git.oblomov.eu Git - linux-2.6/blob - 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  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
   8  *                      Added handling for CPU hotplug
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  *
  14  */
  15
  16 #include <linux/config.h>
  17 #include <linux/kernel.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/notifier.h>
  21 #include <linux/cpufreq.h>
  22 #include <linux/delay.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/device.h>
  26 #include <linux/slab.h>
  27 #include <linux/cpu.h>
  28 #include <linux/completion.h>
  29
  30 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
  31
  32 /**
  33  * The "cpufreq driver" - the arch- or hardware-dependend low
  34  * level driver of CPUFreq support, and its spinlock. This lock
  35  * also protects the cpufreq_cpu_data array.
  36  */
  37 static struct cpufreq_driver    *cpufreq_driver;
  38 static struct cpufreq_policy    *cpufreq_cpu_data[NR_CPUS];
  39 static DEFINE_SPINLOCK(cpufreq_driver_lock);
  40
  41 /* internal prototypes */
  42 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
  43 static void handle_update(void *data);
  44 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci);
  45
  46 /**
  47  * Two notifier lists: the "policy" list is involved in the
  48  * validation process for a new CPU frequency policy; the
  49  * "transition" list for kernel code that needs to handle
  50  * changes to devices when the CPU clock speed changes.
  51  * The mutex locks both lists.
  52  */
  53 static struct notifier_block    *cpufreq_policy_notifier_list;
  54 static struct notifier_block    *cpufreq_transition_notifier_list;
  55 static DECLARE_RWSEM            (cpufreq_notifier_rwsem);
  56
  57
  58 static LIST_HEAD(cpufreq_governor_list);
  59 static DECLARE_MUTEX            (cpufreq_governor_sem);
  60
  61 struct cpufreq_policy * cpufreq_cpu_get(unsigned int cpu)
  62 {
  63         struct cpufreq_policy *data;
  64         unsigned long flags;
  65
  66         if (cpu >= NR_CPUS)
  67                 goto err_out;
  68
  69         /* get the cpufreq driver */
  70         spin_lock_irqsave(&cpufreq_driver_lock, flags);
  71
  72         if (!cpufreq_driver)
  73                 goto err_out_unlock;
  74
  75         if (!try_module_get(cpufreq_driver->owner))
  76                 goto err_out_unlock;
  77
  78
  79         /* get the CPU */
  80         data = cpufreq_cpu_data[cpu];
  81
  82         if (!data)
  83                 goto err_out_put_module;
  84
  85         if (!kobject_get(&data->kobj))
  86                 goto err_out_put_module;
  87
  88
  89         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  90
  91         return data;
  92
  93  err_out_put_module:
  94         module_put(cpufreq_driver->owner);
  95  err_out_unlock:
  96         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  97  err_out:
  98         return NULL;
  99 }
 100 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 101
 102 void cpufreq_cpu_put(struct cpufreq_policy *data)
 103 {
 104         kobject_put(&data->kobj);
 105         module_put(cpufreq_driver->owner);
 106 }
 107 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 108
 109
 110 /*********************************************************************
 111  *                     UNIFIED DEBUG HELPERS                         *
 112  *********************************************************************/
 113 #ifdef CONFIG_CPU_FREQ_DEBUG
 114
 115 /* what part(s) of the CPUfreq subsystem are debugged? */
 116 static unsigned int debug;
 117
 118 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 119  * set or modify this value.
 120  */
 121 static unsigned int debug_ratelimit = 1;
 122
 123 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 124  * loading of a cpufreq driver, temporarily disabled when a new policy
 125  * is set, and disabled upon cpufreq driver removal
 126  */
 127 static unsigned int disable_ratelimit = 1;
 128 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 129
 130 static inline void cpufreq_debug_enable_ratelimit(void)
 131 {
 132         unsigned long flags;
 133
 134         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 135         if (disable_ratelimit)
 136                 disable_ratelimit--;
 137         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 138 }
 139
 140 static inline void cpufreq_debug_disable_ratelimit(void)
 141 {
 142         unsigned long flags;
 143
 144         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 145         disable_ratelimit++;
 146         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 147 }
 148
 149 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
 150 {
 151         char s[256];
 152         va_list args;
 153         unsigned int len;
 154         unsigned long flags;
 155
 156         WARN_ON(!prefix);
 157         if (type & debug) {
 158                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 159                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
 160                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 161                         return;
 162                 }
 163                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 164
 165                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 166
 167                 va_start(args, fmt);
 168                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 169                 va_end(args);
 170
 171                 printk(s);
 172
 173                 WARN_ON(len < 5);
 174         }
 175 }
 176 EXPORT_SYMBOL(cpufreq_debug_printk);
 177
 178
 179 module_param(debug, uint, 0644);
 180 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
 181
 182 module_param(debug_ratelimit, uint, 0644);
 183 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
 184
 185 #else /* !CONFIG_CPU_FREQ_DEBUG */
 186
 187 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 188 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 189
 190 #endif /* CONFIG_CPU_FREQ_DEBUG */
 191
 192
 193 /*********************************************************************
 194  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 195  *********************************************************************/
 196
 197 /**
 198  * adjust_jiffies - adjust the system "loops_per_jiffy"
 199  *
 200  * This function alters the system "loops_per_jiffy" for the clock
 201  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 202  * systems as each CPU might be scaled differently. So, use the arch
 203  * per-CPU loops_per_jiffy value wherever possible.
