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 #include <linux/compiler.h> 26 27 #include "tick-internal.h" 28 #include "ntp_internal.h" 29 #include "timekeeping_internal.h" 30 31 #define TK_CLEAR_NTP (1 << 0) 32 #define TK_MIRROR (1 << 1) 33 #define TK_CLOCK_WAS_SET (1 << 2) 34 35 static struct timekeeper timekeeper; 36 static DEFINE_RAW_SPINLOCK(timekeeper_lock); 37 static seqcount_t timekeeper_seq; 38 static struct timekeeper shadow_timekeeper; 39 40 /* flag for if timekeeping is suspended */ 41 int __read_mostly timekeeping_suspended; 42 43 /* Flag for if there is a persistent clock on this platform */ 44 bool __read_mostly persistent_clock_exist = false; 45 46 static inline void tk_normalize_xtime(struct timekeeper *tk) 47 { 48 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) { 49 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift; 50 tk->xtime_sec++; 51 } 52 } 53 54 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts) 55 { 56 tk->xtime_sec = ts->tv_sec; 57 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift; 58 } 59 60 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts) 61 { 62 tk->xtime_sec += ts->tv_sec; 63 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift; 64 tk_normalize_xtime(tk); 65 } 66 67 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm) 68 { 69 struct timespec tmp; 70 71 /* 72 * Verify consistency of: offset_real = -wall_to_monotonic 73 * before modifying anything 74 */ 75 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec, 76 -tk->wall_to_monotonic.tv_nsec); 77 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64); 78 tk->wall_to_monotonic = wtm; 79 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec); 80 tk->offs_real = timespec_to_ktime(tmp); 81 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0)); 82 } 83 84 static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t) 85 { 86 /* Verify consistency before modifying */ 87 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64); 88 89 tk->total_sleep_time = t; 90 tk->offs_boot = timespec_to_ktime(t); 91 } 92 93 /** 94 * tk_setup_internals - Set up internals to use clocksource clock. 95 * 96 * @tk: The target timekeeper to setup. 97 * @clock: Pointer to clocksource. 98 * 99 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment 100 * pair and interval request. 101 * 102 * Unless you're the timekeeping code, you should not be using this! 103 */ 104 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock) 105 { 106 cycle_t interval; 107 u64 tmp, ntpinterval; 108 struct clocksource *old_clock; 109 110 old_clock = tk->clock; 111 tk->clock = clock; 112 tk->cycle_last = clock->cycle_last = clock->read(clock); 113 114 /* Do the ns -> cycle conversion first, using original mult */ 115 tmp = NTP_INTERVAL_LENGTH; 116 tmp <<= clock->shift; 117 ntpinterval = tmp; 118 tmp += clock->mult/2; 119 do_div(tmp, clock->mult); 120 if (tmp == 0) 121 tmp = 1; 122 123 interval = (cycle_t) tmp; 124 tk->cycle_interval = interval; 125 126 /* Go back from cycles -> shifted ns */ 127 tk->xtime_interval = (u64) interval * clock->mult; 128 tk->xtime_remainder = ntpinterval - tk->xtime_interval; 129 tk->raw_interval = 130 ((u64) interval * clock->mult) >> clock->shift; 131 132 /* if changing clocks, convert xtime_nsec shift units */ 133 if (old_clock) { 134 int shift_change = clock->shift - old_clock->shift; 135 if (shift_change < 0) 136 tk->xtime_nsec >>= -shift_change; 137 else 138 tk->xtime_nsec <<= shift_change; 139 } 140 tk->shift = clock->shift; 141 142 tk->ntp_error = 0; 143 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift; 144 145 /* 146 * The timekeeper keeps its own mult values for the currently 147 * active clocksource. These value will be adjusted via NTP 148 * to counteract clock drifting. 149 */ 150 tk->mult = clock->mult; 151 } 152 153 /* Timekeeper helper functions. */ 154 155 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET 156 u32 (*arch_gettimeoffset)(void); 157 158 u32 get_arch_timeoffset(void) 159 { 160 if (likely(arch_gettimeoffset)) 161 return arch_gettimeoffset(); 162 return 0; 163 } 164 #else 165 static inline u32 get_arch_timeoffset(void) { return 0; } 166 #endif 167 168 static inline s64 timekeeping_get_ns(struct timekeeper *tk) 169 { 170 cycle_t cycle_now, cycle_delta; 171 struct clocksource *clock; 172 s64 nsec; 173 174 /* read clocksource: */ 175 clock = tk->clock; 176 cycle_now = clock->read(clock); 177 178 /* calculate the delta since the last update_wall_time: */ 179 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; 180 181 nsec = cycle_delta * tk->mult + tk->xtime_nsec; 182 nsec >>= tk->shift; 183 184 /* If arch requires, add in get_arch_timeoffset() */ 185 return nsec + get_arch_timeoffset(); 186 } 187 188 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk) 189 { 190 cycle_t cycle_now, cycle_delta; 191 struct clocksource *clock; 192 s64 nsec; 193 194 /* read clocksource: */ 195 clock = tk->clock; 196 cycle_now = clock->read(clock); 197 198 /* calculate the delta since the last update_wall_time: */ 199 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; 200 201 /* convert delta to nanoseconds. */ 202 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); 203 204 /* If arch requires, add in get_arch_timeoffset() */ 205 return nsec + get_arch_timeoffset(); 206 } 207 208 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain); 209 210 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set) 211 { 212 raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk); 213 } 214 215 /** 216 * pvclock_gtod_register_notifier - register a pvclock timedata update listener 217 */ 218 int pvclock_gtod_register_notifier(struct notifier_block *nb) 219 { 220 struct timekeeper *tk = &timekeeper; 221 unsigned long flags; 222 int ret; 223 224 raw_spin_lock_irqsave(&timekeeper_lock, flags); 225 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb); 226 update_pvclock_gtod(tk, true); 227 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 228 229 return ret; 230 } 231 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier); 232 233 /** 234 * pvclock_gtod_unregister_notifier - unregister a pvclock 235 * timedata update listener 236 */ 237 int pvclock_gtod_unregister_notifier(struct notifier_block *nb) 238 { 239 unsigned long flags; 240 int ret; 241 242 raw_spin_lock_irqsave(&timekeeper_lock, flags); 243 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb); 244 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 245 246 return ret; 247 } 248 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier); 249 250 /* must hold timekeeper_lock */ 251 static void timekeeping_update(struct timekeeper *tk, unsigned int action) 252 { 253 if (action & TK_CLEAR_NTP) { 254 tk->ntp_error = 0; 255 ntp_clear(); 256 } 257 update_vsyscall(tk); 258 update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET); 259 260 if (action & TK_MIRROR) 261 memcpy(&shadow_timekeeper, &timekeeper, sizeof(timekeeper)); 262 } 263 264 /** 265 * timekeeping_forward_now - update clock to the current time 266 * 267 * Forward the current clock to update its state since the last call to 268 * update_wall_time(). This is useful before significant clock changes, 269 * as it avoids having to deal with this time offset explicitly. 