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, \
  33                                                 "cpufreq-core", msg)
  34
  35 /**
  36  * The "cpufreq driver" - the arch- or hardware-dependent 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(struct work_struct *work);
  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 struct srcu_notifier_head cpufreq_transition_notifier_list;
  57
  58 static int __init init_cpufreq_transition_notifier_list(void)
  59 {
  60         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
  61         return 0;
  62 }
  63 pure_initcall(init_cpufreq_transition_notifier_list);
  64
  65 static LIST_HEAD(cpufreq_governor_list);
  66 static DEFINE_MUTEX (cpufreq_governor_mutex);
  67
  68 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
  69 {
  70         struct cpufreq_policy *data;
  71         unsigned long flags;
  72
  73         if (cpu >= NR_CPUS)
  74                 goto err_out;
  75
  76         /* get the cpufreq driver */
  77         spin_lock_irqsave(&cpufreq_driver_lock, flags);
  78
  79         if (!cpufreq_driver)
  80                 goto err_out_unlock;
  81
  82         if (!try_module_get(cpufreq_driver->owner))
  83                 goto err_out_unlock;
  84
  85
  86         /* get the CPU */
  87         data = cpufreq_cpu_data[cpu];
  88
  89         if (!data)
  90                 goto err_out_put_module;
  91
  92         if (!kobject_get(&data->kobj))
  93                 goto err_out_put_module;
  94
  95         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  96         return data;
  97
  98 err_out_put_module:
  99         module_put(cpufreq_driver->owner);
 100 err_out_unlock:
 101         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 102 err_out:
 103         return NULL;
 104 }
 105 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 106
 107
 108 void cpufreq_cpu_put(struct cpufreq_policy *data)
 109 {
 110         kobject_put(&data->kobj);
 111         module_put(cpufreq_driver->owner);
 112 }
 113 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 114
 115
 116 /*********************************************************************
 117  *                     UNIFIED DEBUG HELPERS                         *
 118  *********************************************************************/
 119 #ifdef CONFIG_CPU_FREQ_DEBUG
 120
 121 /* what part(s) of the CPUfreq subsystem are debugged? */
 122 static unsigned int debug;
 123
 124 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 125  * set or modify this value.
 126  */
 127 static unsigned int debug_ratelimit = 1;
 128
 129 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 130  * loading of a cpufreq driver, temporarily disabled when a new policy
 131  * is set, and disabled upon cpufreq driver removal
 132  */
 133 static unsigned int disable_ratelimit = 1;
 134 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 135
 136 static void cpufreq_debug_enable_ratelimit(void)
 137 {
 138         unsigned long flags;
 139
 140         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 141         if (disable_ratelimit)
 142                 disable_ratelimit--;
 143         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 144 }
 145
 146 static void cpufreq_debug_disable_ratelimit(void)
 147 {
 148         unsigned long flags;
 149
 150         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 151         disable_ratelimit++;
 152         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 153 }
 154
 155 void cpufreq_debug_printk(unsigned int type, const char *prefix,
 156                                                         const char *fmt, ...)
 157 {
 158         char s[256];
 159         va_list args;
 160         unsigned int len;
 161         unsigned long flags;
 162
 163         WARN_ON(!prefix);
 164         if (type & debug) {
 165                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 166                 if (!disable_ratelimit && debug_ratelimit
 167                                         && !printk_ratelimit()) {
 168                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 169                         return;
 170                 }
 171                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 172
 173                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 174
 175                 va_start(args, fmt);
 176                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 177                 va_end(args);
 178
 179                 printk(s);
 180
 181                 WARN_ON(len < 5);
 182         }
 183 }
 184 EXPORT_SYMBOL(cpufreq_debug_printk);
 185
 186
 187 module_param(debug, uint, 0644);
 188 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core,"
 189                         " 2 to debug drivers, and 4 to debug governors.");
 190
 191 module_param(debug_ratelimit, uint, 0644);
 192 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging:"
 193                                         " set to 0 to disable ratelimiting.");
 194
 195 #else /* !CONFIG_CPU_FREQ_DEBUG */
 196
 197 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 198 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 199
 200 #endif /* CONFIG_CPU_FREQ_DEBUG */
 201
 202
 203 /*********************************************************************
 204  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 205  *********************************************************************/
 206
 207 /**
 208  * adjust_jiffies - adjust the system "loops_per_jiffy"
 209  *
 210  * This function alters the system "loops_per_jiffy" for the clock
 211  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 212  * systems as each CPU might be scaled differently. So, use the arch
 213  * per-CPU loops_per_jiffy value wherever possible.
 214  */
 215 #ifndef CONFIG_SMP
 216 static unsigned long l_p_j_ref;
 217 static unsigned int  l_p_j_ref_freq;
 218
 219 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 220 {
 221         if (ci->flags & CPUFREQ_CONST_LOOPS)
 222                 return;
 223
 224         if (!l_p_j_ref_freq) {
 225                 l_p_j_ref = loops_per_jiffy;
 226                 l_p_j_ref_freq = ci->old;
 227                 dprintk("saving %lu as reference value for loops_per_jiffy;"
 228                         "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 229         }
 230         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 231             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 232             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 233                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
 234                                                                 ci->new);
 235                 dprintk("scaling loops_per_jiffy to %lu"
 236                         "for frequency %u kHz\n", loops_per_jiffy, ci->new);
 237         }
 238 }
 239 #else
 240 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 241 {
 242         return;
 243 }
 244 #endif
 245
 246
 247 /**
 248  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 249  * on frequency transition.
 250  *
 251  * This function calls the transition notifiers and the "adjust_jiffies"
 252  * function. It is called twice on all CPU frequency changes that have
 253  * external effects.
 254  */
 255 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 256 {
 257         struct cpufreq_policy *policy;
 258
 259         BUG_ON(irqs_disabled());
 260
 261         freqs->flags = cpufreq_driver->flags;
 262         dprintk("notification %u of frequency transition to %u kHz\n",
 263                 state, freqs->new);
 264
 265         policy = cpufreq_cpu_data[freqs->cpu];
 266         switch (state) {
 267
 268         case CPUFREQ_PRECHANGE:
 269                 /* detect if the driver reported a value as "old frequency"
 270                  * which is not equal to what the cpufreq core thinks is
 271                  * "old frequency".
