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         cs->flags &= ~CLOCK_SOURCE_WATCHDOG;
  95         list_del(&cs->wd_list);
  96 }
  97
  98 static void clocksource_watchdog(unsigned long data)
  99 {
 100         struct clocksource *cs, *tmp;
 101         cycle_t csnow, wdnow;
 102         int64_t wd_nsec, cs_nsec;
 103         int resumed;
 104
 105         spin_lock(&watchdog_lock);
 106
 107         resumed = test_and_clear_bit(0, &watchdog_resumed);
 108
 109         wdnow = watchdog->read();
 110         wd_nsec = cyc2ns(watchdog, (wdnow - watchdog_last) & watchdog->mask);
 111         watchdog_last = wdnow;
 112
 113         list_for_each_entry_safe(cs, tmp, &watchdog_list, wd_list) {
 114                 csnow = cs->read();
 115
 116                 if (unlikely(resumed)) {
 117                         cs->wd_last = csnow;
 118                         continue;
 119                 }
 120
 121                 /* Initialized ? */
 122                 if (!(cs->flags & CLOCK_SOURCE_WATCHDOG)) {
 123                         if ((cs->flags & CLOCK_SOURCE_IS_CONTINUOUS) &&
 124                             (watchdog->flags & CLOCK_SOURCE_IS_CONTINUOUS)) {
 125                                 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 126                                 /*
 127                                  * We just marked the clocksource as
 128                                  * highres-capable, notify the rest of the
 129                                  * system as well so that we transition
 130                                  * into high-res mode:
 131                                  */
 132                                 tick_clock_notify();
 133                         }
 134                         cs->flags |= CLOCK_SOURCE_WATCHDOG;
 135                         cs->wd_last = csnow;
 136                 } else {
 137                         cs_nsec = cyc2ns(cs, (csnow - cs->wd_last) & cs->mask);
 138                         cs->wd_last = csnow;
 139                         /* Check the delta. Might remove from the list ! */
 140                         clocksource_ratewd(cs, cs_nsec - wd_nsec);
 141                 }
 142         }
 143
 144         if (!list_empty(&watchdog_list)) {
 145                 /* Cycle through CPUs to check if the CPUs stay synchronized to
 146                  * each other. */
 147                 int next_cpu = next_cpu(raw_smp_processor_id(), cpu_online_map);
 148                 if (next_cpu >= NR_CPUS)
 149                         next_cpu = first_cpu(cpu_online_map);
 150                 watchdog_timer.expires += WATCHDOG_INTERVAL;
 151                 add_timer_on(&watchdog_timer, next_cpu);
 152         }
 153         spin_unlock(&watchdog_lock);
 154 }
 155 static void clocksource_resume_watchdog(void)
 156 {
 157         set_bit(0, &watchdog_resumed);
 158 }
 159
 160 static void clocksource_check_watchdog(struct clocksource *cs)
 161 {
 162         struct clocksource *cse;
 163         unsigned long flags;
 164
 165         spin_lock_irqsave(&watchdog_lock, flags);
 166         if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 167                 int started = !list_empty(&watchdog_list);
 168
 169                 list_add(&cs->wd_list, &watchdog_list);
 170                 if (!started && watchdog) {
 171                         watchdog_last = watchdog->read();
 172                         watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
 173                         add_timer_on(&watchdog_timer, first_cpu(cpu_online_map));
 174                 }
 175         } else {
 176                 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 177                         cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 178
 179                 if (!watchdog || cs->rating > watchdog->rating) {
 180                         if (watchdog)
 181                                 del_timer(&watchdog_timer);
 182                         watchdog = cs;
 183                         init_timer_deferrable(&watchdog_timer);
 184                         watchdog_timer.function = clocksource_watchdog;
 185
 186                         /* Reset watchdog cycles */
 187                         list_for_each_entry(cse, &watchdog_list, wd_list)
 188                                 cse->flags &= ~CLOCK_SOURCE_WATCHDOG;
 189                         /* Start if list is not empty */
 190                         if (!list_empty(&watchdog_list)) {
 191                                 watchdog_last = watchdog->read();
 192                                 watchdog_timer.expires =
 193                                         jiffies + WATCHDOG_INTERVAL;
 194                                 add_timer_on(&watchdog_timer,
 195                                                 first_cpu(cpu_online_map));
 196                         }
 197                 }
 198         }
 199         spin_unlock_irqrestore(&watchdog_lock, flags);
 200 }
 201 #else
 202 static void clocksource_check_watchdog(struct clocksource *cs)
 203 {
 204         if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 205                 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 206 }
 207
 208 static inline void clocksource_resume_watchdog(void) { }
 209 #endif
 210
 211 /**
 212  * clocksource_resume - resume the clocksource(s)
 213  */
 214 void clocksource_resume(void)
 215 {
 216         struct clocksource *cs;
 217         unsigned long flags;
 218
 219         spin_lock_irqsave(&clocksource_lock, flags);
 220
 221         list_for_each_entry(cs, &clocksource_list, list) {
 222                 if (cs->resume)
 223                         cs->resume();
 224         }
 225
 226         clocksource_resume_watchdog();
 227
 228         spin_unlock_irqrestore(&clocksource_lock, flags);
 229 }
 230
 231 /**
 232  * clocksource_get_next - Returns the selected clocksource
 233  *
 234  */
 235 struct clocksource *clocksource_get_next(void)
 236 {
 237         unsigned long flags;
 238
 239         spin_lock_irqsave(&clocksource_lock, flags);
 240         if (next_clocksource && finished_booting) {
 241                 curr_clocksource = next_clocksource;
 242                 next_clocksource = NULL;
 243         }
 244         spin_unlock_irqrestore(&clocksource_lock, flags);
 245
 246         return curr_clocksource;
 247 }
 248
 249 /**
 250  * select_clocksource - Selects the best registered clocksource.
