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 static struct cpufreq_governor *__find_governor(const char *str_governor)
 288 {
 289         struct cpufreq_governor *t;
 290
 291         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
 292                 if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN))
 293                         return t;
 294
 295         return NULL;
 296 }
 297
 298 /**
 299  * cpufreq_parse_governor - parse a governor string
 300  */
 301 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 302                                 struct cpufreq_governor **governor)
 303 {
 304         int err = -EINVAL;
 305
 306         if (!cpufreq_driver)
 307                 goto out;
 308
 309         if (cpufreq_driver->setpolicy) {
 310                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 311                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 312                         err = 0;
 313                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 314                         *policy = CPUFREQ_POLICY_POWERSAVE;
 315                         err = 0;
 316                 }
 317         } else if (cpufreq_driver->target) {
 318                 struct cpufreq_governor *t;
 319
 320                 mutex_lock(&cpufreq_governor_mutex);
 321
 322                 t = __find_governor(str_governor);
 323
 324                 if (t != NULL) {
 325                         *governor = t;
 326                         err = 0;
 327                 }
 328
 329                 mutex_unlock(&cpufreq_governor_mutex);
 330         }
 331   out:
 332         return err;
 333 }
 334
 335
 336 /* drivers/base/cpu.c */
 337 extern struct sysdev_class cpu_sysdev_class;
 338
 339
 340 /**
 341  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 342  *
 343  * Write out information from cpufreq_driver->policy[cpu]; object must be
 344  * "unsigned int".
 345  */
 346
 347 #define show_one(file_name, object)                     \
 348 static ssize_t show_##file_name                         \
 349 (struct cpufreq_policy * policy, char *buf)             \
 350 {                                                       \
 351         return sprintf (buf, "%u\n", policy->object);   \
 352 }
 353
 354 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 355 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 356 show_one(scaling_min_freq, min);
 357 show_one(scaling_max_freq, max);
 358 show_one(scaling_cur_freq, cur);
 359
 360 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
 361
 362 /**
 363  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 364  */
 365 #define store_one(file_name, object)                    \
 366 static ssize_t store_##file_name                                        \
 367 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 368 {                                                                       \
 369         unsigned int ret = -EINVAL;                                     \
 370         struct cpufreq_policy new_policy;                               \
 371                                                                         \
 372         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 373         if (ret)                                                        \
 374                 return -EINVAL;                                         \
 375                                                                         \
 376         ret = sscanf (buf, "%u", &new_policy.object);                   \
 377         if (ret != 1)                                                   \
 378                 return -EINVAL;                                         \
 379                                                                         \
 380         lock_cpu_hotplug();                                             \
 381         mutex_lock(&policy->lock);                                      \
 382         ret = __cpufreq_set_policy(policy, &new_policy);                \
 383         policy->user_policy.object = policy->object;                    \
 384         mutex_unlock(&policy->lock);                                    \
 385         unlock_cpu_hotplug();                                           \
 386                                                                         \
 387         return ret ? ret : count;                                       \
 388 }
 389
 390 store_one(scaling_min_freq,min);
 391 store_one(scaling_max_freq,max);
 392
 393 /**
 394  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 395  */
 396 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 397 {
 398         unsigned int cur_freq = cpufreq_get(policy->cpu);
 399         if (!cur_freq)
 400                 return sprintf(buf, "<unknown>");
 401         return sprintf(buf, "%u\n", cur_freq);
 402 }
 403
 404
 405 /**
 406  * show_scaling_governor - show the current policy for the specified CPU
 407  */
 408 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 409 {
 410         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 411                 return sprintf(buf, "powersave\n");
 412         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 413                 return sprintf(buf, "performance\n");
 414         else if (policy->governor)
 415                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 416         return -EINVAL;
 417 }
 418
 419
 420 /**
 421  * store_scaling_governor - store policy for the specified CPU
 422  */
 423 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 424                                        const char *buf, size_t count)