270 */ 271 static void timekeeping_forward_now(struct timekeeper *tk) 272 { 273 cycle_t cycle_now, cycle_delta; 274 struct clocksource *clock; 275 s64 nsec; 276 277 clock = tk->clock; 278 cycle_now = clock->read(clock); 279 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; 280 tk->cycle_last = clock->cycle_last = cycle_now; 281 282 tk->xtime_nsec += cycle_delta * tk->mult; 283 284 /* If arch requires, add in get_arch_timeoffset() */ 285 tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift; 286 287 tk_normalize_xtime(tk); 288 289 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift); 290 timespec_add_ns(&tk->raw_time, nsec); 291 } 292 293 /** 294 * __getnstimeofday - Returns the time of day in a timespec. 295 * @ts: pointer to the timespec to be set 296 * 297 * Updates the time of day in the timespec. 298 * Returns 0 on success, or -ve when suspended (timespec will be undefined). 299 */ 300 int __getnstimeofday(struct timespec *ts) 301 { 302 struct timekeeper *tk = &timekeeper; 303 unsigned long seq; 304 s64 nsecs = 0; 305 306 do { 307 seq = read_seqcount_begin(&timekeeper_seq); 308 309 ts->tv_sec = tk->xtime_sec; 310 nsecs = timekeeping_get_ns(tk); 311 312 } while (read_seqcount_retry(&timekeeper_seq, seq)); 313 314 ts->tv_nsec = 0; 315 timespec_add_ns(ts, nsecs); 316 317 /* 318 * Do not bail out early, in case there were callers still using 319 * the value, even in the face of the WARN_ON. 320 */ 321 if (unlikely(timekeeping_suspended)) 322 return -EAGAIN; 323 return 0; 324 } 325 EXPORT_SYMBOL(__getnstimeofday); 326 327 /** 328 * getnstimeofday - Returns the time of day in a timespec. 329 * @ts: pointer to the timespec to be set 330 * 331 * Returns the time of day in a timespec (WARN if suspended). 332 */ 333 void getnstimeofday(struct timespec *ts) 334 { 335 WARN_ON(__getnstimeofday(ts)); 336 } 337 EXPORT_SYMBOL(getnstimeofday); 338 339 ktime_t ktime_get(void) 340 { 341 struct timekeeper *tk = &timekeeper; 342 unsigned int seq; 343 s64 secs, nsecs; 344 345 WARN_ON(timekeeping_suspended); 346 347 do { 348 seq = read_seqcount_begin(&timekeeper_seq); 349 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec; 350 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec; 351 352 } while (read_seqcount_retry(&timekeeper_seq, seq)); 353 /* 354 * Use ktime_set/ktime_add_ns to create a proper ktime on 355 * 32-bit architectures without CONFIG_KTIME_SCALAR. 356 */ 357 return ktime_add_ns(ktime_set(secs, 0), nsecs); 358 } 359 EXPORT_SYMBOL_GPL(ktime_get); 360 361 /** 362 * ktime_get_ts - get the monotonic clock in timespec format 363 * @ts: pointer to timespec variable 364 * 365 * The function calculates the monotonic clock from the realtime 366 * clock and the wall_to_monotonic offset and stores the result 367 * in normalized timespec format in the variable pointed to by @ts. 368 */ 369 void ktime_get_ts(struct timespec *ts) 370 { 371 struct timekeeper *tk = &timekeeper; 372 struct timespec tomono; 373 s64 nsec; 374 unsigned int seq; 375 376 WARN_ON(timekeeping_suspended); 377 378 do { 379 seq = read_seqcount_begin(&timekeeper_seq); 380 ts->tv_sec = tk->xtime_sec; 381 nsec = timekeeping_get_ns(tk); 382 tomono = tk->wall_to_monotonic; 383 384 } while (read_seqcount_retry(&timekeeper_seq, seq)); 385 386 ts->tv_sec += tomono.tv_sec; 387 ts->tv_nsec = 0; 388 timespec_add_ns(ts, nsec + tomono.tv_nsec); 389 } 390 EXPORT_SYMBOL_GPL(ktime_get_ts); 391 392 393 /** 394 * timekeeping_clocktai - Returns the TAI time of day in a timespec 395 * @ts: pointer to the timespec to be set 396 * 397 * Returns the time of day in a timespec. 398 */ 399 void timekeeping_clocktai(struct timespec *ts) 400 { 401 struct timekeeper *tk = &timekeeper; 402 unsigned long seq; 403 u64 nsecs; 404 405 WARN_ON(timekeeping_suspended); 406 407 do { 408 seq = read_seqcount_begin(&timekeeper_seq); 409 410 ts->tv_sec = tk->xtime_sec + tk->tai_offset; 411 nsecs = timekeeping_get_ns(tk); 412 413 } while (read_seqcount_retry(&timekeeper_seq, seq)); 414 415 ts->tv_nsec = 0; 416 timespec_add_ns(ts, nsecs); 417 418 } 419 EXPORT_SYMBOL(timekeeping_clocktai); 420 421 422 /** 423 * ktime_get_clocktai - Returns the TAI time of day in a ktime 424 * 425 * Returns the time of day in a ktime. 426 */ 427 ktime_t ktime_get_clocktai(void) 428 { 429 struct timespec ts; 430 431 timekeeping_clocktai(&ts); 432 return timespec_to_ktime(ts); 433 } 434 EXPORT_SYMBOL(ktime_get_clocktai); 435 436 #ifdef CONFIG_NTP_PPS 437 438 /** 439 * getnstime_raw_and_real - get day and raw monotonic time in timespec format 440 * @ts_raw: pointer to the timespec to be set to raw monotonic time 441 * @ts_real: pointer to the timespec to be set to the time of day 442 * 443 * This function reads both the time of day and raw monotonic time at the 444 * same time atomically and stores the resulting timestamps in timespec 445 * format. 