1 /* 2 * linux/kernel/time/clocksource.c 3 * 4 * This file contains the functions which manage clocksource drivers. 5 * 6 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com) 7 * 8 * This program is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License as published by 10 * the Free Software Foundation; either version 2 of the License, or 11 * (at your option) any later version. 12 * 13 * This program is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 * GNU General Public License for more details. 17 * 18 * You should have received a copy of the GNU General Public License 19 * along with this program; if not, write to the Free Software 20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 21 * 22 * TODO WishList: 23 * o Allow clocksource drivers to be unregistered 24 */ 25 26 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 27 28 #include <linux/device.h> 29 #include <linux/clocksource.h> 30 #include <linux/init.h> 31 #include <linux/module.h> 32 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */ 33 #include <linux/tick.h> 34 #include <linux/kthread.h> 35 36 #include "tick-internal.h" 37 #include "timekeeping_internal.h" 38 39 /** 40 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks 41 * @mult: pointer to mult variable 42 * @shift: pointer to shift variable 43 * @from: frequency to convert from 44 * @to: frequency to convert to 45 * @maxsec: guaranteed runtime conversion range in seconds 46 * 47 * The function evaluates the shift/mult pair for the scaled math 48 * operations of clocksources and clockevents. 49 * 50 * @to and @from are frequency values in HZ. For clock sources @to is 51 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock 52 * event @to is the counter frequency and @from is NSEC_PER_SEC. 53 * 54 * The @maxsec conversion range argument controls the time frame in 55 * seconds which must be covered by the runtime conversion with the 56 * calculated mult and shift factors. This guarantees that no 64bit 57 * overflow happens when the input value of the conversion is 58 * multiplied with the calculated mult factor. Larger ranges may 59 * reduce the conversion accuracy by chosing smaller mult and shift 60 * factors. 61 */ 62 void 63 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec) 64 { 65 u64 tmp; 66 u32 sft, sftacc= 32; 67 68 /* 69 * Calculate the shift factor which is limiting the conversion 70 * range: 71 */ 72 tmp = ((u64)maxsec * from) >> 32; 73 while (tmp) { 74 tmp >>=1; 75 sftacc--; 76 } 77 78 /* 79 * Find the conversion shift/mult pair which has the best 80 * accuracy and fits the maxsec conversion range: 81 */ 82 for (sft = 32; sft > 0; sft--) { 83 tmp = (u64) to << sft; 84 tmp += from / 2; 85 do_div(tmp, from); 86 if ((tmp >> sftacc) == 0) 87 break; 88 } 89 *mult = tmp; 90 *shift = sft; 91 } 92 93 /*[Clocksource internal variables]--------- 94 * curr_clocksource: 95 * currently selected clocksource. 96 * clocksource_list: 97 * linked list with the registered clocksources 98 * clocksource_mutex: 99 * protects manipulations to curr_clocksource and the clocksource_list 100 * override_name: 101 * Name of the user-specified clocksource. 102 */ 103 static struct clocksource *curr_clocksource; 104 static LIST_HEAD(clocksource_list); 105 static DEFINE_MUTEX(clocksource_mutex); 106 static char override_name[CS_NAME_LEN]; 107 static int finished_booting; 108 109 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG 110 static void clocksource_watchdog_work(struct work_struct *work); 111 static void clocksource_select(void); 112 113 static LIST_HEAD(watchdog_list); 114 static struct clocksource *watchdog; 115 static struct timer_list watchdog_timer; 116 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work); 117 static DEFINE_SPINLOCK(watchdog_lock); 118 static int watchdog_running; 119 static atomic_t watchdog_reset_pending; 120 121 static int clocksource_watchdog_kthread(void *data); 122 static void __clocksource_change_rating(struct clocksource *cs, int rating); 123 124 /* 125 * Interval: 0.5sec Threshold: 0.0625s 126 */ 127 #define WATCHDOG_INTERVAL (HZ >> 1) 128 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4) 129 130 static void clocksource_watchdog_work(struct work_struct *work) 131 { 132 /* 133 * If kthread_run fails the next watchdog scan over the 134 * watchdog_list will find the unstable clock again. 