drivers/cpufreq/cpufreq.c

   1 /*
   2  *  linux/drivers/cpufreq/cpufreq.c
   3  *
   4  *  Copyright (C) 2001 Russell King
   5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
   6  *
   7  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
   8  *      Added handling for CPU hotplug
   9  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
  10  *      Fix handling for CPU hotplug -- affected CPUs
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17
  18 #include <linux/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/notifier.h>
  22 #include <linux/cpufreq.h>
  23 #include <linux/delay.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/device.h>
  27 #include <linux/slab.h>
  28 #include <linux/cpu.h>
  29 #include <linux/completion.h>
  30 #include <linux/mutex.h>
  31
  32 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
  33
  34 /**
  35  * The "cpufreq driver" - the arch- or hardware-dependent low
  36  * level driver of CPUFreq support, and its spinlock. This lock
  37  * also protects the cpufreq_cpu_data array.
  38  */
  39 static struct cpufreq_driver *cpufreq_driver;
  40 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
  41 static DEFINE_SPINLOCK(cpufreq_driver_lock);
  42
  43 /* internal prototypes */
  44 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
  45 static void handle_update(void *data);
  46
  47 /**
  48  * Two notifier lists: the "policy" list is involved in the
  49  * validation process for a new CPU frequency policy; the
  50  * "transition" list for kernel code that needs to handle
  51  * changes to devices when the CPU clock speed changes.
  52  * The mutex locks both lists.
  53  */
  54 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
  55 static struct srcu_notifier_head cpufreq_transition_notifier_list;
  56
  57 static int __init init_cpufreq_transition_notifier_list(void)
  58 {
  59         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
  60         return 0;
  61 }
  62 core_initcall(init_cpufreq_transition_notifier_list);
  63
  64 static LIST_HEAD(cpufreq_governor_list);
  65 static DEFINE_MUTEX (cpufreq_governor_mutex);
  66
  67 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
  68 {
  69         struct cpufreq_policy *data;
  70         unsigned long flags;
  71
  72         if (cpu >= NR_CPUS)
  73                 goto err_out;
  74
  75         /* get the cpufreq driver */
  76         spin_lock_irqsave(&cpufreq_driver_lock, flags);
  77
  78         if (!cpufreq_driver)
  79                 goto err_out_unlock;
  80
  81         if (!try_module_get(cpufreq_driver->owner))
  82                 goto err_out_unlock;
  83
  84
  85         /* get the CPU */
  86         data = cpufreq_cpu_data[cpu];
  87
  88         if (!data)
  89                 goto err_out_put_module;
  90
  91         if (!kobject_get(&data->kobj))
  92                 goto err_out_put_module;
  93
  94         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  95         return data;
  96
  97 err_out_put_module:
  98         module_put(cpufreq_driver->owner);
  99 err_out_unlock:
 100         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 101 err_out:
 102         return NULL;
 103 }
 104 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 105
 106
 107 void cpufreq_cpu_put(struct cpufreq_policy *data)
 108 {
 109         kobject_put(&data->kobj);
 110         module_put(cpufreq_driver->owner);
 111 }
 112 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 113
 114
 115 /*********************************************************************
 116  *                     UNIFIED DEBUG HELPERS                         *
 117  *********************************************************************/
 118 #ifdef CONFIG_CPU_FREQ_DEBUG
 119
 120 /* what part(s) of the CPUfreq subsystem are debugged? */
 121 static unsigned int debug;
 122
 123 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 124  * set or modify this value.