446 */ 447 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real) 448 { 449 struct timekeeper *tk = &timekeeper; 450 unsigned long seq; 451 s64 nsecs_raw, nsecs_real; 452 453 WARN_ON_ONCE(timekeeping_suspended); 454 455 do { 456 seq = read_seqcount_begin(&timekeeper_seq); 457 458 *ts_raw = tk->raw_time; 459 ts_real->tv_sec = tk->xtime_sec; 460 ts_real->tv_nsec = 0; 461 462 nsecs_raw = timekeeping_get_ns_raw(tk); 463 nsecs_real = timekeeping_get_ns(tk); 464 465 } while (read_seqcount_retry(&timekeeper_seq, seq)); 466 467 timespec_add_ns(ts_raw, nsecs_raw); 468 timespec_add_ns(ts_real, nsecs_real); 469 } 470 EXPORT_SYMBOL(getnstime_raw_and_real); 471 472 #endif /* CONFIG_NTP_PPS */ 473 474 /** 475 * do_gettimeofday - Returns the time of day in a timeval 476 * @tv: pointer to the timeval to be set 477 * 478 * NOTE: Users should be converted to using getnstimeofday() 479 */ 480 void do_gettimeofday(struct timeval *tv) 481 { 482 struct timespec now; 483 484 getnstimeofday(&now); 485 tv->tv_sec = now.tv_sec; 486 tv->tv_usec = now.tv_nsec/1000; 487 } 488 EXPORT_SYMBOL(do_gettimeofday); 489 490 /** 491 * do_settimeofday - Sets the time of day 492 * @tv: pointer to the timespec variable containing the new time 493 * 494 * Sets the time of day to the new time and update NTP and notify hrtimers 495 */ 496 int do_settimeofday(const struct timespec *tv) 497 { 498 struct timekeeper *tk = &timekeeper; 499 struct timespec ts_delta, xt; 500 unsigned long flags; 501 502 if (!timespec_valid_strict(tv)) 503 return -EINVAL; 504 505 raw_spin_lock_irqsave(&timekeeper_lock, flags); 506 write_seqcount_begin(&timekeeper_seq); 507 508 timekeeping_forward_now(tk); 509 510 xt = tk_xtime(tk); 511 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec; 512 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec; 513 514 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta)); 515 516 tk_set_xtime(tk, tv); 517 518 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); 519 520 write_seqcount_end(&timekeeper_seq); 521 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 522 523 /* signal hrtimers about time change */ 524 clock_was_set(); 525 526 return 0; 527 } 528 EXPORT_SYMBOL(do_settimeofday); 529 530 /** 531 * timekeeping_inject_offset - Adds or subtracts from the current time. 532 * @tv: pointer to the timespec variable containing the offset 533 * 534 * Adds or subtracts an offset value from the current time. 535 */ 536 int timekeeping_inject_offset(struct timespec *ts) 537 { 538 struct timekeeper *tk = &timekeeper; 539 unsigned long flags; 540 struct timespec tmp; 541 int ret = 0; 542 543 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC) 544 return -EINVAL; 545 546 raw_spin_lock_irqsave(&timekeeper_lock, flags); 547 write_seqcount_begin(&timekeeper_seq); 548 549 timekeeping_forward_now(tk); 550 551 /* Make sure the proposed value is valid */ 552 tmp = timespec_add(tk_xtime(tk), *ts); 553 if (!timespec_valid_strict(&tmp)) { 554 ret = -EINVAL; 555 goto error; 556 } 557 558 tk_xtime_add(tk, ts); 559 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts)); 560 561 error: /* even if we error out, we forwarded the time, so call update */ 562 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); 563 564 write_seqcount_end(&timekeeper_seq); 565 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 566 567 /* signal hrtimers about time change */ 568 clock_was_set(); 569 570 return ret; 571 } 572 EXPORT_SYMBOL(timekeeping_inject_offset); 573 574 575 /** 576 * timekeeping_get_tai_offset - Returns current TAI offset from UTC 577 * 578 */ 579 s32 timekeeping_get_tai_offset(void) 580 { 581 struct timekeeper *tk = &timekeeper; 582 unsigned int seq; 583 s32 ret; 584 585 do { 586 seq = read_seqcount_begin(&timekeeper_seq); 587 ret = tk->tai_offset; 588 } while (read_seqcount_retry(&timekeeper_seq, seq)); 589 590 return ret; 591 } 592 593 /** 594 * __timekeeping_set_tai_offset - Lock free worker function 595 * 596 */ 597 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset) 598 { 599 tk->tai_offset = tai_offset; 600 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0)); 601 } 602 603 /** 604 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC 605 * 606 */ 607 void timekeeping_set_tai_offset(s32 tai_offset) 608 { 609 struct timekeeper *tk = &timekeeper; 610 unsigned long flags; 611 612 raw_spin_lock_irqsave(&timekeeper_lock, flags); 613 write_seqcount_begin(&timekeeper_seq); 614 __timekeeping_set_tai_offset(tk, tai_offset); 615 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); 616 write_seqcount_end(&timekeeper_seq); 617 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 618 clock_was_set(); 619 } 620 621 /** 622 * change_clocksource - Swaps clocksources if a new one is available 623 * 624 * Accumulates current time interval and initializes new clocksource 625 */ 626 static int change_clocksource(void *data) 627 { 628 struct timekeeper *tk = &timekeeper; 629 struct clocksource *new, *old; 630 unsigned long flags; 631 632 new = (struct clocksource *) data; 633 634 raw_spin_lock_irqsave(&timekeeper_lock, flags); 635 write_seqcount_begin(&timekeeper_seq); 636 637 timekeeping_forward_now(tk); 638 /* 639 * If the cs is in module, get a module reference. Succeeds 640 * for built-in code (owner == NULL) as well. 641 */ 642 if (try_module_get(new->owner)) { 643 if (!new->enable || new->enable(new) == 0) { 644 old = tk->clock; 645 tk_setup_internals(tk, new); 646 if (old->disable) 647 old->disable(old); 648 module_put(old->owner); 649 } else { 650 module_put(new->owner); 651 } 652 } 653 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); 654 655 write_seqcount_end(&timekeeper_seq); 656 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 657 658 return 0; 659 } 660 661 /** 662 * timekeeping_notify - Install a new clock source 663 * @clock: pointer to the clock source 664 * 665 * This function is called from clocksource.c after a new, better clock 666 * source has been registered. The caller holds the clocksource_mutex. 667 */ 668 int timekeeping_notify(struct clocksource *clock) 669 { 670 struct timekeeper *tk = &timekeeper; 671 672 if (tk->clock == clock) 673 return 0; 674 stop_machine(change_clocksource, clock, NULL); 675 tick_clock_notify(); 676 return tk->clock == clock ? 