| 1 | |
| 2 | * linux/kernel/time/timekeeping.c
|
| 3 | *
|
| 4 | * Kernel timekeeping code and accessor functions
|
| 5 | *
|
| 6 | * This code was moved from linux/kernel/timer.c.
|
| 7 | * Please see that file for copyright and history logs.
|
| 8 | *
|
| 9 |
|
| 10 |
|
| 11 | #include <linux/timekeeper_internal.h>
|
| 12 | #include <linux/module.h>
|
| 13 | #include <linux/interrupt.h>
|
| 14 | #include <linux/percpu.h>
|
| 15 | #include <linux/init.h>
|
| 16 | #include <linux/mm.h>
|
| 17 | #include <linux/sched.h>
|
| 18 | #include <linux/syscore_ops.h>
|
| 19 | #include <linux/clocksource.h>
|
| 20 | #include <linux/jiffies.h>
|
| 21 | #include <linux/time.h>
|
| 22 | #include <linux/tick.h>
|
| 23 | #include <linux/stop_machine.h>
|
| 24 | #include <linux/pvclock_gtod.h>
|
| 25 |
|
| 26 |
|
| 27 | static struct timekeeper timekeeper;
|
| 28 |
|
| 29 |
|
| 30 | int __read_mostly timekeeping_suspended;
|
| 31 |
|
| 32 | static inline void tk_normalize_xtime(struct timekeeper *tk)
|
| 33 | {
|
| 34 | while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
|
| 35 | tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
|
| 36 | tk->xtime_sec++;
|
| 37 | }
|
| 38 | }
|
| 39 |
|
| 40 | static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
|
| 41 | {
|
| 42 | tk->xtime_sec = ts->tv_sec;
|
| 43 | tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
|
| 44 | }
|
| 45 |
|
| 46 | static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
|
| 47 | {
|
| 48 | tk->xtime_sec += ts->tv_sec;
|
| 49 | tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
|
| 50 | tk_normalize_xtime(tk);
|
| 51 | }
|
| 52 |
|
| 53 | static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
|
| 54 | {
|
| 55 | struct timespec tmp;
|
| 56 |
|
| 57 | |
| 58 | * Verify consistency of: offset_real = -wall_to_monotonic
|
| 59 | * before modifying anything
|
| 60 |
|
| 61 | set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
|
| 62 | -tk->wall_to_monotonic.tv_nsec);
|
| 63 | WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
|
| 64 | tk->wall_to_monotonic = wtm;
|
| 65 | set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
|
| 66 | tk->offs_real = timespec_to_ktime(tmp);
|
| 67 | }
|
| 68 |
|
| 69 | static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
|
| 70 | {
|
| 71 |
|
| 72 | WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
|
| 73 |
|
| 74 | tk->total_sleep_time = t;
|
| 75 | tk->offs_boot = timespec_to_ktime(t);
|
| 76 | }
|
| 77 |
|
| 78 | |
| 79 | * timekeeper_setup_internals - Set up internals to use clocksource clock.
|
| 80 | *
|
| 81 | * @clock: Pointer to clocksource.
|
| 82 | *
|
| 83 | * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
|
| 84 | * pair and interval request.
|
| 85 | *
|
| 86 | * Unless you're the timekeeping code, you should not be using this!
|
| 87 |
|
| 88 | static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
|
| 89 | {
|
| 90 | cycle_t interval;
|
| 91 | u64 tmp, ntpinterval;
|
| 92 | struct clocksource *old_clock;
|
| 93 |
|
| 94 | old_clock = tk->clock;
|
| 95 | tk->clock = clock;
|
| 96 | clock->cycle_last = clock->read(clock);
|
| 97 |
|
| 98 |
|
| 99 | tmp = NTP_INTERVAL_LENGTH;
|
| 100 | tmp <<= clock->shift;
|
| 101 | ntpinterval = tmp;
|
| 102 | tmp += clock->mult/2;
|
| 103 | do_div(tmp, clock->mult);
|
| 104 | if (tmp == 0)
|
| 105 | tmp = 1;
|
| 106 |
|
| 107 | interval = (cycle_t) tmp;
|
| 108 | tk->cycle_interval = interval;
|
| 109 |
|
| 110 |
|
| 111 | tk->xtime_interval = (u64) interval * clock->mult;
|
| 112 | tk->xtime_remainder = ntpinterval - tk->xtime_interval;
|
| 113 | tk->raw_interval =
|
| 114 | ((u64) interval * clock->mult) >> clock->shift;
|
| 115 |
|
| 116 |
|
| 117 | if (old_clock) {
|
| 118 | int shift_change = clock->shift - old_clock->shift;
|
| 119 | if (shift_change < 0)
|
| 120 | tk->xtime_nsec >>= -shift_change;
|
| 121 | else
|
| 122 | tk->xtime_nsec <<= shift_change;
|
| 123 | }
|
| 124 | tk->shift = clock->shift;
|
| 125 |
|
| 126 | tk->ntp_error = 0;
|
| 127 | tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
|
| 128 |
|
| 129 | |
| 130 | * The timekeeper keeps its own mult values for the currently
|
| 131 | * active clocksource. These value will be adjusted via NTP
|
| 132 | * to counteract clock drifting.
|
| 133 |
|
| 134 | tk->mult = clock->mult;
|
| 135 | }
|
| 136 |
|
| 137 |
|
| 138 | static inline s64 timekeeping_get_ns(struct timekeeper *tk)
|
| 139 | {
|
| 140 | cycle_t cycle_now, cycle_delta;
|
| 141 | struct clocksource *clock;
|
| 142 | s64 nsec;
|
| 143 |
|
| 144 |
|
| 145 | clock = tk->clock;
|
| 146 | cycle_now = clock->read(clock);
|
| 147 |
|
| 148 |
|
| 149 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
|
| 150 |
|
| 151 | nsec = cycle_delta * tk->mult + tk->xtime_nsec;
|
| 152 | nsec >>= tk->shift;
|
| 153 |
|
| 154 |
|
| 155 | return nsec + arch_gettimeoffset();
|
| 156 | }
|
| 157 |
|
| 158 | static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
|
| 159 | {
|
| 160 | cycle_t cycle_now, cycle_delta;
|
| 161 | struct clocksource *clock;
|
| 162 | s64 nsec;
|
| 163 |
|
| 164 |
|
| 165 | clock = tk->clock;
|
| 166 | cycle_now = clock->read(clock);
|
| 167 |
|
| 168 |
|
| 169 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
|
| 170 |
|
| 171 |
|
| 172 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
|
| 173 |
|
| 174 |
|
| 175 | return nsec + arch_gettimeoffset();
|
| 176 | }
|
| 177 |
|
| 178 | static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
|
| 179 |
|
| 180 | static void update_pvclock_gtod(struct timekeeper *tk)
|
| 181 | {
|
| 182 | raw_notifier_call_chain(&pvclock_gtod_chain, 0, tk);
|
| 183 | }
|
| 184 |
|
| 185 | |
| 186 | * pvclock_gtod_register_notifier - register a pvclock timedata update listener
|
| 187 | *
|
| 188 | * Must hold write on timekeeper.lock
|
| 189 |
|
| 190 | int pvclock_gtod_register_notifier(struct notifier_block *nb)
|
| 191 | {
|
| 192 | struct timekeeper *tk = &timekeeper;
|
| 193 | unsigned long flags;
|
| 194 | int ret;
|
| 195 |
|
| 196 | write_seqlock_irqsave(&tk->lock, flags);
|
| 197 | ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
|
| 198 |
|
| 199 | update_pvclock_gtod(tk);
|
| 200 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 201 |
|
| 202 | return ret;
|
| 203 | }
|
| 204 | EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
|
| 205 |
|
| 206 | |
| 207 | * pvclock_gtod_unregister_notifier - unregister a pvclock
|
| 208 | * timedata update listener
|
| 209 | *
|
| 210 | * Must hold write on timekeeper.lock
|
| 211 |
|
| 212 | int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
|
| 213 | {
|
| 214 | struct timekeeper *tk = &timekeeper;
|
| 215 | unsigned long flags;
|
| 216 | int ret;
|
| 217 |
|
| 218 | write_seqlock_irqsave(&tk->lock, flags);
|
| 219 | ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
|
| 220 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 221 |
|
| 222 | return ret;
|
| 223 | }
|
| 224 | EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
|
| 225 |
|
| 226 |
|
| 227 | static void timekeeping_update(struct timekeeper *tk, bool clearntp)
|
| 228 | {
|
| 229 | if (clearntp) {
|
| 230 | tk->ntp_error = 0;
|
| 231 | ntp_clear();
|
| 232 | }
|
| 233 | update_vsyscall(tk);
|
| 234 | update_pvclock_gtod(tk);
|
| 235 | }
|
| 236 |
|
| 237 | |
| 238 | * timekeeping_forward_now - update clock to the current time
|
| 239 | *
|
| 240 | * Forward the current clock to update its state since the last call to
|
| 241 | * update_wall_time(). This is useful before significant clock changes,
|
| 242 | * as it avoids having to deal with this time offset explicitly.