 425 {
 426         unsigned int ret = -EINVAL;
 427         char    str_governor[16];
 428         struct cpufreq_policy new_policy;
 429
 430         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 431         if (ret)
 432                 return ret;
 433
 434         ret = sscanf (buf, "%15s", str_governor);
 435         if (ret != 1)
 436                 return -EINVAL;
 437
 438         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 439                 return -EINVAL;
 440
 441         lock_cpu_hotplug();
 442
 443         /* Do not use cpufreq_set_policy here or the user_policy.max
 444            will be wrongly overridden */
 445         mutex_lock(&policy->lock);
 446         ret = __cpufreq_set_policy(policy, &new_policy);
 447
 448         policy->user_policy.policy = policy->policy;
 449         policy->user_policy.governor = policy->governor;
 450         mutex_unlock(&policy->lock);
 451
 452         unlock_cpu_hotplug();
 453
 454         return ret ? ret : count;
 455 }
 456
 457 /**
 458  * show_scaling_driver - show the cpufreq driver currently loaded
 459  */
 460 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 461 {
 462         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 463 }
 464
 465 /**
 466  * show_scaling_available_governors - show the available CPUfreq governors
 467  */
 468 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 469                                 char *buf)
 470 {
 471         ssize_t i = 0;
 472         struct cpufreq_governor *t;
 473
 474         if (!cpufreq_driver->target) {
 475                 i += sprintf(buf, "performance powersave");
 476                 goto out;
 477         }
 478
 479         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 480                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 481                         goto out;
 482                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 483         }
 484 out:
 485         i += sprintf(&buf[i], "\n");
 486         return i;
 487 }
 488 /**
 489  * show_affected_cpus - show the CPUs affected by each transition
 490  */
 491 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 492 {
 493         ssize_t i = 0;
 494         unsigned int cpu;
 495
 496         for_each_cpu_mask(cpu, policy->cpus) {
 497                 if (i)
 498                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 499                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 500                 if (i >= (PAGE_SIZE - 5))
 501                     break;
 502         }
 503         i += sprintf(&buf[i], "\n");
 504         return i;
 505 }
 506
 507
 508 #define define_one_ro(_name) \
 509 static struct freq_attr _name = \
 510 __ATTR(_name, 0444, show_##_name, NULL)
 511
 512 #define define_one_ro0400(_name) \
 513 static struct freq_attr _name = \
 514 __ATTR(_name, 0400, show_##_name, NULL)
 515
 516 #define define_one_rw(_name) \
 517 static struct freq_attr _name = \
 518 __ATTR(_name, 0644, show_##_name, store_##_name)
 519
 520 define_one_ro0400(cpuinfo_cur_freq);
 521 define_one_ro(cpuinfo_min_freq);
 522 define_one_ro(cpuinfo_max_freq);
 523 define_one_ro(scaling_available_governors);
 524 define_one_ro(scaling_driver);
 525 define_one_ro(scaling_cur_freq);
 526 define_one_ro(affected_cpus);
 527 define_one_rw(scaling_min_freq);
 528 define_one_rw(scaling_max_freq);
 529 define_one_rw(scaling_governor);
 530
 531 static struct attribute * default_attrs[] = {
 532         &cpuinfo_min_freq.attr,
 533         &cpuinfo_max_freq.attr,
 534         &scaling_min_freq.attr,
 535         &scaling_max_freq.attr,
 536         &affected_cpus.attr,
 537         &scaling_governor.attr,
 538         &scaling_driver.attr,
 539         &scaling_available_governors.attr,
 540         NULL
 541 };
 542
 543 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 544 #define to_attr(a) container_of(a,struct freq_attr,attr)
 545
 546 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 547 {
 548         struct cpufreq_policy * policy = to_policy(kobj);
 549         struct freq_attr * fattr = to_attr(attr);
 550         ssize_t ret;
 551         policy = cpufreq_cpu_get(policy->cpu);
 552         if (!policy)
 553                 return -EINVAL;
 554         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 555         cpufreq_cpu_put(policy);
 556         return ret;
 557 }
 558
 559 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 560                      const char * buf, size_t count)
 561 {
 562         struct cpufreq_policy * policy = to_policy(kobj);
 563         struct freq_attr * fattr = to_attr(attr);
 564         ssize_t ret;
 565         policy = cpufreq_cpu_get(policy->cpu);
 566         if (!policy)
 567                 return -EINVAL;
 568         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 569         cpufreq_cpu_put(policy);
 570         return ret;
 571 }
 572
 573 static void cpufreq_sysfs_release(struct kobject * kobj)
 574 {
 575         struct cpufreq_policy * policy = to_policy(kobj);
 576         dprintk("last reference is dropped\n");
 577         complete(&policy->kobj_unregister);
 578 }
 579
 580 static struct sysfs_ops sysfs_ops = {
 581         .show   = show,
 582         .store  = store,
 583 };
 584
 585 static struct kobj_type ktype_cpufreq = {
 586         .sysfs_ops      = &sysfs_ops,
 587         .default_attrs  = default_attrs,
 588         .release        = cpufreq_sysfs_release,
 589 };
 590
 591
 592 /**
 593  * cpufreq_add_dev - add a CPU device
 594  *
 595  * Adds the cpufreq interface for a CPU device.