0 : -1; 677 } 678 679 /** 680 * ktime_get_real - get the real (wall-) time in ktime_t format 681 * 682 * returns the time in ktime_t format 683 */ 684 ktime_t ktime_get_real(void) 685 { 686 struct timespec now; 687 688 getnstimeofday(&now); 689 690 return timespec_to_ktime(now); 691 } 692 EXPORT_SYMBOL_GPL(ktime_get_real); 693 694 /** 695 * getrawmonotonic - Returns the raw monotonic time in a timespec 696 * @ts: pointer to the timespec to be set 697 * 698 * Returns the raw monotonic time (completely un-modified by ntp) 699 */ 700 void getrawmonotonic(struct timespec *ts) 701 { 702 struct timekeeper *tk = &timekeeper; 703 unsigned long seq; 704 s64 nsecs; 705 706 do { 707 seq = read_seqcount_begin(&timekeeper_seq); 708 nsecs = timekeeping_get_ns_raw(tk); 709 *ts = tk->raw_time; 710 711 } while (read_seqcount_retry(&timekeeper_seq, seq)); 712 713 timespec_add_ns(ts, nsecs); 714 } 715 EXPORT_SYMBOL(getrawmonotonic); 716 717 /** 718 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres 719 */ 720 int timekeeping_valid_for_hres(void) 721 { 722 struct timekeeper *tk = &timekeeper; 723 unsigned long seq; 724 int ret; 725 726 do { 727 seq = read_seqcount_begin(&timekeeper_seq); 728 729 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; 730 731 } while (read_seqcount_retry(&timekeeper_seq, seq)); 732 733 return ret; 734 } 735 736 /** 737 * timekeeping_max_deferment - Returns max time the clocksource can be deferred 738 */ 739 u64 timekeeping_max_deferment(void) 740 { 741 struct timekeeper *tk = &timekeeper; 742 unsigned long seq; 743 u64 ret; 744 745 do { 746 seq = read_seqcount_begin(&timekeeper_seq); 747 748 ret = tk->clock->max_idle_ns; 749 750 } while (read_seqcount_retry(&timekeeper_seq, seq)); 751 752 return ret; 753 } 754 755 /** 756 * read_persistent_clock - Return time from the persistent clock. 757 * 758 * Weak dummy function for arches that do not yet support it. 759 * Reads the time from the battery backed persistent clock. 760 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. 761 * 762 * XXX - Do be sure to remove it once all arches implement it. 763 */ 764 void __weak read_persistent_clock(struct timespec *ts) 765 { 766 ts->tv_sec = 0; 767 ts->tv_nsec = 0; 768 } 769 770 /** 771 * read_boot_clock - Return time of the system start. 772 * 773 * Weak dummy function for arches that do not yet support it. 774 * Function to read the exact time the system has been started. 775 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported. 776 * 777 * XXX - Do be sure to remove it once all arches implement it. 778 */ 779 void __weak read_boot_clock(struct timespec *ts) 780 { 781 ts->tv_sec = 0; 782 ts->tv_nsec = 0; 783 } 784 785 /* 786 * timekeeping_init - Initializes the clocksource and common timekeeping values 787 */ 788 void __init timekeeping_init(void) 789 { 790 struct timekeeper *tk = &timekeeper; 791 struct clocksource *clock; 792 unsigned long flags; 793 struct timespec now, boot, tmp; 794 795 read_persistent_clock(&now); 796 797 if (!timespec_valid_strict(&now)) { 798 pr_warn("WARNING: Persistent clock returned invalid value!\n" 799 " Check your CMOS/BIOS settings.\n"); 800 now.tv_sec = 0; 801 now.tv_nsec = 0; 802 } else if (now.tv_sec || now.tv_nsec) 803 persistent_clock_exist = true; 804 805 read_boot_clock(&boot); 806 if (!timespec_valid_strict(&boot)) { 807 pr_warn("WARNING: Boot clock returned invalid value!\n" 808 " Check your CMOS/BIOS settings.\n"); 809 boot.tv_sec = 0; 810 boot.tv_nsec = 0; 811 } 812 813 raw_spin_lock_irqsave(&timekeeper_lock, flags); 814 write_seqcount_begin(&timekeeper_seq); 815 ntp_init(); 816 817 clock = clocksource_default_clock(); 818 if (clock->enable) 819 clock->enable(clock); 820 tk_setup_internals(tk, clock); 821 822 tk_set_xtime(tk, &now); 823 tk->raw_time.tv_sec = 0; 824 tk->raw_time.tv_nsec = 0; 825 if (boot.tv_sec == 0 && boot.tv_nsec == 0) 826 boot = tk_xtime(tk); 827 828 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec); 829 tk_set_wall_to_mono(tk, tmp); 830 831 tmp.tv_sec = 0; 832 tmp.tv_nsec = 0; 833 tk_set_sleep_time(tk, tmp); 834 835 memcpy(&shadow_timekeeper, &timekeeper, sizeof(timekeeper)); 836 837 write_seqcount_end(&timekeeper_seq); 838 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 839 } 840 841 /* time in seconds when suspend began */ 842 static struct timespec timekeeping_suspend_time; 843 844 /** 845 * __timekeeping_inject_sleeptime - Internal function to add sleep interval 846 * @delta: pointer to a timespec delta value 847 * 848 * Takes a timespec offset measuring a suspend interval and properly 849 * adds the sleep offset to the timekeeping variables. 850 */ 851 static void __timekeeping_inject_sleeptime(struct timekeeper *tk, 852 struct timespec *delta) 853 { 854 if (!timespec_valid_strict(delta)) { 855 printk_deferred(KERN_WARNING 856 "__timekeeping_inject_sleeptime: Invalid " 857 "sleep delta value!\n"); 858 return; 859 } 860 tk_xtime_add(tk, delta); 861 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta)); 862 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta)); 863 tk_debug_account_sleep_time(delta); 864 } 865 866 /** 867 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values 868 * @delta: pointer to a timespec delta value 869 * 870 * This hook is for architectures that cannot support read_persistent_clock 871 * because their RTC/persistent clock is only accessible when irqs are enabled. 872 * 873 * This function should only be called by rtc_resume(), and allows 874 * a suspend offset to be injected into the timekeeping values. 875 */ 876 void timekeeping_inject_sleeptime(struct timespec *delta) 877 { 878 struct timekeeper *tk = &timekeeper; 879 unsigned long flags; 880 881 /* 882 * Make sure we don't set the clock twice, as timekeeping_resume() 883 * already did it 884 */ 885 if (has_persistent_clock()) 886 return; 887 888 raw_spin_lock_irqsave(&timekeeper_lock, flags); 889 write_seqcount_begin(&timekeeper_seq); 890 891 timekeeping_forward_now(tk); 892 893 __timekeeping_inject_sleeptime(tk, delta); 894 895 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET); 896 897 write_seqcount_end(&timekeeper_seq); 898 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 899 900 /* signal hrtimers about time change */ 901 clock_was_set(); 902 } 903 904 /** 905 * timekeeping_resume - Resumes the generic timekeeping subsystem. 906 * 907 * This is for the generic clocksource timekeeping. 908 * xtime/wall_to_monotonic/jiffies/etc are 909 * still managed by arch specific suspend/resume code. 910 */ 911 static void timekeeping_resume(void) 912 { 913 struct timekeeper *tk = &timekeeper; 914 struct clocksource *clock = tk->clock; 915 unsigned long flags; 916 struct timespec ts_new, ts_delta; 917 cycle_t cycle_now, cycle_delta; 918 bool suspendtime_found = false; 919 920 read_persistent_clock(&ts_new); 921 922 clockevents_resume(); 923 clocksource_resume(); 924 925 raw_spin_lock_irqsave(&timekeeper_lock, flags); 926 write_seqcount_begin(&timekeeper_seq); 927 928 /* 929 * After system resumes, we need to calculate the suspended time and 930 * compensate it for the OS time. There are 3 sources that could be 931 * used: Nonstop clocksource during suspend, persistent clock and rtc 932 * device. 