1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * This file contains the functions which manage clocksource drivers. 4 * 5 * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com) 6 */ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/device.h> 11 #include <linux/clocksource.h> 12 #include <linux/init.h> 13 #include <linux/module.h> 14 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */ 15 #include <linux/tick.h> 16 #include <linux/kthread.h> 17 #include <linux/prandom.h> 18 #include <linux/cpu.h> 19 20 #include "tick-internal.h" 21 #include "timekeeping_internal.h" 22 23 /** 24 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks 25 * @mult: pointer to mult variable 26 * @shift: pointer to shift variable 27 * @from: frequency to convert from 28 * @to: frequency to convert to 29 * @maxsec: guaranteed runtime conversion range in seconds 30 * 31 * The function evaluates the shift/mult pair for the scaled math 32 * operations of clocksources and clockevents. 33 * 34 * @to and @from are frequency values in HZ. For clock sources @to is 35 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock 36 * event @to is the counter frequency and @from is NSEC_PER_SEC. 37 * 38 * The @maxsec conversion range argument controls the time frame in 39 * seconds which must be covered by the runtime conversion with the 40 * calculated mult and shift factors. This guarantees that no 64bit 41 * overflow happens when the input value of the conversion is 42 * multiplied with the calculated mult factor. Larger ranges may 43 * reduce the conversion accuracy by choosing smaller mult and shift 44 * factors. 45 */ 46 void 47 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec) 48 { 49 u64 tmp; 50 u32 sft, sftacc= 32; 51 52 /* 53 * Calculate the shift factor which is limiting the conversion 54 * range: 55 */ 56 tmp = ((u64)maxsec * from) >> 32; 57 while (tmp) { 58 tmp >>=1; 59 sftacc--; 60 } 61 62 /* 63 * Find the conversion shift/mult pair which has the best 64 * accuracy and fits the maxsec conversion range: 65 */ 66 for (sft = 32; sft > 0; sft--) { 67 tmp = (u64) to << sft; 68 tmp += from / 2; 69 do_div(tmp, from); 70 if ((tmp >> sftacc) == 0) 71 break; 72 } 73 *mult = tmp; 74 *shift = sft; 75 } 76 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift); 77 78 /*[Clocksource internal variables]--------- 79 * curr_clocksource: 80 * currently selected clocksource. 81 * suspend_clocksource: 82 * used to calculate the suspend time. 83 * clocksource_list: 84 * linked list with the registered clocksources 85 * clocksource_mutex: 86 * protects manipulations to curr_clocksource and the clocksource_list 87 * override_name: 88 * Name of the user-specified clocksource. 89 */ 90 static struct clocksource *curr_clocksource; 91 static struct clocksource *suspend_clocksource; 92 static LIST_HEAD(clocksource_list); 93 static DEFINE_MUTEX(clocksource_mutex); 94 static char override_name[CS_NAME_LEN]; 95 static int finished_booting; 96 static u64 suspend_start; 97 98 /* 99 * Threshold: 0.0312s, when doubled: 0.0625s. 100 * Also a default for cs->uncertainty_margin when registering clocks. 101 */ 102 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5) 103 104 /* 105 * Maximum permissible delay between two readouts of the watchdog 106 * clocksource surrounding a read of the clocksource being validated. 107 * This delay could be due to SMIs, NMIs, or to VCPU preemptions. Used as 108 * a lower bound for cs->uncertainty_margin values when registering clocks. 109 */ 110 #define WATCHDOG_MAX_SKEW (100 * NSEC_PER_USEC) 111 112 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG 113 static void clocksource_watchdog_work(struct work_struct *work); 114 static void clocksource_select(void); 115 116 static LIST_HEAD(watchdog_list); 117 static struct clocksource *watchdog; 118 static struct timer_list watchdog_timer; 119 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work); 120 static DEFINE_SPINLOCK(watchdog_lock); 121 static int watchdog_running; 122 static atomic_t watchdog_reset_pending; 123 124 static inline void clocksource_watchdog_lock(unsigned long *flags) 125 { 126 spin_lock_irqsave(&watchdog_lock, *flags); 127 } 128 129 static inline void clocksource_watchdog_unlock(unsigned long *flags) 130 { 131 spin_unlock_irqrestore(&watchdog_lock, *flags); 132 } 133 134 static int clocksource_watchdog_kthread(void *data); 135 static void __clocksource_change_rating(struct clocksource *cs, int rating); 136 137 /* 138 * Interval: 0.5sec. 139 */ 140 #define WATCHDOG_INTERVAL (HZ >> 1) 141 142 static void clocksource_watchdog_work(struct work_struct *work) 143 { 144 /* 145 * We cannot directly run clocksource_watchdog_kthread() here, because 146 * clocksource_select() calls timekeeping_notify() which uses 147 * stop_machine(). One cannot use stop_machine() from a workqueue() due 148 * lock inversions wrt CPU hotplug. 149 * 150 * Also, we only ever run this work once or twice during the lifetime 151 * of the kernel, so there is no point in creating a more permanent 152 * kthread for this. 153 * 154 * If kthread_run fails the next watchdog scan over the 155 * watchdog_list will find the unstable clock again. 156 */ 157 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog"); 158 } 159 160 static void __clocksource_unstable(struct clocksource *cs) 161 { 162 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG); 163 cs->flags |= CLOCK_SOURCE_UNSTABLE; 164 165 /* 166 * If the clocksource is registered clocksource_watchdog_kthread() will 167 * re-rate and re-select. 