135 */ 136 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog"); 137 } 138 139 static void __clocksource_unstable(struct clocksource *cs) 140 { 141 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG); 142 cs->flags |= CLOCK_SOURCE_UNSTABLE; 143 if (finished_booting) 144 schedule_work(&watchdog_work); 145 } 146 147 /** 148 * clocksource_mark_unstable - mark clocksource unstable via watchdog 149 * @cs: clocksource to be marked unstable 150 * 151 * This function is called instead of clocksource_change_rating from 152 * cpu hotplug code to avoid a deadlock between the clocksource mutex 153 * and the cpu hotplug mutex. It defers the update of the clocksource 154 * to the watchdog thread. 155 */ 156 void clocksource_mark_unstable(struct clocksource *cs) 157 { 158 unsigned long flags; 159 160 spin_lock_irqsave(&watchdog_lock, flags); 161 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) { 162 if (list_empty(&cs->wd_list)) 163 list_add(&cs->wd_list, &watchdog_list); 164 __clocksource_unstable(cs); 165 } 166 spin_unlock_irqrestore(&watchdog_lock, flags); 167 } 168 169 static void clocksource_watchdog(unsigned long data) 170 { 171 struct clocksource *cs; 172 cycle_t csnow, wdnow, cslast, wdlast, delta; 173 int64_t wd_nsec, cs_nsec; 174 int next_cpu, reset_pending; 175 176 spin_lock(&watchdog_lock); 177 if (!watchdog_running) 178 goto out; 179 180 reset_pending = atomic_read(&watchdog_reset_pending); 181 182 list_for_each_entry(cs, &watchdog_list, wd_list) { 183 184 /* Clocksource already marked unstable? */ 185 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 186 if (finished_booting) 187 schedule_work(&watchdog_work); 188 continue; 189 } 190 191 local_irq_disable(); 192 csnow = cs->read(cs); 193 wdnow = watchdog->read(watchdog); 194 local_irq_enable(); 195 196 /* Clocksource initialized ? */ 197 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) || 198 atomic_read(&watchdog_reset_pending)) { 199 cs->flags |= CLOCK_SOURCE_WATCHDOG; 200 cs->wd_last = wdnow; 201 cs->cs_last = csnow; 202 continue; 203 } 204 205 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask); 206 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult, 207 watchdog->shift); 208 209 delta = clocksource_delta(csnow, cs->cs_last, cs->mask); 210 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift); 211 wdlast = cs->wd_last; /* save these in case we print them */ 212 cslast = cs->cs_last; 213 cs->cs_last = csnow; 214 cs->wd_last = wdnow; 215 216 if (atomic_read(&watchdog_reset_pending)) 217 continue; 218 219 /* Check the deviation from the watchdog clocksource. */ 220 if (abs(cs_nsec - wd_nsec) > WATCHDOG_THRESHOLD) { 221 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n", 222 smp_processor_id(), cs->name); 223 pr_warn(" '%s' wd_now: %llx wd_last: %llx mask: %llx\n", 224 watchdog->name, wdnow, wdlast, watchdog->mask); 225 pr_warn(" '%s' cs_now: %llx cs_last: %llx mask: %llx\n", 226 cs->name, csnow, cslast, cs->mask); 227 __clocksource_unstable(cs); 228 continue; 229 } 230 231 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && 232 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) && 233 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) { 234 /* Mark it valid for high-res. */ 235 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 236 237 /* 238 * clocksource_done_booting() will sort it if 239 * finished_booting is not set yet. 240 */ 241 if (!finished_booting) 242 continue; 243 244 /* 245 * If this is not the current clocksource let 246 * the watchdog thread reselect it. Due to the 247 * change to high res this clocksource might 248 * be preferred now. If it is the current 249 * clocksource let the tick code know about 250 * that change. 