933 * 934 * One specific platform may have 1 or 2 or all of them, and the 935 * preference will be: 936 * suspend-nonstop clocksource -> persistent clock -> rtc 937 * The less preferred source will only be tried if there is no better 938 * usable source. The rtc part is handled separately in rtc core code. 939 */ 940 cycle_now = clock->read(clock); 941 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) && 942 cycle_now > clock->cycle_last) { 943 u64 num, max = ULLONG_MAX; 944 u32 mult = clock->mult; 945 u32 shift = clock->shift; 946 s64 nsec = 0; 947 948 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; 949 950 /* 951 * "cycle_delta * mutl" may cause 64 bits overflow, if the 952 * suspended time is too long. In that case we need do the 953 * 64 bits math carefully 954 */ 955 do_div(max, mult); 956 if (cycle_delta > max) { 957 num = div64_u64(cycle_delta, max); 958 nsec = (((u64) max * mult) >> shift) * num; 959 cycle_delta -= num * max; 960 } 961 nsec += ((u64) cycle_delta * mult) >> shift; 962 963 ts_delta = ns_to_timespec(nsec); 964 suspendtime_found = true; 965 } else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) { 966 ts_delta = timespec_sub(ts_new, timekeeping_suspend_time); 967 suspendtime_found = true; 968 } 969 970 if (suspendtime_found) 971 __timekeeping_inject_sleeptime(tk, &ts_delta); 972 973 /* Re-base the last cycle value */ 974 tk->cycle_last = clock->cycle_last = cycle_now; 975 tk->ntp_error = 0; 976 timekeeping_suspended = 0; 977 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); 978 write_seqcount_end(&timekeeper_seq); 979 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 980 981 touch_softlockup_watchdog(); 982 983 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); 984 985 /* Resume hrtimers */ 986 hrtimers_resume(); 987 } 988 989 static int timekeeping_suspend(void) 990 { 991 struct timekeeper *tk = &timekeeper; 992 unsigned long flags; 993 struct timespec delta, delta_delta; 994 static struct timespec old_delta; 995 996 read_persistent_clock(&timekeeping_suspend_time); 997 998 /* 999 * On some systems the persistent_clock can not be detected at 1000 * timekeeping_init by its return value, so if we see a valid 1001 * value returned, update the persistent_clock_exists flag. 1002 */ 1003 if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec) 1004 persistent_clock_exist = true; 1005 1006 raw_spin_lock_irqsave(&timekeeper_lock, flags); 1007 write_seqcount_begin(&timekeeper_seq); 1008 timekeeping_forward_now(tk); 1009 timekeeping_suspended = 1; 1010 1011 /* 1012 * To avoid drift caused by repeated suspend/resumes, 1013 * which each can add ~1 second drift error, 1014 * try to compensate so the difference in system time 1015 * and persistent_clock time stays close to constant. 1016 */ 1017 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time); 1018 delta_delta = timespec_sub(delta, old_delta); 1019 if (abs(delta_delta.tv_sec) >= 2) { 1020 /* 1021 * if delta_delta is too large, assume time correction 1022 * has occured and set old_delta to the current delta. 1023 */ 1024 old_delta = delta; 1025 } else { 1026 /* Otherwise try to adjust old_system to compensate */ 1027 timekeeping_suspend_time = 1028 timespec_add(timekeeping_suspend_time, delta_delta); 1029 } 1030 1031 timekeeping_update(tk, TK_MIRROR); 1032 write_seqcount_end(&timekeeper_seq); 1033 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 1034 1035 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); 1036 clocksource_suspend(); 1037 clockevents_suspend(); 1038 1039 return 0; 1040 } 1041 1042 /* sysfs resume/suspend bits for timekeeping */ 1043 static struct syscore_ops timekeeping_syscore_ops = { 1044 .resume = timekeeping_resume, 1045 .suspend = timekeeping_suspend, 1046 }; 1047 1048 static int __init timekeeping_init_ops(void) 1049 { 1050 register_syscore_ops(&timekeeping_syscore_ops); 1051 return 0; 1052 } 1053 1054 device_initcall(timekeeping_init_ops); 1055 1056 /* 1057 * If the error is already larger, we look ahead even further 1058 * to compensate for late or lost adjustments. 1059 */ 1060 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk, 1061 s64 error, s64 *interval, 1062 s64 *offset) 1063 { 1064 s64 tick_error, i; 1065 u32 look_ahead, adj; 1066 s32 error2, mult; 1067 1068 /* 1069 * Use the current error value to determine how much to look ahead. 1070 * The larger the error the slower we adjust for it to avoid problems 1071 * with losing too many ticks, otherwise we would overadjust and 1072 * produce an even larger error. The smaller the adjustment the 1073 * faster we try to adjust for it, as lost ticks can do less harm 1074 * here. This is tuned so that an error of about 1 msec is adjusted 1075 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). 1076 */ 1077 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ); 1078 error2 = abs(error2); 1079 for (look_ahead = 0; error2 > 0; look_ahead++) 1080 error2 >>= 2; 1081 1082 /* 1083 * Now calculate the error in (1 << look_ahead) ticks, but first 1084 * remove the single look ahead already included in the error. 1085 */ 1086 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1); 1087 tick_error -= tk->xtime_interval >> 1; 1088 error = ((error - tick_error) >> look_ahead) + tick_error; 1089 1090 /* Finally calculate the adjustment shift value. */ 1091 i = *interval; 1092 mult = 1; 1093 if (error < 0) { 1094 error = -error; 1095 *interval = -*interval; 1096 *offset = -*offset; 1097 mult = -1; 1098 } 1099 for (adj = 0; error > i; adj++) 1100 error >>= 1; 1101 1102 *interval <<= adj; 1103 *offset <<= adj; 1104 return mult << adj; 1105 } 1106 1107 /* 1108 * Adjust the multiplier to reduce the error value, 1109 * this is optimized for the most common adjustments of -1,0,1, 1110 * for other values we can do a bit more work. 1111 */ 1112 static void timekeeping_adjust(struct timekeeper *tk, s64 offset) 1113 { 1114 s64 error, interval = tk->cycle_interval; 1115 int adj; 1116 1117 /* 1118 * The point of this is to check if the error is greater than half 1119 * an interval. 1120 * 1121 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs. 1122 * 1123 * Note we subtract one in the shift, so that error is really error*2. 