|
| 243 |
|
| 244 | static void timekeeping_forward_now(struct timekeeper *tk)
|
| 245 | {
|
| 246 | cycle_t cycle_now, cycle_delta;
|
| 247 | struct clocksource *clock;
|
| 248 | s64 nsec;
|
| 249 |
|
| 250 | clock = tk->clock;
|
| 251 | cycle_now = clock->read(clock);
|
| 252 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
|
| 253 | clock->cycle_last = cycle_now;
|
| 254 |
|
| 255 | tk->xtime_nsec += cycle_delta * tk->mult;
|
| 256 |
|
| 257 |
|
| 258 | tk->xtime_nsec += (u64)arch_gettimeoffset() << tk->shift;
|
| 259 |
|
| 260 | tk_normalize_xtime(tk);
|
| 261 |
|
| 262 | nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
|
| 263 | timespec_add_ns(&tk->raw_time, nsec);
|
| 264 | }
|
| 265 |
|
| 266 | |
| 267 | * getnstimeofday - Returns the time of day in a timespec
|
| 268 | * @ts: pointer to the timespec to be set
|
| 269 | *
|
| 270 | * Returns the time of day in a timespec.
|
| 271 |
|
| 272 | void getnstimeofday(struct timespec *ts)
|
| 273 | {
|
| 274 | struct timekeeper *tk = &timekeeper;
|
| 275 | unsigned long seq;
|
| 276 | s64 nsecs = 0;
|
| 277 |
|
| 278 | WARN_ON(timekeeping_suspended);
|
| 279 |
|
| 280 | do {
|
| 281 | seq = read_seqbegin(&tk->lock);
|
| 282 |
|
| 283 | ts->tv_sec = tk->xtime_sec;
|
| 284 | nsecs = timekeeping_get_ns(tk);
|
| 285 |
|
| 286 | } while (read_seqretry(&tk->lock, seq));
|
| 287 |
|
| 288 | ts->tv_nsec = 0;
|
| 289 | timespec_add_ns(ts, nsecs);
|
| 290 | }
|
| 291 | EXPORT_SYMBOL(getnstimeofday);
|
| 292 |
|
| 293 | ktime_t ktime_get(void)
|
| 294 | {
|
| 295 | struct timekeeper *tk = &timekeeper;
|
| 296 | unsigned int seq;
|
| 297 | s64 secs, nsecs;
|
| 298 |
|
| 299 | WARN_ON(timekeeping_suspended);
|
| 300 |
|
| 301 | do {
|
| 302 | seq = read_seqbegin(&tk->lock);
|
| 303 | secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
|
| 304 | nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
|
| 305 |
|
| 306 | } while (read_seqretry(&tk->lock, seq));
|
| 307 | |
| 308 | * Use ktime_set/ktime_add_ns to create a proper ktime on
|
| 309 | * 32-bit architectures without CONFIG_KTIME_SCALAR.
|
| 310 |
|
| 311 | return ktime_add_ns(ktime_set(secs, 0), nsecs);
|
| 312 | }
|
| 313 | EXPORT_SYMBOL_GPL(ktime_get);
|
| 314 |
|
| 315 | |
| 316 | * ktime_get_ts - get the monotonic clock in timespec format
|
| 317 | * @ts: pointer to timespec variable
|
| 318 | *
|
| 319 | * The function calculates the monotonic clock from the realtime
|
| 320 | * clock and the wall_to_monotonic offset and stores the result
|
| 321 | * in normalized timespec format in the variable pointed to by @ts.
|
| 322 |
|
| 323 | void ktime_get_ts(struct timespec *ts)
|
| 324 | {
|
| 325 | struct timekeeper *tk = &timekeeper;
|
| 326 | struct timespec tomono;
|
| 327 | s64 nsec;
|
| 328 | unsigned int seq;
|
| 329 |
|
| 330 | WARN_ON(timekeeping_suspended);
|
| 331 |
|
| 332 | do {
|
| 333 | seq = read_seqbegin(&tk->lock);
|
| 334 | ts->tv_sec = tk->xtime_sec;
|
| 335 | nsec = timekeeping_get_ns(tk);
|
| 336 | tomono = tk->wall_to_monotonic;
|
| 337 |
|
| 338 | } while (read_seqretry(&tk->lock, seq));
|
| 339 |
|
| 340 | ts->tv_sec += tomono.tv_sec;
|
| 341 | ts->tv_nsec = 0;
|
| 342 | timespec_add_ns(ts, nsec + tomono.tv_nsec);
|
| 343 | }
|
| 344 | EXPORT_SYMBOL_GPL(ktime_get_ts);
|
| 345 |
|
| 346 | #ifdef CONFIG_NTP_PPS
|
| 347 |
|
| 348 | |
| 349 | * getnstime_raw_and_real - get day and raw monotonic time in timespec format
|
| 350 | * @ts_raw: pointer to the timespec to be set to raw monotonic time
|
| 351 | * @ts_real: pointer to the timespec to be set to the time of day
|
| 352 | *
|
| 353 | * This function reads both the time of day and raw monotonic time at the
|
| 354 | * same time atomically and stores the resulting timestamps in timespec
|
| 355 | * format.