 125  */
 126 static unsigned int debug_ratelimit = 1;
 127
 128 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 129  * loading of a cpufreq driver, temporarily disabled when a new policy
 130  * is set, and disabled upon cpufreq driver removal
 131  */
 132 static unsigned int disable_ratelimit = 1;
 133 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 134
 135 static void cpufreq_debug_enable_ratelimit(void)
 136 {
 137         unsigned long flags;
 138
 139         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 140         if (disable_ratelimit)
 141                 disable_ratelimit--;
 142         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 143 }
 144
 145 static void cpufreq_debug_disable_ratelimit(void)
 146 {
 147         unsigned long flags;
 148
 149         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 150         disable_ratelimit++;
 151         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 152 }
 153
 154 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
 155 {
 156         char s[256];
 157         va_list args;
 158         unsigned int len;
 159         unsigned long flags;
 160
 161         WARN_ON(!prefix);
 162         if (type & debug) {
 163                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 164                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
 165                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 166                         return;
 167                 }
 168                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 169
 170                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 171
 172                 va_start(args, fmt);
 173                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 174                 va_end(args);
 175
 176                 printk(s);
 177
 178                 WARN_ON(len < 5);
 179         }
 180 }
 181 EXPORT_SYMBOL(cpufreq_debug_printk);
 182
 183
 184 module_param(debug, uint, 0644);
 185 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
 186
 187 module_param(debug_ratelimit, uint, 0644);
 188 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
 189
 190 #else /* !CONFIG_CPU_FREQ_DEBUG */
 191
 192 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 193 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 194
 195 #endif /* CONFIG_CPU_FREQ_DEBUG */
 196
 197
 198 /*********************************************************************
 199  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 200  *********************************************************************/
 201
 202 /**
 203  * adjust_jiffies - adjust the system "loops_per_jiffy"
 204  *
 205  * This function alters the system "loops_per_jiffy" for the clock
 206  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 207  * systems as each CPU might be scaled differently. So, use the arch
 208  * per-CPU loops_per_jiffy value wherever possible.
 209  */
 210 #ifndef CONFIG_SMP
 211 static unsigned long l_p_j_ref;
 212 static unsigned int  l_p_j_ref_freq;
 213
 214 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 215 {
 216         if (ci->flags & CPUFREQ_CONST_LOOPS)
 217                 return;
 218
 219         if (!l_p_j_ref_freq) {
 220                 l_p_j_ref = loops_per_jiffy;
 221                 l_p_j_ref_freq = ci->old;
 222                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 223         }
 224         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 225             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 226             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 227                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
 228                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
 229         }
 230 }
 231 #else
 232 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
 233 #endif
 234
 235
 236 /**
 237  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 238  * on frequency transition.
 239  *
 240  * This function calls the transition notifiers and the "adjust_jiffies"
 241  * function. It is called twice on all CPU frequency changes that have
 242  * external effects.
 243  */
 244 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 245 {
 246         struct cpufreq_policy *policy;
 247
 248         BUG_ON(irqs_disabled());
 249
 250         freqs->flags = cpufreq_driver->flags;
 251         dprintk("notification %u of frequency transition to %u kHz\n",
 252                 state, freqs->new);
 253
 254         policy = cpufreq_cpu_data[freqs->cpu];
 255         switch (state) {
 256
 257         case CPUFREQ_PRECHANGE:
 258                 /* detect if the driver reported a value as "old frequency"
 259                  * which is not equal to what the cpufreq core thinks is
 260                  * "old frequency".
 261                  */
 262                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 263                         if ((policy) && (policy->cpu == freqs->cpu) &&
 264                             (policy->cur) && (policy->cur != freqs->old)) {
 265                                 dprintk("Warning: CPU frequency is"
 266                                         " %u, cpufreq assumed %u kHz.\n",
 267                                         freqs->old, policy->cur);
 268                                 freqs->old = policy->cur;
 269                         }
 270                 }
 271                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 272                                 CPUFREQ_PRECHANGE, freqs);
 273                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 274                 break;
 275
 276         case CPUFREQ_POSTCHANGE:
 277                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 278                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 279                                 CPUFREQ_POSTCHANGE, freqs);
 280                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 281                         policy->cur = freqs->new;
 282                 break;
 283         }
 284 }
 285 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 286
 287
 288
 289 /*********************************************************************
 290  *                          SYSFS INTERFACE                          *
 291  *********************************************************************/
 292
 293 static struct cpufreq_governor *__find_governor(const char *str_governor)
 294 {
 295         struct cpufreq_governor *t;
 296
 297         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
 298                 if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN))
 299                         return t;
 300
 301         return NULL;
 302 }
 303
 304 /**
 305  * cpufreq_parse_governor - parse a governor string
 306  */
 307 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 308                                 struct cpufreq_governor **governor)
 309 {
 310         int err = -EINVAL;
 311
 312         if (!cpufreq_driver)
 313                 goto out;
 314
 315         if (cpufreq_driver->setpolicy) {
 316                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 317                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 318                         err = 0;
 319                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 320                         *policy = CPUFREQ_POLICY_POWERSAVE;
 321                         err = 0;
 322                 }
 323         } else if (cpufreq_driver->target) {
 324                 struct cpufreq_governor *t;
 325
 326                 mutex_lock(&cpufreq_governor_mutex);
 327
 328                 t = __find_governor(str_governor);
 329
 330                 if (t == NULL) {
 331                         char *name = kasprintf(GFP_KERNEL, "cpufreq_%s", str_governor);
 332
 333                         if (name) {
 334                                 int ret;
 335
 336                                 mutex_unlock(&cpufreq_governor_mutex);
 337                                 ret = request_module(name);
 338                                 mutex_lock(&cpufreq_governor_mutex);
 339
 340                                 if (ret == 0)
 341                                         t = __find_governor(str_governor);
 342                         }
 343
 344                         kfree(name);
 345                 }
 346
 347                 if (t != NULL) {
 348                         *governor = t;
 349                         err = 0;
 350                 }
 351
 352                 mutex_unlock(&cpufreq_governor_mutex);
 353         }
 354   out:
 355         return err;
 356 }
 357
 358
 359 /* drivers/base/cpu.c */
 360 extern struct sysdev_class cpu_sysdev_class;
 361
 362
 363 /**
 364  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 365  *
 366  * Write out information from cpufreq_driver->policy[cpu]; object must be
 367  * "unsigned int".
