drivers/cpufreq/cpufreq.c

   1 /*
   2  *  linux/drivers/cpufreq/cpufreq.c
   3  *
   4  *  Copyright (C) 2001 Russell King
   5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
   6  *
   7  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
   8  *      Added handling for CPU hotplug
   9  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
  10  *      Fix handling for CPU hotplug -- affected CPUs
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17
  18 #include <linux/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/notifier.h>
  22 #include <linux/cpufreq.h>
  23 #include <linux/delay.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/device.h>
  27 #include <linux/slab.h>
  28 #include <linux/cpu.h>
  29 #include <linux/completion.h>
  30 #include <linux/mutex.h>
  31
  32 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
  33
  34 /**
  35  * The "cpufreq driver" - the arch- or hardware-dependend low
  36  * level driver of CPUFreq support, and its spinlock. This lock
  37  * also protects the cpufreq_cpu_data array.
  38  */
  39 static struct cpufreq_driver *cpufreq_driver;
  40 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
  41 static DEFINE_SPINLOCK(cpufreq_driver_lock);
  42
  43 /* internal prototypes */
  44 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
  45 static void handle_update(void *data);
  46
  47 /**
  48  * Two notifier lists: the "policy" list is involved in the
  49  * validation process for a new CPU frequency policy; the
  50  * "transition" list for kernel code that needs to handle
  51  * changes to devices when the CPU clock speed changes.
  52  * The mutex locks both lists.
  53  */
  54 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
  55 static BLOCKING_NOTIFIER_HEAD(cpufreq_transition_notifier_list);
  56
  57
  58 static LIST_HEAD(cpufreq_governor_list);
  59 static DEFINE_MUTEX (cpufreq_governor_mutex);
  60
  61 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
  62 {
  63         struct cpufreq_policy *data;
  64         unsigned long flags;
  65
  66         if (cpu >= NR_CPUS)
  67                 goto err_out;
  68
  69         /* get the cpufreq driver */
  70         spin_lock_irqsave(&cpufreq_driver_lock, flags);
  71
  72         if (!cpufreq_driver)
  73                 goto err_out_unlock;
  74
  75         if (!try_module_get(cpufreq_driver->owner))
  76                 goto err_out_unlock;
  77
  78
  79         /* get the CPU */
  80         data = cpufreq_cpu_data[cpu];
  81
  82         if (!data)
  83                 goto err_out_put_module;
  84
  85         if (!kobject_get(&data->kobj))
  86                 goto err_out_put_module;
  87
  88         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  89         return data;
  90
  91 err_out_put_module:
  92         module_put(cpufreq_driver->owner);
  93 err_out_unlock:
  94         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  95 err_out:
  96         return NULL;
  97 }
  98 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
  99
 100
 101 void cpufreq_cpu_put(struct cpufreq_policy *data)
 102 {
 103         kobject_put(&data->kobj);
 104         module_put(cpufreq_driver->owner);
 105 }
 106 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 107
 108
 109 /*********************************************************************
 110  *                     UNIFIED DEBUG HELPERS                         *
 111  *********************************************************************/
 112 #ifdef CONFIG_CPU_FREQ_DEBUG
 113
 114 /* what part(s) of the CPUfreq subsystem are debugged? */
 115 static unsigned int debug;
 116
 117 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 118  * set or modify this value.
 119  */
 120 static unsigned int debug_ratelimit = 1;
 121
 122 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 123  * loading of a cpufreq driver, temporarily disabled when a new policy
 124  * is set, and disabled upon cpufreq driver removal
 125  */
 126 static unsigned int disable_ratelimit = 1;
 127 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 128
 129 static void cpufreq_debug_enable_ratelimit(void)
 130 {
 131         unsigned long flags;
 132
 133         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 134         if (disable_ratelimit)
 135                 disable_ratelimit--;
 136         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 137 }
 138
 139 static void cpufreq_debug_disable_ratelimit(void)
 140 {
 141         unsigned long flags;
 142
 143         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 144         disable_ratelimit++;
 145         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 146 }
 147
 148 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
 149 {
 150         char s[256];
 151         va_list args;
 152         unsigned int len;
 153         unsigned long flags;
 154
 155         WARN_ON(!prefix);
 156         if (type & debug) {
 157                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 158                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
 159                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 160                         return;
 161                 }
 162                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 163
 164                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 165
 166                 va_start(args, fmt);
 167                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 168                 va_end(args);
 169
 170                 printk(s);
 171
 172                 WARN_ON(len < 5);
 173         }
 174 }
 175 EXPORT_SYMBOL(cpufreq_debug_printk);
 176
 177
 178 module_param(debug, uint, 0644);
 179 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
 180
 181 module_param(debug_ratelimit, uint, 0644);
 182 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
 183
 184 #else /* !CONFIG_CPU_FREQ_DEBUG */
 185
 186 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 187 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 188
 189 #endif /* CONFIG_CPU_FREQ_DEBUG */
 190
 191
 192 /*********************************************************************
 193  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 194  *********************************************************************/
 195
 196 /**
 197  * adjust_jiffies - adjust the system "loops_per_jiffy"
 198  *
 199  * This function alters the system "loops_per_jiffy" for the clock
 200  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 201  * systems as each CPU might be scaled differently. So, use the arch
 202  * per-CPU loops_per_jiffy value wherever possible.
 203  */
 204 #ifndef CONFIG_SMP
 205 static unsigned long l_p_j_ref;
 206 static unsigned int  l_p_j_ref_freq;
 207
 208 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 209 {
 210         if (ci->flags & CPUFREQ_CONST_LOOPS)
 211                 return;
 212
 213         if (!l_p_j_ref_freq) {
 214                 l_p_j_ref = loops_per_jiffy;
 215                 l_p_j_ref_freq = ci->old;
 216                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 217         }
 218         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 219             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 220             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 221                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
 222                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
 223         }
 224 }
 225 #else
 226 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
 227 #endif
 228
 229
 230 /**
 231  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 232  * on frequency transition.