|
| 356 |
|
| 357 | void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
|
| 358 | {
|
| 359 | struct timekeeper *tk = &timekeeper;
|
| 360 | unsigned long seq;
|
| 361 | s64 nsecs_raw, nsecs_real;
|
| 362 |
|
| 363 | WARN_ON_ONCE(timekeeping_suspended);
|
| 364 |
|
| 365 | do {
|
| 366 | seq = read_seqbegin(&tk->lock);
|
| 367 |
|
| 368 | *ts_raw = tk->raw_time;
|
| 369 | ts_real->tv_sec = tk->xtime_sec;
|
| 370 | ts_real->tv_nsec = 0;
|
| 371 |
|
| 372 | nsecs_raw = timekeeping_get_ns_raw(tk);
|
| 373 | nsecs_real = timekeeping_get_ns(tk);
|
| 374 |
|
| 375 | } while (read_seqretry(&tk->lock, seq));
|
| 376 |
|
| 377 | timespec_add_ns(ts_raw, nsecs_raw);
|
| 378 | timespec_add_ns(ts_real, nsecs_real);
|
| 379 | }
|
| 380 | EXPORT_SYMBOL(getnstime_raw_and_real);
|
| 381 |
|
| 382 | #endif
|
| 383 |
|
| 384 | |
| 385 | * do_gettimeofday - Returns the time of day in a timeval
|
| 386 | * @tv: pointer to the timeval to be set
|
| 387 | *
|
| 388 | * NOTE: Users should be converted to using getnstimeofday()
|
| 389 |
|
| 390 | void do_gettimeofday(struct timeval *tv)
|
| 391 | {
|
| 392 | struct timespec now;
|
| 393 |
|
| 394 | getnstimeofday(&now);
|
| 395 | tv->tv_sec = now.tv_sec;
|
| 396 | tv->tv_usec = now.tv_nsec/1000;
|
| 397 | }
|
| 398 | EXPORT_SYMBOL(do_gettimeofday);
|
| 399 |
|
| 400 | |
| 401 | * do_settimeofday - Sets the time of day
|
| 402 | * @tv: pointer to the timespec variable containing the new time
|
| 403 | *
|
| 404 | * Sets the time of day to the new time and update NTP and notify hrtimers
|
| 405 |
|
| 406 | int do_settimeofday(const struct timespec *tv)
|
| 407 | {
|
| 408 | struct timekeeper *tk = &timekeeper;
|
| 409 | struct timespec ts_delta, xt;
|
| 410 | unsigned long flags;
|
| 411 |
|
| 412 | if (!timespec_valid_strict(tv))
|
| 413 | return -EINVAL;
|
| 414 |
|
| 415 | write_seqlock_irqsave(&tk->lock, flags);
|
| 416 |
|
| 417 | timekeeping_forward_now(tk);
|
| 418 |
|
| 419 | xt = tk_xtime(tk);
|
| 420 | ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
|
| 421 | ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
|
| 422 |
|
| 423 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
|
| 424 |
|
| 425 | tk_set_xtime(tk, tv);
|
| 426 |
|
| 427 | timekeeping_update(tk, true);
|
| 428 |
|
| 429 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 430 |
|
| 431 |
|
| 432 | clock_was_set();
|
| 433 |
|
| 434 | return 0;
|
| 435 | }
|
| 436 | EXPORT_SYMBOL(do_settimeofday);
|
| 437 |
|
| 438 | |
| 439 | * timekeeping_inject_offset - Adds or subtracts from the current time.
|
| 440 | * @tv: pointer to the timespec variable containing the offset
|
| 441 | *
|
| 442 | * Adds or subtracts an offset value from the current time.
|
| 443 |
|
| 444 | int timekeeping_inject_offset(struct timespec *ts)
|
| 445 | {
|
| 446 | struct timekeeper *tk = &timekeeper;
|
| 447 | unsigned long flags;
|
| 448 | struct timespec tmp;
|
| 449 | int ret = 0;
|
| 450 |
|
| 451 | if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
|
| 452 | return -EINVAL;
|
| 453 |
|
| 454 | write_seqlock_irqsave(&tk->lock, flags);
|
| 455 |
|
| 456 | timekeeping_forward_now(tk);
|
| 457 |
|
| 458 |
|
| 459 | tmp = timespec_add(tk_xtime(tk), *ts);
|
| 460 | if (!timespec_valid_strict(&tmp)) {
|
| 461 | ret = -EINVAL;
|
| 462 | goto error;
|
| 463 | }
|
| 464 |
|
| 465 | tk_xtime_add(tk, ts);
|
| 466 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
|
| 467 |
|
| 468 | error:
|
| 469 | timekeeping_update(tk, true);
|
| 470 |
|
| 471 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 472 |
|
| 473 |
|
| 474 | clock_was_set();
|
| 475 |
|
| 476 | return ret;
|
| 477 | }
|
| 478 | EXPORT_SYMBOL(timekeeping_inject_offset);
|
| 479 |
|
| 480 | |
| 481 | * change_clocksource - Swaps clocksources if a new one is available
|
| 482 | *
|
| 483 | * Accumulates current time interval and initializes new clocksource
|
| 484 |
|
| 485 | static int change_clocksource(void *data)
|
| 486 | {
|
| 487 | struct timekeeper *tk = &timekeeper;
|
| 488 | struct clocksource *new, *old;
|
| 489 | unsigned long flags;
|
| 490 |
|
| 491 | new = (struct clocksource *) data;
|
| 492 |
|
| 493 | write_seqlock_irqsave(&tk->lock, flags);
|
| 494 |
|
| 495 | timekeeping_forward_now(tk);
|
| 496 | if (!new->enable || new->enable(new) == 0) {
|
| 497 | old = tk->clock;
|
| 498 | tk_setup_internals(tk, new);
|
| 499 | if (old->disable)
|
| 500 | old->disable(old);
|
| 501 | }
|
| 502 | timekeeping_update(tk, true);
|
| 503 |
|
| 504 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 505 |
|
| 506 | return 0;
|
| 507 | }
|
| 508 |
|
| 509 | |
| 510 | * timekeeping_notify - Install a new clock source
|
| 511 | * @clock: pointer to the clock source
|
| 512 | *
|
| 513 | * This function is called from clocksource.c after a new, better clock
|
| 514 | * source has been registered. The caller holds the clocksource_mutex.
|
| 515 |
|
| 516 | void timekeeping_notify(struct clocksource *clock)
|
| 517 | {
|
| 518 | struct timekeeper *tk = &timekeeper;
|
| 519 |
|
| 520 | if (tk->clock == clock)
|
| 521 | return;
|
| 522 | stop_machine(change_clocksource, clock, NULL);
|
| 523 | tick_clock_notify();
|
| 524 | }
|
| 525 |
|
| 526 | |
| 527 | * ktime_get_real - get the real (wall-) time in ktime_t format
|
| 528 | *
|
| 529 | * returns the time in ktime_t format
|
| 530 |
|
| 531 | ktime_t ktime_get_real(void)
|
| 532 | {
|
| 533 | struct timespec now;
|
| 534 |
|
| 535 | getnstimeofday(&now);
|
| 536 |
|
| 537 | return timespec_to_ktime(now);
|
| 538 | }
|
| 539 | EXPORT_SYMBOL_GPL(ktime_get_real);
|
| 540 |
|
| 541 | |
| 542 | * getrawmonotonic - Returns the raw monotonic time in a timespec
|
| 543 | * @ts: pointer to the timespec to be set
|
| 544 | *
|
| 545 | * Returns the raw monotonic time (completely un-modified by ntp)
|
| 546 |
|
| 547 | void getrawmonotonic(struct timespec *ts)
|
| 548 | {
|
| 549 | struct timekeeper *tk = &timekeeper;
|
| 550 | unsigned long seq;
|
| 551 | s64 nsecs;
|
| 552 |
|
| 553 | do {
|
| 554 | seq = read_seqbegin(&tk->lock);
|
| 555 | nsecs = timekeeping_get_ns_raw(tk);
|
| 556 | *ts = tk->raw_time;
|
| 557 |
|
| 558 | } while (read_seqretry(&tk->lock, seq));
|
| 559 |
|
| 560 | timespec_add_ns(ts, nsecs);
|
| 561 | }
|
| 562 | EXPORT_SYMBOL(getrawmonotonic);
|
| 563 |
|
| 564 | |
| 565 | * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
|
| 566 |
|
| 567 | int timekeeping_valid_for_hres(void)
|
| 568 | {
|
| 569 | struct timekeeper *tk = &timekeeper;
|
| 570 | unsigned long seq;
|
| 571 | int ret;
|
| 572 |
|
| 573 | do {
|
| 574 | seq = read_seqbegin(&tk->lock);
|
| 575 |
|
| 576 | ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
|
| 577 |
|
| 578 | } while (read_seqretry(&tk->lock, seq));
|
| 579 |
|
| 580 | return ret;
|
| 581 | }
|
| 582 |
|
| 583 | |
| 584 | * timekeeping_max_deferment - Returns max time the clocksource can be deferred
|
| 585 |
|
| 586 | u64 timekeeping_max_deferment(void)
|
| 587 | {
|
| 588 | struct timekeeper *tk = &timekeeper;
|
| 589 | unsigned long seq;
|
| 590 | u64 ret;
|
| 591 |
|
| 592 | do {
|
| 593 | seq = read_seqbegin(&tk->lock);
|
| 594 |
|
| 595 | ret = tk->clock->max_idle_ns;
|
| 596 |
|
| 597 | } while (read_seqretry(&tk->lock, seq));
|
| 598 |
|
| 599 | return ret;
|
| 600 | }
|
| 601 |
|
| 602 | |
| 603 | * read_persistent_clock - Return time from the persistent clock.
