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  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
  10  *      Fix handling for CPU hotplug -- affected CPUs
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17
  18 #include <linux/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/notifier.h>
  22 #include <linux/cpufreq.h>
  23 #include <linux/delay.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/device.h>
  27 #include <linux/slab.h>
  28 #include <linux/cpu.h>
  29 #include <linux/completion.h>
  30 #include <linux/mutex.h>
  31
  32 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
  33
  34 /**
  35  * The "cpufreq driver" - the arch- or hardware-dependend low
  36  * level driver of CPUFreq support, and its spinlock. This lock
  37  * also protects the cpufreq_cpu_data array.
  38  */
  39 static struct cpufreq_driver *cpufreq_driver;
  40 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
  41 static DEFINE_SPINLOCK(cpufreq_driver_lock);
  42
  43 /* internal prototypes */
  44 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
  45 static void handle_update(void *data);
  46
  47 /**
  48  * Two notifier lists: the "policy" list is involved in the
  49  * validation process for a new CPU frequency policy; the
  50  * "transition" list for kernel code that needs to handle
  51  * changes to devices when the CPU clock speed changes.
  52  * The mutex locks both lists.
  53  */
  54 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
  55 static BLOCKING_NOTIFIER_HEAD(cpufreq_transition_notifier_list);
  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         policy = cpufreq_cpu_data[freqs->cpu];
 249         switch (state) {
 250
 251         case CPUFREQ_PRECHANGE:
 252                 /* detect if the driver reported a value as "old frequency"
 253                  * which is not equal to what the cpufreq core thinks is
 254                  * "old frequency".
 255                  */
 256                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 257                         if ((policy) && (policy->cpu == freqs->cpu) &&
 258                             (policy->cur) && (policy->cur != freqs->old)) {
 259                                 dprintk("Warning: CPU frequency is"
 260                                         " %u, cpufreq assumed %u kHz.\n",
 261                                         freqs->old, policy->cur);
 262                                 freqs->old = policy->cur;
 263                         }
 264                 }
 265                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
 266                                 CPUFREQ_PRECHANGE, freqs);
 267                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 268                 break;
 269
 270         case CPUFREQ_POSTCHANGE:
 271                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 272                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
 273                                 CPUFREQ_POSTCHANGE, freqs);
 274                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 275                         policy->cur = freqs->new;
 276                 break;
 277         }
 278 }
 279 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 280
 281
 282
 283 /*********************************************************************
 284  *                          SYSFS INTERFACE                          *
 285  *********************************************************************/
 286
 287 /**
 288  * cpufreq_parse_governor - parse a governor string
 289  */
 290 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 291                                 struct cpufreq_governor **governor)
 292 {
 293         if (!cpufreq_driver)
 294                 return -EINVAL;
 295         if (cpufreq_driver->setpolicy) {
 296                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 297                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 298                         return 0;
 299                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 300                         *policy = CPUFREQ_POLICY_POWERSAVE;
 301                         return 0;
 302                 }
 303                 return -EINVAL;
 304         } else {
 305                 struct cpufreq_governor *t;
 306                 mutex_lock(&cpufreq_governor_mutex);
 307                 if (!cpufreq_driver || !cpufreq_driver->target)
 308                         goto out;
 309                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 310                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
 311                                 *governor = t;
 312                                 mutex_unlock(&cpufreq_governor_mutex);
 313                                 return 0;
 314                         }
 315                 }
 316 out:
 317                 mutex_unlock(&cpufreq_governor_mutex);
 318         }
 319         return -EINVAL;
 320 }
 321
 322
 323 /* drivers/base/cpu.c */
 324 extern struct sysdev_class cpu_sysdev_class;
 325
 326
 327 /**
 328  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 329  *
 330  * Write out information from cpufreq_driver->policy[cpu]; object must be
 331  * "unsigned int".