 596  */
 597 static int cpufreq_add_dev (struct sys_device * sys_dev)
 598 {
 599         unsigned int cpu = sys_dev->id;
 600         int ret = 0;
 601         struct cpufreq_policy new_policy;
 602         struct cpufreq_policy *policy;
 603         struct freq_attr **drv_attr;
 604         struct sys_device *cpu_sys_dev;
 605         unsigned long flags;
 606         unsigned int j;
 607 #ifdef CONFIG_SMP
 608         struct cpufreq_policy *managed_policy;
 609 #endif
 610
 611         if (cpu_is_offline(cpu))
 612                 return 0;
 613
 614         cpufreq_debug_disable_ratelimit();
 615         dprintk("adding CPU %u\n", cpu);
 616
 617 #ifdef CONFIG_SMP
 618         /* check whether a different CPU already registered this
 619          * CPU because it is in the same boat. */
 620         policy = cpufreq_cpu_get(cpu);
 621         if (unlikely(policy)) {
 622                 cpufreq_cpu_put(policy);
 623                 cpufreq_debug_enable_ratelimit();
 624                 return 0;
 625         }
 626 #endif
 627
 628         if (!try_module_get(cpufreq_driver->owner)) {
 629                 ret = -EINVAL;
 630                 goto module_out;
 631         }
 632
 633         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 634         if (!policy) {
 635                 ret = -ENOMEM;
 636                 goto nomem_out;
 637         }
 638
 639         policy->cpu = cpu;
 640         policy->cpus = cpumask_of_cpu(cpu);
 641
 642         mutex_init(&policy->lock);
 643         mutex_lock(&policy->lock);
 644         init_completion(&policy->kobj_unregister);
 645         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 646
 647         /* call driver. From then on the cpufreq must be able
 648          * to accept all calls to ->verify and ->setpolicy for this CPU
 649          */
 650         ret = cpufreq_driver->init(policy);
 651         if (ret) {
 652                 dprintk("initialization failed\n");
 653                 mutex_unlock(&policy->lock);
 654                 goto err_out;
 655         }
 656
 657 #ifdef CONFIG_SMP
 658         for_each_cpu_mask(j, policy->cpus) {
 659                 if (cpu == j)
 660                         continue;
 661
 662                 /* check for existing affected CPUs.  They may not be aware
 663                  * of it due to CPU Hotplug.
 664                  */
 665                 managed_policy = cpufreq_cpu_get(j);
 666                 if (unlikely(managed_policy)) {
 667                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 668                         managed_policy->cpus = policy->cpus;
 669                         cpufreq_cpu_data[cpu] = managed_policy;
 670                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 671
 672                         dprintk("CPU already managed, adding link\n");
 673                         sysfs_create_link(&sys_dev->kobj,
 674                                           &managed_policy->kobj, "cpufreq");
 675
 676                         cpufreq_debug_enable_ratelimit();
 677                         mutex_unlock(&policy->lock);
 678                         ret = 0;
 679                         goto err_out_driver_exit; /* call driver->exit() */
 680                 }
 681         }
 682 #endif
 683         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 684
 685         /* prepare interface data */
 686         policy->kobj.parent = &sys_dev->kobj;
 687         policy->kobj.ktype = &ktype_cpufreq;
 688         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 689
 690         ret = kobject_register(&policy->kobj);
 691         if (ret) {
 692                 mutex_unlock(&policy->lock);
 693                 goto err_out_driver_exit;
 694         }
 695         /* set up files for this cpu device */
 696         drv_attr = cpufreq_driver->attr;
 697         while ((drv_attr) && (*drv_attr)) {
 698                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 699                 drv_attr++;
 700         }
 701         if (cpufreq_driver->get)
 702                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 703         if (cpufreq_driver->target)
 704                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 705
 706         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 707         for_each_cpu_mask(j, policy->cpus)
 708                 cpufreq_cpu_data[j] = policy;
 709         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 710
 711         /* symlink affected CPUs */
 712         for_each_cpu_mask(j, policy->cpus) {
 713                 if (j == cpu)
 714                         continue;
 715                 if (!cpu_online(j))
 716                         continue;
 717
 718                 dprintk("CPU %u already managed, adding link\n", j);
 719                 cpufreq_cpu_get(cpu);
 720                 cpu_sys_dev = get_cpu_sysdev(j);
 721                 sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 722                                   "cpufreq");
 723         }
 724
 725         policy->governor = NULL; /* to assure that the starting sequence is
 726                                   * run in cpufreq_set_policy */
 727         mutex_unlock(&policy->lock);
 728
 729         /* set default policy */
 730         ret = cpufreq_set_policy(&new_policy);
 731         if (ret) {
 732                 dprintk("setting policy failed\n");
 733                 goto err_out_unregister;
 734         }
 735
 736         module_put(cpufreq_driver->owner);
 737         dprintk("initialization complete\n");
 738         cpufreq_debug_enable_ratelimit();
 739
 740         return 0;
 741
 742
 743 err_out_unregister:
 744         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 745         for_each_cpu_mask(j, policy->cpus)
 746                 cpufreq_cpu_data[j] = NULL;
 747         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 748
 749         kobject_unregister(&policy->kobj);
 750         wait_for_completion(&policy->kobj_unregister);
 751
 752 err_out_driver_exit:
 753         if (cpufreq_driver->exit)
 754                 cpufreq_driver->exit(policy);
 755
 756 err_out:
 757         kfree(policy);
 758
 759 nomem_out:
 760         module_put(cpufreq_driver->owner);
 761 module_out:
 762         cpufreq_debug_enable_ratelimit();
 763         return ret;
 764 }
 765
 766
 767 /**
 768  * cpufreq_remove_dev - remove a CPU device
 769  *
 770  * Removes the cpufreq interface for a CPU device.
