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 struct notifier_block *cpufreq_policy_notifier_list;
  56 static struct notifier_block *cpufreq_transition_notifier_list;
  57 static DECLARE_RWSEM (cpufreq_notifier_rwsem);
  58
  59
  60 static LIST_HEAD(cpufreq_governor_list);
  61 static DEFINE_MUTEX (cpufreq_governor_mutex);
  62
  63 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
  64 {
  65         struct cpufreq_policy *data;
  66         unsigned long flags;
  67
  68         if (cpu >= NR_CPUS)
  69                 goto err_out;
  70
  71         /* get the cpufreq driver */
  72         spin_lock_irqsave(&cpufreq_driver_lock, flags);
  73
  74         if (!cpufreq_driver)
  75                 goto err_out_unlock;
  76
  77         if (!try_module_get(cpufreq_driver->owner))
  78                 goto err_out_unlock;
  79
  80
  81         /* get the CPU */
  82         data = cpufreq_cpu_data[cpu];
  83
  84         if (!data)
  85                 goto err_out_put_module;
  86
  87         if (!kobject_get(&data->kobj))
  88                 goto err_out_put_module;
  89
  90         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  91         return data;
  92
  93 err_out_put_module:
  94         module_put(cpufreq_driver->owner);
  95 err_out_unlock:
  96         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  97 err_out:
  98         return NULL;
  99 }
 100 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 101
 102
 103 void cpufreq_cpu_put(struct cpufreq_policy *data)
 104 {
 105         kobject_put(&data->kobj);
 106         module_put(cpufreq_driver->owner);
 107 }
 108 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 109
 110
 111 /*********************************************************************
 112  *                     UNIFIED DEBUG HELPERS                         *
 113  *********************************************************************/
 114 #ifdef CONFIG_CPU_FREQ_DEBUG
 115
 116 /* what part(s) of the CPUfreq subsystem are debugged? */
 117 static unsigned int debug;
 118
 119 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 120  * set or modify this value.
 121  */
 122 static unsigned int debug_ratelimit = 1;
 123
 124 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 125  * loading of a cpufreq driver, temporarily disabled when a new policy
 126  * is set, and disabled upon cpufreq driver removal
 127  */
 128 static unsigned int disable_ratelimit = 1;
 129 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 130
 131 static void cpufreq_debug_enable_ratelimit(void)
 132 {
 133         unsigned long flags;
 134
 135         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 136         if (disable_ratelimit)
 137                 disable_ratelimit--;
 138         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 139 }
 140
 141 static void cpufreq_debug_disable_ratelimit(void)
 142 {
 143         unsigned long flags;
 144
 145         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 146         disable_ratelimit++;
 147         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 148 }
 149
 150 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
 151 {
 152         char s[256];
 153         va_list args;
 154         unsigned int len;
 155         unsigned long flags;
 156
 157         WARN_ON(!prefix);
 158         if (type & debug) {
 159                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 160                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
 161                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 162                         return;
 163                 }
 164                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 165
 166                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 167
 168                 va_start(args, fmt);
 169                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 170                 va_end(args);
 171
 172                 printk(s);
 173
 174                 WARN_ON(len < 5);
 175         }
 176 }
 177 EXPORT_SYMBOL(cpufreq_debug_printk);
 178
 179
 180 module_param(debug, uint, 0644);
 181 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
 182
 183 module_param(debug_ratelimit, uint, 0644);
 184 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
 185
 186 #else /* !CONFIG_CPU_FREQ_DEBUG */
 187
 188 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 189 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 190
 191 #endif /* CONFIG_CPU_FREQ_DEBUG */
 192
 193
 194 /*********************************************************************
 195  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 196  *********************************************************************/
 197
 198 /**
 199  * adjust_jiffies - adjust the system "loops_per_jiffy"
 200  *
 201  * This function alters the system "loops_per_jiffy" for the clock
 202  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 203  * systems as each CPU might be scaled differently. So, use the arch
 204  * per-CPU loops_per_jiffy value wherever possible.
 205  */
 206 #ifndef CONFIG_SMP
 207 static unsigned long l_p_j_ref;
 208 static unsigned int  l_p_j_ref_freq;
 209
 210 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 211 {
 212         if (ci->flags & CPUFREQ_CONST_LOOPS)
 213                 return;
 214
 215         if (!l_p_j_ref_freq) {
 216                 l_p_j_ref = loops_per_jiffy;
 217                 l_p_j_ref_freq = ci->old;
 218                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 219         }
 220         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 221             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 222             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 223                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
 224                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
 225         }
 226 }
 227 #else
 228 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
 229 #endif
 230
 231
 232 /**
 233  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 234  * on frequency transition.
 235  *
 236  * This function calls the transition notifiers and the "adjust_jiffies"
 237  * function. It is called twice on all CPU frequency changes that have
 238  * external effects.
 239  */
 240 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 241 {
 242         struct cpufreq_policy *policy;
 243
 244         BUG_ON(irqs_disabled());
 245
 246         freqs->flags = cpufreq_driver->flags;
 247         dprintk("notification %u of frequency transition to %u kHz\n",
 248                 state, freqs->new);
 249
 250         down_read(&cpufreq_notifier_rwsem);
 251
 252         policy = cpufreq_cpu_data[freqs->cpu];
 253         switch (state) {
 254
 255         case CPUFREQ_PRECHANGE:
 256                 /* detect if the driver reported a value as "old frequency"
 257                  * which is not equal to what the cpufreq core thinks is
 258                  * "old frequency".
