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 #include <linux/mutex.h>
  30
  31 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
  32
  33 /**
  34  * The "cpufreq driver" - the arch- or hardware-dependend low
  35  * level driver of CPUFreq support, and its spinlock. This lock
  36  * also protects the cpufreq_cpu_data array.
  37  */
  38 static struct cpufreq_driver *cpufreq_driver;
  39 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
  40 static DEFINE_SPINLOCK(cpufreq_driver_lock);
  41
  42 /* internal prototypes */
  43 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
  44 static void handle_update(void *data);
  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 DEFINE_MUTEX (cpufreq_governor_mutex);
  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         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  89         return data;
  90
  91 err_out_put_module:
  92         module_put(cpufreq_driver->owner);
  93 err_out_unlock:
  94         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  95 err_out:
  96         return NULL;
  97 }
  98 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
  99
 100
 101 void cpufreq_cpu_put(struct cpufreq_policy *data)
 102 {
 103         kobject_put(&data->kobj);
 104         module_put(cpufreq_driver->owner);
 105 }
 106 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 107
 108
 109 /*********************************************************************
 110  *                     UNIFIED DEBUG HELPERS                         *
 111  *********************************************************************/
 112 #ifdef CONFIG_CPU_FREQ_DEBUG
 113
 114 /* what part(s) of the CPUfreq subsystem are debugged? */
 115 static unsigned int debug;
 116
 117 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 118  * set or modify this value.
 119  */
 120 static unsigned int debug_ratelimit = 1;
 121
 122 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 123  * loading of a cpufreq driver, temporarily disabled when a new policy
 124  * is set, and disabled upon cpufreq driver removal
 125  */
 126 static unsigned int disable_ratelimit = 1;
 127 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 128
 129 static void cpufreq_debug_enable_ratelimit(void)
 130 {
 131         unsigned long flags;
 132
 133         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 134         if (disable_ratelimit)
 135                 disable_ratelimit--;
 136         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 137 }
 138
 139 static void cpufreq_debug_disable_ratelimit(void)
 140 {
 141         unsigned long flags;
 142
 143         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 144         disable_ratelimit++;
 145         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 146 }
 147
 148 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
 149 {
 150         char s[256];
 151         va_list args;
 152         unsigned int len;
 153         unsigned long flags;
 154
 155         WARN_ON(!prefix);
 156         if (type & debug) {
 157                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 158                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
 159                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 160                         return;
 161                 }
 162                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 163
 164                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 165
 166                 va_start(args, fmt);
 167                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 168                 va_end(args);
 169
 170                 printk(s);
 171
 172                 WARN_ON(len < 5);
 173         }
 174 }
 175 EXPORT_SYMBOL(cpufreq_debug_printk);
 176
 177
 178 module_param(debug, uint, 0644);
 179 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
 180
 181 module_param(debug_ratelimit, uint, 0644);
 182 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
 183
 184 #else /* !CONFIG_CPU_FREQ_DEBUG */
 185
 186 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 187 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 188
 189 #endif /* CONFIG_CPU_FREQ_DEBUG */
 190
 191
 192 /*********************************************************************
 193  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 194  *********************************************************************/
 195
 196 /**
 197  * adjust_jiffies - adjust the system "loops_per_jiffy"
 198  *
 199  * This function alters the system "loops_per_jiffy" for the clock
 200  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 201  * systems as each CPU might be scaled differently. So, use the arch
 202  * per-CPU loops_per_jiffy value wherever possible.
 203  */
 204 #ifndef CONFIG_SMP
 205 static unsigned long l_p_j_ref;
 206 static unsigned int  l_p_j_ref_freq;
 207
 208 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 209 {
 210         if (ci->flags & CPUFREQ_CONST_LOOPS)
 211                 return;
 212
 213         if (!l_p_j_ref_freq) {
 214                 l_p_j_ref = loops_per_jiffy;
 215                 l_p_j_ref_freq = ci->old;
 216                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 217         }
 218         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 219             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 220             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 221                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
 222                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
 223         }
 224 }
 225 #else
 226 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
 227 #endif
 228
 229
 230 /**
 231  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 232  * on frequency transition.
 233  *
 234  * This function calls the transition notifiers and the "adjust_jiffies"
 235  * function. It is called twice on all CPU frequency changes that have
 236  * external effects.
 237  */
 238 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 239 {
 240         struct cpufreq_policy *policy;
 241
 242         BUG_ON(irqs_disabled());
 243
 244         freqs->flags = cpufreq_driver->flags;
 245         dprintk("notification %u of frequency transition to %u kHz\n",
 246                 state, freqs->new);
 247
 248         down_read(&cpufreq_notifier_rwsem);
 249
 250         policy = cpufreq_cpu_data[freqs->cpu];
 251         switch (state) {
 252
 253         case CPUFREQ_PRECHANGE:
 254                 /* detect if the driver reported a value as "old frequency"
 255                  * which is not equal to what the cpufreq core thinks is
 256                  * "old frequency".