 771  */
 772 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 773 {
 774         unsigned int cpu = sys_dev->id;
 775         unsigned long flags;
 776         struct cpufreq_policy *data;
 777 #ifdef CONFIG_SMP
 778         struct sys_device *cpu_sys_dev;
 779         unsigned int j;
 780 #endif
 781
 782         cpufreq_debug_disable_ratelimit();
 783         dprintk("unregistering CPU %u\n", cpu);
 784
 785         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 786         data = cpufreq_cpu_data[cpu];
 787
 788         if (!data) {
 789                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 790                 cpufreq_debug_enable_ratelimit();
 791                 return -EINVAL;
 792         }
 793         cpufreq_cpu_data[cpu] = NULL;
 794
 795
 796 #ifdef CONFIG_SMP
 797         /* if this isn't the CPU which is the parent of the kobj, we
 798          * only need to unlink, put and exit
 799          */
 800         if (unlikely(cpu != data->cpu)) {
 801                 dprintk("removing link\n");
 802                 cpu_clear(cpu, data->cpus);
 803                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 804                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 805                 cpufreq_cpu_put(data);
 806                 cpufreq_debug_enable_ratelimit();
 807                 return 0;
 808         }
 809 #endif
 810
 811
 812         if (!kobject_get(&data->kobj)) {
 813                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 814                 cpufreq_debug_enable_ratelimit();
 815                 return -EFAULT;
 816         }
 817
 818 #ifdef CONFIG_SMP
 819         /* if we have other CPUs still registered, we need to unlink them,
 820          * or else wait_for_completion below will lock up. Clean the
 821          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 822          * links afterwards.
 823          */
 824         if (unlikely(cpus_weight(data->cpus) > 1)) {
 825                 for_each_cpu_mask(j, data->cpus) {
 826                         if (j == cpu)
 827                                 continue;
 828                         cpufreq_cpu_data[j] = NULL;
 829                 }
 830         }
 831
 832         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 833
 834         if (unlikely(cpus_weight(data->cpus) > 1)) {
 835                 for_each_cpu_mask(j, data->cpus) {
 836                         if (j == cpu)
 837                                 continue;
 838                         dprintk("removing link for cpu %u\n", j);
 839                         cpu_sys_dev = get_cpu_sysdev(j);
 840                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 841                         cpufreq_cpu_put(data);
 842                 }
 843         }
 844 #else
 845         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 846 #endif
 847
 848         mutex_lock(&data->lock);
 849         if (cpufreq_driver->target)
 850                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 851         mutex_unlock(&data->lock);
 852
 853         kobject_unregister(&data->kobj);
 854
 855         kobject_put(&data->kobj);
 856
 857         /* we need to make sure that the underlying kobj is actually
 858          * not referenced anymore by anybody before we proceed with
 859          * unloading.
 860          */
 861         dprintk("waiting for dropping of refcount\n");
 862         wait_for_completion(&data->kobj_unregister);
 863         dprintk("wait complete\n");
 864
 865         if (cpufreq_driver->exit)
 866                 cpufreq_driver->exit(data);
 867
 868         kfree(data);
 869
 870         cpufreq_debug_enable_ratelimit();
 871         return 0;
 872 }
 873
 874
 875 static void handle_update(void *data)
 876 {
 877         unsigned int cpu = (unsigned int)(long)data;
 878         dprintk("handle_update for cpu %u called\n", cpu);
 879         cpufreq_update_policy(cpu);
 880 }
 881
 882 /**
 883  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 884  *      @cpu: cpu number
 885  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 886  *      @new_freq: CPU frequency the CPU actually runs at
 887  *
 888  *      We adjust to current frequency first, and need to clean up later. So either call
 889  *      to cpufreq_update_policy() or schedule handle_update()).
 890  */
 891 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 892 {
 893         struct cpufreq_freqs freqs;
 894
 895         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
 896                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 897
 898         freqs.cpu = cpu;
 899         freqs.old = old_freq;
 900         freqs.new = new_freq;
 901         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 902         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 903 }
 904
 905
 906 /**
 907  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
 908  * @cpu: CPU number
 909  *
 910  * This is the last known freq, without actually getting it from the driver.
 911  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 912  */
 913 unsigned int cpufreq_quick_get(unsigned int cpu)
 914 {
 915         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 916         unsigned int ret = 0;
 917
 918         if (policy) {
 919                 mutex_lock(&policy->lock);
 920                 ret = policy->cur;
 921                 mutex_unlock(&policy->lock);
 922                 cpufreq_cpu_put(policy);
 923         }
 924
 925         return (ret);
 926 }
 927 EXPORT_SYMBOL(cpufreq_quick_get);
 928
 929
 930 /**
 931  * cpufreq_get - get the current CPU frequency (in kHz)
 932  * @cpu: CPU number
 933  *
 934  * Get the CPU current (static) CPU frequency
 935  */
 936 unsigned int cpufreq_get(unsigned int cpu)
 937 {
 938         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 939         unsigned int ret = 0;
 940
 941         if (!policy)
 942                 return 0;
 943
 944         if (!cpufreq_driver->get)
 945                 goto out;
 946
 947         mutex_lock(&policy->lock);
 948
 949         ret = cpufreq_driver->get(cpu);
 950
 951         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 952                 /* verify no discrepancy between actual and saved value exists */
 953                 if (unlikely(ret != policy->cur)) {
 954                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 955                         schedule_work(&policy->update);
 956                 }
 957         }
 958
 959         mutex_unlock(&policy->lock);
 960
 961 out:
 962         cpufreq_cpu_put(policy);
 963
 964         return (ret);
 965 }
 966 EXPORT_SYMBOL(cpufreq_get);
 967
 968
 969 /**
 970  *      cpufreq_suspend - let the low level driver prepare for suspend
 971  */
 972
 973 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 974 {
 975         int cpu = sysdev->id;
 976         unsigned int ret = 0;
 977         unsigned int cur_freq = 0;
 978         struct cpufreq_policy *cpu_policy;
 979
 980         dprintk("resuming cpu %u\n", cpu);
 981
 982         if (!cpu_online(cpu))
 983                 return 0;
 984
 985         /* we may be lax here as interrupts are off. Nonetheless
 986          * we need to grab the correct cpu policy, as to check
 987          * whether we really run on this CPU.
