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 = cpumask_next(raw_smp_processor_id(),
 149                                             cpu_online_mask);
 150
 151                 if (next_cpu >= nr_cpu_ids)
 152                         next_cpu = cpumask_first(cpu_online_mask);
 153                 watchdog_timer.expires += WATCHDOG_INTERVAL;
 154                 add_timer_on(&watchdog_timer, next_cpu);
 155         }
 156         spin_unlock(&watchdog_lock);
 157 }
 158 static void clocksource_resume_watchdog(void)
 159 {
 160         set_bit(0, &watchdog_resumed);
 161 }
 162
 163 static void clocksource_check_watchdog(struct clocksource *cs)
 164 {
 165         struct clocksource *cse;
 166         unsigned long flags;
 167
 168         spin_lock_irqsave(&watchdog_lock, flags);
 169         if (cs->flags & CLOCK_SOURCE_MUST_VERIFY) {
 170                 int started = !list_empty(&watchdog_list);
 171
 172                 list_add(&cs->wd_list, &watchdog_list);
 173                 if (!started && watchdog) {
 174                         watchdog_last = watchdog->read();
 175                         watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
 176                         add_timer_on(&watchdog_timer,
 177                                      cpumask_first(cpu_online_mask));
 178                 }
 179         } else {
 180                 if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 181                         cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 182
 183                 if (!watchdog || cs->rating > watchdog->rating) {
 184                         if (watchdog)
 185                                 del_timer(&watchdog_timer);
 186                         watchdog = cs;
 187                         init_timer(&watchdog_timer);
 188                         watchdog_timer.function = clocksource_watchdog;
 189
 190                         /* Reset watchdog cycles */
 191                         list_for_each_entry(cse, &watchdog_list, wd_list)
 192                                 cse->flags &= ~CLOCK_SOURCE_WATCHDOG;
 193                         /* Start if list is not empty */
 194                         if (!list_empty(&watchdog_list)) {
 195                                 watchdog_last = watchdog->read();
 196                                 watchdog_timer.expires =
 197                                         jiffies + WATCHDOG_INTERVAL;
 198                                 add_timer_on(&watchdog_timer,
 199                                              cpumask_first(cpu_online_mask));
 200                         }
 201                 }
 202         }
 203         spin_unlock_irqrestore(&watchdog_lock, flags);
 204 }
 205 #else
 206 static void clocksource_check_watchdog(struct clocksource *cs)
 207 {
 208         if (cs->flags & CLOCK_SOURCE_IS_CONTINUOUS)
 209                 cs->flags |= CLOCK_SOURCE_VALID_FOR_HRES;
 210 }
 211
 212 static inline void clocksource_resume_watchdog(void) { }
 213 #endif
 214
 215 /**
 216  * clocksource_resume - resume the clocksource(s)
 217  */
 218 void clocksource_resume(void)
 219 {
 220         struct clocksource *cs;
 221         unsigned long flags;
 222
 223         spin_lock_irqsave(&clocksource_lock, flags);
 224
 225         list_for_each_entry(cs, &clocksource_list, list) {
 226                 if (cs->resume)
 227                         cs->resume();
 228         }
 229
 230         clocksource_resume_watchdog();
 231
 232         spin_unlock_irqrestore(&clocksource_lock, flags);
 233 }
 234
 235 /**
 236  * clocksource_touch_watchdog - Update watchdog
 237  *
 238  * Update the watchdog after exception contexts such as kgdb so as not
 239  * to incorrectly trip the watchdog.
 240  *
 241  */
 242 void clocksource_touch_watchdog(void)
 243 {
 244         clocksource_resume_watchdog();
 245 }
 246
 247 /**
 248  * clocksource_get_next - Returns the selected clocksource
 249  *
 250  */
 251 struct clocksource *clocksource_get_next(void)
 252 {
 253         unsigned long flags;
 254
 255         spin_lock_irqsave(&clocksource_lock, flags);
 256         if (next_clocksource && finished_booting) {
 257                 curr_clocksource = next_clocksource;
 258                 next_clocksource = NULL;
 259         }
 260         spin_unlock_irqrestore(&clocksource_lock, flags);
 261
 262         return curr_clocksource;
 263 }
 264
 265 /**
 266  * select_clocksource - Selects the best registered clocksource.
 267  *
 268  * Private function. Must hold clocksource_lock when called.
 269  *
 270  * Select the clocksource with the best rating, or the clocksource,
 271  * which is selected by userspace override.