 272                  */
 273                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 274                         if ((policy) && (policy->cpu == freqs->cpu) &&
 275                             (policy->cur) && (policy->cur != freqs->old)) {
 276                                 dprintk("Warning: CPU frequency is"
 277                                         " %u, cpufreq assumed %u kHz.\n",
 278                                         freqs->old, policy->cur);
 279                                 freqs->old = policy->cur;
 280                         }
 281                 }
 282                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 283                                 CPUFREQ_PRECHANGE, freqs);
 284                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 285                 break;
 286
 287         case CPUFREQ_POSTCHANGE:
 288                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 289                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 290                                 CPUFREQ_POSTCHANGE, freqs);
 291                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 292                         policy->cur = freqs->new;
 293                 break;
 294         }
 295 }
 296 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 297
 298
 299
 300 /*********************************************************************
 301  *                          SYSFS INTERFACE                          *
 302  *********************************************************************/
 303
 304 static struct cpufreq_governor *__find_governor(const char *str_governor)
 305 {
 306         struct cpufreq_governor *t;
 307
 308         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
 309                 if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN))
 310                         return t;
 311
 312         return NULL;
 313 }
 314
 315 /**
 316  * cpufreq_parse_governor - parse a governor string
 317  */
 318 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 319                                 struct cpufreq_governor **governor)
 320 {
 321         int err = -EINVAL;
 322
 323         if (!cpufreq_driver)
 324                 goto out;
 325
 326         if (cpufreq_driver->setpolicy) {
 327                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 328                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 329                         err = 0;
 330                 } else if (!strnicmp(str_governor, "powersave",
 331                                                 CPUFREQ_NAME_LEN)) {
 332                         *policy = CPUFREQ_POLICY_POWERSAVE;
 333                         err = 0;
 334                 }
 335         } else if (cpufreq_driver->target) {
 336                 struct cpufreq_governor *t;
 337
 338                 mutex_lock(&cpufreq_governor_mutex);
 339
 340                 t = __find_governor(str_governor);
 341
 342                 if (t == NULL) {
 343                         char *name = kasprintf(GFP_KERNEL, "cpufreq_%s",
 344                                                                 str_governor);
 345
 346                         if (name) {
 347                                 int ret;
 348
 349                                 mutex_unlock(&cpufreq_governor_mutex);
 350                                 ret = request_module(name);
 351                                 mutex_lock(&cpufreq_governor_mutex);
 352
 353                                 if (ret == 0)
 354                                         t = __find_governor(str_governor);
 355                         }
 356
 357                         kfree(name);
 358                 }
 359
 360                 if (t != NULL) {
 361                         *governor = t;
 362                         err = 0;
 363                 }
 364
 365                 mutex_unlock(&cpufreq_governor_mutex);
 366         }
 367   out:
 368         return err;
 369 }
 370
 371
 372 /* drivers/base/cpu.c */
 373 extern struct sysdev_class cpu_sysdev_class;
 374
 375
 376 /**
 377  * cpufreq_per_cpu_attr_read() / show_##file_name() -
 378  * print out cpufreq information
 379  *
 380  * Write out information from cpufreq_driver->policy[cpu]; object must be
 381  * "unsigned int".
 382  */
 383
 384 #define show_one(file_name, object)                     \
 385 static ssize_t show_##file_name                         \
 386 (struct cpufreq_policy * policy, char *buf)             \
 387 {                                                       \
 388         return sprintf (buf, "%u\n", policy->object);   \
 389 }
 390
 391 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 392 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 393 show_one(scaling_min_freq, min);
 394 show_one(scaling_max_freq, max);
 395 show_one(scaling_cur_freq, cur);
 396
 397 static int __cpufreq_set_policy(struct cpufreq_policy *data,
 398                                 struct cpufreq_policy *policy);
 399
 400 /**
 401  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 402  */
 403 #define store_one(file_name, object)                    \
 404 static ssize_t store_##file_name                                        \
 405 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 406 {                                                                       \
 407         unsigned int ret = -EINVAL;                                     \
 408         struct cpufreq_policy new_policy;                               \
 409                                                                         \
 410         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 411         if (ret)                                                        \
 412                 return -EINVAL;                                         \
 413                                                                         \
 414         ret = sscanf (buf, "%u", &new_policy.object);                   \
 415         if (ret != 1)                                                   \
 416                 return -EINVAL;                                         \
 417                                                                         \
 418         lock_cpu_hotplug();                                             \
 419         mutex_lock(&policy->lock);                                      \
 420         ret = __cpufreq_set_policy(policy, &new_policy);                \
 421         policy->user_policy.object = policy->object;                    \
 422         mutex_unlock(&policy->lock);                                    \
 423         unlock_cpu_hotplug();                                           \
 424                                                                         \
 425         return ret ? ret : count;                                       \
 426 }
 427
 428 store_one(scaling_min_freq,min);
 429 store_one(scaling_max_freq,max);
 430
 431 /**
 432  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 433  */
 434 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy,
 435                                                         char *buf)
 436 {
 437         unsigned int cur_freq = cpufreq_get(policy->cpu);
 438         if (!cur_freq)
 439                 return sprintf(buf, "<unknown>");
 440         return sprintf(buf, "%u\n", cur_freq);
 441 }
 442
 443
 444 /**
 445  * show_scaling_governor - show the current policy for the specified CPU
 446  */
 447 static ssize_t show_scaling_governor (struct cpufreq_policy * policy,
 448                                                         char *buf)
 449 {
 450         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 451                 return sprintf(buf, "powersave\n");
 452         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 453                 return sprintf(buf, "performance\n");
 454         else if (policy->governor)
 455                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 456         return -EINVAL;
 457 }
 458
 459
 460 /**
 461  * store_scaling_governor - store policy for the specified CPU
 462  */
 463 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 464                                        const char *buf, size_t count)
 465 {
