drivers/cpufreq/cpufreq.c

   1 /*
   2  *  linux/drivers/cpufreq/cpufreq.c
   3  *
   4  *  Copyright (C) 2001 Russell King
   5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  *
  11  */
  12
  13 #include <linux/config.h>
  14 #include <linux/kernel.h>
  15 #include <linux/module.h>
  16 #include <linux/init.h>
  17 #include <linux/notifier.h>
  18 #include <linux/cpufreq.h>
  19 #include <linux/delay.h>
  20 #include <linux/interrupt.h>
  21 #include <linux/spinlock.h>
  22 #include <linux/device.h>
  23 #include <linux/slab.h>
  24 #include <linux/cpu.h>
  25 #include <linux/completion.h>
  26
  27 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
  28
  29 /**
  30  * The "cpufreq driver" - the arch- or hardware-dependend low
  31  * level driver of CPUFreq support, and its spinlock. This lock
  32  * also protects the cpufreq_cpu_data array.
  33  */
  34 static struct cpufreq_driver    *cpufreq_driver;
  35 static struct cpufreq_policy    *cpufreq_cpu_data[NR_CPUS];
  36 static DEFINE_SPINLOCK(cpufreq_driver_lock);
  37
  38
  39 /* we keep a copy of all ->add'ed CPU's struct sys_device here;
  40  * as it is only accessed in ->add and ->remove, no lock or reference
  41  * count is necessary.
  42  */
  43 static struct sys_device        *cpu_sys_devices[NR_CPUS];
  44
  45
  46 /* internal prototypes */
  47 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
  48 static void handle_update(void *data);
  49 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci);
  50
  51 /**
  52  * Two notifier lists: the "policy" list is involved in the
  53  * validation process for a new CPU frequency policy; the
  54  * "transition" list for kernel code that needs to handle
  55  * changes to devices when the CPU clock speed changes.
  56  * The mutex locks both lists.
  57  */
  58 static struct notifier_block    *cpufreq_policy_notifier_list;
  59 static struct notifier_block    *cpufreq_transition_notifier_list;
  60 static DECLARE_RWSEM            (cpufreq_notifier_rwsem);
  61
  62
  63 static LIST_HEAD(cpufreq_governor_list);
  64 static DECLARE_MUTEX            (cpufreq_governor_sem);
  65
  66 struct cpufreq_policy * cpufreq_cpu_get(unsigned int cpu)
  67 {
  68         struct cpufreq_policy *data;
  69         unsigned long flags;
  70
  71         if (cpu >= NR_CPUS)
  72                 goto err_out;
  73
  74         /* get the cpufreq driver */
  75         spin_lock_irqsave(&cpufreq_driver_lock, flags);
  76
  77         if (!cpufreq_driver)
  78                 goto err_out_unlock;
  79
  80         if (!try_module_get(cpufreq_driver->owner))
  81                 goto err_out_unlock;
  82
  83
  84         /* get the CPU */
  85         data = cpufreq_cpu_data[cpu];
  86
  87         if (!data)
  88                 goto err_out_put_module;
  89
  90         if (!kobject_get(&data->kobj))
  91                 goto err_out_put_module;
  92
  93
  94         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  95
  96         return data;
  97
  98  err_out_put_module:
  99         module_put(cpufreq_driver->owner);
 100  err_out_unlock:
 101         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 102  err_out:
 103         return NULL;
 104 }
 105 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 106
 107 void cpufreq_cpu_put(struct cpufreq_policy *data)
 108 {
 109         kobject_put(&data->kobj);
 110         module_put(cpufreq_driver->owner);
 111 }
 112 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 113
 114
 115 /*********************************************************************
 116  *                     UNIFIED DEBUG HELPERS                         *
 117  *********************************************************************/
 118 #ifdef CONFIG_CPU_FREQ_DEBUG
 119
 120 /* what part(s) of the CPUfreq subsystem are debugged? */
 121 static unsigned int debug;
 122
 123 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 124  * set or modify this value.
 125  */
 126 static unsigned int debug_ratelimit = 1;
 127
 128 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 129  * loading of a cpufreq driver, temporarily disabled when a new policy
 130  * is set, and disabled upon cpufreq driver removal
 131  */
 132 static unsigned int disable_ratelimit = 1;
 133 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 134
 135 static inline void cpufreq_debug_enable_ratelimit(void)
 136 {
 137         unsigned long flags;
 138
 139         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 140         if (disable_ratelimit)
 141                 disable_ratelimit--;
 142         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 143 }
 144
 145 static inline void cpufreq_debug_disable_ratelimit(void)
 146 {
 147         unsigned long flags;
 148
 149         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 150         disable_ratelimit++;
 151         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 152 }
 153
 154 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
 155 {
 156         char s[256];
 157         va_list args;
 158         unsigned int len;
 159         unsigned long flags;
 160
 161         WARN_ON(!prefix);
 162         if (type & debug) {
 163                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 164                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
 165                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 166                         return;
 167                 }
 168                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 169
 170                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 171
 172                 va_start(args, fmt);
 173                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 174                 va_end(args);
 175
 176                 printk(s);
 177
 178                 WARN_ON(len < 5);
 179         }
 180 }
 181 EXPORT_SYMBOL(cpufreq_debug_printk);
 182
 183
 184 module_param(debug, uint, 0644);
 185 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
 186
 187 module_param(debug_ratelimit, uint, 0644);
 188 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
 189
 190 #else /* !CONFIG_CPU_FREQ_DEBUG */
 191
 192 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 193 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 194
 195 #endif /* CONFIG_CPU_FREQ_DEBUG */
 196
 197
 198 /*********************************************************************
 199  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 200  *********************************************************************/
 201
 202 /**
 203  * adjust_jiffies - adjust the system "loops_per_jiffy"
 204  *
 205  * This function alters the system "loops_per_jiffy" for the clock
 206  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 207  * systems as each CPU might be scaled differently. So, use the arch
 208  * per-CPU loops_per_jiffy value wherever possible.