 988          */
 989
 990         cpu_policy = cpufreq_cpu_get(cpu);
 991         if (!cpu_policy)
 992                 return -EINVAL;
 993
 994         /* only handle each CPU group once */
 995         if (unlikely(cpu_policy->cpu != cpu)) {
 996                 cpufreq_cpu_put(cpu_policy);
 997                 return 0;
 998         }
 999
1000         if (cpufreq_driver->suspend) {
1001                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1002                 if (ret) {
1003                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1004                                         "step on CPU %u\n", cpu_policy->cpu);
1005                         cpufreq_cpu_put(cpu_policy);
1006                         return ret;
1007                 }
1008         }
1009
1010
1011         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1012                 goto out;
1013
1014         if (cpufreq_driver->get)
1015                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1016
1017         if (!cur_freq || !cpu_policy->cur) {
1018                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1019                        "frequency is what timing core thinks it is.\n");
1020                 goto out;
1021         }
1022
1023         if (unlikely(cur_freq != cpu_policy->cur)) {
1024                 struct cpufreq_freqs freqs;
1025
1026                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1027                         dprintk("Warning: CPU frequency is %u, "
1028                                "cpufreq assumed %u kHz.\n",
1029                                cur_freq, cpu_policy->cur);
1030
1031                 freqs.cpu = cpu;
1032                 freqs.old = cpu_policy->cur;
1033                 freqs.new = cur_freq;
1034
1035                 blocking_notifier_call_chain(&cpufreq_transition_notifier_list,
1036                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1037                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1038
1039                 cpu_policy->cur = cur_freq;
1040         }
1041
1042 out:
1043         cpufreq_cpu_put(cpu_policy);
1044         return 0;
1045 }
1046
1047 /**
1048  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1049  *
1050  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1051  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1052  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1053  *          restored.
1054  */
1055 static int cpufreq_resume(struct sys_device * sysdev)
1056 {
1057         int cpu = sysdev->id;
1058         unsigned int ret = 0;
1059         struct cpufreq_policy *cpu_policy;
1060
1061         dprintk("resuming cpu %u\n", cpu);
1062
1063         if (!cpu_online(cpu))
1064                 return 0;
1065
1066         /* we may be lax here as interrupts are off. Nonetheless
1067          * we need to grab the correct cpu policy, as to check
1068          * whether we really run on this CPU.
1069          */
1070
1071         cpu_policy = cpufreq_cpu_get(cpu);
1072         if (!cpu_policy)
1073                 return -EINVAL;
1074
1075         /* only handle each CPU group once */
1076         if (unlikely(cpu_policy->cpu != cpu)) {
1077                 cpufreq_cpu_put(cpu_policy);
1078                 return 0;
1079         }
1080
1081         if (cpufreq_driver->resume) {
1082                 ret = cpufreq_driver->resume(cpu_policy);
1083                 if (ret) {
1084                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1085                                         "step on CPU %u\n", cpu_policy->cpu);
1086                         cpufreq_cpu_put(cpu_policy);
1087                         return ret;
1088                 }
1089         }
1090
1091         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1092                 unsigned int cur_freq = 0;
1093
1094                 if (cpufreq_driver->get)
1095                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1096
1097                 if (!cur_freq || !cpu_policy->cur) {
1098                         printk(KERN_ERR "cpufreq: resume failed to assert "
1099                                         "current frequency is what timing core "
1100                                         "thinks it is.\n");
1101                         goto out;
1102                 }
1103
1104                 if (unlikely(cur_freq != cpu_policy->cur)) {
1105                         struct cpufreq_freqs freqs;
1106
1107                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1108                                 dprintk("Warning: CPU frequency"
1109                                        "is %u, cpufreq assumed %u kHz.\n",
1110                                        cur_freq, cpu_policy->cur);
1111
1112                         freqs.cpu = cpu;
1113                         freqs.old = cpu_policy->cur;
1114                         freqs.new = cur_freq;
1115
1116                         blocking_notifier_call_chain(
1117                                         &cpufreq_transition_notifier_list,
1118                                         CPUFREQ_RESUMECHANGE, &freqs);
1119                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1120
1121                         cpu_policy->cur = cur_freq;
1122                 }
1123         }
1124
1125 out:
1126         schedule_work(&cpu_policy->update);
1127         cpufreq_cpu_put(cpu_policy);
1128         return ret;
1129 }
1130
1131 static struct sysdev_driver cpufreq_sysdev_driver = {
1132         .add            = cpufreq_add_dev,
1133         .remove         = cpufreq_remove_dev,
1134         .suspend        = cpufreq_suspend,
1135         .resume         = cpufreq_resume,
1136 };
1137
1138
1139 /*********************************************************************
1140  *                     NOTIFIER LISTS INTERFACE                      *
1141  *********************************************************************/
1142
1143 /**
1144  *      cpufreq_register_notifier - register a driver with cpufreq
1145  *      @nb: notifier function to register
1146  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1147  *
1148  *      Add a driver to one of two lists: either a list of drivers that
1149  *      are notified about clock rate changes (once before and once after
1150  *      the transition), or a list of drivers that are notified about
1151  *      changes in cpufreq policy.