 466         unsigned int ret = -EINVAL;
 467         char    str_governor[16];
 468         struct cpufreq_policy new_policy;
 469
 470         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 471         if (ret)
 472                 return ret;
 473
 474         ret = sscanf (buf, "%15s", str_governor);
 475         if (ret != 1)
 476                 return -EINVAL;
 477
 478         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
 479                                                 &new_policy.governor))
 480                 return -EINVAL;
 481
 482         lock_cpu_hotplug();
 483
 484         /* Do not use cpufreq_set_policy here or the user_policy.max
 485            will be wrongly overridden */
 486         mutex_lock(&policy->lock);
 487         ret = __cpufreq_set_policy(policy, &new_policy);
 488
 489         policy->user_policy.policy = policy->policy;
 490         policy->user_policy.governor = policy->governor;
 491         mutex_unlock(&policy->lock);
 492
 493         unlock_cpu_hotplug();
 494
 495         if (ret)
 496                 return ret;
 497         else
 498                 return count;
 499 }
 500
 501 /**
 502  * show_scaling_driver - show the cpufreq driver currently loaded
 503  */
 504 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 505 {
 506         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 507 }
 508
 509 /**
 510  * show_scaling_available_governors - show the available CPUfreq governors
 511  */
 512 static ssize_t show_scaling_available_governors (struct cpufreq_policy *policy,
 513                                 char *buf)
 514 {
 515         ssize_t i = 0;
 516         struct cpufreq_governor *t;
 517
 518         if (!cpufreq_driver->target) {
 519                 i += sprintf(buf, "performance powersave");
 520                 goto out;
 521         }
 522
 523         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 524                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 525                         goto out;
 526                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 527         }
 528 out:
 529         i += sprintf(&buf[i], "\n");
 530         return i;
 531 }
 532 /**
 533  * show_affected_cpus - show the CPUs affected by each transition
 534  */
 535 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 536 {
 537         ssize_t i = 0;
 538         unsigned int cpu;
 539
 540         for_each_cpu_mask(cpu, policy->cpus) {
 541                 if (i)
 542                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 543                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 544                 if (i >= (PAGE_SIZE - 5))
 545                     break;
 546         }
 547         i += sprintf(&buf[i], "\n");
 548         return i;
 549 }
 550
 551
 552 #define define_one_ro(_name) \
 553 static struct freq_attr _name = \
 554 __ATTR(_name, 0444, show_##_name, NULL)
 555
 556 #define define_one_ro0400(_name) \
 557 static struct freq_attr _name = \
 558 __ATTR(_name, 0400, show_##_name, NULL)
 559
 560 #define define_one_rw(_name) \
 561 static struct freq_attr _name = \
 562 __ATTR(_name, 0644, show_##_name, store_##_name)
 563
 564 define_one_ro0400(cpuinfo_cur_freq);
 565 define_one_ro(cpuinfo_min_freq);
 566 define_one_ro(cpuinfo_max_freq);
 567 define_one_ro(scaling_available_governors);
 568 define_one_ro(scaling_driver);
 569 define_one_ro(scaling_cur_freq);
 570 define_one_ro(affected_cpus);
 571 define_one_rw(scaling_min_freq);
 572 define_one_rw(scaling_max_freq);
 573 define_one_rw(scaling_governor);
 574
 575 static struct attribute * default_attrs[] = {
 576         &cpuinfo_min_freq.attr,
 577         &cpuinfo_max_freq.attr,
 578         &scaling_min_freq.attr,
 579         &scaling_max_freq.attr,
 580         &affected_cpus.attr,
 581         &scaling_governor.attr,
 582         &scaling_driver.attr,
 583         &scaling_available_governors.attr,
 584         NULL
 585 };
 586
 587 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 588 #define to_attr(a) container_of(a,struct freq_attr,attr)
 589
 590 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 591 {
 592         struct cpufreq_policy * policy = to_policy(kobj);
 593         struct freq_attr * fattr = to_attr(attr);
 594         ssize_t ret;
 595         policy = cpufreq_cpu_get(policy->cpu);
 596         if (!policy)
 597                 return -EINVAL;
 598         if (fattr->show)
 599                 ret = fattr->show(policy, buf);
 600         else
 601                 ret = -EIO;
 602
 603         cpufreq_cpu_put(policy);
 604         return ret;
 605 }
 606
 607 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 608                      const char * buf, size_t count)
 609 {
 610         struct cpufreq_policy * policy = to_policy(kobj);
 611         struct freq_attr * fattr = to_attr(attr);
 612         ssize_t ret;
 613         policy = cpufreq_cpu_get(policy->cpu);
 614         if (!policy)
 615                 return -EINVAL;
 616         if (fattr->store)
 617                 ret = fattr->store(policy, buf, count);
 618         else
 619                 ret = -EIO;
 620
 621         cpufreq_cpu_put(policy);
 622         return ret;
 623 }
 624
 625 static void cpufreq_sysfs_release(struct kobject * kobj)
 626 {
 627         struct cpufreq_policy * policy = to_policy(kobj);
 628         dprintk("last reference is dropped\n");
 629         complete(&policy->kobj_unregister);
 630 }
 631
 632 static struct sysfs_ops sysfs_ops = {
 633         .show   = show,
 634         .store  = store,
 635 };
 636
 637 static struct kobj_type ktype_cpufreq = {
 638         .sysfs_ops      = &sysfs_ops,
 639         .default_attrs  = default_attrs,
 640         .release        = cpufreq_sysfs_release,
 641 };
 642
 643
 644 /**
 645  * cpufreq_add_dev - add a CPU device
 646  *
 647  * Adds the cpufreq interface for a CPU device.
 648  */
 649 static int cpufreq_add_dev (struct sys_device * sys_dev)
 650 {
 651         unsigned int cpu = sys_dev->id;
 652         int ret = 0;
 653         struct cpufreq_policy new_policy;
 654         struct cpufreq_policy *policy;
 655         struct freq_attr **drv_attr;
 656         struct sys_device *cpu_sys_dev;
 657         unsigned long flags;
 658         unsigned int j;
 659 #ifdef CONFIG_SMP
 660         struct cpufreq_policy *managed_policy;
 661 #endif
 662
 663         if (cpu_is_offline(cpu))
 664                 return 0;
 665
 666         cpufreq_debug_disable_ratelimit();
 667         dprintk("adding CPU %u\n", cpu);
 668
 669 #ifdef CONFIG_SMP
 670         /* check whether a different CPU already registered this
 671          * CPU because it is in the same boat. */
 672         policy = cpufreq_cpu_get(cpu);
 673         if (unlikely(policy)) {
 674                 cpufreq_cpu_put(policy);
 675                 cpufreq_debug_enable_ratelimit();
 676                 return 0;
 677         }
 678 #endif
 679
 680         if (!try_module_get(cpufreq_driver->owner)) {
 681                 ret = -EINVAL;
 682                 goto module_out;
 683         }
 684
 685         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 686         if (!policy) {
 687                 ret = -ENOMEM;
 688                 goto nomem_out;
 689         }
 690
 691         policy->cpu = cpu;
 692         policy->cpus = cpumask_of_cpu(cpu);
 693
 694         mutex_init(&policy->lock);
 695         mutex_lock(&policy->lock);
 696         init_completion(&policy->kobj_unregister);
 697         INIT_WORK(&policy->update, handle_update);
 698
 699         /* call driver. From then on the cpufreq must be able
 700          * to accept all calls to ->verify and ->setpolicy for this CPU
 701          */
 702         ret = cpufreq_driver->init(policy);
 703         if (ret) {
 704                 dprintk("initialization failed\n");
 705                 mutex_unlock(&policy->lock);
 706                 goto err_out;
 707         }
 708
 709 #ifdef CONFIG_SMP
 710         for_each_cpu_mask(j, policy->cpus) {
 711                 if (cpu == j)
 712                         continue;
 713
 714                 /* check for existing affected CPUs.  They may not be aware
 715                  * of it due to CPU Hotplug.