 209  */
 210 #ifndef CONFIG_SMP
 211 static unsigned long l_p_j_ref;
 212 static unsigned int  l_p_j_ref_freq;
 213
 214 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 215 {
 216         if (ci->flags & CPUFREQ_CONST_LOOPS)
 217                 return;
 218
 219         if (!l_p_j_ref_freq) {
 220                 l_p_j_ref = loops_per_jiffy;
 221                 l_p_j_ref_freq = ci->old;
 222                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 223         }
 224         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 225             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 226             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 227                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
 228                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
 229         }
 230 }
 231 #else
 232 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
 233 #endif
 234
 235
 236 /**
 237  * cpufreq_notify_transition - call notifier chain and adjust_jiffies on frequency transition
 238  *
 239  * This function calls the transition notifiers and the "adjust_jiffies" function. It is called
 240  * twice on all CPU frequency changes that have external effects.
 241  */
 242 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 243 {
 244         BUG_ON(irqs_disabled());
 245
 246         freqs->flags = cpufreq_driver->flags;
 247         dprintk("notification %u of frequency transition to %u kHz\n", state, freqs->new);
 248
 249         down_read(&cpufreq_notifier_rwsem);
 250         switch (state) {
 251         case CPUFREQ_PRECHANGE:
 252                 /* detect if the driver reported a value as "old frequency" which
 253                  * is not equal to what the cpufreq core thinks is "old frequency".
 254                  */
 255                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 256                         if ((likely(cpufreq_cpu_data[freqs->cpu])) &&
 257                             (likely(cpufreq_cpu_data[freqs->cpu]->cpu == freqs->cpu)) &&
 258                             (likely(cpufreq_cpu_data[freqs->cpu]->cur)) &&
 259                             (unlikely(freqs->old != cpufreq_cpu_data[freqs->cpu]->cur)))
 260                         {
 261                                 dprintk(KERN_WARNING "Warning: CPU frequency is %u, "
 262                                        "cpufreq assumed %u kHz.\n", freqs->old, cpufreq_cpu_data[freqs->cpu]->cur);
 263                                 freqs->old = cpufreq_cpu_data[freqs->cpu]->cur;
 264                         }
 265                 }
 266                 notifier_call_chain(&cpufreq_transition_notifier_list, CPUFREQ_PRECHANGE, freqs);
 267                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 268                 break;
 269         case CPUFREQ_POSTCHANGE:
 270                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 271                 notifier_call_chain(&cpufreq_transition_notifier_list, CPUFREQ_POSTCHANGE, freqs);
 272                 if ((likely(cpufreq_cpu_data[freqs->cpu])) &&
 273                     (likely(cpufreq_cpu_data[freqs->cpu]->cpu == freqs->cpu)))
 274                         cpufreq_cpu_data[freqs->cpu]->cur = freqs->new;
 275                 break;
 276         }
 277         up_read(&cpufreq_notifier_rwsem);
 278 }
 279 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 280
 281
 282
 283 /*********************************************************************
 284  *                          SYSFS INTERFACE                          *
 285  *********************************************************************/
 286
 287 /**
 288  * cpufreq_parse_governor - parse a governor string
 289  */
 290 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 291                                 struct cpufreq_governor **governor)
 292 {
 293         if (!cpufreq_driver)
 294                 return -EINVAL;
 295         if (cpufreq_driver->setpolicy) {
 296                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 297                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 298                         return 0;
 299                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 300                         *policy = CPUFREQ_POLICY_POWERSAVE;
 301                         return 0;
 302                 }
 303                 return -EINVAL;
 304         } else {
 305                 struct cpufreq_governor *t;
 306                 down(&cpufreq_governor_sem);
 307                 if (!cpufreq_driver || !cpufreq_driver->target)
 308                         goto out;
 309                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 310                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
 311                                 *governor = t;
 312                                 up(&cpufreq_governor_sem);
 313                                 return 0;
 314                         }
 315                 }
 316         out:
 317                 up(&cpufreq_governor_sem);
 318         }
 319         return -EINVAL;
 320 }
 321 EXPORT_SYMBOL_GPL(cpufreq_parse_governor);
 322
 323
 324 /* drivers/base/cpu.c */
 325 extern struct sysdev_class cpu_sysdev_class;
 326
 327
 328 /**
 329  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 330  *
 331  * Write out information from cpufreq_driver->policy[cpu]; object must be
 332  * "unsigned int".