251 */ 252 if (cs != curr_clocksource) { 253 cs->flags |= CLOCK_SOURCE_RESELECT; 254 schedule_work(&watchdog_work); 255 } else { 256 tick_clock_notify(); 257 } 258 } 259 } 260 261 /* 262 * We only clear the watchdog_reset_pending, when we did a 263 * full cycle through all clocksources. 264 */ 265 if (reset_pending) 266 atomic_dec(&watchdog_reset_pending); 267 268 /* 269 * Cycle through CPUs to check if the CPUs stay synchronized 270 * to each other. 271 */ 272 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask); 273 if (next_cpu >= nr_cpu_ids) 274 next_cpu = cpumask_first(cpu_online_mask); 275 watchdog_timer.expires += WATCHDOG_INTERVAL; 276 add_timer_on(&watchdog_timer, next_cpu); 277 out: 278 spin_unlock(&watchdog_lock); 279 } 280 281 static inline void clocksource_start_watchdog(void) 282 { 283 if (watchdog_running || !watchdog || list_empty(&watchdog_list)) 284 return; 285 init_timer(&watchdog_timer); 286 watchdog_timer.function = clocksource_watchdog; 287 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL; 288 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask)); 289 watchdog_running = 1; 290 } 291 292 static inline void clocksource_stop_watchdog(void) 293 { 294 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list))) 295 return; 296 del_timer(&watchdog_timer); 297 watchdog_running = 0; 298 } 299 300 static inline void clocksource_reset_watchdog(void) 301 { 302 struct clocksource *cs; 303 304 list_for_each_entry(cs, &watchdog_list, wd_list) 305 cs->flags &= ~CLOCK_SOURCE_WATCHDOG; 306 } 307 308 static void clocksource_resume_watchdog(void) 309 { 310 atomic_inc(&watchdog_reset_pending); 311 } 312 313 static void clocksource_enqueue_watchdog(struct clocksource *cs) 314 { 315 unsigned long flags; 316 317 spin_lock_irqsave(&watchdog_lock, flags); 318 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { 319 /* cs is a clocksource to be watched. */ 320 list_add(&cs->wd_list, &watchdog_list); 321 cs->flags &= ~CLOCK_SOURCE_WATCHDOG; 322 } else { 323 /* cs is a watchdog. */ 324 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) 325 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 326 } 327 spin_unlock_irqrestore(&watchdog_lock, flags); 328 } 329 330 static void clocksource_select_watchdog(bool fallback) 331 { 332 struct clocksource *cs, *old_wd; 333 unsigned long flags; 334 335 spin_lock_irqsave(&watchdog_lock, flags); 336 /* save current watchdog */ 337 old_wd = watchdog; 338 if (fallback) 339 watchdog = NULL; 340 341 list_for_each_entry(cs, &clocksource_list, list) { 342 /* cs is a clocksource to be watched. */ 343 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) 344 continue; 345 346 /* Skip current if we were requested for a fallback. */ 347 if (fallback && cs == old_wd) 348 continue; 349 350 /* Pick the best watchdog. */ 351 if (!watchdog || cs->rating > watchdog->rating) 352 watchdog = cs; 353 } 354 /* If we failed to find a fallback restore the old one. */ 355 if (!watchdog) 356 watchdog = old_wd; 357 358 /* If we changed the watchdog we need to reset cycles. */ 359 if (watchdog != old_wd) 360 clocksource_reset_watchdog(); 361 362 /* Check if the watchdog timer needs to be started. */ 363 clocksource_start_watchdog(); 364 spin_unlock_irqrestore(&watchdog_lock, flags); 365 } 366 367 static void clocksource_dequeue_watchdog(struct clocksource *cs) 368 { 369 unsigned long flags; 370 371 spin_lock_irqsave(&watchdog_lock, flags); 372 if (cs != watchdog) { 373 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { 374 /* cs is a watched clocksource. */ 375 list_del_init(&cs->wd_list); 376 /* Check if the watchdog timer needs to be stopped. */ 377 clocksource_stop_watchdog(); 378 } 379 } 380 spin_unlock_irqrestore(&watchdog_lock, flags); 381 } 382 383 static int __clocksource_watchdog_kthread(void) 384 { 385 struct clocksource *cs, *tmp; 