168 */ 169 if (list_empty(&cs->list)) { 170 cs->rating = 0; 171 return; 172 } 173 174 if (cs->mark_unstable) 175 cs->mark_unstable(cs); 176 177 /* kick clocksource_watchdog_kthread() */ 178 if (finished_booting) 179 schedule_work(&watchdog_work); 180 } 181 182 /** 183 * clocksource_mark_unstable - mark clocksource unstable via watchdog 184 * @cs: clocksource to be marked unstable 185 * 186 * This function is called by the x86 TSC code to mark clocksources as unstable; 187 * it defers demotion and re-selection to a kthread. 188 */ 189 void clocksource_mark_unstable(struct clocksource *cs) 190 { 191 unsigned long flags; 192 193 spin_lock_irqsave(&watchdog_lock, flags); 194 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) { 195 if (!list_empty(&cs->list) && list_empty(&cs->wd_list)) 196 list_add(&cs->wd_list, &watchdog_list); 197 __clocksource_unstable(cs); 198 } 199 spin_unlock_irqrestore(&watchdog_lock, flags); 200 } 201 202 ulong max_cswd_read_retries = 2; 203 module_param(max_cswd_read_retries, ulong, 0644); 204 EXPORT_SYMBOL_GPL(max_cswd_read_retries); 205 static int verify_n_cpus = 8; 206 module_param(verify_n_cpus, int, 0644); 207 208 enum wd_read_status { 209 WD_READ_SUCCESS, 210 WD_READ_UNSTABLE, 211 WD_READ_SKIP 212 }; 213 214 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow) 215 { 216 unsigned int nretries; 217 u64 wd_end, wd_end2, wd_delta; 218 int64_t wd_delay, wd_seq_delay; 219 220 for (nretries = 0; nretries <= max_cswd_read_retries; nretries++) { 221 local_irq_disable(); 222 *wdnow = watchdog->read(watchdog); 223 *csnow = cs->read(cs); 224 wd_end = watchdog->read(watchdog); 225 wd_end2 = watchdog->read(watchdog); 226 local_irq_enable(); 227 228 wd_delta = clocksource_delta(wd_end, *wdnow, watchdog->mask); 229 wd_delay = clocksource_cyc2ns(wd_delta, watchdog->mult, 230 watchdog->shift); 231 if (wd_delay <= WATCHDOG_MAX_SKEW) { 232 if (nretries > 1 || nretries >= max_cswd_read_retries) { 233 pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n", 234 smp_processor_id(), watchdog->name, nretries); 235 } 236 return WD_READ_SUCCESS; 237 } 238 239 /* 240 * Now compute delay in consecutive watchdog read to see if 241 * there is too much external interferences that cause 242 * significant delay in reading both clocksource and watchdog. 243 * 244 * If consecutive WD read-back delay > WATCHDOG_MAX_SKEW/2, 245 * report system busy, reinit the watchdog and skip the current 246 * watchdog test. 247 */ 248 wd_delta = clocksource_delta(wd_end2, wd_end, watchdog->mask); 249 wd_seq_delay = clocksource_cyc2ns(wd_delta, watchdog->mult, watchdog->shift); 250 if (wd_seq_delay > WATCHDOG_MAX_SKEW/2) 251 goto skip_test; 252 } 253 254 pr_warn("timekeeping watchdog on CPU%d: %s read-back delay of %lldns, attempt %d, marking unstable\n", 255 smp_processor_id(), watchdog->name, wd_delay, nretries); 256 return WD_READ_UNSTABLE; 257 258 skip_test: 259 pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n", 260 smp_processor_id(), watchdog->name, wd_seq_delay); 261 pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n", 262 cs->name, wd_delay); 263 return WD_READ_SKIP; 264 } 265 266 static u64 csnow_mid; 267 static cpumask_t cpus_ahead; 268 static cpumask_t cpus_behind; 269 static cpumask_t cpus_chosen; 270 271 static void clocksource_verify_choose_cpus(void) 272 { 273 int cpu, i, n = verify_n_cpus; 274 275 if (n < 0) { 276 /* Check all of the CPUs. */ 277 cpumask_copy(&cpus_chosen, cpu_online_mask); 278 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen); 279 return; 280 } 281 282 /* If no checking desired, or no other CPU to check, leave. */ 283 cpumask_clear(&cpus_chosen); 284 if (n == 0 || num_online_cpus() <= 1) 285 return; 286 287 /* Make sure to select at least one CPU other than the current CPU. */ 288 cpu = cpumask_first(cpu_online_mask); 289 if (cpu == smp_processor_id()) 290 cpu = cpumask_next(cpu, cpu_online_mask); 291 if (WARN_ON_ONCE(cpu >= nr_cpu_ids)) 292 return; 293 cpumask_set_cpu(cpu, &cpus_chosen); 294 295 /* Force a sane value for the boot parameter. */ 296 if (n > nr_cpu_ids) 297 n = nr_cpu_ids; 298 299 /* 300 * Randomly select the specified number of CPUs. If the same 301 * CPU is selected multiple times, that CPU is checked only once, 302 * and no replacement CPU is selected. This gracefully handles 303 * situations where verify_n_cpus is greater than the number of 304 * CPUs that are currently online. 305 */ 306 for (i = 1; i < n; i++) { 307 cpu = prandom_u32() % nr_cpu_ids; 308 cpu = cpumask_next(cpu - 1, cpu_online_mask); 309 if (cpu >= nr_cpu_ids) 310 cpu = cpumask_first(cpu_online_mask); 311 if (!WARN_ON_ONCE(cpu >= nr_cpu_ids)) 312 cpumask_set_cpu(cpu, &cpus_chosen); 313 } 314 315 /* Don't verify ourselves. */ 316 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen); 317 } 318 319 static void clocksource_verify_one_cpu(void *csin) 320 { 321 struct clocksource *cs = (struct clocksource *)csin; 322 323 csnow_mid = cs->read(cs); 324 } 325 326 void clocksource_verify_percpu(struct clocksource *cs) 327 { 328 int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX; 329 u64 csnow_begin, csnow_end; 330 int cpu, testcpu; 331 s64 delta; 332 333 if (verify_n_cpus == 0) 334 return; 335 cpumask_clear(&cpus_ahead); 336 cpumask_clear(&cpus_behind); 337 cpus_read_lock(); 338 preempt_disable(); 339 clocksource_verify_choose_cpus(); 340 if (cpumask_weight(&cpus_chosen) == 0) { 341 preempt_enable(); 342 cpus_read_unlock(); 343 pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name); 344 return; 345 } 346 testcpu = smp_processor_id(); 347 pr_warn("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", cs->name, testcpu, cpumask_pr_args(&cpus_chosen)); 348 for_each_cpu(cpu, &cpus_chosen) { 349 if (cpu == testcpu) 350 continue; 351 csnow_begin = cs->read(cs); 352 smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1); 353 csnow_end = cs->read(cs); 354 delta = (s64)((csnow_mid - csnow_begin) & cs->mask); 355 if (delta < 0) 356 cpumask_set_cpu(cpu, &cpus_behind); 357 delta = (csnow_end - csnow_mid) & cs->mask; 358 if (delta < 0) 359 cpumask_set_cpu(cpu, &cpus_ahead); 360 delta = clocksource_delta(csnow_end, csnow_begin, cs->mask); 361 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift); 362 if (cs_nsec > cs_nsec_max) 363 cs_nsec_max = cs_nsec; 364 if (cs_nsec < cs_nsec_min) 365 cs_nsec_min = cs_nsec; 366 } 367 preempt_enable(); 368 cpus_read_unlock(); 369 if (!cpumask_empty(&cpus_ahead)) 370 pr_warn(" CPUs %*pbl ahead of CPU %d for clocksource %s.