 333  */
 334
 335 #define show_one(file_name, object)                                     \
 336 static ssize_t show_##file_name                                         \
 337 (struct cpufreq_policy * policy, char *buf)                             \
 338 {                                                                       \
 339         return sprintf (buf, "%u\n", policy->object);                   \
 340 }
 341
 342 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 343 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 344 show_one(scaling_min_freq, min);
 345 show_one(scaling_max_freq, max);
 346 show_one(scaling_cur_freq, cur);
 347
 348 /**
 349  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 350  */
 351 #define store_one(file_name, object)                    \
 352 static ssize_t store_##file_name                                        \
 353 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 354 {                                                                       \
 355         unsigned int ret = -EINVAL;                                     \
 356         struct cpufreq_policy new_policy;                               \
 357                                                                         \
 358         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 359         if (ret)                                                        \
 360                 return -EINVAL;                                         \
 361                                                                         \
 362         ret = sscanf (buf, "%u", &new_policy.object);                   \
 363         if (ret != 1)                                                   \
 364                 return -EINVAL;                                         \
 365                                                                         \
 366         ret = cpufreq_set_policy(&new_policy);                          \
 367                                                                         \
 368         return ret ? ret : count;                                       \
 369 }
 370
 371 store_one(scaling_min_freq,min);
 372 store_one(scaling_max_freq,max);
 373
 374 /**
 375  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 376  */
 377 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 378 {
 379         unsigned int cur_freq = cpufreq_get(policy->cpu);
 380         if (!cur_freq)
 381                 return sprintf(buf, "<unknown>");
 382         return sprintf(buf, "%u\n", cur_freq);
 383 }
 384
 385
 386 /**
 387  * show_scaling_governor - show the current policy for the specified CPU
 388  */
 389 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 390 {
 391         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 392                 return sprintf(buf, "powersave\n");
 393         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 394                 return sprintf(buf, "performance\n");
 395         else if (policy->governor)
 396                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 397         return -EINVAL;
 398 }
 399
 400
 401 /**
 402  * store_scaling_governor - store policy for the specified CPU
 403  */
 404 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 405                                        const char *buf, size_t count)
 406 {
 407         unsigned int ret = -EINVAL;
 408         char    str_governor[16];
 409         struct cpufreq_policy new_policy;
 410
 411         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 412         if (ret)
 413                 return ret;
 414
 415         ret = sscanf (buf, "%15s", str_governor);
 416         if (ret != 1)
 417                 return -EINVAL;
 418
 419         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 420                 return -EINVAL;
 421
 422         ret = cpufreq_set_policy(&new_policy);
 423
 424         return ret ? ret : count;
 425 }
 426
 427 /**
 428  * show_scaling_driver - show the cpufreq driver currently loaded
 429  */
 430 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 431 {
 432         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 433 }
 434
 435 /**
 436  * show_scaling_available_governors - show the available CPUfreq governors
 437  */
 438 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 439                                 char *buf)
 440 {
 441         ssize_t i = 0;
 442         struct cpufreq_governor *t;
 443
 444         if (!cpufreq_driver->target) {
 445                 i += sprintf(buf, "performance powersave");
 446                 goto out;
 447         }
 448
 449         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 450                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 451                         goto out;
 452                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 453         }
 454  out:
 455         i += sprintf(&buf[i], "\n");
 456         return i;
 457 }
 458 /**
 459  * show_affected_cpus - show the CPUs affected by each transition
 460  */
 461 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 462 {
 463         ssize_t i = 0;
 464         unsigned int cpu;
 465
 466         for_each_cpu_mask(cpu, policy->cpus) {
 467                 if (i)
 468                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 469                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 470                 if (i >= (PAGE_SIZE - 5))
 471                     break;
 472         }
 473         i += sprintf(&buf[i], "\n");
 474         return i;
 475 }
 476
 477
 478 #define define_one_ro(_name) \
 479 static struct freq_attr _name = \
 480 __ATTR(_name, 0444, show_##_name, NULL)
 481
 482 #define define_one_ro0400(_name) \
 483 static struct freq_attr _name = \
 484 __ATTR(_name, 0400, show_##_name, NULL)
 485
 486 #define define_one_rw(_name) \
 487 static struct freq_attr _name = \
 488 __ATTR(_name, 0644, show_##_name, store_##_name)
 489
 490 define_one_ro0400(cpuinfo_cur_freq);
 491 define_one_ro(cpuinfo_min_freq);
 492 define_one_ro(cpuinfo_max_freq);
 493 define_one_ro(scaling_available_governors);
 494 define_one_ro(scaling_driver);
 495 define_one_ro(scaling_cur_freq);
 496 define_one_ro(affected_cpus);
 497 define_one_rw(scaling_min_freq);
 498 define_one_rw(scaling_max_freq);
 499 define_one_rw(scaling_governor);
 500
 501 static struct attribute * default_attrs[] = {
 502         &cpuinfo_min_freq.attr,
 503         &cpuinfo_max_freq.attr,
 504         &scaling_min_freq.attr,
 505         &scaling_max_freq.attr,
 506         &affected_cpus.attr,
 507         &scaling_governor.attr,
 508         &scaling_driver.attr,
 509         &scaling_available_governors.attr,
 510         NULL
 511 };
 512
 513 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 514 #define to_attr(a) container_of(a,struct freq_attr,attr)
 515
 516 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 517 {
 518         struct cpufreq_policy * policy = to_policy(kobj);
 519         struct freq_attr * fattr = to_attr(attr);
 520         ssize_t ret;
 521         policy = cpufreq_cpu_get(policy->cpu);
 522         if (!policy)
 523                 return -EINVAL;
 524         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 525         cpufreq_cpu_put(policy);
 526         return ret;
 527 }
 528
 529 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 530                      const char * buf, size_t count)
 531 {
 532         struct cpufreq_policy * policy = to_policy(kobj);
 533         struct freq_attr * fattr = to_attr(attr);
 534         ssize_t ret;
 535         policy = cpufreq_cpu_get(policy->cpu);
 536         if (!policy)
 537                 return -EINVAL;
 538         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 539         cpufreq_cpu_put(policy);
 540         return ret;
 541 }
 542
 543 static void cpufreq_sysfs_release(struct kobject * kobj)
 544 {
 545         struct cpufreq_policy * policy = to_policy(kobj);
 546         dprintk("last reference is dropped\n");
 547         complete(&policy->kobj_unregister);
 548 }
 549
 550 static struct sysfs_ops sysfs_ops = {
 551         .show   = show,
 552         .store  = store,
 553 };
 554
 555 static struct kobj_type ktype_cpufreq = {
 556         .sysfs_ops      = &sysfs_ops,
 557         .default_attrs  = default_attrs,
 558         .release        = cpufreq_sysfs_release,
 559 };
 560
 561
 562 /**
 563  * cpufreq_add_dev - add a CPU device
 564  *
 565  * Adds the cpufreq interface for a CPU device.
