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                         ret = sysfs_create_link(&sys_dev->kobj,
 726                                                 &managed_policy->kobj,
 727                                                 "cpufreq");
 728                         if (ret) {
 729                                 mutex_unlock(&policy->lock);
 730                                 goto err_out_driver_exit;
 731                         }
 732
 733                         cpufreq_debug_enable_ratelimit();
 734                         mutex_unlock(&policy->lock);
 735                         ret = 0;
 736                         goto err_out_driver_exit; /* call driver->exit() */
 737                 }
 738         }
 739 #endif
 740         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 741
 742         /* prepare interface data */
 743         policy->kobj.parent = &sys_dev->kobj;
 744         policy->kobj.ktype = &ktype_cpufreq;
 745         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 746
 747         ret = kobject_register(&policy->kobj);
 748         if (ret) {
 749                 mutex_unlock(&policy->lock);
 750                 goto err_out_driver_exit;
 751         }
 752         /* set up files for this cpu device */
 753         drv_attr = cpufreq_driver->attr;
 754         while ((drv_attr) && (*drv_attr)) {
 755                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 756                 drv_attr++;
 757         }
 758         if (cpufreq_driver->get)
 759                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 760         if (cpufreq_driver->target)
 761                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 762
 763         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 764         for_each_cpu_mask(j, policy->cpus)
 765                 cpufreq_cpu_data[j] = policy;
 766         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 767
 768         /* symlink affected CPUs */
 769         for_each_cpu_mask(j, policy->cpus) {
 770                 if (j == cpu)
 771                         continue;
 772                 if (!cpu_online(j))
 773                         continue;
 774
 775                 dprintk("CPU %u already managed, adding link\n", j);
 776                 cpufreq_cpu_get(cpu);
 777                 cpu_sys_dev = get_cpu_sysdev(j);
 778                 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 779                                         "cpufreq");
 780                 if (ret) {
 781                         mutex_unlock(&policy->lock);
 782                         goto err_out_unregister;
 783                 }
 784         }
 785
 786         policy->governor = NULL; /* to assure that the starting sequence is
 787                                   * run in cpufreq_set_policy */
 788         mutex_unlock(&policy->lock);
 789
 790         /* set default policy */
 791         ret = cpufreq_set_policy(&new_policy);
 792         if (ret) {
 793                 dprintk("setting policy failed\n");
 794                 goto err_out_unregister;
 795         }
 796
 797         module_put(cpufreq_driver->owner);
 798         dprintk("initialization complete\n");
 799         cpufreq_debug_enable_ratelimit();
 800
 801         return 0;
 802
 803
 804 err_out_unregister:
 805         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 806         for_each_cpu_mask(j, policy->cpus)
 807                 cpufreq_cpu_data[j] = NULL;
 808         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 809
 810         kobject_unregister(&policy->kobj);
 811         wait_for_completion(&policy->kobj_unregister);
 812
 813 err_out_driver_exit:
 814         if (cpufreq_driver->exit)
 815                 cpufreq_driver->exit(policy);
 816
 817 err_out:
 818         kfree(policy);
 819
 820 nomem_out:
 821         module_put(cpufreq_driver->owner);
 822 module_out:
 823         cpufreq_debug_enable_ratelimit();
 824         return ret;
 825 }
 826
 827
 828 /**
 829  * cpufreq_remove_dev - remove a CPU device
 830  *
 831  * Removes the cpufreq interface for a CPU device.