 716                  */
 717                 managed_policy = cpufreq_cpu_get(j);
 718                 if (unlikely(managed_policy)) {
 719                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 720                         managed_policy->cpus = policy->cpus;
 721                         cpufreq_cpu_data[cpu] = managed_policy;
 722                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 723
 724                         dprintk("CPU already managed, adding link\n");
 725                         sysfs_create_link(&sys_dev->kobj,
 726                                           &managed_policy->kobj, "cpufreq");
 727
 728                         cpufreq_debug_enable_ratelimit();
 729                         mutex_unlock(&policy->lock);
 730                         ret = 0;
 731                         goto err_out_driver_exit; /* call driver->exit() */
 732                 }
 733         }
 734 #endif
 735         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 736
 737         /* prepare interface data */
 738         policy->kobj.parent = &sys_dev->kobj;
 739         policy->kobj.ktype = &ktype_cpufreq;
 740         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 741
 742         ret = kobject_register(&policy->kobj);
 743         if (ret) {
 744                 mutex_unlock(&policy->lock);
 745                 goto err_out_driver_exit;
 746         }
 747         /* set up files for this cpu device */
 748         drv_attr = cpufreq_driver->attr;
 749         while ((drv_attr) && (*drv_attr)) {
 750                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 751                 drv_attr++;
 752         }
 753         if (cpufreq_driver->get)
 754                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 755         if (cpufreq_driver->target)
 756                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 757
 758         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 759         for_each_cpu_mask(j, policy->cpus)
 760                 cpufreq_cpu_data[j] = policy;
 761         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 762
 763         /* symlink affected CPUs */
 764         for_each_cpu_mask(j, policy->cpus) {
 765                 if (j == cpu)
 766                         continue;
 767                 if (!cpu_online(j))
 768                         continue;
 769
 770                 dprintk("CPU %u already managed, adding link\n", j);
 771                 cpufreq_cpu_get(cpu);
 772                 cpu_sys_dev = get_cpu_sysdev(j);
 773                 sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 774                                   "cpufreq");
 775         }
 776
 777         policy->governor = NULL; /* to assure that the starting sequence is
 778                                   * run in cpufreq_set_policy */
 779         mutex_unlock(&policy->lock);
 780
 781         /* set default policy */
 782         ret = cpufreq_set_policy(&new_policy);
 783         if (ret) {
 784                 dprintk("setting policy failed\n");
 785                 goto err_out_unregister;
 786         }
 787
 788         module_put(cpufreq_driver->owner);
 789         dprintk("initialization complete\n");
 790         cpufreq_debug_enable_ratelimit();
 791
 792         return 0;
 793
 794
 795 err_out_unregister:
 796         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 797         for_each_cpu_mask(j, policy->cpus)
 798                 cpufreq_cpu_data[j] = NULL;
 799         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 800
 801         kobject_unregister(&policy->kobj);
 802         wait_for_completion(&policy->kobj_unregister);
 803
 804 err_out_driver_exit:
 805         if (cpufreq_driver->exit)
 806                 cpufreq_driver->exit(policy);
 807
 808 err_out:
 809         kfree(policy);
 810
 811 nomem_out:
 812         module_put(cpufreq_driver->owner);
 813 module_out:
 814         cpufreq_debug_enable_ratelimit();
 815         return ret;
 816 }
 817
 818
 819 /**
 820  * cpufreq_remove_dev - remove a CPU device
 821  *
 822  * Removes the cpufreq interface for a CPU device.
 823  */
 824 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 825 {
 826         unsigned int cpu = sys_dev->id;
 827         unsigned long flags;
 828         struct cpufreq_policy *data;
 829 #ifdef CONFIG_SMP
 830         struct sys_device *cpu_sys_dev;
 831         unsigned int j;
 832 #endif
 833
 834         cpufreq_debug_disable_ratelimit();
 835         dprintk("unregistering CPU %u\n", cpu);
 836
 837         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 838         data = cpufreq_cpu_data[cpu];
 839
 840         if (!data) {
 841                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 842                 cpufreq_debug_enable_ratelimit();
 843                 return -EINVAL;
 844         }
 845         cpufreq_cpu_data[cpu] = NULL;
 846
 847
 848 #ifdef CONFIG_SMP
 849         /* if this isn't the CPU which is the parent of the kobj, we
 850          * only need to unlink, put and exit
 851          */
 852         if (unlikely(cpu != data->cpu)) {
 853                 dprintk("removing link\n");
 854                 cpu_clear(cpu, data->cpus);
 855                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 856                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 857                 cpufreq_cpu_put(data);
 858                 cpufreq_debug_enable_ratelimit();
 859                 return 0;
 860         }
 861 #endif
 862
 863
 864         if (!kobject_get(&data->kobj)) {
 865                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 866                 cpufreq_debug_enable_ratelimit();
 867                 return -EFAULT;
 868         }
 869
 870 #ifdef CONFIG_SMP
 871         /* if we have other CPUs still registered, we need to unlink them,
 872          * or else wait_for_completion below will lock up. Clean the
 873          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 874          * links afterwards.