 272  */
 273 static struct clocksource *select_clocksource(void)
 274 {
 275         struct clocksource *next;
 276
 277         if (list_empty(&clocksource_list))
 278                 return NULL;
 279
 280         if (clocksource_override)
 281                 next = clocksource_override;
 282         else
 283                 next = list_entry(clocksource_list.next, struct clocksource,
 284                                   list);
 285
 286         if (next == curr_clocksource)
 287                 return NULL;
 288
 289         return next;
 290 }
 291
 292 /*
 293  * Enqueue the clocksource sorted by rating
 294  */
 295 static int clocksource_enqueue(struct clocksource *c)
 296 {
 297         struct list_head *tmp, *entry = &clocksource_list;
 298
 299         list_for_each(tmp, &clocksource_list) {
 300                 struct clocksource *cs;
 301
 302                 cs = list_entry(tmp, struct clocksource, list);
 303                 if (cs == c)
 304                         return -EBUSY;
 305                 /* Keep track of the place, where to insert */
 306                 if (cs->rating >= c->rating)
 307                         entry = tmp;
 308         }
 309         list_add(&c->list, entry);
 310
 311         if (strlen(c->name) == strlen(override_name) &&
 312             !strcmp(c->name, override_name))
 313                 clocksource_override = c;
 314
 315         return 0;
 316 }
 317
 318 /**
 319  * clocksource_register - Used to install new clocksources
 320  * @t:          clocksource to be registered
 321  *
 322  * Returns -EBUSY if registration fails, zero otherwise.
 323  */
 324 int clocksource_register(struct clocksource *c)
 325 {
 326         unsigned long flags;
 327         int ret;
 328
 329         /* save mult_orig on registration */
 330         c->mult_orig = c->mult;
 331
 332         spin_lock_irqsave(&clocksource_lock, flags);
 333         ret = clocksource_enqueue(c);
 334         if (!ret)
 335                 next_clocksource = select_clocksource();
 336         spin_unlock_irqrestore(&clocksource_lock, flags);
 337         if (!ret)
 338                 clocksource_check_watchdog(c);
 339         return ret;
 340 }
 341 EXPORT_SYMBOL(clocksource_register);
 342
 343 /**
 344  * clocksource_change_rating - Change the rating of a registered clocksource
 345  *
 346  */
 347 void clocksource_change_rating(struct clocksource *cs, int rating)
 348 {
 349         unsigned long flags;
 350
 351         spin_lock_irqsave(&clocksource_lock, flags);
 352         list_del(&cs->list);
 353         cs->rating = rating;
 354         clocksource_enqueue(cs);
 355         next_clocksource = select_clocksource();
 356         spin_unlock_irqrestore(&clocksource_lock, flags);
 357 }
 358
 359 /**
 360  * clocksource_unregister - remove a registered clocksource
 361  */
 362 void clocksource_unregister(struct clocksource *cs)
 363 {
 364         unsigned long flags;
 365
 366         spin_lock_irqsave(&clocksource_lock, flags);
 367         list_del(&cs->list);
 368         if (clocksource_override == cs)
 369                 clocksource_override = NULL;
 370         next_clocksource = select_clocksource();
 371         spin_unlock_irqrestore(&clocksource_lock, flags);
 372 }
 373
 374 #ifdef CONFIG_SYSFS
 375 /**
 376  * sysfs_show_current_clocksources - sysfs interface for current clocksource
 377  * @dev:        unused
 378  * @buf:        char buffer to be filled with clocksource list
 379  *
 380  * Provides sysfs interface for listing current clocksource.
 381  */
 382 static ssize_t
 383 sysfs_show_current_clocksources(struct sys_device *dev,
 384                                 struct sysdev_attribute *attr, char *buf)
 385 {
 386         ssize_t count = 0;
 387
 388         spin_lock_irq(&clocksource_lock);
 389         count = snprintf(buf, PAGE_SIZE, "%s\n", curr_clocksource->name);
 390         spin_unlock_irq(&clocksource_lock);
 391
 392         return count;
 393 }
 394
 395 /**
 396  * sysfs_override_clocksource - interface for manually overriding clocksource
 397  * @dev:        unused
 398  * @buf:        name of override clocksource
 399  * @count:      length of buffer
 400  *
 401  * Takes input from sysfs interface for manually overriding the default
 402  * clocksource selction.