|
| 604 | *
|
| 605 | * Weak dummy function for arches that do not yet support it.
|
| 606 | * Reads the time from the battery backed persistent clock.
|
| 607 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
|
| 608 | *
|
| 609 | * XXX - Do be sure to remove it once all arches implement it.
|
| 610 |
|
| 611 | void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
|
| 612 | {
|
| 613 | ts->tv_sec = 0;
|
| 614 | ts->tv_nsec = 0;
|
| 615 | }
|
| 616 |
|
| 617 | |
| 618 | * read_boot_clock - Return time of the system start.
|
| 619 | *
|
| 620 | * Weak dummy function for arches that do not yet support it.
|
| 621 | * Function to read the exact time the system has been started.
|
| 622 | * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
|
| 623 | *
|
| 624 | * XXX - Do be sure to remove it once all arches implement it.
|
| 625 |
|
| 626 | void __attribute__((weak)) read_boot_clock(struct timespec *ts)
|
| 627 | {
|
| 628 | ts->tv_sec = 0;
|
| 629 | ts->tv_nsec = 0;
|
| 630 | }
|
| 631 |
|
| 632 | |
| 633 | * timekeeping_init - Initializes the clocksource and common timekeeping values
|
| 634 |
|
| 635 | void __init timekeeping_init(void)
|
| 636 | {
|
| 637 | struct timekeeper *tk = &timekeeper;
|
| 638 | struct clocksource *clock;
|
| 639 | unsigned long flags;
|
| 640 | struct timespec now, boot, tmp;
|
| 641 |
|
| 642 | read_persistent_clock(&now);
|
| 643 | if (!timespec_valid_strict(&now)) {
|
| 644 | pr_warn("WARNING: Persistent clock returned invalid value!\n"
|
| 645 | " Check your CMOS/BIOS settings.\n");
|
| 646 | now.tv_sec = 0;
|
| 647 | now.tv_nsec = 0;
|
| 648 | }
|
| 649 |
|
| 650 | read_boot_clock(&boot);
|
| 651 | if (!timespec_valid_strict(&boot)) {
|
| 652 | pr_warn("WARNING: Boot clock returned invalid value!\n"
|
| 653 | " Check your CMOS/BIOS settings.\n");
|
| 654 | boot.tv_sec = 0;
|
| 655 | boot.tv_nsec = 0;
|
| 656 | }
|
| 657 |
|
| 658 | seqlock_init(&tk->lock);
|
| 659 |
|
| 660 | ntp_init();
|
| 661 |
|
| 662 | write_seqlock_irqsave(&tk->lock, flags);
|
| 663 | clock = clocksource_default_clock();
|
| 664 | if (clock->enable)
|
| 665 | clock->enable(clock);
|
| 666 | tk_setup_internals(tk, clock);
|
| 667 |
|
| 668 | tk_set_xtime(tk, &now);
|
| 669 | tk->raw_time.tv_sec = 0;
|
| 670 | tk->raw_time.tv_nsec = 0;
|
| 671 | if (boot.tv_sec == 0 && boot.tv_nsec == 0)
|
| 672 | boot = tk_xtime(tk);
|
| 673 |
|
| 674 | set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
|
| 675 | tk_set_wall_to_mono(tk, tmp);
|
| 676 |
|
| 677 | tmp.tv_sec = 0;
|
| 678 | tmp.tv_nsec = 0;
|
| 679 | tk_set_sleep_time(tk, tmp);
|
| 680 |
|
| 681 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 682 | }
|
| 683 |
|
| 684 |
|
| 685 | static struct timespec timekeeping_suspend_time;
|
| 686 |
|
| 687 | |
| 688 | * __timekeeping_inject_sleeptime - Internal function to add sleep interval
|
| 689 | * @delta: pointer to a timespec delta value
|
| 690 | *
|
| 691 | * Takes a timespec offset measuring a suspend interval and properly
|
| 692 | * adds the sleep offset to the timekeeping variables.
|
| 693 |
|
| 694 | static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
|
| 695 | struct timespec *delta)
|
| 696 | {
|
| 697 | if (!timespec_valid_strict(delta)) {
|
| 698 | printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
|
| 699 | "sleep delta value!\n");
|
| 700 | return;
|
| 701 | }
|
| 702 | tk_xtime_add(tk, delta);
|
| 703 | tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
|
| 704 | tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
|
| 705 | }
|
| 706 |
|
| 707 | |
| 708 | * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
|
| 709 | * @delta: pointer to a timespec delta value
|
| 710 | *
|
| 711 | * This hook is for architectures that cannot support read_persistent_clock
|
| 712 | * because their RTC/persistent clock is only accessible when irqs are enabled.
|
| 713 | *
|
| 714 | * This function should only be called by rtc_resume(), and allows
|
| 715 | * a suspend offset to be injected into the timekeeping values.
|
| 716 |
|
| 717 | void timekeeping_inject_sleeptime(struct timespec *delta)
|
| 718 | {
|
| 719 | struct timekeeper *tk = &timekeeper;
|
| 720 | unsigned long flags;
|
| 721 | struct timespec ts;
|
| 722 |
|
| 723 |
|
| 724 | read_persistent_clock(&ts);
|
| 725 | if (!(ts.tv_sec == 0 && ts.tv_nsec == 0))
|
| 726 | return;
|
| 727 |
|
| 728 | write_seqlock_irqsave(&tk->lock, flags);
|
| 729 |
|
| 730 | timekeeping_forward_now(tk);
|
| 731 |
|
| 732 | __timekeeping_inject_sleeptime(tk, delta);
|
| 733 |
|
| 734 | timekeeping_update(tk, true);
|
| 735 |
|
| 736 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 737 |
|
| 738 |
|
| 739 | clock_was_set();
|
| 740 | }
|
| 741 |
|
| 742 | |
| 743 | * timekeeping_resume - Resumes the generic timekeeping subsystem.
|
| 744 | *
|
| 745 | * This is for the generic clocksource timekeeping.
|
| 746 | * xtime/wall_to_monotonic/jiffies/etc are
|
| 747 | * still managed by arch specific suspend/resume code.