 566  */
 567 static int cpufreq_add_dev (struct sys_device * sys_dev)
 568 {
 569         unsigned int cpu = sys_dev->id;
 570         int ret = 0;
 571         struct cpufreq_policy new_policy;
 572         struct cpufreq_policy *policy;
 573         struct freq_attr **drv_attr;
 574         unsigned long flags;
 575         unsigned int j;
 576
 577         cpufreq_debug_disable_ratelimit();
 578         dprintk("adding CPU %u\n", cpu);
 579
 580 #ifdef CONFIG_SMP
 581         /* check whether a different CPU already registered this
 582          * CPU because it is in the same boat. */
 583         policy = cpufreq_cpu_get(cpu);
 584         if (unlikely(policy)) {
 585                 cpu_sys_devices[cpu] = sys_dev;
 586                 dprintk("CPU already managed, adding link\n");
 587                 sysfs_create_link(&sys_dev->kobj, &policy->kobj, "cpufreq");
 588                 cpufreq_debug_enable_ratelimit();
 589                 return 0;
 590         }
 591 #endif
 592
 593         if (!try_module_get(cpufreq_driver->owner)) {
 594                 ret = -EINVAL;
 595                 goto module_out;
 596         }
 597
 598         policy = kmalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 599         if (!policy) {
 600                 ret = -ENOMEM;
 601                 goto nomem_out;
 602         }
 603         memset(policy, 0, sizeof(struct cpufreq_policy));
 604
 605         policy->cpu = cpu;
 606         policy->cpus = cpumask_of_cpu(cpu);
 607
 608         init_MUTEX_LOCKED(&policy->lock);
 609         init_completion(&policy->kobj_unregister);
 610         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 611
 612         /* call driver. From then on the cpufreq must be able
 613          * to accept all calls to ->verify and ->setpolicy for this CPU
 614          */
 615         ret = cpufreq_driver->init(policy);
 616         if (ret) {
 617                 dprintk("initialization failed\n");
 618                 goto err_out;
 619         }
 620
 621         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 622
 623         /* prepare interface data */
 624         policy->kobj.parent = &sys_dev->kobj;
 625         policy->kobj.ktype = &ktype_cpufreq;
 626         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 627
 628         ret = kobject_register(&policy->kobj);
 629         if (ret)
 630                 goto err_out_driver_exit;
 631
 632         /* set up files for this cpu device */
 633         drv_attr = cpufreq_driver->attr;
 634         while ((drv_attr) && (*drv_attr)) {
 635                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 636                 drv_attr++;
 637         }
 638         if (cpufreq_driver->get)
 639                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 640         if (cpufreq_driver->target)
 641                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 642
 643         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 644         for_each_cpu_mask(j, policy->cpus)
 645                 cpufreq_cpu_data[j] = policy;
 646         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 647         policy->governor = NULL; /* to assure that the starting sequence is
 648                                   * run in cpufreq_set_policy */
 649         up(&policy->lock);
 650
 651         /* set default policy */
 652
 653         ret = cpufreq_set_policy(&new_policy);
 654         if (ret) {
 655                 dprintk("setting policy failed\n");
 656                 goto err_out_unregister;
 657         }
 658
 659         module_put(cpufreq_driver->owner);
 660         cpu_sys_devices[cpu] = sys_dev;
 661         dprintk("initialization complete\n");
 662         cpufreq_debug_enable_ratelimit();
 663
 664         return 0;
 665
 666
 667 err_out_unregister:
 668         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 669         for_each_cpu_mask(j, policy->cpus)
 670                 cpufreq_cpu_data[j] = NULL;
 671         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 672
 673         kobject_unregister(&policy->kobj);
 674         wait_for_completion(&policy->kobj_unregister);
 675
 676 err_out_driver_exit:
 677         if (cpufreq_driver->exit)
 678                 cpufreq_driver->exit(policy);
 679
 680 err_out:
 681         kfree(policy);
 682
 683 nomem_out:
 684         module_put(cpufreq_driver->owner);
 685  module_out:
 686         cpufreq_debug_enable_ratelimit();
 687         return ret;
 688 }
 689
 690
 691 /**
 692  * cpufreq_remove_dev - remove a CPU device
 693  *
 694  * Removes the cpufreq interface for a CPU device.
 695  */
 696 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 697 {
 698         unsigned int cpu = sys_dev->id;
 699         unsigned long flags;
 700         struct cpufreq_policy *data;
 701 #ifdef CONFIG_SMP
 702         unsigned int j;
 703 #endif
 704
 705         cpufreq_debug_disable_ratelimit();
 706         dprintk("unregistering CPU %u\n", cpu);
 707
 708         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 709         data = cpufreq_cpu_data[cpu];
 710
 711         if (!data) {
 712                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 713                 cpu_sys_devices[cpu] = NULL;
 714                 cpufreq_debug_enable_ratelimit();
 715                 return -EINVAL;
 716         }
 717         cpufreq_cpu_data[cpu] = NULL;
 718
 719
 720 #ifdef CONFIG_SMP
 721         /* if this isn't the CPU which is the parent of the kobj, we
 722          * only need to unlink, put and exit
 723          */
 724         if (unlikely(cpu != data->cpu)) {
 725                 dprintk("removing link\n");
 726                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 727                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 728                 cpu_sys_devices[cpu] = NULL;
 729                 cpufreq_cpu_put(data);
 730                 cpufreq_debug_enable_ratelimit();
 731                 return 0;
 732         }
 733 #endif
 734
 735         cpu_sys_devices[cpu] = NULL;
 736
 737         if (!kobject_get(&data->kobj)) {
 738                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 739                 cpufreq_debug_enable_ratelimit();
 740                 return -EFAULT;
 741         }
 742
 743 #ifdef CONFIG_SMP
 744         /* if we have other CPUs still registered, we need to unlink them,
 745          * or else wait_for_completion below will lock up. Clean the
 746          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 747          * links afterwards.