 204  */
 205 #ifndef CONFIG_SMP
 206 static unsigned long l_p_j_ref;
 207 static unsigned int  l_p_j_ref_freq;
 208
 209 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 210 {
 211         if (ci->flags & CPUFREQ_CONST_LOOPS)
 212                 return;
 213
 214         if (!l_p_j_ref_freq) {
 215                 l_p_j_ref = loops_per_jiffy;
 216                 l_p_j_ref_freq = ci->old;
 217                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 218         }
 219         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 220             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 221             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 222                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
 223                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
 224         }
 225 }
 226 #else
 227 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
 228 #endif
 229
 230
 231 /**
 232  * cpufreq_notify_transition - call notifier chain and adjust_jiffies on frequency transition
 233  *
 234  * This function calls the transition notifiers and the "adjust_jiffies" function. It is called
 235  * twice on all CPU frequency changes that have external effects.
 236  */
 237 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 238 {
 239         BUG_ON(irqs_disabled());
 240
 241         freqs->flags = cpufreq_driver->flags;
 242         dprintk("notification %u of frequency transition to %u kHz\n", state, freqs->new);
 243
 244         down_read(&cpufreq_notifier_rwsem);
 245         switch (state) {
 246         case CPUFREQ_PRECHANGE:
 247                 /* detect if the driver reported a value as "old frequency" which
 248                  * is not equal to what the cpufreq core thinks is "old frequency".
 249                  */
 250                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 251                         if ((likely(cpufreq_cpu_data[freqs->cpu])) &&
 252                             (likely(cpufreq_cpu_data[freqs->cpu]->cpu == freqs->cpu)) &&
 253                             (likely(cpufreq_cpu_data[freqs->cpu]->cur)) &&
 254                             (unlikely(freqs->old != cpufreq_cpu_data[freqs->cpu]->cur)))
 255                         {
 256                                 dprintk(KERN_WARNING "Warning: CPU frequency is %u, "
 257                                        "cpufreq assumed %u kHz.\n", freqs->old, cpufreq_cpu_data[freqs->cpu]->cur);
 258                                 freqs->old = cpufreq_cpu_data[freqs->cpu]->cur;
 259                         }
 260                 }
 261                 notifier_call_chain(&cpufreq_transition_notifier_list, CPUFREQ_PRECHANGE, freqs);
 262                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 263                 break;
 264         case CPUFREQ_POSTCHANGE:
 265                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 266                 notifier_call_chain(&cpufreq_transition_notifier_list, CPUFREQ_POSTCHANGE, freqs);
 267                 if ((likely(cpufreq_cpu_data[freqs->cpu])) &&
 268                     (likely(cpufreq_cpu_data[freqs->cpu]->cpu == freqs->cpu)))
 269                         cpufreq_cpu_data[freqs->cpu]->cur = freqs->new;
 270                 break;
 271         }
 272         up_read(&cpufreq_notifier_rwsem);
 273 }
 274 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 275
 276
 277
 278 /*********************************************************************
 279  *                          SYSFS INTERFACE                          *
 280  *********************************************************************/
 281
 282 /**
 283  * cpufreq_parse_governor - parse a governor string
 284  */
 285 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 286                                 struct cpufreq_governor **governor)
 287 {
 288         if (!cpufreq_driver)
 289                 return -EINVAL;
 290         if (cpufreq_driver->setpolicy) {
 291                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 292                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 293                         return 0;
 294                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 295                         *policy = CPUFREQ_POLICY_POWERSAVE;
 296                         return 0;
 297                 }
 298                 return -EINVAL;
 299         } else {
 300                 struct cpufreq_governor *t;
 301                 down(&cpufreq_governor_sem);
 302                 if (!cpufreq_driver || !cpufreq_driver->target)
 303                         goto out;
 304                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 305                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
 306                                 *governor = t;
 307                                 up(&cpufreq_governor_sem);
 308                                 return 0;
 309                         }
 310                 }
 311         out:
 312                 up(&cpufreq_governor_sem);
 313         }
 314         return -EINVAL;
 315 }
 316 EXPORT_SYMBOL_GPL(cpufreq_parse_governor);
 317
 318
 319 /* drivers/base/cpu.c */
 320 extern struct sysdev_class cpu_sysdev_class;
 321
 322
 323 /**
 324  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 325  *
 326  * Write out information from cpufreq_driver->policy[cpu]; object must be
 327  * "unsigned int".
 328  */
 329
 330 #define show_one(file_name, object)                                     \
 331 static ssize_t show_##file_name                                         \
 332 (struct cpufreq_policy * policy, char *buf)                             \
 333 {                                                                       \
 334         return sprintf (buf, "%u\n", policy->object);                   \
 335 }
 336
 337 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 338 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 339 show_one(scaling_min_freq, min);
 340 show_one(scaling_max_freq, max);
 341 show_one(scaling_cur_freq, cur);
 342
 343 /**
 344  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 345  */
 346 #define store_one(file_name, object)                    \
 347 static ssize_t store_##file_name                                        \
 348 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 349 {                                                                       \
 350         unsigned int ret = -EINVAL;                                     \
 351         struct cpufreq_policy new_policy;                               \
 352                                                                         \
 353         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 354         if (ret)                                                        \
 355                 return -EINVAL;                                         \
 356                                                                         \