 259                  */
 260                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 261                         if ((policy) && (policy->cpu == freqs->cpu) &&
 262                             (policy->cur) && (policy->cur != freqs->old)) {
 263                                 dprintk(KERN_WARNING "Warning: CPU frequency is"
 264                                         " %u, cpufreq assumed %u kHz.\n",
 265                                         freqs->old, policy->cur);
 266                                 freqs->old = policy->cur;
 267                         }
 268                 }
 269                 notifier_call_chain(&cpufreq_transition_notifier_list,
 270                                         CPUFREQ_PRECHANGE, freqs);
 271                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 272                 break;
 273
 274         case CPUFREQ_POSTCHANGE:
 275                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 276                 notifier_call_chain(&cpufreq_transition_notifier_list,
 277                                         CPUFREQ_POSTCHANGE, freqs);
 278                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 279                         policy->cur = freqs->new;
 280                 break;
 281         }
 282         up_read(&cpufreq_notifier_rwsem);
 283 }
 284 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 285
 286
 287
 288 /*********************************************************************
 289  *                          SYSFS INTERFACE                          *
 290  *********************************************************************/
 291
 292 /**
 293  * cpufreq_parse_governor - parse a governor string
 294  */
 295 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 296                                 struct cpufreq_governor **governor)
 297 {
 298         if (!cpufreq_driver)
 299                 return -EINVAL;
 300         if (cpufreq_driver->setpolicy) {
 301                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 302                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 303                         return 0;
 304                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 305                         *policy = CPUFREQ_POLICY_POWERSAVE;
 306                         return 0;
 307                 }
 308                 return -EINVAL;
 309         } else {
 310                 struct cpufreq_governor *t;
 311                 mutex_lock(&cpufreq_governor_mutex);
 312                 if (!cpufreq_driver || !cpufreq_driver->target)
 313                         goto out;
 314                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 315                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
 316                                 *governor = t;
 317                                 mutex_unlock(&cpufreq_governor_mutex);
 318                                 return 0;
 319                         }
 320                 }
 321 out:
 322                 mutex_unlock(&cpufreq_governor_mutex);
 323         }
 324         return -EINVAL;
 325 }
 326 EXPORT_SYMBOL_GPL(cpufreq_parse_governor);
 327
 328
 329 /* drivers/base/cpu.c */
 330 extern struct sysdev_class cpu_sysdev_class;
 331
 332
 333 /**
 334  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 335  *
 336  * Write out information from cpufreq_driver->policy[cpu]; object must be
 337  * "unsigned int".
 338  */
 339
 340 #define show_one(file_name, object)                     \
 341 static ssize_t show_##file_name                         \
 342 (struct cpufreq_policy * policy, char *buf)             \
 343 {                                                       \
 344         return sprintf (buf, "%u\n", policy->object);   \
 345 }
 346
 347 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 348 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 349 show_one(scaling_min_freq, min);
 350 show_one(scaling_max_freq, max);
 351 show_one(scaling_cur_freq, cur);
 352
 353 /**
 354  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 355  */
 356 #define store_one(file_name, object)                    \
 357 static ssize_t store_##file_name                                        \
 358 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 359 {                                                                       \
 360         unsigned int ret = -EINVAL;                                     \
 361         struct cpufreq_policy new_policy;                               \
 362                                                                         \
 363         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 364         if (ret)                                                        \
 365                 return -EINVAL;                                         \
 366                                                                         \
 367         ret = sscanf (buf, "%u", &new_policy.object);                   \
 368         if (ret != 1)                                                   \
 369                 return -EINVAL;                                         \
 370                                                                         \
 371         ret = cpufreq_set_policy(&new_policy);                          \
 372                                                                         \
 373         return ret ? ret : count;                                       \
 374 }
 375
 376 store_one(scaling_min_freq,min);
 377 store_one(scaling_max_freq,max);
 378
 379 /**
 380  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 381  */
 382 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 383 {
 384         unsigned int cur_freq = cpufreq_get(policy->cpu);
 385         if (!cur_freq)
 386                 return sprintf(buf, "<unknown>");
 387         return sprintf(buf, "%u\n", cur_freq);
 388 }
 389
 390
 391 /**
 392  * show_scaling_governor - show the current policy for the specified CPU
 393  */
 394 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 395 {
 396         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 397                 return sprintf(buf, "powersave\n");
 398         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 399                 return sprintf(buf, "performance\n");
 400         else if (policy->governor)
 401                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 402         return -EINVAL;
 403 }
 404
 405
 406 /**
 407  * store_scaling_governor - store policy for the specified CPU
 408  */
 409 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 410                                        const char *buf, size_t count)
 411 {
 412         unsigned int ret = -EINVAL;
 413         char    str_governor[16];
 414         struct cpufreq_policy new_policy;
 415
 416         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 417         if (ret)
 418                 return ret;
 419
 420         ret = sscanf (buf, "%15s", str_governor);
 421         if (ret != 1)
 422                 return -EINVAL;
 423