 748          */
 749         if (unlikely(cpus_weight(data->cpus) > 1)) {
 750                 for_each_cpu_mask(j, data->cpus) {
 751                         if (j == cpu)
 752                                 continue;
 753                         cpufreq_cpu_data[j] = NULL;
 754                 }
 755         }
 756
 757         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 758
 759         if (unlikely(cpus_weight(data->cpus) > 1)) {
 760                 for_each_cpu_mask(j, data->cpus) {
 761                         if (j == cpu)
 762                                 continue;
 763                         dprintk("removing link for cpu %u\n", j);
 764                         sysfs_remove_link(&cpu_sys_devices[j]->kobj, "cpufreq");
 765                         cpufreq_cpu_put(data);
 766                 }
 767         }
 768 #else
 769         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 770 #endif
 771
 772         down(&data->lock);
 773         if (cpufreq_driver->target)
 774                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 775         cpufreq_driver->target = NULL;
 776         up(&data->lock);
 777
 778         kobject_unregister(&data->kobj);
 779
 780         kobject_put(&data->kobj);
 781
 782         /* we need to make sure that the underlying kobj is actually
 783          * not referenced anymore by anybody before we proceed with
 784          * unloading.
 785          */
 786         dprintk("waiting for dropping of refcount\n");
 787         wait_for_completion(&data->kobj_unregister);
 788         dprintk("wait complete\n");
 789
 790         if (cpufreq_driver->exit)
 791                 cpufreq_driver->exit(data);
 792
 793         kfree(data);
 794
 795         cpufreq_debug_enable_ratelimit();
 796
 797         return 0;
 798 }
 799
 800
 801 static void handle_update(void *data)
 802 {
 803         unsigned int cpu = (unsigned int)(long)data;
 804         dprintk("handle_update for cpu %u called\n", cpu);
 805         cpufreq_update_policy(cpu);
 806 }
 807
 808 /**
 809  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 810  *      @cpu: cpu number
 811  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 812  *      @new_freq: CPU frequency the CPU actually runs at
 813  *
 814  *      We adjust to current frequency first, and need to clean up later. So either call
 815  *      to cpufreq_update_policy() or schedule handle_update()).
 816  */
 817 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 818 {
 819         struct cpufreq_freqs freqs;
 820
 821         dprintk(KERN_WARNING "Warning: CPU frequency out of sync: cpufreq and timing "
 822                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 823
 824         freqs.cpu = cpu;
 825         freqs.old = old_freq;
 826         freqs.new = new_freq;
 827         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 828         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 829 }
 830
 831
 832 /**
 833  * cpufreq_get - get the current CPU frequency (in kHz)
 834  * @cpu: CPU number
 835  *
 836  * Get the CPU current (static) CPU frequency
 837  */
 838 unsigned int cpufreq_get(unsigned int cpu)
 839 {
 840         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 841         unsigned int ret = 0;
 842
 843         if (!policy)
 844                 return 0;
 845
 846         if (!cpufreq_driver->get)
 847                 goto out;
 848
 849         down(&policy->lock);
 850
 851         ret = cpufreq_driver->get(cpu);
 852
 853         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS))
 854         {
 855                 /* verify no discrepancy between actual and saved value exists */
 856                 if (unlikely(ret != policy->cur)) {
 857                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 858                         schedule_work(&policy->update);
 859                 }
 860         }
 861
 862         up(&policy->lock);
 863
 864  out:
 865         cpufreq_cpu_put(policy);
 866
 867         return (ret);
 868 }
 869 EXPORT_SYMBOL(cpufreq_get);
 870
 871
 872 /**
 873  *      cpufreq_suspend - let the low level driver prepare for suspend
 874  */
 875
 876 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 877 {
 878         int cpu = sysdev->id;
 879         unsigned int ret = 0;
 880         unsigned int cur_freq = 0;
 881         struct cpufreq_policy *cpu_policy;
 882
 883         dprintk("resuming cpu %u\n", cpu);
 884
 885         if (!cpu_online(cpu))
 886                 return 0;
 887
 888         /* we may be lax here as interrupts are off. Nonetheless
 889          * we need to grab the correct cpu policy, as to check
 890          * whether we really run on this CPU.
 891          */
 892
 893         cpu_policy = cpufreq_cpu_get(cpu);
 894         if (!cpu_policy)
 895                 return -EINVAL;
 896
 897         /* only handle each CPU group once */
 898         if (unlikely(cpu_policy->cpu != cpu)) {
 899                 cpufreq_cpu_put(cpu_policy);
 900                 return 0;
 901         }
 902
 903         if (cpufreq_driver->suspend) {
 904                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
 905                 if (ret) {
 906                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
 907                                         "step on CPU %u\n", cpu_policy->cpu);
 908                         cpufreq_cpu_put(cpu_policy);
 909                         return ret;
 910                 }
 911         }
 912
 913
 914         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
 915                 goto out;
 916
 917         if (cpufreq_driver->get)
 918                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
 919
 920         if (!cur_freq || !cpu_policy->cur) {
 921                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
 922                        "frequency is what timing core thinks it is.\n");
 923                 goto out;
 924         }
 925
 926         if (unlikely(cur_freq != cpu_policy->cur)) {
 927                 struct cpufreq_freqs freqs;
 928
 929                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
 930                         dprintk(KERN_DEBUG "Warning: CPU frequency is %u, "
 931                                "cpufreq assumed %u kHz.\n",
 932                                cur_freq, cpu_policy->cur);
 933
 934                 freqs.cpu = cpu;
 935                 freqs.old = cpu_policy->cur;
 936                 freqs.new = cur_freq;
 937
 938                 notifier_call_chain(&cpufreq_transition_notifier_list,
 939                                     CPUFREQ_SUSPENDCHANGE, &freqs);
 940                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
 941
 942                 cpu_policy->cur = cur_freq;
 943         }
 944
 945  out:
 946         cpufreq_cpu_put(cpu_policy);
 947         return 0;
 948 }
 949
 950 /**
 951  *      cpufreq_resume -  restore proper CPU frequency handling after resume
 952  *
 953  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
 954  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
 955  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
 956  *          restored.
