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 static void clocksource_enqueue(struct clocksource *cs); 24 25 static noinline u64 cycles_to_nsec_safe(struct clocksource *cs, u64 start, u64 end) 26 { 27 u64 delta = clocksource_delta(end, start, cs->mask, cs->max_raw_delta); 28 29 if (likely(delta < cs->max_cycles)) 30 return clocksource_cyc2ns(delta, cs->mult, cs->shift); 31 32 return mul_u64_u32_shr(delta, cs->mult, cs->shift); 33 } 34 35 /** 36 * clocks_calc_mult_shift - calculate mult/shift factors for scaled math of clocks 37 * @mult: pointer to mult variable 38 * @shift: pointer to shift variable 39 * @from: frequency to convert from 40 * @to: frequency to convert to 41 * @maxsec: guaranteed runtime conversion range in seconds 42 * 43 * The function evaluates the shift/mult pair for the scaled math 44 * operations of clocksources and clockevents. 45 * 46 * @to and @from are frequency values in HZ. For clock sources @to is 47 * NSEC_PER_SEC == 1GHz and @from is the counter frequency. For clock 48 * event @to is the counter frequency and @from is NSEC_PER_SEC. 49 * 50 * The @maxsec conversion range argument controls the time frame in 51 * seconds which must be covered by the runtime conversion with the 52 * calculated mult and shift factors. This guarantees that no 64bit 53 * overflow happens when the input value of the conversion is 54 * multiplied with the calculated mult factor. Larger ranges may 55 * reduce the conversion accuracy by choosing smaller mult and shift 56 * factors. 57 */ 58 void 59 clocks_calc_mult_shift(u32 *mult, u32 *shift, u32 from, u32 to, u32 maxsec) 60 { 61 u64 tmp; 62 u32 sft, sftacc= 32; 63 64 /* 65 * Calculate the shift factor which is limiting the conversion 66 * range: 67 */ 68 tmp = ((u64)maxsec * from) >> 32; 69 while (tmp) { 70 tmp >>=1; 71 sftacc--; 72 } 73 74 /* 75 * Find the conversion shift/mult pair which has the best 76 * accuracy and fits the maxsec conversion range: 77 */ 78 for (sft = 32; sft > 0; sft--) { 79 tmp = (u64) to << sft; 80 tmp += from / 2; 81 do_div(tmp, from); 82 if ((tmp >> sftacc) == 0) 83 break; 84 } 85 *mult = tmp; 86 *shift = sft; 87 } 88 EXPORT_SYMBOL_GPL(clocks_calc_mult_shift); 89 90 /*[Clocksource internal variables]--------- 91 * curr_clocksource: 92 * currently selected clocksource. 93 * suspend_clocksource: 94 * used to calculate the suspend time. 95 * clocksource_list: 96 * linked list with the registered clocksources 97 * clocksource_mutex: 98 * protects manipulations to curr_clocksource and the clocksource_list 99 * override_name: 100 * Name of the user-specified clocksource. 101 */ 102 static struct clocksource *curr_clocksource; 103 static struct clocksource *suspend_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 static u64 suspend_start; 109 110 /* 111 * Interval: 0.5sec. 112 */ 113 #define WATCHDOG_INTERVAL (HZ >> 1) 114 #define WATCHDOG_INTERVAL_MAX_NS ((2 * WATCHDOG_INTERVAL) * (NSEC_PER_SEC / HZ)) 115 116 /* 117 * Threshold: 0.0312s, when doubled: 0.0625s. 118 */ 119 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 5) 120 121 /* 122 * Maximum permissible delay between two readouts of the watchdog 123 * clocksource surrounding a read of the clocksource being validated. 124 * This delay could be due to SMIs, NMIs, or to VCPU preemptions. Used as 125 * a lower bound for cs->uncertainty_margin values when registering clocks. 126 * 127 * The default of 500 parts per million is based on NTP's limits. 128 * If a clocksource is good enough for NTP, it is good enough for us! 129 * 130 * In other words, by default, even if a clocksource is extremely 131 * precise (for example, with a sub-nanosecond period), the maximum 132 * permissible skew between the clocksource watchdog and the clocksource 133 * under test is not permitted to go below the 500ppm minimum defined 134 * by MAX_SKEW_USEC. This 500ppm minimum may be overridden using the 135 * CLOCKSOURCE_WATCHDOG_MAX_SKEW_US Kconfig option. 136 */ 137 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US 138 #define MAX_SKEW_USEC CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US 139 #else 140 #define MAX_SKEW_USEC (125 * WATCHDOG_INTERVAL / HZ) 141 #endif 142 143 /* 144 * Default for maximum permissible skew when cs->uncertainty_margin is 145 * not specified, and the lower bound even when cs->uncertainty_margin 146 * is specified. This is also the default that is used when registering 147 * clocks with unspecifed cs->uncertainty_margin, so this macro is used 148 * even in CONFIG_CLOCKSOURCE_WATCHDOG=n kernels. 149 */ 150 #define WATCHDOG_MAX_SKEW (MAX_SKEW_USEC * NSEC_PER_USEC) 151 152 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG 153 static void clocksource_watchdog_work(struct work_struct *work); 154 static void clocksource_select(void); 155 156 static LIST_HEAD(watchdog_list); 157 static struct clocksource *watchdog; 158 static struct timer_list watchdog_timer; 159 static DECLARE_WORK(watchdog_work, clocksource_watchdog_work); 160 static DEFINE_SPINLOCK(watchdog_lock); 161 static int watchdog_running; 162 static atomic_t watchdog_reset_pending; 163 static int64_t watchdog_max_interval; 164 165 static inline void clocksource_watchdog_lock(unsigned long *flags) 166 { 167 spin_lock_irqsave(&watchdog_lock, *flags); 168 } 169 170 static inline void clocksource_watchdog_unlock(unsigned long *flags) 171 { 172 spin_unlock_irqrestore(&watchdog_lock, *flags); 173 } 174 175 static int clocksource_watchdog_kthread(void *data); 176 177 static void clocksource_watchdog_work(struct work_struct *work) 178 { 179 /* 180 * We cannot directly run clocksource_watchdog_kthread() here, because 181 * clocksource_select() calls timekeeping_notify() which uses 182 * stop_machine(). One cannot use stop_machine() from a workqueue() due 183 * lock inversions wrt CPU hotplug. 184 * 185 * Also, we only ever run this work once or twice during the lifetime 186 * of the kernel, so there is no point in creating a more permanent 187 * kthread for this. 188 * 189 * If kthread_run fails the next watchdog scan over the 190 * watchdog_list will find the unstable clock again. 