|
| 748 |
|
| 749 | static void timekeeping_resume(void)
|
| 750 | {
|
| 751 | struct timekeeper *tk = &timekeeper;
|
| 752 | unsigned long flags;
|
| 753 | struct timespec ts;
|
| 754 |
|
| 755 | read_persistent_clock(&ts);
|
| 756 |
|
| 757 | clockevents_resume();
|
| 758 | clocksource_resume();
|
| 759 |
|
| 760 | write_seqlock_irqsave(&tk->lock, flags);
|
| 761 |
|
| 762 | if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
|
| 763 | ts = timespec_sub(ts, timekeeping_suspend_time);
|
| 764 | __timekeeping_inject_sleeptime(tk, &ts);
|
| 765 | }
|
| 766 |
|
| 767 | tk->clock->cycle_last = tk->clock->read(tk->clock);
|
| 768 | tk->ntp_error = 0;
|
| 769 | timekeeping_suspended = 0;
|
| 770 | timekeeping_update(tk, false);
|
| 771 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 772 |
|
| 773 | touch_softlockup_watchdog();
|
| 774 |
|
| 775 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
|
| 776 |
|
| 777 |
|
| 778 | hrtimers_resume();
|
| 779 | }
|
| 780 |
|
| 781 | static int timekeeping_suspend(void)
|
| 782 | {
|
| 783 | struct timekeeper *tk = &timekeeper;
|
| 784 | unsigned long flags;
|
| 785 | struct timespec delta, delta_delta;
|
| 786 | static struct timespec old_delta;
|
| 787 |
|
| 788 | read_persistent_clock(&timekeeping_suspend_time);
|
| 789 |
|
| 790 | write_seqlock_irqsave(&tk->lock, flags);
|
| 791 | timekeeping_forward_now(tk);
|
| 792 | timekeeping_suspended = 1;
|
| 793 |
|
| 794 | |
| 795 | * To avoid drift caused by repeated suspend/resumes,
|
| 796 | * which each can add ~1 second drift error,
|
| 797 | * try to compensate so the difference in system time
|
| 798 | * and persistent_clock time stays close to constant.
|
| 799 |
|
| 800 | delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
|
| 801 | delta_delta = timespec_sub(delta, old_delta);
|
| 802 | if (abs(delta_delta.tv_sec) >= 2) {
|
| 803 | |
| 804 | * if delta_delta is too large, assume time correction
|
| 805 | * has occured and set old_delta to the current delta.
|
| 806 |
|
| 807 | old_delta = delta;
|
| 808 | } else {
|
| 809 |
|
| 810 | timekeeping_suspend_time =
|
| 811 | timespec_add(timekeeping_suspend_time, delta_delta);
|
| 812 | }
|
| 813 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 814 |
|
| 815 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
|
| 816 | clocksource_suspend();
|
| 817 | clockevents_suspend();
|
| 818 |
|
| 819 | return 0;
|
| 820 | }
|
| 821 |
|
| 822 |
|
| 823 | static struct syscore_ops timekeeping_syscore_ops = {
|
| 824 | .resume = timekeeping_resume,
|
| 825 | .suspend = timekeeping_suspend,
|
| 826 | };
|
| 827 |
|
| 828 | static int __init timekeeping_init_ops(void)
|
| 829 | {
|
| 830 | register_syscore_ops(&timekeeping_syscore_ops);
|
| 831 | return 0;
|
| 832 | }
|
| 833 |
|
| 834 | device_initcall(timekeeping_init_ops);
|
| 835 |
|
| 836 | |
| 837 | * If the error is already larger, we look ahead even further
|
| 838 | * to compensate for late or lost adjustments.
|
| 839 |
|
| 840 | static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
|
| 841 | s64 error, s64 *interval,
|
| 842 | s64 *offset)
|
| 843 | {
|
| 844 | s64 tick_error, i;
|
| 845 | u32 look_ahead, adj;
|
| 846 | s32 error2, mult;
|
| 847 |
|
| 848 | |
| 849 | * Use the current error value to determine how much to look ahead.
|
| 850 | * The larger the error the slower we adjust for it to avoid problems
|
| 851 | * with losing too many ticks, otherwise we would overadjust and
|
| 852 | * produce an even larger error. The smaller the adjustment the
|
| 853 | * faster we try to adjust for it, as lost ticks can do less harm
|
| 854 | * here. This is tuned so that an error of about 1 msec is adjusted
|
| 855 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
|
| 856 |
|
| 857 | error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
|
| 858 | error2 = abs(error2);
|
| 859 | for (look_ahead = 0; error2 > 0; look_ahead++)
|
| 860 | error2 >>= 2;
|
| 861 |
|
| 862 | |
| 863 | * Now calculate the error in (1 << look_ahead) ticks, but first
|
| 864 | * remove the single look ahead already included in the error.
|
| 865 |
|
| 866 | tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
|
| 867 | tick_error -= tk->xtime_interval >> 1;
|
| 868 | error = ((error - tick_error) >> look_ahead) + tick_error;
|
| 869 |
|
| 870 |
|
| 871 | i = *interval;
|
| 872 | mult = 1;
|
| 873 | if (error < 0) {
|
| 874 | error = -error;
|
| 875 | *interval = -*interval;
|
| 876 | *offset = -*offset;
|
| 877 | mult = -1;
|
| 878 | }
|
| 879 | for (adj = 0; error > i; adj++)
|
| 880 | error >>= 1;
|
| 881 |
|
| 882 | *interval <<= adj;
|
| 883 | *offset <<= adj;
|
| 884 | return mult << adj;
|
| 885 | }
|
| 886 |
|
| 887 | |
| 888 | * Adjust the multiplier to reduce the error value,
|
| 889 | * this is optimized for the most common adjustments of -1,0,1,
|
| 890 | * for other values we can do a bit more work.
|
| 891 |
|
| 892 | static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
|
| 893 | {
|
| 894 | s64 error, interval = tk->cycle_interval;
|
| 895 | int adj;
|
| 896 |
|
| 897 | |
| 898 | * The point of this is to check if the error is greater than half
|
| 899 | * an interval.
|
| 900 | *
|
| 901 | * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
|
| 902 | *
|
| 903 | * Note we subtract one in the shift, so that error is really error*2.
|
| 904 | * This "saves" dividing(shifting) interval twice, but keeps the
|
| 905 | * (error > interval) comparison as still measuring if error is
|
| 906 | * larger than half an interval.
|
| 907 | *
|
| 908 | * Note: It does not "save" on aggravation when reading the code.
|
| 909 |
|
| 910 | error = tk->ntp_error >> (tk->ntp_error_shift - 1);
|
| 911 | if (error > interval) {
|
| 912 | |
| 913 | * We now divide error by 4(via shift), which checks if
|
| 914 | * the error is greater than twice the interval.
|
| 915 | * If it is greater, we need a bigadjust, if its smaller,
|
| 916 | * we can adjust by 1.
|
| 917 |
|
| 918 | error >>= 2;
|
| 919 | |
| 920 | * XXX - In update_wall_time, we round up to the next
|
| 921 | * nanosecond, and store the amount rounded up into
|
| 922 | * the error. This causes the likely below to be unlikely.
|
| 923 | *
|
| 924 | * The proper fix is to avoid rounding up by using
|
| 925 | * the high precision tk->xtime_nsec instead of
|
| 926 | * xtime.tv_nsec everywhere. Fixing this will take some
|
| 927 | * time.
