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