 403  */
 404 static ssize_t sysfs_override_clocksource(struct sys_device *dev,
 405                                           struct sysdev_attribute *attr,
 406                                           const char *buf, size_t count)
 407 {
 408         struct clocksource *ovr = NULL;
 409         size_t ret = count;
 410         int len;
 411
 412         /* strings from sysfs write are not 0 terminated! */
 413         if (count >= sizeof(override_name))
 414                 return -EINVAL;
 415
 416         /* strip of \n: */
 417         if (buf[count-1] == '\n')
 418                 count--;
 419
 420         spin_lock_irq(&clocksource_lock);
 421
 422         if (count > 0)
 423                 memcpy(override_name, buf, count);
 424         override_name[count] = 0;
 425
 426         len = strlen(override_name);
 427         if (len) {
 428                 struct clocksource *cs;
 429
 430                 ovr = clocksource_override;
 431                 /* try to select it: */
 432                 list_for_each_entry(cs, &clocksource_list, list) {
 433                         if (strlen(cs->name) == len &&
 434                             !strcmp(cs->name, override_name))
 435                                 ovr = cs;
 436                 }
 437         }
 438
 439         /* Reselect, when the override name has changed */
 440         if (ovr != clocksource_override) {
 441                 clocksource_override = ovr;
 442                 next_clocksource = select_clocksource();
 443         }
 444
 445         spin_unlock_irq(&clocksource_lock);
 446
 447         return ret;
 448 }
 449
 450 /**
 451  * sysfs_show_available_clocksources - sysfs interface for listing clocksource
 452  * @dev:        unused
 453  * @buf:        char buffer to be filled with clocksource list
 454  *
 455  * Provides sysfs interface for listing registered clocksources
 456  */
 457 static ssize_t
 458 sysfs_show_available_clocksources(struct sys_device *dev,
 459                                   struct sysdev_attribute *attr,
 460                                   char *buf)
 461 {
 462         struct clocksource *src;
 463         ssize_t count = 0;
 464
 465         spin_lock_irq(&clocksource_lock);
 466         list_for_each_entry(src, &clocksource_list, list) {
 467                 count += snprintf(buf + count,
 468                                   max((ssize_t)PAGE_SIZE - count, (ssize_t)0),
 469                                   "%s ", src->name);
 470         }
 471         spin_unlock_irq(&clocksource_lock);
 472
 473         count += snprintf(buf + count,
 474                           max((ssize_t)PAGE_SIZE - count, (ssize_t)0), "\n");
 475
 476         return count;
 477 }
 478
 479 /*
 480  * Sysfs setup bits:
 481  */
 482 static SYSDEV_ATTR(current_clocksource, 0644, sysfs_show_current_clocksources,
 483                    sysfs_override_clocksource);
 484
 485 static SYSDEV_ATTR(available_clocksource, 0444,
 486                    sysfs_show_available_clocksources, NULL);
 487
 488 static struct sysdev_class clocksource_sysclass = {
 489         .name = "clocksource",
 490 };
 491
 492 static struct sys_device device_clocksource = {
 493         .id     = 0,
 494         .cls    = &clocksource_sysclass,
 495 };
 496
 497 static int __init init_clocksource_sysfs(void)
 498 {
 499         int error = sysdev_class_register(&clocksource_sysclass);
 500
 501         if (!error)
 502                 error = sysdev_register(&device_clocksource);
 503         if (!error)
 504                 error = sysdev_create_file(
 505                                 &device_clocksource,
 506                                 &attr_current_clocksource);
 507         if (!error)
 508                 error = sysdev_create_file(
 509                                 &device_clocksource,
 510                                 &attr_available_clocksource);
 511         return error;
 512 }
 513
 514 device_initcall(init_clocksource_sysfs);
 515 #endif /* CONFIG_SYSFS */
 516
 517 /**
 518  * boot_override_clocksource - boot clock override
 519  * @str:        override name
 520  *
 521  * Takes a clocksource= boot argument and uses it
 522  * as the clocksource override name.
 523  */
 524 static int __init boot_override_clocksource(char* str)
 525 {
 526         unsigned long flags;
 527         spin_lock_irqsave(&clocksource_lock, flags);
 528         if (str)
 529                 strlcpy(override_name, str, sizeof(override_name));
 530         spin_unlock_irqrestore(&clocksource_lock, flags);
 531         return 1;
 532 }
 533
 534 __setup("clocksource=", boot_override_clocksource);
 535
 536 /**
 537  * boot_override_clock - Compatibility layer for deprecated boot option
 538  * @str:        override name
 539  *
 540  * DEPRECATED! Takes a clock= boot argument and uses it
 541  * as the clocksource override name
 542  */
 543 static int __init boot_override_clock(char* str)
 544 {
 545         if (!strcmp(str, "pmtmr")) {
 546                 printk("Warning: clock=pmtmr is deprecated. "
 547                         "Use clocksource=acpi_pm.\n");
 548                 return boot_override_clocksource("acpi_pm");
 549         }
 550         printk("Warning! clock= boot option is deprecated. "
 551                 "Use clocksource=xyz\n");
 552         return boot_override_clocksource(str);
 553 }
 554
 555 __setup("clock=", boot_override_clock);