 832  */
 833 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 834 {
 835         unsigned int cpu = sys_dev->id;
 836         unsigned long flags;
 837         struct cpufreq_policy *data;
 838 #ifdef CONFIG_SMP
 839         struct sys_device *cpu_sys_dev;
 840         unsigned int j;
 841 #endif
 842
 843         cpufreq_debug_disable_ratelimit();
 844         dprintk("unregistering CPU %u\n", cpu);
 845
 846         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 847         data = cpufreq_cpu_data[cpu];
 848
 849         if (!data) {
 850                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 851                 cpufreq_debug_enable_ratelimit();
 852                 return -EINVAL;
 853         }
 854         cpufreq_cpu_data[cpu] = NULL;
 855
 856
 857 #ifdef CONFIG_SMP
 858         /* if this isn't the CPU which is the parent of the kobj, we
 859          * only need to unlink, put and exit
 860          */
 861         if (unlikely(cpu != data->cpu)) {
 862                 dprintk("removing link\n");
 863                 cpu_clear(cpu, data->cpus);
 864                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 865                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 866                 cpufreq_cpu_put(data);
 867                 cpufreq_debug_enable_ratelimit();
 868                 return 0;
 869         }
 870 #endif
 871
 872
 873         if (!kobject_get(&data->kobj)) {
 874                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 875                 cpufreq_debug_enable_ratelimit();
 876                 return -EFAULT;
 877         }
 878
 879 #ifdef CONFIG_SMP
 880         /* if we have other CPUs still registered, we need to unlink them,
 881          * or else wait_for_completion below will lock up. Clean the
 882          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 883          * links afterwards.
 884          */
 885         if (unlikely(cpus_weight(data->cpus) > 1)) {
 886                 for_each_cpu_mask(j, data->cpus) {
 887                         if (j == cpu)
 888                                 continue;
 889                         cpufreq_cpu_data[j] = NULL;
 890                 }
 891         }
 892
 893         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 894
 895         if (unlikely(cpus_weight(data->cpus) > 1)) {
 896                 for_each_cpu_mask(j, data->cpus) {
 897                         if (j == cpu)
 898                                 continue;
 899                         dprintk("removing link for cpu %u\n", j);
 900                         cpu_sys_dev = get_cpu_sysdev(j);
 901                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 902                         cpufreq_cpu_put(data);
 903                 }
 904         }
 905 #else
 906         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 907 #endif
 908
 909         mutex_lock(&data->lock);
 910         if (cpufreq_driver->target)
 911                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 912         mutex_unlock(&data->lock);
 913
 914         kobject_unregister(&data->kobj);
 915
 916         kobject_put(&data->kobj);
 917
 918         /* we need to make sure that the underlying kobj is actually
 919          * not referenced anymore by anybody before we proceed with
 920          * unloading.
 921          */
 922         dprintk("waiting for dropping of refcount\n");
 923         wait_for_completion(&data->kobj_unregister);
 924         dprintk("wait complete\n");
 925
 926         if (cpufreq_driver->exit)
 927                 cpufreq_driver->exit(data);
 928
 929         kfree(data);
 930
 931         cpufreq_debug_enable_ratelimit();
 932         return 0;
 933 }
 934
 935
 936 static void handle_update(struct work_struct *work)
 937 {
 938         struct cpufreq_policy *policy =
 939                 container_of(work, struct cpufreq_policy, update);
 940         unsigned int cpu = policy->cpu;
 941         dprintk("handle_update for cpu %u called\n", cpu);
 942         cpufreq_update_policy(cpu);
 943 }
 944
 945 /**
 946  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 947  *      @cpu: cpu number
 948  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 949  *      @new_freq: CPU frequency the CPU actually runs at
 950  *
 951  *      We adjust to current frequency first, and need to clean up later. So either call
 952  *      to cpufreq_update_policy() or schedule handle_update()).
 953  */
 954 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
 955                                 unsigned int new_freq)
 956 {
 957         struct cpufreq_freqs freqs;
 958
 959         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
 960                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 961
 962         freqs.cpu = cpu;
 963         freqs.old = old_freq;
 964         freqs.new = new_freq;
 965         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 966         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 967 }
 968
 969
 970 /**
 971  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
 972  * @cpu: CPU number
 973  *
 974  * This is the last known freq, without actually getting it from the driver.
