|
| 567 | if (!finished_booting || list_empty(&clocksource_list))
|
| 568 | return NULL;
|
| 569 |
|
| 570 | |
| 571 | * We pick the clocksource with the highest rating. If oneshot
|
| 572 | * mode is active, we pick the highres valid clocksource with
|
| 573 | * the best rating.
|
| 574 |
|
| 575 | list_for_each_entry(cs, &clocksource_list, list) {
|
| 576 | if (skipcur && cs == curr_clocksource)
|
| 577 | continue;
|
| 578 | if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES))
|
| 579 | continue;
|
| 580 | return cs;
|
| 581 | }
|
| 582 | return NULL;
|
| 583 | }
|
| 584 |
|
| 585 | static void __clocksource_select(bool skipcur)
|
| 586 | {
|
| 587 | bool oneshot = tick_oneshot_mode_active();
|
| 588 | struct clocksource *best, *cs;
|
| 589 |
|
| 590 |
|
| 591 | best = clocksource_find_best(oneshot, skipcur);
|
| 592 | if (!best)
|
| 593 | return;
|
| 594 |
|
| 595 | if (!strlen(override_name))
|
| 596 | goto found;
|
| 597 |
|
| 598 |
|
| 599 | list_for_each_entry(cs, &clocksource_list, list) {
|
| 600 | if (skipcur && cs == curr_clocksource)
|
| 601 | continue;
|
| 602 | if (strcmp(cs->name, override_name) != 0)
|
| 603 | continue;
|
| 604 | |
| 605 | * Check to make sure we don't switch to a non-highres
|
| 606 | * capable clocksource if the tick code is in oneshot
|
| 607 | * mode (highres or nohz)
|
| 608 |
|
| 609 | if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) {
|
| 610 |
|
| 611 | if (cs->flags & CLOCK_SOURCE_UNSTABLE) {
|
| 612 | pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n",
|
| 613 | cs->name);
|
| 614 | override_name[0] = 0;
|
| 615 | } else {
|
| 616 | |
| 617 | * The override cannot be currently verified.
|
| 618 | * Deferring to let the watchdog check.
|
| 619 |
|
| 620 | pr_info("Override clocksource %s is not currently HRT compatible - deferring\n",
|
| 621 | cs->name);
|
| 622 | }
|
| 623 | } else
|
| 624 |
|
| 625 | best = cs;
|
| 626 | break;
|
| 627 | }
|
| 628 |
|
| 629 | found:
|
| 630 | if (curr_clocksource != best && !timekeeping_notify(best)) {
|
| 631 | pr_info("Switched to clocksource %s\n", best->name);
|
| 632 | curr_clocksource = best;
|
| 633 | }
|
| 634 | }
|
| 635 |
|
| 636 | |
| 637 | * clocksource_select - Select the best clocksource available
|
| 638 | *
|
| 639 | * Private function. Must hold clocksource_mutex when called.
|
| 640 | *
|
| 641 | * Select the clocksource with the best rating, or the clocksource,
|
| 642 | * which is selected by userspace override.
|
| 643 |
|
| 644 | static void clocksource_select(void)
|
| 645 | {
|
| 646 | __clocksource_select(false);
|
| 647 | }
|
| 648 |
|
| 649 | static void clocksource_select_fallback(void)
|
| 650 | {
|
| 651 | __clocksource_select(true);
|
| 652 | }
|
| 653 |
|
| 654 | #else
|
| 655 | static inline void clocksource_select(void) { }
|
| 656 | static inline void clocksource_select_fallback(void) { }
|
| 657 |
|
| 658 | #endif
|
| 659 |
|
| 660 | |
| 661 | * clocksource_done_booting - Called near the end of core bootup
|
| 662 | *
|
| 663 | * Hack to avoid lots of clocksource churn at boot time.
|
| 664 | * We use fs_initcall because we want this to start before
|
| 665 | * device_initcall but after subsys_initcall.