 957  */
 958 static int cpufreq_resume(struct sys_device * sysdev)
 959 {
 960         int cpu = sysdev->id;
 961         unsigned int ret = 0;
 962         struct cpufreq_policy *cpu_policy;
 963
 964         dprintk("resuming cpu %u\n", cpu);
 965
 966         if (!cpu_online(cpu))
 967                 return 0;
 968
 969         /* we may be lax here as interrupts are off. Nonetheless
 970          * we need to grab the correct cpu policy, as to check
 971          * whether we really run on this CPU.
 972          */
 973
 974         cpu_policy = cpufreq_cpu_get(cpu);
 975         if (!cpu_policy)
 976                 return -EINVAL;
 977
 978         /* only handle each CPU group once */
 979         if (unlikely(cpu_policy->cpu != cpu)) {
 980                 cpufreq_cpu_put(cpu_policy);
 981                 return 0;
 982         }
 983
 984         if (cpufreq_driver->resume) {
 985                 ret = cpufreq_driver->resume(cpu_policy);
 986                 if (ret) {
 987                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
 988                                         "step on CPU %u\n", cpu_policy->cpu);
 989                         cpufreq_cpu_put(cpu_policy);
 990                         return ret;
 991                 }
 992         }
 993
 994         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 995                 unsigned int cur_freq = 0;
 996
 997                 if (cpufreq_driver->get)
 998                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
 999
1000                 if (!cur_freq || !cpu_policy->cur) {
1001                         printk(KERN_ERR "cpufreq: resume failed to assert "
1002                                         "current frequency is what timing core "
1003                                         "thinks it is.\n");
1004                         goto out;
1005                 }
1006
1007                 if (unlikely(cur_freq != cpu_policy->cur)) {
1008                         struct cpufreq_freqs freqs;
1009
1010                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1011                                 dprintk(KERN_WARNING "Warning: CPU frequency"
1012                                        "is %u, cpufreq assumed %u kHz.\n",
1013                                        cur_freq, cpu_policy->cur);
1014
1015                         freqs.cpu = cpu;
1016                         freqs.old = cpu_policy->cur;
1017                         freqs.new = cur_freq;
1018
1019                         notifier_call_chain(&cpufreq_transition_notifier_list,
1020                                         CPUFREQ_RESUMECHANGE, &freqs);
1021                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1022
1023                         cpu_policy->cur = cur_freq;
1024                 }
1025         }
1026
1027 out:
1028         schedule_work(&cpu_policy->update);
1029         cpufreq_cpu_put(cpu_policy);
1030         return ret;
1031 }
1032
1033 static struct sysdev_driver cpufreq_sysdev_driver = {
1034         .add            = cpufreq_add_dev,
1035         .remove         = cpufreq_remove_dev,
1036         .suspend        = cpufreq_suspend,
1037         .resume         = cpufreq_resume,
1038 };
1039
1040
1041 /*********************************************************************
1042  *                     NOTIFIER LISTS INTERFACE                      *
1043  *********************************************************************/
1044
1045 /**
1046  *      cpufreq_register_notifier - register a driver with cpufreq
1047  *      @nb: notifier function to register
1048  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1049  *
1050  *      Add a driver to one of two lists: either a list of drivers that
1051  *      are notified about clock rate changes (once before and once after
1052  *      the transition), or a list of drivers that are notified about
1053  *      changes in cpufreq policy.
1054  *
1055  *      This function may sleep, and has the same return conditions as
1056  *      notifier_chain_register.
1057  */
1058 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1059 {
1060         int ret;
1061
1062         down_write(&cpufreq_notifier_rwsem);
1063         switch (list) {
1064         case CPUFREQ_TRANSITION_NOTIFIER:
1065                 ret = notifier_chain_register(&cpufreq_transition_notifier_list, nb);
1066                 break;
1067         case CPUFREQ_POLICY_NOTIFIER:
1068                 ret = notifier_chain_register(&cpufreq_policy_notifier_list, nb);
1069                 break;
1070         default:
1071                 ret = -EINVAL;
1072         }
1073         up_write(&cpufreq_notifier_rwsem);
1074
1075         return ret;
1076 }
1077 EXPORT_SYMBOL(cpufreq_register_notifier);
1078
1079
1080 /**
1081  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1082  *      @nb: notifier block to be unregistered
1083  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1084  *
1085  *      Remove a driver from the CPU frequency notifier list.
1086  *
1087  *      This function may sleep, and has the same return conditions as
1088  *      notifier_chain_unregister.