\n", 371 cpumask_pr_args(&cpus_ahead), testcpu, cs->name); 372 if (!cpumask_empty(&cpus_behind)) 373 pr_warn(" CPUs %*pbl behind CPU %d for clocksource %s.\n", 374 cpumask_pr_args(&cpus_behind), testcpu, cs->name); 375 if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind)) 376 pr_warn(" CPU %d check durations %lldns - %lldns for clocksource %s.\n", 377 testcpu, cs_nsec_min, cs_nsec_max, cs->name); 378 } 379 EXPORT_SYMBOL_GPL(clocksource_verify_percpu); 380 381 static void clocksource_watchdog(struct timer_list *unused) 382 { 383 u64 csnow, wdnow, cslast, wdlast, delta; 384 int next_cpu, reset_pending; 385 int64_t wd_nsec, cs_nsec; 386 struct clocksource *cs; 387 enum wd_read_status read_ret; 388 u32 md; 389 390 spin_lock(&watchdog_lock); 391 if (!watchdog_running) 392 goto out; 393 394 reset_pending = atomic_read(&watchdog_reset_pending); 395 396 list_for_each_entry(cs, &watchdog_list, wd_list) { 397 398 /* Clocksource already marked unstable? */ 399 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 400 if (finished_booting) 401 schedule_work(&watchdog_work); 402 continue; 403 } 404 405 read_ret = cs_watchdog_read(cs, &csnow, &wdnow); 406 407 if (read_ret != WD_READ_SUCCESS) { 408 if (read_ret == WD_READ_UNSTABLE) 409 /* Clock readout unreliable, so give it up. */ 410 __clocksource_unstable(cs); 411 continue; 412 } 413 414 /* Clocksource initialized ? */ 415 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) || 416 atomic_read(&watchdog_reset_pending)) { 417 cs->flags |= CLOCK_SOURCE_WATCHDOG; 418 cs->wd_last = wdnow; 419 cs->cs_last = csnow; 420 continue; 421 } 422 423 delta = clocksource_delta(wdnow, cs->wd_last, watchdog->mask); 424 wd_nsec = clocksource_cyc2ns(delta, watchdog->mult, 425 watchdog->shift); 426 427 delta = clocksource_delta(csnow, cs->cs_last, cs->mask); 428 cs_nsec = clocksource_cyc2ns(delta, cs->mult, cs->shift); 429 wdlast = cs->wd_last; /* save these in case we print them */ 430 cslast = cs->cs_last; 431 cs->cs_last = csnow; 432 cs->wd_last = wdnow; 433 434 if (atomic_read(&watchdog_reset_pending)) 435 continue; 436 437 /* Check the deviation from the watchdog clocksource. */ 438 md = cs->uncertainty_margin + watchdog->uncertainty_margin; 439 if (abs(cs_nsec - wd_nsec) > md) { 440 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n", 441 smp_processor_id(), cs->name); 442 pr_warn(" '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n", 443 watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask); 444 pr_warn(" '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n", 445 cs->name, cs_nsec, csnow, cslast, cs->mask); 446 if (curr_clocksource == cs) 447 pr_warn(" '%s' is current clocksource.\n", cs->name); 448 else if (curr_clocksource) 449 pr_warn(" '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name); 450 else 451 pr_warn(" No current clocksource.\n"); 452 __clocksource_unstable(cs); 453 continue; 454 } 455 456 if (cs == curr_clocksource && cs->tick_stable) 457 cs->tick_stable(cs); 458 459 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && 460 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) && 461 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) { 462 /* Mark it valid for high-res. */ 463 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 464 465 /* 466 * clocksource_done_booting() will sort it if 467 * finished_booting is not set yet. 468 */ 469 if (!finished_booting) 470 continue; 471 472 /* 473 * If this is not the current clocksource let 474 * the watchdog thread reselect it. Due to the 475 * change to high res this clocksource might 476 * be preferred now. If it is the current 477 * clocksource let the tick code know about 478 * that change. 479 */ 480 if (cs != curr_clocksource) { 481 cs->flags |= CLOCK_SOURCE_RESELECT; 482 schedule_work(&watchdog_work); 483 } else { 484 tick_clock_notify(); 485 } 486 } 487 } 488 489 /* 490 * We only clear the watchdog_reset_pending, when we did a 491 * full cycle through all clocksources. 492 */ 493 if (reset_pending) 494 atomic_dec(&watchdog_reset_pending); 495 496 /* 497 * Cycle through CPUs to check if the CPUs stay synchronized 498 * to each other. 499 */ 500 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask); 501 if (next_cpu >= nr_cpu_ids) 502 next_cpu = cpumask_first(cpu_online_mask); 503 504 /* 505 * Arm timer if not already pending: could race with concurrent 506 * pair clocksource_stop_watchdog() clocksource_start_watchdog(). 