 332  */
 333
 334 #define show_one(file_name, object)                     \
 335 static ssize_t show_##file_name                         \
 336 (struct cpufreq_policy * policy, char *buf)             \
 337 {                                                       \
 338         return sprintf (buf, "%u\n", policy->object);   \
 339 }
 340
 341 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 342 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 343 show_one(scaling_min_freq, min);
 344 show_one(scaling_max_freq, max);
 345 show_one(scaling_cur_freq, cur);
 346
 347 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
 348
 349 /**
 350  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 351  */
 352 #define store_one(file_name, object)                    \
 353 static ssize_t store_##file_name                                        \
 354 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 355 {                                                                       \
 356         unsigned int ret = -EINVAL;                                     \
 357         struct cpufreq_policy new_policy;                               \
 358                                                                         \
 359         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 360         if (ret)                                                        \
 361                 return -EINVAL;                                         \
 362                                                                         \
 363         ret = sscanf (buf, "%u", &new_policy.object);                   \
 364         if (ret != 1)                                                   \
 365                 return -EINVAL;                                         \
 366                                                                         \
 367         mutex_lock(&policy->lock);                                      \
 368         ret = __cpufreq_set_policy(policy, &new_policy);                \
 369         policy->user_policy.object = policy->object;                    \
 370         mutex_unlock(&policy->lock);                                    \
 371                                                                         \
 372         return ret ? ret : count;                                       \
 373 }
 374
 375 store_one(scaling_min_freq,min);
 376 store_one(scaling_max_freq,max);
 377
 378 /**
 379  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 380  */
 381 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 382 {
 383         unsigned int cur_freq = cpufreq_get(policy->cpu);
 384         if (!cur_freq)
 385                 return sprintf(buf, "<unknown>");
 386         return sprintf(buf, "%u\n", cur_freq);
 387 }
 388
 389
 390 /**
 391  * show_scaling_governor - show the current policy for the specified CPU
 392  */
 393 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 394 {
 395         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 396                 return sprintf(buf, "powersave\n");
 397         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 398                 return sprintf(buf, "performance\n");
 399         else if (policy->governor)
 400                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 401         return -EINVAL;
 402 }
 403
 404
 405 /**
 406  * store_scaling_governor - store policy for the specified CPU
 407  */
 408 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 409                                        const char *buf, size_t count)
 410 {
 411         unsigned int ret = -EINVAL;
 412         char    str_governor[16];
 413         struct cpufreq_policy new_policy;
 414
 415         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 416         if (ret)
 417                 return ret;
 418
 419         ret = sscanf (buf, "%15s", str_governor);
 420         if (ret != 1)
 421                 return -EINVAL;
 422
 423         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 424                 return -EINVAL;
 425
 426         /* Do not use cpufreq_set_policy here or the user_policy.max
 427            will be wrongly overridden */
 428         mutex_lock(&policy->lock);
 429         ret = __cpufreq_set_policy(policy, &new_policy);
 430
 431         policy->user_policy.policy = policy->policy;
 432         policy->user_policy.governor = policy->governor;
 433         mutex_unlock(&policy->lock);
 434
 435         return ret ? ret : count;
 436 }
 437
 438 /**
 439  * show_scaling_driver - show the cpufreq driver currently loaded
 440  */
 441 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 442 {
 443         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 444 }
 445
 446 /**
 447  * show_scaling_available_governors - show the available CPUfreq governors
 448  */
 449 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 450                                 char *buf)
 451 {
 452         ssize_t i = 0;
 453         struct cpufreq_governor *t;
 454
 455         if (!cpufreq_driver->target) {
 456                 i += sprintf(buf, "performance powersave");
 457                 goto out;
 458         }
 459
 460         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 461                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 462                         goto out;
 463                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 464         }
 465 out:
 466         i += sprintf(&buf[i], "\n");
 467         return i;
 468 }
 469 /**
 470  * show_affected_cpus - show the CPUs affected by each transition
 471  */
 472 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 473 {
 474         ssize_t i = 0;
 475         unsigned int cpu;
 476
 477         for_each_cpu_mask(cpu, policy->cpus) {
 478                 if (i)
 479                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 480                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 481                 if (i >= (PAGE_SIZE - 5))
 482                     break;
 483         }
 484         i += sprintf(&buf[i], "\n");
 485         return i;
 486 }
 487
 488
 489 #define define_one_ro(_name) \
 490 static struct freq_attr _name = \
 491 __ATTR(_name, 0444, show_##_name, NULL)
 492
 493 #define define_one_ro0400(_name) \
 494 static struct freq_attr _name = \
 495 __ATTR(_name, 0400, show_##_name, NULL)
 496
 497 #define define_one_rw(_name) \
 498 static struct freq_attr _name = \
 499 __ATTR(_name, 0644, show_##_name, store_##_name)
 500
 501 define_one_ro0400(cpuinfo_cur_freq);
 502 define_one_ro(cpuinfo_min_freq);
 503 define_one_ro(cpuinfo_max_freq);
 504 define_one_ro(scaling_available_governors);
 505 define_one_ro(scaling_driver);
 506 define_one_ro(scaling_cur_freq);
 507 define_one_ro(affected_cpus);
 508 define_one_rw(scaling_min_freq);
 509 define_one_rw(scaling_max_freq);
 510 define_one_rw(scaling_governor);
 511
 512 static struct attribute * default_attrs[] = {
 513         &cpuinfo_min_freq.attr,
 514         &cpuinfo_max_freq.attr,
 515         &scaling_min_freq.attr,
 516         &scaling_max_freq.attr,
 517         &affected_cpus.attr,
 518         &scaling_governor.attr,
 519         &scaling_driver.attr,
 520         &scaling_available_governors.attr,
 521         NULL
 522 };
 523
 524 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 525 #define to_attr(a) container_of(a,struct freq_attr,attr)
 526
 527 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 528 {
 529         struct cpufreq_policy * policy = to_policy(kobj);
 530         struct freq_attr * fattr = to_attr(attr);
 531         ssize_t ret;
 532         policy = cpufreq_cpu_get(policy->cpu);
 533         if (!policy)
 534                 return -EINVAL;
 535         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 536         cpufreq_cpu_put(policy);
 537         return ret;
 538 }
 539
 540 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 541                      const char * buf, size_t count)
 542 {
 543         struct cpufreq_policy * policy = to_policy(kobj);
 544         struct freq_attr * fattr = to_attr(attr);
 545         ssize_t ret;
 546         policy = cpufreq_cpu_get(policy->cpu);
 547         if (!policy)
 548                 return -EINVAL;
 549         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 550         cpufreq_cpu_put(policy);
 551         return ret;
 552 }
 553
 554 static void cpufreq_sysfs_release(struct kobject * kobj)
 555 {
 556         struct cpufreq_policy * policy = to_policy(kobj);
 557         dprintk("last reference is dropped\n");
 558         complete(&policy->kobj_unregister);
 559 }
 560
 561 static struct sysfs_ops sysfs_ops = {
 562         .show   = show,
 563         .store  = store,
 564 };
 565
 566 static struct kobj_type ktype_cpufreq = {
 567         .sysfs_ops      = &sysfs_ops,
 568         .default_attrs  = default_attrs,
 569         .release        = cpufreq_sysfs_release,
 570 };
 571
 572
 573 /**
 574  * cpufreq_add_dev - add a CPU device
 575  *
 576  * Adds the cpufreq interface for a CPU device.