1089  */
1090 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1091 {
1092         int ret;
1093
1094         down_write(&cpufreq_notifier_rwsem);
1095         switch (list) {
1096         case CPUFREQ_TRANSITION_NOTIFIER:
1097                 ret = notifier_chain_unregister(&cpufreq_transition_notifier_list, nb);
1098                 break;
1099         case CPUFREQ_POLICY_NOTIFIER:
1100                 ret = notifier_chain_unregister(&cpufreq_policy_notifier_list, nb);
1101                 break;
1102         default:
1103                 ret = -EINVAL;
1104         }
1105         up_write(&cpufreq_notifier_rwsem);
1106
1107         return ret;
1108 }
1109 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1110
1111
1112 /*********************************************************************
1113  *                              GOVERNORS                            *
1114  *********************************************************************/
1115
1116
1117 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1118                             unsigned int target_freq,
1119                             unsigned int relation)
1120 {
1121         int retval = -EINVAL;
1122         lock_cpu_hotplug();
1123         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1124                 target_freq, relation);
1125         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1126                 retval = cpufreq_driver->target(policy, target_freq, relation);
1127         unlock_cpu_hotplug();
1128         return retval;
1129 }
1130 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1131
1132
1133 int cpufreq_driver_target(struct cpufreq_policy *policy,
1134                           unsigned int target_freq,
1135                           unsigned int relation)
1136 {
1137         int ret;
1138
1139         policy = cpufreq_cpu_get(policy->cpu);
1140         if (!policy)
1141                 return -EINVAL;
1142
1143         down(&policy->lock);
1144
1145         ret = __cpufreq_driver_target(policy, target_freq, relation);
1146
1147         up(&policy->lock);
1148
1149         cpufreq_cpu_put(policy);
1150
1151         return ret;
1152 }
1153 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1154
1155
1156 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1157 {
1158         int ret;
1159
1160         if (!try_module_get(policy->governor->owner))
1161                 return -EINVAL;
1162
1163         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1164         ret = policy->governor->governor(policy, event);
1165
1166         /* we keep one module reference alive for each CPU governed by this CPU */
1167         if ((event != CPUFREQ_GOV_START) || ret)
1168                 module_put(policy->governor->owner);
1169         if ((event == CPUFREQ_GOV_STOP) && !ret)
1170                 module_put(policy->governor->owner);
1171
1172         return ret;
1173 }
1174
1175
1176 int cpufreq_governor(unsigned int cpu, unsigned int event)
1177 {
1178         int ret = 0;
1179         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1180
1181         if (!policy)
1182                 return -EINVAL;
1183
1184         down(&policy->lock);
1185         ret = __cpufreq_governor(policy, event);
1186         up(&policy->lock);
1187
1188         cpufreq_cpu_put(policy);
1189
1190         return ret;
1191 }
1192 EXPORT_SYMBOL_GPL(cpufreq_governor);
1193
1194
1195 int cpufreq_register_governor(struct cpufreq_governor *governor)
1196 {
1197         struct cpufreq_governor *t;
1198
1199         if (!governor)
1200                 return -EINVAL;
1201
1202         down(&cpufreq_governor_sem);
1203
1204         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1205                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1206                         up(&cpufreq_governor_sem);
1207                         return -EBUSY;
1208                 }
1209         }
1210         list_add(&governor->governor_list, &cpufreq_governor_list);
1211
1212         up(&cpufreq_governor_sem);
1213
1214         return 0;
1215 }
1216 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1217
1218
1219 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1220 {
1221         if (!governor)
1222                 return;
1223
1224         down(&cpufreq_governor_sem);
1225         list_del(&governor->governor_list);
1226         up(&cpufreq_governor_sem);
1227         return;
1228 }
1229 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1230
1231
1232
1233 /*********************************************************************
1234  *                          POLICY INTERFACE                         *
1235  *********************************************************************/
1236
1237 /**
1238  * cpufreq_get_policy - get the current cpufreq_policy
1239  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1240  *
1241  * Reads the current cpufreq policy.
1242  */
1243 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1244 {
1245         struct cpufreq_policy *cpu_policy;
1246         if (!policy)
1247                 return -EINVAL;
1248
1249         cpu_policy = cpufreq_cpu_get(cpu);
1250         if (!cpu_policy)
1251                 return -EINVAL;
1252
1253         down(&cpu_policy->lock);
1254         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1255         up(&cpu_policy->lock);
1256
1257         cpufreq_cpu_put(cpu_policy);
1258
1259         return 0;
1260 }
1261 EXPORT_SYMBOL(cpufreq_get_policy);
1262
1263
1264 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1265 {
1266         int ret = 0;
1267
1268         cpufreq_debug_disable_ratelimit();
1269         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1270                 policy->min, policy->max);
1271
1272         memcpy(&policy->cpuinfo,
1273                &data->cpuinfo,
1274                sizeof(struct cpufreq_cpuinfo));
1275
1276         /* verify the cpu speed can be set within this limit */
1277         ret = cpufreq_driver->verify(policy);
1278         if (ret)
1279                 goto error_out;
1280
1281         down_read(&cpufreq_notifier_rwsem);
1282
1283         /* adjust if necessary - all reasons */
1284         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_ADJUST,
1285                             policy);
1286
1287         /* adjust if necessary - hardware incompatibility*/
1288         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_INCOMPATIBLE,
1289                             policy);
1290
1291         /* verify the cpu speed can be set within this limit,
1292            which might be different to the first one */
1293         ret = cpufreq_driver->verify(policy);
1294         if (ret) {
1295                 up_read(&cpufreq_notifier_rwsem);
1296                 goto error_out;
1297         }
1298
1299         /* notification of the new policy */
1300         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_NOTIFY,
1301                             policy);
1302
1303         up_read(&cpufreq_notifier_rwsem);
1304
1305         data->min    = policy->min;
1306         data->max    = policy->max;
1307
1308         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1309
1310         if (cpufreq_driver->setpolicy) {
1311                 data->policy = policy->policy;
1312                 dprintk("setting range\n");
1313                 ret = cpufreq_driver->setpolicy(policy);
1314         } else {
1315                 if (policy->governor != data->governor) {
1316                         /* save old, working values */
1317                         struct cpufreq_governor *old_gov = data->governor;
1318
1319                         dprintk("governor switch\n");
1320
1321                         /* end old governor */
1322                         if (data->governor)
1323                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1324
1325                         /* start new governor */
1326                         data->governor = policy->governor;
1327                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1328                                 /* new governor failed, so re-start old one */
1329                                 dprintk("starting governor %s failed\n", data->governor->name);
1330                                 if (old_gov) {
1331                                         data->governor = old_gov;
1332                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1333                                 }
1334                                 ret = -EINVAL;
1335                                 goto error_out;
1336                         }
1337                         /* might be a policy change, too, so fall through */
1338                 }
1339                 dprintk("governor: change or update limits\n");
1340                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1341         }
1342
1343  error_out:
1344         cpufreq_debug_enable_ratelimit();
1345         return ret;
1346 }
1347
1348 /**
1349  *      cpufreq_set_policy - set a new CPUFreq policy
1350  *      @policy: policy to be set.