 975  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 976  */
 977 unsigned int cpufreq_quick_get(unsigned int cpu)
 978 {
 979         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 980         unsigned int ret_freq = 0;
 981
 982         if (policy) {
 983                 mutex_lock(&policy->lock);
 984                 ret_freq = policy->cur;
 985                 mutex_unlock(&policy->lock);
 986                 cpufreq_cpu_put(policy);
 987         }
 988
 989         return (ret_freq);
 990 }
 991 EXPORT_SYMBOL(cpufreq_quick_get);
 992
 993
 994 /**
 995  * cpufreq_get - get the current CPU frequency (in kHz)
 996  * @cpu: CPU number
 997  *
 998  * Get the CPU current (static) CPU frequency
 999  */
1000 unsigned int cpufreq_get(unsigned int cpu)
1001 {
1002         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1003         unsigned int ret_freq = 0;
1004
1005         if (!policy)
1006                 return 0;
1007
1008         if (!cpufreq_driver->get)
1009                 goto out;
1010
1011         mutex_lock(&policy->lock);
1012
1013         ret_freq = cpufreq_driver->get(cpu);
1014
1015         if (ret_freq && policy->cur &&
1016                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1017                 /* verify no discrepancy between actual and
1018                                         saved value exists */
1019                 if (unlikely(ret_freq != policy->cur)) {
1020                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1021                         schedule_work(&policy->update);
1022                 }
1023         }
1024
1025         mutex_unlock(&policy->lock);
1026
1027 out:
1028         cpufreq_cpu_put(policy);
1029
1030         return (ret_freq);
1031 }
1032 EXPORT_SYMBOL(cpufreq_get);
1033
1034
1035 /**
1036  *      cpufreq_suspend - let the low level driver prepare for suspend
1037  */
1038
1039 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
1040 {
1041         int cpu = sysdev->id;
1042         int ret = 0;
1043         unsigned int cur_freq = 0;
1044         struct cpufreq_policy *cpu_policy;
1045
1046         dprintk("suspending cpu %u\n", cpu);
1047
1048         if (!cpu_online(cpu))
1049                 return 0;
1050
1051         /* we may be lax here as interrupts are off. Nonetheless
1052          * we need to grab the correct cpu policy, as to check
1053          * whether we really run on this CPU.
1054          */
1055
1056         cpu_policy = cpufreq_cpu_get(cpu);
1057         if (!cpu_policy)
1058                 return -EINVAL;
1059
1060         /* only handle each CPU group once */
1061         if (unlikely(cpu_policy->cpu != cpu)) {
1062                 cpufreq_cpu_put(cpu_policy);
1063                 return 0;
1064         }
1065
1066         if (cpufreq_driver->suspend) {
1067                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1068                 if (ret) {
1069                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1070                                         "step on CPU %u\n", cpu_policy->cpu);
1071                         cpufreq_cpu_put(cpu_policy);
1072                         return ret;
1073                 }
1074         }
1075
1076
1077         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1078                 goto out;
1079
1080         if (cpufreq_driver->get)
1081                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1082
1083         if (!cur_freq || !cpu_policy->cur) {
1084                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1085                        "frequency is what timing core thinks it is.\n");
1086                 goto out;
1087         }
1088
1089         if (unlikely(cur_freq != cpu_policy->cur)) {
1090                 struct cpufreq_freqs freqs;
1091
1092                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1093                         dprintk("Warning: CPU frequency is %u, "
1094                                "cpufreq assumed %u kHz.\n",
1095                                cur_freq, cpu_policy->cur);
1096
1097                 freqs.cpu = cpu;
1098                 freqs.old = cpu_policy->cur;
1099                 freqs.new = cur_freq;
1100
1101                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
1102                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1103                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1104
1105                 cpu_policy->cur = cur_freq;
1106         }
1107
1108 out:
1109         cpufreq_cpu_put(cpu_policy);
1110         return 0;
1111 }
1112
1113 /**
1114  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1115  *
1116  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1117  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1118  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1119  *          restored.