507 */ 508 if (!timer_pending(&watchdog_timer)) { 509 watchdog_timer.expires += WATCHDOG_INTERVAL; 510 add_timer_on(&watchdog_timer, next_cpu); 511 } 512 out: 513 spin_unlock(&watchdog_lock); 514 } 515 516 static inline void clocksource_start_watchdog(void) 517 { 518 if (watchdog_running || !watchdog || list_empty(&watchdog_list)) 519 return; 520 timer_setup(&watchdog_timer, clocksource_watchdog, 0); 521 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL; 522 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask)); 523 watchdog_running = 1; 524 } 525 526 static inline void clocksource_stop_watchdog(void) 527 { 528 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list))) 529 return; 530 del_timer(&watchdog_timer); 531 watchdog_running = 0; 532 } 533 534 static inline void clocksource_reset_watchdog(void) 535 { 536 struct clocksource *cs; 537 538 list_for_each_entry(cs, &watchdog_list, wd_list) 539 cs->flags &= ~CLOCK_SOURCE_WATCHDOG; 540 } 541 542 static void clocksource_resume_watchdog(void) 543 { 544 atomic_inc(&watchdog_reset_pending); 545 } 546 547 static void clocksource_enqueue_watchdog(struct clocksource *cs) 548 { 549 INIT_LIST_HEAD(&cs->wd_list); 550 551 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { 552 /* cs is a clocksource to be watched. */ 553 list_add(&cs->wd_list, &watchdog_list); 554 cs->flags &= ~CLOCK_SOURCE_WATCHDOG; 555 } else { 556 /* cs is a watchdog. */ 557 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) 558 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 559 } 560 } 561 562 static void clocksource_select_watchdog(bool fallback) 563 { 564 struct clocksource *cs, *old_wd; 565 unsigned long flags; 566 567 spin_lock_irqsave(&watchdog_lock, flags); 568 /* save current watchdog */ 569 old_wd = watchdog; 570 if (fallback) 571 watchdog = NULL; 572 573 list_for_each_entry(cs, &clocksource_list, list) { 574 /* cs is a clocksource to be watched. */ 575 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) 576 continue; 577 578 /* Skip current if we were requested for a fallback. */ 579 if (fallback && cs == old_wd) 580 continue; 581 582 /* Pick the best watchdog. */ 583 if (!watchdog || cs->rating > watchdog->rating) 584 watchdog = cs; 585 } 586 /* If we failed to find a fallback restore the old one. */ 587 if (!watchdog) 588 watchdog = old_wd; 589 590 /* If we changed the watchdog we need to reset cycles. */ 591 if (watchdog != old_wd) 592 clocksource_reset_watchdog(); 593 594 /* Check if the watchdog timer needs to be started. */ 595 clocksource_start_watchdog(); 596 spin_unlock_irqrestore(&watchdog_lock, flags); 597 } 598 599 static void clocksource_dequeue_watchdog(struct clocksource *cs) 600 { 601 if (cs != watchdog) { 602 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { 603 /* cs is a watched clocksource. */ 604 list_del_init(&cs->wd_list); 605 /* Check if the watchdog timer needs to be stopped. */ 606 clocksource_stop_watchdog(); 607 } 608 } 609 } 610 611 static int __clocksource_watchdog_kthread(void) 612 { 613 struct clocksource *cs, *tmp; 614 unsigned long flags; 615 int select = 0; 616 617 /* Do any required per-CPU skew verification. */ 618 if (curr_clocksource && 619 curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE && 620 curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU) 621 clocksource_verify_percpu(curr_clocksource); 622 623 spin_lock_irqsave(&watchdog_lock, flags); 624 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) { 625 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 626 list_del_init(&cs->wd_list); 627 __clocksource_change_rating(cs, 0); 628 select = 1; 629 } 630 if (cs->flags & CLOCK_SOURCE_RESELECT) { 631 cs->flags &= ~CLOCK_SOURCE_RESELECT; 632 select = 1; 633 } 634 } 635 /* Check if the watchdog timer needs to be stopped. */ 636 clocksource_stop_watchdog(); 637 spin_unlock_irqrestore(&watchdog_lock, flags); 638 639 return select; 640 } 641 642 static int clocksource_watchdog_kthread(void *data) 643 { 644 mutex_lock(&clocksource_mutex); 645 if (__clocksource_watchdog_kthread()) 646 clocksource_select(); 647 mutex_unlock(&clocksource_mutex); 648 return 0; 649 } 650 651 static bool clocksource_is_watchdog(struct clocksource *cs) 652 { 653 return cs == watchdog; 654 } 655 656 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */ 657 658 static void clocksource_enqueue_watchdog(struct clocksource *cs) 659 { 660 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) 661 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 662 } 663 664 static void clocksource_select_watchdog(bool fallback) { } 665 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { } 666 static inline void clocksource_resume_watchdog(void) { } 667 static inline int __clocksource_watchdog_kthread(void) { return 0; } 668 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; } 669 void clocksource_mark_unstable(struct clocksource *cs) { } 670 671 static inline void clocksource_watchdog_lock(unsigned long *flags) { } 672 static inline void clocksource_watchdog_unlock(unsigned long *flags) { } 673 674 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */ 675 676 static bool clocksource_is_suspend(struct clocksource *cs) 677 { 678 return cs == suspend_clocksource; 679 } 680 681 static void __clocksource_suspend_select(struct clocksource *cs) 682 { 683 /* 684 * Skip the clocksource which will be stopped in suspend state. 685 */ 686 if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP)) 687 return; 688 689 /* 690 * The nonstop clocksource can be selected as the suspend clocksource to 691 * calculate the suspend time, so it should not supply suspend/resume 692 * interfaces to suspend the nonstop clocksource when system suspends. 693 */ 694 if (cs->suspend || cs->resume) { 695 pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n", 696 cs->name); 697 } 698 699 /* Pick the best rating. */ 700 if (!suspend_clocksource || cs->rating > suspend_clocksource->rating) 701 suspend_clocksource = cs; 702 } 703 704 /** 705 * clocksource_suspend_select - Select the best clocksource for suspend timing 706 * @fallback: if select a fallback clocksource 707 */ 708 static void clocksource_suspend_select(bool fallback) 709 { 710 struct clocksource *cs, *old_suspend; 711 712 old_suspend = suspend_clocksource; 713 if (fallback) 714 suspend_clocksource = NULL; 715 716 list_for_each_entry(cs, &clocksource_list, list) { 717 /* Skip current if we were requested for a fallback. */ 718 if (fallback && cs == old_suspend) 719 continue; 720 721 __clocksource_suspend_select(cs); 722 } 723 } 724 725 /** 726 * clocksource_start_suspend_timing - Start measuring the suspend timing 727 * @cs: current clocksource from timekeeping 728 * @start_cycles: current cycles from timekeeping 729 * 730 * This function will save the start cycle values of suspend timer to calculate 731 * the suspend time when resuming system. 