kernel/time/clocksource.c

   1 /*
   2  * linux/kernel/time/clocksource.c
   3  *
   4  * This file contains the functions which manage clocksource drivers.
   5  *
   6  * Copyright (C) 2004, 2005 IBM, John Stultz (johnstul@us.ibm.com)
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 2 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; if not, write to the Free Software
  20  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  21  *
  22  * TODO WishList:
  23  *   o Allow clocksource drivers to be unregistered
  24  *   o get rid of clocksource_jiffies extern
  25  */
  26
  27 #include <linux/clocksource.h>
  28 #include <linux/sysdev.h>
  29 #include <linux/init.h>
  30 #include <linux/module.h>
  31 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
  32 #include <linux/tick.h>
  33
  34 /* XXX - Would like a better way for initializing curr_clocksource */
  35 extern struct clocksource clocksource_jiffies;
  36
  37 /*[Clocksource internal variables]---------
  38  * curr_clocksource:
  39  *      currently selected clocksource. Initialized to clocksource_jiffies.
  40  * next_clocksource:
  41  *      pending next selected clocksource.
  42  * clocksource_list:
  43  *      linked list with the registered clocksources
  44  * clocksource_lock:
  45  *      protects manipulations to curr_clocksource and next_clocksource
  46  *      and the clocksource_list
  47  * override_name:
  48  *      Name of the user-specified clocksource.
  49  */
  50 static struct clocksource *curr_clocksource = &clocksource_jiffies;
  51 static struct clocksource *next_clocksource;
  52 static struct clocksource *clocksource_override;
  53 static LIST_HEAD(clocksource_list);
  54 static DEFINE_SPINLOCK(clocksource_lock);
  55 static char override_name[32];
  56 static int finished_booting;
  57
  58 /* clocksource_done_booting - Called near the end of core bootup
  59  *
  60  * Hack to avoid lots of clocksource churn at boot time.
  61  * We use fs_initcall because we want this to start before
  62  * device_initcall but after subsys_initcall.
  63  */
  64 static int __init clocksource_done_booting(void)
  65 {
  66         finished_booting = 1;
  67         return 0;
  68 }
  69 fs_initcall(clocksource_done_booting);
  70
  71 #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
  72 static LIST_HEAD(watchdog_list);
  73 static struct clocksource *watchdog;
  74 static struct timer_list watchdog_timer;
  75 static DEFINE_SPINLOCK(watchdog_lock);
  76 static cycle_t watchdog_last;
  77 static unsigned long watchdog_resumed;
  78
  79 /*
  80  * Interval: 0.5sec Threshold: 0.0625s
  81  */
  82 #define WATCHDOG_INTERVAL (HZ >> 1)
  83 #define WATCHDOG_THRESHOLD (NSEC_PER_SEC >> 4)
  84
  85 static void clocksource_ratewd(struct clocksource *cs, int64_t delta)
  86 {
  87         if (delta > -WATCHDOG_THRESHOLD && delta < WATCHDOG_THRESHOLD)
  88                 return;
  89
  90         printk(KERN_WARNING "Clocksource %s unstable (delta = %Ld ns)\n",
  91                cs->name, delta);
  92         cs->flags &= ~(CLOCK_SOURCE_VALID_FOR_HRES | CLOCK_SOURCE_WATCHDOG);
  93         clocksource_change_rating(cs, 0);
  94         list_del(&cs->wd_list);
  95 }
  96
  97 static void clocksource_watchdog(unsigned long data)
  98 {
  99         struct clocksource *cs, *tmp;
 100         cycle_t csnow, wdnow;
 101         int64_t wd_nsec, cs_nsec;
 102         int resumed;
 103
 104         spin_lock(&watchdog_lock);
 105
 106         resumed = test_and_clear_bit(0, &watchdog_resumed);
 107
 108         wdnow = watchdog->read();
 109         wd_nsec = cyc2ns(watchdog, (wdnow - watchdog_last) & watchdog->mask);
 110         watchdog_last = wdnow;
 111
 112         list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
 113                 csnow = cs->read();
 114
 115                 if (unlikely(resumed)) {
 116                         cs->wd_last = csnow;
 117                         continue;
 118                 }
 119
 120                 /* Initialized ? */
 121                 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG)) {
 122                         if ((cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
 123                             (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
 124                                 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 125                                 /*
 126                                  * We just marked the clocksource as
 127                                  * highres-capable, notify the rest of the
 128                                  * system as well so that we transition
 129                                  * into high-res mode:
 130                                  */
 131                                 tick_clock_notify();
 132                         }
 133                         cs->flags |= CLOCK_SOURCE_WATCHDOG;
 134                         cs->wd_last = csnow;
 135                 } else {
 136                         cs_nsec = cyc2ns(cs, (csnow - cs->wd_last) & cs->mask);
 137                         cs->wd_last = csnow;
 138                         /* Check the delta. Might remove from the list ! */
 139                         clocksource_ratewd(cs, cs_nsec - wd_nsec);
 140                 }
 141         }
 142
 143         if (!list_empty(&watchdog_list)) {
 144                 /*
 145                  * Cycle through CPUs to check if the CPUs stay
 146                  * synchronized to each other.