1152  *
1153  *      This function may sleep, and has the same return conditions as
1154  *      blocking_notifier_chain_register.
1155  */
1156 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1157 {
1158         int ret;
1159
1160         switch (list) {
1161         case CPUFREQ_TRANSITION_NOTIFIER:
1162                 ret = blocking_notifier_chain_register(
1163                                 &cpufreq_transition_notifier_list, nb);
1164                 break;
1165         case CPUFREQ_POLICY_NOTIFIER:
1166                 ret = blocking_notifier_chain_register(
1167                                 &cpufreq_policy_notifier_list, nb);
1168                 break;
1169         default:
1170                 ret = -EINVAL;
1171         }
1172
1173         return ret;
1174 }
1175 EXPORT_SYMBOL(cpufreq_register_notifier);
1176
1177
1178 /**
1179  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1180  *      @nb: notifier block to be unregistered
1181  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1182  *
1183  *      Remove a driver from the CPU frequency notifier list.
1184  *
1185  *      This function may sleep, and has the same return conditions as
1186  *      blocking_notifier_chain_unregister.
1187  */
1188 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1189 {
1190         int ret;
1191
1192         switch (list) {
1193         case CPUFREQ_TRANSITION_NOTIFIER:
1194                 ret = blocking_notifier_chain_unregister(
1195                                 &cpufreq_transition_notifier_list, nb);
1196                 break;
1197         case CPUFREQ_POLICY_NOTIFIER:
1198                 ret = blocking_notifier_chain_unregister(
1199                                 &cpufreq_policy_notifier_list, nb);
1200                 break;
1201         default:
1202                 ret = -EINVAL;
1203         }
1204
1205         return ret;
1206 }
1207 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1208
1209
1210 /*********************************************************************
1211  *                              GOVERNORS                            *
1212  *********************************************************************/
1213
1214
1215 /* Must be called with lock_cpu_hotplug held */
1216 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1217                             unsigned int target_freq,
1218                             unsigned int relation)
1219 {
1220         int retval = -EINVAL;
1221
1222         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1223                 target_freq, relation);
1224         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1225                 retval = cpufreq_driver->target(policy, target_freq, relation);
1226
1227         return retval;
1228 }
1229 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1230
1231 int cpufreq_driver_target(struct cpufreq_policy *policy,
1232                           unsigned int target_freq,
1233                           unsigned int relation)
1234 {
1235         int ret;
1236
1237         policy = cpufreq_cpu_get(policy->cpu);
1238         if (!policy)
1239                 return -EINVAL;
1240
1241         lock_cpu_hotplug();
1242         mutex_lock(&policy->lock);
1243
1244         ret = __cpufreq_driver_target(policy, target_freq, relation);
1245
1246         mutex_unlock(&policy->lock);
1247         unlock_cpu_hotplug();
1248
1249         cpufreq_cpu_put(policy);
1250         return ret;
1251 }
1252 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1253
1254 /*
1255  * Locking: Must be called with the lock_cpu_hotplug() lock held
1256  * when "event" is CPUFREQ_GOV_LIMITS
1257  */
1258
1259 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1260 {
1261         int ret;
1262
1263         if (!try_module_get(policy->governor->owner))
1264                 return -EINVAL;
1265
1266         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1267         ret = policy->governor->governor(policy, event);
1268
1269         /* we keep one module reference alive for each CPU governed by this CPU */
1270         if ((event != CPUFREQ_GOV_START) || ret)
1271                 module_put(policy->governor->owner);
1272         if ((event == CPUFREQ_GOV_STOP) && !ret)
1273                 module_put(policy->governor->owner);
1274
1275         return ret;
1276 }
1277
1278
1279 int cpufreq_register_governor(struct cpufreq_governor *governor)
1280 {
1281         int err;
1282
1283         if (!governor)
1284                 return -EINVAL;
1285
1286         mutex_lock(&cpufreq_governor_mutex);
1287
1288         err = -EBUSY;
1289         if (__find_governor(governor->name) == NULL) {
1290                 err = 0;
1291                 list_add(&governor->governor_list, &cpufreq_governor_list);
1292         }
1293
1294         mutex_unlock(&cpufreq_governor_mutex);
1295         return err;
1296 }
1297 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1298
1299
1300 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1301 {
1302         if (!governor)
1303                 return;
1304
1305         mutex_lock(&cpufreq_governor_mutex);
1306         list_del(&governor->governor_list);
1307         mutex_unlock(&cpufreq_governor_mutex);
1308         return;
1309 }
1310 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1311
1312
1313
1314 /*********************************************************************
1315  *                          POLICY INTERFACE                         *
1316  *********************************************************************/
1317
1318 /**
1319  * cpufreq_get_policy - get the current cpufreq_policy
1320  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1321  *
1322  * Reads the current cpufreq policy.