732 * 733 * This function is called late in the suspend process from timekeeping_suspend(), 734 * that means processes are frozen, non-boot cpus and interrupts are disabled 735 * now. It is therefore possible to start the suspend timer without taking the 736 * clocksource mutex. 737 */ 738 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles) 739 { 740 if (!suspend_clocksource) 741 return; 742 743 /* 744 * If current clocksource is the suspend timer, we should use the 745 * tkr_mono.cycle_last value as suspend_start to avoid same reading 746 * from suspend timer. 747 */ 748 if (clocksource_is_suspend(cs)) { 749 suspend_start = start_cycles; 750 return; 751 } 752 753 if (suspend_clocksource->enable && 754 suspend_clocksource->enable(suspend_clocksource)) { 755 pr_warn_once("Failed to enable the non-suspend-able clocksource.\n"); 756 return; 757 } 758 759 suspend_start = suspend_clocksource->read(suspend_clocksource); 760 } 761 762 /** 763 * clocksource_stop_suspend_timing - Stop measuring the suspend timing 764 * @cs: current clocksource from timekeeping 765 * @cycle_now: current cycles from timekeeping 766 * 767 * This function will calculate the suspend time from suspend timer. 768 * 769 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource. 770 * 771 * This function is called early in the resume process from timekeeping_resume(), 772 * that means there is only one cpu, no processes are running and the interrupts 773 * are disabled. It is therefore possible to stop the suspend timer without 774 * taking the clocksource mutex. 775 */ 776 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now) 777 { 778 u64 now, delta, nsec = 0; 779 780 if (!suspend_clocksource) 781 return 0; 782 783 /* 784 * If current clocksource is the suspend timer, we should use the 785 * tkr_mono.cycle_last value from timekeeping as current cycle to 786 * avoid same reading from suspend timer. 787 */ 788 if (clocksource_is_suspend(cs)) 789 now = cycle_now; 790 else 791 now = suspend_clocksource->read(suspend_clocksource); 792 793 if (now > suspend_start) { 794 delta = clocksource_delta(now, suspend_start, 795 suspend_clocksource->mask); 796 nsec = mul_u64_u32_shr(delta, suspend_clocksource->mult, 797 suspend_clocksource->shift); 798 } 799 800 /* 801 * Disable the suspend timer to save power if current clocksource is 802 * not the suspend timer. 803 */ 804 if (!clocksource_is_suspend(cs) && suspend_clocksource->disable) 805 suspend_clocksource->disable(suspend_clocksource); 806 807 return nsec; 808 } 809 810 /** 811 * clocksource_suspend - suspend the clocksource(s) 812 */ 813 void clocksource_suspend(void) 814 { 815 struct clocksource *cs; 816 817 list_for_each_entry_reverse(cs, &clocksource_list, list) 818 if (cs->suspend) 819 cs->suspend(cs); 820 } 821 822 /** 823 * clocksource_resume - resume the clocksource(s) 824 */ 825 void clocksource_resume(void) 826 { 827 struct clocksource *cs; 828 829 list_for_each_entry(cs, &clocksource_list, list) 830 if (cs->resume) 831 cs->resume(cs); 832 833 clocksource_resume_watchdog(); 834 } 835 836 /** 837 * clocksource_touch_watchdog - Update watchdog 838 * 839 * Update the watchdog after exception contexts such as kgdb so as not 840 * to incorrectly trip the watchdog. This might fail when the kernel 841 * was stopped in code which holds watchdog_lock. 842 */ 843 void clocksource_touch_watchdog(void) 844 { 845 clocksource_resume_watchdog(); 846 } 847 848 /** 849 * clocksource_max_adjustment- Returns max adjustment amount 850 * @cs: Pointer to clocksource 851 * 852 */ 853 static u32 clocksource_max_adjustment(struct clocksource *cs) 854 { 855 u64 ret; 856 /* 857 * We won't try to correct for more than 11% adjustments (110,000 ppm), 858 */ 859 ret = (u64)cs->mult * 11; 860 do_div(ret,100); 861 return (u32)ret; 862 } 863 864 /** 865 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted 866 * @mult: cycle to nanosecond multiplier 867 * @shift: cycle to nanosecond divisor (power of two) 868 * @maxadj: maximum adjustment value to mult (~11%) 869 * @mask: bitmask for two's complement subtraction of non 64 bit counters 870 * @max_cyc: maximum cycle value before potential overflow (does not include 871 * any safety margin) 872 * 873 * NOTE: This function includes a safety margin of 50%, in other words, we 874 * return half the number of nanoseconds the hardware counter can technically 875 * cover. This is done so that we can potentially detect problems caused by 876 * delayed timers or bad hardware, which might result in time intervals that 877 * are larger than what the math used can handle without overflows. 878 */ 879 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc) 880 { 881 u64 max_nsecs, max_cycles; 882 883 /* 884 * Calculate the maximum number of cycles that we can pass to the 885 * cyc2ns() function without overflowing a 64-bit result. 886 */ 887 max_cycles = ULLONG_MAX; 888 do_div(max_cycles, mult+maxadj); 889 890 /* 891 * The actual maximum number of cycles we can defer the clocksource is 892 * determined by the minimum of max_cycles and mask. 893 * Note: Here we subtract the maxadj to make sure we don't sleep for 894 * too long if there's a large negative adjustment. 