|
| 666 |
|
| 667 | static int __init clocksource_done_booting(void)
|
| 668 | {
|
| 669 | mutex_lock(&clocksource_mutex);
|
| 670 | curr_clocksource = clocksource_default_clock();
|
| 671 | finished_booting = 1;
|
| 672 | |
| 673 | * Run the watchdog first to eliminate unstable clock sources
|
| 674 |
|
| 675 | __clocksource_watchdog_work();
|
| 676 | clocksource_select();
|
| 677 | mutex_unlock(&clocksource_mutex);
|
| 678 | return 0;
|
| 679 | }
|
| 680 | fs_initcall(clocksource_done_booting);
|
| 681 |
|
| 682 | |
| 683 | * Enqueue the clocksource sorted by rating
|
| 684 |
|
| 685 | static void clocksource_enqueue(struct clocksource *cs)
|
| 686 | {
|
| 687 | struct list_head *entry = &clocksource_list;
|
| 688 | struct clocksource *tmp;
|
| 689 |
|
| 690 | list_for_each_entry(tmp, &clocksource_list, list) {
|
| 691 |
|
| 692 | if (tmp->rating < cs->rating)
|
| 693 | break;
|
| 694 | entry = &tmp->list;
|
| 695 | }
|
| 696 | list_add(&cs->list, entry);
|
| 697 | }
|
| 698 |
|
| 699 | |
| 700 | * __clocksource_update_freq_scale - Used update clocksource with new freq
|
| 701 | * @cs: clocksource to be registered
|
| 702 | * @scale: Scale factor multiplied against freq to get clocksource hz
|
| 703 | * @freq: clocksource frequency (cycles per second) divided by scale
|
| 704 | *
|
| 705 | * This should only be called from the clocksource->enable() method.
|
| 706 | *
|
| 707 | * This *SHOULD NOT* be called directly! Please use the
|
| 708 | * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper
|
| 709 | * functions.
|
| 710 |
|
| 711 | void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq)
|
| 712 | {
|
| 713 | u64 sec;
|
| 714 |
|
| 715 | |
| 716 | * Default clocksources are *special* and self-define their mult/shift.
|
| 717 | * But, you're not special, so you should specify a freq value.
|
| 718 |
|
| 719 | if (freq) {
|
| 720 | |
| 721 | * Calc the maximum number of seconds which we can run before
|
| 722 | * wrapping around. For clocksources which have a mask > 32-bit
|
| 723 | * we need to limit the max sleep time to have a good
|
| 724 | * conversion precision. 10 minutes is still a reasonable
|
| 725 | * amount. That results in a shift value of 24 for a
|
| 726 | * clocksource with mask >= 40-bit and f >= 4GHz. That maps to
|
| 727 | * ~ 0.06ppm granularity for NTP.
|
| 728 |
|
| 729 | sec = cs->mask;
|
| 730 | do_div(sec, freq);
|
| 731 | do_div(sec, scale);
|
| 732 | if (!sec)
|
| 733 | sec = 1;
|
| 734 | else if (sec > 600 && cs->mask > UINT_MAX)
|
| 735 | sec = 600;
|
| 736 |
|
| 737 | clocks_calc_mult_shift(&cs->mult, &cs->shift, freq,
|
| 738 | NSEC_PER_SEC / scale, sec * scale);
|
| 739 | }
|
| 740 | |
| 741 | * Ensure clocksources that have large 'mult' values don't overflow
|
| 742 | * when adjusted.
|
| 743 |
|
| 744 | cs->maxadj = clocksource_max_adjustment(cs);
|
| 745 | while (freq && ((cs->mult + cs->maxadj < cs->mult)
|
| 746 | || (cs->mult - cs->maxadj > cs->mult))) {
|
| 747 | cs->mult >>= 1;
|
| 748 | cs->shift--;
|
| 749 | cs->maxadj = clocksource_max_adjustment(cs);
|
| 750 | }
|
| 751 |
|
| 752 | |
| 753 | * Only warn for *special* clocksources that self-define
|
| 754 | * their mult/shift values and don't specify a freq.
|
| 755 |
|
| 756 | WARN_ONCE(cs->mult + cs->maxadj < cs->mult,
|
| 757 | "timekeeping: Clocksource %s might overflow on 11%% adjustment\n",
|
| 758 | cs->name);
|
| 759 |
|
| 760 | clocksource_update_max_deferment(cs);
|
| 761 |
|
| 762 | pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n",
|
| 763 | cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns);
|
| 764 | }
|
| 765 | EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale);
|
| 766 |
|
| 767 | |
| 768 | * __clocksource_register_scale - Used to install new clocksources
|
| 769 | * @cs: clocksource to be registered
|
| 770 | * @scale: Scale factor multiplied against freq to get clocksource hz
|
| 771 | * @freq: clocksource frequency (cycles per second) divided by scale
|
| 772 | *
|
| 773 | * Returns -EBUSY if registration fails, zero otherwise.