 875          */
 876         if (unlikely(cpus_weight(data->cpus) > 1)) {
 877                 for_each_cpu_mask(j, data->cpus) {
 878                         if (j == cpu)
 879                                 continue;
 880                         cpufreq_cpu_data[j] = NULL;
 881                 }
 882         }
 883
 884         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 885
 886         if (unlikely(cpus_weight(data->cpus) > 1)) {
 887                 for_each_cpu_mask(j, data->cpus) {
 888                         if (j == cpu)
 889                                 continue;
 890                         dprintk("removing link for cpu %u\n", j);
 891                         cpu_sys_dev = get_cpu_sysdev(j);
 892                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 893                         cpufreq_cpu_put(data);
 894                 }
 895         }
 896 #else
 897         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 898 #endif
 899
 900         mutex_lock(&data->lock);
 901         if (cpufreq_driver->target)
 902                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 903         mutex_unlock(&data->lock);
 904
 905         kobject_unregister(&data->kobj);
 906
 907         kobject_put(&data->kobj);
 908
 909         /* we need to make sure that the underlying kobj is actually
 910          * not referenced anymore by anybody before we proceed with
 911          * unloading.
 912          */
 913         dprintk("waiting for dropping of refcount\n");
 914         wait_for_completion(&data->kobj_unregister);
 915         dprintk("wait complete\n");
 916
 917         if (cpufreq_driver->exit)
 918                 cpufreq_driver->exit(data);
 919
 920         kfree(data);
 921
 922         cpufreq_debug_enable_ratelimit();
 923         return 0;
 924 }
 925
 926
 927 static void handle_update(struct work_struct *work)
 928 {
 929         struct cpufreq_policy *policy =
 930                 container_of(work, struct cpufreq_policy, update);
 931         unsigned int cpu = policy->cpu;
 932         dprintk("handle_update for cpu %u called\n", cpu);
 933         cpufreq_update_policy(cpu);
 934 }
 935
 936 /**
 937  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 938  *      @cpu: cpu number
 939  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 940  *      @new_freq: CPU frequency the CPU actually runs at
 941  *
 942  *      We adjust to current frequency first, and need to clean up later. So either call
 943  *      to cpufreq_update_policy() or schedule handle_update()).
 944  */
 945 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
 946                                 unsigned int new_freq)
 947 {
 948         struct cpufreq_freqs freqs;
 949
 950         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
 951                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 952
 953         freqs.cpu = cpu;
 954         freqs.old = old_freq;
 955         freqs.new = new_freq;
 956         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 957         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 958 }
 959
 960
 961 /**
 962  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
 963  * @cpu: CPU number
 964  *
 965  * This is the last known freq, without actually getting it from the driver.
 966  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 967  */
 968 unsigned int cpufreq_quick_get(unsigned int cpu)
 969 {
 970         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 971         unsigned int ret_freq = 0;
 972
 973         if (policy) {
 974                 mutex_lock(&policy->lock);
 975                 ret_freq = policy->cur;
 976                 mutex_unlock(&policy->lock);
 977                 cpufreq_cpu_put(policy);
 978         }
 979
 980         return (ret_freq);
 981 }
 982 EXPORT_SYMBOL(cpufreq_quick_get);
 983
 984
 985 /**
 986  * cpufreq_get - get the current CPU frequency (in kHz)
 987  * @cpu: CPU number
 988  *
 989  * Get the CPU current (static) CPU frequency
 990  */
 991 unsigned int cpufreq_get(unsigned int cpu)
 992 {
 993         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 994         unsigned int ret_freq = 0;
 995
 996         if (!policy)
 997                 return 0;
 998
 999         if (!cpufreq_driver->get)
1000                 goto out;
1001
1002         mutex_lock(&policy->lock);
1003
1004         ret_freq = cpufreq_driver->get(cpu);
1005
1006         if (ret_freq && policy->cur &&
1007                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1008                 /* verify no discrepancy between actual and
1009                                         saved value exists */
1010                 if (unlikely(ret_freq != policy->cur)) {
1011                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1012                         schedule_work(&policy->update);
1013                 }
1014         }
1015
1016         mutex_unlock(&policy->lock);
1017
1018 out:
1019         cpufreq_cpu_put(policy);
1020
1021         return (ret_freq);
1022 }
1023 EXPORT_SYMBOL(cpufreq_get);
1024
1025
1026 /**
1027  *      cpufreq_suspend - let the low level driver prepare for suspend
1028  */
1029
1030 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
1031 {
1032         int cpu = sysdev->id;
1033         int ret = 0;
1034         unsigned int cur_freq = 0;
1035         struct cpufreq_policy *cpu_policy;
1036
1037         dprintk("suspending cpu %u\n", cpu);
1038
1039         if (!cpu_online(cpu))
1040                 return 0;
1041
1042         /* we may be lax here as interrupts are off. Nonetheless
1043          * we need to grab the correct cpu policy, as to check
1044          * whether we really run on this CPU.
1045          */
1046
1047         cpu_policy = cpufreq_cpu_get(cpu);
1048         if (!cpu_policy)
1049                 return -EINVAL;
1050
1051         /* only handle each CPU group once */
1052         if (unlikely(cpu_policy->cpu != cpu)) {
1053                 cpufreq_cpu_put(cpu_policy);
1054                 return 0;
1055         }
1056
1057         if (cpufreq_driver->suspend) {
1058                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1059                 if (ret) {
1060                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1061                                         "step on CPU %u\n", cpu_policy->cpu);
1062                         cpufreq_cpu_put(cpu_policy);
1063                         return ret;
1064                 }
1065         }
1066
1067
1068         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1069                 goto out;
1070
1071         if (cpufreq_driver->get)
1072                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1073
1074         if (!cur_freq || !cpu_policy->cur) {
1075                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1076                        "frequency is what timing core thinks it is.\n");
1077                 goto out;
1078         }
1079
1080         if (unlikely(cur_freq != cpu_policy->cur)) {
1081                 struct cpufreq_freqs freqs;
1082
1083                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1084                         dprintk("Warning: CPU frequency is %u, "
1085                                "cpufreq assumed %u kHz.\n",
1086                                cur_freq, cpu_policy->cur);
1087
1088                 freqs.cpu = cpu;
1089                 freqs.old = cpu_policy->cur;
1090                 freqs.new = cur_freq;
1091
1092                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
1093                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1094                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1095
1096                 cpu_policy->cur = cur_freq;
1097         }
1098
1099 out:
1100         cpufreq_cpu_put(cpu_policy);
1101         return 0;
1102 }
1103
1104 /**
1105  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1106  *
1107  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1108  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1109  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1110  *          restored.