 424         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 425                 return -EINVAL;
 426
 427         ret = cpufreq_set_policy(&new_policy);
 428         return ret ? ret : count;
 429 }
 430
 431 /**
 432  * show_scaling_driver - show the cpufreq driver currently loaded
 433  */
 434 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 435 {
 436         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 437 }
 438
 439 /**
 440  * show_scaling_available_governors - show the available CPUfreq governors
 441  */
 442 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 443                                 char *buf)
 444 {
 445         ssize_t i = 0;
 446         struct cpufreq_governor *t;
 447
 448         if (!cpufreq_driver->target) {
 449                 i += sprintf(buf, "performance powersave");
 450                 goto out;
 451         }
 452
 453         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 454                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 455                         goto out;
 456                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 457         }
 458 out:
 459         i += sprintf(&buf[i], "\n");
 460         return i;
 461 }
 462 /**
 463  * show_affected_cpus - show the CPUs affected by each transition
 464  */
 465 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 466 {
 467         ssize_t i = 0;
 468         unsigned int cpu;
 469
 470         for_each_cpu_mask(cpu, policy->cpus) {
 471                 if (i)
 472                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 473                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 474                 if (i >= (PAGE_SIZE - 5))
 475                     break;
 476         }
 477         i += sprintf(&buf[i], "\n");
 478         return i;
 479 }
 480
 481
 482 #define define_one_ro(_name) \
 483 static struct freq_attr _name = \
 484 __ATTR(_name, 0444, show_##_name, NULL)
 485
 486 #define define_one_ro0400(_name) \
 487 static struct freq_attr _name = \
 488 __ATTR(_name, 0400, show_##_name, NULL)
 489
 490 #define define_one_rw(_name) \
 491 static struct freq_attr _name = \
 492 __ATTR(_name, 0644, show_##_name, store_##_name)
 493
 494 define_one_ro0400(cpuinfo_cur_freq);
 495 define_one_ro(cpuinfo_min_freq);
 496 define_one_ro(cpuinfo_max_freq);
 497 define_one_ro(scaling_available_governors);
 498 define_one_ro(scaling_driver);
 499 define_one_ro(scaling_cur_freq);
 500 define_one_ro(affected_cpus);
 501 define_one_rw(scaling_min_freq);
 502 define_one_rw(scaling_max_freq);
 503 define_one_rw(scaling_governor);
 504
 505 static struct attribute * default_attrs[] = {
 506         &cpuinfo_min_freq.attr,
 507         &cpuinfo_max_freq.attr,
 508         &scaling_min_freq.attr,
 509         &scaling_max_freq.attr,
 510         &affected_cpus.attr,
 511         &scaling_governor.attr,
 512         &scaling_driver.attr,
 513         &scaling_available_governors.attr,
 514         NULL
 515 };
 516
 517 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 518 #define to_attr(a) container_of(a,struct freq_attr,attr)
 519
 520 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 521 {
 522         struct cpufreq_policy * policy = to_policy(kobj);
 523         struct freq_attr * fattr = to_attr(attr);
 524         ssize_t ret;
 525         policy = cpufreq_cpu_get(policy->cpu);
 526         if (!policy)
 527                 return -EINVAL;
 528         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 529         cpufreq_cpu_put(policy);
 530         return ret;
 531 }
 532
 533 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 534                      const char * buf, size_t count)
 535 {
 536         struct cpufreq_policy * policy = to_policy(kobj);
 537         struct freq_attr * fattr = to_attr(attr);
 538         ssize_t ret;
 539         policy = cpufreq_cpu_get(policy->cpu);
 540         if (!policy)
 541                 return -EINVAL;
 542         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 543         cpufreq_cpu_put(policy);
 544         return ret;
 545 }
 546
 547 static void cpufreq_sysfs_release(struct kobject * kobj)
 548 {
 549         struct cpufreq_policy * policy = to_policy(kobj);
 550         dprintk("last reference is dropped\n");
 551         complete(&policy->kobj_unregister);
 552 }
 553
 554 static struct sysfs_ops sysfs_ops = {
 555         .show   = show,
 556         .store  = store,
 557 };
 558
 559 static struct kobj_type ktype_cpufreq = {
 560         .sysfs_ops      = &sysfs_ops,
 561         .default_attrs  = default_attrs,
 562         .release        = cpufreq_sysfs_release,
 563 };
 564
 565
 566 /**
 567  * cpufreq_add_dev - add a CPU device
 568  *
 569  * Adds the cpufreq interface for a CPU device.
 570  */
 571 static int cpufreq_add_dev (struct sys_device * sys_dev)
 572 {
 573         unsigned int cpu = sys_dev->id;
 574         int ret = 0;
 575         struct cpufreq_policy new_policy;
 576         struct cpufreq_policy *policy;
 577         struct freq_attr **drv_attr;
 578         struct sys_device *cpu_sys_dev;
 579         unsigned long flags;
 580         unsigned int j;
 581 #ifdef CONFIG_SMP
 582         struct cpufreq_policy *managed_policy;
 583 #endif
 584
 585         if (cpu_is_offline(cpu))
 586                 return 0;
 587
 588         cpufreq_debug_disable_ratelimit();
 589         dprintk("adding CPU %u\n", cpu);
 590
 591 #ifdef CONFIG_SMP
 592         /* check whether a different CPU already registered this
 593          * CPU because it is in the same boat. */
 594         policy = cpufreq_cpu_get(cpu);
 595         if (unlikely(policy)) {
 596                 cpufreq_cpu_put(policy);
 597                 cpufreq_debug_enable_ratelimit();
 598                 return 0;
 599         }
 600 #endif
 601
 602         if (!try_module_get(cpufreq_driver->owner)) {
 603                 ret = -EINVAL;
 604                 goto module_out;
 605         }
 606
 607         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 608         if (!policy) {
 609                 ret = -ENOMEM;
 610                 goto nomem_out;
 611         }
 612
 613         policy->cpu = cpu;
 614         policy->cpus = cpumask_of_cpu(cpu);
 615
 616         mutex_init(&policy->lock);
 617         mutex_lock(&policy->lock);
 618         init_completion(&policy->kobj_unregister);
 619         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 620
 621         /* call driver. From then on the cpufreq must be able
 622          * to accept all calls to ->verify and ->setpolicy for this CPU
 623          */
 624         ret = cpufreq_driver->init(policy);
 625         if (ret) {
 626                 dprintk("initialization failed\n");
 627                 mutex_unlock(&policy->lock);
 628                 goto err_out;
 629         }
 630
 631 #ifdef CONFIG_SMP
 632         for_each_cpu_mask(j, policy->cpus) {
 633                 if (cpu == j)
 634                         continue;
 635
 636                 /* check for existing affected CPUs.  They may not be aware
 637                  * of it due to CPU Hotplug.