 577  */
 578 static int cpufreq_add_dev (struct sys_device * sys_dev)
 579 {
 580         unsigned int cpu = sys_dev->id;
 581         int ret = 0;
 582         struct cpufreq_policy new_policy;
 583         struct cpufreq_policy *policy;
 584         struct freq_attr **drv_attr;
 585         struct sys_device *cpu_sys_dev;
 586         unsigned long flags;
 587         unsigned int j;
 588 #ifdef CONFIG_SMP
 589         struct cpufreq_policy *managed_policy;
 590 #endif
 591
 592         if (cpu_is_offline(cpu))
 593                 return 0;
 594
 595         cpufreq_debug_disable_ratelimit();
 596         dprintk("adding CPU %u\n", cpu);
 597
 598 #ifdef CONFIG_SMP
 599         /* check whether a different CPU already registered this
 600          * CPU because it is in the same boat. */
 601         policy = cpufreq_cpu_get(cpu);
 602         if (unlikely(policy)) {
 603                 cpufreq_cpu_put(policy);
 604                 cpufreq_debug_enable_ratelimit();
 605                 return 0;
 606         }
 607 #endif
 608
 609         if (!try_module_get(cpufreq_driver->owner)) {
 610                 ret = -EINVAL;
 611                 goto module_out;
 612         }
 613
 614         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 615         if (!policy) {
 616                 ret = -ENOMEM;
 617                 goto nomem_out;
 618         }
 619
 620         policy->cpu = cpu;
 621         policy->cpus = cpumask_of_cpu(cpu);
 622
 623         mutex_init(&policy->lock);
 624         mutex_lock(&policy->lock);
 625         init_completion(&policy->kobj_unregister);
 626         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 627
 628         /* call driver. From then on the cpufreq must be able
 629          * to accept all calls to ->verify and ->setpolicy for this CPU
 630          */
 631         ret = cpufreq_driver->init(policy);
 632         if (ret) {
 633                 dprintk("initialization failed\n");
 634                 mutex_unlock(&policy->lock);
 635                 goto err_out;
 636         }
 637
 638 #ifdef CONFIG_SMP
 639         for_each_cpu_mask(j, policy->cpus) {
 640                 if (cpu == j)
 641                         continue;
 642
 643                 /* check for existing affected CPUs.  They may not be aware
 644                  * of it due to CPU Hotplug.
 645                  */
 646                 managed_policy = cpufreq_cpu_get(j);
 647                 if (unlikely(managed_policy)) {
 648                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 649                         managed_policy->cpus = policy->cpus;
 650                         cpufreq_cpu_data[cpu] = managed_policy;
 651                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 652
 653                         dprintk("CPU already managed, adding link\n");
 654                         sysfs_create_link(&sys_dev->kobj,
 655                                           &managed_policy->kobj, "cpufreq");
 656
 657                         cpufreq_debug_enable_ratelimit();
 658                         mutex_unlock(&policy->lock);
 659                         ret = 0;
 660                         goto err_out_driver_exit; /* call driver->exit() */
 661                 }
 662         }
 663 #endif
 664         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 665
 666         /* prepare interface data */
 667         policy->kobj.parent = &sys_dev->kobj;
 668         policy->kobj.ktype = &ktype_cpufreq;
 669         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 670
 671         ret = kobject_register(&policy->kobj);
 672         if (ret) {
 673                 mutex_unlock(&policy->lock);
 674                 goto err_out_driver_exit;
 675         }
 676         /* set up files for this cpu device */
 677         drv_attr = cpufreq_driver->attr;
 678         while ((drv_attr) && (*drv_attr)) {
 679                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 680                 drv_attr++;
 681         }
 682         if (cpufreq_driver->get)
 683                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 684         if (cpufreq_driver->target)
 685                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 686
 687         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 688         for_each_cpu_mask(j, policy->cpus)
 689                 cpufreq_cpu_data[j] = policy;
 690         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 691
 692         /* symlink affected CPUs */
 693         for_each_cpu_mask(j, policy->cpus) {
 694                 if (j == cpu)
 695                         continue;
 696                 if (!cpu_online(j))
 697                         continue;
 698
 699                 dprintk("CPU %u already managed, adding link\n", j);
 700                 cpufreq_cpu_get(cpu);
 701                 cpu_sys_dev = get_cpu_sysdev(j);
 702                 sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 703                                   "cpufreq");
 704         }
 705
 706         policy->governor = NULL; /* to assure that the starting sequence is
 707                                   * run in cpufreq_set_policy */
 708         mutex_unlock(&policy->lock);
 709
 710         /* set default policy */
 711         ret = cpufreq_set_policy(&new_policy);
 712         if (ret) {
 713                 dprintk("setting policy failed\n");
 714                 goto err_out_unregister;
 715         }
 716
 717         module_put(cpufreq_driver->owner);
 718         dprintk("initialization complete\n");
 719         cpufreq_debug_enable_ratelimit();
 720
 721         return 0;
 722
 723
 724 err_out_unregister:
 725         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 726         for_each_cpu_mask(j, policy->cpus)
 727                 cpufreq_cpu_data[j] = NULL;
 728         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 729
 730         kobject_unregister(&policy->kobj);
 731         wait_for_completion(&policy->kobj_unregister);
 732
 733 err_out_driver_exit:
 734         if (cpufreq_driver->exit)
 735                 cpufreq_driver->exit(policy);
 736
 737 err_out:
 738         kfree(policy);
 739
 740 nomem_out:
 741         module_put(cpufreq_driver->owner);
 742 module_out:
 743         cpufreq_debug_enable_ratelimit();
 744         return ret;
 745 }
 746
 747
 748 /**
 749  * cpufreq_remove_dev - remove a CPU device
 750  *
 751  * Removes the cpufreq interface for a CPU device.