|
| 774 | *
|
| 775 | * This *SHOULD NOT* be called directly! Please use the
|
| 776 | * clocksource_register_hz() or clocksource_register_khz helper functions.
|
| 777 |
|
| 778 | int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq)
|
| 779 | {
|
| 780 | unsigned long flags;
|
| 781 |
|
| 782 |
|
| 783 | __clocksource_update_freq_scale(cs, scale, freq);
|
| 784 |
|
| 785 |
|
| 786 | mutex_lock(&clocksource_mutex);
|
| 787 |
|
| 788 | clocksource_watchdog_lock(&flags);
|
| 789 | clocksource_enqueue(cs);
|
| 790 | clocksource_enqueue_watchdog(cs);
|
| 791 | clocksource_watchdog_unlock(&flags);
|
| 792 |
|
| 793 | clocksource_select();
|
| 794 | clocksource_select_watchdog(false);
|
| 795 | mutex_unlock(&clocksource_mutex);
|
| 796 | return 0;
|
| 797 | }
|
| 798 | EXPORT_SYMBOL_GPL(__clocksource_register_scale);
|
| 799 |
|
| 800 | static void __clocksource_change_rating(struct clocksource *cs, int rating)
|
| 801 | {
|
| 802 | list_del(&cs->list);
|
| 803 | cs->rating = rating;
|
| 804 | clocksource_enqueue(cs);
|
| 805 | }
|
| 806 |
|
| 807 | |
| 808 | * clocksource_change_rating - Change the rating of a registered clocksource
|
| 809 | * @cs: clocksource to be changed
|
| 810 | * @rating: new rating
|
| 811 |
|
| 812 | void clocksource_change_rating(struct clocksource *cs, int rating)
|
| 813 | {
|
| 814 | unsigned long flags;
|
| 815 |
|
| 816 | mutex_lock(&clocksource_mutex);
|
| 817 | clocksource_watchdog_lock(&flags);
|
| 818 | __clocksource_change_rating(cs, rating);
|
| 819 | clocksource_watchdog_unlock(&flags);
|
| 820 |
|
| 821 | clocksource_select();
|
| 822 | clocksource_select_watchdog(false);
|
| 823 | mutex_unlock(&clocksource_mutex);
|
| 824 | }
|
| 825 | EXPORT_SYMBOL(clocksource_change_rating);
|
| 826 |
|
| 827 | |
| 828 | * Unbind clocksource @cs. Called with clocksource_mutex held
|
| 829 |
|
| 830 | static int clocksource_unbind(struct clocksource *cs)
|
| 831 | {
|
| 832 | unsigned long flags;
|
| 833 |
|
| 834 | if (clocksource_is_watchdog(cs)) {
|
| 835 |
|
| 836 | clocksource_select_watchdog(true);
|
| 837 | if (clocksource_is_watchdog(cs))
|
| 838 | return -EBUSY;
|
| 839 | }
|
| 840 |
|
| 841 | if (cs == curr_clocksource) {
|
| 842 |
|
| 843 | clocksource_select_fallback();
|
| 844 | if (curr_clocksource == cs)
|
| 845 | return -EBUSY;
|
| 846 | }
|
| 847 |
|
| 848 | clocksource_watchdog_lock(&flags);
|
| 849 | clocksource_dequeue_watchdog(cs);
|
| 850 | list_del_init(&cs->list);
|
| 851 | clocksource_watchdog_unlock(&flags);
|
| 852 |
|
| 853 | return 0;
|
| 854 | }
|
| 855 |
|
| 856 | |
| 857 | * clocksource_unregister - remove a registered clocksource
|
| 858 | * @cs: clocksource to be unregistered
|
| 859 |
|
| 860 | int clocksource_unregister(struct clocksource *cs)
|
| 861 | {
|
| 862 | int ret = 0;
|
| 863 |
|
| 864 | mutex_lock(&clocksource_mutex);
|
| 865 | if (!list_empty(&cs->list))
|
| 866 | ret = clocksource_unbind(cs);
|
| 867 | mutex_unlock(&clocksource_mutex);
|
| 868 | return ret;
|
| 869 | }
|
| 870 | EXPORT_SYMBOL(clocksource_unregister);
|
| 871 |
|
| 872 | #ifdef CONFIG_SYSFS
|
| 873 | |
| 874 | * current_clocksource_show - sysfs interface for current clocksource
|
| 875 | * @dev: unused
|
| 876 | * @attr: unused
|
| 877 | * @buf: char buffer to be filled with clocksource list
|
| 878 | *
|
| 879 | * Provides sysfs interface for listing current clocksource.