 638                  */
 639                 managed_policy = cpufreq_cpu_get(j);
 640                 if (unlikely(managed_policy)) {
 641                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 642                         managed_policy->cpus = policy->cpus;
 643                         cpufreq_cpu_data[cpu] = managed_policy;
 644                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 645
 646                         dprintk("CPU already managed, adding link\n");
 647                         sysfs_create_link(&sys_dev->kobj,
 648                                           &managed_policy->kobj, "cpufreq");
 649
 650                         cpufreq_debug_enable_ratelimit();
 651                         mutex_unlock(&policy->lock);
 652                         ret = 0;
 653                         goto err_out_driver_exit; /* call driver->exit() */
 654                 }
 655         }
 656 #endif
 657         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 658
 659         /* prepare interface data */
 660         policy->kobj.parent = &sys_dev->kobj;
 661         policy->kobj.ktype = &ktype_cpufreq;
 662         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 663
 664         ret = kobject_register(&policy->kobj);
 665         if (ret) {
 666                 mutex_unlock(&policy->lock);
 667                 goto err_out_driver_exit;
 668         }
 669         /* set up files for this cpu device */
 670         drv_attr = cpufreq_driver->attr;
 671         while ((drv_attr) && (*drv_attr)) {
 672                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 673                 drv_attr++;
 674         }
 675         if (cpufreq_driver->get)
 676                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 677         if (cpufreq_driver->target)
 678                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 679
 680         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 681         for_each_cpu_mask(j, policy->cpus)
 682                 cpufreq_cpu_data[j] = policy;
 683         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 684
 685         /* symlink affected CPUs */
 686         for_each_cpu_mask(j, policy->cpus) {
 687                 if (j == cpu)
 688                         continue;
 689                 if (!cpu_online(j))
 690                         continue;
 691
 692                 dprintk("CPU already managed, adding link\n");
 693                 cpufreq_cpu_get(cpu);
 694                 cpu_sys_dev = get_cpu_sysdev(j);
 695                 sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 696                                   "cpufreq");
 697         }
 698
 699         policy->governor = NULL; /* to assure that the starting sequence is
 700                                   * run in cpufreq_set_policy */
 701         mutex_unlock(&policy->lock);
 702
 703         /* set default policy */
 704
 705         ret = cpufreq_set_policy(&new_policy);
 706         if (ret) {
 707                 dprintk("setting policy failed\n");
 708                 goto err_out_unregister;
 709         }
 710
 711         module_put(cpufreq_driver->owner);
 712         dprintk("initialization complete\n");
 713         cpufreq_debug_enable_ratelimit();
 714
 715         return 0;
 716
 717
 718 err_out_unregister:
 719         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 720         for_each_cpu_mask(j, policy->cpus)
 721                 cpufreq_cpu_data[j] = NULL;
 722         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 723
 724         kobject_unregister(&policy->kobj);
 725         wait_for_completion(&policy->kobj_unregister);
 726
 727 err_out_driver_exit:
 728         if (cpufreq_driver->exit)
 729                 cpufreq_driver->exit(policy);
 730
 731 err_out:
 732         kfree(policy);
 733
 734 nomem_out:
 735         module_put(cpufreq_driver->owner);
 736 module_out:
 737         cpufreq_debug_enable_ratelimit();
 738         return ret;
 739 }
 740
 741
 742 /**
 743  * cpufreq_remove_dev - remove a CPU device
 744  *
 745  * Removes the cpufreq interface for a CPU device.
 746  */
 747 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 748 {
 749         unsigned int cpu = sys_dev->id;
 750         unsigned long flags;
 751         struct cpufreq_policy *data;
 752 #ifdef CONFIG_SMP
 753         struct sys_device *cpu_sys_dev;
 754         unsigned int j;
 755 #endif
 756
 757         cpufreq_debug_disable_ratelimit();
 758         dprintk("unregistering CPU %u\n", cpu);
 759
 760         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 761         data = cpufreq_cpu_data[cpu];
 762
 763         if (!data) {
 764                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 765                 cpufreq_debug_enable_ratelimit();
 766                 return -EINVAL;
 767         }
 768         cpufreq_cpu_data[cpu] = NULL;
 769
 770
 771 #ifdef CONFIG_SMP
 772         /* if this isn't the CPU which is the parent of the kobj, we
 773          * only need to unlink, put and exit
 774          */
 775         if (unlikely(cpu != data->cpu)) {
 776                 dprintk("removing link\n");
 777                 cpu_clear(cpu, data->cpus);
 778                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 779                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 780                 cpufreq_cpu_put(data);
 781                 cpufreq_debug_enable_ratelimit();
 782                 return 0;
 783         }
 784 #endif
 785
 786
 787         if (!kobject_get(&data->kobj)) {
 788                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 789                 cpufreq_debug_enable_ratelimit();
 790                 return -EFAULT;
 791         }
 792
 793 #ifdef CONFIG_SMP
 794         /* if we have other CPUs still registered, we need to unlink them,
 795          * or else wait_for_completion below will lock up. Clean the
 796          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 797          * links afterwards.
 798          */
 799         if (unlikely(cpus_weight(data->cpus) > 1)) {
 800                 for_each_cpu_mask(j, data->cpus) {
 801                         if (j == cpu)
 802                                 continue;
 803                         cpufreq_cpu_data[j] = NULL;
 804                 }
 805         }
 806
 807         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 808
 809         if (unlikely(cpus_weight(data->cpus) > 1)) {
 810                 for_each_cpu_mask(j, data->cpus) {
 811                         if (j == cpu)
 812                                 continue;
 813                         dprintk("removing link for cpu %u\n", j);
 814                         cpu_sys_dev = get_cpu_sysdev(j);
 815                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 816                         cpufreq_cpu_put(data);
 817                 }
 818         }
 819 #else
 820         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 821 #endif
 822
 823         mutex_lock(&data->lock);
 824         if (cpufreq_driver->target)
 825                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 826         mutex_unlock(&data->lock);
 827
 828         kobject_unregister(&data->kobj);
 829
 830         kobject_put(&data->kobj);
 831
 832         /* we need to make sure that the underlying kobj is actually
 833          * not referenced anymore by anybody before we proceed with
 834          * unloading.