386 unsigned long flags; 387 LIST_HEAD(unstable); 388 int select = 0; 389 390 spin_lock_irqsave(&watchdog_lock, flags); 391 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) { 392 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 393 list_del_init(&cs->wd_list); 394 list_add(&cs->wd_list, &unstable); 395 select = 1; 396 } 397 if (cs->flags & CLOCK_SOURCE_RESELECT) { 398 cs->flags &= ~CLOCK_SOURCE_RESELECT; 399 select = 1; 400 } 401 } 402 /* Check if the watchdog timer needs to be stopped. */ 403 clocksource_stop_watchdog(); 404 spin_unlock_irqrestore(&watchdog_lock, flags); 405 406 /* Needs to be done outside of watchdog lock */ 407 list_for_each_entry_safe(cs, tmp, &unstable, wd_list) { 408 list_del_init(&cs->wd_list); 409 __clocksource_change_rating(cs, 0); 410 } 411 return select; 412 } 413 414 static int clocksource_watchdog_kthread(void *data) 415 { 416 mutex_lock(&clocksource_mutex); 417 if (__clocksource_watchdog_kthread()) 418 clocksource_select(); 419 mutex_unlock(&clocksource_mutex); 420 return 0; 421 } 422 423 static bool clocksource_is_watchdog(struct clocksource *cs) 424 { 425 return cs == watchdog; 426 } 427 428 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */ 429 430 static void clocksource_enqueue_watchdog(struct clocksource *cs) 431 { 432 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) 433 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 434 } 435 436 static void clocksource_select_watchdog(bool fallback) { } 437 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { } 438 static inline void clocksource_resume_watchdog(void) { } 439 static inline int __clocksource_watchdog_kthread(void) { return 0; } 440 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; } 441 void clocksource_mark_unstable(struct clocksource *cs) { } 442 443 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */ 444 445 /** 446 * clocksource_suspend - suspend the clocksource(s) 447 */ 448 void clocksource_suspend(void) 449 { 450 struct clocksource *cs; 451 452 list_for_each_entry_reverse(cs, &clocksource_list, list) 453 if (cs->suspend) 454 cs->suspend(cs); 455 } 456 457 /** 458 * clocksource_resume - resume the clocksource(s) 459 */ 460 void clocksource_resume(void) 461 { 462 struct clocksource *cs; 463 464 list_for_each_entry(cs, &clocksource_list, list) 465 if (cs->resume) 466 cs->resume(cs); 467 468 clocksource_resume_watchdog(); 469 } 470 471 /** 472 * clocksource_touch_watchdog - Update watchdog 473 * 474 * Update the watchdog after exception contexts such as kgdb so as not 475 * to incorrectly trip the watchdog. This might fail when the kernel 476 * was stopped in code which holds watchdog_lock. 477 */ 478 void clocksource_touch_watchdog(void) 479 { 480 clocksource_resume_watchdog(); 481 } 482 483 /** 484 * clocksource_max_adjustment- Returns max adjustment amount 485 * @cs: Pointer to clocksource 486 * 487 */ 488 static u32 clocksource_max_adjustment(struct clocksource *cs) 489 { 490 u64 ret; 491 /* 492 * We won't try to correct for more than 11% adjustments (110,000 ppm), 493 */ 494 ret = (u64)cs->mult * 11; 495 do_div(ret,100); 496 return (u32)ret; 497 } 498 499 /** 500 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted 501 * @mult: cycle to nanosecond multiplier 502 * @shift: cycle to nanosecond divisor (power of two) 503 * @maxadj: maximum adjustment value to mult (~11%) 504 * @mask: bitmask for two's complement subtraction of non 64 bit counters 505 * @max_cyc: maximum cycle value before potential overflow (does not include 506 * any safety margin) 507 * 508 * NOTE: This function includes a safety margin of 50%, in other words, we 509 * return half the number of nanoseconds the hardware counter can technically 510 * cover. This is done so that we can potentially detect problems caused by 511 * delayed timers or bad hardware, which might result in time intervals that 512 * are larger than what the math used can handle without overflows. 