 233  *
 234  * This function calls the transition notifiers and the "adjust_jiffies"
 235  * function. It is called twice on all CPU frequency changes that have
 236  * external effects.
 237  */
 238 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 239 {
 240         struct cpufreq_policy *policy;
 241
 242         BUG_ON(irqs_disabled());
 243
 244         freqs->flags = cpufreq_driver->flags;
 245         dprintk("notification %u of frequency transition to %u kHz\n",
 246                 state, freqs->new);
 247
 248         policy = cpufreq_cpu_data[freqs->cpu];
 249         switch (state) {
 250
 251         case CPUFREQ_PRECHANGE:
 252                 /* detect if the driver reported a value as "old frequency"
 253                  * which is not equal to what the cpufreq core thinks is
 254                  * "old frequency".
 255                  */
 256                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 257                         if ((policy) && (policy->cpu == freqs->cpu) &&
 258                             (policy->cur) && (policy->cur != freqs->old)) {
 259                                 dprintk("Warning: CPU frequency is"
 260                                         " %u, cpufreq assumed %u kHz.\n",
 261                                         freqs->old, policy->cur);
 262                                 freqs->old = policy->cur;
 263                         }
 264                 }
 265                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
 266                                 CPUFREQ_PRECHANGE, freqs);
 267                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 268                 break;
 269
 270         case CPUFREQ_POSTCHANGE:
 271                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 272                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
 273                                 CPUFREQ_POSTCHANGE, freqs);
 274                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 275                         policy->cur = freqs->new;
 276                 break;
 277         }
 278 }
 279 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 280
 281
 282
 283 /*********************************************************************
 284  *                          SYSFS INTERFACE                          *
 285  *********************************************************************/
 286
 287 /**
 288  * cpufreq_parse_governor - parse a governor string
 289  */
 290 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 291                                 struct cpufreq_governor **governor)
 292 {
 293         if (!cpufreq_driver)
 294                 return -EINVAL;
 295         if (cpufreq_driver->setpolicy) {
 296                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 297                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 298                         return 0;
 299                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 300                         *policy = CPUFREQ_POLICY_POWERSAVE;
 301                         return 0;
 302                 }
 303                 return -EINVAL;
 304         } else {
 305                 struct cpufreq_governor *t;
 306                 mutex_lock(&cpufreq_governor_mutex);
 307                 if (!cpufreq_driver || !cpufreq_driver->target)
 308                         goto out;
 309                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 310                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
 311                                 *governor = t;
 312                                 mutex_unlock(&cpufreq_governor_mutex);
 313                                 return 0;
 314                         }
 315                 }
 316 out:
 317                 mutex_unlock(&cpufreq_governor_mutex);
 318         }
 319         return -EINVAL;
 320 }
 321
 322
 323 /* drivers/base/cpu.c */
 324 extern struct sysdev_class cpu_sysdev_class;
 325
 326
 327 /**
 328  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 329  *
 330  * Write out information from cpufreq_driver->policy[cpu]; object must be
 331  * "unsigned int".
 332  */
 333
 334 #define show_one(file_name, object)                     \
 335 static ssize_t show_##file_name                         \
 336 (struct cpufreq_policy * policy, char *buf)             \
 337 {                                                       \
 338         return sprintf (buf, "%u\n", policy->object);   \
 339 }
 340
 341 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 342 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 343 show_one(scaling_min_freq, min);
 344 show_one(scaling_max_freq, max);
 345 show_one(scaling_cur_freq, cur);
 346
 347 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
 348
 349 /**
 350  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 351  */
 352 #define store_one(file_name, object)                    \
 353 static ssize_t store_##file_name                                        \
 354 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 355 {                                                                       \
 356         unsigned int ret = -EINVAL;                                     \
 357         struct cpufreq_policy new_policy;                               \
 358                                                                         \
 359         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 360         if (ret)                                                        \
 361                 return -EINVAL;                                         \
 362                                                                         \
 363         ret = sscanf (buf, "%u", &new_policy.object);                   \
 364         if (ret != 1)                                                   \
 365                 return -EINVAL;                                         \
 366                                                                         \
 367         mutex_lock(&policy->lock);                                      \
 368         ret = __cpufreq_set_policy(policy, &new_policy);                \
 369         policy->user_policy.object = policy->object;                    \
 370         mutex_unlock(&policy->lock);                                    \
 371                                                                         \
 372         return ret ? ret : count;                                       \
 373 }
 374
 375 store_one(scaling_min_freq,min);
 376 store_one(scaling_max_freq,max);
 377
 378 /**
 379  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 380  */
 381 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 382 {
 383         unsigned int cur_freq = cpufreq_get(policy->cpu);
 384         if (!cur_freq)
 385                 return sprintf(buf, "<unknown>");
 386         return sprintf(buf, "%u\n", cur_freq);
 387 }
 388
 389
 390 /**
 391  * show_scaling_governor - show the current policy for the specified CPU
 392  */
 393 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 394 {
 395         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 396                 return sprintf(buf, "powersave\n");
 397         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 398                 return sprintf(buf, "performance\n");
 399         else if (policy->governor)
 400                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 401         return -EINVAL;
 402 }
 403
 404
 405 /**
 406  * store_scaling_governor - store policy for the specified CPU
 407  */
 408 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 409                                        const char *buf, size_t count)
 410 {
 411         unsigned int ret = -EINVAL;
 412         char    str_governor[16];
 413         struct cpufreq_policy new_policy;