 251  *
 252  * Private function. Must hold clocksource_lock when called.
 253  *
 254  * Select the clocksource with the best rating, or the clocksource,
 255  * which is selected by userspace override.
 256  */
 257 static struct clocksource *select_clocksource(void)
 258 {
 259         struct clocksource *next;
 260
 261         if (list_empty(&clocksource_list))
 262                 return NULL;
 263
 264         if (clocksource_override)
 265                 next = clocksource_override;
 266         else
 267                 next = list_entry(clocksource_list.next, struct clocksource,
 268                                   list);
 269
 270         if (next == curr_clocksource)
 271                 return NULL;
 272
 273         return next;
 274 }
 275
 276 /*
 277  * Enqueue the clocksource sorted by rating
 278  */
 279 static int clocksource_enqueue(struct clocksource *c)
 280 {
 281         struct list_head *tmp, *entry = &clocksource_list;
 282
 283         list_for_each(tmp, &clocksource_list) {
 284                 struct clocksource *cs;
 285
 286                 cs = list_entry(tmp, struct clocksource, list);
 287                 if (cs == c)
 288                         return -EBUSY;
 289                 /* Keep track of the place, where to insert */
 290                 if (cs->rating >= c->rating)
 291                         entry = tmp;
 292         }
 293         list_add(&c->list, entry);
 294
 295         if (strlen(c->name) == strlen(override_name) &&
 296             !strcmp(c->name, override_name))
 297                 clocksource_override = c;
 298
 299         return 0;
 300 }
 301
 302 /**
 303  * clocksource_register - Used to install new clocksources
 304  * @t:          clocksource to be registered
 305  *
 306  * Returns -EBUSY if registration fails, zero otherwise.
 307  */
 308 int clocksource_register(struct clocksource *c)
 309 {
 310         unsigned long flags;
 311         int ret;
 312
 313         spin_lock_irqsave(&clocksource_lock, flags);
 314         ret = clocksource_enqueue(c);
 315         if (!ret)
 316                 next_clocksource = select_clocksource();
 317         spin_unlock_irqrestore(&clocksource_lock, flags);
 318         if (!ret)
 319                 clocksource_check_watchdog(c);
 320         return ret;
 321 }
 322 EXPORT_SYMBOL(clocksource_register);
 323
 324 /**
 325  * clocksource_change_rating - Change the rating of a registered clocksource
 326  *
 327  */
 328 void clocksource_change_rating(struct clocksource *cs, int rating)
 329 {
 330         unsigned long flags;
 331
 332         spin_lock_irqsave(&clocksource_lock, flags);
 333         list_del(&cs->list);
 334         cs->rating = rating;
 335         clocksource_enqueue(cs);
 336         next_clocksource = select_clocksource();
 337         spin_unlock_irqrestore(&clocksource_lock, flags);
 338 }
 339
 340 /**
 341  * clocksource_unregister - remove a registered clocksource
 342  */
 343 void clocksource_unregister(struct clocksource *cs)
 344 {
 345         unsigned long flags;
 346
 347         spin_lock_irqsave(&clocksource_lock, flags);
 348         list_del(&cs->list);
 349         if (clocksource_override == cs)
 350                 clocksource_override = NULL;
 351         next_clocksource = select_clocksource();
 352         spin_unlock_irqrestore(&clocksource_lock, flags);
 353 }
 354
 355 #ifdef CONFIG_SYSFS
 356 /**
 357  * sysfs_show_current_clocksources - sysfs interface for current clocksource
 358  * @dev:        unused
 359  * @buf:        char buffer to be filled with clocksource list
 360  *
 361  * Provides sysfs interface for listing current clocksource.
 362  */
 363 static ssize_t
 364 sysfs_show_current_clocksources(struct sys_device *dev, char *buf)
 365 {
 366         char *curr = buf;
 367
 368         spin_lock_irq(&clocksource_lock);
 369         curr += sprintf(curr, "%s ", curr_clocksource->name);
 370         spin_unlock_irq(&clocksource_lock);
 371
 372         curr += sprintf(curr, "\n");
 373
 374         return curr - buf;
 375 }
 376
 377 /**
 378  * sysfs_override_clocksource - interface for manually overriding clocksource
 379  * @dev:        unused
 380  * @buf:        name of override clocksource
 381  * @count:      length of buffer
 382  *
 383  * Takes input from sysfs interface for manually overriding the default
 384  * clocksource selction.