895 */ 896 max_cycles = min(max_cycles, mask); 897 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift); 898 899 /* return the max_cycles value as well if requested */ 900 if (max_cyc) 901 *max_cyc = max_cycles; 902 903 /* Return 50% of the actual maximum, so we can detect bad values */ 904 max_nsecs >>= 1; 905 906 return max_nsecs; 907 } 908 909 /** 910 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles 911 * @cs: Pointer to clocksource to be updated 912 * 913 */ 914 static inline void clocksource_update_max_deferment(struct clocksource *cs) 915 { 916 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift, 917 cs->maxadj, cs->mask, 918 &cs->max_cycles); 919 } 920 921 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur) 922 { 923 struct clocksource *cs; 924 925 if (!finished_booting || list_empty(&clocksource_list)) 926 return NULL; 927 928 /* 929 * We pick the clocksource with the highest rating. If oneshot 930 * mode is active, we pick the highres valid clocksource with 931 * the best rating. 932 */ 933 list_for_each_entry(cs, &clocksource_list, list) { 934 if (skipcur && cs == curr_clocksource) 935 continue; 936 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES)) 937 continue; 938 return cs; 939 } 940 return NULL; 941 } 942 943 static void __clocksource_select(bool skipcur) 944 { 945 bool oneshot = tick_oneshot_mode_active(); 946 struct clocksource *best, *cs; 947 948 /* Find the best suitable clocksource */ 949 best = clocksource_find_best(oneshot, skipcur); 950 if (!best) 951 return; 952 953 if (!strlen(override_name)) 954 goto found; 955 956 /* Check for the override clocksource. */ 957 list_for_each_entry(cs, &clocksource_list, list) { 958 if (skipcur && cs == curr_clocksource) 959 continue; 960 if (strcmp(cs->name, override_name) != 0) 961 continue; 962 /* 963 * Check to make sure we don't switch to a non-highres 964 * capable clocksource if the tick code is in oneshot 965 * mode (highres or nohz) 966 */ 967 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) { 968 /* Override clocksource cannot be used. */ 969 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 970 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n", 971 cs->name); 972 override_name[0] = 0; 973 } else { 974 /* 975 * The override cannot be currently verified. 976 * Deferring to let the watchdog check. 977 */ 978 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n", 979 cs->name); 980 } 981 } else 982 /* Override clocksource can be used. */ 983 best = cs; 984 break; 985 } 986 987 found: 988 if (curr_clocksource != best && !timekeeping_notify(best)) { 989 pr_info("Switched to clocksource %s\n", best->name); 990 curr_clocksource = best; 991 } 992 } 993 994 /** 995 * clocksource_select - Select the best clocksource available 996 * 997 * Private function. Must hold clocksource_mutex when called. 998 * 999 * Select the clocksource with the best rating, or the clocksource, 1000 * which is selected by userspace override. 1001 */ 1002 static void clocksource_select(void) 1003 { 1004 __clocksource_select(false); 1005 } 1006 1007 static void clocksource_select_fallback(void) 1008 { 1009 __clocksource_select(true); 1010 } 1011 1012 /* 1013 * clocksource_done_booting - Called near the end of core bootup 1014 * 1015 * Hack to avoid lots of clocksource churn at boot time. 1016 * We use fs_initcall because we want this to start before 1017 * device_initcall but after subsys_initcall. 1018 */ 1019 static int __init clocksource_done_booting(void) 1020 { 1021 mutex_lock(&clocksource_mutex); 1022 curr_clocksource = clocksource_default_clock(); 1023 finished_booting = 1; 1024 /* 1025 * Run the watchdog first to eliminate unstable clock sources 1026 */ 1027 __clocksource_watchdog_kthread(); 1028 clocksource_select(); 1029 mutex_unlock(&clocksource_mutex); 1030 return 0; 1031 } 1032 fs_initcall(clocksource_done_booting); 1033 1034 /* 1035 * Enqueue the clocksource sorted by rating 1036 */ 1037 static void clocksource_enqueue(struct clocksource *cs) 1038 { 1039 struct list_head *entry = &clocksource_list; 1040 struct clocksource *tmp; 1041 1042 list_for_each_entry(tmp, &clocksource_list, list) { 1043 /* Keep track of the place, where to insert */ 1044 if (tmp->rating < cs->rating) 1045 break; 1046 entry = &tmp->list; 1047 } 1048 list_add(&cs->list, entry); 1049 } 1050 1051 /** 1052 * __clocksource_update_freq_scale - Used update clocksource with new freq 1053 * @cs: clocksource to be registered 1054 * @scale: Scale factor multiplied against freq to get clocksource hz 1055 * @freq: clocksource frequency (cycles per second) divided by scale 1056 * 1057 * This should only be called from the clocksource->enable() method. 1058 * 1059 * This *SHOULD NOT* be called directly! Please use the 1060 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper 1061 * functions. 1062 */ 1063 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq) 1064 { 1065 u64 sec; 1066 1067 /* 1068 * Default clocksources are *special* and self-define their mult/shift. 1069 * But, you're not special, so you should specify a freq value. 1070 */ 1071 if (freq) { 1072 /* 1073 * Calc the maximum number of seconds which we can run before 1074 * wrapping around. For clocksources which have a mask > 32-bit 1075 * we need to limit the max sleep time to have a good 1076 * conversion precision. 10 minutes is still a reasonable 1077 * amount. That results in a shift value of 24 for a 1078 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to 1079 * ~ 0.06ppm granularity for NTP. 1080 */ 1081 sec = cs->mask; 1082 do_div(sec, freq); 1083 do_div(sec, scale); 1084 if (!sec) 1085 sec = 1; 1086 else if (sec > 600 && cs->mask > UINT_MAX) 1087 sec = 600; 1088 1089 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq, 1090 NSEC_PER_SEC / scale, sec * scale); 1091 } 1092 1093 /* 1094 * If the uncertainty margin is not specified, calculate it. 1095 * If both scale and freq are non-zero, calculate the clock 1096 * period, but bound below at 2*WATCHDOG_MAX_SKEW. However, 1097 * if either of scale or freq is zero, be very conservative and 1098 * take the tens-of-milliseconds WATCHDOG_THRESHOLD value for the 1099 * uncertainty margin. Allow stupidly small uncertainty margins 1100 * to be specified by the caller for testing purposes, but warn 1101 * to discourage production use of this capability. 