|
| 928 |
|
| 929 | if (likely(error <= interval))
|
| 930 | adj = 1;
|
| 931 | else
|
| 932 | adj = timekeeping_bigadjust(tk, error, &interval, &offset);
|
| 933 | } else {
|
| 934 | if (error < -interval) {
|
| 935 |
|
| 936 | error >>= 2;
|
| 937 | if (likely(error >= -interval)) {
|
| 938 | adj = -1;
|
| 939 | interval = -interval;
|
| 940 | offset = -offset;
|
| 941 | } else {
|
| 942 | adj = timekeeping_bigadjust(tk, error, &interval, &offset);
|
| 943 | }
|
| 944 | } else {
|
| 945 | goto out_adjust;
|
| 946 | }
|
| 947 | }
|
| 948 |
|
| 949 | if (unlikely(tk->clock->maxadj &&
|
| 950 | (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
|
| 951 | printk_once(KERN_WARNING
|
| 952 | "Adjusting %s more than 11%% (%ld vs %ld)\n",
|
| 953 | tk->clock->name, (long)tk->mult + adj,
|
| 954 | (long)tk->clock->mult + tk->clock->maxadj);
|
| 955 | }
|
| 956 | |
| 957 | * So the following can be confusing.
|
| 958 | *
|
| 959 | * To keep things simple, lets assume adj == 1 for now.
|
| 960 | *
|
| 961 | * When adj != 1, remember that the interval and offset values
|
| 962 | * have been appropriately scaled so the math is the same.
|
| 963 | *
|
| 964 | * The basic idea here is that we're increasing the multiplier
|
| 965 | * by one, this causes the xtime_interval to be incremented by
|
| 966 | * one cycle_interval. This is because:
|
| 967 | * xtime_interval = cycle_interval * mult
|
| 968 | * So if mult is being incremented by one:
|
| 969 | * xtime_interval = cycle_interval * (mult + 1)
|
| 970 | * Its the same as:
|
| 971 | * xtime_interval = (cycle_interval * mult) + cycle_interval
|
| 972 | * Which can be shortened to:
|
| 973 | * xtime_interval += cycle_interval
|
| 974 | *
|
| 975 | * So offset stores the non-accumulated cycles. Thus the current
|
| 976 | * time (in shifted nanoseconds) is:
|
| 977 | * now = (offset * adj) + xtime_nsec
|
| 978 | * Now, even though we're adjusting the clock frequency, we have
|
| 979 | * to keep time consistent. In other words, we can't jump back
|
| 980 | * in time, and we also want to avoid jumping forward in time.
|
| 981 | *
|
| 982 | * So given the same offset value, we need the time to be the same
|
| 983 | * both before and after the freq adjustment.
|
| 984 | * now = (offset * adj_1) + xtime_nsec_1
|
| 985 | * now = (offset * adj_2) + xtime_nsec_2
|
| 986 | * So:
|
| 987 | * (offset * adj_1) + xtime_nsec_1 =
|
| 988 | * (offset * adj_2) + xtime_nsec_2
|
| 989 | * And we know:
|
| 990 | * adj_2 = adj_1 + 1
|
| 991 | * So:
|
| 992 | * (offset * adj_1) + xtime_nsec_1 =
|
| 993 | * (offset * (adj_1+1)) + xtime_nsec_2
|
| 994 | * (offset * adj_1) + xtime_nsec_1 =
|
| 995 | * (offset * adj_1) + offset + xtime_nsec_2
|
| 996 | * Canceling the sides:
|
| 997 | * xtime_nsec_1 = offset + xtime_nsec_2
|
| 998 | * Which gives us:
|
| 999 | * xtime_nsec_2 = xtime_nsec_1 - offset
|
| 1000 | * Which simplfies to:
|
| 1001 | * xtime_nsec -= offset
|
| 1002 | *
|
| 1003 | * XXX - TODO: Doc ntp_error calculation.
|
| 1004 |
|
| 1005 | tk->mult += adj;
|
| 1006 | tk->xtime_interval += interval;
|
| 1007 | tk->xtime_nsec -= offset;
|
| 1008 | tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
|
| 1009 |
|
| 1010 | out_adjust:
|
| 1011 | |
| 1012 | * It may be possible that when we entered this function, xtime_nsec
|
| 1013 | * was very small. Further, if we're slightly speeding the clocksource
|
| 1014 | * in the code above, its possible the required corrective factor to
|
| 1015 | * xtime_nsec could cause it to underflow.
|
| 1016 | *
|
| 1017 | * Now, since we already accumulated the second, cannot simply roll
|
| 1018 | * the accumulated second back, since the NTP subsystem has been
|
| 1019 | * notified via second_overflow. So instead we push xtime_nsec forward
|
| 1020 | * by the amount we underflowed, and add that amount into the error.
|
| 1021 | *
|
| 1022 | * We'll correct this error next time through this function, when
|
| 1023 | * xtime_nsec is not as small.
|
| 1024 |
|
| 1025 | if (unlikely((s64)tk->xtime_nsec < 0)) {
|
| 1026 | s64 neg = -(s64)tk->xtime_nsec;
|
| 1027 | tk->xtime_nsec = 0;
|
| 1028 | tk->ntp_error += neg << tk->ntp_error_shift;
|
| 1029 | }
|
| 1030 |
|
| 1031 | }
|
| 1032 |
|
| 1033 | |
| 1034 | * accumulate_nsecs_to_secs - Accumulates nsecs into secs
|
| 1035 | *
|
| 1036 | * Helper function that accumulates a the nsecs greater then a second
|
| 1037 | * from the xtime_nsec field to the xtime_secs field.
|
| 1038 | * It also calls into the NTP code to handle leapsecond processing.
|
| 1039 | *
|
| 1040 |
|
| 1041 | static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
|
| 1042 | {
|
| 1043 | u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
|
| 1044 |
|
| 1045 | while (tk->xtime_nsec >= nsecps) {
|
| 1046 | int leap;
|
| 1047 |
|
| 1048 | tk->xtime_nsec -= nsecps;
|
| 1049 | tk->xtime_sec++;
|
| 1050 |
|
| 1051 |
|
| 1052 | leap = second_overflow(tk->xtime_sec);
|
| 1053 | if (unlikely(leap)) {
|
| 1054 | struct timespec ts;
|
| 1055 |
|
| 1056 | tk->xtime_sec += leap;
|
| 1057 |
|
| 1058 | ts.tv_sec = leap;
|
| 1059 | ts.tv_nsec = 0;
|
| 1060 | tk_set_wall_to_mono(tk,
|
| 1061 | timespec_sub(tk->wall_to_monotonic, ts));
|
| 1062 |
|
| 1063 | clock_was_set_delayed();
|
| 1064 | }
|
| 1065 | }
|
| 1066 | }
|
| 1067 |
|
| 1068 | |
| 1069 | * logarithmic_accumulation - shifted accumulation of cycles
|
| 1070 | *
|
| 1071 | * This functions accumulates a shifted interval of cycles into
|
| 1072 | * into a shifted interval nanoseconds. Allows for O(log) accumulation
|
| 1073 | * loop.
|
| 1074 | *
|
| 1075 | * Returns the unconsumed cycles.
|
| 1076 |
|
| 1077 | static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
|
| 1078 | u32 shift)
|
| 1079 | {
|
| 1080 | u64 raw_nsecs;
|
| 1081 |
|
| 1082 |
|
| 1083 | if (offset < tk->cycle_interval<<shift)
|
| 1084 | return offset;
|
| 1085 |
|
| 1086 |
|
| 1087 | offset -= tk->cycle_interval << shift;
|
| 1088 | tk->clock->cycle_last += tk->cycle_interval << shift;
|
| 1089 |
|
| 1090 | tk->xtime_nsec += tk->xtime_interval << shift;
|
| 1091 | accumulate_nsecs_to_secs(tk);
|
| 1092 |
|
| 1093 |
|
| 1094 | raw_nsecs = (u64)tk->raw_interval << shift;
|
| 1095 | raw_nsecs += tk->raw_time.tv_nsec;
|
| 1096 | if (raw_nsecs >= NSEC_PER_SEC) {
|
| 1097 | u64 raw_secs = raw_nsecs;
|
| 1098 | raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
|
| 1099 | tk->raw_time.tv_sec += raw_secs;
|
| 1100 | }
|
| 1101 | tk->raw_time.tv_nsec = raw_nsecs;
|
| 1102 |
|
| 1103 |
|
| 1104 | tk->ntp_error += ntp_tick_length() << shift;
|
| 1105 | tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
|
| 1106 | (tk->ntp_error_shift + shift);
|
| 1107 |
|
| 1108 | return offset;
|
| 1109 | }
|
| 1110 |
|
| 1111 | #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
|
| 1112 | static inline void old_vsyscall_fixup(struct timekeeper *tk)
|
| 1113 | {
|
| 1114 | s64 remainder;
|
| 1115 |
|
| 1116 | |
| 1117 | * Store only full nanoseconds into xtime_nsec after rounding
|
| 1118 | * it up and add the remainder to the error difference.