 257                  */
 258                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 259                         if ((policy) && (policy->cpu == freqs->cpu) &&
 260                             (policy->cur) && (policy->cur != freqs->old)) {
 261                                 dprintk(KERN_WARNING "Warning: CPU frequency is"
 262                                         " %u, cpufreq assumed %u kHz.\n",
 263                                         freqs->old, policy->cur);
 264                                 freqs->old = policy->cur;
 265                         }
 266                 }
 267                 notifier_call_chain(&cpufreq_transition_notifier_list,
 268                                         CPUFREQ_PRECHANGE, freqs);
 269                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 270                 break;
 271
 272         case CPUFREQ_POSTCHANGE:
 273                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 274                 notifier_call_chain(&cpufreq_transition_notifier_list,
 275                                         CPUFREQ_POSTCHANGE, freqs);
 276                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 277                         policy->cur = freqs->new;
 278                 break;
 279         }
 280         up_read(&cpufreq_notifier_rwsem);
 281 }
 282 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 283
 284
 285
 286 /*********************************************************************
 287  *                          SYSFS INTERFACE                          *
 288  *********************************************************************/
 289
 290 /**
 291  * cpufreq_parse_governor - parse a governor string
 292  */
 293 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 294                                 struct cpufreq_governor **governor)
 295 {
 296         if (!cpufreq_driver)
 297                 return -EINVAL;
 298         if (cpufreq_driver->setpolicy) {
 299                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 300                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 301                         return 0;
 302                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 303                         *policy = CPUFREQ_POLICY_POWERSAVE;
 304                         return 0;
 305                 }
 306                 return -EINVAL;
 307         } else {
 308                 struct cpufreq_governor *t;
 309                 mutex_lock(&cpufreq_governor_mutex);
 310                 if (!cpufreq_driver || !cpufreq_driver->target)
 311                         goto out;
 312                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 313                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
 314                                 *governor = t;
 315                                 mutex_unlock(&cpufreq_governor_mutex);
 316                                 return 0;
 317                         }
 318                 }
 319 out:
 320                 mutex_unlock(&cpufreq_governor_mutex);
 321         }
 322         return -EINVAL;
 323 }
 324 EXPORT_SYMBOL_GPL(cpufreq_parse_governor);
 325
 326
 327 /* drivers/base/cpu.c */
 328 extern struct sysdev_class cpu_sysdev_class;
 329
 330
 331 /**
 332  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 333  *
 334  * Write out information from cpufreq_driver->policy[cpu]; object must be
 335  * "unsigned int".
 336  */
 337
 338 #define show_one(file_name, object)                                     \
 339 static ssize_t show_##file_name                                         \
 340 (struct cpufreq_policy * policy, char *buf)                             \
 341 {                                                                       \
 342         return sprintf (buf, "%u\n", policy->object);                   \
 343 }
 344
 345 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 346 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 347 show_one(scaling_min_freq, min);
 348 show_one(scaling_max_freq, max);
 349 show_one(scaling_cur_freq, cur);
 350
 351 /**
 352  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 353  */
 354 #define store_one(file_name, object)                    \
 355 static ssize_t store_##file_name                                        \
 356 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 357 {                                                                       \
 358         unsigned int ret = -EINVAL;                                     \
 359         struct cpufreq_policy new_policy;                               \
 360                                                                         \
 361         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 362         if (ret)                                                        \
 363                 return -EINVAL;                                         \
 364                                                                         \
 365         ret = sscanf (buf, "%u", &new_policy.object);                   \
 366         if (ret != 1)                                                   \
 367                 return -EINVAL;                                         \
 368                                                                         \
 369         ret = cpufreq_set_policy(&new_policy);                          \
 370                                                                         \
 371         return ret ? ret : count;                                       \
 372 }
 373
 374 store_one(scaling_min_freq,min);
 375 store_one(scaling_max_freq,max);
 376
 377 /**
 378  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 379  */
 380 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 381 {
 382         unsigned int cur_freq = cpufreq_get(policy->cpu);
 383         if (!cur_freq)
 384                 return sprintf(buf, "<unknown>");
 385         return sprintf(buf, "%u\n", cur_freq);
 386 }
 387
 388
 389 /**
 390  * show_scaling_governor - show the current policy for the specified CPU
 391  */
 392 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 393 {
 394         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 395                 return sprintf(buf, "powersave\n");
 396         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 397                 return sprintf(buf, "performance\n");
 398         else if (policy->governor)
 399                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 400         return -EINVAL;
 401 }
 402
 403
 404 /**
 405  * store_scaling_governor - store policy for the specified CPU
 406  */
 407 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 408                                        const char *buf, size_t count)
 409 {
 410         unsigned int ret = -EINVAL;
 411         char    str_governor[16];
 412         struct cpufreq_policy new_policy;
 413
 414         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 415         if (ret)