191 */ 192 kthread_run(clocksource_watchdog_kthread, NULL, "kwatchdog"); 193 } 194 195 static void clocksource_change_rating(struct clocksource *cs, int rating) 196 { 197 list_del(&cs->list); 198 cs->rating = rating; 199 clocksource_enqueue(cs); 200 } 201 202 static void __clocksource_unstable(struct clocksource *cs) 203 { 204 cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG); 205 cs->flags |= CLOCK_SOURCE_UNSTABLE; 206 207 /* 208 * If the clocksource is registered clocksource_watchdog_kthread() will 209 * re-rate and re-select. 210 */ 211 if (list_empty(&cs->list)) { 212 cs->rating = 0; 213 return; 214 } 215 216 if (cs->mark_unstable) 217 cs->mark_unstable(cs); 218 219 /* kick clocksource_watchdog_kthread() */ 220 if (finished_booting) 221 schedule_work(&watchdog_work); 222 } 223 224 /** 225 * clocksource_mark_unstable - mark clocksource unstable via watchdog 226 * @cs: clocksource to be marked unstable 227 * 228 * This function is called by the x86 TSC code to mark clocksources as unstable; 229 * it defers demotion and re-selection to a kthread. 230 */ 231 void clocksource_mark_unstable(struct clocksource *cs) 232 { 233 unsigned long flags; 234 235 spin_lock_irqsave(&watchdog_lock, flags); 236 if (!(cs->flags & CLOCK_SOURCE_UNSTABLE)) { 237 if (!list_empty(&cs->list) && list_empty(&cs->wd_list)) 238 list_add(&cs->wd_list, &watchdog_list); 239 __clocksource_unstable(cs); 240 } 241 spin_unlock_irqrestore(&watchdog_lock, flags); 242 } 243 244 static int verify_n_cpus = 8; 245 module_param(verify_n_cpus, int, 0644); 246 247 enum wd_read_status { 248 WD_READ_SUCCESS, 249 WD_READ_UNSTABLE, 250 WD_READ_SKIP 251 }; 252 253 static enum wd_read_status cs_watchdog_read(struct clocksource *cs, u64 *csnow, u64 *wdnow) 254 { 255 int64_t md = 2 * watchdog->uncertainty_margin; 256 unsigned int nretries, max_retries; 257 int64_t wd_delay, wd_seq_delay; 258 u64 wd_end, wd_end2; 259 260 max_retries = clocksource_get_max_watchdog_retry(); 261 for (nretries = 0; nretries <= max_retries; nretries++) { 262 local_irq_disable(); 263 *wdnow = watchdog->read(watchdog); 264 *csnow = cs->read(cs); 265 wd_end = watchdog->read(watchdog); 266 wd_end2 = watchdog->read(watchdog); 267 local_irq_enable(); 268 269 wd_delay = cycles_to_nsec_safe(watchdog, *wdnow, wd_end); 270 if (wd_delay <= md + cs->uncertainty_margin) { 271 if (nretries > 1 && nretries >= max_retries) { 272 pr_warn("timekeeping watchdog on CPU%d: %s retried %d times before success\n", 273 smp_processor_id(), watchdog->name, nretries); 274 } 275 return WD_READ_SUCCESS; 276 } 277 278 /* 279 * Now compute delay in consecutive watchdog read to see if 280 * there is too much external interferences that cause 281 * significant delay in reading both clocksource and watchdog. 282 * 283 * If consecutive WD read-back delay > md, report 284 * system busy, reinit the watchdog and skip the current 285 * watchdog test. 286 */ 287 wd_seq_delay = cycles_to_nsec_safe(watchdog, wd_end, wd_end2); 288 if (wd_seq_delay > md) 289 goto skip_test; 290 } 291 292 pr_warn("timekeeping watchdog on CPU%d: wd-%s-wd excessive read-back delay of %lldns vs. limit of %ldns, wd-wd read-back delay only %lldns, attempt %d, marking %s unstable\n", 293 smp_processor_id(), cs->name, wd_delay, WATCHDOG_MAX_SKEW, wd_seq_delay, nretries, cs->name); 294 return WD_READ_UNSTABLE; 295 296 skip_test: 297 pr_info("timekeeping watchdog on CPU%d: %s wd-wd read-back delay of %lldns\n", 298 smp_processor_id(), watchdog->name, wd_seq_delay); 299 pr_info("wd-%s-wd read-back delay of %lldns, clock-skew test skipped!\n", 300 cs->name, wd_delay); 301 return WD_READ_SKIP; 302 } 303 304 static u64 csnow_mid; 305 static cpumask_t cpus_ahead; 306 static cpumask_t cpus_behind; 307 static cpumask_t cpus_chosen; 308 309 static void clocksource_verify_choose_cpus(void) 310 { 311 int cpu, i, n = verify_n_cpus; 312 313 if (n < 0) { 314 /* Check all of the CPUs. */ 315 cpumask_copy(&cpus_chosen, cpu_online_mask); 316 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen); 317 return; 318 } 319 320 /* If no checking desired, or no other CPU to check, leave. */ 321 cpumask_clear(&cpus_chosen); 322 if (n == 0 || num_online_cpus() <= 1) 323 return; 324 325 /* Make sure to select at least one CPU other than the current CPU. */ 326 cpu = cpumask_first(cpu_online_mask); 327 if (cpu == smp_processor_id()) 328 cpu = cpumask_next(cpu, cpu_online_mask); 329 if (WARN_ON_ONCE(cpu >= nr_cpu_ids)) 330 return; 331 cpumask_set_cpu(cpu, &cpus_chosen); 332 333 /* Force a sane value for the boot parameter. */ 334 if (n > nr_cpu_ids) 335 n = nr_cpu_ids; 336 337 /* 338 * Randomly select the specified number of CPUs. If the same 339 * CPU is selected multiple times, that CPU is checked only once, 340 * and no replacement CPU is selected. This gracefully handles 341 * situations where verify_n_cpus is greater than the number of 342 * CPUs that are currently online. 343 */ 344 for (i = 1; i < n; i++) { 345 cpu = get_random_u32_below(nr_cpu_ids); 346 cpu = cpumask_next(cpu - 1, cpu_online_mask); 347 if (cpu >= nr_cpu_ids) 348 cpu = cpumask_first(cpu_online_mask); 349 if (!WARN_ON_ONCE(cpu >= nr_cpu_ids)) 350 cpumask_set_cpu(cpu, &cpus_chosen); 351 } 352 353 /* Don't verify ourselves. */ 354 cpumask_clear_cpu(smp_processor_id(), &cpus_chosen); 355 } 356 357 static void clocksource_verify_one_cpu(void *csin) 358 { 359 struct clocksource *cs = (struct clocksource *)csin; 360 361 csnow_mid = cs->read(cs); 362 } 363 364 void clocksource_verify_percpu(struct clocksource *cs) 365 { 366 int64_t cs_nsec, cs_nsec_max = 0, cs_nsec_min = LLONG_MAX; 367 u64 csnow_begin, csnow_end; 368 int cpu, testcpu; 369 s64 delta; 370 371 if (verify_n_cpus == 0) 372 return; 373 cpumask_clear(&cpus_ahead); 374 cpumask_clear(&cpus_behind); 375 cpus_read_lock(); 376 preempt_disable(); 377 clocksource_verify_choose_cpus(); 378 if (cpumask_empty(&cpus_chosen)) { 379 preempt_enable(); 380 cpus_read_unlock(); 381 pr_warn("Not enough CPUs to check clocksource '%s'.\n", cs->name); 382 return; 383 } 384 testcpu = smp_processor_id(); 385 pr_info("Checking clocksource %s synchronization from CPU %d to CPUs %*pbl.