1124 * This "saves" dividing(shifting) interval twice, but keeps the 1125 * (error > interval) comparison as still measuring if error is 1126 * larger than half an interval. 1127 * 1128 * Note: It does not "save" on aggravation when reading the code. 1129 */ 1130 error = tk->ntp_error >> (tk->ntp_error_shift - 1); 1131 if (error > interval) { 1132 /* 1133 * We now divide error by 4(via shift), which checks if 1134 * the error is greater than twice the interval. 1135 * If it is greater, we need a bigadjust, if its smaller, 1136 * we can adjust by 1. 1137 */ 1138 error >>= 2; 1139 if (likely(error <= interval)) 1140 adj = 1; 1141 else 1142 adj = timekeeping_bigadjust(tk, error, &interval, &offset); 1143 } else { 1144 if (error < -interval) { 1145 /* See comment above, this is just switched for the negative */ 1146 error >>= 2; 1147 if (likely(error >= -interval)) { 1148 adj = -1; 1149 interval = -interval; 1150 offset = -offset; 1151 } else { 1152 adj = timekeeping_bigadjust(tk, error, &interval, &offset); 1153 } 1154 } else { 1155 goto out_adjust; 1156 } 1157 } 1158 1159 if (unlikely(tk->clock->maxadj && 1160 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) { 1161 printk_deferred_once(KERN_WARNING 1162 "Adjusting %s more than 11%% (%ld vs %ld)\n", 1163 tk->clock->name, (long)tk->mult + adj, 1164 (long)tk->clock->mult + tk->clock->maxadj); 1165 } 1166 /* 1167 * So the following can be confusing. 1168 * 1169 * To keep things simple, lets assume adj == 1 for now. 1170 * 1171 * When adj != 1, remember that the interval and offset values 1172 * have been appropriately scaled so the math is the same. 1173 * 1174 * The basic idea here is that we're increasing the multiplier 1175 * by one, this causes the xtime_interval to be incremented by 1176 * one cycle_interval. This is because: 1177 * xtime_interval = cycle_interval * mult 1178 * So if mult is being incremented by one: 1179 * xtime_interval = cycle_interval * (mult + 1) 1180 * Its the same as: 1181 * xtime_interval = (cycle_interval * mult) + cycle_interval 1182 * Which can be shortened to: 1183 * xtime_interval += cycle_interval 1184 * 1185 * So offset stores the non-accumulated cycles. Thus the current 1186 * time (in shifted nanoseconds) is: 1187 * now = (offset * adj) + xtime_nsec 1188 * Now, even though we're adjusting the clock frequency, we have 1189 * to keep time consistent. In other words, we can't jump back 1190 * in time, and we also want to avoid jumping forward in time. 1191 * 1192 * So given the same offset value, we need the time to be the same 1193 * both before and after the freq adjustment. 1194 * now = (offset * adj_1) + xtime_nsec_1 1195 * now = (offset * adj_2) + xtime_nsec_2 1196 * So: 1197 * (offset * adj_1) + xtime_nsec_1 = 1198 * (offset * adj_2) + xtime_nsec_2 1199 * And we know: 1200 * adj_2 = adj_1 + 1 1201 * So: 1202 * (offset * adj_1) + xtime_nsec_1 = 1203 * (offset * (adj_1+1)) + xtime_nsec_2 1204 * (offset * adj_1) + xtime_nsec_1 = 1205 * (offset * adj_1) + offset + xtime_nsec_2 1206 * Canceling the sides: 1207 * xtime_nsec_1 = offset + xtime_nsec_2 1208 * Which gives us: 1209 * xtime_nsec_2 = xtime_nsec_1 - offset 1210 * Which simplfies to: 1211 * xtime_nsec -= offset 1212 * 1213 * XXX - TODO: Doc ntp_error calculation. 1214 */ 1215 tk->mult += adj; 1216 tk->xtime_interval += interval; 1217 tk->xtime_nsec -= offset; 1218 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift; 1219 1220 out_adjust: 1221 /* 1222 * It may be possible that when we entered this function, xtime_nsec 1223 * was very small. Further, if we're slightly speeding the clocksource 1224 * in the code above, its possible the required corrective factor to 1225 * xtime_nsec could cause it to underflow. 1226 * 1227 * Now, since we already accumulated the second, cannot simply roll 1228 * the accumulated second back, since the NTP subsystem has been 1229 * notified via second_overflow. So instead we push xtime_nsec forward 1230 * by the amount we underflowed, and add that amount into the error. 1231 * 1232 * We'll correct this error next time through this function, when 1233 * xtime_nsec is not as small. 1234 */ 1235 if (unlikely((s64)tk->xtime_nsec < 0)) { 1236 s64 neg = -(s64)tk->xtime_nsec; 1237 tk->xtime_nsec = 0; 1238 tk->ntp_error += neg << tk->ntp_error_shift; 1239 } 1240 1241 } 1242 1243 /** 1244 * accumulate_nsecs_to_secs - Accumulates nsecs into secs 1245 * 1246 * Helper function that accumulates a the nsecs greater then a second 1247 * from the xtime_nsec field to the xtime_secs field. 1248 * It also calls into the NTP code to handle leapsecond processing. 1249 * 1250 */ 1251 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk) 1252 { 1253 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift; 1254 unsigned int clock_set = 0; 1255 1256 while (tk->xtime_nsec >= nsecps) { 1257 int leap; 1258 1259 tk->xtime_nsec -= nsecps; 1260 tk->xtime_sec++; 1261 1262 /* Figure out if its a leap sec and apply if needed */ 1263 leap = second_overflow(tk->xtime_sec); 1264 if (unlikely(leap)) { 1265 struct timespec ts; 1266 1267 tk->xtime_sec += leap; 1268 1269 ts.tv_sec = leap; 1270 ts.tv_nsec = 0; 1271 tk_set_wall_to_mono(tk, 1272 timespec_sub(tk->wall_to_monotonic, ts)); 1273 1274 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap); 1275 1276 clock_set = TK_CLOCK_WAS_SET; 1277 } 1278 } 1279 return clock_set; 1280 } 1281 1282 /** 1283 * logarithmic_accumulation - shifted accumulation of cycles 1284 * 1285 * This functions accumulates a shifted interval of cycles into 1286 * into a shifted interval nanoseconds. Allows for O(log) accumulation 1287 * loop. 1288 * 1289 * Returns the unconsumed cycles. 1290 */ 1291 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset, 1292 u32 shift, 1293 unsigned int *clock_set) 1294 { 1295 cycle_t interval = tk->cycle_interval << shift; 1296 u64 raw_nsecs; 1297 1298 /* If the offset is smaller then a shifted interval, do nothing */ 1299 if (offset < interval) 1300 return offset; 1301 1302 /* Accumulate one shifted interval */ 1303 offset -= interval; 1304 tk->cycle_last += interval; 1305 1306 tk->xtime_nsec += tk->xtime_interval << shift; 1307 *clock_set |= accumulate_nsecs_to_secs(tk); 1308 1309 /* Accumulate raw time */ 1310 raw_nsecs = (u64)tk->raw_interval << shift; 1311 raw_nsecs += tk->raw_time.tv_nsec; 1312 if (raw_nsecs >= NSEC_PER_SEC) { 1313 u64 raw_secs = raw_nsecs; 1314 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC); 1315 tk->raw_time.tv_sec += raw_secs; 1316 } 1317 tk->raw_time.tv_nsec = raw_nsecs; 1318 1319 /* Accumulate error between NTP and clock interval */ 1320 tk->ntp_error += ntp_tick_length() << shift; 1321 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) << 1322 (tk->ntp_error_shift + shift); 1323 1324 return offset; 1325 } 1326 1327 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD 1328 static inline void old_vsyscall_fixup(struct timekeeper *tk) 1329 { 1330 s64 remainder; 1331 1332 /* 1333 * Store only full nanoseconds into xtime_nsec after rounding 1334 * it up and add the remainder to the error difference. 