1102 */ 1103 if (scale && freq && !cs->uncertainty_margin) { 1104 cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq); 1105 if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW) 1106 cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW; 1107 } else if (!cs->uncertainty_margin) { 1108 cs->uncertainty_margin = WATCHDOG_THRESHOLD; 1109 } 1110 WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW); 1111 1112 /* 1113 * Ensure clocksources that have large 'mult' values don't overflow 1114 * when adjusted. 1115 */ 1116 cs->maxadj = clocksource_max_adjustment(cs); 1117 while (freq && ((cs->mult + cs->maxadj < cs->mult) 1118 || (cs->mult - cs->maxadj > cs->mult))) { 1119 cs->mult >>= 1; 1120 cs->shift--; 1121 cs->maxadj = clocksource_max_adjustment(cs); 1122 } 1123 1124 /* 1125 * Only warn for *special* clocksources that self-define 1126 * their mult/shift values and don't specify a freq. 1127 */ 1128 WARN_ONCE(cs->mult + cs->maxadj < cs->mult, 1129 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n", 1130 cs->name); 1131 1132 clocksource_update_max_deferment(cs); 1133 1134 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n", 1135 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns); 1136 } 1137 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale); 1138 1139 /** 1140 * __clocksource_register_scale - Used to install new clocksources 1141 * @cs: clocksource to be registered 1142 * @scale: Scale factor multiplied against freq to get clocksource hz 1143 * @freq: clocksource frequency (cycles per second) divided by scale 1144 * 1145 * Returns -EBUSY if registration fails, zero otherwise. 1146 * 1147 * This *SHOULD NOT* be called directly! Please use the 1148 * clocksource_register_hz() or clocksource_register_khz helper functions. 1149 */ 1150 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq) 1151 { 1152 unsigned long flags; 1153 1154 clocksource_arch_init(cs); 1155 1156 if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX)) 1157 cs->id = CSID_GENERIC; 1158 if (cs->vdso_clock_mode < 0 || 1159 cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) { 1160 pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n", 1161 cs->name, cs->vdso_clock_mode); 1162 cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE; 1163 } 1164 1165 /* Initialize mult/shift and max_idle_ns */ 1166 __clocksource_update_freq_scale(cs, scale, freq); 1167 1168 /* Add clocksource to the clocksource list */ 1169 mutex_lock(&clocksource_mutex); 1170 1171 clocksource_watchdog_lock(&flags); 1172 clocksource_enqueue(cs); 1173 clocksource_enqueue_watchdog(cs); 1174 clocksource_watchdog_unlock(&flags); 1175 1176 clocksource_select(); 1177 clocksource_select_watchdog(false); 1178 __clocksource_suspend_select(cs); 1179 mutex_unlock(&clocksource_mutex); 1180 return 0; 1181 } 1182 EXPORT_SYMBOL_GPL(__clocksource_register_scale); 1183 1184 static void __clocksource_change_rating(struct clocksource *cs, int rating) 1185 { 1186 list_del(&cs->list); 1187 cs->rating = rating; 1188 clocksource_enqueue(cs); 1189 } 1190 1191 /** 1192 * clocksource_change_rating - Change the rating of a registered clocksource 1193 * @cs: clocksource to be changed 1194 * @rating: new rating 1195 */ 1196 void clocksource_change_rating(struct clocksource *cs, int rating) 1197 { 1198 unsigned long flags; 1199 1200 mutex_lock(&clocksource_mutex); 1201 clocksource_watchdog_lock(&flags); 1202 __clocksource_change_rating(cs, rating); 1203 clocksource_watchdog_unlock(&flags); 1204 1205 clocksource_select(); 1206 clocksource_select_watchdog(false); 1207 clocksource_suspend_select(false); 1208 mutex_unlock(&clocksource_mutex); 1209 } 1210 EXPORT_SYMBOL(clocksource_change_rating); 1211 1212 /* 1213 * Unbind clocksource @cs. Called with clocksource_mutex held 1214 */ 1215 static int clocksource_unbind(struct clocksource *cs) 1216 { 1217 unsigned long flags; 1218 1219 if (clocksource_is_watchdog(cs)) { 1220 /* Select and try to install a replacement watchdog. */ 1221 clocksource_select_watchdog(true); 1222 if (clocksource_is_watchdog(cs)) 1223 return -EBUSY; 1224 } 1225 1226 if (cs == curr_clocksource) { 1227 /* Select and try to install a replacement clock source */ 1228 clocksource_select_fallback(); 1229 if (curr_clocksource == cs) 1230 return -EBUSY; 1231 } 1232 1233 if (clocksource_is_suspend(cs)) { 1234 /* 1235 * Select and try to install a replacement suspend clocksource. 1236 * If no replacement suspend clocksource, we will just let the 1237 * clocksource go and have no suspend clocksource. 1238 */ 1239 clocksource_suspend_select(true); 1240 } 1241 1242 clocksource_watchdog_lock(&flags); 1243 clocksource_dequeue_watchdog(cs); 1244 list_del_init(&cs->list); 1245 clocksource_watchdog_unlock(&flags); 1246 1247 return 0; 1248 } 1249 1250 /** 1251 * clocksource_unregister - remove a registered clocksource 1252 * @cs: clocksource to be unregistered 1253 */ 1254 int clocksource_unregister(struct clocksource *cs) 1255 { 1256 int ret = 0; 1257 1258 mutex_lock(&clocksource_mutex); 1259 if (!list_empty(&cs->list)) 1260 ret = clocksource_unbind(cs); 1261 mutex_unlock(&clocksource_mutex); 1262 return ret; 1263 } 1264 EXPORT_SYMBOL(clocksource_unregister); 1265 1266 #ifdef CONFIG_SYSFS 1267 /** 1268 * current_clocksource_show - sysfs interface for current clocksource 1269 * @dev: unused 1270 * @attr: unused 1271 * @buf: char buffer to be filled with clocksource list 1272 * 1273 * Provides sysfs interface for listing current clocksource. 