 414
 415         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 416         if (ret)
 417                 return ret;
 418
 419         ret = sscanf (buf, "%15s", str_governor);
 420         if (ret != 1)
 421                 return -EINVAL;
 422
 423         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 424                 return -EINVAL;
 425
 426         lock_cpu_hotplug();
 427
 428         /* Do not use cpufreq_set_policy here or the user_policy.max
 429            will be wrongly overridden */
 430         mutex_lock(&policy->lock);
 431         ret = __cpufreq_set_policy(policy, &new_policy);
 432
 433         policy->user_policy.policy = policy->policy;
 434         policy->user_policy.governor = policy->governor;
 435         mutex_unlock(&policy->lock);
 436
 437         unlock_cpu_hotplug();
 438
 439         return ret ? ret : count;
 440 }
 441
 442 /**
 443  * show_scaling_driver - show the cpufreq driver currently loaded
 444  */
 445 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 446 {
 447         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 448 }
 449
 450 /**
 451  * show_scaling_available_governors - show the available CPUfreq governors
 452  */
 453 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 454                                 char *buf)
 455 {
 456         ssize_t i = 0;
 457         struct cpufreq_governor *t;
 458
 459         if (!cpufreq_driver->target) {
 460                 i += sprintf(buf, "performance powersave");
 461                 goto out;
 462         }
 463
 464         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 465                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 466                         goto out;
 467                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 468         }
 469 out:
 470         i += sprintf(&buf[i], "\n");
 471         return i;
 472 }
 473 /**
 474  * show_affected_cpus - show the CPUs affected by each transition
 475  */
 476 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 477 {
 478         ssize_t i = 0;
 479         unsigned int cpu;
 480
 481         for_each_cpu_mask(cpu, policy->cpus) {
 482                 if (i)
 483                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 484                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 485                 if (i >= (PAGE_SIZE - 5))
 486                     break;
 487         }
 488         i += sprintf(&buf[i], "\n");
 489         return i;
 490 }
 491
 492
 493 #define define_one_ro(_name) \
 494 static struct freq_attr _name = \
 495 __ATTR(_name, 0444, show_##_name, NULL)
 496
 497 #define define_one_ro0400(_name) \
 498 static struct freq_attr _name = \
 499 __ATTR(_name, 0400, show_##_name, NULL)
 500
 501 #define define_one_rw(_name) \
 502 static struct freq_attr _name = \
 503 __ATTR(_name, 0644, show_##_name, store_##_name)
 504
 505 define_one_ro0400(cpuinfo_cur_freq);
 506 define_one_ro(cpuinfo_min_freq);
 507 define_one_ro(cpuinfo_max_freq);
 508 define_one_ro(scaling_available_governors);
 509 define_one_ro(scaling_driver);
 510 define_one_ro(scaling_cur_freq);
 511 define_one_ro(affected_cpus);
 512 define_one_rw(scaling_min_freq);
 513 define_one_rw(scaling_max_freq);
 514 define_one_rw(scaling_governor);
 515
 516 static struct attribute * default_attrs[] = {
 517         &cpuinfo_min_freq.attr,
 518         &cpuinfo_max_freq.attr,
 519         &scaling_min_freq.attr,
 520         &scaling_max_freq.attr,
 521         &affected_cpus.attr,
 522         &scaling_governor.attr,
 523         &scaling_driver.attr,
 524         &scaling_available_governors.attr,
 525         NULL
 526 };
 527
 528 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 529 #define to_attr(a) container_of(a,struct freq_attr,attr)
 530
 531 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 532 {
 533         struct cpufreq_policy * policy = to_policy(kobj);
 534         struct freq_attr * fattr = to_attr(attr);
 535         ssize_t ret;
 536         policy = cpufreq_cpu_get(policy->cpu);
 537         if (!policy)
 538                 return -EINVAL;
 539         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 540         cpufreq_cpu_put(policy);
 541         return ret;
 542 }
 543
 544 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 545                      const char * buf, size_t count)
 546 {
 547         struct cpufreq_policy * policy = to_policy(kobj);
 548         struct freq_attr * fattr = to_attr(attr);
 549         ssize_t ret;
 550         policy = cpufreq_cpu_get(policy->cpu);
 551         if (!policy)
 552                 return -EINVAL;
 553         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 554         cpufreq_cpu_put(policy);
 555         return ret;
 556 }
 557
 558 static void cpufreq_sysfs_release(struct kobject * kobj)
 559 {
 560         struct cpufreq_policy * policy = to_policy(kobj);
 561         dprintk("last reference is dropped\n");
 562         complete(&policy->kobj_unregister);
 563 }
 564
 565 static struct sysfs_ops sysfs_ops = {
 566         .show   = show,
 567         .store  = store,
 568 };
 569
 570 static struct kobj_type ktype_cpufreq = {
 571         .sysfs_ops      = &sysfs_ops,
 572         .default_attrs  = default_attrs,
 573         .release        = cpufreq_sysfs_release,
 574 };
 575
 576
 577 /**
 578  * cpufreq_add_dev - add a CPU device
 579  *
 580  * Adds the cpufreq interface for a CPU device.
 581  */
 582 static int cpufreq_add_dev (struct sys_device * sys_dev)
 583 {
 584         unsigned int cpu = sys_dev->id;
 585         int ret = 0;
 586         struct cpufreq_policy new_policy;
 587         struct cpufreq_policy *policy;
 588         struct freq_attr **drv_attr;
 589         struct sys_device *cpu_sys_dev;
 590         unsigned long flags;
 591         unsigned int j;
 592 #ifdef CONFIG_SMP
 593         struct cpufreq_policy *managed_policy;
 594 #endif
 595
 596         if (cpu_is_offline(cpu))
 597                 return 0;
 598
 599         cpufreq_debug_disable_ratelimit();
 600         dprintk("adding CPU %u\n", cpu);
 601
 602 #ifdef CONFIG_SMP
 603         /* check whether a different CPU already registered this
 604          * CPU because it is in the same boat. */
 605         policy = cpufreq_cpu_get(cpu);
 606         if (unlikely(policy)) {
 607                 cpufreq_cpu_put(policy);
 608                 cpufreq_debug_enable_ratelimit();
 609                 return 0;
 610         }
 611 #endif
 612
 613         if (!try_module_get(cpufreq_driver->owner)) {
 614                 ret = -EINVAL;
 615                 goto module_out;
 616         }
 617
 618         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 619         if (!policy) {
 620                 ret = -ENOMEM;
 621                 goto nomem_out;
 622         }
 623
 624         policy->cpu = cpu;
 625         policy->cpus = cpumask_of_cpu(cpu);
 626
 627         mutex_init(&policy->lock);
 628         mutex_lock(&policy->lock);
 629         init_completion(&policy->kobj_unregister);
 630         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 631
 632         /* call driver. From then on the cpufreq must be able
 633          * to accept all calls to ->verify and ->setpolicy for this CPU
 634          */
 635         ret = cpufreq_driver->init(policy);
 636         if (ret) {
 637                 dprintk("initialization failed\n");
 638                 mutex_unlock(&policy->lock);
 639                 goto err_out;
 640         }
 641
 642 #ifdef CONFIG_SMP
 643         for_each_cpu_mask(j, policy->cpus) {
 644                 if (cpu == j)
 645                         continue;
 646
 647                 /* check for existing affected CPUs.  They may not be aware
 648                  * of it due to CPU Hotplug.