1323  */
1324 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1325 {
1326         struct cpufreq_policy *cpu_policy;
1327         if (!policy)
1328                 return -EINVAL;
1329
1330         cpu_policy = cpufreq_cpu_get(cpu);
1331         if (!cpu_policy)
1332                 return -EINVAL;
1333
1334         mutex_lock(&cpu_policy->lock);
1335         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1336         mutex_unlock(&cpu_policy->lock);
1337
1338         cpufreq_cpu_put(cpu_policy);
1339         return 0;
1340 }
1341 EXPORT_SYMBOL(cpufreq_get_policy);
1342
1343
1344 /*
1345  * Locking: Must be called with the lock_cpu_hotplug() lock held
1346  */
1347 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1348 {
1349         int ret = 0;
1350
1351         cpufreq_debug_disable_ratelimit();
1352         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1353                 policy->min, policy->max);
1354
1355         memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
1356
1357         if (policy->min > data->min && policy->min > policy->max) {
1358                 ret = -EINVAL;
1359                 goto error_out;
1360         }
1361
1362         /* verify the cpu speed can be set within this limit */
1363         ret = cpufreq_driver->verify(policy);
1364         if (ret)
1365                 goto error_out;
1366
1367         /* adjust if necessary - all reasons */
1368         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1369                         CPUFREQ_ADJUST, policy);
1370
1371         /* adjust if necessary - hardware incompatibility*/
1372         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1373                         CPUFREQ_INCOMPATIBLE, policy);
1374
1375         /* verify the cpu speed can be set within this limit,
1376            which might be different to the first one */
1377         ret = cpufreq_driver->verify(policy);
1378         if (ret)
1379                 goto error_out;
1380
1381         /* notification of the new policy */
1382         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1383                         CPUFREQ_NOTIFY, policy);
1384
1385         data->min = policy->min;
1386         data->max = policy->max;
1387
1388         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1389
1390         if (cpufreq_driver->setpolicy) {
1391                 data->policy = policy->policy;
1392                 dprintk("setting range\n");
1393                 ret = cpufreq_driver->setpolicy(policy);
1394         } else {
1395                 if (policy->governor != data->governor) {
1396                         /* save old, working values */
1397                         struct cpufreq_governor *old_gov = data->governor;
1398
1399                         dprintk("governor switch\n");
1400
1401                         /* end old governor */
1402                         if (data->governor)
1403                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1404
1405                         /* start new governor */
1406                         data->governor = policy->governor;
1407                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1408                                 /* new governor failed, so re-start old one */
1409                                 dprintk("starting governor %s failed\n", data->governor->name);
1410                                 if (old_gov) {
1411                                         data->governor = old_gov;
1412                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1413                                 }
1414                                 ret = -EINVAL;
1415                                 goto error_out;
1416                         }
1417                         /* might be a policy change, too, so fall through */
1418                 }
1419                 dprintk("governor: change or update limits\n");
1420                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1421         }
1422
1423 error_out:
1424         cpufreq_debug_enable_ratelimit();
1425         return ret;
1426 }
1427
1428 /**
1429  *      cpufreq_set_policy - set a new CPUFreq policy
1430  *      @policy: policy to be set.
1431  *
1432  *      Sets a new CPU frequency and voltage scaling policy.
1433  */
1434 int cpufreq_set_policy(struct cpufreq_policy *policy)
1435 {
1436         int ret = 0;
1437         struct cpufreq_policy *data;
1438
1439         if (!policy)
1440                 return -EINVAL;
1441
1442         data = cpufreq_cpu_get(policy->cpu);
1443         if (!data)
1444                 return -EINVAL;
1445
1446         lock_cpu_hotplug();
1447
1448         /* lock this CPU */
1449         mutex_lock(&data->lock);
1450
1451         ret = __cpufreq_set_policy(data, policy);
1452         data->user_policy.min = data->min;
1453         data->user_policy.max = data->max;
1454         data->user_policy.policy = data->policy;
1455         data->user_policy.governor = data->governor;
1456
1457         mutex_unlock(&data->lock);
1458
1459         unlock_cpu_hotplug();
1460         cpufreq_cpu_put(data);
1461
1462         return ret;
1463 }
1464 EXPORT_SYMBOL(cpufreq_set_policy);
1465
1466
1467 /**
1468  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1469  *      @cpu: CPU which shall be re-evaluated
1470  *
1471  *      Usefull for policy notifiers which have different necessities
1472  *      at different times.