1335 * XXX - This is necessary to avoid small 1ns inconsistnecies caused 1336 * by truncating the remainder in vsyscalls. However, it causes 1337 * additional work to be done in timekeeping_adjust(). Once 1338 * the vsyscall implementations are converted to use xtime_nsec 1339 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD 1340 * users are removed, this can be killed. 1341 */ 1342 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1); 1343 tk->xtime_nsec -= remainder; 1344 tk->xtime_nsec += 1ULL << tk->shift; 1345 tk->ntp_error += remainder << tk->ntp_error_shift; 1346 tk->ntp_error -= (1ULL << tk->shift) << tk->ntp_error_shift; 1347 } 1348 #else 1349 #define old_vsyscall_fixup(tk) 1350 #endif 1351 1352 1353 1354 /** 1355 * update_wall_time - Uses the current clocksource to increment the wall time 1356 * 1357 */ 1358 void update_wall_time(void) 1359 { 1360 struct clocksource *clock; 1361 struct timekeeper *real_tk = &timekeeper; 1362 struct timekeeper *tk = &shadow_timekeeper; 1363 cycle_t offset; 1364 int shift = 0, maxshift; 1365 unsigned int clock_set = 0; 1366 unsigned long flags; 1367 1368 raw_spin_lock_irqsave(&timekeeper_lock, flags); 1369 1370 /* Make sure we're fully resumed: */ 1371 if (unlikely(timekeeping_suspended)) 1372 goto out; 1373 1374 clock = real_tk->clock; 1375 1376 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET 1377 offset = real_tk->cycle_interval; 1378 #else 1379 offset = (clock->read(clock) - clock->cycle_last) & clock->mask; 1380 #endif 1381 1382 /* Check if there's really nothing to do */ 1383 if (offset < real_tk->cycle_interval) 1384 goto out; 1385 1386 /* 1387 * With NO_HZ we may have to accumulate many cycle_intervals 1388 * (think "ticks") worth of time at once. To do this efficiently, 1389 * we calculate the largest doubling multiple of cycle_intervals 1390 * that is smaller than the offset. We then accumulate that 1391 * chunk in one go, and then try to consume the next smaller 1392 * doubled multiple. 1393 */ 1394 shift = ilog2(offset) - ilog2(tk->cycle_interval); 1395 shift = max(0, shift); 1396 /* Bound shift to one less than what overflows tick_length */ 1397 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1; 1398 shift = min(shift, maxshift); 1399 while (offset >= tk->cycle_interval) { 1400 offset = logarithmic_accumulation(tk, offset, shift, 1401 &clock_set); 1402 if (offset < tk->cycle_interval<<shift) 1403 shift--; 1404 } 1405 1406 /* correct the clock when NTP error is too big */ 1407 timekeeping_adjust(tk, offset); 1408 1409 /* 1410 * XXX This can be killed once everyone converts 1411 * to the new update_vsyscall. 1412 */ 1413 old_vsyscall_fixup(tk); 1414 1415 /* 1416 * Finally, make sure that after the rounding 1417 * xtime_nsec isn't larger than NSEC_PER_SEC 1418 */ 1419 clock_set |= accumulate_nsecs_to_secs(tk); 1420 1421 write_seqcount_begin(&timekeeper_seq); 1422 /* Update clock->cycle_last with the new value */ 1423 clock->cycle_last = tk->cycle_last; 1424 /* 1425 * Update the real timekeeper. 1426 * 1427 * We could avoid this memcpy by switching pointers, but that 1428 * requires changes to all other timekeeper usage sites as 1429 * well, i.e. move the timekeeper pointer getter into the 1430 * spinlocked/seqcount protected sections. And we trade this 1431 * memcpy under the timekeeper_seq against one before we start 1432 * updating. 1433 */ 1434 memcpy(real_tk, tk, sizeof(*tk)); 1435 timekeeping_update(real_tk, clock_set); 1436 write_seqcount_end(&timekeeper_seq); 1437 out: 1438 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 1439 if (clock_set) 1440 /* Have to call _delayed version, since in irq context*/ 1441 clock_was_set_delayed(); 1442 } 1443 1444 /** 1445 * getboottime - Return the real time of system boot. 1446 * @ts: pointer to the timespec to be set 1447 * 1448 * Returns the wall-time of boot in a timespec. 1449 * 1450 * This is based on the wall_to_monotonic offset and the total suspend 1451 * time. Calls to settimeofday will affect the value returned (which 1452 * basically means that however wrong your real time clock is at boot time, 1453 * you get the right time here). 1454 */ 1455 void getboottime(struct timespec *ts) 1456 { 1457 struct timekeeper *tk = &timekeeper; 1458 struct timespec boottime = { 1459 .tv_sec = tk->wall_to_monotonic.tv_sec + 1460 tk->total_sleep_time.tv_sec, 1461 .tv_nsec = tk->wall_to_monotonic.tv_nsec + 1462 tk->total_sleep_time.tv_nsec 1463 }; 1464 1465 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec); 1466 } 1467 EXPORT_SYMBOL_GPL(getboottime); 1468 1469 /** 1470 * get_monotonic_boottime - Returns monotonic time since boot 1471 * @ts: pointer to the timespec to be set 1472 * 1473 * Returns the monotonic time since boot in a timespec. 1474 * 1475 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also 1476 * includes the time spent in suspend. 1477 */ 1478 void get_monotonic_boottime(struct timespec *ts) 1479 { 1480 struct timekeeper *tk = &timekeeper; 1481 struct timespec tomono, sleep; 1482 s64 nsec; 1483 unsigned int seq; 1484 1485 WARN_ON(timekeeping_suspended); 1486 1487 do { 1488 seq = read_seqcount_begin(&timekeeper_seq); 1489 ts->tv_sec = tk->xtime_sec; 1490 nsec = timekeeping_get_ns(tk); 1491 tomono = tk->wall_to_monotonic; 1492 sleep = tk->total_sleep_time; 1493 1494 } while (read_seqcount_retry(&timekeeper_seq, seq)); 1495 1496 ts->tv_sec += tomono.tv_sec + sleep.tv_sec; 1497 ts->tv_nsec = 0; 1498 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec); 1499 } 1500 EXPORT_SYMBOL_GPL(get_monotonic_boottime); 1501 1502 /** 1503 * ktime_get_boottime - Returns monotonic time since boot in a ktime 1504 * 1505 * Returns the monotonic time since boot in a ktime 1506 * 1507 * This is similar to CLOCK_MONTONIC/ktime_get, but also 1508 * includes the time spent in suspend. 