 649                  */
 650                 managed_policy = cpufreq_cpu_get(j);
 651                 if (unlikely(managed_policy)) {
 652                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 653                         managed_policy->cpus = policy->cpus;
 654                         cpufreq_cpu_data[cpu] = managed_policy;
 655                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 656
 657                         dprintk("CPU already managed, adding link\n");
 658                         sysfs_create_link(&sys_dev->kobj,
 659                                           &managed_policy->kobj, "cpufreq");
 660
 661                         cpufreq_debug_enable_ratelimit();
 662                         mutex_unlock(&policy->lock);
 663                         ret = 0;
 664                         goto err_out_driver_exit; /* call driver->exit() */
 665                 }
 666         }
 667 #endif
 668         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 669
 670         /* prepare interface data */
 671         policy->kobj.parent = &sys_dev->kobj;
 672         policy->kobj.ktype = &ktype_cpufreq;
 673         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 674
 675         ret = kobject_register(&policy->kobj);
 676         if (ret) {
 677                 mutex_unlock(&policy->lock);
 678                 goto err_out_driver_exit;
 679         }
 680         /* set up files for this cpu device */
 681         drv_attr = cpufreq_driver->attr;
 682         while ((drv_attr) && (*drv_attr)) {
 683                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 684                 drv_attr++;
 685         }
 686         if (cpufreq_driver->get)
 687                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 688         if (cpufreq_driver->target)
 689                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 690
 691         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 692         for_each_cpu_mask(j, policy->cpus)
 693                 cpufreq_cpu_data[j] = policy;
 694         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 695
 696         /* symlink affected CPUs */
 697         for_each_cpu_mask(j, policy->cpus) {
 698                 if (j == cpu)
 699                         continue;
 700                 if (!cpu_online(j))
 701                         continue;
 702
 703                 dprintk("CPU %u already managed, adding link\n", j);
 704                 cpufreq_cpu_get(cpu);
 705                 cpu_sys_dev = get_cpu_sysdev(j);
 706                 sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 707                                   "cpufreq");
 708         }
 709
 710         policy->governor = NULL; /* to assure that the starting sequence is
 711                                   * run in cpufreq_set_policy */
 712         mutex_unlock(&policy->lock);
 713
 714         /* set default policy */
 715         ret = cpufreq_set_policy(&new_policy);
 716         if (ret) {
 717                 dprintk("setting policy failed\n");
 718                 goto err_out_unregister;
 719         }
 720
 721         module_put(cpufreq_driver->owner);
 722         dprintk("initialization complete\n");
 723         cpufreq_debug_enable_ratelimit();
 724
 725         return 0;
 726
 727
 728 err_out_unregister:
 729         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 730         for_each_cpu_mask(j, policy->cpus)
 731                 cpufreq_cpu_data[j] = NULL;
 732         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 733
 734         kobject_unregister(&policy->kobj);
 735         wait_for_completion(&policy->kobj_unregister);
 736
 737 err_out_driver_exit:
 738         if (cpufreq_driver->exit)
 739                 cpufreq_driver->exit(policy);
 740
 741 err_out:
 742         kfree(policy);
 743
 744 nomem_out:
 745         module_put(cpufreq_driver->owner);
 746 module_out:
 747         cpufreq_debug_enable_ratelimit();
 748         return ret;
 749 }
 750
 751
 752 /**
 753  * cpufreq_remove_dev - remove a CPU device
 754  *
 755  * Removes the cpufreq interface for a CPU device.
 756  */
 757 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 758 {
 759         unsigned int cpu = sys_dev->id;
 760         unsigned long flags;
 761         struct cpufreq_policy *data;
 762 #ifdef CONFIG_SMP
 763         struct sys_device *cpu_sys_dev;
 764         unsigned int j;
 765 #endif
 766
 767         cpufreq_debug_disable_ratelimit();
 768         dprintk("unregistering CPU %u\n", cpu);
 769
 770         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 771         data = cpufreq_cpu_data[cpu];
 772
 773         if (!data) {
 774                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 775                 cpufreq_debug_enable_ratelimit();
 776                 return -EINVAL;
 777         }
 778         cpufreq_cpu_data[cpu] = NULL;
 779
 780
 781 #ifdef CONFIG_SMP
 782         /* if this isn't the CPU which is the parent of the kobj, we
 783          * only need to unlink, put and exit
 784          */
 785         if (unlikely(cpu != data->cpu)) {
 786                 dprintk("removing link\n");
 787                 cpu_clear(cpu, data->cpus);
 788                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 789                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 790                 cpufreq_cpu_put(data);
 791                 cpufreq_debug_enable_ratelimit();
 792                 return 0;
 793         }
 794 #endif
 795
 796
 797         if (!kobject_get(&data->kobj)) {
 798                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 799                 cpufreq_debug_enable_ratelimit();
 800                 return -EFAULT;
 801         }
 802
 803 #ifdef CONFIG_SMP
 804         /* if we have other CPUs still registered, we need to unlink them,
 805          * or else wait_for_completion below will lock up. Clean the
 806          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 807          * links afterwards.