\n", 386 cs->name, testcpu, cpumask_pr_args(&cpus_chosen)); 387 for_each_cpu(cpu, &cpus_chosen) { 388 if (cpu == testcpu) 389 continue; 390 csnow_begin = cs->read(cs); 391 smp_call_function_single(cpu, clocksource_verify_one_cpu, cs, 1); 392 csnow_end = cs->read(cs); 393 delta = (s64)((csnow_mid - csnow_begin) & cs->mask); 394 if (delta < 0) 395 cpumask_set_cpu(cpu, &cpus_behind); 396 delta = (csnow_end - csnow_mid) & cs->mask; 397 if (delta < 0) 398 cpumask_set_cpu(cpu, &cpus_ahead); 399 cs_nsec = cycles_to_nsec_safe(cs, csnow_begin, csnow_end); 400 if (cs_nsec > cs_nsec_max) 401 cs_nsec_max = cs_nsec; 402 if (cs_nsec < cs_nsec_min) 403 cs_nsec_min = cs_nsec; 404 } 405 preempt_enable(); 406 cpus_read_unlock(); 407 if (!cpumask_empty(&cpus_ahead)) 408 pr_warn(" CPUs %*pbl ahead of CPU %d for clocksource %s.\n", 409 cpumask_pr_args(&cpus_ahead), testcpu, cs->name); 410 if (!cpumask_empty(&cpus_behind)) 411 pr_warn(" CPUs %*pbl behind CPU %d for clocksource %s.\n", 412 cpumask_pr_args(&cpus_behind), testcpu, cs->name); 413 if (!cpumask_empty(&cpus_ahead) || !cpumask_empty(&cpus_behind)) 414 pr_warn(" CPU %d check durations %lldns - %lldns for clocksource %s.\n", 415 testcpu, cs_nsec_min, cs_nsec_max, cs->name); 416 } 417 EXPORT_SYMBOL_GPL(clocksource_verify_percpu); 418 419 static inline void clocksource_reset_watchdog(void) 420 { 421 struct clocksource *cs; 422 423 list_for_each_entry(cs, &watchdog_list, wd_list) 424 cs->flags &= ~CLOCK_SOURCE_WATCHDOG; 425 } 426 427 428 static void clocksource_watchdog(struct timer_list *unused) 429 { 430 int64_t wd_nsec, cs_nsec, interval; 431 u64 csnow, wdnow, cslast, wdlast; 432 int next_cpu, reset_pending; 433 struct clocksource *cs; 434 enum wd_read_status read_ret; 435 unsigned long extra_wait = 0; 436 u32 md; 437 438 spin_lock(&watchdog_lock); 439 if (!watchdog_running) 440 goto out; 441 442 reset_pending = atomic_read(&watchdog_reset_pending); 443 444 list_for_each_entry(cs, &watchdog_list, wd_list) { 445 446 /* Clocksource already marked unstable? */ 447 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 448 if (finished_booting) 449 schedule_work(&watchdog_work); 450 continue; 451 } 452 453 read_ret = cs_watchdog_read(cs, &csnow, &wdnow); 454 455 if (read_ret == WD_READ_UNSTABLE) { 456 /* Clock readout unreliable, so give it up. */ 457 __clocksource_unstable(cs); 458 continue; 459 } 460 461 /* 462 * When WD_READ_SKIP is returned, it means the system is likely 463 * under very heavy load, where the latency of reading 464 * watchdog/clocksource is very big, and affect the accuracy of 465 * watchdog check. So give system some space and suspend the 466 * watchdog check for 5 minutes. 467 */ 468 if (read_ret == WD_READ_SKIP) { 469 /* 470 * As the watchdog timer will be suspended, and 471 * cs->last could keep unchanged for 5 minutes, reset 472 * the counters. 473 */ 474 clocksource_reset_watchdog(); 475 extra_wait = HZ * 300; 476 break; 477 } 478 479 /* Clocksource initialized ? */ 480 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG) || 481 atomic_read(&watchdog_reset_pending)) { 482 cs->flags |= CLOCK_SOURCE_WATCHDOG; 483 cs->wd_last = wdnow; 484 cs->cs_last = csnow; 485 continue; 486 } 487 488 wd_nsec = cycles_to_nsec_safe(watchdog, cs->wd_last, wdnow); 489 cs_nsec = cycles_to_nsec_safe(cs, cs->cs_last, csnow); 490 wdlast = cs->wd_last; /* save these in case we print them */ 491 cslast = cs->cs_last; 492 cs->cs_last = csnow; 493 cs->wd_last = wdnow; 494 495 if (atomic_read(&watchdog_reset_pending)) 496 continue; 497 498 /* 499 * The processing of timer softirqs can get delayed (usually 500 * on account of ksoftirqd not getting to run in a timely 501 * manner), which causes the watchdog interval to stretch. 502 * Skew detection may fail for longer watchdog intervals 503 * on account of fixed margins being used. 504 * Some clocksources, e.g. acpi_pm, cannot tolerate 505 * watchdog intervals longer than a few seconds. 506 */ 507 interval = max(cs_nsec, wd_nsec); 508 if (unlikely(interval > WATCHDOG_INTERVAL_MAX_NS)) { 509 if (system_state > SYSTEM_SCHEDULING && 510 interval > 2 * watchdog_max_interval) { 511 watchdog_max_interval = interval; 512 pr_warn("Long readout interval, skipping watchdog check: cs_nsec: %lld wd_nsec: %lld\n", 513 cs_nsec, wd_nsec); 514 } 515 watchdog_timer.expires = jiffies; 516 continue; 517 } 518 519 /* Check the deviation from the watchdog clocksource. */ 520 md = cs->uncertainty_margin + watchdog->uncertainty_margin; 521 if (abs(cs_nsec - wd_nsec) > md) { 522 s64 cs_wd_msec; 523 s64 wd_msec; 524 u32 wd_rem; 525 526 pr_warn("timekeeping watchdog on CPU%d: Marking clocksource '%s' as unstable because the skew is too large:\n", 527 smp_processor_id(), cs->name); 528 pr_warn(" '%s' wd_nsec: %lld wd_now: %llx wd_last: %llx mask: %llx\n", 529 watchdog->name, wd_nsec, wdnow, wdlast, watchdog->mask); 530 pr_warn(" '%s' cs_nsec: %lld cs_now: %llx cs_last: %llx mask: %llx\n", 531 cs->name, cs_nsec, csnow, cslast, cs->mask); 532 cs_wd_msec = div_s64_rem(cs_nsec - wd_nsec, 1000 * 1000, &wd_rem); 533 wd_msec = div_s64_rem(wd_nsec, 1000 * 1000, &wd_rem); 534 pr_warn(" Clocksource '%s' skewed %lld ns (%lld ms) over watchdog '%s' interval of %lld ns (%lld ms)\n", 535 cs->name, cs_nsec - wd_nsec, cs_wd_msec, watchdog->name, wd_nsec, wd_msec); 536 if (curr_clocksource == cs) 537 pr_warn(" '%s' is current clocksource.\n", cs->name); 538 else if (curr_clocksource) 539 pr_warn(" '%s' (not '%s') is current clocksource.\n", curr_clocksource->name, cs->name); 540 else 541 pr_warn(" No current clocksource.\n"); 542 __clocksource_unstable(cs); 543 continue; 544 } 545 546 if (cs == curr_clocksource && cs->tick_stable) 547 cs->tick_stable(cs); 548 549 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && 550 (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) && 551 (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) { 552 /* Mark it valid for high-res. */ 553 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 554 555 /* 556 * clocksource_done_booting() will sort it if 557 * finished_booting is not set yet. 558 */ 559 if (!finished_booting) 560 continue; 561 562 /* 563 * If this is not the current clocksource let 564 * the watchdog thread reselect it. Due to the 565 * change to high res this clocksource might 566 * be preferred now. If it is the current 567 * clocksource let the tick code know about 568 * that change. 