 835          */
 836         dprintk("waiting for dropping of refcount\n");
 837         wait_for_completion(&data->kobj_unregister);
 838         dprintk("wait complete\n");
 839
 840         if (cpufreq_driver->exit)
 841                 cpufreq_driver->exit(data);
 842
 843         kfree(data);
 844
 845         cpufreq_debug_enable_ratelimit();
 846         return 0;
 847 }
 848
 849
 850 static void handle_update(void *data)
 851 {
 852         unsigned int cpu = (unsigned int)(long)data;
 853         dprintk("handle_update for cpu %u called\n", cpu);
 854         cpufreq_update_policy(cpu);
 855 }
 856
 857 /**
 858  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 859  *      @cpu: cpu number
 860  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 861  *      @new_freq: CPU frequency the CPU actually runs at
 862  *
 863  *      We adjust to current frequency first, and need to clean up later. So either call
 864  *      to cpufreq_update_policy() or schedule handle_update()).
 865  */
 866 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 867 {
 868         struct cpufreq_freqs freqs;
 869
 870         dprintk(KERN_WARNING "Warning: CPU frequency out of sync: cpufreq and timing "
 871                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 872
 873         freqs.cpu = cpu;
 874         freqs.old = old_freq;
 875         freqs.new = new_freq;
 876         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 877         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 878 }
 879
 880
 881 /**
 882  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
 883  * @cpu: CPU number
 884  *
 885  * This is the last known freq, without actually getting it from the driver.
 886  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 887  */
 888 unsigned int cpufreq_quick_get(unsigned int cpu)
 889 {
 890         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 891         unsigned int ret = 0;
 892
 893         if (policy) {
 894                 mutex_lock(&policy->lock);
 895                 ret = policy->cur;
 896                 mutex_unlock(&policy->lock);
 897                 cpufreq_cpu_put(policy);
 898         }
 899
 900         return (ret);
 901 }
 902 EXPORT_SYMBOL(cpufreq_quick_get);
 903
 904
 905 /**
 906  * cpufreq_get - get the current CPU frequency (in kHz)
 907  * @cpu: CPU number
 908  *
 909  * Get the CPU current (static) CPU frequency
 910  */
 911 unsigned int cpufreq_get(unsigned int cpu)
 912 {
 913         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 914         unsigned int ret = 0;
 915
 916         if (!policy)
 917                 return 0;
 918
 919         if (!cpufreq_driver->get)
 920                 goto out;
 921
 922         mutex_lock(&policy->lock);
 923
 924         ret = cpufreq_driver->get(cpu);
 925
 926         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 927                 /* verify no discrepancy between actual and saved value exists */
 928                 if (unlikely(ret != policy->cur)) {
 929                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 930                         schedule_work(&policy->update);
 931                 }
 932         }
 933
 934         mutex_unlock(&policy->lock);
 935
 936 out:
 937         cpufreq_cpu_put(policy);
 938
 939         return (ret);
 940 }
 941 EXPORT_SYMBOL(cpufreq_get);
 942
 943
 944 /**
 945  *      cpufreq_suspend - let the low level driver prepare for suspend
 946  */
 947
 948 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 949 {
 950         int cpu = sysdev->id;
 951         unsigned int ret = 0;
 952         unsigned int cur_freq = 0;
 953         struct cpufreq_policy *cpu_policy;
 954
 955         dprintk("resuming cpu %u\n", cpu);
 956
 957         if (!cpu_online(cpu))
 958                 return 0;
 959
 960         /* we may be lax here as interrupts are off. Nonetheless
 961          * we need to grab the correct cpu policy, as to check
 962          * whether we really run on this CPU.
 963          */
 964
 965         cpu_policy = cpufreq_cpu_get(cpu);
 966         if (!cpu_policy)
 967                 return -EINVAL;
 968
 969         /* only handle each CPU group once */
 970         if (unlikely(cpu_policy->cpu != cpu)) {
 971                 cpufreq_cpu_put(cpu_policy);
 972                 return 0;
 973         }
 974
 975         if (cpufreq_driver->suspend) {
 976                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
 977                 if (ret) {
 978                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
 979                                         "step on CPU %u\n", cpu_policy->cpu);
 980                         cpufreq_cpu_put(cpu_policy);
 981                         return ret;
 982                 }
 983         }
 984
 985
 986         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
 987                 goto out;
 988
 989         if (cpufreq_driver->get)
 990                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
 991
 992         if (!cur_freq || !cpu_policy->cur) {
 993                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
 994                        "frequency is what timing core thinks it is.\n");
 995                 goto out;
 996         }
 997
 998         if (unlikely(cur_freq != cpu_policy->cur)) {
 999                 struct cpufreq_freqs freqs;
1000
1001                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1002                         dprintk(KERN_DEBUG "Warning: CPU frequency is %u, "
1003                                "cpufreq assumed %u kHz.\n",
1004                                cur_freq, cpu_policy->cur);
1005
1006                 freqs.cpu = cpu;
1007                 freqs.old = cpu_policy->cur;
1008                 freqs.new = cur_freq;
1009
1010                 notifier_call_chain(&cpufreq_transition_notifier_list,
1011                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1012                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1013
1014                 cpu_policy->cur = cur_freq;
1015         }
1016
1017 out:
1018         cpufreq_cpu_put(cpu_policy);
1019         return 0;
1020 }
1021
1022 /**
1023  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1024  *
1025  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1026  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1027  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1028  *          restored.
1029  */
1030 static int cpufreq_resume(struct sys_device * sysdev)
1031 {
1032         int cpu = sysdev->id;
1033         unsigned int ret = 0;
1034         struct cpufreq_policy *cpu_policy;
1035
1036         dprintk("resuming cpu %u\n", cpu);
1037
1038         if (!cpu_online(cpu))
1039                 return 0;
1040
1041         /* we may be lax here as interrupts are off. Nonetheless
1042          * we need to grab the correct cpu policy, as to check
1043          * whether we really run on this CPU.