 808          */
 809         if (unlikely(cpus_weight(data->cpus) > 1)) {
 810                 for_each_cpu_mask(j, data->cpus) {
 811                         if (j == cpu)
 812                                 continue;
 813                         cpufreq_cpu_data[j] = NULL;
 814                 }
 815         }
 816
 817         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 818
 819         if (unlikely(cpus_weight(data->cpus) > 1)) {
 820                 for_each_cpu_mask(j, data->cpus) {
 821                         if (j == cpu)
 822                                 continue;
 823                         dprintk("removing link for cpu %u\n", j);
 824                         cpu_sys_dev = get_cpu_sysdev(j);
 825                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 826                         cpufreq_cpu_put(data);
 827                 }
 828         }
 829 #else
 830         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 831 #endif
 832
 833         mutex_lock(&data->lock);
 834         if (cpufreq_driver->target)
 835                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 836         mutex_unlock(&data->lock);
 837
 838         kobject_unregister(&data->kobj);
 839
 840         kobject_put(&data->kobj);
 841
 842         /* we need to make sure that the underlying kobj is actually
 843          * not referenced anymore by anybody before we proceed with
 844          * unloading.
 845          */
 846         dprintk("waiting for dropping of refcount\n");
 847         wait_for_completion(&data->kobj_unregister);
 848         dprintk("wait complete\n");
 849
 850         if (cpufreq_driver->exit)
 851                 cpufreq_driver->exit(data);
 852
 853         kfree(data);
 854
 855         cpufreq_debug_enable_ratelimit();
 856         return 0;
 857 }
 858
 859
 860 static void handle_update(void *data)
 861 {
 862         unsigned int cpu = (unsigned int)(long)data;
 863         dprintk("handle_update for cpu %u called\n", cpu);
 864         cpufreq_update_policy(cpu);
 865 }
 866
 867 /**
 868  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 869  *      @cpu: cpu number
 870  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 871  *      @new_freq: CPU frequency the CPU actually runs at
 872  *
 873  *      We adjust to current frequency first, and need to clean up later. So either call
 874  *      to cpufreq_update_policy() or schedule handle_update()).
 875  */
 876 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 877 {
 878         struct cpufreq_freqs freqs;
 879
 880         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
 881                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 882
 883         freqs.cpu = cpu;
 884         freqs.old = old_freq;
 885         freqs.new = new_freq;
 886         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 887         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 888 }
 889
 890
 891 /**
 892  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
 893  * @cpu: CPU number
 894  *
 895  * This is the last known freq, without actually getting it from the driver.
 896  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 897  */
 898 unsigned int cpufreq_quick_get(unsigned int cpu)
 899 {
 900         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 901         unsigned int ret = 0;
 902
 903         if (policy) {
 904                 mutex_lock(&policy->lock);
 905                 ret = policy->cur;
 906                 mutex_unlock(&policy->lock);
 907                 cpufreq_cpu_put(policy);
 908         }
 909
 910         return (ret);
 911 }
 912 EXPORT_SYMBOL(cpufreq_quick_get);
 913
 914
 915 /**
 916  * cpufreq_get - get the current CPU frequency (in kHz)
 917  * @cpu: CPU number
 918  *
 919  * Get the CPU current (static) CPU frequency
 920  */
 921 unsigned int cpufreq_get(unsigned int cpu)
 922 {
 923         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 924         unsigned int ret = 0;
 925
 926         if (!policy)
 927                 return 0;
 928
 929         if (!cpufreq_driver->get)
 930                 goto out;
 931
 932         mutex_lock(&policy->lock);
 933
 934         ret = cpufreq_driver->get(cpu);
 935
 936         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 937                 /* verify no discrepancy between actual and saved value exists */
 938                 if (unlikely(ret != policy->cur)) {
 939                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 940                         schedule_work(&policy->update);
 941                 }
 942         }
 943
 944         mutex_unlock(&policy->lock);
 945
 946 out:
 947         cpufreq_cpu_put(policy);
 948
 949         return (ret);
 950 }
 951 EXPORT_SYMBOL(cpufreq_get);
 952
 953
 954 /**
 955  *      cpufreq_suspend - let the low level driver prepare for suspend
 956  */
 957
 958 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 959 {
 960         int cpu = sysdev->id;
 961         unsigned int ret = 0;
 962         unsigned int cur_freq = 0;
 963         struct cpufreq_policy *cpu_policy;
 964
 965         dprintk("resuming cpu %u\n", cpu);
 966
 967         if (!cpu_online(cpu))
 968                 return 0;
 969
 970         /* we may be lax here as interrupts are off. Nonetheless
 971          * we need to grab the correct cpu policy, as to check
 972          * whether we really run on this CPU.
 973          */
 974
 975         cpu_policy = cpufreq_cpu_get(cpu);
 976         if (!cpu_policy)
 977                 return -EINVAL;
 978
 979         /* only handle each CPU group once */
 980         if (unlikely(cpu_policy->cpu != cpu)) {
 981                 cpufreq_cpu_put(cpu_policy);
 982                 return 0;
 983         }
 984
 985         if (cpufreq_driver->suspend) {
 986                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
 987                 if (ret) {
 988                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
 989                                         "step on CPU %u\n", cpu_policy->cpu);
 990                         cpufreq_cpu_put(cpu_policy);
 991                         return ret;
 992                 }
 993         }
 994
 995
 996         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
 997                 goto out;
 998
 999         if (cpufreq_driver->get)
1000                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1001
1002         if (!cur_freq || !cpu_policy->cur) {
1003                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1004                        "frequency is what timing core thinks it is.\n");
1005                 goto out;
1006         }
1007
1008         if (unlikely(cur_freq != cpu_policy->cur)) {
1009                 struct cpufreq_freqs freqs;
1010
1011                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1012                         dprintk("Warning: CPU frequency is %u, "
1013                                "cpufreq assumed %u kHz.\n",
1014                                cur_freq, cpu_policy->cur);
1015
1016                 freqs.cpu = cpu;
1017                 freqs.old = cpu_policy->cur;
1018                 freqs.new = cur_freq;
1019
1020                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
1021                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1022                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1023
1024                 cpu_policy->cur = cur_freq;
1025         }
1026
1027 out:
1028         cpufreq_cpu_put(cpu_policy);
1029         return 0;
1030 }
1031
1032 /**
1033  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1034  *
1035  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1036  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1037  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1038  *          restored.