 357         ret = sscanf (buf, "%u", &new_policy.object);                   \
 358         if (ret != 1)                                                   \
 359                 return -EINVAL;                                         \
 360                                                                         \
 361         ret = cpufreq_set_policy(&new_policy);                          \
 362                                                                         \
 363         return ret ? ret : count;                                       \
 364 }
 365
 366 store_one(scaling_min_freq,min);
 367 store_one(scaling_max_freq,max);
 368
 369 /**
 370  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 371  */
 372 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 373 {
 374         unsigned int cur_freq = cpufreq_get(policy->cpu);
 375         if (!cur_freq)
 376                 return sprintf(buf, "<unknown>");
 377         return sprintf(buf, "%u\n", cur_freq);
 378 }
 379
 380
 381 /**
 382  * show_scaling_governor - show the current policy for the specified CPU
 383  */
 384 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 385 {
 386         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 387                 return sprintf(buf, "powersave\n");
 388         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 389                 return sprintf(buf, "performance\n");
 390         else if (policy->governor)
 391                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 392         return -EINVAL;
 393 }
 394
 395
 396 /**
 397  * store_scaling_governor - store policy for the specified CPU
 398  */
 399 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 400                                        const char *buf, size_t count)
 401 {
 402         unsigned int ret = -EINVAL;
 403         char    str_governor[16];
 404         struct cpufreq_policy new_policy;
 405
 406         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 407         if (ret)
 408                 return ret;
 409
 410         ret = sscanf (buf, "%15s", str_governor);
 411         if (ret != 1)
 412                 return -EINVAL;
 413
 414         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 415                 return -EINVAL;
 416
 417         ret = cpufreq_set_policy(&new_policy);
 418
 419         return ret ? ret : count;
 420 }
 421
 422 /**
 423  * show_scaling_driver - show the cpufreq driver currently loaded
 424  */
 425 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 426 {
 427         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 428 }
 429
 430 /**
 431  * show_scaling_available_governors - show the available CPUfreq governors
 432  */
 433 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 434                                 char *buf)
 435 {
 436         ssize_t i = 0;
 437         struct cpufreq_governor *t;
 438
 439         if (!cpufreq_driver->target) {
 440                 i += sprintf(buf, "performance powersave");
 441                 goto out;
 442         }
 443
 444         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 445                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 446                         goto out;
 447                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 448         }
 449  out:
 450         i += sprintf(&buf[i], "\n");
 451         return i;
 452 }
 453 /**
 454  * show_affected_cpus - show the CPUs affected by each transition
 455  */
 456 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 457 {
 458         ssize_t i = 0;
 459         unsigned int cpu;
 460
 461         for_each_cpu_mask(cpu, policy->cpus) {
 462                 if (i)
 463                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 464                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 465                 if (i >= (PAGE_SIZE - 5))
 466                     break;
 467         }
 468         i += sprintf(&buf[i], "\n");
 469         return i;
 470 }
 471
 472
 473 #define define_one_ro(_name) \
 474 static struct freq_attr _name = \
 475 __ATTR(_name, 0444, show_##_name, NULL)
 476
 477 #define define_one_ro0400(_name) \
 478 static struct freq_attr _name = \
 479 __ATTR(_name, 0400, show_##_name, NULL)
 480
 481 #define define_one_rw(_name) \
 482 static struct freq_attr _name = \
 483 __ATTR(_name, 0644, show_##_name, store_##_name)
 484
 485 define_one_ro0400(cpuinfo_cur_freq);
 486 define_one_ro(cpuinfo_min_freq);
 487 define_one_ro(cpuinfo_max_freq);
 488 define_one_ro(scaling_available_governors);
 489 define_one_ro(scaling_driver);
 490 define_one_ro(scaling_cur_freq);
 491 define_one_ro(affected_cpus);
 492 define_one_rw(scaling_min_freq);
 493 define_one_rw(scaling_max_freq);
 494 define_one_rw(scaling_governor);
 495
 496 static struct attribute * default_attrs[] = {
 497         &cpuinfo_min_freq.attr,
 498         &cpuinfo_max_freq.attr,
 499         &scaling_min_freq.attr,
 500         &scaling_max_freq.attr,
 501         &affected_cpus.attr,
 502         &scaling_governor.attr,
 503         &scaling_driver.attr,
 504         &scaling_available_governors.attr,
 505         NULL
 506 };
 507
 508 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 509 #define to_attr(a) container_of(a,struct freq_attr,attr)
 510
 511 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 512 {
 513         struct cpufreq_policy * policy = to_policy(kobj);
 514         struct freq_attr * fattr = to_attr(attr);
 515         ssize_t ret;
 516         policy = cpufreq_cpu_get(policy->cpu);
 517         if (!policy)
 518                 return -EINVAL;
 519         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 520         cpufreq_cpu_put(policy);
 521         return ret;
 522 }
 523
 524 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 525                      const char * buf, size_t count)
 526 {
 527         struct cpufreq_policy * policy = to_policy(kobj);
 528         struct freq_attr * fattr = to_attr(attr);
 529         ssize_t ret;
 530         policy = cpufreq_cpu_get(policy->cpu);
 531         if (!policy)
 532                 return -EINVAL;
 533         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 534         cpufreq_cpu_put(policy);
 535         return ret;
 536 }
 537
 538 static void cpufreq_sysfs_release(struct kobject * kobj)
 539 {
 540         struct cpufreq_policy * policy = to_policy(kobj);
 541         dprintk("last reference is dropped\n");
 542         complete(&policy->kobj_unregister);
 543 }
 544
 545 static struct sysfs_ops sysfs_ops = {
 546         .show   = show,
 547         .store  = store,
 548 };
 549
 550 static struct kobj_type ktype_cpufreq = {
 551         .sysfs_ops      = &sysfs_ops,
 552         .default_attrs  = default_attrs,
 553         .release        = cpufreq_sysfs_release,
 554 };
 555
 556
 557 /**
 558  * cpufreq_add_dev - add a CPU device
 559  *
 560  * Adds the cpufreq interface for a CPU device.