 147                  */
 148                 int next_cpu = next_cpu(raw_smp_processor_id(), cpu_online_map);
 149
 150                 if (next_cpu >= NR_CPUS)
 151                         next_cpu = first_cpu(cpu_online_map);
 152                 watchdog_timer.expires += WATCHDOG_INTERVAL;
 153                 add_timer_on(&watchdog_timer, next_cpu);
 154         }
 155         spin_unlock(&watchdog_lock);
 156 }
 157 static void clocksource_resume_watchdog(void)
 158 {
 159         set_bit(0, &watchdog_resumed);
 160 }
 161
 162 static void clocksource_check_watchdog(struct clocksource *cs)
 163 {
 164         struct clocksource *cse;
 165         unsigned long flags;
 166
 167         spin_lock_irqsave(&watchdog_lock, flags);
 168         if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 169                 int started = !list_empty(&watchdog_list);
 170
 171                 list_add(&cs->wd_list, &watchdog_list);
 172                 if (!started && watchdog) {
 173                         watchdog_last = watchdog->read();
 174                         watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
 175                         add_timer_on(&watchdog_timer,
 176                                      first_cpu(cpu_online_map));
 177                 }
 178         } else {
 179                 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 180                         cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 181
 182                 if (!watchdog || cs->rating > watchdog->rating) {
 183                         if (watchdog)
 184                                 del_timer(&watchdog_timer);
 185                         watchdog = cs;
 186                         init_timer(&watchdog_timer);
 187                         watchdog_timer.function = clocksource_watchdog;
 188
 189                         /* Reset watchdog cycles */
 190                         list_for_each_entry(cse, &watchdog_list, wd_list)
 191                                 cse->flags &= ~CLOCK_SOURCE_WATCHDOG;
 192                         /* Start if list is not empty */
 193                         if (!list_empty(&watchdog_list)) {
 194                                 watchdog_last = watchdog->read();
 195                                 watchdog_timer.expires =
 196                                         jiffies + WATCHDOG_INTERVAL;
 197                                 add_timer_on(&watchdog_timer,
 198                                              first_cpu(cpu_online_map));
 199                         }
 200                 }
 201         }
 202         spin_unlock_irqrestore(&watchdog_lock, flags);
 203 }
 204 #else
 205 static void clocksource_check_watchdog(struct clocksource *cs)
 206 {
 207         if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 208                 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 209 }
 210
 211 static inline void clocksource_resume_watchdog(void) { }
 212 #endif
 213
 214 /**
 215  * clocksource_resume - resume the clocksource(s)
 216  */
 217 void clocksource_resume(void)
 218 {
 219         struct clocksource *cs;
 220         unsigned long flags;
 221
 222         spin_lock_irqsave(&clocksource_lock, flags);
 223
 224         list_for_each_entry(cs, &clocksource_list, list) {
 225                 if (cs->resume)
 226                         cs->resume();
 227         }
 228
 229         clocksource_resume_watchdog();
 230
 231         spin_unlock_irqrestore(&clocksource_lock, flags);
 232 }
 233
 234 /**
 235  * clocksource_get_next - Returns the selected clocksource
 236  *
 237  */
 238 struct clocksource *clocksource_get_next(void)
 239 {
 240         unsigned long flags;
 241
 242         spin_lock_irqsave(&clocksource_lock, flags);
 243         if (next_clocksource && finished_booting) {
 244                 curr_clocksource = next_clocksource;
 245                 next_clocksource = NULL;
 246         }
 247         spin_unlock_irqrestore(&clocksource_lock, flags);
 248
 249         return curr_clocksource;
 250 }
 251
 252 /**
 253  * select_clocksource - Selects the best registered clocksource.
 254  *
 255  * Private function. Must hold clocksource_lock when called.
 256  *
 257  * Select the clocksource with the best rating, or the clocksource,
 258  * which is selected by userspace override.