 752  */
 753 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 754 {
 755         unsigned int cpu = sys_dev->id;
 756         unsigned long flags;
 757         struct cpufreq_policy *data;
 758 #ifdef CONFIG_SMP
 759         struct sys_device *cpu_sys_dev;
 760         unsigned int j;
 761 #endif
 762
 763         cpufreq_debug_disable_ratelimit();
 764         dprintk("unregistering CPU %u\n", cpu);
 765
 766         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 767         data = cpufreq_cpu_data[cpu];
 768
 769         if (!data) {
 770                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 771                 cpufreq_debug_enable_ratelimit();
 772                 return -EINVAL;
 773         }
 774         cpufreq_cpu_data[cpu] = NULL;
 775
 776
 777 #ifdef CONFIG_SMP
 778         /* if this isn't the CPU which is the parent of the kobj, we
 779          * only need to unlink, put and exit
 780          */
 781         if (unlikely(cpu != data->cpu)) {
 782                 dprintk("removing link\n");
 783                 cpu_clear(cpu, data->cpus);
 784                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 785                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 786                 cpufreq_cpu_put(data);
 787                 cpufreq_debug_enable_ratelimit();
 788                 return 0;
 789         }
 790 #endif
 791
 792
 793         if (!kobject_get(&data->kobj)) {
 794                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 795                 cpufreq_debug_enable_ratelimit();
 796                 return -EFAULT;
 797         }
 798
 799 #ifdef CONFIG_SMP
 800         /* if we have other CPUs still registered, we need to unlink them,
 801          * or else wait_for_completion below will lock up. Clean the
 802          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 803          * links afterwards.
 804          */
 805         if (unlikely(cpus_weight(data->cpus) > 1)) {
 806                 for_each_cpu_mask(j, data->cpus) {
 807                         if (j == cpu)
 808                                 continue;
 809                         cpufreq_cpu_data[j] = NULL;
 810                 }
 811         }
 812
 813         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 814
 815         if (unlikely(cpus_weight(data->cpus) > 1)) {
 816                 for_each_cpu_mask(j, data->cpus) {
 817                         if (j == cpu)
 818                                 continue;
 819                         dprintk("removing link for cpu %u\n", j);
 820                         cpu_sys_dev = get_cpu_sysdev(j);
 821                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 822                         cpufreq_cpu_put(data);
 823                 }
 824         }
 825 #else
 826         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 827 #endif
 828
 829         mutex_lock(&data->lock);
 830         if (cpufreq_driver->target)
 831                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 832         mutex_unlock(&data->lock);
 833
 834         kobject_unregister(&data->kobj);
 835
 836         kobject_put(&data->kobj);
 837
 838         /* we need to make sure that the underlying kobj is actually
 839          * not referenced anymore by anybody before we proceed with
 840          * unloading.
 841          */
 842         dprintk("waiting for dropping of refcount\n");
 843         wait_for_completion(&data->kobj_unregister);
 844         dprintk("wait complete\n");
 845
 846         if (cpufreq_driver->exit)
 847                 cpufreq_driver->exit(data);
 848
 849         kfree(data);
 850
 851         cpufreq_debug_enable_ratelimit();
 852         return 0;
 853 }
 854
 855
 856 static void handle_update(void *data)
 857 {
 858         unsigned int cpu = (unsigned int)(long)data;
 859         dprintk("handle_update for cpu %u called\n", cpu);
 860         cpufreq_update_policy(cpu);
 861 }
 862
 863 /**
 864  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 865  *      @cpu: cpu number
 866  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 867  *      @new_freq: CPU frequency the CPU actually runs at
 868  *
 869  *      We adjust to current frequency first, and need to clean up later. So either call
 870  *      to cpufreq_update_policy() or schedule handle_update()).
 871  */
 872 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 873 {
 874         struct cpufreq_freqs freqs;
 875
 876         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
 877                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 878
 879         freqs.cpu = cpu;
 880         freqs.old = old_freq;
 881         freqs.new = new_freq;
 882         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 883         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 884 }
 885
 886
 887 /**
 888  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
 889  * @cpu: CPU number
 890  *
 891  * This is the last known freq, without actually getting it from the driver.