1120  */
1121 static int cpufreq_resume(struct sys_device * sysdev)
1122 {
1123         int cpu = sysdev->id;
1124         int ret = 0;
1125         struct cpufreq_policy *cpu_policy;
1126
1127         dprintk("resuming cpu %u\n", cpu);
1128
1129         if (!cpu_online(cpu))
1130                 return 0;
1131
1132         /* we may be lax here as interrupts are off. Nonetheless
1133          * we need to grab the correct cpu policy, as to check
1134          * whether we really run on this CPU.
1135          */
1136
1137         cpu_policy = cpufreq_cpu_get(cpu);
1138         if (!cpu_policy)
1139                 return -EINVAL;
1140
1141         /* only handle each CPU group once */
1142         if (unlikely(cpu_policy->cpu != cpu)) {
1143                 cpufreq_cpu_put(cpu_policy);
1144                 return 0;
1145         }
1146
1147         if (cpufreq_driver->resume) {
1148                 ret = cpufreq_driver->resume(cpu_policy);
1149                 if (ret) {
1150                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1151                                         "step on CPU %u\n", cpu_policy->cpu);
1152                         cpufreq_cpu_put(cpu_policy);
1153                         return ret;
1154                 }
1155         }
1156
1157         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1158                 unsigned int cur_freq = 0;
1159
1160                 if (cpufreq_driver->get)
1161                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1162
1163                 if (!cur_freq || !cpu_policy->cur) {
1164                         printk(KERN_ERR "cpufreq: resume failed to assert "
1165                                         "current frequency is what timing core "
1166                                         "thinks it is.\n");
1167                         goto out;
1168                 }
1169
1170                 if (unlikely(cur_freq != cpu_policy->cur)) {
1171                         struct cpufreq_freqs freqs;
1172
1173                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1174                                 dprintk("Warning: CPU frequency"
1175                                        "is %u, cpufreq assumed %u kHz.\n",
1176                                        cur_freq, cpu_policy->cur);
1177
1178                         freqs.cpu = cpu;
1179                         freqs.old = cpu_policy->cur;
1180                         freqs.new = cur_freq;
1181
1182                         srcu_notifier_call_chain(
1183                                         &cpufreq_transition_notifier_list,
1184                                         CPUFREQ_RESUMECHANGE, &freqs);
1185                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1186
1187                         cpu_policy->cur = cur_freq;
1188                 }
1189         }
1190
1191 out:
1192         schedule_work(&cpu_policy->update);
1193         cpufreq_cpu_put(cpu_policy);
1194         return ret;
1195 }
1196
1197 static struct sysdev_driver cpufreq_sysdev_driver = {
1198         .add            = cpufreq_add_dev,
1199         .remove         = cpufreq_remove_dev,
1200         .suspend        = cpufreq_suspend,
1201         .resume         = cpufreq_resume,
1202 };
1203
1204
1205 /*********************************************************************
1206  *                     NOTIFIER LISTS INTERFACE                      *
1207  *********************************************************************/
1208
1209 /**
1210  *      cpufreq_register_notifier - register a driver with cpufreq
1211  *      @nb: notifier function to register
1212  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1213  *
1214  *      Add a driver to one of two lists: either a list of drivers that
1215  *      are notified about clock rate changes (once before and once after
1216  *      the transition), or a list of drivers that are notified about
1217  *      changes in cpufreq policy.
1218  *
1219  *      This function may sleep, and has the same return conditions as
1220  *      blocking_notifier_chain_register.