1351  *
1352  *      Sets a new CPU frequency and voltage scaling policy.
1353  */
1354 int cpufreq_set_policy(struct cpufreq_policy *policy)
1355 {
1356         int ret = 0;
1357         struct cpufreq_policy *data;
1358
1359         if (!policy)
1360                 return -EINVAL;
1361
1362         data = cpufreq_cpu_get(policy->cpu);
1363         if (!data)
1364                 return -EINVAL;
1365
1366         /* lock this CPU */
1367         down(&data->lock);
1368
1369         ret = __cpufreq_set_policy(data, policy);
1370         data->user_policy.min = data->min;
1371         data->user_policy.max = data->max;
1372         data->user_policy.policy = data->policy;
1373         data->user_policy.governor = data->governor;
1374
1375         up(&data->lock);
1376         cpufreq_cpu_put(data);
1377
1378         return ret;
1379 }
1380 EXPORT_SYMBOL(cpufreq_set_policy);
1381
1382
1383 /**
1384  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1385  *      @cpu: CPU which shall be re-evaluated
1386  *
1387  *      Usefull for policy notifiers which have different necessities
1388  *      at different times.
1389  */
1390 int cpufreq_update_policy(unsigned int cpu)
1391 {
1392         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1393         struct cpufreq_policy policy;
1394         int ret = 0;
1395
1396         if (!data)
1397                 return -ENODEV;
1398
1399         down(&data->lock);
1400
1401         dprintk("updating policy for CPU %u\n", cpu);
1402         memcpy(&policy,
1403                data,
1404                sizeof(struct cpufreq_policy));
1405         policy.min = data->user_policy.min;
1406         policy.max = data->user_policy.max;
1407         policy.policy = data->user_policy.policy;
1408         policy.governor = data->user_policy.governor;
1409
1410         ret = __cpufreq_set_policy(data, &policy);
1411
1412         up(&data->lock);
1413
1414         cpufreq_cpu_put(data);
1415         return ret;
1416 }
1417 EXPORT_SYMBOL(cpufreq_update_policy);
1418
1419
1420 /*********************************************************************
1421  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1422  *********************************************************************/
1423
1424 /**
1425  * cpufreq_register_driver - register a CPU Frequency driver
1426  * @driver_data: A struct cpufreq_driver containing the values#
1427  * submitted by the CPU Frequency driver.
1428  *
1429  *   Registers a CPU Frequency driver to this core code. This code
1430  * returns zero on success, -EBUSY when another driver got here first
1431  * (and isn't unregistered in the meantime).
1432  *
1433  */
1434 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1435 {
1436         unsigned long flags;
1437         int ret;
1438
1439         if (!driver_data || !driver_data->verify || !driver_data->init ||
1440             ((!driver_data->setpolicy) && (!driver_data->target)))
1441                 return -EINVAL;
1442
1443         dprintk("trying to register driver %s\n", driver_data->name);
1444
1445         if (driver_data->setpolicy)
1446                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1447
1448         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1449         if (cpufreq_driver) {
1450                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1451                 return -EBUSY;
1452         }
1453         cpufreq_driver = driver_data;
1454         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1455
1456         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1457
1458         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1459                 int i;
1460                 ret = -ENODEV;
1461
1462                 /* check for at least one working CPU */
1463                 for (i=0; i<NR_CPUS; i++)
1464                         if (cpufreq_cpu_data[i])
1465                                 ret = 0;
1466
1467                 /* if all ->init() calls failed, unregister */
1468                 if (ret) {
1469                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1470                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1471
1472                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1473                         cpufreq_driver = NULL;
1474                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1475                 }
1476         }
1477
1478         if (!ret) {
1479                 dprintk("driver %s up and running\n", driver_data->name);
1480                 cpufreq_debug_enable_ratelimit();
1481         }
1482
1483         return (ret);
1484 }
1485 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1486
1487
1488 /**
1489  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1490  *
1491  *    Unregister the current CPUFreq driver. Only call this if you have
1492  * the right to do so, i.e. if you have succeeded in initialising before!
1493  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1494  * currently not initialised.
1495  */
1496 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1497 {
1498         unsigned long flags;
1499
1500         cpufreq_debug_disable_ratelimit();
1501
1502         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1503                 cpufreq_debug_enable_ratelimit();
1504                 return -EINVAL;
1505         }
1506
1507         dprintk("unregistering driver %s\n", driver->name);
1508
1509         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1510
1511         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1512         cpufreq_driver = NULL;
1513         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1514
1515         return 0;
1516 }
1517 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);