|
| 1119 | * XXX - This is necessary to avoid small 1ns inconsistnecies caused
|
| 1120 | * by truncating the remainder in vsyscalls. However, it causes
|
| 1121 | * additional work to be done in timekeeping_adjust(). Once
|
| 1122 | * the vsyscall implementations are converted to use xtime_nsec
|
| 1123 | * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
|
| 1124 | * users are removed, this can be killed.
|
| 1125 |
|
| 1126 | remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
|
| 1127 | tk->xtime_nsec -= remainder;
|
| 1128 | tk->xtime_nsec += 1ULL << tk->shift;
|
| 1129 | tk->ntp_error += remainder << tk->ntp_error_shift;
|
| 1130 |
|
| 1131 | }
|
| 1132 | #else
|
| 1133 | #define old_vsyscall_fixup(tk)
|
| 1134 | #endif
|
| 1135 |
|
| 1136 |
|
| 1137 |
|
| 1138 | |
| 1139 | * update_wall_time - Uses the current clocksource to increment the wall time
|
| 1140 | *
|
| 1141 |
|
| 1142 | static void update_wall_time(void)
|
| 1143 | {
|
| 1144 | struct clocksource *clock;
|
| 1145 | struct timekeeper *tk = &timekeeper;
|
| 1146 | cycle_t offset;
|
| 1147 | int shift = 0, maxshift;
|
| 1148 | unsigned long flags;
|
| 1149 |
|
| 1150 | write_seqlock_irqsave(&tk->lock, flags);
|
| 1151 |
|
| 1152 |
|
| 1153 | if (unlikely(timekeeping_suspended))
|
| 1154 | goto out;
|
| 1155 |
|
| 1156 | clock = tk->clock;
|
| 1157 |
|
| 1158 | #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
|
| 1159 | offset = tk->cycle_interval;
|
| 1160 | #else
|
| 1161 | offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
|
| 1162 | #endif
|
| 1163 |
|
| 1164 |
|
| 1165 | if (offset < tk->cycle_interval)
|
| 1166 | goto out;
|
| 1167 |
|
| 1168 | |
| 1169 | * With NO_HZ we may have to accumulate many cycle_intervals
|
| 1170 | * (think "ticks") worth of time at once. To do this efficiently,
|
| 1171 | * we calculate the largest doubling multiple of cycle_intervals
|
| 1172 | * that is smaller than the offset. We then accumulate that
|
| 1173 | * chunk in one go, and then try to consume the next smaller
|
| 1174 | * doubled multiple.
|
| 1175 |
|
| 1176 | shift = ilog2(offset) - ilog2(tk->cycle_interval);
|
| 1177 | shift = max(0, shift);
|
| 1178 |
|
| 1179 | maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
|
| 1180 | shift = min(shift, maxshift);
|
| 1181 | while (offset >= tk->cycle_interval) {
|
| 1182 | offset = logarithmic_accumulation(tk, offset, shift);
|
| 1183 | if (offset < tk->cycle_interval<<shift)
|
| 1184 | shift--;
|
| 1185 | }
|
| 1186 |
|
| 1187 |
|
| 1188 | timekeeping_adjust(tk, offset);
|
| 1189 |
|
| 1190 | |
| 1191 | * XXX This can be killed once everyone converts
|
| 1192 | * to the new update_vsyscall.
|
| 1193 |
|
| 1194 | old_vsyscall_fixup(tk);
|
| 1195 |
|
| 1196 | |
| 1197 | * Finally, make sure that after the rounding
|
| 1198 | * xtime_nsec isn't larger than NSEC_PER_SEC
|
| 1199 |
|
| 1200 | accumulate_nsecs_to_secs(tk);
|
| 1201 |
|
| 1202 | timekeeping_update(tk, false);
|
| 1203 |
|
| 1204 | out:
|
| 1205 | write_sequnlock_irqrestore(&tk->lock, flags);
|
| 1206 |
|
| 1207 | }
|
| 1208 |
|
| 1209 | |
| 1210 | * getboottime - Return the real time of system boot.
|
| 1211 | * @ts: pointer to the timespec to be set
|
| 1212 | *
|
| 1213 | * Returns the wall-time of boot in a timespec.
|
| 1214 | *
|
| 1215 | * This is based on the wall_to_monotonic offset and the total suspend
|
| 1216 | * time. Calls to settimeofday will affect the value returned (which
|
| 1217 | * basically means that however wrong your real time clock is at boot time,
|
| 1218 | * you get the right time here).
|
| 1219 |
|
| 1220 | void getboottime(struct timespec *ts)
|
| 1221 | {
|
| 1222 | struct timekeeper *tk = &timekeeper;
|
| 1223 | struct timespec boottime = {
|
| 1224 | .tv_sec = tk->wall_to_monotonic.tv_sec +
|
| 1225 | tk->total_sleep_time.tv_sec,
|
| 1226 | .tv_nsec = tk->wall_to_monotonic.tv_nsec +
|
| 1227 | tk->total_sleep_time.tv_nsec
|
| 1228 | };
|
| 1229 |
|
| 1230 | set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
|
| 1231 | }
|
| 1232 | EXPORT_SYMBOL_GPL(getboottime);
|
| 1233 |
|
| 1234 | |
| 1235 | * get_monotonic_boottime - Returns monotonic time since boot
|
| 1236 | * @ts: pointer to the timespec to be set
|
| 1237 | *
|
| 1238 | * Returns the monotonic time since boot in a timespec.
|
| 1239 | *
|
| 1240 | * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
|
| 1241 | * includes the time spent in suspend.
|
| 1242 |
|
| 1243 | void get_monotonic_boottime(struct timespec *ts)
|
| 1244 | {
|
| 1245 | struct timekeeper *tk = &timekeeper;
|
| 1246 | struct timespec tomono, sleep;
|
| 1247 | s64 nsec;
|
| 1248 | unsigned int seq;
|
| 1249 |
|
| 1250 | WARN_ON(timekeeping_suspended);
|
| 1251 |
|
| 1252 | do {
|
| 1253 | seq = read_seqbegin(&tk->lock);
|
| 1254 | ts->tv_sec = tk->xtime_sec;
|
| 1255 | nsec = timekeeping_get_ns(tk);
|
| 1256 | tomono = tk->wall_to_monotonic;
|
| 1257 | sleep = tk->total_sleep_time;
|
| 1258 |
|
| 1259 | } while (read_seqretry(&tk->lock, seq));
|
| 1260 |
|
| 1261 | ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
|
| 1262 | ts->tv_nsec = 0;
|
| 1263 | timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
|
| 1264 | }
|
| 1265 | EXPORT_SYMBOL_GPL(get_monotonic_boottime);
|
| 1266 |
|
| 1267 | |
| 1268 | * ktime_get_boottime - Returns monotonic time since boot in a ktime
|
| 1269 | *
|
| 1270 | * Returns the monotonic time since boot in a ktime
|
| 1271 | *
|
| 1272 | * This is similar to CLOCK_MONTONIC/ktime_get, but also
|
| 1273 | * includes the time spent in suspend.