 561  */
 562 static int cpufreq_add_dev (struct sys_device * sys_dev)
 563 {
 564         unsigned int cpu = sys_dev->id;
 565         int ret = 0;
 566         struct cpufreq_policy new_policy;
 567         struct cpufreq_policy *policy;
 568         struct freq_attr **drv_attr;
 569         unsigned long flags;
 570         unsigned int j;
 571
 572         if (cpu_is_offline(cpu))
 573                 return 0;
 574
 575         cpufreq_debug_disable_ratelimit();
 576         dprintk("adding CPU %u\n", cpu);
 577
 578 #ifdef CONFIG_SMP
 579         /* check whether a different CPU already registered this
 580          * CPU because it is in the same boat. */
 581         policy = cpufreq_cpu_get(cpu);
 582         if (unlikely(policy)) {
 583                 dprintk("CPU already managed, adding link\n");
 584                 sysfs_create_link(&sys_dev->kobj, &policy->kobj, "cpufreq");
 585                 cpufreq_debug_enable_ratelimit();
 586                 return 0;
 587         }
 588 #endif
 589
 590         if (!try_module_get(cpufreq_driver->owner)) {
 591                 ret = -EINVAL;
 592                 goto module_out;
 593         }
 594
 595         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 596         if (!policy) {
 597                 ret = -ENOMEM;
 598                 goto nomem_out;
 599         }
 600
 601         policy->cpu = cpu;
 602         policy->cpus = cpumask_of_cpu(cpu);
 603
 604         init_MUTEX_LOCKED(&policy->lock);
 605         init_completion(&policy->kobj_unregister);
 606         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 607
 608         /* call driver. From then on the cpufreq must be able
 609          * to accept all calls to ->verify and ->setpolicy for this CPU
 610          */
 611         ret = cpufreq_driver->init(policy);
 612         if (ret) {
 613                 dprintk("initialization failed\n");
 614                 goto err_out;
 615         }
 616
 617         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 618
 619         /* prepare interface data */
 620         policy->kobj.parent = &sys_dev->kobj;
 621         policy->kobj.ktype = &ktype_cpufreq;
 622         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 623
 624         ret = kobject_register(&policy->kobj);
 625         if (ret)
 626                 goto err_out_driver_exit;
 627
 628         /* set up files for this cpu device */
 629         drv_attr = cpufreq_driver->attr;
 630         while ((drv_attr) && (*drv_attr)) {
 631                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 632                 drv_attr++;
 633         }
 634         if (cpufreq_driver->get)
 635                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 636         if (cpufreq_driver->target)
 637                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 638
 639         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 640         for_each_cpu_mask(j, policy->cpus)
 641                 cpufreq_cpu_data[j] = policy;
 642         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 643         policy->governor = NULL; /* to assure that the starting sequence is
 644                                   * run in cpufreq_set_policy */
 645         up(&policy->lock);
 646
 647         /* set default policy */
 648
 649         ret = cpufreq_set_policy(&new_policy);
 650         if (ret) {
 651                 dprintk("setting policy failed\n");
 652                 goto err_out_unregister;
 653         }
 654
 655         module_put(cpufreq_driver->owner);
 656         dprintk("initialization complete\n");
 657         cpufreq_debug_enable_ratelimit();
 658
 659         return 0;
 660
 661
 662 err_out_unregister:
 663         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 664         for_each_cpu_mask(j, policy->cpus)
 665                 cpufreq_cpu_data[j] = NULL;
 666         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 667
 668         kobject_unregister(&policy->kobj);
 669         wait_for_completion(&policy->kobj_unregister);
 670
 671 err_out_driver_exit:
 672         if (cpufreq_driver->exit)
 673                 cpufreq_driver->exit(policy);
 674
 675 err_out:
 676         kfree(policy);
 677
 678 nomem_out:
 679         module_put(cpufreq_driver->owner);
 680 module_out:
 681         cpufreq_debug_enable_ratelimit();
 682         return ret;
 683 }
 684
 685
 686 /**
 687  * cpufreq_remove_dev - remove a CPU device
 688  *
 689  * Removes the cpufreq interface for a CPU device.
 690  */
 691 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 692 {
 693         unsigned int cpu = sys_dev->id;
 694         unsigned long flags;
 695         struct cpufreq_policy *data;
 696 #ifdef CONFIG_SMP
 697         struct sys_device *cpu_sys_dev;
 698         unsigned int j;
 699 #endif
 700
 701         cpufreq_debug_disable_ratelimit();
 702         dprintk("unregistering CPU %u\n", cpu);
 703
 704         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 705         data = cpufreq_cpu_data[cpu];
 706
 707         if (!data) {
 708                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 709                 cpufreq_debug_enable_ratelimit();
 710                 return -EINVAL;
 711         }
 712         cpufreq_cpu_data[cpu] = NULL;
 713
 714
 715 #ifdef CONFIG_SMP
 716         /* if this isn't the CPU which is the parent of the kobj, we
 717          * only need to unlink, put and exit
 718          */
 719         if (unlikely(cpu != data->cpu)) {
 720                 dprintk("removing link\n");
 721                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 722                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 723                 cpufreq_cpu_put(data);
 724                 cpufreq_debug_enable_ratelimit();
 725                 return 0;
 726         }
 727 #endif
 728
 729
 730         if (!kobject_get(&data->kobj)) {
 731                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 732                 cpufreq_debug_enable_ratelimit();
 733                 return -EFAULT;
 734         }
 735
 736 #ifdef CONFIG_SMP
 737         /* if we have other CPUs still registered, we need to unlink them,
 738          * or else wait_for_completion below will lock up. Clean the
 739          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 740          * links afterwards.
 741          */
 742         if (unlikely(cpus_weight(data->cpus) > 1)) {
 743                 for_each_cpu_mask(j, data->cpus) {
 744                         if (j == cpu)
 745                                 continue;
 746                         cpufreq_cpu_data[j] = NULL;
 747                 }
 748         }
 749
 750         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 751
 752         if (unlikely(cpus_weight(data->cpus) > 1)) {
 753                 for_each_cpu_mask(j, data->cpus) {
 754                         if (j == cpu)
 755                                 continue;
 756                         dprintk("removing link for cpu %u\n", j);
 757                         cpu_sys_dev = get_cpu_sysdev(j);
 758                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 759                         cpufreq_cpu_put(data);
 760                 }
 761         }
 762 #else
 763         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 764 #endif
 765
 766         down(&data->lock);
 767         if (cpufreq_driver->target)
 768                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 769         up(&data->lock);
 770
 771         kobject_unregister(&data->kobj);
 772
 773         kobject_put(&data->kobj);
 774
 775         /* we need to make sure that the underlying kobj is actually
 776          * not referenced anymore by anybody before we proceed with
 777          * unloading.