 416                 return ret;
 417
 418         ret = sscanf (buf, "%15s", str_governor);
 419         if (ret != 1)
 420                 return -EINVAL;
 421
 422         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 423                 return -EINVAL;
 424
 425         ret = cpufreq_set_policy(&new_policy);
 426         return ret ? ret : count;
 427 }
 428
 429 /**
 430  * show_scaling_driver - show the cpufreq driver currently loaded
 431  */
 432 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 433 {
 434         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 435 }
 436
 437 /**
 438  * show_scaling_available_governors - show the available CPUfreq governors
 439  */
 440 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 441                                 char *buf)
 442 {
 443         ssize_t i = 0;
 444         struct cpufreq_governor *t;
 445
 446         if (!cpufreq_driver->target) {
 447                 i += sprintf(buf, "performance powersave");
 448                 goto out;
 449         }
 450
 451         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 452                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 453                         goto out;
 454                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 455         }
 456 out:
 457         i += sprintf(&buf[i], "\n");
 458         return i;
 459 }
 460 /**
 461  * show_affected_cpus - show the CPUs affected by each transition
 462  */
 463 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 464 {
 465         ssize_t i = 0;
 466         unsigned int cpu;
 467
 468         for_each_cpu_mask(cpu, policy->cpus) {
 469                 if (i)
 470                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 471                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 472                 if (i >= (PAGE_SIZE - 5))
 473                     break;
 474         }
 475         i += sprintf(&buf[i], "\n");
 476         return i;
 477 }
 478
 479
 480 #define define_one_ro(_name) \
 481 static struct freq_attr _name = \
 482 __ATTR(_name, 0444, show_##_name, NULL)
 483
 484 #define define_one_ro0400(_name) \
 485 static struct freq_attr _name = \
 486 __ATTR(_name, 0400, show_##_name, NULL)
 487
 488 #define define_one_rw(_name) \
 489 static struct freq_attr _name = \
 490 __ATTR(_name, 0644, show_##_name, store_##_name)
 491
 492 define_one_ro0400(cpuinfo_cur_freq);
 493 define_one_ro(cpuinfo_min_freq);
 494 define_one_ro(cpuinfo_max_freq);
 495 define_one_ro(scaling_available_governors);
 496 define_one_ro(scaling_driver);
 497 define_one_ro(scaling_cur_freq);
 498 define_one_ro(affected_cpus);
 499 define_one_rw(scaling_min_freq);
 500 define_one_rw(scaling_max_freq);
 501 define_one_rw(scaling_governor);
 502
 503 static struct attribute * default_attrs[] = {
 504         &cpuinfo_min_freq.attr,
 505         &cpuinfo_max_freq.attr,
 506         &scaling_min_freq.attr,
 507         &scaling_max_freq.attr,
 508         &affected_cpus.attr,
 509         &scaling_governor.attr,
 510         &scaling_driver.attr,
 511         &scaling_available_governors.attr,
 512         NULL
 513 };
 514
 515 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 516 #define to_attr(a) container_of(a,struct freq_attr,attr)
 517
 518 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 519 {
 520         struct cpufreq_policy * policy = to_policy(kobj);
 521         struct freq_attr * fattr = to_attr(attr);
 522         ssize_t ret;
 523         policy = cpufreq_cpu_get(policy->cpu);
 524         if (!policy)
 525                 return -EINVAL;
 526         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 527         cpufreq_cpu_put(policy);
 528         return ret;
 529 }
 530
 531 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 532                      const char * buf, size_t count)
 533 {
 534         struct cpufreq_policy * policy = to_policy(kobj);
 535         struct freq_attr * fattr = to_attr(attr);
 536         ssize_t ret;
 537         policy = cpufreq_cpu_get(policy->cpu);
 538         if (!policy)
 539                 return -EINVAL;
 540         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 541         cpufreq_cpu_put(policy);
 542         return ret;
 543 }
 544
 545 static void cpufreq_sysfs_release(struct kobject * kobj)
 546 {
 547         struct cpufreq_policy * policy = to_policy(kobj);
 548         dprintk("last reference is dropped\n");
 549         complete(&policy->kobj_unregister);
 550 }
 551
 552 static struct sysfs_ops sysfs_ops = {
 553         .show   = show,
 554         .store  = store,
 555 };
 556
 557 static struct kobj_type ktype_cpufreq = {
 558         .sysfs_ops      = &sysfs_ops,
 559         .default_attrs  = default_attrs,
 560         .release        = cpufreq_sysfs_release,
 561 };
 562
 563
 564 /**
 565  * cpufreq_add_dev - add a CPU device
 566  *
 567  * Adds the cpufreq interface for a CPU device.
 568  */
 569 static int cpufreq_add_dev (struct sys_device * sys_dev)
 570 {
 571         unsigned int cpu = sys_dev->id;
 572         int ret = 0;
 573         struct cpufreq_policy new_policy;
 574         struct cpufreq_policy *policy;
 575         struct freq_attr **drv_attr;
 576         unsigned long flags;
 577         unsigned int j;
 578
 579         if (cpu_is_offline(cpu))
 580                 return 0;
 581
 582         cpufreq_debug_disable_ratelimit();
 583         dprintk("adding CPU %u\n", cpu);
 584
 585 #ifdef CONFIG_SMP
 586         /* check whether a different CPU already registered this
 587          * CPU because it is in the same boat. */
 588         policy = cpufreq_cpu_get(cpu);
 589         if (unlikely(policy)) {
 590                 dprintk("CPU already managed, adding link\n");
 591                 sysfs_create_link(&sys_dev->kobj, &policy->kobj, "cpufreq");
 592                 cpufreq_debug_enable_ratelimit();
 593                 return 0;
 594         }
 595 #endif
 596
 597         if (!try_module_get(cpufreq_driver->owner)) {
 598                 ret = -EINVAL;
 599                 goto module_out;
 600         }
 601
 602         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 603         if (!policy) {
 604                 ret = -ENOMEM;
 605                 goto nomem_out;
 606         }
 607
 608         policy->cpu = cpu;
 609         policy->cpus = cpumask_of_cpu(cpu);
 610
 611         mutex_init(&policy->lock);
 612         mutex_lock(&policy->lock);
 613         init_completion(&policy->kobj_unregister);
 614         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 615
 616         /* call driver. From then on the cpufreq must be able
 617          * to accept all calls to ->verify and ->setpolicy for this CPU
 618          */
 619         ret = cpufreq_driver->init(policy);
 620         if (ret) {
 621                 dprintk("initialization failed\n");
 622                 mutex_unlock(&policy->lock);
 623                 goto err_out;