1039  */
1040 static int cpufreq_resume(struct sys_device * sysdev)
1041 {
1042         int cpu = sysdev->id;
1043         unsigned int ret = 0;
1044         struct cpufreq_policy *cpu_policy;
1045
1046         dprintk("resuming cpu %u\n", cpu);
1047
1048         if (!cpu_online(cpu))
1049                 return 0;
1050
1051         /* we may be lax here as interrupts are off. Nonetheless
1052          * we need to grab the correct cpu policy, as to check
1053          * whether we really run on this CPU.
1054          */
1055
1056         cpu_policy = cpufreq_cpu_get(cpu);
1057         if (!cpu_policy)
1058                 return -EINVAL;
1059
1060         /* only handle each CPU group once */
1061         if (unlikely(cpu_policy->cpu != cpu)) {
1062                 cpufreq_cpu_put(cpu_policy);
1063                 return 0;
1064         }
1065
1066         if (cpufreq_driver->resume) {
1067                 ret = cpufreq_driver->resume(cpu_policy);
1068                 if (ret) {
1069                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1070                                         "step on CPU %u\n", cpu_policy->cpu);
1071                         cpufreq_cpu_put(cpu_policy);
1072                         return ret;
1073                 }
1074         }
1075
1076         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1077                 unsigned int cur_freq = 0;
1078
1079                 if (cpufreq_driver->get)
1080                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1081
1082                 if (!cur_freq || !cpu_policy->cur) {
1083                         printk(KERN_ERR "cpufreq: resume failed to assert "
1084                                         "current frequency is what timing core "
1085                                         "thinks it is.\n");
1086                         goto out;
1087                 }
1088
1089                 if (unlikely(cur_freq != cpu_policy->cur)) {
1090                         struct cpufreq_freqs freqs;
1091
1092                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1093                                 dprintk("Warning: CPU frequency"
1094                                        "is %u, cpufreq assumed %u kHz.\n",
1095                                        cur_freq, cpu_policy->cur);
1096
1097                         freqs.cpu = cpu;
1098                         freqs.old = cpu_policy->cur;
1099                         freqs.new = cur_freq;
1100
1101                         blocking_notifier_call_chain(
1102                                         &cpufreq_transition_notifier_list,
1103                                         CPUFREQ_RESUMECHANGE, &freqs);
1104                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1105
1106                         cpu_policy->cur = cur_freq;
1107                 }
1108         }
1109
1110 out:
1111         schedule_work(&cpu_policy->update);
1112         cpufreq_cpu_put(cpu_policy);
1113         return ret;
1114 }
1115
1116 static struct sysdev_driver cpufreq_sysdev_driver = {
1117         .add            = cpufreq_add_dev,
1118         .remove         = cpufreq_remove_dev,
1119         .suspend        = cpufreq_suspend,
1120         .resume         = cpufreq_resume,
1121 };
1122
1123
1124 /*********************************************************************
1125  *                     NOTIFIER LISTS INTERFACE                      *
1126  *********************************************************************/
1127
1128 /**
1129  *      cpufreq_register_notifier - register a driver with cpufreq
1130  *      @nb: notifier function to register
1131  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1132  *
1133  *      Add a driver to one of two lists: either a list of drivers that
1134  *      are notified about clock rate changes (once before and once after
1135  *      the transition), or a list of drivers that are notified about
1136  *      changes in cpufreq policy.
1137  *
1138  *      This function may sleep, and has the same return conditions as
1139  *      blocking_notifier_chain_register.
1140  */
1141 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1142 {
1143         int ret;
1144
1145         switch (list) {
1146         case CPUFREQ_TRANSITION_NOTIFIER:
1147                 ret = blocking_notifier_chain_register(
1148                                 &cpufreq_transition_notifier_list, nb);
1149                 break;
1150         case CPUFREQ_POLICY_NOTIFIER:
1151                 ret = blocking_notifier_chain_register(
1152                                 &cpufreq_policy_notifier_list, nb);
1153                 break;
1154         default:
1155                 ret = -EINVAL;
1156         }
1157
1158         return ret;
1159 }
1160 EXPORT_SYMBOL(cpufreq_register_notifier);
1161
1162
1163 /**
1164  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1165  *      @nb: notifier block to be unregistered
1166  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1167  *
1168  *      Remove a driver from the CPU frequency notifier list.
1169  *
1170  *      This function may sleep, and has the same return conditions as
1171  *      blocking_notifier_chain_unregister.