1111  */
1112 static int cpufreq_resume(struct sys_device * sysdev)
1113 {
1114         int cpu = sysdev->id;
1115         int ret = 0;
1116         struct cpufreq_policy *cpu_policy;
1117
1118         dprintk("resuming cpu %u\n", cpu);
1119
1120         if (!cpu_online(cpu))
1121                 return 0;
1122
1123         /* we may be lax here as interrupts are off. Nonetheless
1124          * we need to grab the correct cpu policy, as to check
1125          * whether we really run on this CPU.
1126          */
1127
1128         cpu_policy = cpufreq_cpu_get(cpu);
1129         if (!cpu_policy)
1130                 return -EINVAL;
1131
1132         /* only handle each CPU group once */
1133         if (unlikely(cpu_policy->cpu != cpu)) {
1134                 cpufreq_cpu_put(cpu_policy);
1135                 return 0;
1136         }
1137
1138         if (cpufreq_driver->resume) {
1139                 ret = cpufreq_driver->resume(cpu_policy);
1140                 if (ret) {
1141                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1142                                         "step on CPU %u\n", cpu_policy->cpu);
1143                         cpufreq_cpu_put(cpu_policy);
1144                         return ret;
1145                 }
1146         }
1147
1148         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1149                 unsigned int cur_freq = 0;
1150
1151                 if (cpufreq_driver->get)
1152                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1153
1154                 if (!cur_freq || !cpu_policy->cur) {
1155                         printk(KERN_ERR "cpufreq: resume failed to assert "
1156                                         "current frequency is what timing core "
1157                                         "thinks it is.\n");
1158                         goto out;
1159                 }
1160
1161                 if (unlikely(cur_freq != cpu_policy->cur)) {
1162                         struct cpufreq_freqs freqs;
1163
1164                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1165                                 dprintk("Warning: CPU frequency"
1166                                        "is %u, cpufreq assumed %u kHz.\n",
1167                                        cur_freq, cpu_policy->cur);
1168
1169                         freqs.cpu = cpu;
1170                         freqs.old = cpu_policy->cur;
1171                         freqs.new = cur_freq;
1172
1173                         srcu_notifier_call_chain(
1174                                         &cpufreq_transition_notifier_list,
1175                                         CPUFREQ_RESUMECHANGE, &freqs);
1176                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1177
1178                         cpu_policy->cur = cur_freq;
1179                 }
1180         }
1181
1182 out:
1183         schedule_work(&cpu_policy->update);
1184         cpufreq_cpu_put(cpu_policy);
1185         return ret;
1186 }
1187
1188 static struct sysdev_driver cpufreq_sysdev_driver = {
1189         .add            = cpufreq_add_dev,
1190         .remove         = cpufreq_remove_dev,
1191         .suspend        = cpufreq_suspend,
1192         .resume         = cpufreq_resume,
1193 };
1194
1195
1196 /*********************************************************************
1197  *                     NOTIFIER LISTS INTERFACE                      *
1198  *********************************************************************/
1199
1200 /**
1201  *      cpufreq_register_notifier - register a driver with cpufreq
1202  *      @nb: notifier function to register
1203  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1204  *
1205  *      Add a driver to one of two lists: either a list of drivers that
1206  *      are notified about clock rate changes (once before and once after
1207  *      the transition), or a list of drivers that are notified about
1208  *      changes in cpufreq policy.
1209  *
1210  *      This function may sleep, and has the same return conditions as
1211  *      blocking_notifier_chain_register.
1212  */
1213 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1214 {
1215         int ret;
1216
1217         switch (list) {
1218         case CPUFREQ_TRANSITION_NOTIFIER:
1219                 ret = srcu_notifier_chain_register(
1220                                 &cpufreq_transition_notifier_list, nb);
1221                 break;
1222         case CPUFREQ_POLICY_NOTIFIER:
1223                 ret = blocking_notifier_chain_register(
1224                                 &cpufreq_policy_notifier_list, nb);
1225                 break;
1226         default:
1227                 ret = -EINVAL;
1228         }
1229
1230         return ret;
1231 }
1232 EXPORT_SYMBOL(cpufreq_register_notifier);
1233
1234
1235 /**
1236  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1237  *      @nb: notifier block to be unregistered
1238  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1239  *
1240  *      Remove a driver from the CPU frequency notifier list.
1241  *
1242  *      This function may sleep, and has the same return conditions as
1243  *      blocking_notifier_chain_unregister.
1244  */
1245 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1246 {
1247         int ret;
1248
1249         switch (list) {
1250         case CPUFREQ_TRANSITION_NOTIFIER:
1251                 ret = srcu_notifier_chain_unregister(
1252                                 &cpufreq_transition_notifier_list, nb);
1253                 break;
1254         case CPUFREQ_POLICY_NOTIFIER:
1255                 ret = blocking_notifier_chain_unregister(
1256                                 &cpufreq_policy_notifier_list, nb);
1257                 break;
1258         default:
1259                 ret = -EINVAL;
1260         }
1261
1262         return ret;
1263 }
1264 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1265
1266
1267 /*********************************************************************
1268  *                              GOVERNORS                            *
1269  *********************************************************************/
1270
1271
1272 /* Must be called with lock_cpu_hotplug held */
1273 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1274                             unsigned int target_freq,
1275                             unsigned int relation)
1276 {
1277         int retval = -EINVAL;
1278
1279         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1280                 target_freq, relation);
1281         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1282                 retval = cpufreq_driver->target(policy, target_freq, relation);
1283
1284         return retval;
1285 }
1286 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1287
1288 int cpufreq_driver_target(struct cpufreq_policy *policy,
1289                           unsigned int target_freq,
1290                           unsigned int relation)
1291 {
1292         int ret;
1293
1294         policy = cpufreq_cpu_get(policy->cpu);
1295         if (!policy)
1296                 return -EINVAL;
1297
1298         lock_cpu_hotplug();
1299         mutex_lock(&policy->lock);
1300
1301         ret = __cpufreq_driver_target(policy, target_freq, relation);
1302
1303         mutex_unlock(&policy->lock);
1304         unlock_cpu_hotplug();
1305
1306         cpufreq_cpu_put(policy);
1307         return ret;
1308 }
1309 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1310
1311 int cpufreq_driver_getavg(struct cpufreq_policy *policy)
1312 {