 624         }
 625
 626         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 627
 628         /* prepare interface data */
 629         policy->kobj.parent = &sys_dev->kobj;
 630         policy->kobj.ktype = &ktype_cpufreq;
 631         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 632
 633         ret = kobject_register(&policy->kobj);
 634         if (ret) {
 635                 mutex_unlock(&policy->lock);
 636                 goto err_out_driver_exit;
 637         }
 638         /* set up files for this cpu device */
 639         drv_attr = cpufreq_driver->attr;
 640         while ((drv_attr) && (*drv_attr)) {
 641                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 642                 drv_attr++;
 643         }
 644         if (cpufreq_driver->get)
 645                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 646         if (cpufreq_driver->target)
 647                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 648
 649         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 650         for_each_cpu_mask(j, policy->cpus)
 651                 cpufreq_cpu_data[j] = policy;
 652         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 653         policy->governor = NULL; /* to assure that the starting sequence is
 654                                   * run in cpufreq_set_policy */
 655         mutex_unlock(&policy->lock);
 656
 657         /* set default policy */
 658
 659         ret = cpufreq_set_policy(&new_policy);
 660         if (ret) {
 661                 dprintk("setting policy failed\n");
 662                 goto err_out_unregister;
 663         }
 664
 665         module_put(cpufreq_driver->owner);
 666         dprintk("initialization complete\n");
 667         cpufreq_debug_enable_ratelimit();
 668
 669         return 0;
 670
 671
 672 err_out_unregister:
 673         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 674         for_each_cpu_mask(j, policy->cpus)
 675                 cpufreq_cpu_data[j] = NULL;
 676         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 677
 678         kobject_unregister(&policy->kobj);
 679         wait_for_completion(&policy->kobj_unregister);
 680
 681 err_out_driver_exit:
 682         if (cpufreq_driver->exit)
 683                 cpufreq_driver->exit(policy);
 684
 685 err_out:
 686         kfree(policy);
 687
 688 nomem_out:
 689         module_put(cpufreq_driver->owner);
 690 module_out:
 691         cpufreq_debug_enable_ratelimit();
 692         return ret;
 693 }
 694
 695
 696 /**
 697  * cpufreq_remove_dev - remove a CPU device
 698  *
 699  * Removes the cpufreq interface for a CPU device.
 700  */
 701 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 702 {
 703         unsigned int cpu = sys_dev->id;
 704         unsigned long flags;
 705         struct cpufreq_policy *data;
 706 #ifdef CONFIG_SMP
 707         struct sys_device *cpu_sys_dev;
 708         unsigned int j;
 709 #endif
 710
 711         cpufreq_debug_disable_ratelimit();
 712         dprintk("unregistering CPU %u\n", cpu);
 713
 714         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 715         data = cpufreq_cpu_data[cpu];
 716
 717         if (!data) {
 718                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 719                 cpufreq_debug_enable_ratelimit();
 720                 return -EINVAL;
 721         }
 722         cpufreq_cpu_data[cpu] = NULL;
 723
 724
 725 #ifdef CONFIG_SMP
 726         /* if this isn't the CPU which is the parent of the kobj, we
 727          * only need to unlink, put and exit
 728          */
 729         if (unlikely(cpu != data->cpu)) {
 730                 dprintk("removing link\n");
 731                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 732                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 733                 cpufreq_cpu_put(data);
 734                 cpufreq_debug_enable_ratelimit();
 735                 return 0;
 736         }
 737 #endif
 738
 739
 740         if (!kobject_get(&data->kobj)) {
 741                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 742                 cpufreq_debug_enable_ratelimit();
 743                 return -EFAULT;
 744         }
 745
 746 #ifdef CONFIG_SMP
 747         /* if we have other CPUs still registered, we need to unlink them,
 748          * or else wait_for_completion below will lock up. Clean the
 749          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 750          * links afterwards.
 751          */
 752         if (unlikely(cpus_weight(data->cpus) > 1)) {
 753                 for_each_cpu_mask(j, data->cpus) {
 754                         if (j == cpu)
 755                                 continue;
 756                         cpufreq_cpu_data[j] = NULL;
 757                 }
 758         }
 759
 760         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 761
 762         if (unlikely(cpus_weight(data->cpus) > 1)) {
 763                 for_each_cpu_mask(j, data->cpus) {
 764                         if (j == cpu)
 765                                 continue;
 766                         dprintk("removing link for cpu %u\n", j);
 767                         cpu_sys_dev = get_cpu_sysdev(j);
 768                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 769                         cpufreq_cpu_put(data);
 770                 }
 771         }
 772 #else
 773         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 774 #endif
 775
 776         mutex_lock(&data->lock);
 777         if (cpufreq_driver->target)
 778                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 779         mutex_unlock(&data->lock);
 780
 781         kobject_unregister(&data->kobj);
 782
 783         kobject_put(&data->kobj);
 784
 785         /* we need to make sure that the underlying kobj is actually
 786          * not referenced anymore by anybody before we proceed with
 787          * unloading.
 788          */
 789         dprintk("waiting for dropping of refcount\n");
 790         wait_for_completion(&data->kobj_unregister);
 791         dprintk("wait complete\n");
 792
 793         if (cpufreq_driver->exit)
 794                 cpufreq_driver->exit(data);
 795
 796         kfree(data);
 797
 798         cpufreq_debug_enable_ratelimit();
 799         return 0;
 800 }
 801
 802
 803 static void handle_update(void *data)
 804 {
 805         unsigned int cpu = (unsigned int)(long)data;
 806         dprintk("handle_update for cpu %u called\n", cpu);
 807         cpufreq_update_policy(cpu);
 808 }
 809
 810 /**
 811  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 812  *      @cpu: cpu number
 813  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 814  *      @new_freq: CPU frequency the CPU actually runs at
 815  *
 816  *      We adjust to current frequency first, and need to clean up later. So either call
 817  *      to cpufreq_update_policy() or schedule handle_update()).