1509 */ 1510 ktime_t ktime_get_boottime(void) 1511 { 1512 struct timespec ts; 1513 1514 get_monotonic_boottime(&ts); 1515 return timespec_to_ktime(ts); 1516 } 1517 EXPORT_SYMBOL_GPL(ktime_get_boottime); 1518 1519 /** 1520 * monotonic_to_bootbased - Convert the monotonic time to boot based. 1521 * @ts: pointer to the timespec to be converted 1522 */ 1523 void monotonic_to_bootbased(struct timespec *ts) 1524 { 1525 struct timekeeper *tk = &timekeeper; 1526 1527 *ts = timespec_add(*ts, tk->total_sleep_time); 1528 } 1529 EXPORT_SYMBOL_GPL(monotonic_to_bootbased); 1530 1531 unsigned long get_seconds(void) 1532 { 1533 struct timekeeper *tk = &timekeeper; 1534 1535 return tk->xtime_sec; 1536 } 1537 EXPORT_SYMBOL(get_seconds); 1538 1539 struct timespec __current_kernel_time(void) 1540 { 1541 struct timekeeper *tk = &timekeeper; 1542 1543 return tk_xtime(tk); 1544 } 1545 1546 struct timespec current_kernel_time(void) 1547 { 1548 struct timekeeper *tk = &timekeeper; 1549 struct timespec now; 1550 unsigned long seq; 1551 1552 do { 1553 seq = read_seqcount_begin(&timekeeper_seq); 1554 1555 now = tk_xtime(tk); 1556 } while (read_seqcount_retry(&timekeeper_seq, seq)); 1557 1558 return now; 1559 } 1560 EXPORT_SYMBOL(current_kernel_time); 1561 1562 struct timespec get_monotonic_coarse(void) 1563 { 1564 struct timekeeper *tk = &timekeeper; 1565 struct timespec now, mono; 1566 unsigned long seq; 1567 1568 do { 1569 seq = read_seqcount_begin(&timekeeper_seq); 1570 1571 now = tk_xtime(tk); 1572 mono = tk->wall_to_monotonic; 1573 } while (read_seqcount_retry(&timekeeper_seq, seq)); 1574 1575 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec, 1576 now.tv_nsec + mono.tv_nsec); 1577 return now; 1578 } 1579 1580 /* 1581 * Must hold jiffies_lock 1582 */ 1583 void do_timer(unsigned long ticks) 1584 { 1585 jiffies_64 += ticks; 1586 calc_global_load(ticks); 1587 } 1588 1589 /** 1590 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic, 1591 * and sleep offsets. 1592 * @xtim: pointer to timespec to be set with xtime 1593 * @wtom: pointer to timespec to be set with wall_to_monotonic 1594 * @sleep: pointer to timespec to be set with time in suspend 1595 */ 1596 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim, 1597 struct timespec *wtom, struct timespec *sleep) 1598 { 1599 struct timekeeper *tk = &timekeeper; 1600 unsigned long seq; 1601 1602 do { 1603 seq = read_seqcount_begin(&timekeeper_seq); 1604 *xtim = tk_xtime(tk); 1605 *wtom = tk->wall_to_monotonic; 1606 *sleep = tk->total_sleep_time; 1607 } while (read_seqcount_retry(&timekeeper_seq, seq)); 1608 } 1609 1610 #ifdef CONFIG_HIGH_RES_TIMERS 1611 /** 1612 * ktime_get_update_offsets - hrtimer helper 1613 * @offs_real: pointer to storage for monotonic -> realtime offset 1614 * @offs_boot: pointer to storage for monotonic -> boottime offset 1615 * @offs_tai: pointer to storage for monotonic -> clock tai offset 1616 * 1617 * Returns current monotonic time and updates the offsets 1618 * Called from hrtimer_interrupt() or retrigger_next_event() 1619 */ 1620 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot, 1621 ktime_t *offs_tai) 1622 { 1623 struct timekeeper *tk = &timekeeper; 1624 ktime_t now; 1625 unsigned int seq; 1626 u64 secs, nsecs; 1627 1628 do { 1629 seq = read_seqcount_begin(&timekeeper_seq); 1630 1631 secs = tk->xtime_sec; 1632 nsecs = timekeeping_get_ns(tk); 1633 1634 *offs_real = tk->offs_real; 1635 *offs_boot = tk->offs_boot; 1636 *offs_tai = tk->offs_tai; 1637 } while (read_seqcount_retry(&timekeeper_seq, seq)); 1638 1639 now = ktime_add_ns(ktime_set(secs, 0), nsecs); 1640 now = ktime_sub(now, *offs_real); 1641 return now; 1642 } 1643 #endif 1644 1645 /** 1646 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format 1647 */ 1648 ktime_t ktime_get_monotonic_offset(void) 1649 { 1650 struct timekeeper *tk = &timekeeper; 1651 unsigned long seq; 1652 struct timespec wtom; 1653 1654 do { 1655 seq = read_seqcount_begin(&timekeeper_seq); 1656 wtom = tk->wall_to_monotonic; 1657 } while (read_seqcount_retry(&timekeeper_seq, seq)); 1658 1659 return timespec_to_ktime(wtom); 1660 } 1661 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset); 1662 1663 /** 1664 * do_adjtimex() - Accessor function to NTP __do_adjtimex function 1665 */ 1666 int do_adjtimex(struct timex *txc) 1667 { 1668 struct timekeeper *tk = &timekeeper; 1669 unsigned long flags; 1670 struct timespec ts; 1671 s32 orig_tai, tai; 1672 int ret; 1673 1674 /* Validate the data before disabling interrupts */ 1675 ret = ntp_validate_timex(txc); 1676 if (ret) 1677 return ret; 1678 1679 if (txc->modes & ADJ_SETOFFSET) { 1680 struct timespec delta; 1681 delta.tv_sec = txc->time.tv_sec; 1682 delta.tv_nsec = txc->time.tv_usec; 1683 if (!(txc->modes & ADJ_NANO)) 1684 delta.tv_nsec *= 1000; 1685 ret = timekeeping_inject_offset(&delta); 1686 if (ret) 1687 return ret; 1688 } 1689 1690 getnstimeofday(&ts); 1691 1692 raw_spin_lock_irqsave(&timekeeper_lock, flags); 1693 write_seqcount_begin(&timekeeper_seq); 1694 1695 orig_tai = tai = tk->tai_offset; 1696 ret = __do_adjtimex(txc, &ts, &tai); 1697 1698 if (tai != orig_tai) { 1699 __timekeeping_set_tai_offset(tk, tai); 1700 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET); 1701 } 1702 write_seqcount_end(&timekeeper_seq); 1703 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 1704 1705 if (tai != orig_tai) 1706 clock_was_set(); 1707 1708 ntp_notify_cmos_timer(); 1709 1710 return ret; 1711 } 1712 1713 #ifdef CONFIG_NTP_PPS 1714 /** 1715 * hardpps() - Accessor function to NTP __hardpps function 1716 */ 1717 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts) 1718 { 1719 unsigned long flags; 1720 1721 raw_spin_lock_irqsave(&timekeeper_lock, flags); 1722 write_seqcount_begin(&timekeeper_seq); 1723 1724 __hardpps(phase_ts, raw_ts); 1725 1726 write_seqcount_end(&timekeeper_seq); 1727 raw_spin_unlock_irqrestore(&timekeeper_lock, flags); 1728 } 1729 EXPORT_SYMBOL(hardpps); 1730 #endif 1731 1732 /** 1733 * xtime_update() - advances the timekeeping infrastructure 1734 * @ticks: number of ticks, that have elapsed since the last call. 1735 * 1736 * Must be called with interrupts disabled. 1737 */ 1738 void xtime_update(unsigned long ticks) 1739 { 1740 write_seqlock(&jiffies_lock); 1741 do_timer(ticks); 1742 write_sequnlock(&jiffies_lock); 1743 update_wall_time(); 1744 } 1745