 385  */
 386 static ssize_t sysfs_override_clocksource(struct sys_device *dev,
 387                                           const char *buf, size_t count)
 388 {
 389         struct clocksource *ovr = NULL;
 390         size_t ret = count;
 391         int len;
 392
 393         /* strings from sysfs write are not 0 terminated! */
 394         if (count >= sizeof(override_name))
 395                 return -EINVAL;
 396
 397         /* strip of \n: */
 398         if (buf[count-1] == '\n')
 399                 count--;
 400
 401         spin_lock_irq(&clocksource_lock);
 402
 403         if (count > 0)
 404                 memcpy(override_name, buf, count);
 405         override_name[count] = 0;
 406
 407         len = strlen(override_name);
 408         if (len) {
 409                 struct clocksource *cs;
 410
 411                 ovr = clocksource_override;
 412                 /* try to select it: */
 413                 list_for_each_entry(cs, &clocksource_list, list) {
 414                         if (strlen(cs->name) == len &&
 415                             !strcmp(cs->name, override_name))
 416                                 ovr = cs;
 417                 }
 418         }
 419
 420         /* Reselect, when the override name has changed */
 421         if (ovr != clocksource_override) {
 422                 clocksource_override = ovr;
 423                 next_clocksource = select_clocksource();
 424         }
 425
 426         spin_unlock_irq(&clocksource_lock);
 427
 428         return ret;
 429 }
 430
 431 /**
 432  * sysfs_show_available_clocksources - sysfs interface for listing clocksource
 433  * @dev:        unused
 434  * @buf:        char buffer to be filled with clocksource list
 435  *
 436  * Provides sysfs interface for listing registered clocksources
 437  */
 438 static ssize_t
 439 sysfs_show_available_clocksources(struct sys_device *dev, char *buf)
 440 {
 441         struct clocksource *src;
 442         char *curr = buf;
 443
 444         spin_lock_irq(&clocksource_lock);
 445         list_for_each_entry(src, &clocksource_list, list) {
 446                 curr += sprintf(curr, "%s ", src->name);
 447         }
 448         spin_unlock_irq(&clocksource_lock);
 449
 450         curr += sprintf(curr, "\n");
 451
 452         return curr - buf;
 453 }
 454
 455 /*
 456  * Sysfs setup bits:
 457  */
 458 static SYSDEV_ATTR(current_clocksource, 0600, sysfs_show_current_clocksources,
 459                    sysfs_override_clocksource);
 460
 461 static SYSDEV_ATTR(available_clocksource, 0600,
 462                    sysfs_show_available_clocksources, NULL);
 463
 464 static struct sysdev_class clocksource_sysclass = {
 465         .name = "clocksource",
 466 };
 467
 468 static struct sys_device device_clocksource = {
 469         .id     = 0,
 470         .cls    = &clocksource_sysclass,
 471 };
 472
 473 static int __init init_clocksource_sysfs(void)
 474 {
 475         int error = sysdev_class_register(&clocksource_sysclass);
 476
 477         if (!error)
 478                 error = sysdev_register(&device_clocksource);
 479         if (!error)
 480                 error = sysdev_create_file(
 481                                 &device_clocksource,
 482                                 &attr_current_clocksource);
 483         if (!error)
 484                 error = sysdev_create_file(
 485                                 &device_clocksource,
 486                                 &attr_available_clocksource);
 487         return error;
 488 }
 489
 490 device_initcall(init_clocksource_sysfs);
 491 #endif /* CONFIG_SYSFS */
 492
 493 /**
 494  * boot_override_clocksource - boot clock override
 495  * @str:        override name
 496  *
 497  * Takes a clocksource= boot argument and uses it
 498  * as the clocksource override name.
 499  */
 500 static int __init boot_override_clocksource(char* str)
 501 {
 502         unsigned long flags;
 503         spin_lock_irqsave(&clocksource_lock, flags);
 504         if (str)
 505                 strlcpy(override_name, str, sizeof(override_name));
 506         spin_unlock_irqrestore(&clocksource_lock, flags);
 507         return 1;
 508 }
 509
 510 __setup("clocksource=", boot_override_clocksource);
 511
 512 /**
 513  * boot_override_clock - Compatibility layer for deprecated boot option
 514  * @str:        override name
 515  *
 516  * DEPRECATED! Takes a clock= boot argument and uses it
 517  * as the clocksource override name
 518  */
 519 static int __init boot_override_clock(char* str)
 520 {
 521         if (!strcmp(str, "pmtmr")) {
 522                 printk("Warning: clock=pmtmr is deprecated. "
 523                         "Use clocksource=acpi_pm.\n");
 524                 return boot_override_clocksource("acpi_pm");
 525         }
 526         printk("Warning! clock= boot option is deprecated. "
 527                 "Use clocksource=xyz\n");
 528         return boot_override_clocksource(str);
 529 }
 530
 531 __setup("clock=", boot_override_clock);