|
| 880 |
|
| 881 | static ssize_t current_clocksource_show(struct device *dev,
|
| 882 | struct device_attribute *attr,
|
| 883 | char *buf)
|
| 884 | {
|
| 885 | ssize_t count = 0;
|
| 886 |
|
| 887 | mutex_lock(&clocksource_mutex);
|
| 888 | count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
|
| 889 | mutex_unlock(&clocksource_mutex);
|
| 890 |
|
| 891 | return count;
|
| 892 | }
|
| 893 |
|
| 894 | ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt)
|
| 895 | {
|
| 896 | size_t ret = cnt;
|
| 897 |
|
| 898 |
|
| 899 | if (!cnt || cnt >= CS_NAME_LEN)
|
| 900 | return -EINVAL;
|
| 901 |
|
| 902 |
|
| 903 | if (buf[cnt-1] == '\n')
|
| 904 | cnt--;
|
| 905 | if (cnt > 0)
|
| 906 | memcpy(dst, buf, cnt);
|
| 907 | dst[cnt] = 0;
|
| 908 | return ret;
|
| 909 | }
|
| 910 |
|
| 911 | |
| 912 | * current_clocksource_store - interface for manually overriding clocksource
|
| 913 | * @dev: unused
|
| 914 | * @attr: unused
|
| 915 | * @buf: name of override clocksource
|
| 916 | * @count: length of buffer
|
| 917 | *
|
| 918 | * Takes input from sysfs interface for manually overriding the default
|
| 919 | * clocksource selection.
|
| 920 |
|
| 921 | static ssize_t current_clocksource_store(struct device *dev,
|
| 922 | struct device_attribute *attr,
|
| 923 | const char *buf, size_t count)
|
| 924 | {
|
| 925 | ssize_t ret;
|
| 926 |
|
| 927 | mutex_lock(&clocksource_mutex);
|
| 928 |
|
| 929 | ret = sysfs_get_uname(buf, override_name, count);
|
| 930 | if (ret >= 0)
|
| 931 | clocksource_select();
|
| 932 |
|
| 933 | mutex_unlock(&clocksource_mutex);
|
| 934 |
|
| 935 | return ret;
|
| 936 | }
|
| 937 | static DEVICE_ATTR_RW(current_clocksource);
|
| 938 |
|
| 939 | |
| 940 | * unbind_clocksource_store - interface for manually unbinding clocksource
|
| 941 | * @dev: unused
|
| 942 | * @attr: unused
|
| 943 | * @buf: unused
|
| 944 | * @count: length of buffer
|
| 945 | *
|
| 946 | * Takes input from sysfs interface for manually unbinding a clocksource.
|
| 947 |
|
| 948 | static ssize_t unbind_clocksource_store(struct device *dev,
|
| 949 | struct device_attribute *attr,
|
| 950 | const char *buf, size_t count)
|
| 951 | {
|
| 952 | struct clocksource *cs;
|
| 953 | char name[CS_NAME_LEN];
|
| 954 | ssize_t ret;
|
| 955 |
|
| 956 | ret = sysfs_get_uname(buf, name, count);
|
| 957 | if (ret < 0)
|
| 958 | return ret;
|
| 959 |
|
| 960 | ret = -ENODEV;
|
| 961 | mutex_lock(&clocksource_mutex);
|
| 962 | list_for_each_entry(cs, &clocksource_list, list) {
|
| 963 | if (strcmp(cs->name, name))
|
| 964 | continue;
|
| 965 | ret = clocksource_unbind(cs);
|
| 966 | break;
|
| 967 | }
|
| 968 | mutex_unlock(&clocksource_mutex);
|
| 969 |
|
| 970 | return ret ? ret : count;
|
| 971 | }
|
| 972 | static DEVICE_ATTR_WO(unbind_clocksource);
|
| 973 |
|
| 974 | |
| 975 | * available_clocksource_show - sysfs interface for listing clocksource
|
| 976 | * @dev: unused
|
| 977 | * @attr: unused
|
| 978 | * @buf: char buffer to be filled with clocksource list
|
| 979 | *
|
| 980 | * Provides sysfs interface for listing registered clocksources
|
| 981 |
|
| 982 | static ssize_t available_clocksource_show(struct device *dev,
|
| 983 | struct device_attribute *attr,