 778          */
 779         dprintk("waiting for dropping of refcount\n");
 780         wait_for_completion(&data->kobj_unregister);
 781         dprintk("wait complete\n");
 782
 783         if (cpufreq_driver->exit)
 784                 cpufreq_driver->exit(data);
 785
 786         kfree(data);
 787
 788         cpufreq_debug_enable_ratelimit();
 789
 790         return 0;
 791 }
 792
 793
 794 static void handle_update(void *data)
 795 {
 796         unsigned int cpu = (unsigned int)(long)data;
 797         dprintk("handle_update for cpu %u called\n", cpu);
 798         cpufreq_update_policy(cpu);
 799 }
 800
 801 /**
 802  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 803  *      @cpu: cpu number
 804  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 805  *      @new_freq: CPU frequency the CPU actually runs at
 806  *
 807  *      We adjust to current frequency first, and need to clean up later. So either call
 808  *      to cpufreq_update_policy() or schedule handle_update()).
 809  */
 810 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 811 {
 812         struct cpufreq_freqs freqs;
 813
 814         dprintk(KERN_WARNING "Warning: CPU frequency out of sync: cpufreq and timing "
 815                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 816
 817         freqs.cpu = cpu;
 818         freqs.old = old_freq;
 819         freqs.new = new_freq;
 820         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 821         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 822 }
 823
 824
 825 /**
 826  * cpufreq_get - get the current CPU frequency (in kHz)
 827  * @cpu: CPU number
 828  *
 829  * Get the CPU current (static) CPU frequency
 830  */
 831 unsigned int cpufreq_get(unsigned int cpu)
 832 {
 833         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 834         unsigned int ret = 0;
 835
 836         if (!policy)
 837                 return 0;
 838
 839         if (!cpufreq_driver->get)
 840                 goto out;
 841
 842         down(&policy->lock);
 843
 844         ret = cpufreq_driver->get(cpu);
 845
 846         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS))
 847         {
 848                 /* verify no discrepancy between actual and saved value exists */
 849                 if (unlikely(ret != policy->cur)) {
 850                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 851                         schedule_work(&policy->update);
 852                 }
 853         }
 854
 855         up(&policy->lock);
 856
 857  out:
 858         cpufreq_cpu_put(policy);
 859
 860         return (ret);
 861 }
 862 EXPORT_SYMBOL(cpufreq_get);
 863
 864
 865 /**
 866  *      cpufreq_suspend - let the low level driver prepare for suspend
 867  */
 868
 869 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 870 {
 871         int cpu = sysdev->id;
 872         unsigned int ret = 0;
 873         unsigned int cur_freq = 0;
 874         struct cpufreq_policy *cpu_policy;
 875
 876         dprintk("resuming cpu %u\n", cpu);
 877
 878         if (!cpu_online(cpu))
 879                 return 0;
 880
 881         /* we may be lax here as interrupts are off. Nonetheless
 882          * we need to grab the correct cpu policy, as to check
 883          * whether we really run on this CPU.
 884          */
 885
 886         cpu_policy = cpufreq_cpu_get(cpu);
 887         if (!cpu_policy)
 888                 return -EINVAL;
 889
 890         /* only handle each CPU group once */
 891         if (unlikely(cpu_policy->cpu != cpu)) {
 892                 cpufreq_cpu_put(cpu_policy);
 893                 return 0;
 894         }
 895
 896         if (cpufreq_driver->suspend) {
 897                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
 898                 if (ret) {
 899                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
 900                                         "step on CPU %u\n", cpu_policy->cpu);
 901                         cpufreq_cpu_put(cpu_policy);
 902                         return ret;
 903                 }
 904         }
 905
 906
 907         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
 908                 goto out;
 909
 910         if (cpufreq_driver->get)
 911                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
 912
 913         if (!cur_freq || !cpu_policy->cur) {
 914                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
 915                        "frequency is what timing core thinks it is.\n");
 916                 goto out;
 917         }
 918
 919         if (unlikely(cur_freq != cpu_policy->cur)) {
 920                 struct cpufreq_freqs freqs;
 921
 922                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
 923                         dprintk(KERN_DEBUG "Warning: CPU frequency is %u, "
 924                                "cpufreq assumed %u kHz.\n",
 925                                cur_freq, cpu_policy->cur);
 926
 927                 freqs.cpu = cpu;
 928                 freqs.old = cpu_policy->cur;
 929                 freqs.new = cur_freq;
 930
 931                 notifier_call_chain(&cpufreq_transition_notifier_list,
 932                                     CPUFREQ_SUSPENDCHANGE, &freqs);
 933                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
 934
 935                 cpu_policy->cur = cur_freq;
 936         }
 937
 938  out:
 939         cpufreq_cpu_put(cpu_policy);
 940         return 0;
 941 }
 942
 943 /**
 944  *      cpufreq_resume -  restore proper CPU frequency handling after resume
 945  *
 946  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
 947  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
 948  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
 949  *          restored.
 950  */
 951 static int cpufreq_resume(struct sys_device * sysdev)
 952 {
 953         int cpu = sysdev->id;
 954         unsigned int ret = 0;
 955         struct cpufreq_policy *cpu_policy;
 956
 957         dprintk("resuming cpu %u\n", cpu);
 958
 959         if (!cpu_online(cpu))
 960                 return 0;
 961
 962         /* we may be lax here as interrupts are off. Nonetheless
 963          * we need to grab the correct cpu policy, as to check
 964          * whether we really run on this CPU.