1172  */
1173 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1174 {
1175         int ret;
1176
1177         switch (list) {
1178         case CPUFREQ_TRANSITION_NOTIFIER:
1179                 ret = blocking_notifier_chain_unregister(
1180                                 &cpufreq_transition_notifier_list, nb);
1181                 break;
1182         case CPUFREQ_POLICY_NOTIFIER:
1183                 ret = blocking_notifier_chain_unregister(
1184                                 &cpufreq_policy_notifier_list, nb);
1185                 break;
1186         default:
1187                 ret = -EINVAL;
1188         }
1189
1190         return ret;
1191 }
1192 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1193
1194
1195 /*********************************************************************
1196  *                              GOVERNORS                            *
1197  *********************************************************************/
1198
1199
1200 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1201                             unsigned int target_freq,
1202                             unsigned int relation)
1203 {
1204         int retval = -EINVAL;
1205
1206         lock_cpu_hotplug();
1207         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1208                 target_freq, relation);
1209         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1210                 retval = cpufreq_driver->target(policy, target_freq, relation);
1211
1212         unlock_cpu_hotplug();
1213
1214         return retval;
1215 }
1216 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1217
1218 int cpufreq_driver_target(struct cpufreq_policy *policy,
1219                           unsigned int target_freq,
1220                           unsigned int relation)
1221 {
1222         int ret;
1223
1224         policy = cpufreq_cpu_get(policy->cpu);
1225         if (!policy)
1226                 return -EINVAL;
1227
1228         mutex_lock(&policy->lock);
1229
1230         ret = __cpufreq_driver_target(policy, target_freq, relation);
1231
1232         mutex_unlock(&policy->lock);
1233
1234         cpufreq_cpu_put(policy);
1235         return ret;
1236 }
1237 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1238
1239
1240 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1241 {
1242         int ret;
1243
1244         if (!try_module_get(policy->governor->owner))
1245                 return -EINVAL;
1246
1247         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1248         ret = policy->governor->governor(policy, event);
1249
1250         /* we keep one module reference alive for each CPU governed by this CPU */
1251         if ((event != CPUFREQ_GOV_START) || ret)
1252                 module_put(policy->governor->owner);
1253         if ((event == CPUFREQ_GOV_STOP) && !ret)
1254                 module_put(policy->governor->owner);
1255
1256         return ret;
1257 }
1258
1259
1260 int cpufreq_governor(unsigned int cpu, unsigned int event)
1261 {
1262         int ret = 0;
1263         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1264
1265         if (!policy)
1266                 return -EINVAL;
1267
1268         mutex_lock(&policy->lock);
1269         ret = __cpufreq_governor(policy, event);
1270         mutex_unlock(&policy->lock);
1271
1272         cpufreq_cpu_put(policy);
1273         return ret;
1274 }
1275 EXPORT_SYMBOL_GPL(cpufreq_governor);
1276
1277
1278 int cpufreq_register_governor(struct cpufreq_governor *governor)
1279 {
1280         struct cpufreq_governor *t;
1281
1282         if (!governor)
1283                 return -EINVAL;
1284
1285         mutex_lock(&cpufreq_governor_mutex);
1286
1287         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1288                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1289                         mutex_unlock(&cpufreq_governor_mutex);
1290                         return -EBUSY;
1291                 }
1292         }
1293         list_add(&governor->governor_list, &cpufreq_governor_list);
1294
1295         mutex_unlock(&cpufreq_governor_mutex);
1296         return 0;
1297 }
1298 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1299
1300
1301 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1302 {
1303         if (!governor)
1304                 return;
1305
1306         mutex_lock(&cpufreq_governor_mutex);
1307         list_del(&governor->governor_list);
1308         mutex_unlock(&cpufreq_governor_mutex);
1309         return;
1310 }
1311 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1312
1313
1314
1315 /*********************************************************************
1316  *                          POLICY INTERFACE                         *
1317  *********************************************************************/
1318
1319 /**
1320  * cpufreq_get_policy - get the current cpufreq_policy
1321  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1322  *
1323  * Reads the current cpufreq policy.
1324  */
1325 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1326 {
1327         struct cpufreq_policy *cpu_policy;
1328         if (!policy)
1329                 return -EINVAL;
1330
1331         cpu_policy = cpufreq_cpu_get(cpu);
1332         if (!cpu_policy)
1333                 return -EINVAL;
1334
1335         mutex_lock(&cpu_policy->lock);
1336         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1337         mutex_unlock(&cpu_policy->lock);
1338
1339         cpufreq_cpu_put(cpu_policy);
1340         return 0;
1341 }
1342 EXPORT_SYMBOL(cpufreq_get_policy);
1343
1344
1345 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1346 {
1347         int ret = 0;
1348
1349         cpufreq_debug_disable_ratelimit();
1350         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1351                 policy->min, policy->max);
1352
1353         memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
1354
1355         /* verify the cpu speed can be set within this limit */
1356         ret = cpufreq_driver->verify(policy);
1357         if (ret)
1358                 goto error_out;
1359
1360         /* adjust if necessary - all reasons */
1361         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1362                         CPUFREQ_ADJUST, policy);
1363
1364         /* adjust if necessary - hardware incompatibility*/
1365         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1366                         CPUFREQ_INCOMPATIBLE, policy);
1367
1368         /* verify the cpu speed can be set within this limit,
1369            which might be different to the first one */
1370         ret = cpufreq_driver->verify(policy);
1371         if (ret)
1372                 goto error_out;
1373
1374         /* notification of the new policy */
1375         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1376                         CPUFREQ_NOTIFY, policy);
1377
1378         data->min = policy->min;
1379         data->max = policy->max;
1380
1381         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1382
1383         if (cpufreq_driver->setpolicy) {
1384                 data->policy = policy->policy;
1385                 dprintk("setting range\n");
1386                 ret = cpufreq_driver->setpolicy(policy);
1387         } else {
1388                 if (policy->governor != data->governor) {
1389                         /* save old, working values */
1390                         struct cpufreq_governor *old_gov = data->governor;
1391
1392                         dprintk("governor switch\n");
1393
1394                         /* end old governor */
1395                         if (data->governor)
1396                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1397
1398                         /* start new governor */
1399                         data->governor = policy->governor;
1400                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1401                                 /* new governor failed, so re-start old one */
1402                                 dprintk("starting governor %s failed\n", data->governor->name);
1403                                 if (old_gov) {
1404                                         data->governor = old_gov;
1405                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1406                                 }
1407                                 ret = -EINVAL;
1408                                 goto error_out;
1409                         }
1410                         /* might be a policy change, too, so fall through */
1411                 }
1412                 dprintk("governor: change or update limits\n");
1413                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1414         }
1415
1416 error_out:
1417         cpufreq_debug_enable_ratelimit();
1418         return ret;
1419 }
1420
1421 /**
1422  *      cpufreq_set_policy - set a new CPUFreq policy
1423  *      @policy: policy to be set.
1424  *
1425  *      Sets a new CPU frequency and voltage scaling policy.