1044          */
1045
1046         cpu_policy = cpufreq_cpu_get(cpu);
1047         if (!cpu_policy)
1048                 return -EINVAL;
1049
1050         /* only handle each CPU group once */
1051         if (unlikely(cpu_policy->cpu != cpu)) {
1052                 cpufreq_cpu_put(cpu_policy);
1053                 return 0;
1054         }
1055
1056         if (cpufreq_driver->resume) {
1057                 ret = cpufreq_driver->resume(cpu_policy);
1058                 if (ret) {
1059                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1060                                         "step on CPU %u\n", cpu_policy->cpu);
1061                         cpufreq_cpu_put(cpu_policy);
1062                         return ret;
1063                 }
1064         }
1065
1066         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1067                 unsigned int cur_freq = 0;
1068
1069                 if (cpufreq_driver->get)
1070                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1071
1072                 if (!cur_freq || !cpu_policy->cur) {
1073                         printk(KERN_ERR "cpufreq: resume failed to assert "
1074                                         "current frequency is what timing core "
1075                                         "thinks it is.\n");
1076                         goto out;
1077                 }
1078
1079                 if (unlikely(cur_freq != cpu_policy->cur)) {
1080                         struct cpufreq_freqs freqs;
1081
1082                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1083                                 dprintk(KERN_WARNING "Warning: CPU frequency"
1084                                        "is %u, cpufreq assumed %u kHz.\n",
1085                                        cur_freq, cpu_policy->cur);
1086
1087                         freqs.cpu = cpu;
1088                         freqs.old = cpu_policy->cur;
1089                         freqs.new = cur_freq;
1090
1091                         notifier_call_chain(&cpufreq_transition_notifier_list,
1092                                         CPUFREQ_RESUMECHANGE, &freqs);
1093                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1094
1095                         cpu_policy->cur = cur_freq;
1096                 }
1097         }
1098
1099 out:
1100         schedule_work(&cpu_policy->update);
1101         cpufreq_cpu_put(cpu_policy);
1102         return ret;
1103 }
1104
1105 static struct sysdev_driver cpufreq_sysdev_driver = {
1106         .add            = cpufreq_add_dev,
1107         .remove         = cpufreq_remove_dev,
1108         .suspend        = cpufreq_suspend,
1109         .resume         = cpufreq_resume,
1110 };
1111
1112
1113 /*********************************************************************
1114  *                     NOTIFIER LISTS INTERFACE                      *
1115  *********************************************************************/
1116
1117 /**
1118  *      cpufreq_register_notifier - register a driver with cpufreq
1119  *      @nb: notifier function to register
1120  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1121  *
1122  *      Add a driver to one of two lists: either a list of drivers that
1123  *      are notified about clock rate changes (once before and once after
1124  *      the transition), or a list of drivers that are notified about
1125  *      changes in cpufreq policy.
1126  *
1127  *      This function may sleep, and has the same return conditions as
1128  *      notifier_chain_register.
1129  */
1130 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1131 {
1132         int ret;
1133
1134         down_write(&cpufreq_notifier_rwsem);
1135         switch (list) {
1136         case CPUFREQ_TRANSITION_NOTIFIER:
1137                 ret = notifier_chain_register(&cpufreq_transition_notifier_list, nb);
1138                 break;
1139         case CPUFREQ_POLICY_NOTIFIER:
1140                 ret = notifier_chain_register(&cpufreq_policy_notifier_list, nb);
1141                 break;
1142         default:
1143                 ret = -EINVAL;
1144         }
1145         up_write(&cpufreq_notifier_rwsem);
1146
1147         return ret;
1148 }
1149 EXPORT_SYMBOL(cpufreq_register_notifier);
1150
1151
1152 /**
1153  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1154  *      @nb: notifier block to be unregistered
1155  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1156  *
1157  *      Remove a driver from the CPU frequency notifier list.
1158  *
1159  *      This function may sleep, and has the same return conditions as
1160  *      notifier_chain_unregister.
1161  */
1162 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1163 {
1164         int ret;
1165
1166         down_write(&cpufreq_notifier_rwsem);
1167         switch (list) {
1168         case CPUFREQ_TRANSITION_NOTIFIER:
1169                 ret = notifier_chain_unregister(&cpufreq_transition_notifier_list, nb);
1170                 break;
1171         case CPUFREQ_POLICY_NOTIFIER:
1172                 ret = notifier_chain_unregister(&cpufreq_policy_notifier_list, nb);
1173                 break;
1174         default:
1175                 ret = -EINVAL;
1176         }
1177         up_write(&cpufreq_notifier_rwsem);
1178
1179         return ret;
1180 }
1181 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1182
1183
1184 /*********************************************************************
1185  *                              GOVERNORS                            *
1186  *********************************************************************/
1187
1188
1189 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1190                             unsigned int target_freq,
1191                             unsigned int relation)
1192 {
1193         int retval = -EINVAL;
1194
1195         lock_cpu_hotplug();
1196         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1197                 target_freq, relation);
1198         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1199                 retval = cpufreq_driver->target(policy, target_freq, relation);
1200
1201         unlock_cpu_hotplug();
1202
1203         return retval;
1204 }
1205 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1206
1207 int cpufreq_driver_target(struct cpufreq_policy *policy,
1208                           unsigned int target_freq,
1209                           unsigned int relation)
1210 {
1211         int ret;
1212
1213         policy = cpufreq_cpu_get(policy->cpu);
1214         if (!policy)
1215                 return -EINVAL;
1216
1217         mutex_lock(&policy->lock);
1218
1219         ret = __cpufreq_driver_target(policy, target_freq, relation);
1220
1221         mutex_unlock(&policy->lock);
1222
1223         cpufreq_cpu_put(policy);
1224         return ret;
1225 }
1226 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1227
1228
