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