 965          */
 966
 967         cpu_policy = cpufreq_cpu_get(cpu);
 968         if (!cpu_policy)
 969                 return -EINVAL;
 970
 971         /* only handle each CPU group once */
 972         if (unlikely(cpu_policy->cpu != cpu)) {
 973                 cpufreq_cpu_put(cpu_policy);
 974                 return 0;
 975         }
 976
 977         if (cpufreq_driver->resume) {
 978                 ret = cpufreq_driver->resume(cpu_policy);
 979                 if (ret) {
 980                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
 981                                         "step on CPU %u\n", cpu_policy->cpu);
 982                         cpufreq_cpu_put(cpu_policy);
 983                         return ret;
 984                 }
 985         }
 986
 987         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 988                 unsigned int cur_freq = 0;
 989
 990                 if (cpufreq_driver->get)
 991                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
 992
 993                 if (!cur_freq || !cpu_policy->cur) {
 994                         printk(KERN_ERR "cpufreq: resume failed to assert "
 995                                         "current frequency is what timing core "
 996                                         "thinks it is.\n");
 997                         goto out;
 998                 }
 999
1000                 if (unlikely(cur_freq != cpu_policy->cur)) {
1001                         struct cpufreq_freqs freqs;
1002
1003                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1004                                 dprintk(KERN_WARNING "Warning: CPU frequency"
1005                                        "is %u, cpufreq assumed %u kHz.\n",
1006                                        cur_freq, cpu_policy->cur);
1007
1008                         freqs.cpu = cpu;
1009                         freqs.old = cpu_policy->cur;
1010                         freqs.new = cur_freq;
1011
1012                         notifier_call_chain(&cpufreq_transition_notifier_list,
1013                                         CPUFREQ_RESUMECHANGE, &freqs);
1014                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1015
1016                         cpu_policy->cur = cur_freq;
1017                 }
1018         }
1019
1020 out:
1021         schedule_work(&cpu_policy->update);
1022         cpufreq_cpu_put(cpu_policy);
1023         return ret;
1024 }
1025
1026 static struct sysdev_driver cpufreq_sysdev_driver = {
1027         .add            = cpufreq_add_dev,
1028         .remove         = cpufreq_remove_dev,
1029         .suspend        = cpufreq_suspend,
1030         .resume         = cpufreq_resume,
1031 };
1032
1033
1034 /*********************************************************************
1035  *                     NOTIFIER LISTS INTERFACE                      *
1036  *********************************************************************/
1037
1038 /**
1039  *      cpufreq_register_notifier - register a driver with cpufreq
1040  *      @nb: notifier function to register
1041  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1042  *
1043  *      Add a driver to one of two lists: either a list of drivers that
1044  *      are notified about clock rate changes (once before and once after
1045  *      the transition), or a list of drivers that are notified about
1046  *      changes in cpufreq policy.
1047  *
1048  *      This function may sleep, and has the same return conditions as
1049  *      notifier_chain_register.
1050  */
1051 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1052 {
1053         int ret;
1054
1055         down_write(&cpufreq_notifier_rwsem);
1056         switch (list) {
1057         case CPUFREQ_TRANSITION_NOTIFIER:
1058                 ret = notifier_chain_register(&cpufreq_transition_notifier_list, nb);
1059                 break;
1060         case CPUFREQ_POLICY_NOTIFIER:
1061                 ret = notifier_chain_register(&cpufreq_policy_notifier_list, nb);
1062                 break;
1063         default:
1064                 ret = -EINVAL;
1065         }
1066         up_write(&cpufreq_notifier_rwsem);
1067
1068         return ret;
1069 }
1070 EXPORT_SYMBOL(cpufreq_register_notifier);
1071
1072
1073 /**
1074  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1075  *      @nb: notifier block to be unregistered
1076  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1077  *
1078  *      Remove a driver from the CPU frequency notifier list.
1079  *
1080  *      This function may sleep, and has the same return conditions as
1081  *      notifier_chain_unregister.