1229 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1230 {
1231         int ret;
1232
1233         if (!try_module_get(policy->governor->owner))
1234                 return -EINVAL;
1235
1236         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1237         ret = policy->governor->governor(policy, event);
1238
1239         /* we keep one module reference alive for each CPU governed by this CPU */
1240         if ((event != CPUFREQ_GOV_START) || ret)
1241                 module_put(policy->governor->owner);
1242         if ((event == CPUFREQ_GOV_STOP) && !ret)
1243                 module_put(policy->governor->owner);
1244
1245         return ret;
1246 }
1247
1248
1249 int cpufreq_governor(unsigned int cpu, unsigned int event)
1250 {
1251         int ret = 0;
1252         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1253
1254         if (!policy)
1255                 return -EINVAL;
1256
1257         mutex_lock(&policy->lock);
1258         ret = __cpufreq_governor(policy, event);
1259         mutex_unlock(&policy->lock);
1260
1261         cpufreq_cpu_put(policy);
1262         return ret;
1263 }
1264 EXPORT_SYMBOL_GPL(cpufreq_governor);
1265
1266
1267 int cpufreq_register_governor(struct cpufreq_governor *governor)
1268 {
1269         struct cpufreq_governor *t;
1270
1271         if (!governor)
1272                 return -EINVAL;
1273
1274         mutex_lock(&cpufreq_governor_mutex);
1275
1276         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1277                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1278                         mutex_unlock(&cpufreq_governor_mutex);
1279                         return -EBUSY;
1280                 }
1281         }
1282         list_add(&governor->governor_list, &cpufreq_governor_list);
1283
1284         mutex_unlock(&cpufreq_governor_mutex);
1285         return 0;
1286 }
1287 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1288
1289
1290 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1291 {
1292         if (!governor)
1293                 return;
1294
1295         mutex_lock(&cpufreq_governor_mutex);
1296         list_del(&governor->governor_list);
1297         mutex_unlock(&cpufreq_governor_mutex);
1298         return;
1299 }
1300 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1301
1302
1303
1304 /*********************************************************************
1305  *                          POLICY INTERFACE                         *
1306  *********************************************************************/
1307
1308 /**
1309  * cpufreq_get_policy - get the current cpufreq_policy
1310  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1311  *
1312  * Reads the current cpufreq policy.
1313  */
1314 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1315 {
1316         struct cpufreq_policy *cpu_policy;
1317         if (!policy)
1318                 return -EINVAL;
1319
1320         cpu_policy = cpufreq_cpu_get(cpu);
1321         if (!cpu_policy)
1322                 return -EINVAL;
1323
1324         mutex_lock(&cpu_policy->lock);
1325         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1326         mutex_unlock(&cpu_policy->lock);
1327
1328         cpufreq_cpu_put(cpu_policy);
1329         return 0;
1330 }
1331 EXPORT_SYMBOL(cpufreq_get_policy);
1332
1333
1334 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1335 {
1336         int ret = 0;
1337
1338         cpufreq_debug_disable_ratelimit();
1339         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1340                 policy->min, policy->max);
1341
1342         memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
1343
1344         /* verify the cpu speed can be set within this limit */
1345         ret = cpufreq_driver->verify(policy);
1346         if (ret)
1347                 goto error_out;
1348
1349         down_read(&cpufreq_notifier_rwsem);
1350
1351         /* adjust if necessary - all reasons */
1352         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_ADJUST,
1353                             policy);
1354
1355         /* adjust if necessary - hardware incompatibility*/
1356         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_INCOMPATIBLE,
1357                             policy);
1358
1359         /* verify the cpu speed can be set within this limit,
1360            which might be different to the first one */
1361         ret = cpufreq_driver->verify(policy);
1362         if (ret) {
1363                 up_read(&cpufreq_notifier_rwsem);
1364                 goto error_out;
1365         }
1366
1367         /* notification of the new policy */
1368         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_NOTIFY,
1369                             policy);
1370
1371         up_read(&cpufreq_notifier_rwsem);
1372
1373         data->min = policy->min;
1374         data->max = policy->max;
1375
1376         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1377
1378         if (cpufreq_driver->setpolicy) {
1379                 data->policy = policy->policy;
1380                 dprintk("setting range\n");
1381                 ret = cpufreq_driver->setpolicy(policy);
1382         } else {
1383                 if (policy->governor != data->governor) {
1384                         /* save old, working values */
1385                         struct cpufreq_governor *old_gov = data->governor;
1386
1387                         dprintk("governor switch\n");
1388
1389                         /* end old governor */
1390                         if (data->governor)
1391                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1392
1393                         /* start new governor */
1394                         data->governor = policy->governor;
1395                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1396                                 /* new governor failed, so re-start old one */
1397                                 dprintk("starting governor %s failed\n", data->governor->name);
1398                                 if (old_gov) {
1399                                         data->governor = old_gov;
1400                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1401                                 }
1402                                 ret = -EINVAL;
1403                                 goto error_out;
1404                         }
1405                         /* might be a policy change, too, so fall through */
1406                 }
1407                 dprintk("governor: change or update limits\n");
1408                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1409         }
1410
1411 error_out:
1412         cpufreq_debug_enable_ratelimit();
1413         return ret;
1414 }
1415
1416 /**
1417  *      cpufreq_set_policy - set a new CPUFreq policy
1418  *      @policy: policy to be set.
1419  *
1420  *      Sets a new CPU frequency and voltage scaling policy.