1313         int ret = 0;
1314
1315         policy = cpufreq_cpu_get(policy->cpu);
1316         if (!policy)
1317                 return -EINVAL;
1318
1319         mutex_lock(&policy->lock);
1320
1321         if (cpu_online(policy->cpu) && cpufreq_driver->getavg)
1322                 ret = cpufreq_driver->getavg(policy->cpu);
1323
1324         mutex_unlock(&policy->lock);
1325
1326         cpufreq_cpu_put(policy);
1327         return ret;
1328 }
1329 EXPORT_SYMBOL_GPL(cpufreq_driver_getavg);
1330
1331 /*
1332  * Locking: Must be called with the lock_cpu_hotplug() lock held
1333  * when "event" is CPUFREQ_GOV_LIMITS
1334  */
1335
1336 static int __cpufreq_governor(struct cpufreq_policy *policy,
1337                                         unsigned int event)
1338 {
1339         int ret;
1340
1341         if (!try_module_get(policy->governor->owner))
1342                 return -EINVAL;
1343
1344         dprintk("__cpufreq_governor for CPU %u, event %u\n",
1345                                                 policy->cpu, event);
1346         ret = policy->governor->governor(policy, event);
1347
1348         /* we keep one module reference alive for
1349                         each CPU governed by this CPU */
1350         if ((event != CPUFREQ_GOV_START) || ret)
1351                 module_put(policy->governor->owner);
1352         if ((event == CPUFREQ_GOV_STOP) && !ret)
1353                 module_put(policy->governor->owner);
1354
1355         return ret;
1356 }
1357
1358
1359 int cpufreq_register_governor(struct cpufreq_governor *governor)
1360 {
1361         int err;
1362
1363         if (!governor)
1364                 return -EINVAL;
1365
1366         mutex_lock(&cpufreq_governor_mutex);
1367
1368         err = -EBUSY;
1369         if (__find_governor(governor->name) == NULL) {
1370                 err = 0;
1371                 list_add(&governor->governor_list, &cpufreq_governor_list);
1372         }
1373
1374         mutex_unlock(&cpufreq_governor_mutex);
1375         return err;
1376 }
1377 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1378
1379
1380 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1381 {
1382         if (!governor)
1383                 return;
1384
1385         mutex_lock(&cpufreq_governor_mutex);
1386         list_del(&governor->governor_list);
1387         mutex_unlock(&cpufreq_governor_mutex);
1388         return;
1389 }
1390 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1391
1392
1393
1394 /*********************************************************************
1395  *                          POLICY INTERFACE                         *
1396  *********************************************************************/
1397
1398 /**
1399  * cpufreq_get_policy - get the current cpufreq_policy
1400  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1401  *
1402  * Reads the current cpufreq policy.
1403  */
1404 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1405 {
1406         struct cpufreq_policy *cpu_policy;
1407         if (!policy)
1408                 return -EINVAL;
1409
1410         cpu_policy = cpufreq_cpu_get(cpu);
1411         if (!cpu_policy)
1412                 return -EINVAL;
1413
1414         mutex_lock(&cpu_policy->lock);
1415         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1416         mutex_unlock(&cpu_policy->lock);
1417
1418         cpufreq_cpu_put(cpu_policy);
1419         return 0;
1420 }
1421 EXPORT_SYMBOL(cpufreq_get_policy);
1422
1423
1424 /*
1425  * data   : current policy.
1426  * policy : policy to be set.
1427  * Locking: Must be called with the lock_cpu_hotplug() lock held
1428  */
1429 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1430                                 struct cpufreq_policy *policy)
1431 {
1432         int ret = 0;
1433
1434         cpufreq_debug_disable_ratelimit();
1435         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1436                 policy->min, policy->max);
1437
1438         memcpy(&policy->cpuinfo, &data->cpuinfo,
1439                                 sizeof(struct cpufreq_cpuinfo));
1440
1441         if (policy->min > data->min && policy->min > policy->max) {
1442                 ret = -EINVAL;
1443                 goto error_out;
1444         }
1445
1446         /* verify the cpu speed can be set within this limit */
1447         ret = cpufreq_driver->verify(policy);
1448         if (ret)
1449                 goto error_out;
1450
1451         /* adjust if necessary - all reasons */
1452         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1453                         CPUFREQ_ADJUST, policy);
1454
1455         /* adjust if necessary - hardware incompatibility*/
1456         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1457                         CPUFREQ_INCOMPATIBLE, policy);
1458
1459         /* verify the cpu speed can be set within this limit,
1460            which might be different to the first one */
1461         ret = cpufreq_driver->verify(policy);
1462         if (ret)
1463                 goto error_out;
1464
1465         /* notification of the new policy */
1466         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1467                         CPUFREQ_NOTIFY, policy);
1468
1469         data->min = policy->min;
1470         data->max = policy->max;
1471
1472         dprintk("new min and max freqs are %u - %u kHz\n",
1473                                         data->min, data->max);
1474
1475         if (cpufreq_driver->setpolicy) {
1476                 data->policy = policy->policy;
1477                 dprintk("setting range\n");
1478                 ret = cpufreq_driver->setpolicy(policy);
1479         } else {
1480                 if (policy->governor != data->governor) {
1481                         /* save old, working values */
1482                         struct cpufreq_governor *old_gov = data->governor;
1483
1484                         dprintk("governor switch\n");
1485
1486                         /* end old governor */
1487                         if (data->governor)
1488                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1489
1490                         /* start new governor */
1491                         data->governor = policy->governor;
1492                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1493                                 /* new governor failed, so re-start old one */
1494                                 dprintk("starting governor %s failed\n",
1495                                                         data->governor->name);
1496                                 if (old_gov) {
1497                                         data->governor = old_gov;
1498                                         __cpufreq_governor(data,
1499                                                            CPUFREQ_GOV_START);
1500                                 }
1501                                 ret = -EINVAL;
1502                                 goto error_out;
1503                         }
1504                         /* might be a policy change, too, so fall through */
1505                 }
1506                 dprintk("governor: change or update limits\n");
1507                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1508         }
1509
1510 error_out:
1511         cpufreq_debug_enable_ratelimit();
1512         return ret;
1513 }
1514
1515 /**
1516  *      cpufreq_set_policy - set a new CPUFreq policy
1517  *      @policy: policy to be set.
1518  *
1519  *      Sets a new CPU frequency and voltage scaling policy.