 818  */
 819 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 820 {
 821         struct cpufreq_freqs freqs;
 822
 823         dprintk(KERN_WARNING "Warning: CPU frequency out of sync: cpufreq and timing "
 824                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 825
 826         freqs.cpu = cpu;
 827         freqs.old = old_freq;
 828         freqs.new = new_freq;
 829         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 830         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 831 }
 832
 833
 834 /**
 835  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
 836  * @cpu: CPU number
 837  *
 838  * This is the last known freq, without actually getting it from the driver.
 839  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 840  */
 841 unsigned int cpufreq_quick_get(unsigned int cpu)
 842 {
 843         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 844         unsigned int ret = 0;
 845
 846         if (policy) {
 847                 mutex_lock(&policy->lock);
 848                 ret = policy->cur;
 849                 mutex_unlock(&policy->lock);
 850                 cpufreq_cpu_put(policy);
 851         }
 852
 853         return (ret);
 854 }
 855 EXPORT_SYMBOL(cpufreq_quick_get);
 856
 857
 858 /**
 859  * cpufreq_get - get the current CPU frequency (in kHz)
 860  * @cpu: CPU number
 861  *
 862  * Get the CPU current (static) CPU frequency
 863  */
 864 unsigned int cpufreq_get(unsigned int cpu)
 865 {
 866         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 867         unsigned int ret = 0;
 868
 869         if (!policy)
 870                 return 0;
 871
 872         if (!cpufreq_driver->get)
 873                 goto out;
 874
 875         mutex_lock(&policy->lock);
 876
 877         ret = cpufreq_driver->get(cpu);
 878
 879         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 880                 /* verify no discrepancy between actual and saved value exists */
 881                 if (unlikely(ret != policy->cur)) {
 882                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 883                         schedule_work(&policy->update);
 884                 }
 885         }
 886
 887         mutex_unlock(&policy->lock);
 888
 889 out:
 890         cpufreq_cpu_put(policy);
 891
 892         return (ret);
 893 }
 894 EXPORT_SYMBOL(cpufreq_get);
 895
 896
 897 /**
 898  *      cpufreq_suspend - let the low level driver prepare for suspend
 899  */
 900
 901 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 902 {
 903         int cpu = sysdev->id;
 904         unsigned int ret = 0;
 905         unsigned int cur_freq = 0;
 906         struct cpufreq_policy *cpu_policy;
 907
 908         dprintk("resuming cpu %u\n", cpu);
 909
 910         if (!cpu_online(cpu))
 911                 return 0;
 912
 913         /* we may be lax here as interrupts are off. Nonetheless
 914          * we need to grab the correct cpu policy, as to check
 915          * whether we really run on this CPU.
 916          */
 917
 918         cpu_policy = cpufreq_cpu_get(cpu);
 919         if (!cpu_policy)
 920                 return -EINVAL;
 921
 922         /* only handle each CPU group once */
 923         if (unlikely(cpu_policy->cpu != cpu)) {
 924                 cpufreq_cpu_put(cpu_policy);
 925                 return 0;
 926         }
 927
 928         if (cpufreq_driver->suspend) {
 929                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
 930                 if (ret) {
 931                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
 932                                         "step on CPU %u\n", cpu_policy->cpu);
 933                         cpufreq_cpu_put(cpu_policy);
 934                         return ret;
 935                 }
 936         }
 937
 938
 939         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
 940                 goto out;
 941
 942         if (cpufreq_driver->get)
 943                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
 944
 945         if (!cur_freq || !cpu_policy->cur) {
 946                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
 947                        "frequency is what timing core thinks it is.\n");
 948                 goto out;
 949         }
 950
 951         if (unlikely(cur_freq != cpu_policy->cur)) {
 952                 struct cpufreq_freqs freqs;
 953
 954                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
 955                         dprintk(KERN_DEBUG "Warning: CPU frequency is %u, "
 956                                "cpufreq assumed %u kHz.\n",
 957                                cur_freq, cpu_policy->cur);
 958
 959                 freqs.cpu = cpu;
 960                 freqs.old = cpu_policy->cur;
 961                 freqs.new = cur_freq;
 962
 963                 notifier_call_chain(&cpufreq_transition_notifier_list,
 964                                     CPUFREQ_SUSPENDCHANGE, &freqs);
 965                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
 966
 967                 cpu_policy->cur = cur_freq;
 968         }
 969
 970 out:
 971         cpufreq_cpu_put(cpu_policy);
 972         return 0;
 973 }
 974
 975 /**
 976  *      cpufreq_resume -  restore proper CPU frequency handling after resume
 977  *
 978  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
 979  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
 980  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
 981  *          restored.
 982  */
 983 static int cpufreq_resume(struct sys_device * sysdev)
 984 {
 985         int cpu = sysdev->id;
 986         unsigned int ret = 0;
 987         struct cpufreq_policy *cpu_policy;
 988
 989         dprintk("resuming cpu %u\n", cpu);
 990
 991         if (!cpu_online(cpu))
 992                 return 0;
 993
 994         /* we may be lax here as interrupts are off. Nonetheless
 995          * we need to grab the correct cpu policy, as to check
 996          * whether we really run on this CPU.