569 */ 570 if (cs != curr_clocksource) { 571 cs->flags |= CLOCK_SOURCE_RESELECT; 572 schedule_work(&watchdog_work); 573 } else { 574 tick_clock_notify(); 575 } 576 } 577 } 578 579 /* 580 * We only clear the watchdog_reset_pending, when we did a 581 * full cycle through all clocksources. 582 */ 583 if (reset_pending) 584 atomic_dec(&watchdog_reset_pending); 585 586 /* 587 * Cycle through CPUs to check if the CPUs stay synchronized 588 * to each other. 589 */ 590 next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask); 591 if (next_cpu >= nr_cpu_ids) 592 next_cpu = cpumask_first(cpu_online_mask); 593 594 /* 595 * Arm timer if not already pending: could race with concurrent 596 * pair clocksource_stop_watchdog() clocksource_start_watchdog(). 597 */ 598 if (!timer_pending(&watchdog_timer)) { 599 watchdog_timer.expires += WATCHDOG_INTERVAL + extra_wait; 600 add_timer_on(&watchdog_timer, next_cpu); 601 } 602 out: 603 spin_unlock(&watchdog_lock); 604 } 605 606 static inline void clocksource_start_watchdog(void) 607 { 608 if (watchdog_running || !watchdog || list_empty(&watchdog_list)) 609 return; 610 timer_setup(&watchdog_timer, clocksource_watchdog, 0); 611 watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL; 612 add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask)); 613 watchdog_running = 1; 614 } 615 616 static inline void clocksource_stop_watchdog(void) 617 { 618 if (!watchdog_running || (watchdog && !list_empty(&watchdog_list))) 619 return; 620 del_timer(&watchdog_timer); 621 watchdog_running = 0; 622 } 623 624 static void clocksource_resume_watchdog(void) 625 { 626 atomic_inc(&watchdog_reset_pending); 627 } 628 629 static void clocksource_enqueue_watchdog(struct clocksource *cs) 630 { 631 INIT_LIST_HEAD(&cs->wd_list); 632 633 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { 634 /* cs is a clocksource to be watched. */ 635 list_add(&cs->wd_list, &watchdog_list); 636 cs->flags &= ~CLOCK_SOURCE_WATCHDOG; 637 } else { 638 /* cs is a watchdog. */ 639 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) 640 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 641 } 642 } 643 644 static void clocksource_select_watchdog(bool fallback) 645 { 646 struct clocksource *cs, *old_wd; 647 unsigned long flags; 648 649 spin_lock_irqsave(&watchdog_lock, flags); 650 /* save current watchdog */ 651 old_wd = watchdog; 652 if (fallback) 653 watchdog = NULL; 654 655 list_for_each_entry(cs, &clocksource_list, list) { 656 /* cs is a clocksource to be watched. */ 657 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) 658 continue; 659 660 /* Skip current if we were requested for a fallback. */ 661 if (fallback && cs == old_wd) 662 continue; 663 664 /* Pick the best watchdog. */ 665 if (!watchdog || cs->rating > watchdog->rating) 666 watchdog = cs; 667 } 668 /* If we failed to find a fallback restore the old one. */ 669 if (!watchdog) 670 watchdog = old_wd; 671 672 /* If we changed the watchdog we need to reset cycles. */ 673 if (watchdog != old_wd) 674 clocksource_reset_watchdog(); 675 676 /* Check if the watchdog timer needs to be started. */ 677 clocksource_start_watchdog(); 678 spin_unlock_irqrestore(&watchdog_lock, flags); 679 } 680 681 static void clocksource_dequeue_watchdog(struct clocksource *cs) 682 { 683 if (cs != watchdog) { 684 if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) { 685 /* cs is a watched clocksource. */ 686 list_del_init(&cs->wd_list); 687 /* Check if the watchdog timer needs to be stopped. */ 688 clocksource_stop_watchdog(); 689 } 690 } 691 } 692 693 static int __clocksource_watchdog_kthread(void) 694 { 695 struct clocksource *cs, *tmp; 696 unsigned long flags; 697 int select = 0; 698 699 /* Do any required per-CPU skew verification. */ 700 if (curr_clocksource && 701 curr_clocksource->flags & CLOCK_SOURCE_UNSTABLE && 702 curr_clocksource->flags & CLOCK_SOURCE_VERIFY_PERCPU) 703 clocksource_verify_percpu(curr_clocksource); 704 705 spin_lock_irqsave(&watchdog_lock, flags); 706 list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) { 707 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 708 list_del_init(&cs->wd_list); 709 clocksource_change_rating(cs, 0); 710 select = 1; 711 } 712 if (cs->flags & CLOCK_SOURCE_RESELECT) { 713 cs->flags &= ~CLOCK_SOURCE_RESELECT; 714 select = 1; 715 } 716 } 717 /* Check if the watchdog timer needs to be stopped. */ 718 clocksource_stop_watchdog(); 719 spin_unlock_irqrestore(&watchdog_lock, flags); 720 721 return select; 722 } 723 724 static int clocksource_watchdog_kthread(void *data) 725 { 726 mutex_lock(&clocksource_mutex); 727 if (__clocksource_watchdog_kthread()) 728 clocksource_select(); 729 mutex_unlock(&clocksource_mutex); 730 return 0; 731 } 732 733 static bool clocksource_is_watchdog(struct clocksource *cs) 734 { 735 return cs == watchdog; 736 } 737 738 #else /* CONFIG_CLOCKSOURCE_WATCHDOG */ 739 740 static void clocksource_enqueue_watchdog(struct clocksource *cs) 741 { 742 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) 743 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES; 744 } 745 746 static void clocksource_select_watchdog(bool fallback) { } 747 static inline void clocksource_dequeue_watchdog(struct clocksource *cs) { } 748 static inline void clocksource_resume_watchdog(void) { } 749 static inline int __clocksource_watchdog_kthread(void) { return 0; } 750 static bool clocksource_is_watchdog(struct clocksource *cs) { return false; } 751 void clocksource_mark_unstable(struct clocksource *cs) { } 752 753 static inline void clocksource_watchdog_lock(unsigned long *flags) { } 754 static inline void clocksource_watchdog_unlock(unsigned long *flags) { } 755 756 #endif /* CONFIG_CLOCKSOURCE_WATCHDOG */ 757 758 static bool clocksource_is_suspend(struct clocksource *cs) 759 { 760 return cs == suspend_clocksource; 761 } 762 763 static void __clocksource_suspend_select(struct clocksource *cs) 764 { 765 /* 766 * Skip the clocksource which will be stopped in suspend state. 