513 */ 514 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc) 515 { 516 u64 max_nsecs, max_cycles; 517 518 /* 519 * Calculate the maximum number of cycles that we can pass to the 520 * cyc2ns() function without overflowing a 64-bit result. 521 */ 522 max_cycles = ULLONG_MAX; 523 do_div(max_cycles, mult+maxadj); 524 525 /* 526 * The actual maximum number of cycles we can defer the clocksource is 527 * determined by the minimum of max_cycles and mask. 528 * Note: Here we subtract the maxadj to make sure we don't sleep for 529 * too long if there's a large negative adjustment. 530 */ 531 max_cycles = min(max_cycles, mask); 532 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift); 533 534 /* return the max_cycles value as well if requested */ 535 if (max_cyc) 536 *max_cyc = max_cycles; 537 538 /* Return 50% of the actual maximum, so we can detect bad values */ 539 max_nsecs >>= 1; 540 541 return max_nsecs; 542 } 543 544 /** 545 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles 546 * @cs: Pointer to clocksource to be updated 547 * 548 */ 549 static inline void clocksource_update_max_deferment(struct clocksource *cs) 550 { 551 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift, 552 cs->maxadj, cs->mask, 553 &cs->max_cycles); 554 } 555 556 #ifndef CONFIG_ARCH_USES_GETTIMEOFFSET 557 558 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur) 559 { 560 struct clocksource *cs; 561 562 if (!finished_booting || list_empty(&clocksource_list)) 563 return NULL; 564 565 /* 566 * We pick the clocksource with the highest rating. If oneshot 567 * mode is active, we pick the highres valid clocksource with 568 * the best rating. 569 */ 570 list_for_each_entry(cs, &clocksource_list, list) { 571 if (skipcur && cs == curr_clocksource) 572 continue; 573 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES)) 574 continue; 575 return cs; 576 } 577 return NULL; 578 } 579 580 static void __clocksource_select(bool skipcur) 581 { 582 bool oneshot = tick_oneshot_mode_active(); 583 struct clocksource *best, *cs; 584 585 /* Find the best suitable clocksource */ 586 best = clocksource_find_best(oneshot, skipcur); 587 if (!best) 588 return; 589 590 /* Check for the override clocksource. */ 591 list_for_each_entry(cs, &clocksource_list, list) { 592 if (skipcur && cs == curr_clocksource) 593 continue; 594 if (strcmp(cs->name, override_name) != 0) 595 continue; 596 /* 597 * Check to make sure we don't switch to a non-highres 598 * capable clocksource if the tick code is in oneshot 599 * mode (highres or nohz) 600 */ 601 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) { 602 /* Override clocksource cannot be used. */ 603 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 604 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n", 605 cs->name); 606 override_name[0] = 0; 607 } else { 608 /* 609 * The override cannot be currently verified. 610 * Deferring to let the watchdog check. 611 */ 612 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n", 613 cs->name); 614 } 615 } else 616 /* Override clocksource can be used. */ 617 best = cs; 618 break; 619 } 620 621 if (curr_clocksource != best && !timekeeping_notify(best)) { 622 pr_info("Switched to clocksource %s\n", best->name); 623 curr_clocksource = best; 624 } 625 } 626 627 /** 628 * clocksource_select - Select the best clocksource available 629 * 630 * Private function. Must hold clocksource_mutex when called. 631 * 632 * Select the clocksource with the best rating, or the clocksource, 633 * which is selected by userspace override. 634 */ 635 static void clocksource_select(void) 636 { 637 __clocksource_select(false); 638 } 639 640 static void clocksource_select_fallback(void) 641 { 642 __clocksource_select(true); 643 } 644 645 #else /* !CONFIG_ARCH_USES_GETTIMEOFFSET */ 646 static inline void clocksource_select(void) { } 647 static inline void clocksource_select_fallback(void) { } 648 649 #endif 650 651 /* 652 * clocksource_done_booting - Called near the end of core bootup 653 * 654 * Hack to avoid lots of clocksource churn at boot time. 655 * We use fs_initcall because we want this to start before 656 * device_initcall but after subsys_initcall. 