 892  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 893  */
 894 unsigned int cpufreq_quick_get(unsigned int cpu)
 895 {
 896         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 897         unsigned int ret = 0;
 898
 899         if (policy) {
 900                 mutex_lock(&policy->lock);
 901                 ret = policy->cur;
 902                 mutex_unlock(&policy->lock);
 903                 cpufreq_cpu_put(policy);
 904         }
 905
 906         return (ret);
 907 }
 908 EXPORT_SYMBOL(cpufreq_quick_get);
 909
 910
 911 /**
 912  * cpufreq_get - get the current CPU frequency (in kHz)
 913  * @cpu: CPU number
 914  *
 915  * Get the CPU current (static) CPU frequency
 916  */
 917 unsigned int cpufreq_get(unsigned int cpu)
 918 {
 919         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 920         unsigned int ret = 0;
 921
 922         if (!policy)
 923                 return 0;
 924
 925         if (!cpufreq_driver->get)
 926                 goto out;
 927
 928         mutex_lock(&policy->lock);
 929
 930         ret = cpufreq_driver->get(cpu);
 931
 932         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 933                 /* verify no discrepancy between actual and saved value exists */
 934                 if (unlikely(ret != policy->cur)) {
 935                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 936                         schedule_work(&policy->update);
 937                 }
 938         }
 939
 940         mutex_unlock(&policy->lock);
 941
 942 out:
 943         cpufreq_cpu_put(policy);
 944
 945         return (ret);
 946 }
 947 EXPORT_SYMBOL(cpufreq_get);
 948
 949
 950 /**
 951  *      cpufreq_suspend - let the low level driver prepare for suspend
 952  */
 953
 954 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 955 {
 956         int cpu = sysdev->id;
 957         unsigned int ret = 0;
 958         unsigned int cur_freq = 0;
 959         struct cpufreq_policy *cpu_policy;
 960
 961         dprintk("resuming cpu %u\n", cpu);
 962
 963         if (!cpu_online(cpu))
 964                 return 0;
 965
 966         /* we may be lax here as interrupts are off. Nonetheless
 967          * we need to grab the correct cpu policy, as to check
 968          * whether we really run on this CPU.
 969          */
 970
 971         cpu_policy = cpufreq_cpu_get(cpu);
 972         if (!cpu_policy)
 973                 return -EINVAL;
 974
 975         /* only handle each CPU group once */
 976         if (unlikely(cpu_policy->cpu != cpu)) {
 977                 cpufreq_cpu_put(cpu_policy);
 978                 return 0;
 979         }
 980
 981         if (cpufreq_driver->suspend) {
 982                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
 983                 if (ret) {
 984                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
 985                                         "step on CPU %u\n", cpu_policy->cpu);
 986                         cpufreq_cpu_put(cpu_policy);
 987                         return ret;
 988                 }
 989         }
 990
 991
 992         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
 993                 goto out;
 994
 995         if (cpufreq_driver->get)
 996                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
 997
 998         if (!cur_freq || !cpu_policy->cur) {
 999                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1000                        "frequency is what timing core thinks it is.\n");
1001                 goto out;
1002         }
1003
1004         if (unlikely(cur_freq != cpu_policy->cur)) {
1005                 struct cpufreq_freqs freqs;
1006
1007                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1008                         dprintk("Warning: CPU frequency is %u, "
1009                                "cpufreq assumed %u kHz.\n",
1010                                cur_freq, cpu_policy->cur);
1011
1012                 freqs.cpu = cpu;
1013                 freqs.old = cpu_policy->cur;
1014                 freqs.new = cur_freq;
1015
1016                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
1017                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1018                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1019
1020                 cpu_policy->cur = cur_freq;
1021         }
1022
1023 out:
1024         cpufreq_cpu_put(cpu_policy);
1025         return 0;
1026 }
1027
1028 /**
1029  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1030  *
1031  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1032  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1033  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1034  *          restored.
1035  */
1036 static int cpufreq_resume(struct sys_device * sysdev)
1037 {
1038         int cpu = sysdev->id;
1039         unsigned int ret = 0;
1040         struct cpufreq_policy *cpu_policy;
1041
1042         dprintk("resuming cpu %u\n", cpu);
1043
1044         if (!cpu_online(cpu))
1045                 return 0;
1046
1047         /* we may be lax here as interrupts are off. Nonetheless
1048          * we need to grab the correct cpu policy, as to check
1049          * whether we really run on this CPU.