|
| 984 | char *buf)
|
| 985 | {
|
| 986 | struct clocksource *src;
|
| 987 | ssize_t count = 0;
|
| 988 |
|
| 989 | mutex_lock(&clocksource_mutex);
|
| 990 | list_for_each_entry(src, &clocksource_list, list) {
|
| 991 | |
| 992 | * Don't show non-HRES clocksource if the tick code is
|
| 993 | * in one shot mode (highres=on or nohz=on)
|
| 994 |
|
| 995 | if (!tick_oneshot_mode_active() ||
|
| 996 | (src->flags & CLOCK_SOURCE_VALID_FOR_HRES))
|
| 997 | count += snprintf(buf + count,
|
| 998 | max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
|
| 999 | "%s ", src->name);
|
| 1000 | }
|
| 1001 | mutex_unlock(&clocksource_mutex);
|
| 1002 |
|
| 1003 | count += snprintf(buf + count,
|
| 1004 | max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
|
| 1005 |
|
| 1006 | return count;
|
| 1007 | }
|
| 1008 | static DEVICE_ATTR_RO(available_clocksource);
|
| 1009 |
|
| 1010 | static struct attribute *clocksource_attrs[] = {
|
| 1011 | &dev_attr_current_clocksource.attr,
|
| 1012 | &dev_attr_unbind_clocksource.attr,
|
| 1013 | &dev_attr_available_clocksource.attr,
|
| 1014 | NULL
|
| 1015 | };
|
| 1016 | ATTRIBUTE_GROUPS(clocksource);
|
| 1017 |
|
| 1018 | static struct bus_type clocksource_subsys = {
|
| 1019 | .name = "clocksource",
|
| 1020 | .dev_name = "clocksource",
|
| 1021 | };
|
| 1022 |
|
| 1023 | static struct device device_clocksource = {
|
| 1024 | .id = 0,
|
| 1025 | .bus = &clocksource_subsys,
|
| 1026 | .groups = clocksource_groups,
|
| 1027 | };
|
| 1028 |
|
| 1029 | static int __init init_clocksource_sysfs(void)
|
| 1030 | {
|
| 1031 | int error = subsys_system_register(&clocksource_subsys, NULL);
|
| 1032 |
|
| 1033 | if (!error)
|
| 1034 | error = device_register(&device_clocksource);
|
| 1035 |
|
| 1036 | return error;
|
| 1037 | }
|
| 1038 |
|
| 1039 | device_initcall(init_clocksource_sysfs);
|
| 1040 | #endif
|
| 1041 |
|
| 1042 | |
| 1043 | * boot_override_clocksource - boot clock override
|
| 1044 | * @str: override name
|
| 1045 | *
|
| 1046 | * Takes a clocksource= boot argument and uses it
|
| 1047 | * as the clocksource override name.
|
| 1048 |
|
| 1049 | static int __init boot_override_clocksource(char* str)
|
| 1050 | {
|
| 1051 | mutex_lock(&clocksource_mutex);
|
| 1052 | if (str)
|
| 1053 | strlcpy(override_name, str, sizeof(override_name));
|
| 1054 | mutex_unlock(&clocksource_mutex);
|
| 1055 | return 1;
|
| 1056 | }
|
| 1057 |
|
| 1058 | __setup("clocksource=", boot_override_clocksource);
|
| 1059 |
|
| 1060 | |
| 1061 | * boot_override_clock - Compatibility layer for deprecated boot option
|
| 1062 | * @str: override name
|
| 1063 | *
|
| 1064 | * DEPRECATED! Takes a clock= boot argument and uses it
|
| 1065 | * as the clocksource override name
|
| 1066 |
|
| 1067 | static int __init boot_override_clock(char* str)
|
| 1068 | {
|
| 1069 | if (!strcmp(str, "pmtmr")) {
|
| 1070 | pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n");
|
| 1071 | return boot_override_clocksource("acpi_pm");
|
| 1072 | }
|
| 1073 | pr_warn("clock= boot option is deprecated - use clocksource=xyz\n");
|
| 1074 | return boot_override_clocksource(str);
|
| 1075 | }
|
| 1076 |
|
| 1077 | __setup("clock=", boot_override_clock);
|