|
| 1274 |
|
| 1275 | ktime_t ktime_get_boottime(void)
|
| 1276 | {
|
| 1277 | struct timespec ts;
|
| 1278 |
|
| 1279 | get_monotonic_boottime(&ts);
|
| 1280 | return timespec_to_ktime(ts);
|
| 1281 | }
|
| 1282 | EXPORT_SYMBOL_GPL(ktime_get_boottime);
|
| 1283 |
|
| 1284 | |
| 1285 | * monotonic_to_bootbased - Convert the monotonic time to boot based.
|
| 1286 | * @ts: pointer to the timespec to be converted
|
| 1287 |
|
| 1288 | void monotonic_to_bootbased(struct timespec *ts)
|
| 1289 | {
|
| 1290 | struct timekeeper *tk = &timekeeper;
|
| 1291 |
|
| 1292 | *ts = timespec_add(*ts, tk->total_sleep_time);
|
| 1293 | }
|
| 1294 | EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
|
| 1295 |
|
| 1296 | unsigned long get_seconds(void)
|
| 1297 | {
|
| 1298 | struct timekeeper *tk = &timekeeper;
|
| 1299 |
|
| 1300 | return tk->xtime_sec;
|
| 1301 | }
|
| 1302 | EXPORT_SYMBOL(get_seconds);
|
| 1303 |
|
| 1304 | struct timespec __current_kernel_time(void)
|
| 1305 | {
|
| 1306 | struct timekeeper *tk = &timekeeper;
|
| 1307 |
|
| 1308 | return tk_xtime(tk);
|
| 1309 | }
|
| 1310 |
|
| 1311 | struct timespec current_kernel_time(void)
|
| 1312 | {
|
| 1313 | struct timekeeper *tk = &timekeeper;
|
| 1314 | struct timespec now;
|
| 1315 | unsigned long seq;
|
| 1316 |
|
| 1317 | do {
|
| 1318 | seq = read_seqbegin(&tk->lock);
|
| 1319 |
|
| 1320 | now = tk_xtime(tk);
|
| 1321 | } while (read_seqretry(&tk->lock, seq));
|
| 1322 |
|
| 1323 | return now;
|
| 1324 | }
|
| 1325 | EXPORT_SYMBOL(current_kernel_time);
|
| 1326 |
|
| 1327 | struct timespec get_monotonic_coarse(void)
|
| 1328 | {
|
| 1329 | struct timekeeper *tk = &timekeeper;
|
| 1330 | struct timespec now, mono;
|
| 1331 | unsigned long seq;
|
| 1332 |
|
| 1333 | do {
|
| 1334 | seq = read_seqbegin(&tk->lock);
|
| 1335 |
|
| 1336 | now = tk_xtime(tk);
|
| 1337 | mono = tk->wall_to_monotonic;
|
| 1338 | } while (read_seqretry(&tk->lock, seq));
|
| 1339 |
|
| 1340 | set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
|
| 1341 | now.tv_nsec + mono.tv_nsec);
|
| 1342 | return now;
|
| 1343 | }
|
| 1344 |
|
| 1345 | |
| 1346 | * Must hold jiffies_lock
|
| 1347 |
|
| 1348 | void do_timer(unsigned long ticks)
|
| 1349 | {
|
| 1350 | jiffies_64 += ticks;
|
| 1351 | update_wall_time();
|
| 1352 | calc_global_load(ticks);
|
| 1353 | }
|
| 1354 |
|
| 1355 | |
| 1356 | * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
|
| 1357 | * and sleep offsets.
|
| 1358 | * @xtim: pointer to timespec to be set with xtime
|
| 1359 | * @wtom: pointer to timespec to be set with wall_to_monotonic
|
| 1360 | * @sleep: pointer to timespec to be set with time in suspend
|
| 1361 |
|
| 1362 | void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
|
| 1363 | struct timespec *wtom, struct timespec *sleep)
|
| 1364 | {
|
| 1365 | struct timekeeper *tk = &timekeeper;
|
| 1366 | unsigned long seq;
|
| 1367 |
|
| 1368 | do {
|
| 1369 | seq = read_seqbegin(&tk->lock);
|
| 1370 | *xtim = tk_xtime(tk);
|
| 1371 | *wtom = tk->wall_to_monotonic;
|
| 1372 | *sleep = tk->total_sleep_time;
|
| 1373 | } while (read_seqretry(&tk->lock, seq));
|
| 1374 | }
|
| 1375 |
|
| 1376 | #ifdef CONFIG_HIGH_RES_TIMERS
|
| 1377 | |
| 1378 | * ktime_get_update_offsets - hrtimer helper
|
| 1379 | * @offs_real: pointer to storage for monotonic -> realtime offset
|
| 1380 | * @offs_boot: pointer to storage for monotonic -> boottime offset
|
| 1381 | *
|
| 1382 | * Returns current monotonic time and updates the offsets
|
| 1383 | * Called from hrtimer_interupt() or retrigger_next_event()
|
| 1384 |
|
| 1385 | ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot)
|
| 1386 | {
|
| 1387 | struct timekeeper *tk = &timekeeper;
|
| 1388 | ktime_t now;
|
| 1389 | unsigned int seq;
|
| 1390 | u64 secs, nsecs;
|
| 1391 |
|
| 1392 | do {
|
| 1393 | seq = read_seqbegin(&tk->lock);
|
| 1394 |
|
| 1395 | secs = tk->xtime_sec;
|
| 1396 | nsecs = timekeeping_get_ns(tk);
|
| 1397 |
|
| 1398 | *offs_real = tk->offs_real;
|
| 1399 | *offs_boot = tk->offs_boot;
|
| 1400 | } while (read_seqretry(&tk->lock, seq));
|
| 1401 |
|
| 1402 | now = ktime_add_ns(ktime_set(secs, 0), nsecs);
|
| 1403 | now = ktime_sub(now, *offs_real);
|
| 1404 | return now;
|
| 1405 | }
|
| 1406 | #endif
|
| 1407 |
|
| 1408 | |
| 1409 | * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
|
| 1410 |
|
| 1411 | ktime_t ktime_get_monotonic_offset(void)
|
| 1412 | {
|
| 1413 | struct timekeeper *tk = &timekeeper;
|
| 1414 | unsigned long seq;
|
| 1415 | struct timespec wtom;
|
| 1416 |
|
| 1417 | do {
|
| 1418 | seq = read_seqbegin(&tk->lock);
|
| 1419 | wtom = tk->wall_to_monotonic;
|
| 1420 | } while (read_seqretry(&tk->lock, seq));
|
| 1421 |
|
| 1422 | return timespec_to_ktime(wtom);
|
| 1423 | }
|
| 1424 | EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
|
| 1425 |
|
| 1426 | |
| 1427 | * xtime_update() - advances the timekeeping infrastructure
|
| 1428 | * @ticks: number of ticks, that have elapsed since the last call.
|
| 1429 | *
|
| 1430 | * Must be called with interrupts disabled.
|
| 1431 |
|
| 1432 | void xtime_update(unsigned long ticks)
|
| 1433 | {
|
| 1434 | write_seqlock(&jiffies_lock);
|
| 1435 | do_timer(ticks);
|
| 1436 | write_sequnlock(&jiffies_lock);
|
| 1437 | }
|