 997          */
 998
 999         cpu_policy = cpufreq_cpu_get(cpu);
1000         if (!cpu_policy)
1001                 return -EINVAL;
1002
1003         /* only handle each CPU group once */
1004         if (unlikely(cpu_policy->cpu != cpu)) {
1005                 cpufreq_cpu_put(cpu_policy);
1006                 return 0;
1007         }
1008
1009         if (cpufreq_driver->resume) {
1010                 ret = cpufreq_driver->resume(cpu_policy);
1011                 if (ret) {
1012                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1013                                         "step on CPU %u\n", cpu_policy->cpu);
1014                         cpufreq_cpu_put(cpu_policy);
1015                         return ret;
1016                 }
1017         }
1018
1019         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1020                 unsigned int cur_freq = 0;
1021
1022                 if (cpufreq_driver->get)
1023                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1024
1025                 if (!cur_freq || !cpu_policy->cur) {
1026                         printk(KERN_ERR "cpufreq: resume failed to assert "
1027                                         "current frequency is what timing core "
1028                                         "thinks it is.\n");
1029                         goto out;
1030                 }
1031
1032                 if (unlikely(cur_freq != cpu_policy->cur)) {
1033                         struct cpufreq_freqs freqs;
1034
1035                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1036                                 dprintk(KERN_WARNING "Warning: CPU frequency"
1037                                        "is %u, cpufreq assumed %u kHz.\n",
1038                                        cur_freq, cpu_policy->cur);
1039
1040                         freqs.cpu = cpu;
1041                         freqs.old = cpu_policy->cur;
1042                         freqs.new = cur_freq;
1043
1044                         notifier_call_chain(&cpufreq_transition_notifier_list,
1045                                         CPUFREQ_RESUMECHANGE, &freqs);
1046                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1047
1048                         cpu_policy->cur = cur_freq;
1049                 }
1050         }
1051
1052 out:
1053         schedule_work(&cpu_policy->update);
1054         cpufreq_cpu_put(cpu_policy);
1055         return ret;
1056 }
1057
1058 static struct sysdev_driver cpufreq_sysdev_driver = {
1059         .add            = cpufreq_add_dev,
1060         .remove         = cpufreq_remove_dev,
1061         .suspend        = cpufreq_suspend,
1062         .resume         = cpufreq_resume,
1063 };
1064
1065
1066 /*********************************************************************
1067  *                     NOTIFIER LISTS INTERFACE                      *
1068  *********************************************************************/
1069
1070 /**
1071  *      cpufreq_register_notifier - register a driver with cpufreq
1072  *      @nb: notifier function to register
1073  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1074  *
1075  *      Add a driver to one of two lists: either a list of drivers that
1076  *      are notified about clock rate changes (once before and once after
1077  *      the transition), or a list of drivers that are notified about
1078  *      changes in cpufreq policy.
1079  *
1080  *      This function may sleep, and has the same return conditions as
1081  *      notifier_chain_register.
1082  */
1083 int cpufreq_register_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_register(&cpufreq_transition_notifier_list, nb);
1091                 break;
1092         case CPUFREQ_POLICY_NOTIFIER:
1093                 ret = notifier_chain_register(&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_register_notifier);
1103
1104
1105 /**
1106  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1107  *      @nb: notifier block to be unregistered
1108  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1109  *
1110  *      Remove a driver from the CPU frequency notifier list.
1111  *
1112  *      This function may sleep, and has the same return conditions as
1113  *      notifier_chain_unregister.
1114  */
1115 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1116 {
1117         int ret;
1118
1119         down_write(&cpufreq_notifier_rwsem);
1120         switch (list) {
1121         case CPUFREQ_TRANSITION_NOTIFIER:
1122                 ret = notifier_chain_unregister(&cpufreq_transition_notifier_list, nb);
1123                 break;
1124         case CPUFREQ_POLICY_NOTIFIER:
1125                 ret = notifier_chain_unregister(&cpufreq_policy_notifier_list, nb);
1126                 break;
1127         default:
1128                 ret = -EINVAL;
1129         }
1130         up_write(&cpufreq_notifier_rwsem);
1131
1132         return ret;
1133 }
1134 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1135
1136
1137 /*********************************************************************
1138  *                              GOVERNORS                            *
1139  *********************************************************************/
1140
1141
1142 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1143                             unsigned int target_freq,
1144                             unsigned int relation)
1145 {
1146         int retval = -EINVAL;
1147
1148         lock_cpu_hotplug();
1149         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1150                 target_freq, relation);
1151         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1152                 retval = cpufreq_driver->target(policy, target_freq, relation);
1153
1154         unlock_cpu_hotplug();
1155
1156         return retval;
1157 }
1158 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1159
1160 int cpufreq_driver_target(struct cpufreq_policy *policy,
1161                           unsigned int target_freq,
1162                           unsigned int relation)
1163 {
1164         int ret;
1165
1166         policy = cpufreq_cpu_get(policy->cpu);
1167         if (!policy)
1168                 return -EINVAL;
1169
1170         mutex_lock(&policy->lock);
1171
1172         ret = __cpufreq_driver_target(policy, target_freq, relation);
1173
1174         mutex_unlock(&policy->lock);
1175
1176         cpufreq_cpu_put(policy);
1177         return ret;
1178 }
1179 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1180
1181
1182 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1183 {
1184         int ret;
1185
1186         if (!try_module_get(policy->governor->owner))
1187                 return -EINVAL;
1188
1189         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1190         ret = policy->governor->governor(policy, event);
1191
1192         /* we keep one module reference alive for each CPU governed by this CPU */
1193         if ((event != CPUFREQ_GOV_START) || ret)
1194                 module_put(policy->governor->owner);
1195         if ((event == CPUFREQ_GOV_STOP) && !ret)
1196                 module_put(policy->governor->owner);
1197
1198         return ret;
1199 }
1200
1201
1202 int cpufreq_governor(unsigned int cpu, unsigned int event)
1203 {
1204         int ret = 0;
1205         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1206
1207         if (!policy)
1208                 return -EINVAL;
1209
1210         mutex_lock(&policy->lock);
1211         ret = __cpufreq_governor(policy, event);
1212         mutex_unlock(&policy->lock);
1213
1214         cpufreq_cpu_put(policy);
1215         return ret;
1216 }
1217 EXPORT_SYMBOL_GPL(cpufreq_governor);
1218
1219
1220 int cpufreq_register_governor(struct cpufreq_governor *governor)
1221 {
1222         struct cpufreq_governor *t;
1223
1224         if (!governor)
1225                 return -EINVAL;
1226
1227         mutex_lock(&cpufreq_governor_mutex);
1228
1229         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1230                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1231                         mutex_unlock(&cpufreq_governor_mutex);
1232                         return -EBUSY;
1233                 }
1234         }
1235         list_add(&governor->governor_list, &cpufreq_governor_list);
1236
1237         mutex_unlock(&cpufreq_governor_mutex);
1238         return 0;
1239 }
1240 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1241
1242
1243 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1244 {
1245         if (!governor)
1246                 return;
1247
1248         mutex_lock(&cpufreq_governor_mutex);
1249         list_del(&governor->governor_list);
1250         mutex_unlock(&cpufreq_governor_mutex);
1251         return;
1252 }
1253 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1254
1255
1256
1257 /*********************************************************************
1258  *                          POLICY INTERFACE                         *
1259  *********************************************************************/
1260
1261 /**
1262  * cpufreq_get_policy - get the current cpufreq_policy
1263  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1264  *
1265  * Reads the current cpufreq policy.