 368  */
 369
 370 #define show_one(file_name, object)                     \
 371 static ssize_t show_##file_name                         \
 372 (struct cpufreq_policy * policy, char *buf)             \
 373 {                                                       \
 374         return sprintf (buf, "%u\n", policy->object);   \
 375 }
 376
 377 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 378 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 379 show_one(scaling_min_freq, min);
 380 show_one(scaling_max_freq, max);
 381 show_one(scaling_cur_freq, cur);
 382
 383 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
 384
 385 /**
 386  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 387  */
 388 #define store_one(file_name, object)                    \
 389 static ssize_t store_##file_name                                        \
 390 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 391 {                                                                       \
 392         unsigned int ret = -EINVAL;                                     \
 393         struct cpufreq_policy new_policy;                               \
 394                                                                         \
 395         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 396         if (ret)                                                        \
 397                 return -EINVAL;                                         \
 398                                                                         \
 399         ret = sscanf (buf, "%u", &new_policy.object);                   \
 400         if (ret != 1)                                                   \
 401                 return -EINVAL;                                         \
 402                                                                         \
 403         lock_cpu_hotplug();                                             \
 404         mutex_lock(&policy->lock);                                      \
 405         ret = __cpufreq_set_policy(policy, &new_policy);                \
 406         policy->user_policy.object = policy->object;                    \
 407         mutex_unlock(&policy->lock);                                    \
 408         unlock_cpu_hotplug();                                           \
 409                                                                         \
 410         return ret ? ret : count;                                       \
 411 }
 412
 413 store_one(scaling_min_freq,min);
 414 store_one(scaling_max_freq,max);
 415
 416 /**
 417  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 418  */
 419 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 420 {
 421         unsigned int cur_freq = cpufreq_get(policy->cpu);
 422         if (!cur_freq)
 423                 return sprintf(buf, "<unknown>");
 424         return sprintf(buf, "%u\n", cur_freq);
 425 }
 426
 427
 428 /**
 429  * show_scaling_governor - show the current policy for the specified CPU
 430  */
 431 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 432 {
 433         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 434                 return sprintf(buf, "powersave\n");
 435         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 436                 return sprintf(buf, "performance\n");
 437         else if (policy->governor)
 438                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 439         return -EINVAL;
 440 }
 441
 442
 443 /**
 444  * store_scaling_governor - store policy for the specified CPU
 445  */
 446 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 447                                        const char *buf, size_t count)
 448 {
 449         unsigned int ret = -EINVAL;
 450         char    str_governor[16];
 451         struct cpufreq_policy new_policy;
 452
 453         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 454         if (ret)
 455                 return ret;
 456
 457         ret = sscanf (buf, "%15s", str_governor);
 458         if (ret != 1)
 459                 return -EINVAL;
 460
 461         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 462                 return -EINVAL;
 463
 464         lock_cpu_hotplug();
 465
 466         /* Do not use cpufreq_set_policy here or the user_policy.max
 467            will be wrongly overridden */
 468         mutex_lock(&policy->lock);
 469         ret = __cpufreq_set_policy(policy, &new_policy);
 470
 471         policy->user_policy.policy = policy->policy;
 472         policy->user_policy.governor = policy->governor;
 473         mutex_unlock(&policy->lock);
 474
 475         unlock_cpu_hotplug();