1520  */
1521 int cpufreq_set_policy(struct cpufreq_policy *policy)
1522 {
1523         int ret = 0;
1524         struct cpufreq_policy *data;
1525
1526         if (!policy)
1527                 return -EINVAL;
1528
1529         data = cpufreq_cpu_get(policy->cpu);
1530         if (!data)
1531                 return -EINVAL;
1532
1533         lock_cpu_hotplug();
1534
1535         /* lock this CPU */
1536         mutex_lock(&data->lock);
1537
1538         ret = __cpufreq_set_policy(data, policy);
1539         data->user_policy.min = data->min;
1540         data->user_policy.max = data->max;
1541         data->user_policy.policy = data->policy;
1542         data->user_policy.governor = data->governor;
1543
1544         mutex_unlock(&data->lock);
1545
1546         unlock_cpu_hotplug();
1547         cpufreq_cpu_put(data);
1548
1549         return ret;
1550 }
1551 EXPORT_SYMBOL(cpufreq_set_policy);
1552
1553
1554 /**
1555  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1556  *      @cpu: CPU which shall be re-evaluated
1557  *
1558  *      Usefull for policy notifiers which have different necessities
1559  *      at different times.
1560  */
1561 int cpufreq_update_policy(unsigned int cpu)
1562 {
1563         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1564         struct cpufreq_policy policy;
1565         int ret = 0;
1566
1567         if (!data)
1568                 return -ENODEV;
1569
1570         lock_cpu_hotplug();
1571         mutex_lock(&data->lock);
1572
1573         dprintk("updating policy for CPU %u\n", cpu);
1574         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1575         policy.min = data->user_policy.min;
1576         policy.max = data->user_policy.max;
1577         policy.policy = data->user_policy.policy;
1578         policy.governor = data->user_policy.governor;
1579
1580         /* BIOS might change freq behind our back
1581           -> ask driver for current freq and notify governors about a change */
1582         if (cpufreq_driver->get) {
1583                 policy.cur = cpufreq_driver->get(cpu);
1584                 if (!data->cur) {
1585                         dprintk("Driver did not initialize current freq");
1586                         data->cur = policy.cur;
1587                 } else {
1588                         if (data->cur != policy.cur)
1589                                 cpufreq_out_of_sync(cpu, data->cur,
1590                                                                 policy.cur);
1591                 }
1592         }
1593
1594         ret = __cpufreq_set_policy(data, &policy);
1595
1596         mutex_unlock(&data->lock);
1597         unlock_cpu_hotplug();
1598         cpufreq_cpu_put(data);
1599         return ret;
1600 }
1601 EXPORT_SYMBOL(cpufreq_update_policy);
1602
1603 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1604                                         unsigned long action, void *hcpu)
1605 {
1606         unsigned int cpu = (unsigned long)hcpu;
1607         struct cpufreq_policy *policy;
1608         struct sys_device *sys_dev;
1609
1610         sys_dev = get_cpu_sysdev(cpu);
1611
1612         if (sys_dev) {
1613                 switch (action) {
1614                 case CPU_ONLINE:
1615                         cpufreq_add_dev(sys_dev);
1616                         break;
1617                 case CPU_DOWN_PREPARE:
1618                         /*
1619                          * We attempt to put this cpu in lowest frequency
1620                          * possible before going down. This will permit
1621                          * hardware-managed P-State to switch other related
1622                          * threads to min or higher speeds if possible.
1623                          */
1624                         policy = cpufreq_cpu_data[cpu];
1625                         if (policy) {
1626                                 cpufreq_driver_target(policy, policy->min,
1627                                                 CPUFREQ_RELATION_H);
1628                         }
1629                         break;
1630                 case CPU_DEAD:
1631                         cpufreq_remove_dev(sys_dev);
1632                         break;
1633                 }
1634         }
1635         return NOTIFY_OK;
1636 }
1637
1638 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1639 {
1640     .notifier_call = cpufreq_cpu_callback,
1641 };
1642
1643 /*********************************************************************
1644  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1645  *********************************************************************/
1646
1647 /**
1648  * cpufreq_register_driver - register a CPU Frequency driver
1649  * @driver_data: A struct cpufreq_driver containing the values#
1650  * submitted by the CPU Frequency driver.
1651  *
1652  *   Registers a CPU Frequency driver to this core code. This code
1653  * returns zero on success, -EBUSY when another driver got here first
1654  * (and isn't unregistered in the meantime).
1655  *
1656  */
1657 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1658 {
1659         unsigned long flags;
1660         int ret;
1661
1662         if (!driver_data || !driver_data->verify || !driver_data->init ||
1663             ((!driver_data->setpolicy) && (!driver_data->target)))
1664                 return -EINVAL;
1665
1666         dprintk("trying to register driver %s\n", driver_data->name);
1667
1668         if (driver_data->setpolicy)
1669                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1670
1671         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1672         if (cpufreq_driver) {
1673                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1674                 return -EBUSY;
1675         }
1676         cpufreq_driver = driver_data;
1677         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1678
1679         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1680
1681         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1682                 int i;
1683                 ret = -ENODEV;
1684
1685                 /* check for at least one working CPU */
1686                 for (i=0; i<NR_CPUS; i++)
1687                         if (cpufreq_cpu_data[i])
1688                                 ret = 0;
1689
1690                 /* if all ->init() calls failed, unregister */
1691                 if (ret) {
1692                         dprintk("no CPU initialized for driver %s\n",
1693                                                         driver_data->name);
1694                         sysdev_driver_unregister(&cpu_sysdev_class,
1695                                                 &cpufreq_sysdev_driver);
1696
1697                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1698                         cpufreq_driver = NULL;
1699                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1700                 }
1701         }
1702
1703         if (!ret) {
1704                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1705                 dprintk("driver %s up and running\n", driver_data->name);
1706                 cpufreq_debug_enable_ratelimit();
1707         }
1708
1709         return (ret);
1710 }
1711 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1712
1713
1714 /**
1715  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1716  *
1717  *    Unregister the current CPUFreq driver. Only call this if you have
1718  * the right to do so, i.e. if you have succeeded in initialising before!
1719  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1720  * currently not initialised.
1721  */
1722 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1723 {
1724         unsigned long flags;
1725
1726         cpufreq_debug_disable_ratelimit();
1727
1728         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1729                 cpufreq_debug_enable_ratelimit();
1730                 return -EINVAL;
1731         }
1732
1733         dprintk("unregistering driver %s\n", driver->name);
1734
1735         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1736         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1737
1738         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1739         cpufreq_driver = NULL;
1740         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1741
1742         return 0;
1743 }
1744 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);