 259  */
 260 static struct clocksource *select_clocksource(void)
 261 {
 262         struct clocksource *next;
 263
 264         if (list_empty(&clocksource_list))
 265                 return NULL;
 266
 267         if (clocksource_override)
 268                 next = clocksource_override;
 269         else
 270                 next = list_entry(clocksource_list.next, struct clocksource,
 271                                   list);
 272
 273         if (next == curr_clocksource)
 274                 return NULL;
 275
 276         return next;
 277 }
 278
 279 /*
 280  * Enqueue the clocksource sorted by rating
 281  */
 282 static int clocksource_enqueue(struct clocksource *c)
 283 {
 284         struct list_head *tmp, *entry = &clocksource_list;
 285
 286         list_for_each(tmp, &clocksource_list) {
 287                 struct clocksource *cs;
 288
 289                 cs = list_entry(tmp, struct clocksource, list);
 290                 if (cs == c)
 291                         return -EBUSY;
 292                 /* Keep track of the place, where to insert */
 293                 if (cs->rating >= c->rating)
 294                         entry = tmp;
 295         }
 296         list_add(&c->list, entry);
 297
 298         if (strlen(c->name) == strlen(override_name) &&
 299             !strcmp(c->name, override_name))
 300                 clocksource_override = c;
 301
 302         return 0;
 303 }
 304
 305 /**
 306  * clocksource_register - Used to install new clocksources
 307  * @t:          clocksource to be registered
 308  *
 309  * Returns -EBUSY if registration fails, zero otherwise.
 310  */
 311 int clocksource_register(struct clocksource *c)
 312 {
 313         unsigned long flags;
 314         int ret;
 315
 316         spin_lock_irqsave(&clocksource_lock, flags);
 317         ret = clocksource_enqueue(c);
 318         if (!ret)
 319                 next_clocksource = select_clocksource();
 320         spin_unlock_irqrestore(&clocksource_lock, flags);
 321         if (!ret)
 322                 clocksource_check_watchdog(c);
 323         return ret;
 324 }
 325 EXPORT_SYMBOL(clocksource_register);
 326
 327 /**
 328  * clocksource_change_rating - Change the rating of a registered clocksource
 329  *
 330  */
 331 void clocksource_change_rating(struct clocksource *cs, int rating)
 332 {
 333         unsigned long flags;
 334
 335         spin_lock_irqsave(&clocksource_lock, flags);
 336         list_del(&cs->list);
 337         cs->rating = rating;
 338         clocksource_enqueue(cs);
 339         next_clocksource = select_clocksource();
 340         spin_unlock_irqrestore(&clocksource_lock, flags);
 341 }
 342
 343 /**
 344  * clocksource_unregister - remove a registered clocksource
 345  */
 346 void clocksource_unregister(struct clocksource *cs)
 347 {
 348         unsigned long flags;
 349
 350         spin_lock_irqsave(&clocksource_lock, flags);
 351         list_del(&cs->list);
 352         if (clocksource_override == cs)
 353                 clocksource_override = NULL;
 354         next_clocksource = select_clocksource();
 355         spin_unlock_irqrestore(&clocksource_lock, flags);
 356 }
 357
 358 #ifdef CONFIG_SYSFS
 359 /**
 360  * sysfs_show_current_clocksources - sysfs interface for current clocksource
 361  * @dev:        unused
 362  * @buf:        char buffer to be filled with clocksource list
 363  *
 364  * Provides sysfs interface for listing current clocksource.
 365  */
 366 static ssize_t
 367 sysfs_show_current_clocksources(struct sys_device *dev, char *buf)
 368 {
 369         ssize_t count = 0;
 370
 371         spin_lock_irq(&clocksource_lock);
 372         count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
 373         spin_unlock_irq(&clocksource_lock);
 374
 375         return count;
 376 }
 377
 378 /**
 379  * sysfs_override_clocksource - interface for manually overriding clocksource
 380  * @dev:        unused
 381  * @buf:        name of override clocksource
 382  * @count:      length of buffer
 383  *
 384  * Takes input from sysfs interface for manually overriding the default
 385  * clocksource selction.