 476
 477         return ret ? ret : count;
 478 }
 479
 480 /**
 481  * show_scaling_driver - show the cpufreq driver currently loaded
 482  */
 483 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 484 {
 485         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 486 }
 487
 488 /**
 489  * show_scaling_available_governors - show the available CPUfreq governors
 490  */
 491 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 492                                 char *buf)
 493 {
 494         ssize_t i = 0;
 495         struct cpufreq_governor *t;
 496
 497         if (!cpufreq_driver->target) {
 498                 i += sprintf(buf, "performance powersave");
 499                 goto out;
 500         }
 501
 502         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 503                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 504                         goto out;
 505                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 506         }
 507 out:
 508         i += sprintf(&buf[i], "\n");
 509         return i;
 510 }
 511 /**
 512  * show_affected_cpus - show the CPUs affected by each transition
 513  */
 514 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 515 {
 516         ssize_t i = 0;
 517         unsigned int cpu;
 518
 519         for_each_cpu_mask(cpu, policy->cpus) {
 520                 if (i)
 521                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 522                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 523                 if (i >= (PAGE_SIZE - 5))
 524                     break;
 525         }
 526         i += sprintf(&buf[i], "\n");
 527         return i;
 528 }
 529
 530
 531 #define define_one_ro(_name) \
 532 static struct freq_attr _name = \
 533 __ATTR(_name, 0444, show_##_name, NULL)
 534
 535 #define define_one_ro0400(_name) \
 536 static struct freq_attr _name = \
 537 __ATTR(_name, 0400, show_##_name, NULL)
 538
 539 #define define_one_rw(_name) \
 540 static struct freq_attr _name = \
 541 __ATTR(_name, 0644, show_##_name, store_##_name)
 542
 543 define_one_ro0400(cpuinfo_cur_freq);
 544 define_one_ro(cpuinfo_min_freq);
 545 define_one_ro(cpuinfo_max_freq);
 546 define_one_ro(scaling_available_governors);
 547 define_one_ro(scaling_driver);
 548 define_one_ro(scaling_cur_freq);
 549 define_one_ro(affected_cpus);
 550 define_one_rw(scaling_min_freq);
 551 define_one_rw(scaling_max_freq);
 552 define_one_rw(scaling_governor);
 553
 554 static struct attribute * default_attrs[] = {
 555         &cpuinfo_min_freq.attr,
 556         &cpuinfo_max_freq.attr,
 557         &scaling_min_freq.attr,
 558         &scaling_max_freq.attr,
 559         &affected_cpus.attr,
 560         &scaling_governor.attr,
 561         &scaling_driver.attr,
 562         &scaling_available_governors.attr,
 563         NULL
 564 };
 565
 566 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 567 #define to_attr(a) container_of(a,struct freq_attr,attr)
 568
 569 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 570 {
 571         struct cpufreq_policy * policy = to_policy(kobj);
 572         struct freq_attr * fattr = to_attr(attr);
 573         ssize_t ret;
 574         policy = cpufreq_cpu_get(policy->cpu);
 575         if (!policy)
 576                 return -EINVAL;
 577         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 578         cpufreq_cpu_put(policy);
 579         return ret;
 580 }
 581
 582 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 583                      const char * buf, size_t count)
 584 {
 585         struct cpufreq_policy * policy = to_policy(kobj);
 586         struct freq_attr * fattr = to_attr(attr);
 587         ssize_t ret;
 588         policy = cpufreq_cpu_get(policy->cpu);
 589         if (!policy)
 590                 return -EINVAL;
 591         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 592         cpufreq_cpu_put(policy);
 593         return ret;
 594 }
 595
 596 static void cpufreq_sysfs_release(struct kobject * kobj)
 597 {
 598         struct cpufreq_policy * policy = to_policy(kobj);
 599         dprintk("last reference is dropped\n");
 600         complete(&policy->kobj_unregister);
 601 }
 602
 603 static struct sysfs_ops sysfs_ops = {
 604         .show   = show,
 605         .store  = store,
 606 };
 607
 608 static struct kobj_type ktype_cpufreq = {
 609         .sysfs_ops      = &sysfs_ops,
 610         .default_attrs  = default_attrs,
 611         .release        = cpufreq_sysfs_release,
 612 };
 613
 614
 615 /**
 616  * cpufreq_add_dev - add a CPU device
 617  *
 618  * Adds the cpufreq interface for a CPU device.