767 */ 768 if (!(cs->flags & CLOCK_SOURCE_SUSPEND_NONSTOP)) 769 return; 770 771 /* 772 * The nonstop clocksource can be selected as the suspend clocksource to 773 * calculate the suspend time, so it should not supply suspend/resume 774 * interfaces to suspend the nonstop clocksource when system suspends. 775 */ 776 if (cs->suspend || cs->resume) { 777 pr_warn("Nonstop clocksource %s should not supply suspend/resume interfaces\n", 778 cs->name); 779 } 780 781 /* Pick the best rating. */ 782 if (!suspend_clocksource || cs->rating > suspend_clocksource->rating) 783 suspend_clocksource = cs; 784 } 785 786 /** 787 * clocksource_suspend_select - Select the best clocksource for suspend timing 788 * @fallback: if select a fallback clocksource 789 */ 790 static void clocksource_suspend_select(bool fallback) 791 { 792 struct clocksource *cs, *old_suspend; 793 794 old_suspend = suspend_clocksource; 795 if (fallback) 796 suspend_clocksource = NULL; 797 798 list_for_each_entry(cs, &clocksource_list, list) { 799 /* Skip current if we were requested for a fallback. */ 800 if (fallback && cs == old_suspend) 801 continue; 802 803 __clocksource_suspend_select(cs); 804 } 805 } 806 807 /** 808 * clocksource_start_suspend_timing - Start measuring the suspend timing 809 * @cs: current clocksource from timekeeping 810 * @start_cycles: current cycles from timekeeping 811 * 812 * This function will save the start cycle values of suspend timer to calculate 813 * the suspend time when resuming system. 814 * 815 * This function is called late in the suspend process from timekeeping_suspend(), 816 * that means processes are frozen, non-boot cpus and interrupts are disabled 817 * now. It is therefore possible to start the suspend timer without taking the 818 * clocksource mutex. 819 */ 820 void clocksource_start_suspend_timing(struct clocksource *cs, u64 start_cycles) 821 { 822 if (!suspend_clocksource) 823 return; 824 825 /* 826 * If current clocksource is the suspend timer, we should use the 827 * tkr_mono.cycle_last value as suspend_start to avoid same reading 828 * from suspend timer. 829 */ 830 if (clocksource_is_suspend(cs)) { 831 suspend_start = start_cycles; 832 return; 833 } 834 835 if (suspend_clocksource->enable && 836 suspend_clocksource->enable(suspend_clocksource)) { 837 pr_warn_once("Failed to enable the non-suspend-able clocksource.\n"); 838 return; 839 } 840 841 suspend_start = suspend_clocksource->read(suspend_clocksource); 842 } 843 844 /** 845 * clocksource_stop_suspend_timing - Stop measuring the suspend timing 846 * @cs: current clocksource from timekeeping 847 * @cycle_now: current cycles from timekeeping 848 * 849 * This function will calculate the suspend time from suspend timer. 850 * 851 * Returns nanoseconds since suspend started, 0 if no usable suspend clocksource. 852 * 853 * This function is called early in the resume process from timekeeping_resume(), 854 * that means there is only one cpu, no processes are running and the interrupts 855 * are disabled. It is therefore possible to stop the suspend timer without 856 * taking the clocksource mutex. 857 */ 858 u64 clocksource_stop_suspend_timing(struct clocksource *cs, u64 cycle_now) 859 { 860 u64 now, nsec = 0; 861 862 if (!suspend_clocksource) 863 return 0; 864 865 /* 866 * If current clocksource is the suspend timer, we should use the 867 * tkr_mono.cycle_last value from timekeeping as current cycle to 868 * avoid same reading from suspend timer. 869 */ 870 if (clocksource_is_suspend(cs)) 871 now = cycle_now; 872 else 873 now = suspend_clocksource->read(suspend_clocksource); 874 875 if (now > suspend_start) 876 nsec = cycles_to_nsec_safe(suspend_clocksource, suspend_start, now); 877 878 /* 879 * Disable the suspend timer to save power if current clocksource is 880 * not the suspend timer. 881 */ 882 if (!clocksource_is_suspend(cs) && suspend_clocksource->disable) 883 suspend_clocksource->disable(suspend_clocksource); 884 885 return nsec; 886 } 887 888 /** 889 * clocksource_suspend - suspend the clocksource(s) 890 */ 891 void clocksource_suspend(void) 892 { 893 struct clocksource *cs; 894 895 list_for_each_entry_reverse(cs, &clocksource_list, list) 896 if (cs->suspend) 897 cs->suspend(cs); 898 } 899 900 /** 901 * clocksource_resume - resume the clocksource(s) 902 */ 903 void clocksource_resume(void) 904 { 905 struct clocksource *cs; 906 907 list_for_each_entry(cs, &clocksource_list, list) 908 if (cs->resume) 909 cs->resume(cs); 910 911 clocksource_resume_watchdog(); 912 } 913 914 /** 915 * clocksource_touch_watchdog - Update watchdog 916 * 917 * Update the watchdog after exception contexts such as kgdb so as not 918 * to incorrectly trip the watchdog. This might fail when the kernel 919 * was stopped in code which holds watchdog_lock. 920 */ 921 void clocksource_touch_watchdog(void) 922 { 923 clocksource_resume_watchdog(); 924 } 925 926 /** 927 * clocksource_max_adjustment- Returns max adjustment amount 928 * @cs: Pointer to clocksource 929 * 930 */ 931 static u32 clocksource_max_adjustment(struct clocksource *cs) 932 { 933 u64 ret; 934 /* 935 * We won't try to correct for more than 11% adjustments (110,000 ppm), 936 */ 937 ret = (u64)cs->mult * 11; 938 do_div(ret,100); 939 return (u32)ret; 940 } 941 942 /** 943 * clocks_calc_max_nsecs - Returns maximum nanoseconds that can be converted 944 * @mult: cycle to nanosecond multiplier 945 * @shift: cycle to nanosecond divisor (power of two) 946 * @maxadj: maximum adjustment value to mult (~11%) 947 * @mask: bitmask for two's complement subtraction of non 64 bit counters 948 * @max_cyc: maximum cycle value before potential overflow (does not include 949 * any safety margin) 950 * 951 * NOTE: This function includes a safety margin of 50%, in other words, we 952 * return half the number of nanoseconds the hardware counter can technically 953 * cover. This is done so that we can potentially detect problems caused by 954 * delayed timers or bad hardware, which might result in time intervals that 955 * are larger than what the math used can handle without overflows. 