 386  */
 387 static ssize_t sysfs_override_clocksource(struct sys_device *dev,
 388                                           const char *buf, size_t count)
 389 {
 390         struct clocksource *ovr = NULL;
 391         size_t ret = count;
 392         int len;
 393
 394         /* strings from sysfs write are not 0 terminated! */
 395         if (count >= sizeof(override_name))
 396                 return -EINVAL;
 397
 398         /* strip of \n: */
 399         if (buf[count-1] == '\n')
 400                 count--;
 401
 402         spin_lock_irq(&clocksource_lock);
 403
 404         if (count > 0)
 405                 memcpy(override_name, buf, count);
 406         override_name[count] = 0;
 407
 408         len = strlen(override_name);
 409         if (len) {
 410                 struct clocksource *cs;
 411
 412                 ovr = clocksource_override;
 413                 /* try to select it: */
 414                 list_for_each_entry(cs, &clocksource_list, list) {
 415                         if (strlen(cs->name) == len &&
 416                             !strcmp(cs->name, override_name))
 417                                 ovr = cs;
 418                 }
 419         }
 420
 421         /* Reselect, when the override name has changed */
 422         if (ovr != clocksource_override) {
 423                 clocksource_override = ovr;
 424                 next_clocksource = select_clocksource();
 425         }
 426
 427         spin_unlock_irq(&clocksource_lock);
 428
 429         return ret;
 430 }
 431
 432 /**
 433  * sysfs_show_available_clocksources - sysfs interface for listing clocksource
 434  * @dev:        unused
 435  * @buf:        char buffer to be filled with clocksource list
 436  *
 437  * Provides sysfs interface for listing registered clocksources
 438  */
 439 static ssize_t
 440 sysfs_show_available_clocksources(struct sys_device *dev, char *buf)
 441 {
 442         struct clocksource *src;
 443         ssize_t count = 0;
 444
 445         spin_lock_irq(&clocksource_lock);
 446         list_for_each_entry(src, &clocksource_list, list) {
 447                 count += snprintf(buf + count,
 448                                   max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
 449                                   "%s ", src->name);
 450         }
 451         spin_unlock_irq(&clocksource_lock);
 452
 453         count += snprintf(buf + count,
 454                           max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
 455
 456         return count;
 457 }
 458
 459 /*
 460  * Sysfs setup bits:
 461  */
 462 static SYSDEV_ATTR(current_clocksource, 0600, sysfs_show_current_clocksources,
 463                    sysfs_override_clocksource);
 464
 465 static SYSDEV_ATTR(available_clocksource, 0600,
 466                    sysfs_show_available_clocksources, NULL);
 467
 468 static struct sysdev_class clocksource_sysclass = {
 469         .name = "clocksource",
 470 };
 471
 472 static struct sys_device device_clocksource = {
 473         .id     = 0,
 474         .cls    = &clocksource_sysclass,
 475 };
 476
 477 static int __init init_clocksource_sysfs(void)
 478 {
 479         int error = sysdev_class_register(&clocksource_sysclass);
 480
 481         if (!error)
 482                 error = sysdev_register(&device_clocksource);
 483         if (!error)
 484                 error = sysdev_create_file(
 485                                 &device_clocksource,
 486                                 &attr_current_clocksource);
 487         if (!error)
 488                 error = sysdev_create_file(
 489                                 &device_clocksource,
 490                                 &attr_available_clocksource);
 491         return error;
 492 }
 493
 494 device_initcall(init_clocksource_sysfs);
 495 #endif /* CONFIG_SYSFS */
 496
 497 /**
 498  * boot_override_clocksource - boot clock override
 499  * @str:        override name
 500  *
 501  * Takes a clocksource= boot argument and uses it
 502  * as the clocksource override name.
 503  */
 504 static int __init boot_override_clocksource(char* str)
 505 {
 506         unsigned long flags;
 507         spin_lock_irqsave(&clocksource_lock, flags);
 508         if (str)
 509                 strlcpy(override_name, str, sizeof(override_name));
 510         spin_unlock_irqrestore(&clocksource_lock, flags);
 511         return 1;
 512 }
 513
 514 __setup("clocksource=", boot_override_clocksource);
 515
 516 /**
 517  * boot_override_clock - Compatibility layer for deprecated boot option
 518  * @str:        override name
 519  *
 520  * DEPRECATED! Takes a clock= boot argument and uses it
 521  * as the clocksource override name
 522  */
 523 static int __init boot_override_clock(char* str)
 524 {
 525         if (!strcmp(str, "pmtmr")) {
 526                 printk("Warning: clock=pmtmr is deprecated. "
 527                         "Use clocksource=acpi_pm.\n");
 528                 return boot_override_clocksource("acpi_pm");
 529         }
 530         printk("Warning! clock= boot option is deprecated. "
 531                 "Use clocksource=xyz\n");
 532         return boot_override_clocksource(str);
 533 }
 534
 535 __setup("clock=", boot_override_clock);