1266  */
1267 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1268 {
1269         struct cpufreq_policy *cpu_policy;
1270         if (!policy)
1271                 return -EINVAL;
1272
1273         cpu_policy = cpufreq_cpu_get(cpu);
1274         if (!cpu_policy)
1275                 return -EINVAL;
1276
1277         mutex_lock(&cpu_policy->lock);
1278         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1279         mutex_unlock(&cpu_policy->lock);
1280
1281         cpufreq_cpu_put(cpu_policy);
1282         return 0;
1283 }
1284 EXPORT_SYMBOL(cpufreq_get_policy);
1285
1286
1287 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1288 {
1289         int ret = 0;
1290
1291         cpufreq_debug_disable_ratelimit();
1292         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1293                 policy->min, policy->max);
1294
1295         memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
1296
1297         /* verify the cpu speed can be set within this limit */
1298         ret = cpufreq_driver->verify(policy);
1299         if (ret)
1300                 goto error_out;
1301
1302         down_read(&cpufreq_notifier_rwsem);
1303
1304         /* adjust if necessary - all reasons */
1305         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_ADJUST,
1306                             policy);
1307
1308         /* adjust if necessary - hardware incompatibility*/
1309         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_INCOMPATIBLE,
1310                             policy);
1311
1312         /* verify the cpu speed can be set within this limit,
1313            which might be different to the first one */
1314         ret = cpufreq_driver->verify(policy);
1315         if (ret) {
1316                 up_read(&cpufreq_notifier_rwsem);
1317                 goto error_out;
1318         }
1319
1320         /* notification of the new policy */
1321         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_NOTIFY,
1322                             policy);
1323
1324         up_read(&cpufreq_notifier_rwsem);
1325
1326         data->min = policy->min;
1327         data->max = policy->max;
1328
1329         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1330
1331         if (cpufreq_driver->setpolicy) {
1332                 data->policy = policy->policy;
1333                 dprintk("setting range\n");
1334                 ret = cpufreq_driver->setpolicy(policy);
1335         } else {
1336                 if (policy->governor != data->governor) {
1337                         /* save old, working values */
1338                         struct cpufreq_governor *old_gov = data->governor;
1339
1340                         dprintk("governor switch\n");
1341
1342                         /* end old governor */
1343                         if (data->governor)
1344                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1345
1346                         /* start new governor */
1347                         data->governor = policy->governor;
1348                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1349                                 /* new governor failed, so re-start old one */
1350                                 dprintk("starting governor %s failed\n", data->governor->name);
1351                                 if (old_gov) {
1352                                         data->governor = old_gov;
1353                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1354                                 }
1355                                 ret = -EINVAL;
1356                                 goto error_out;
1357                         }
1358                         /* might be a policy change, too, so fall through */
1359                 }
1360                 dprintk("governor: change or update limits\n");
1361                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1362         }
1363
1364 error_out:
1365         cpufreq_debug_enable_ratelimit();
1366         return ret;
1367 }
1368
1369 /**
1370  *      cpufreq_set_policy - set a new CPUFreq policy
1371  *      @policy: policy to be set.
1372  *
1373  *      Sets a new CPU frequency and voltage scaling policy.
1374  */
1375 int cpufreq_set_policy(struct cpufreq_policy *policy)
1376 {
1377         int ret = 0;
1378         struct cpufreq_policy *data;
1379
1380         if (!policy)
1381                 return -EINVAL;
1382
1383         data = cpufreq_cpu_get(policy->cpu);
1384         if (!data)
1385                 return -EINVAL;
1386
1387         /* lock this CPU */
1388         mutex_lock(&data->lock);
1389
1390         ret = __cpufreq_set_policy(data, policy);
1391         data->user_policy.min = data->min;
1392         data->user_policy.max = data->max;
1393         data->user_policy.policy = data->policy;
1394         data->user_policy.governor = data->governor;
1395
1396         mutex_unlock(&data->lock);
1397         cpufreq_cpu_put(data);
1398
1399         return ret;
1400 }
1401 EXPORT_SYMBOL(cpufreq_set_policy);
1402
1403
1404 /**
1405  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1406  *      @cpu: CPU which shall be re-evaluated
1407  *
1408  *      Usefull for policy notifiers which have different necessities
1409  *      at different times.