1221  */
1222 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1223 {
1224         int ret;
1225
1226         switch (list) {
1227         case CPUFREQ_TRANSITION_NOTIFIER:
1228                 ret = srcu_notifier_chain_register(
1229                                 &cpufreq_transition_notifier_list, nb);
1230                 break;
1231         case CPUFREQ_POLICY_NOTIFIER:
1232                 ret = blocking_notifier_chain_register(
1233                                 &cpufreq_policy_notifier_list, nb);
1234                 break;
1235         default:
1236                 ret = -EINVAL;
1237         }
1238
1239         return ret;
1240 }
1241 EXPORT_SYMBOL(cpufreq_register_notifier);
1242
1243
1244 /**
1245  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1246  *      @nb: notifier block to be unregistered
1247  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1248  *
1249  *      Remove a driver from the CPU frequency notifier list.
1250  *
1251  *      This function may sleep, and has the same return conditions as
1252  *      blocking_notifier_chain_unregister.
1253  */
1254 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1255 {
1256         int ret;
1257
1258         switch (list) {
1259         case CPUFREQ_TRANSITION_NOTIFIER:
1260                 ret = srcu_notifier_chain_unregister(
1261                                 &cpufreq_transition_notifier_list, nb);
1262                 break;
1263         case CPUFREQ_POLICY_NOTIFIER:
1264                 ret = blocking_notifier_chain_unregister(
1265                                 &cpufreq_policy_notifier_list, nb);
1266                 break;
1267         default:
1268                 ret = -EINVAL;
1269         }
1270
1271         return ret;
1272 }
1273 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1274
1275
1276 /*********************************************************************
1277  *                              GOVERNORS                            *
1278  *********************************************************************/
1279
1280
1281 /* Must be called with lock_cpu_hotplug held */
1282 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1283                             unsigned int target_freq,
1284                             unsigned int relation)
1285 {
1286         int retval = -EINVAL;
1287
1288         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1289                 target_freq, relation);
1290         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1291                 retval = cpufreq_driver->target(policy, target_freq, relation);
1292
1293         return retval;
1294 }
1295 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1296
1297 int cpufreq_driver_target(struct cpufreq_policy *policy,
1298                           unsigned int target_freq,
1299                           unsigned int relation)
1300 {
1301         int ret;
1302
1303         policy = cpufreq_cpu_get(policy->cpu);
1304         if (!policy)
1305                 return -EINVAL;
1306
1307         lock_cpu_hotplug();
1308         mutex_lock(&policy->lock);
1309
1310         ret = __cpufreq_driver_target(policy, target_freq, relation);
1311
1312         mutex_unlock(&policy->lock);
1313         unlock_cpu_hotplug();
1314
1315         cpufreq_cpu_put(policy);
1316         return ret;
1317 }
1318 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1319
1320 int cpufreq_driver_getavg(struct cpufreq_policy *policy)
1321 {
1322         int ret = 0;
1323
1324         policy = cpufreq_cpu_get(policy->cpu);
1325         if (!policy)
1326                 return -EINVAL;
1327
1328         mutex_lock(&policy->lock);
1329
1330         if (cpu_online(policy->cpu) && cpufreq_driver->getavg)
1331                 ret = cpufreq_driver->getavg(policy->cpu);
1332
1333         mutex_unlock(&policy->lock);
1334
1335         cpufreq_cpu_put(policy);
1336         return ret;
1337 }
1338 EXPORT_SYMBOL_GPL(cpufreq_driver_getavg);
1339
1340 /*
1341  * Locking: Must be called with the lock_cpu_hotplug() lock held
1342  * when "event" is CPUFREQ_GOV_LIMITS
1343  */
1344
1345 static int __cpufreq_governor(struct cpufreq_policy *policy,
1346                                         unsigned int event)
1347 {
1348         int ret;
1349
1350         if (!try_module_get(policy->governor->owner))
1351                 return -EINVAL;
1352
1353         dprintk("__cpufreq_governor for CPU %u, event %u\n",
1354                                                 policy->cpu, event);
1355         ret = policy->governor->governor(policy, event);
1356
1357         /* we keep one module reference alive for
1358                         each CPU governed by this CPU */
1359         if ((event != CPUFREQ_GOV_START) || ret)
1360                 module_put(policy->governor->owner);
1361         if ((event == CPUFREQ_GOV_STOP) && !ret)
1362                 module_put(policy->governor->owner);
1363
1364         return ret;
1365 }
1366
1367
1368 int cpufreq_register_governor(struct cpufreq_governor *governor)
1369 {
1370         int err;
1371
1372         if (!governor)
1373                 return -EINVAL;
1374
1375         mutex_lock(&cpufreq_governor_mutex);
1376
1377         err = -EBUSY;
1378         if (__find_governor(governor->name) == NULL) {
1379                 err = 0;
1380                 list_add(&governor->governor_list, &cpufreq_governor_list);
1381         }
1382
1383         mutex_unlock(&cpufreq_governor_mutex);
1384         return err;