1082  */
1083 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1084 {
1085         int ret;
1086
1087         down_write(&cpufreq_notifier_rwsem);
1088         switch (list) {
1089         case CPUFREQ_TRANSITION_NOTIFIER:
1090                 ret = notifier_chain_unregister(&cpufreq_transition_notifier_list, nb);
1091                 break;
1092         case CPUFREQ_POLICY_NOTIFIER:
1093                 ret = notifier_chain_unregister(&cpufreq_policy_notifier_list, nb);
1094                 break;
1095         default:
1096                 ret = -EINVAL;
1097         }
1098         up_write(&cpufreq_notifier_rwsem);
1099
1100         return ret;
1101 }
1102 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1103
1104
1105 /*********************************************************************
1106  *                              GOVERNORS                            *
1107  *********************************************************************/
1108
1109
1110 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1111                             unsigned int target_freq,
1112                             unsigned int relation)
1113 {
1114         int retval = -EINVAL;
1115
1116         lock_cpu_hotplug();
1117         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1118                 target_freq, relation);
1119         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1120                 retval = cpufreq_driver->target(policy, target_freq, relation);
1121
1122         unlock_cpu_hotplug();
1123
1124         return retval;
1125 }
1126 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1127
1128 int cpufreq_driver_target(struct cpufreq_policy *policy,
1129                           unsigned int target_freq,
1130                           unsigned int relation)
1131 {
1132         int ret;
1133
1134         policy = cpufreq_cpu_get(policy->cpu);
1135         if (!policy)
1136                 return -EINVAL;
1137
1138         down(&policy->lock);
1139
1140         ret = __cpufreq_driver_target(policy, target_freq, relation);
1141
1142         up(&policy->lock);
1143
1144         cpufreq_cpu_put(policy);
1145
1146         return ret;
1147 }
1148 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1149
1150
1151 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1152 {
1153         int ret;
1154
1155         if (!try_module_get(policy->governor->owner))
1156                 return -EINVAL;
1157
1158         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1159         ret = policy->governor->governor(policy, event);
1160
1161         /* we keep one module reference alive for each CPU governed by this CPU */
1162         if ((event != CPUFREQ_GOV_START) || ret)
1163                 module_put(policy->governor->owner);
1164         if ((event == CPUFREQ_GOV_STOP) && !ret)
1165                 module_put(policy->governor->owner);
1166
1167         return ret;
1168 }
1169
1170
1171 int cpufreq_governor(unsigned int cpu, unsigned int event)
1172 {
1173         int ret = 0;
1174         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1175
1176         if (!policy)
1177                 return -EINVAL;
1178
1179         down(&policy->lock);
1180         ret = __cpufreq_governor(policy, event);
1181         up(&policy->lock);
1182
1183         cpufreq_cpu_put(policy);
1184
1185         return ret;
1186 }
1187 EXPORT_SYMBOL_GPL(cpufreq_governor);
1188
1189
1190 int cpufreq_register_governor(struct cpufreq_governor *governor)
1191 {
1192         struct cpufreq_governor *t;
1193
1194         if (!governor)
1195                 return -EINVAL;
1196
1197         down(&cpufreq_governor_sem);
1198
1199         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1200                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1201                         up(&cpufreq_governor_sem);
1202                         return -EBUSY;
1203                 }
1204         }
1205         list_add(&governor->governor_list, &cpufreq_governor_list);
1206
1207         up(&cpufreq_governor_sem);
1208
1209         return 0;
1210 }
1211 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1212
1213
1214 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1215 {
1216         if (!governor)
1217                 return;
1218
1219         down(&cpufreq_governor_sem);
1220         list_del(&governor->governor_list);
1221         up(&cpufreq_governor_sem);
1222         return;
1223 }
1224 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1225
1226
1227
1228 /*********************************************************************
1229  *                          POLICY INTERFACE                         *
1230  *********************************************************************/
1231
1232 /**
1233  * cpufreq_get_policy - get the current cpufreq_policy
1234  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1235  *
1236  * Reads the current cpufreq policy.
1237  */
1238 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1239 {
1240         struct cpufreq_policy *cpu_policy;
1241         if (!policy)
1242                 return -EINVAL;
1243
1244         cpu_policy = cpufreq_cpu_get(cpu);
1245         if (!cpu_policy)
1246                 return -EINVAL;
1247
1248         down(&cpu_policy->lock);
1249         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1250         up(&cpu_policy->lock);
1251
1252         cpufreq_cpu_put(cpu_policy);
1253
1254         return 0;
1255 }
1256 EXPORT_SYMBOL(cpufreq_get_policy);
1257
1258
1259 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1260 {
1261         int ret = 0;
1262
1263         cpufreq_debug_disable_ratelimit();
1264         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1265                 policy->min, policy->max);
1266
1267         memcpy(&policy->cpuinfo,
1268                &data->cpuinfo,
1269                sizeof(struct cpufreq_cpuinfo));
1270
1271         /* verify the cpu speed can be set within this limit */
1272         ret = cpufreq_driver->verify(policy);
1273         if (ret)
1274                 goto error_out;
1275
1276         down_read(&cpufreq_notifier_rwsem);
1277
1278         /* adjust if necessary - all reasons */
1279         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_ADJUST,
1280                             policy);
1281
1282         /* adjust if necessary - hardware incompatibility*/
1283         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_INCOMPATIBLE,
1284                             policy);
1285
1286         /* verify the cpu speed can be set within this limit,
1287            which might be different to the first one */
1288         ret = cpufreq_driver->verify(policy);
1289         if (ret) {
1290                 up_read(&cpufreq_notifier_rwsem);
1291                 goto error_out;
1292         }
1293
1294         /* notification of the new policy */
1295         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_NOTIFY,
1296                             policy);
1297
1298         up_read(&cpufreq_notifier_rwsem);
1299
1300         data->min    = policy->min;
1301         data->max    = policy->max;
1302
1303         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1304
1305         if (cpufreq_driver->setpolicy) {
1306                 data->policy = policy->policy;
1307                 dprintk("setting range\n");
1308                 ret = cpufreq_driver->setpolicy(policy);
1309         } else {
1310                 if (policy->governor != data->governor) {
1311                         /* save old, working values */
1312                         struct cpufreq_governor *old_gov = data->governor;
1313
1314                         dprintk("governor switch\n");
1315
1316                         /* end old governor */
1317                         if (data->governor)
1318                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1319
1320                         /* start new governor */
1321                         data->governor = policy->governor;
1322                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1323                                 /* new governor failed, so re-start old one */
1324                                 dprintk("starting governor %s failed\n", data->governor->name);
1325                                 if (old_gov) {
1326                                         data->governor = old_gov;
1327                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1328                                 }
1329                                 ret = -EINVAL;
1330                                 goto error_out;
1331                         }
1332                         /* might be a policy change, too, so fall through */
1333                 }
1334                 dprintk("governor: change or update limits\n");
1335                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1336         }
1337
1338  error_out:
1339         cpufreq_debug_enable_ratelimit();
1340         return ret;
1341 }
1342
1343 /**
1344  *      cpufreq_set_policy - set a new CPUFreq policy
1345  *      @policy: policy to be set.
1346  *
1347  *      Sets a new CPU frequency and voltage scaling policy.