956 */ 957 u64 clocks_calc_max_nsecs(u32 mult, u32 shift, u32 maxadj, u64 mask, u64 *max_cyc) 958 { 959 u64 max_nsecs, max_cycles; 960 961 /* 962 * Calculate the maximum number of cycles that we can pass to the 963 * cyc2ns() function without overflowing a 64-bit result. 964 */ 965 max_cycles = ULLONG_MAX; 966 do_div(max_cycles, mult+maxadj); 967 968 /* 969 * The actual maximum number of cycles we can defer the clocksource is 970 * determined by the minimum of max_cycles and mask. 971 * Note: Here we subtract the maxadj to make sure we don't sleep for 972 * too long if there's a large negative adjustment. 973 */ 974 max_cycles = min(max_cycles, mask); 975 max_nsecs = clocksource_cyc2ns(max_cycles, mult - maxadj, shift); 976 977 /* return the max_cycles value as well if requested */ 978 if (max_cyc) 979 *max_cyc = max_cycles; 980 981 /* Return 50% of the actual maximum, so we can detect bad values */ 982 max_nsecs >>= 1; 983 984 return max_nsecs; 985 } 986 987 /** 988 * clocksource_update_max_deferment - Updates the clocksource max_idle_ns & max_cycles 989 * @cs: Pointer to clocksource to be updated 990 * 991 */ 992 static inline void clocksource_update_max_deferment(struct clocksource *cs) 993 { 994 cs->max_idle_ns = clocks_calc_max_nsecs(cs->mult, cs->shift, 995 cs->maxadj, cs->mask, 996 &cs->max_cycles); 997 998 /* 999 * Threshold for detecting negative motion in clocksource_delta(). 1000 * 1001 * Allow for 0.875 of the counter width so that overly long idle 1002 * sleeps, which go slightly over mask/2, do not trigger the 1003 * negative motion detection. 1004 */ 1005 cs->max_raw_delta = (cs->mask >> 1) + (cs->mask >> 2) + (cs->mask >> 3); 1006 } 1007 1008 static struct clocksource *clocksource_find_best(bool oneshot, bool skipcur) 1009 { 1010 struct clocksource *cs; 1011 1012 if (!finished_booting || list_empty(&clocksource_list)) 1013 return NULL; 1014 1015 /* 1016 * We pick the clocksource with the highest rating. If oneshot 1017 * mode is active, we pick the highres valid clocksource with 1018 * the best rating. 1019 */ 1020 list_for_each_entry(cs, &clocksource_list, list) { 1021 if (skipcur && cs == curr_clocksource) 1022 continue; 1023 if (oneshot && !(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES)) 1024 continue; 1025 return cs; 1026 } 1027 return NULL; 1028 } 1029 1030 static void __clocksource_select(bool skipcur) 1031 { 1032 bool oneshot = tick_oneshot_mode_active(); 1033 struct clocksource *best, *cs; 1034 1035 /* Find the best suitable clocksource */ 1036 best = clocksource_find_best(oneshot, skipcur); 1037 if (!best) 1038 return; 1039 1040 if (!strlen(override_name)) 1041 goto found; 1042 1043 /* Check for the override clocksource. */ 1044 list_for_each_entry(cs, &clocksource_list, list) { 1045 if (skipcur && cs == curr_clocksource) 1046 continue; 1047 if (strcmp(cs->name, override_name) != 0) 1048 continue; 1049 /* 1050 * Check to make sure we don't switch to a non-highres 1051 * capable clocksource if the tick code is in oneshot 1052 * mode (highres or nohz) 1053 */ 1054 if (!(cs->flags & CLOCK_SOURCE_VALID_FOR_HRES) && oneshot) { 1055 /* Override clocksource cannot be used. */ 1056 if (cs->flags & CLOCK_SOURCE_UNSTABLE) { 1057 pr_warn("Override clocksource %s is unstable and not HRT compatible - cannot switch while in HRT/NOHZ mode\n", 1058 cs->name); 1059 override_name[0] = 0; 1060 } else { 1061 /* 1062 * The override cannot be currently verified. 1063 * Deferring to let the watchdog check. 1064 */ 1065 pr_info("Override clocksource %s is not currently HRT compatible - deferring\n", 1066 cs->name); 1067 } 1068 } else 1069 /* Override clocksource can be used. */ 1070 best = cs; 1071 break; 1072 } 1073 1074 found: 1075 if (curr_clocksource != best && !timekeeping_notify(best)) { 1076 pr_info("Switched to clocksource %s\n", best->name); 1077 curr_clocksource = best; 1078 } 1079 } 1080 1081 /** 1082 * clocksource_select - Select the best clocksource available 1083 * 1084 * Private function. Must hold clocksource_mutex when called. 1085 * 1086 * Select the clocksource with the best rating, or the clocksource, 1087 * which is selected by userspace override. 1088 */ 1089 static void clocksource_select(void) 1090 { 1091 __clocksource_select(false); 1092 } 1093 1094 static void clocksource_select_fallback(void) 1095 { 1096 __clocksource_select(true); 1097 } 1098 1099 /* 1100 * clocksource_done_booting - Called near the end of core bootup 1101 * 1102 * Hack to avoid lots of clocksource churn at boot time. 1103 * We use fs_initcall because we want this to start before 1104 * device_initcall but after subsys_initcall. 1105 */ 1106 static int __init clocksource_done_booting(void) 1107 { 1108 mutex_lock(&clocksource_mutex); 1109 curr_clocksource = clocksource_default_clock(); 1110 finished_booting = 1; 1111 /* 1112 * Run the watchdog first to eliminate unstable clock sources 1113 */ 1114 __clocksource_watchdog_kthread(); 1115 clocksource_select(); 1116 mutex_unlock(&clocksource_mutex); 1117 return 0; 1118 } 1119 fs_initcall(clocksource_done_booting); 1120 1121 /* 1122 * Enqueue the clocksource sorted by rating 1123 */ 1124 static void clocksource_enqueue(struct clocksource *cs) 1125 { 1126 struct list_head *entry = &clocksource_list; 1127 struct clocksource *tmp; 1128 1129 list_for_each_entry(tmp, &clocksource_list, list) { 1130 /* Keep track of the place, where to insert */ 1131 if (tmp->rating < cs->rating) 1132 break; 1133 entry = &tmp->list; 1134 } 1135 list_add(&cs->list, entry); 1136 } 1137 1138 /** 1139 * __clocksource_update_freq_scale - Used update clocksource with new freq 1140 * @cs: clocksource to be registered 1141 * @scale: Scale factor multiplied against freq to get clocksource hz 1142 * @freq: clocksource frequency (cycles per second) divided by scale 1143 * 1144 * This should only be called from the clocksource->enable() method. 