1385 }
1386 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1387
1388
1389 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1390 {
1391         if (!governor)
1392                 return;
1393
1394         mutex_lock(&cpufreq_governor_mutex);
1395         list_del(&governor->governor_list);
1396         mutex_unlock(&cpufreq_governor_mutex);
1397         return;
1398 }
1399 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1400
1401
1402
1403 /*********************************************************************
1404  *                          POLICY INTERFACE                         *
1405  *********************************************************************/
1406
1407 /**
1408  * cpufreq_get_policy - get the current cpufreq_policy
1409  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1410  *
1411  * Reads the current cpufreq policy.
1412  */
1413 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1414 {
1415         struct cpufreq_policy *cpu_policy;
1416         if (!policy)
1417                 return -EINVAL;
1418
1419         cpu_policy = cpufreq_cpu_get(cpu);
1420         if (!cpu_policy)
1421                 return -EINVAL;
1422
1423         mutex_lock(&cpu_policy->lock);
1424         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1425         mutex_unlock(&cpu_policy->lock);
1426
1427         cpufreq_cpu_put(cpu_policy);
1428         return 0;
1429 }
1430 EXPORT_SYMBOL(cpufreq_get_policy);
1431
1432
1433 /*
1434  * data   : current policy.
1435  * policy : policy to be set.
1436  * Locking: Must be called with the lock_cpu_hotplug() lock held
1437  */
1438 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1439                                 struct cpufreq_policy *policy)
1440 {
1441         int ret = 0;
1442
1443         cpufreq_debug_disable_ratelimit();
1444         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1445                 policy->min, policy->max);
1446
1447         memcpy(&policy->cpuinfo, &data->cpuinfo,
1448                                 sizeof(struct cpufreq_cpuinfo));
1449
1450         if (policy->min > data->min && policy->min > policy->max) {
1451                 ret = -EINVAL;
1452                 goto error_out;
1453         }
1454
1455         /* verify the cpu speed can be set within this limit */
1456         ret = cpufreq_driver->verify(policy);
1457         if (ret)
1458                 goto error_out;
1459
1460         /* adjust if necessary - all reasons */
1461         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1462                         CPUFREQ_ADJUST, policy);
1463
1464         /* adjust if necessary - hardware incompatibility*/
1465         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1466                         CPUFREQ_INCOMPATIBLE, policy);
1467
1468         /* verify the cpu speed can be set within this limit,
1469            which might be different to the first one */
1470         ret = cpufreq_driver->verify(policy);
1471         if (ret)
1472                 goto error_out;
1473
1474         /* notification of the new policy */
1475         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1476                         CPUFREQ_NOTIFY, policy);
1477
1478         data->min = policy->min;
1479         data->max = policy->max;
1480
1481         dprintk("new min and max freqs are %u - %u kHz\n",
1482                                         data->min, data->max);
1483
1484         if (cpufreq_driver->setpolicy) {
1485                 data->policy = policy->policy;
1486                 dprintk("setting range\n");
1487                 ret = cpufreq_driver->setpolicy(policy);
1488         } else {
1489                 if (policy->governor != data->governor) {
1490                         /* save old, working values */
1491                         struct cpufreq_governor *old_gov = data->governor;
1492
1493                         dprintk("governor switch\n");
1494
1495                         /* end old governor */
1496                         if (data->governor)
1497                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1498
1499                         /* start new governor */
1500                         data->governor = policy->governor;
1501                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1502                                 /* new governor failed, so re-start old one */
1503                                 dprintk("starting governor %s failed\n",
1504                                                         data->governor->name);
1505                                 if (old_gov) {
1506                                         data->governor = old_gov;
1507                                         __cpufreq_governor(data,
1508                                                            CPUFREQ_GOV_START);
1509                                 }
1510                                 ret = -EINVAL;
1511                                 goto error_out;
1512                         }
1513                         /* might be a policy change, too, so fall through */
1514                 }
1515                 dprintk("governor: change or update limits\n");
1516                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1517         }
1518
1519 error_out:
1520         cpufreq_debug_enable_ratelimit();
1521         return ret;
1522 }
1523
1524 /**
1525  *      cpufreq_set_policy - set a new CPUFreq policy
1526  *      @policy: policy to be set.