1050          */
1051
1052         cpu_policy = cpufreq_cpu_get(cpu);
1053         if (!cpu_policy)
1054                 return -EINVAL;
1055
1056         /* only handle each CPU group once */
1057         if (unlikely(cpu_policy->cpu != cpu)) {
1058                 cpufreq_cpu_put(cpu_policy);
1059                 return 0;
1060         }
1061
1062         if (cpufreq_driver->resume) {
1063                 ret = cpufreq_driver->resume(cpu_policy);
1064                 if (ret) {
1065                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1066                                         "step on CPU %u\n", cpu_policy->cpu);
1067                         cpufreq_cpu_put(cpu_policy);
1068                         return ret;
1069                 }
1070         }
1071
1072         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1073                 unsigned int cur_freq = 0;
1074
1075                 if (cpufreq_driver->get)
1076                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1077
1078                 if (!cur_freq || !cpu_policy->cur) {
1079                         printk(KERN_ERR "cpufreq: resume failed to assert "
1080                                         "current frequency is what timing core "
1081                                         "thinks it is.\n");
1082                         goto out;
1083                 }
1084
1085                 if (unlikely(cur_freq != cpu_policy->cur)) {
1086                         struct cpufreq_freqs freqs;
1087
1088                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1089                                 dprintk("Warning: CPU frequency"
1090                                        "is %u, cpufreq assumed %u kHz.\n",
1091                                        cur_freq, cpu_policy->cur);
1092
1093                         freqs.cpu = cpu;
1094                         freqs.old = cpu_policy->cur;
1095                         freqs.new = cur_freq;
1096
1097                         blocking_notifier_call_chain(
1098                                         &cpufreq_transition_notifier_list,
1099                                         CPUFREQ_RESUMECHANGE, &freqs);
1100                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1101
1102                         cpu_policy->cur = cur_freq;
1103                 }
1104         }
1105
1106 out:
1107         schedule_work(&cpu_policy->update);
1108         cpufreq_cpu_put(cpu_policy);
1109         return ret;
1110 }
1111
1112 static struct sysdev_driver cpufreq_sysdev_driver = {
1113         .add            = cpufreq_add_dev,
1114         .remove         = cpufreq_remove_dev,
1115         .suspend        = cpufreq_suspend,
1116         .resume         = cpufreq_resume,
1117 };
1118
1119
1120 /*********************************************************************
1121  *                     NOTIFIER LISTS INTERFACE                      *
1122  *********************************************************************/
1123
1124 /**
1125  *      cpufreq_register_notifier - register a driver with cpufreq
1126  *      @nb: notifier function to register
1127  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1128  *
1129  *      Add a driver to one of two lists: either a list of drivers that
1130  *      are notified about clock rate changes (once before and once after
1131  *      the transition), or a list of drivers that are notified about
1132  *      changes in cpufreq policy.
1133  *
1134  *      This function may sleep, and has the same return conditions as
1135  *      blocking_notifier_chain_register.
1136  */
1137 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1138 {
1139         int ret;
1140
1141         switch (list) {
1142         case CPUFREQ_TRANSITION_NOTIFIER:
1143                 ret = blocking_notifier_chain_register(
1144                                 &cpufreq_transition_notifier_list, nb);
1145                 break;
1146         case CPUFREQ_POLICY_NOTIFIER:
1147                 ret = blocking_notifier_chain_register(
1148                                 &cpufreq_policy_notifier_list, nb);
1149                 break;
1150         default:
1151                 ret = -EINVAL;
1152         }
1153
1154         return ret;
1155 }
1156 EXPORT_SYMBOL(cpufreq_register_notifier);
1157
1158
1159 /**
1160  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1161  *      @nb: notifier block to be unregistered
1162  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1163  *
1164  *      Remove a driver from the CPU frequency notifier list.
1165  *
1166  *      This function may sleep, and has the same return conditions as
1167  *      blocking_notifier_chain_unregister.
1168  */
1169 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1170 {
1171         int ret;
1172
1173         switch (list) {
1174         case CPUFREQ_TRANSITION_NOTIFIER:
1175                 ret = blocking_notifier_chain_unregister(
1176                                 &cpufreq_transition_notifier_list, nb);
1177                 break;
1178         case CPUFREQ_POLICY_NOTIFIER:
1179                 ret = blocking_notifier_chain_unregister(
1180                                 &cpufreq_policy_notifier_list, nb);
1181                 break;
1182         default:
1183                 ret = -EINVAL;
1184         }
1185
1186         return ret;
1187 }
1188 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1189
1190
1191 /*********************************************************************
1192  *                              GOVERNORS                            *
1193  *********************************************************************/
1194
1195
1196 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1197                             unsigned int target_freq,
1198                             unsigned int relation)
1199 {
1200         int retval = -EINVAL;
1201
1202         lock_cpu_hotplug();
1203         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1204                 target_freq, relation);
1205         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1206                 retval = cpufreq_driver->target(policy, target_freq, relation);
1207
1208         unlock_cpu_hotplug();
1209
1210         return retval;
1211 }
1212 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1213
1214 int cpufreq_driver_target(struct cpufreq_policy *policy,
1215                           unsigned int target_freq,
1216                           unsigned int relation)
1217 {
1218         int ret;
1219
1220         policy = cpufreq_cpu_get(policy->cpu);
1221         if (!policy)
1222                 return -EINVAL;
1223
1224         mutex_lock(&policy->lock);
1225
1226         ret = __cpufreq_driver_target(policy, target_freq, relation);
1227
1228         mutex_unlock(&policy->lock);
1229
1230         cpufreq_cpu_put(policy);
1231         return ret;
1232 }
1233 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1234
1235
1236 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1237 {
1238         int ret;
1239
1240         if (!try_module_get(policy->governor->owner))
1241                 return -EINVAL;
1242
1243         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1244         ret = policy->governor->governor(policy, event);
1245
1246         /* we keep one module reference alive for each CPU governed by this CPU */
1247         if ((event != CPUFREQ_GOV_START) || ret)
1248                 module_put(policy->governor->owner);
1249         if ((event == CPUFREQ_GOV_STOP) && !ret)
1250                 module_put(policy->governor->owner);
1251
1252         return ret;
1253 }
1254
1255
1256 int cpufreq_governor(unsigned int cpu, unsigned int event)
1257 {
1258         int ret = 0;
1259         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1260
1261         if (!policy)
1262                 return -EINVAL;
1263
1264         mutex_lock(&policy->lock);
1265         ret = __cpufreq_governor(policy, event);
1266         mutex_unlock(&policy->lock);
1267
1268         cpufreq_cpu_put(policy);
1269         return ret;
1270 }
1271 EXPORT_SYMBOL_GPL(cpufreq_governor);
1272
1273
1274 int cpufreq_register_governor(struct cpufreq_governor *governor)
1275 {
1276         struct cpufreq_governor *t;
1277
1278         if (!governor)
1279                 return -EINVAL;
1280
1281         mutex_lock(&cpufreq_governor_mutex);
1282
1283         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1284                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1285                         mutex_unlock(&cpufreq_governor_mutex);
1286                         return -EBUSY;
1287                 }
1288         }
1289         list_add(&governor->governor_list, &cpufreq_governor_list);
1290
1291         mutex_unlock(&cpufreq_governor_mutex);
1292         return 0;
1293 }
1294 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1295
1296
1297 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1298 {
1299         if (!governor)
1300                 return;
1301
1302         mutex_lock(&cpufreq_governor_mutex);
1303         list_del(&governor->governor_list);
1304         mutex_unlock(&cpufreq_governor_mutex);
1305         return;
1306 }
1307 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1308
1309
1310
1311 /*********************************************************************
1312  *                          POLICY INTERFACE                         *
1313  *********************************************************************/
1314
1315 /**
1316  * cpufreq_get_policy - get the current cpufreq_policy
1317  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1318  *
1319  * Reads the current cpufreq policy.