1421  */
1422 int cpufreq_set_policy(struct cpufreq_policy *policy)
1423 {
1424         int ret = 0;
1425         struct cpufreq_policy *data;
1426
1427         if (!policy)
1428                 return -EINVAL;
1429
1430         data = cpufreq_cpu_get(policy->cpu);
1431         if (!data)
1432                 return -EINVAL;
1433
1434         /* lock this CPU */
1435         mutex_lock(&data->lock);
1436
1437         ret = __cpufreq_set_policy(data, policy);
1438         data->user_policy.min = data->min;
1439         data->user_policy.max = data->max;
1440         data->user_policy.policy = data->policy;
1441         data->user_policy.governor = data->governor;
1442
1443         mutex_unlock(&data->lock);
1444         cpufreq_cpu_put(data);
1445
1446         return ret;
1447 }
1448 EXPORT_SYMBOL(cpufreq_set_policy);
1449
1450
1451 /**
1452  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1453  *      @cpu: CPU which shall be re-evaluated
1454  *
1455  *      Usefull for policy notifiers which have different necessities
1456  *      at different times.
1457  */
1458 int cpufreq_update_policy(unsigned int cpu)
1459 {
1460         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1461         struct cpufreq_policy policy;
1462         int ret = 0;
1463
1464         if (!data)
1465                 return -ENODEV;
1466
1467         mutex_lock(&data->lock);
1468
1469         dprintk("updating policy for CPU %u\n", cpu);
1470         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1471         policy.min = data->user_policy.min;
1472         policy.max = data->user_policy.max;
1473         policy.policy = data->user_policy.policy;
1474         policy.governor = data->user_policy.governor;
1475
1476         /* BIOS might change freq behind our back
1477           -> ask driver for current freq and notify governors about a change */
1478         if (cpufreq_driver->get) {
1479                 policy.cur = cpufreq_driver->get(cpu);
1480                 if (!data->cur) {
1481                         dprintk("Driver did not initialize current freq");
1482                         data->cur = policy.cur;
1483                 } else {
1484                         if (data->cur != policy.cur)
1485                                 cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1486                 }
1487         }
1488
1489         ret = __cpufreq_set_policy(data, &policy);
1490
1491         mutex_unlock(&data->lock);
1492
1493         cpufreq_cpu_put(data);
1494         return ret;
1495 }
1496 EXPORT_SYMBOL(cpufreq_update_policy);
1497
1498 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1499                                         unsigned long action, void *hcpu)
1500 {
1501         unsigned int cpu = (unsigned long)hcpu;
1502         struct cpufreq_policy *policy;
1503         struct sys_device *sys_dev;
1504
1505         sys_dev = get_cpu_sysdev(cpu);
1506
1507         if (sys_dev) {
1508                 switch (action) {
1509                 case CPU_ONLINE:
1510                         cpufreq_add_dev(sys_dev);
1511                         break;
1512                 case CPU_DOWN_PREPARE:
1513                         /*
1514                          * We attempt to put this cpu in lowest frequency
1515                          * possible before going down. This will permit
1516                          * hardware-managed P-State to switch other related
1517                          * threads to min or higher speeds if possible.
1518                          */
1519                         policy = cpufreq_cpu_data[cpu];
1520                         if (policy) {
1521                                 cpufreq_driver_target(policy, policy->min,
1522                                                 CPUFREQ_RELATION_H);
1523                         }
1524                         break;
1525                 case CPU_DEAD:
1526                         cpufreq_remove_dev(sys_dev);
1527                         break;
1528                 }
1529         }
1530         return NOTIFY_OK;
1531 }
1532
1533 static struct notifier_block cpufreq_cpu_notifier =
1534 {
1535     .notifier_call = cpufreq_cpu_callback,
1536 };
1537
1538 /*********************************************************************
1539  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1540  *********************************************************************/
1541
1542 /**
1543  * cpufreq_register_driver - register a CPU Frequency driver
1544  * @driver_data: A struct cpufreq_driver containing the values#
1545  * submitted by the CPU Frequency driver.
1546  *
1547  *   Registers a CPU Frequency driver to this core code. This code
1548  * returns zero on success, -EBUSY when another driver got here first
1549  * (and isn't unregistered in the meantime).
1550  *
1551  */
1552 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1553 {
1554         unsigned long flags;
1555         int ret;
1556
1557         if (!driver_data || !driver_data->verify || !driver_data->init ||
1558             ((!driver_data->setpolicy) && (!driver_data->target)))
1559                 return -EINVAL;
1560
1561         dprintk("trying to register driver %s\n", driver_data->name);
1562
1563         if (driver_data->setpolicy)
1564                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1565
1566         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1567         if (cpufreq_driver) {
1568                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1569                 return -EBUSY;
1570         }
1571         cpufreq_driver = driver_data;
1572         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1573
1574         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1575
1576         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1577                 int i;
1578                 ret = -ENODEV;
1579
1580                 /* check for at least one working CPU */
1581                 for (i=0; i<NR_CPUS; i++)
1582                         if (cpufreq_cpu_data[i])
1583                                 ret = 0;
1584
1585                 /* if all ->init() calls failed, unregister */
1586                 if (ret) {
1587                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1588                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1589
1590                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1591                         cpufreq_driver = NULL;
1592                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1593                 }
1594         }
1595
1596         if (!ret) {
1597                 register_cpu_notifier(&cpufreq_cpu_notifier);
1598                 dprintk("driver %s up and running\n", driver_data->name);
1599                 cpufreq_debug_enable_ratelimit();
1600         }
1601
1602         return (ret);
1603 }
1604 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1605
1606
1607 /**
1608  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1609  *
1610  *    Unregister the current CPUFreq driver. Only call this if you have
1611  * the right to do so, i.e. if you have succeeded in initialising before!
1612  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1613  * currently not initialised.
1614  */
1615 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1616 {
1617         unsigned long flags;
1618
1619         cpufreq_debug_disable_ratelimit();
1620
1621         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1622                 cpufreq_debug_enable_ratelimit();
1623                 return -EINVAL;
1624         }
1625
1626         dprintk("unregistering driver %s\n", driver->name);
1627
1628         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1629         unregister_cpu_notifier(&cpufreq_cpu_notifier);
1630
1631         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1632         cpufreq_driver = NULL;
1633         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1634
1635         return 0;
1636 }
1637 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);