1145 * 1146 * This *SHOULD NOT* be called directly! Please use the 1147 * __clocksource_update_freq_hz() or __clocksource_update_freq_khz() helper 1148 * functions. 1149 */ 1150 void __clocksource_update_freq_scale(struct clocksource *cs, u32 scale, u32 freq) 1151 { 1152 u64 sec; 1153 1154 /* 1155 * Default clocksources are *special* and self-define their mult/shift. 1156 * But, you're not special, so you should specify a freq value. 1157 */ 1158 if (freq) { 1159 /* 1160 * Calc the maximum number of seconds which we can run before 1161 * wrapping around. For clocksources which have a mask > 32-bit 1162 * we need to limit the max sleep time to have a good 1163 * conversion precision. 10 minutes is still a reasonable 1164 * amount. That results in a shift value of 24 for a 1165 * clocksource with mask >= 40-bit and f >= 4GHz. That maps to 1166 * ~ 0.06ppm granularity for NTP. 1167 */ 1168 sec = cs->mask; 1169 do_div(sec, freq); 1170 do_div(sec, scale); 1171 if (!sec) 1172 sec = 1; 1173 else if (sec > 600 && cs->mask > UINT_MAX) 1174 sec = 600; 1175 1176 clocks_calc_mult_shift(&cs->mult, &cs->shift, freq, 1177 NSEC_PER_SEC / scale, sec * scale); 1178 } 1179 1180 /* 1181 * If the uncertainty margin is not specified, calculate it. If 1182 * both scale and freq are non-zero, calculate the clock period, but 1183 * bound below at 2*WATCHDOG_MAX_SKEW, that is, 500ppm by default. 1184 * However, if either of scale or freq is zero, be very conservative 1185 * and take the tens-of-milliseconds WATCHDOG_THRESHOLD value 1186 * for the uncertainty margin. Allow stupidly small uncertainty 1187 * margins to be specified by the caller for testing purposes, 1188 * but warn to discourage production use of this capability. 1189 * 1190 * Bottom line: The sum of the uncertainty margins of the 1191 * watchdog clocksource and the clocksource under test will be at 1192 * least 500ppm by default. For more information, please see the 1193 * comment preceding CONFIG_CLOCKSOURCE_WATCHDOG_MAX_SKEW_US above. 1194 */ 1195 if (scale && freq && !cs->uncertainty_margin) { 1196 cs->uncertainty_margin = NSEC_PER_SEC / (scale * freq); 1197 if (cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW) 1198 cs->uncertainty_margin = 2 * WATCHDOG_MAX_SKEW; 1199 } else if (!cs->uncertainty_margin) { 1200 cs->uncertainty_margin = WATCHDOG_THRESHOLD; 1201 } 1202 WARN_ON_ONCE(cs->uncertainty_margin < 2 * WATCHDOG_MAX_SKEW); 1203 1204 /* 1205 * Ensure clocksources that have large 'mult' values don't overflow 1206 * when adjusted. 1207 */ 1208 cs->maxadj = clocksource_max_adjustment(cs); 1209 while (freq && ((cs->mult + cs->maxadj < cs->mult) 1210 || (cs->mult - cs->maxadj > cs->mult))) { 1211 cs->mult >>= 1; 1212 cs->shift--; 1213 cs->maxadj = clocksource_max_adjustment(cs); 1214 } 1215 1216 /* 1217 * Only warn for *special* clocksources that self-define 1218 * their mult/shift values and don't specify a freq. 1219 */ 1220 WARN_ONCE(cs->mult + cs->maxadj < cs->mult, 1221 "timekeeping: Clocksource %s might overflow on 11%% adjustment\n", 1222 cs->name); 1223 1224 clocksource_update_max_deferment(cs); 1225 1226 pr_info("%s: mask: 0x%llx max_cycles: 0x%llx, max_idle_ns: %lld ns\n", 1227 cs->name, cs->mask, cs->max_cycles, cs->max_idle_ns); 1228 } 1229 EXPORT_SYMBOL_GPL(__clocksource_update_freq_scale); 1230 1231 /** 1232 * __clocksource_register_scale - Used to install new clocksources 1233 * @cs: clocksource to be registered 1234 * @scale: Scale factor multiplied against freq to get clocksource hz 1235 * @freq: clocksource frequency (cycles per second) divided by scale 1236 * 1237 * Returns -EBUSY if registration fails, zero otherwise. 1238 * 1239 * This *SHOULD NOT* be called directly! Please use the 1240 * clocksource_register_hz() or clocksource_register_khz helper functions. 1241 */ 1242 int __clocksource_register_scale(struct clocksource *cs, u32 scale, u32 freq) 1243 { 1244 unsigned long flags; 1245 1246 clocksource_arch_init(cs); 1247 1248 if (WARN_ON_ONCE((unsigned int)cs->id >= CSID_MAX)) 1249 cs->id = CSID_GENERIC; 1250 if (cs->vdso_clock_mode < 0 || 1251 cs->vdso_clock_mode >= VDSO_CLOCKMODE_MAX) { 1252 pr_warn("clocksource %s registered with invalid VDSO mode %d. Disabling VDSO support.\n", 1253 cs->name, cs->vdso_clock_mode); 1254 cs->vdso_clock_mode = VDSO_CLOCKMODE_NONE; 1255 } 1256 1257 /* Initialize mult/shift and max_idle_ns */ 1258 __clocksource_update_freq_scale(cs, scale, freq); 1259 1260 /* Add clocksource to the clocksource list */ 1261 mutex_lock(&clocksource_mutex); 1262 1263 clocksource_watchdog_lock(&flags); 1264 clocksource_enqueue(cs); 1265 clocksource_enqueue_watchdog(cs); 1266 clocksource_watchdog_unlock(&flags); 1267 1268 clocksource_select(); 1269 clocksource_select_watchdog(false); 1270 __clocksource_suspend_select(cs); 1271 mutex_unlock(&clocksource_mutex); 1272 return 0; 1273 } 1274 EXPORT_SYMBOL_GPL(__clocksource_register_scale); 1275 1276 /* 1277 * Unbind clocksource @cs. Called with clocksource_mutex held 1278 */ 1279 static int clocksource_unbind(struct clocksource *cs) 1280 { 1281 unsigned long flags; 1282 1283 if (clocksource_is_watchdog(cs)) { 1284 /* Select and try to install a replacement watchdog. */ 1285 clocksource_select_watchdog(true); 1286 if (clocksource_is_watchdog(cs)) 1287 return -EBUSY; 1288 } 1289 1290 if (cs == curr_clocksource) { 1291 /* Select and try to install a replacement clock source */ 1292 clocksource_select_fallback(); 1293 if (curr_clocksource == cs) 1294 return -EBUSY; 1295 } 1296 1297 if (clocksource_is_suspend(cs)) { 1298 /* 1299 * Select and try to install a replacement suspend clocksource. 1300 * If no replacement suspend clocksource, we will just let the 1301 * clocksource go and have no suspend clocksource. 