1410  */
1411 int cpufreq_update_policy(unsigned int cpu)
1412 {
1413         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1414         struct cpufreq_policy policy;
1415         int ret = 0;
1416
1417         if (!data)
1418                 return -ENODEV;
1419
1420         mutex_lock(&data->lock);
1421
1422         dprintk("updating policy for CPU %u\n", cpu);
1423         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1424         policy.min = data->user_policy.min;
1425         policy.max = data->user_policy.max;
1426         policy.policy = data->user_policy.policy;
1427         policy.governor = data->user_policy.governor;
1428
1429         /* BIOS might change freq behind our back
1430           -> ask driver for current freq and notify governors about a change */
1431         if (cpufreq_driver->get) {
1432                 policy.cur = cpufreq_driver->get(cpu);
1433                 if (!data->cur) {
1434                         dprintk("Driver did not initialize current freq");
1435                         data->cur = policy.cur;
1436                 } else {
1437                         if (data->cur != policy.cur)
1438                                 cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1439                 }
1440         }
1441
1442         ret = __cpufreq_set_policy(data, &policy);
1443
1444         mutex_unlock(&data->lock);
1445
1446         cpufreq_cpu_put(data);
1447         return ret;
1448 }
1449 EXPORT_SYMBOL(cpufreq_update_policy);
1450
1451 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1452                                         unsigned long action, void *hcpu)
1453 {
1454         unsigned int cpu = (unsigned long)hcpu;
1455         struct cpufreq_policy *policy;
1456         struct sys_device *sys_dev;
1457
1458         sys_dev = get_cpu_sysdev(cpu);
1459
1460         if (sys_dev) {
1461                 switch (action) {
1462                 case CPU_ONLINE:
1463                         cpufreq_add_dev(sys_dev);
1464                         break;
1465                 case CPU_DOWN_PREPARE:
1466                         /*
1467                          * We attempt to put this cpu in lowest frequency
1468                          * possible before going down. This will permit
1469                          * hardware-managed P-State to switch other related
1470                          * threads to min or higher speeds if possible.
1471                          */
1472                         policy = cpufreq_cpu_data[cpu];
1473                         if (policy) {
1474                                 cpufreq_driver_target(policy, policy->min,
1475                                                 CPUFREQ_RELATION_H);
1476                         }
1477                         break;
1478                 case CPU_DEAD:
1479                         cpufreq_remove_dev(sys_dev);
1480                         break;
1481                 }
1482         }
1483         return NOTIFY_OK;
1484 }
1485
1486 static struct notifier_block cpufreq_cpu_notifier =
1487 {
1488     .notifier_call = cpufreq_cpu_callback,
1489 };
1490
1491 /*********************************************************************
1492  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1493  *********************************************************************/
1494
1495 /**
1496  * cpufreq_register_driver - register a CPU Frequency driver
1497  * @driver_data: A struct cpufreq_driver containing the values#
1498  * submitted by the CPU Frequency driver.
1499  *
1500  *   Registers a CPU Frequency driver to this core code. This code
1501  * returns zero on success, -EBUSY when another driver got here first
1502  * (and isn't unregistered in the meantime).
1503  *
1504  */
1505 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1506 {
1507         unsigned long flags;
1508         int ret;
1509
1510         if (!driver_data || !driver_data->verify || !driver_data->init ||
1511             ((!driver_data->setpolicy) && (!driver_data->target)))
1512                 return -EINVAL;
1513
1514         dprintk("trying to register driver %s\n", driver_data->name);
1515
1516         if (driver_data->setpolicy)
1517                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1518
1519         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1520         if (cpufreq_driver) {
1521                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1522                 return -EBUSY;
1523         }
1524         cpufreq_driver = driver_data;
1525         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1526
1527         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1528
1529         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1530                 int i;
1531                 ret = -ENODEV;
1532
1533                 /* check for at least one working CPU */
1534                 for (i=0; i<NR_CPUS; i++)
1535                         if (cpufreq_cpu_data[i])
1536                                 ret = 0;
1537
1538                 /* if all ->init() calls failed, unregister */
1539                 if (ret) {
1540                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1541                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1542
1543                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1544                         cpufreq_driver = NULL;
1545                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1546                 }
1547         }
1548
1549         if (!ret) {
1550                 register_cpu_notifier(&cpufreq_cpu_notifier);
1551                 dprintk("driver %s up and running\n", driver_data->name);
1552                 cpufreq_debug_enable_ratelimit();
1553         }
1554
1555         return (ret);
1556 }
1557 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1558
1559
1560 /**
1561  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1562  *
1563  *    Unregister the current CPUFreq driver. Only call this if you have
1564  * the right to do so, i.e. if you have succeeded in initialising before!
1565  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1566  * currently not initialised.
1567  */
1568 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1569 {
1570         unsigned long flags;
1571
1572         cpufreq_debug_disable_ratelimit();
1573
1574         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1575                 cpufreq_debug_enable_ratelimit();
1576                 return -EINVAL;
1577         }
1578
1579         dprintk("unregistering driver %s\n", driver->name);
1580
1581         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1582         unregister_cpu_notifier(&cpufreq_cpu_notifier);
1583
1584         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1585         cpufreq_driver = NULL;
1586         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1587
1588         return 0;
1589 }
1590 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);