657 */ 658 static int __init clocksource_done_booting(void) 659 { 660 mutex_lock(&clocksource_mutex); 661 curr_clocksource = clocksource_default_clock(); 662 finished_booting = 1; 663 /* 664 * Run the watchdog first to eliminate unstable clock sources 665 */ 666 __clocksource_watchdog_kthread(); 667 clocksource_select(); 668 mutex_unlock(&clocksource_mutex); 669 return 0; 670 } 671 fs_initcall(clocksource_done_booting); 672 673 /* 674 * Enqueue the clocksource sorted by rating 675 */ 676 static void clocksource_enqueue(struct clocksource *cs) 677 { 678 struct list_head *entry = &clocksource_list; 679 struct clocksource *tmp; 680 681 list_for_each_entry(tmp, &clocksource_list, list) { 682 /* Keep track of the place, where to insert */ 683 if (tmp->rating < cs->rating) 684 break; 685 entry = &tmp->list; 686 } 687 list_add(&cs->list, entry); 688 } 689 690 /** 691 * __clocksource_update_freq_scale - Used update clocksource with new freq 692 * @cs: clocksource to be registered 693 * @scale: Scale factor multiplied against freq to get clocksource hz 694 * @freq: clocksource frequency (cycles per second) divided by scale 695 * 696 * This should only be called from the clocksource->enable() method. 697 * 698 * This *SHOULD NOT* be called directly! Please use the 699 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper 700 * functions. 701 */ 702 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq) 703 { 704 u64 sec; 705 706 /* 707 * Default clocksources are *special* and self-define their mult/shift. 708 * But, you're not special, so you should specify a freq value. 709 */ 710 if (freq) { 711 /* 712 * Calc the maximum number of seconds which we can run before 713 * wrapping around. For clocksources which have a mask > 32-bit 714 * we need to limit the max sleep time to have a good 715 * conversion precision. 10 minutes is still a reasonable 716 * amount. That results in a shift value of 24 for a 717 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to 718 * ~ 0.06ppm granularity for NTP. 719 */ 720 sec = cs->mask; 721 do_div(sec, freq); 722 do_div(sec, scale); 723 if (!sec) 724 sec = 1; 725 else if (sec > 600 && cs->mask > UINT_MAX) 726 sec = 600; 727 728 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq, 729 NSEC_PER_SEC / scale, sec * scale); 730 } 731 /* 732 * Ensure clocksources that have large 'mult' values don't overflow 733 * when adjusted. 734 */ 735 cs->maxadj = clocksource_max_adjustment(cs); 736 while (freq && ((cs->mult + cs->maxadj < cs->mult) 737 || (cs->mult - cs->maxadj > cs->mult))) { 738 cs->mult >>= 1; 739 cs->shift--; 740 cs->maxadj = clocksource_max_adjustment(cs); 741 } 742 743 /* 744 * Only warn for *special* clocksources that self-define 745 * their mult/shift values and don't specify a freq. 746 */ 747 WARN_ONCE(cs->mult + cs->maxadj < cs->mult, 748 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n", 749 cs->name); 750 751 clocksource_update_max_deferment(cs); 752 753 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n", 754 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns); 755 } 756 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale); 757 758 /** 759 * __clocksource_register_scale - Used to install new clocksources 760 * @cs: clocksource to be registered 761 * @scale: Scale factor multiplied against freq to get clocksource hz 762 * @freq: clocksource frequency (cycles per second) divided by scale 763 * 764 * Returns -EBUSY if registration fails, zero otherwise. 765 * 766 * This *SHOULD NOT* be called directly! Please use the 767 * clocksource_register_hz() or clocksource_register_khz helper functions. 