1320  */
1321 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1322 {
1323         struct cpufreq_policy *cpu_policy;
1324         if (!policy)
1325                 return -EINVAL;
1326
1327         cpu_policy = cpufreq_cpu_get(cpu);
1328         if (!cpu_policy)
1329                 return -EINVAL;
1330
1331         mutex_lock(&cpu_policy->lock);
1332         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1333         mutex_unlock(&cpu_policy->lock);
1334
1335         cpufreq_cpu_put(cpu_policy);
1336         return 0;
1337 }
1338 EXPORT_SYMBOL(cpufreq_get_policy);
1339
1340
1341 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1342 {
1343         int ret = 0;
1344
1345         cpufreq_debug_disable_ratelimit();
1346         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1347                 policy->min, policy->max);
1348
1349         memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
1350
1351         /* verify the cpu speed can be set within this limit */
1352         ret = cpufreq_driver->verify(policy);
1353         if (ret)
1354                 goto error_out;
1355
1356         /* adjust if necessary - all reasons */
1357         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1358                         CPUFREQ_ADJUST, policy);
1359
1360         /* adjust if necessary - hardware incompatibility*/
1361         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1362                         CPUFREQ_INCOMPATIBLE, policy);
1363
1364         /* verify the cpu speed can be set within this limit,
1365            which might be different to the first one */
1366         ret = cpufreq_driver->verify(policy);
1367         if (ret)
1368                 goto error_out;
1369
1370         /* notification of the new policy */
1371         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1372                         CPUFREQ_NOTIFY, policy);
1373
1374         data->min = policy->min;
1375         data->max = policy->max;
1376
1377         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1378
1379         if (cpufreq_driver->setpolicy) {
1380                 data->policy = policy->policy;
1381                 dprintk("setting range\n");
1382                 ret = cpufreq_driver->setpolicy(policy);
1383         } else {
1384                 if (policy->governor != data->governor) {
1385                         /* save old, working values */
1386                         struct cpufreq_governor *old_gov = data->governor;
1387
1388                         dprintk("governor switch\n");
1389
1390                         /* end old governor */
1391                         if (data->governor)
1392                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1393
1394                         /* start new governor */
1395                         data->governor = policy->governor;
1396                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1397                                 /* new governor failed, so re-start old one */
1398                                 dprintk("starting governor %s failed\n", data->governor->name);
1399                                 if (old_gov) {
1400                                         data->governor = old_gov;
1401                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1402                                 }
1403                                 ret = -EINVAL;
1404                                 goto error_out;
1405                         }
1406                         /* might be a policy change, too, so fall through */
1407                 }
1408                 dprintk("governor: change or update limits\n");
1409                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1410         }
1411
1412 error_out:
1413         cpufreq_debug_enable_ratelimit();
1414         return ret;
1415 }
1416
1417 /**
1418  *      cpufreq_set_policy - set a new CPUFreq policy
1419  *      @policy: policy to be set.
1420  *
1421  *      Sets a new CPU frequency and voltage scaling policy.
1422  */
1423 int cpufreq_set_policy(struct cpufreq_policy *policy)
1424 {
1425         int ret = 0;
1426         struct cpufreq_policy *data;
1427
1428         if (!policy)
1429                 return -EINVAL;
1430
1431         data = cpufreq_cpu_get(policy->cpu);
1432         if (!data)
1433                 return -EINVAL;
1434
1435         /* lock this CPU */
1436         mutex_lock(&data->lock);
1437
1438         ret = __cpufreq_set_policy(data, policy);
1439         data->user_policy.min = data->min;
1440         data->user_policy.max = data->max;
1441         data->user_policy.policy = data->policy;
1442         data->user_policy.governor = data->governor;
1443
1444         mutex_unlock(&data->lock);
1445         cpufreq_cpu_put(data);
1446
1447         return ret;
1448 }
1449 EXPORT_SYMBOL(cpufreq_set_policy);
1450
1451
1452 /**
1453  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1454  *      @cpu: CPU which shall be re-evaluated
1455  *
1456  *      Usefull for policy notifiers which have different necessities
1457  *      at different times.