1274 */ 1275 static ssize_t current_clocksource_show(struct device *dev, 1276 struct device_attribute *attr, 1277 char *buf) 1278 { 1279 ssize_t count = 0; 1280 1281 mutex_lock(&clocksource_mutex); 1282 count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name); 1283 mutex_unlock(&clocksource_mutex); 1284 1285 return count; 1286 } 1287 1288 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt) 1289 { 1290 size_t ret = cnt; 1291 1292 /* strings from sysfs write are not 0 terminated! */ 1293 if (!cnt || cnt >= CS_NAME_LEN) 1294 return -EINVAL; 1295 1296 /* strip of \n: */ 1297 if (buf[cnt-1] == '\n') 1298 cnt--; 1299 if (cnt > 0) 1300 memcpy(dst, buf, cnt); 1301 dst[cnt] = 0; 1302 return ret; 1303 } 1304 1305 /** 1306 * current_clocksource_store - interface for manually overriding clocksource 1307 * @dev: unused 1308 * @attr: unused 1309 * @buf: name of override clocksource 1310 * @count: length of buffer 1311 * 1312 * Takes input from sysfs interface for manually overriding the default 1313 * clocksource selection. 1314 */ 1315 static ssize_t current_clocksource_store(struct device *dev, 1316 struct device_attribute *attr, 1317 const char *buf, size_t count) 1318 { 1319 ssize_t ret; 1320 1321 mutex_lock(&clocksource_mutex); 1322 1323 ret = sysfs_get_uname(buf, override_name, count); 1324 if (ret >= 0) 1325 clocksource_select(); 1326 1327 mutex_unlock(&clocksource_mutex); 1328 1329 return ret; 1330 } 1331 static DEVICE_ATTR_RW(current_clocksource); 1332 1333 /** 1334 * unbind_clocksource_store - interface for manually unbinding clocksource 1335 * @dev: unused 1336 * @attr: unused 1337 * @buf: unused 1338 * @count: length of buffer 1339 * 1340 * Takes input from sysfs interface for manually unbinding a clocksource. 1341 */ 1342 static ssize_t unbind_clocksource_store(struct device *dev, 1343 struct device_attribute *attr, 1344 const char *buf, size_t count) 1345 { 1346 struct clocksource *cs; 1347 char name[CS_NAME_LEN]; 1348 ssize_t ret; 1349 1350 ret = sysfs_get_uname(buf, name, count); 1351 if (ret < 0) 1352 return ret; 1353 1354 ret = -ENODEV; 1355 mutex_lock(&clocksource_mutex); 1356 list_for_each_entry(cs, &clocksource_list, list) { 1357 if (strcmp(cs->name, name)) 1358 continue; 1359 ret = clocksource_unbind(cs); 1360 break; 1361 } 1362 mutex_unlock(&clocksource_mutex); 1363 1364 return ret ? ret : count; 1365 } 1366 static DEVICE_ATTR_WO(unbind_clocksource); 1367 1368 /** 1369 * available_clocksource_show - sysfs interface for listing clocksource 1370 * @dev: unused 1371 * @attr: unused 1372 * @buf: char buffer to be filled with clocksource list 1373 * 1374 * Provides sysfs interface for listing registered clocksources 1375 */ 1376 static ssize_t available_clocksource_show(struct device *dev, 1377 struct device_attribute *attr, 1378 char *buf) 1379 { 1380 struct clocksource *src; 1381 ssize_t count = 0; 1382 1383 mutex_lock(&clocksource_mutex); 1384 list_for_each_entry(src, &clocksource_list, list) { 1385 /* 1386 * Don't show non-HRES clocksource if the tick code is 1387 * in one shot mode (highres=on or nohz=on) 1388 */ 1389 if (!tick_oneshot_mode_active() || 1390 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES)) 1391 count += snprintf(buf + count, 1392 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), 1393 "%s ", src->name); 1394 } 1395 mutex_unlock(&clocksource_mutex); 1396 1397 count += snprintf(buf + count, 1398 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n"); 1399 1400 return count; 1401 } 1402 static DEVICE_ATTR_RO(available_clocksource); 1403 1404 static struct attribute *clocksource_attrs[] = { 1405 &dev_attr_current_clocksource.attr, 1406 &dev_attr_unbind_clocksource.attr, 1407 &dev_attr_available_clocksource.attr, 1408 NULL 1409 }; 1410 ATTRIBUTE_GROUPS(clocksource); 1411 1412 static struct bus_type clocksource_subsys = { 1413 .name = "clocksource", 1414 .dev_name = "clocksource", 1415 }; 1416 1417 static struct device device_clocksource = { 1418 .id = 0, 1419 .bus = &clocksource_subsys, 1420 .groups = clocksource_groups, 1421 }; 1422 1423 static int __init init_clocksource_sysfs(void) 1424 { 1425 int error = subsys_system_register(&clocksource_subsys, NULL); 1426 1427 if (!error) 1428 error = device_register(&device_clocksource); 1429 1430 return error; 1431 } 1432 1433 device_initcall(init_clocksource_sysfs); 1434 #endif /* CONFIG_SYSFS */ 1435 1436 /** 1437 * boot_override_clocksource - boot clock override 1438 * @str: override name 1439 * 1440 * Takes a clocksource= boot argument and uses it 1441 * as the clocksource override name. 1442 */ 1443 static int __init boot_override_clocksource(char* str) 1444 { 1445 mutex_lock(&clocksource_mutex); 1446 if (str) 1447 strlcpy(override_name, str, sizeof(override_name)); 1448 mutex_unlock(&clocksource_mutex); 1449 return 1; 1450 } 1451 1452 __setup("clocksource=", boot_override_clocksource); 1453 1454 /** 1455 * boot_override_clock - Compatibility layer for deprecated boot option 1456 * @str: override name 1457 * 1458 * DEPRECATED! Takes a clock= boot argument and uses it 1459 * as the clocksource override name 1460 */ 1461 static int __init boot_override_clock(char* str) 1462 { 1463 if (!strcmp(str, "pmtmr")) { 1464 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n"); 1465 return boot_override_clocksource("acpi_pm"); 1466 } 1467 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n"); 1468 return boot_override_clocksource(str); 1469 } 1470 1471 __setup("clock=", boot_override_clock); 1472