768 */ 769 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq) 770 { 771 772 /* Initialize mult/shift and max_idle_ns */ 773 __clocksource_update_freq_scale(cs, scale, freq); 774 775 /* Add clocksource to the clocksource list */ 776 mutex_lock(&clocksource_mutex); 777 clocksource_enqueue(cs); 778 clocksource_enqueue_watchdog(cs); 779 clocksource_select(); 780 clocksource_select_watchdog(false); 781 mutex_unlock(&clocksource_mutex); 782 return 0; 783 } 784 EXPORT_SYMBOL_GPL(__clocksource_register_scale); 785 786 static void __clocksource_change_rating(struct clocksource *cs, int rating) 787 { 788 list_del(&cs->list); 789 cs->rating = rating; 790 clocksource_enqueue(cs); 791 } 792 793 /** 794 * clocksource_change_rating - Change the rating of a registered clocksource 795 * @cs: clocksource to be changed 796 * @rating: new rating 797 */ 798 void clocksource_change_rating(struct clocksource *cs, int rating) 799 { 800 mutex_lock(&clocksource_mutex); 801 __clocksource_change_rating(cs, rating); 802 clocksource_select(); 803 clocksource_select_watchdog(false); 804 mutex_unlock(&clocksource_mutex); 805 } 806 EXPORT_SYMBOL(clocksource_change_rating); 807 808 /* 809 * Unbind clocksource @cs. Called with clocksource_mutex held 810 */ 811 static int clocksource_unbind(struct clocksource *cs) 812 { 813 if (clocksource_is_watchdog(cs)) { 814 /* Select and try to install a replacement watchdog. */ 815 clocksource_select_watchdog(true); 816 if (clocksource_is_watchdog(cs)) 817 return -EBUSY; 818 } 819 820 if (cs == curr_clocksource) { 821 /* Select and try to install a replacement clock source */ 822 clocksource_select_fallback(); 823 if (curr_clocksource == cs) 824 return -EBUSY; 825 } 826 clocksource_dequeue_watchdog(cs); 827 list_del_init(&cs->list); 828 return 0; 829 } 830 831 /** 832 * clocksource_unregister - remove a registered clocksource 833 * @cs: clocksource to be unregistered 834 */ 835 int clocksource_unregister(struct clocksource *cs) 836 { 837 int ret = 0; 838 839 mutex_lock(&clocksource_mutex); 840 if (!list_empty(&cs->list)) 841 ret = clocksource_unbind(cs); 842 mutex_unlock(&clocksource_mutex); 843 return ret; 844 } 845 EXPORT_SYMBOL(clocksource_unregister); 846 847 #ifdef CONFIG_SYSFS 848 /** 849 * sysfs_show_current_clocksources - sysfs interface for current clocksource 850 * @dev: unused 851 * @attr: unused 852 * @buf: char buffer to be filled with clocksource list 853 * 854 * Provides sysfs interface for listing current clocksource. 855 */ 856 static ssize_t 857 sysfs_show_current_clocksources(struct device *dev, 858 struct device_attribute *attr, char *buf) 859 { 860 ssize_t count = 0; 861 862 mutex_lock(&clocksource_mutex); 863 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name); 864 mutex_unlock(&clocksource_mutex); 865 866 return count; 867 } 868 869 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt) 870 { 871 size_t ret = cnt; 872 873 /* strings from sysfs write are not 0 terminated! */ 874 if (!cnt || cnt >= CS_NAME_LEN) 875 return -EINVAL; 876 877 /* strip of \n: */ 878 if (buf[cnt-1] == '\n') 879 cnt--; 880 if (cnt > 0) 881 memcpy(dst, buf, cnt); 882 dst[cnt] = 0; 883 return ret; 884 } 885 886 /** 887 * sysfs_override_clocksource - interface for manually overriding clocksource 888 * @dev: unused 889 * @attr: unused 890 * @buf: name of override clocksource 891 * @count: length of buffer 892 * 893 * Takes input from sysfs interface for manually overriding the default 894 * clocksource selection. 895 */ 896 static ssize_t sysfs_override_clocksource(struct device *dev, 897 struct device_attribute *attr, 898 const char *buf, size_t count) 899 { 900 ssize_t ret; 901 902 mutex_lock(&clocksource_mutex); 903 904 ret = sysfs_get_uname(buf, override_name, count); 905 if (ret >= 0) 906 clocksource_select(); 907 908 mutex_unlock(&clocksource_mutex); 909 910 return ret; 911 } 912 913 /** 914 * sysfs_unbind_current_clocksource - interface for manually unbinding clocksource 915 * @dev: unused 916 * @attr: unused 917 * @buf: unused 918 * @count: length of buffer 919 * 920 * Takes input from sysfs interface for manually unbinding a clocksource. 