1458  */
1459 int cpufreq_update_policy(unsigned int cpu)
1460 {
1461         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1462         struct cpufreq_policy policy;
1463         int ret = 0;
1464
1465         if (!data)
1466                 return -ENODEV;
1467
1468         mutex_lock(&data->lock);
1469
1470         dprintk("updating policy for CPU %u\n", cpu);
1471         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1472         policy.min = data->user_policy.min;
1473         policy.max = data->user_policy.max;
1474         policy.policy = data->user_policy.policy;
1475         policy.governor = data->user_policy.governor;
1476
1477         /* BIOS might change freq behind our back
1478           -> ask driver for current freq and notify governors about a change */
1479         if (cpufreq_driver->get) {
1480                 policy.cur = cpufreq_driver->get(cpu);
1481                 if (!data->cur) {
1482                         dprintk("Driver did not initialize current freq");
1483                         data->cur = policy.cur;
1484                 } else {
1485                         if (data->cur != policy.cur)
1486                                 cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1487                 }
1488         }
1489
1490         ret = __cpufreq_set_policy(data, &policy);
1491
1492         mutex_unlock(&data->lock);
1493
1494         cpufreq_cpu_put(data);
1495         return ret;
1496 }
1497 EXPORT_SYMBOL(cpufreq_update_policy);
1498
1499 #ifdef CONFIG_HOTPLUG_CPU
1500 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1501                                         unsigned long action, void *hcpu)
1502 {
1503         unsigned int cpu = (unsigned long)hcpu;
1504         struct cpufreq_policy *policy;
1505         struct sys_device *sys_dev;
1506
1507         sys_dev = get_cpu_sysdev(cpu);
1508
1509         if (sys_dev) {
1510                 switch (action) {
1511                 case CPU_ONLINE:
1512                         cpufreq_add_dev(sys_dev);
1513                         break;
1514                 case CPU_DOWN_PREPARE:
1515                         /*
1516                          * We attempt to put this cpu in lowest frequency
1517                          * possible before going down. This will permit
1518                          * hardware-managed P-State to switch other related
1519                          * threads to min or higher speeds if possible.
1520                          */
1521                         policy = cpufreq_cpu_data[cpu];
1522                         if (policy) {
1523                                 cpufreq_driver_target(policy, policy->min,
1524                                                 CPUFREQ_RELATION_H);
1525                         }
1526                         break;
1527                 case CPU_DEAD:
1528                         cpufreq_remove_dev(sys_dev);
1529                         break;
1530                 }
1531         }
1532         return NOTIFY_OK;
1533 }
1534
1535 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1536 {
1537     .notifier_call = cpufreq_cpu_callback,
1538 };
1539 #endif /* CONFIG_HOTPLUG_CPU */
1540
1541 /*********************************************************************
1542  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1543  *********************************************************************/
1544
1545 /**
1546  * cpufreq_register_driver - register a CPU Frequency driver
1547  * @driver_data: A struct cpufreq_driver containing the values#
1548  * submitted by the CPU Frequency driver.
1549  *
1550  *   Registers a CPU Frequency driver to this core code. This code
1551  * returns zero on success, -EBUSY when another driver got here first
1552  * (and isn't unregistered in the meantime).
1553  *
1554  */
1555 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1556 {
1557         unsigned long flags;
1558         int ret;
1559
1560         if (!driver_data || !driver_data->verify || !driver_data->init ||
1561             ((!driver_data->setpolicy) && (!driver_data->target)))
1562                 return -EINVAL;
1563
1564         dprintk("trying to register driver %s\n", driver_data->name);
1565
1566         if (driver_data->setpolicy)
1567                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1568
1569         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1570         if (cpufreq_driver) {
1571                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1572                 return -EBUSY;
1573         }
1574         cpufreq_driver = driver_data;
1575         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1576
1577         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1578
1579         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1580                 int i;
1581                 ret = -ENODEV;
1582
1583                 /* check for at least one working CPU */
1584                 for (i=0; i<NR_CPUS; i++)
1585                         if (cpufreq_cpu_data[i])
1586                                 ret = 0;
1587
1588                 /* if all ->init() calls failed, unregister */
1589                 if (ret) {
1590                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1591                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1592
1593                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1594                         cpufreq_driver = NULL;
1595                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1596                 }
1597         }
1598
1599         if (!ret) {
1600                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1601                 dprintk("driver %s up and running\n", driver_data->name);
1602                 cpufreq_debug_enable_ratelimit();
1603         }
1604
1605         return (ret);
1606 }
1607 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1608
1609
1610 /**
1611  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1612  *
1613  *    Unregister the current CPUFreq driver. Only call this if you have
1614  * the right to do so, i.e. if you have succeeded in initialising before!
1615  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1616  * currently not initialised.
1617  */
1618 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1619 {
1620         unsigned long flags;
1621
1622         cpufreq_debug_disable_ratelimit();
1623
1624         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1625                 cpufreq_debug_enable_ratelimit();
1626                 return -EINVAL;
1627         }
1628
1629         dprintk("unregistering driver %s\n", driver->name);
1630
1631         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1632         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1633
1634         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1635         cpufreq_driver = NULL;
1636         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1637
1638         return 0;
1639 }
1640 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);