1473  */
1474 int cpufreq_update_policy(unsigned int cpu)
1475 {
1476         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1477         struct cpufreq_policy policy;
1478         int ret = 0;
1479
1480         if (!data)
1481                 return -ENODEV;
1482
1483         lock_cpu_hotplug();
1484         mutex_lock(&data->lock);
1485
1486         dprintk("updating policy for CPU %u\n", cpu);
1487         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1488         policy.min = data->user_policy.min;
1489         policy.max = data->user_policy.max;
1490         policy.policy = data->user_policy.policy;
1491         policy.governor = data->user_policy.governor;
1492
1493         /* BIOS might change freq behind our back
1494           -> ask driver for current freq and notify governors about a change */
1495         if (cpufreq_driver->get) {
1496                 policy.cur = cpufreq_driver->get(cpu);
1497                 if (!data->cur) {
1498                         dprintk("Driver did not initialize current freq");
1499                         data->cur = policy.cur;
1500                 } else {
1501                         if (data->cur != policy.cur)
1502                                 cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1503                 }
1504         }
1505
1506         ret = __cpufreq_set_policy(data, &policy);
1507
1508         mutex_unlock(&data->lock);
1509         unlock_cpu_hotplug();
1510         cpufreq_cpu_put(data);
1511         return ret;
1512 }
1513 EXPORT_SYMBOL(cpufreq_update_policy);
1514
1515 #ifdef CONFIG_HOTPLUG_CPU
1516 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1517                                         unsigned long action, void *hcpu)
1518 {
1519         unsigned int cpu = (unsigned long)hcpu;
1520         struct cpufreq_policy *policy;
1521         struct sys_device *sys_dev;
1522
1523         sys_dev = get_cpu_sysdev(cpu);
1524
1525         if (sys_dev) {
1526                 switch (action) {
1527                 case CPU_ONLINE:
1528                         cpufreq_add_dev(sys_dev);
1529                         break;
1530                 case CPU_DOWN_PREPARE:
1531                         /*
1532                          * We attempt to put this cpu in lowest frequency
1533                          * possible before going down. This will permit
1534                          * hardware-managed P-State to switch other related
1535                          * threads to min or higher speeds if possible.
1536                          */
1537                         policy = cpufreq_cpu_data[cpu];
1538                         if (policy) {
1539                                 cpufreq_driver_target(policy, policy->min,
1540                                                 CPUFREQ_RELATION_H);
1541                         }
1542                         break;
1543                 case CPU_DEAD:
1544                         cpufreq_remove_dev(sys_dev);
1545                         break;
1546                 }
1547         }
1548         return NOTIFY_OK;
1549 }
1550
1551 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1552 {
1553     .notifier_call = cpufreq_cpu_callback,
1554 };
1555 #endif /* CONFIG_HOTPLUG_CPU */
1556
1557 /*********************************************************************
1558  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1559  *********************************************************************/
1560
1561 /**
1562  * cpufreq_register_driver - register a CPU Frequency driver
1563  * @driver_data: A struct cpufreq_driver containing the values#
1564  * submitted by the CPU Frequency driver.
1565  *
1566  *   Registers a CPU Frequency driver to this core code. This code
1567  * returns zero on success, -EBUSY when another driver got here first
1568  * (and isn't unregistered in the meantime).
1569  *
1570  */
1571 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1572 {
1573         unsigned long flags;
1574         int ret;
1575
1576         if (!driver_data || !driver_data->verify || !driver_data->init ||
1577             ((!driver_data->setpolicy) && (!driver_data->target)))
1578                 return -EINVAL;
1579
1580         dprintk("trying to register driver %s\n", driver_data->name);
1581
1582         if (driver_data->setpolicy)
1583                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1584
1585         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1586         if (cpufreq_driver) {
1587                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1588                 return -EBUSY;
1589         }
1590         cpufreq_driver = driver_data;
1591         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1592
1593         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1594
1595         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1596                 int i;
1597                 ret = -ENODEV;
1598
1599                 /* check for at least one working CPU */
1600                 for (i=0; i<NR_CPUS; i++)
1601                         if (cpufreq_cpu_data[i])
1602                                 ret = 0;
1603
1604                 /* if all ->init() calls failed, unregister */
1605                 if (ret) {
1606                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1607                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1608
1609                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1610                         cpufreq_driver = NULL;
1611                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1612                 }
1613         }
1614
1615         if (!ret) {
1616                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1617                 dprintk("driver %s up and running\n", driver_data->name);
1618                 cpufreq_debug_enable_ratelimit();
1619         }
1620
1621         return (ret);
1622 }
1623 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1624
1625
1626 /**
1627  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1628  *
1629  *    Unregister the current CPUFreq driver. Only call this if you have
1630  * the right to do so, i.e. if you have succeeded in initialising before!
1631  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1632  * currently not initialised.
1633  */
1634 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1635 {
1636         unsigned long flags;
1637
1638         cpufreq_debug_disable_ratelimit();
1639
1640         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1641                 cpufreq_debug_enable_ratelimit();
1642                 return -EINVAL;
1643         }
1644
1645         dprintk("unregistering driver %s\n", driver->name);
1646
1647         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1648         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1649
1650         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1651         cpufreq_driver = NULL;
1652         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1653
1654         return 0;
1655 }
1656 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);