1348  */
1349 int cpufreq_set_policy(struct cpufreq_policy *policy)
1350 {
1351         int ret = 0;
1352         struct cpufreq_policy *data;
1353
1354         if (!policy)
1355                 return -EINVAL;
1356
1357         data = cpufreq_cpu_get(policy->cpu);
1358         if (!data)
1359                 return -EINVAL;
1360
1361         /* lock this CPU */
1362         down(&data->lock);
1363
1364         ret = __cpufreq_set_policy(data, policy);
1365         data->user_policy.min = data->min;
1366         data->user_policy.max = data->max;
1367         data->user_policy.policy = data->policy;
1368         data->user_policy.governor = data->governor;
1369
1370         up(&data->lock);
1371         cpufreq_cpu_put(data);
1372
1373         return ret;
1374 }
1375 EXPORT_SYMBOL(cpufreq_set_policy);
1376
1377
1378 /**
1379  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1380  *      @cpu: CPU which shall be re-evaluated
1381  *
1382  *      Usefull for policy notifiers which have different necessities
1383  *      at different times.
1384  */
1385 int cpufreq_update_policy(unsigned int cpu)
1386 {
1387         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1388         struct cpufreq_policy policy;
1389         int ret = 0;
1390
1391         if (!data)
1392                 return -ENODEV;
1393
1394         down(&data->lock);
1395
1396         dprintk("updating policy for CPU %u\n", cpu);
1397         memcpy(&policy,
1398                data,
1399                sizeof(struct cpufreq_policy));
1400         policy.min = data->user_policy.min;
1401         policy.max = data->user_policy.max;
1402         policy.policy = data->user_policy.policy;
1403         policy.governor = data->user_policy.governor;
1404
1405         ret = __cpufreq_set_policy(data, &policy);
1406
1407         up(&data->lock);
1408
1409         cpufreq_cpu_put(data);
1410         return ret;
1411 }
1412 EXPORT_SYMBOL(cpufreq_update_policy);
1413
1414 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1415                                         unsigned long action, void *hcpu)
1416 {
1417         unsigned int cpu = (unsigned long)hcpu;
1418         struct cpufreq_policy *policy;
1419         struct sys_device *sys_dev;
1420
1421         sys_dev = get_cpu_sysdev(cpu);
1422
1423         if (sys_dev) {
1424                 switch (action) {
1425                 case CPU_ONLINE:
1426                         cpufreq_add_dev(sys_dev);
1427                         break;
1428                 case CPU_DOWN_PREPARE:
1429                         /*
1430                          * We attempt to put this cpu in lowest frequency
1431                          * possible before going down. This will permit
1432                          * hardware-managed P-State to switch other related
1433                          * threads to min or higher speeds if possible.
1434                          */
1435                         policy = cpufreq_cpu_data[cpu];
1436                         if (policy) {
1437                                 cpufreq_driver_target(policy, policy->min,
1438                                                 CPUFREQ_RELATION_H);
1439                         }
1440                         break;
1441                 case CPU_DEAD:
1442                         cpufreq_remove_dev(sys_dev);
1443                         break;
1444                 }
1445         }
1446         return NOTIFY_OK;
1447 }
1448
1449 static struct notifier_block cpufreq_cpu_notifier =
1450 {
1451     .notifier_call = cpufreq_cpu_callback,
1452 };
1453
1454 /*********************************************************************
1455  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1456  *********************************************************************/
1457
1458 /**
1459  * cpufreq_register_driver - register a CPU Frequency driver
1460  * @driver_data: A struct cpufreq_driver containing the values#
1461  * submitted by the CPU Frequency driver.
1462  *
1463  *   Registers a CPU Frequency driver to this core code. This code
1464  * returns zero on success, -EBUSY when another driver got here first
1465  * (and isn't unregistered in the meantime).
1466  *
1467  */
1468 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1469 {
1470         unsigned long flags;
1471         int ret;
1472
1473         if (!driver_data || !driver_data->verify || !driver_data->init ||
1474             ((!driver_data->setpolicy) && (!driver_data->target)))
1475                 return -EINVAL;
1476
1477         dprintk("trying to register driver %s\n", driver_data->name);
1478
1479         if (driver_data->setpolicy)
1480                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1481
1482         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1483         if (cpufreq_driver) {
1484                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1485                 return -EBUSY;
1486         }
1487         cpufreq_driver = driver_data;
1488         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1489
1490         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1491
1492         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1493                 int i;
1494                 ret = -ENODEV;
1495
1496                 /* check for at least one working CPU */
1497                 for (i=0; i<NR_CPUS; i++)
1498                         if (cpufreq_cpu_data[i])
1499                                 ret = 0;
1500
1501                 /* if all ->init() calls failed, unregister */
1502                 if (ret) {
1503                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1504                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1505
1506                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1507                         cpufreq_driver = NULL;
1508                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1509                 }
1510         }
1511
1512         if (!ret) {
1513                 register_cpu_notifier(&cpufreq_cpu_notifier);
1514                 dprintk("driver %s up and running\n", driver_data->name);
1515                 cpufreq_debug_enable_ratelimit();
1516         }
1517
1518         return (ret);
1519 }
1520 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1521
1522
1523 /**
1524  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1525  *
1526  *    Unregister the current CPUFreq driver. Only call this if you have
1527  * the right to do so, i.e. if you have succeeded in initialising before!
1528  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1529  * currently not initialised.
1530  */
1531 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1532 {
1533         unsigned long flags;
1534
1535         cpufreq_debug_disable_ratelimit();
1536
1537         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1538                 cpufreq_debug_enable_ratelimit();
1539                 return -EINVAL;
1540         }
1541
1542         dprintk("unregistering driver %s\n", driver->name);
1543
1544         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1545         unregister_cpu_notifier(&cpufreq_cpu_notifier);
1546
1547         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1548         cpufreq_driver = NULL;
1549         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1550
1551         return 0;
1552 }
1553 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);