1527  *
1528  *      Sets a new CPU frequency and voltage scaling policy.
1529  */
1530 int cpufreq_set_policy(struct cpufreq_policy *policy)
1531 {
1532         int ret = 0;
1533         struct cpufreq_policy *data;
1534
1535         if (!policy)
1536                 return -EINVAL;
1537
1538         data = cpufreq_cpu_get(policy->cpu);
1539         if (!data)
1540                 return -EINVAL;
1541
1542         lock_cpu_hotplug();
1543
1544         /* lock this CPU */
1545         mutex_lock(&data->lock);
1546
1547         ret = __cpufreq_set_policy(data, policy);
1548         data->user_policy.min = data->min;
1549         data->user_policy.max = data->max;
1550         data->user_policy.policy = data->policy;
1551         data->user_policy.governor = data->governor;
1552
1553         mutex_unlock(&data->lock);
1554
1555         unlock_cpu_hotplug();
1556         cpufreq_cpu_put(data);
1557
1558         return ret;
1559 }
1560 EXPORT_SYMBOL(cpufreq_set_policy);
1561
1562
1563 /**
1564  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1565  *      @cpu: CPU which shall be re-evaluated
1566  *
1567  *      Usefull for policy notifiers which have different necessities
1568  *      at different times.
1569  */
1570 int cpufreq_update_policy(unsigned int cpu)
1571 {
1572         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1573         struct cpufreq_policy policy;
1574         int ret = 0;
1575
1576         if (!data)
1577                 return -ENODEV;
1578
1579         lock_cpu_hotplug();
1580         mutex_lock(&data->lock);
1581
1582         dprintk("updating policy for CPU %u\n", cpu);
1583         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1584         policy.min = data->user_policy.min;
1585         policy.max = data->user_policy.max;
1586         policy.policy = data->user_policy.policy;
1587         policy.governor = data->user_policy.governor;
1588
1589         /* BIOS might change freq behind our back
1590           -> ask driver for current freq and notify governors about a change */
1591         if (cpufreq_driver->get) {
1592                 policy.cur = cpufreq_driver->get(cpu);
1593                 if (!data->cur) {
1594                         dprintk("Driver did not initialize current freq");
1595                         data->cur = policy.cur;
1596                 } else {
1597                         if (data->cur != policy.cur)
1598                                 cpufreq_out_of_sync(cpu, data->cur,
1599                                                                 policy.cur);
1600                 }
1601         }
1602
1603         ret = __cpufreq_set_policy(data, &policy);
1604
1605         mutex_unlock(&data->lock);
1606         unlock_cpu_hotplug();
1607         cpufreq_cpu_put(data);
1608         return ret;
1609 }
1610 EXPORT_SYMBOL(cpufreq_update_policy);
1611
1612 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1613                                         unsigned long action, void *hcpu)
1614 {
1615         unsigned int cpu = (unsigned long)hcpu;
1616         struct cpufreq_policy *policy;
1617         struct sys_device *sys_dev;
1618
1619         sys_dev = get_cpu_sysdev(cpu);
1620
1621         if (sys_dev) {
1622                 switch (action) {
1623                 case CPU_ONLINE:
1624                         cpufreq_add_dev(sys_dev);
1625                         break;
1626                 case CPU_DOWN_PREPARE:
1627                         /*
1628                          * We attempt to put this cpu in lowest frequency
1629                          * possible before going down. This will permit
1630                          * hardware-managed P-State to switch other related
1631                          * threads to min or higher speeds if possible.