1302 */ 1303 clocksource_suspend_select(true); 1304 } 1305 1306 clocksource_watchdog_lock(&flags); 1307 clocksource_dequeue_watchdog(cs); 1308 list_del_init(&cs->list); 1309 clocksource_watchdog_unlock(&flags); 1310 1311 return 0; 1312 } 1313 1314 /** 1315 * clocksource_unregister - remove a registered clocksource 1316 * @cs: clocksource to be unregistered 1317 */ 1318 int clocksource_unregister(struct clocksource *cs) 1319 { 1320 int ret = 0; 1321 1322 mutex_lock(&clocksource_mutex); 1323 if (!list_empty(&cs->list)) 1324 ret = clocksource_unbind(cs); 1325 mutex_unlock(&clocksource_mutex); 1326 return ret; 1327 } 1328 EXPORT_SYMBOL(clocksource_unregister); 1329 1330 #ifdef CONFIG_SYSFS 1331 /** 1332 * current_clocksource_show - sysfs interface for current clocksource 1333 * @dev: unused 1334 * @attr: unused 1335 * @buf: char buffer to be filled with clocksource list 1336 * 1337 * Provides sysfs interface for listing current clocksource. 1338 */ 1339 static ssize_t current_clocksource_show(struct device *dev, 1340 struct device_attribute *attr, 1341 char *buf) 1342 { 1343 ssize_t count = 0; 1344 1345 mutex_lock(&clocksource_mutex); 1346 count = sysfs_emit(buf, "%s\n", curr_clocksource->name); 1347 mutex_unlock(&clocksource_mutex); 1348 1349 return count; 1350 } 1351 1352 ssize_t sysfs_get_uname(const char *buf, char *dst, size_t cnt) 1353 { 1354 size_t ret = cnt; 1355 1356 /* strings from sysfs write are not 0 terminated! */ 1357 if (!cnt || cnt >= CS_NAME_LEN) 1358 return -EINVAL; 1359 1360 /* strip of \n: */ 1361 if (buf[cnt-1] == '\n') 1362 cnt--; 1363 if (cnt > 0) 1364 memcpy(dst, buf, cnt); 1365 dst[cnt] = 0; 1366 return ret; 1367 } 1368 1369 /** 1370 * current_clocksource_store - interface for manually overriding clocksource 1371 * @dev: unused 1372 * @attr: unused 1373 * @buf: name of override clocksource 1374 * @count: length of buffer 1375 * 1376 * Takes input from sysfs interface for manually overriding the default 1377 * clocksource selection. 1378 */ 1379 static ssize_t current_clocksource_store(struct device *dev, 1380 struct device_attribute *attr, 1381 const char *buf, size_t count) 1382 { 1383 ssize_t ret; 1384 1385 mutex_lock(&clocksource_mutex); 1386 1387 ret = sysfs_get_uname(buf, override_name, count); 1388 if (ret >= 0) 1389 clocksource_select(); 1390 1391 mutex_unlock(&clocksource_mutex); 1392 1393 return ret; 1394 } 1395 static DEVICE_ATTR_RW(current_clocksource); 1396 1397 /** 1398 * unbind_clocksource_store - interface for manually unbinding clocksource 1399 * @dev: unused 1400 * @attr: unused 1401 * @buf: unused 1402 * @count: length of buffer 1403 * 1404 * Takes input from sysfs interface for manually unbinding a clocksource. 1405 */ 1406 static ssize_t unbind_clocksource_store(struct device *dev, 1407 struct device_attribute *attr, 1408 const char *buf, size_t count) 1409 { 1410 struct clocksource *cs; 1411 char name[CS_NAME_LEN]; 1412 ssize_t ret; 1413 1414 ret = sysfs_get_uname(buf, name, count); 1415 if (ret < 0) 1416 return ret; 1417 1418 ret = -ENODEV; 1419 mutex_lock(&clocksource_mutex); 1420 list_for_each_entry(cs, &clocksource_list, list) { 1421 if (strcmp(cs->name, name)) 1422 continue; 1423 ret = clocksource_unbind(cs); 1424 break; 1425 } 1426 mutex_unlock(&clocksource_mutex); 1427 1428 return ret ? ret : count; 1429 } 1430 static DEVICE_ATTR_WO(unbind_clocksource); 1431 1432 /** 1433 * available_clocksource_show - sysfs interface for listing clocksource 1434 * @dev: unused 1435 * @attr: unused 1436 * @buf: char buffer to be filled with clocksource list 1437 * 1438 * Provides sysfs interface for listing registered clocksources 1439 */ 1440 static ssize_t available_clocksource_show(struct device *dev, 1441 struct device_attribute *attr, 1442 char *buf) 1443 { 1444 struct clocksource *src; 1445 ssize_t count = 0; 1446 1447 mutex_lock(&clocksource_mutex); 1448 list_for_each_entry(src, &clocksource_list, list) { 1449 /* 1450 * Don't show non-HRES clocksource if the tick code is 1451 * in one shot mode (highres=on or nohz=on) 1452 */ 1453 if (!tick_oneshot_mode_active() || 1454 (src->flags & CLOCK_SOURCE_VALID_FOR_HRES)) 1455 count += snprintf(buf + count, 1456 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), 1457 "%s ", src->name); 1458 } 1459 mutex_unlock(&clocksource_mutex); 1460 1461 count += snprintf(buf + count, 1462 max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n"); 1463 1464 return count; 1465 } 1466 static DEVICE_ATTR_RO(available_clocksource); 1467 1468 static struct attribute *clocksource_attrs[] = { 1469 &dev_attr_current_clocksource.attr, 1470 &dev_attr_unbind_clocksource.attr, 1471 &dev_attr_available_clocksource.attr, 1472 NULL 1473 }; 1474 ATTRIBUTE_GROUPS(clocksource); 1475 1476 static const struct bus_type clocksource_subsys = { 1477 .name = "clocksource", 1478 .dev_name = "clocksource", 1479 }; 1480 1481 static struct device device_clocksource = { 1482 .id = 0, 1483 .bus = &clocksource_subsys, 1484 .groups = clocksource_groups, 1485 }; 1486 1487 static int __init init_clocksource_sysfs(void) 1488 { 1489 int error = subsys_system_register(&clocksource_subsys, NULL); 1490 1491 if (!error) 1492 error = device_register(&device_clocksource); 1493 1494 return error; 1495 } 1496 1497 device_initcall(init_clocksource_sysfs); 1498 #endif /* CONFIG_SYSFS */ 1499 1500 /** 1501 * boot_override_clocksource - boot clock override 1502 * @str: override name 1503 * 1504 * Takes a clocksource= boot argument and uses it 1505 * as the clocksource override name. 1506 */ 1507 static int __init boot_override_clocksource(char* str) 1508 { 1509 mutex_lock(&clocksource_mutex); 1510 if (str) 1511 strscpy(override_name, str, sizeof(override_name)); 1512 mutex_unlock(&clocksource_mutex); 1513 return 1; 1514 } 1515 1516 __setup("clocksource=", boot_override_clocksource); 1517 1518 /** 1519 * boot_override_clock - Compatibility layer for deprecated boot option 1520 * @str: override name 1521 * 1522 * DEPRECATED! Takes a clock= boot argument and uses it 1523 * as the clocksource override name 1524 */ 1525 static int __init boot_override_clock(char* str) 1526 { 1527 if (!strcmp(str, "pmtmr")) { 1528 pr_warn("clock=pmtmr is deprecated - use clocksource=acpi_pm\n"); 1529 return boot_override_clocksource("acpi_pm"); 1530 } 1531 pr_warn("clock= boot option is deprecated - use clocksource=xyz\n"); 1532 return boot_override_clocksource(str); 1533 } 1534 1535 __setup("clock=", boot_override_clock); 1536