921 */ 922 static ssize_t sysfs_unbind_clocksource(struct device *dev, 923 struct device_attribute *attr, 924 const char *buf, size_t count) 925 { 926 struct clocksource *cs; 927 char name[CS_NAME_LEN]; 928 ssize_t ret; 929 930 ret = sysfs_get_uname(buf, name, count); 931 if (ret < 0) 932 return ret; 933 934 ret = -ENODEV; 935 mutex_lock(&clocksource_mutex); 936 list_for_each_entry(cs, &clocksource_list, list) { 937 if (strcmp(cs->name, name)) 938 continue; 939 ret = clocksource_unbind(cs); 940 break; 941 } 942 mutex_unlock(&clocksource_mutex); 943 944 return ret ? ret : count; 945 } 946 947 /** 948 * sysfs_show_available_clocksources - sysfs interface for listing clocksource 949 * @dev: unused 950 * @attr: unused 951 * @buf: char buffer to be filled with clocksource list 952 * 953 * Provides sysfs interface for listing registered clocksources 954 */ 955 static ssize_t 956 sysfs_show_available_clocksources(struct device *dev, 957 struct device_attribute *attr, 958 char *buf) 959 { 960 struct clocksource *src; 961 ssize_t count = 0; 962 963 mutex_lock(&clocksource_mutex); 964 list_for_each_entry(src, &clocksource_list, list) { 965 /* 966 * Don't show non-HRES clocksource if the tick code is 967 * in one shot mode (highres=on or nohz=on) 968 */ 969 if (!tick_oneshot_mode_active() || 970 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES)) 971 count += snprintf(buf + count, 972 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), 973 "%s ", src->name); 974 } 975 mutex_unlock(&clocksource_mutex); 976 977 count += snprintf(buf + count, 978 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n"); 979 980 return count; 981 } 982 983 /* 984 * Sysfs setup bits: 985 */ 986 static DEVICE_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources, 987 sysfs_override_clocksource); 988 989 static DEVICE_ATTR(unbind_clocksource, 0200, NULL, sysfs_unbind_clocksource); 990 991 static DEVICE_ATTR(available_clocksource, 0444, 992 sysfs_show_available_clocksources, NULL); 993 994 static struct bus_type clocksource_subsys = { 995 .name = "clocksource", 996 .dev_name = "clocksource", 997 }; 998 999 static struct device device_clocksource = { 1000 .id = 0, 1001 .bus = &clocksource_subsys, 1002 }; 1003 1004 static int __init init_clocksource_sysfs(void) 1005 { 1006 int error = subsys_system_register(&clocksource_subsys, NULL); 1007 1008 if (!error) 1009 error = device_register(&device_clocksource); 1010 if (!error) 1011 error = device_create_file( 1012 &device_clocksource, 1013 &dev_attr_current_clocksource); 1014 if (!error) 1015 error = device_create_file(&device_clocksource, 1016 &dev_attr_unbind_clocksource); 1017 if (!error) 1018 error = device_create_file( 1019 &device_clocksource, 1020 &dev_attr_available_clocksource); 1021 return error; 1022 } 1023 1024 device_initcall(init_clocksource_sysfs); 1025 #endif /* CONFIG_SYSFS */ 1026 1027 /** 1028 * boot_override_clocksource - boot clock override 1029 * @str: override name 1030 * 1031 * Takes a clocksource= boot argument and uses it 1032 * as the clocksource override name. 1033 */ 1034 static int __init boot_override_clocksource(char* str) 1035 { 1036 mutex_lock(&clocksource_mutex); 1037 if (str) 1038 strlcpy(override_name, str, sizeof(override_name)); 1039 mutex_unlock(&clocksource_mutex); 1040 return 1; 1041 } 1042 1043 __setup("clocksource=", boot_override_clocksource); 1044 1045 /** 1046 * boot_override_clock - Compatibility layer for deprecated boot option 1047 * @str: override name 1048 * 1049 * DEPRECATED! Takes a clock= boot argument and uses it 1050 * as the clocksource override name 1051 */ 1052 static int __init boot_override_clock(char* str) 1053 { 1054 if (!strcmp(str, "pmtmr")) { 1055 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n"); 1056 return boot_override_clocksource("acpi_pm"); 1057 } 1058 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n"); 1059 return boot_override_clocksource(str); 1060 } 1061 1062 __setup("clock=", boot_override_clock); 1063