1632                          */
1633                         policy = cpufreq_cpu_data[cpu];
1634                         if (policy) {
1635                                 cpufreq_driver_target(policy, policy->min,
1636                                                 CPUFREQ_RELATION_H);
1637                         }
1638                         break;
1639                 case CPU_DEAD:
1640                         cpufreq_remove_dev(sys_dev);
1641                         break;
1642                 }
1643         }
1644         return NOTIFY_OK;
1645 }
1646
1647 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1648 {
1649     .notifier_call = cpufreq_cpu_callback,
1650 };
1651
1652 /*********************************************************************
1653  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1654  *********************************************************************/
1655
1656 /**
1657  * cpufreq_register_driver - register a CPU Frequency driver
1658  * @driver_data: A struct cpufreq_driver containing the values#
1659  * submitted by the CPU Frequency driver.
1660  *
1661  *   Registers a CPU Frequency driver to this core code. This code
1662  * returns zero on success, -EBUSY when another driver got here first
1663  * (and isn't unregistered in the meantime).
1664  *
1665  */
1666 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1667 {
1668         unsigned long flags;
1669         int ret;
1670
1671         if (!driver_data || !driver_data->verify || !driver_data->init ||
1672             ((!driver_data->setpolicy) && (!driver_data->target)))
1673                 return -EINVAL;
1674
1675         dprintk("trying to register driver %s\n", driver_data->name);
1676
1677         if (driver_data->setpolicy)
1678                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1679
1680         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1681         if (cpufreq_driver) {
1682                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1683                 return -EBUSY;
1684         }
1685         cpufreq_driver = driver_data;
1686         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1687
1688         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1689
1690         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1691                 int i;
1692                 ret = -ENODEV;
1693
1694                 /* check for at least one working CPU */
1695                 for (i=0; i<NR_CPUS; i++)
1696                         if (cpufreq_cpu_data[i])
1697                                 ret = 0;
1698
1699                 /* if all ->init() calls failed, unregister */
1700                 if (ret) {
1701                         dprintk("no CPU initialized for driver %s\n",
1702                                                         driver_data->name);
1703                         sysdev_driver_unregister(&cpu_sysdev_class,
1704                                                 &cpufreq_sysdev_driver);
1705
1706                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1707                         cpufreq_driver = NULL;
1708                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1709                 }
1710         }
1711
1712         if (!ret) {
1713                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1714                 dprintk("driver %s up and running\n", driver_data->name);
1715                 cpufreq_debug_enable_ratelimit();
1716         }
1717
1718         return (ret);
1719 }
1720 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1721
1722
1723 /**
1724  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1725  *
1726  *    Unregister the current CPUFreq driver. Only call this if you have
1727  * the right to do so, i.e. if you have succeeded in initialising before!
1728  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1729  * currently not initialised.
1730  */
1731 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1732 {
1733         unsigned long flags;
1734
1735         cpufreq_debug_disable_ratelimit();
1736
1737         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1738                 cpufreq_debug_enable_ratelimit();
1739                 return -EINVAL;
1740         }
1741
1742         dprintk("unregistering driver %s\n", driver->name);
1743
1744         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1745         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1746
1747         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1748         cpufreq_driver = NULL;
1749         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1750
1751         return 0;
1752 }
1753 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);