1426  */
1427 int cpufreq_set_policy(struct cpufreq_policy *policy)
1428 {
1429         int ret = 0;
1430         struct cpufreq_policy *data;
1431
1432         if (!policy)
1433                 return -EINVAL;
1434
1435         data = cpufreq_cpu_get(policy->cpu);
1436         if (!data)
1437                 return -EINVAL;
1438
1439         /* lock this CPU */
1440         mutex_lock(&data->lock);
1441
1442         ret = __cpufreq_set_policy(data, policy);
1443         data->user_policy.min = data->min;
1444         data->user_policy.max = data->max;
1445         data->user_policy.policy = data->policy;
1446         data->user_policy.governor = data->governor;
1447
1448         mutex_unlock(&data->lock);
1449         cpufreq_cpu_put(data);
1450
1451         return ret;
1452 }
1453 EXPORT_SYMBOL(cpufreq_set_policy);
1454
1455
1456 /**
1457  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1458  *      @cpu: CPU which shall be re-evaluated
1459  *
1460  *      Usefull for policy notifiers which have different necessities
1461  *      at different times.
1462  */
1463 int cpufreq_update_policy(unsigned int cpu)
1464 {
1465         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1466         struct cpufreq_policy policy;
1467         int ret = 0;
1468
1469         if (!data)
1470                 return -ENODEV;
1471
1472         mutex_lock(&data->lock);
1473
1474         dprintk("updating policy for CPU %u\n", cpu);
1475         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1476         policy.min = data->user_policy.min;
1477         policy.max = data->user_policy.max;
1478         policy.policy = data->user_policy.policy;
1479         policy.governor = data->user_policy.governor;
1480
1481         /* BIOS might change freq behind our back
1482           -> ask driver for current freq and notify governors about a change */
1483         if (cpufreq_driver->get) {
1484                 policy.cur = cpufreq_driver->get(cpu);
1485                 if (!data->cur) {
1486                         dprintk("Driver did not initialize current freq");
1487                         data->cur = policy.cur;
1488                 } else {
1489                         if (data->cur != policy.cur)
1490                                 cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1491                 }
1492         }
1493
1494         ret = __cpufreq_set_policy(data, &policy);
1495
1496         mutex_unlock(&data->lock);
1497
1498         cpufreq_cpu_put(data);
1499         return ret;
1500 }
1501 EXPORT_SYMBOL(cpufreq_update_policy);
1502
1503 #ifdef CONFIG_HOTPLUG_CPU
1504 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1505                                         unsigned long action, void *hcpu)
1506 {
1507         unsigned int cpu = (unsigned long)hcpu;
1508         struct cpufreq_policy *policy;
1509         struct sys_device *sys_dev;
1510
1511         sys_dev = get_cpu_sysdev(cpu);
1512
1513         if (sys_dev) {
1514                 switch (action) {
1515                 case CPU_ONLINE:
1516                         cpufreq_add_dev(sys_dev);
1517                         break;
1518                 case CPU_DOWN_PREPARE:
1519                         /*
1520                          * We attempt to put this cpu in lowest frequency
1521                          * possible before going down. This will permit
1522                          * hardware-managed P-State to switch other related
1523                          * threads to min or higher speeds if possible.
1524                          */
1525                         policy = cpufreq_cpu_data[cpu];
1526                         if (policy) {
1527                                 cpufreq_driver_target(policy, policy->min,
1528                                                 CPUFREQ_RELATION_H);
1529                         }
1530                         break;
1531                 case CPU_DEAD:
1532                         cpufreq_remove_dev(sys_dev);
1533                         break;
1534                 }
1535         }
1536         return NOTIFY_OK;
1537 }
1538
1539 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1540 {
1541     .notifier_call = cpufreq_cpu_callback,
1542 };
1543 #endif /* CONFIG_HOTPLUG_CPU */
1544
1545 /*********************************************************************
1546  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1547  *********************************************************************/
1548
1549 /**
1550  * cpufreq_register_driver - register a CPU Frequency driver
1551  * @driver_data: A struct cpufreq_driver containing the values#
1552  * submitted by the CPU Frequency driver.
1553  *
1554  *   Registers a CPU Frequency driver to this core code. This code
1555  * returns zero on success, -EBUSY when another driver got here first
1556  * (and isn't unregistered in the meantime).
1557  *
1558  */
1559 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1560 {
1561         unsigned long flags;
1562         int ret;
1563
1564         if (!driver_data || !driver_data->verify || !driver_data->init ||
1565             ((!driver_data->setpolicy) && (!driver_data->target)))
1566                 return -EINVAL;
1567
1568         dprintk("trying to register driver %s\n", driver_data->name);
1569
1570         if (driver_data->setpolicy)
1571                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1572
1573         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1574         if (cpufreq_driver) {
1575                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1576                 return -EBUSY;
1577         }
1578         cpufreq_driver = driver_data;
1579         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1580
1581         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1582
1583         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1584                 int i;
1585                 ret = -ENODEV;
1586
1587                 /* check for at least one working CPU */
1588                 for (i=0; i<NR_CPUS; i++)
1589                         if (cpufreq_cpu_data[i])
1590                                 ret = 0;
1591
1592                 /* if all ->init() calls failed, unregister */
1593                 if (ret) {
1594                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1595                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1596
1597                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1598                         cpufreq_driver = NULL;
1599                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1600                 }
1601         }
1602
1603         if (!ret) {
1604                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1605                 dprintk("driver %s up and running\n", driver_data->name);
1606                 cpufreq_debug_enable_ratelimit();
1607         }
1608
1609         return (ret);
1610 }
1611 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1612
1613
1614 /**
1615  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1616  *
1617  *    Unregister the current CPUFreq driver. Only call this if you have
1618  * the right to do so, i.e. if you have succeeded in initialising before!
1619  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1620  * currently not initialised.
1621  */
1622 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1623 {
1624         unsigned long flags;
1625
1626         cpufreq_debug_disable_ratelimit();
1627
1628         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1629                 cpufreq_debug_enable_ratelimit();
1630                 return -EINVAL;
1631         }
1632
1633         dprintk("unregistering driver %s\n", driver->name);
1634
1635         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1636         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1637
1638         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1639         cpufreq_driver = NULL;
1640         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1641
1642         return 0;
1643 }
1644 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);