SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/cpufreq/cpufreq.c

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