 619  */
 620 static int cpufreq_add_dev (struct sys_device * sys_dev)
 621 {
 622         unsigned int cpu = sys_dev->id;
 623         int ret = 0;
 624         struct cpufreq_policy new_policy;
 625         struct cpufreq_policy *policy;
 626         struct freq_attr **drv_attr;
 627         struct sys_device *cpu_sys_dev;
 628         unsigned long flags;
 629         unsigned int j;
 630 #ifdef CONFIG_SMP
 631         struct cpufreq_policy *managed_policy;
 632 #endif
 633
 634         if (cpu_is_offline(cpu))
 635                 return 0;
 636
 637         cpufreq_debug_disable_ratelimit();
 638         dprintk("adding CPU %u\n", cpu);
 639
 640 #ifdef CONFIG_SMP
 641         /* check whether a different CPU already registered this
 642          * CPU because it is in the same boat. */
 643         policy = cpufreq_cpu_get(cpu);
 644         if (unlikely(policy)) {
 645                 cpufreq_cpu_put(policy);
 646                 cpufreq_debug_enable_ratelimit();
 647                 return 0;
 648         }
 649 #endif
 650
 651         if (!try_module_get(cpufreq_driver->owner)) {
 652                 ret = -EINVAL;
 653                 goto module_out;
 654         }
 655
 656         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 657         if (!policy) {
 658                 ret = -ENOMEM;
 659                 goto nomem_out;
 660         }
 661
 662         policy->cpu = cpu;
 663         policy->cpus = cpumask_of_cpu(cpu);
 664
 665         mutex_init(&policy->lock);
 666         mutex_lock(&policy->lock);
 667         init_completion(&policy->kobj_unregister);
 668         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 669
 670         /* call driver. From then on the cpufreq must be able
 671          * to accept all calls to ->verify and ->setpolicy for this CPU
 672          */
 673         ret = cpufreq_driver->init(policy);
 674         if (ret) {
 675                 dprintk("initialization failed\n");
 676                 mutex_unlock(&policy->lock);
 677                 goto err_out;
 678         }
 679
 680 #ifdef CONFIG_SMP
 681         for_each_cpu_mask(j, policy->cpus) {
 682                 if (cpu == j)
 683                         continue;
 684
 685                 /* check for existing affected CPUs.  They may not be aware
 686                  * of it due to CPU Hotplug.
 687                  */
 688                 managed_policy = cpufreq_cpu_get(j);
 689                 if (unlikely(managed_policy)) {
 690                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 691                         managed_policy->cpus = policy->cpus;
 692                         cpufreq_cpu_data[cpu] = managed_policy;
 693                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 694
 695                         dprintk("CPU already managed, adding link\n");
 696                         sysfs_create_link(&sys_dev->kobj,
 697                                           &managed_policy->kobj, "cpufreq");
 698
 699                         cpufreq_debug_enable_ratelimit();
 700                         mutex_unlock(&policy->lock);
 701                         ret = 0;
 702                         goto err_out_driver_exit; /* call driver->exit() */
 703                 }
 704         }
 705 #endif
 706         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 707
 708         /* prepare interface data */
 709         policy->kobj.parent = &sys_dev->kobj;
 710         policy->kobj.ktype = &ktype_cpufreq;
 711         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 712
 713         ret = kobject_register(&policy->kobj);
 714         if (ret) {
 715                 mutex_unlock(&policy->lock);
 716                 goto err_out_driver_exit;
 717         }
 718         /* set up files for this cpu device */
 719         drv_attr = cpufreq_driver->attr;
 720         while ((drv_attr) && (*drv_attr)) {
 721                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 722                 drv_attr++;
 723         }
 724         if (cpufreq_driver->get)
 725                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 726         if (cpufreq_driver->target)
 727                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 728
 729         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 730         for_each_cpu_mask(j, policy->cpus)
 731                 cpufreq_cpu_data[j] = policy;
 732         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 733
 734         /* symlink affected CPUs */
 735         for_each_cpu_mask(j, policy->cpus) {
 736                 if (j == cpu)
 737                         continue;
 738                 if (!cpu_online(j))
 739                         continue;
 740
 741                 dprintk("CPU %u already managed, adding link\n", j);
 742                 cpufreq_cpu_get(cpu);
 743                 cpu_sys_dev = get_cpu_sysdev(j);
 744                 sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 745                                   "cpufreq");
 746         }
 747
 748         policy->governor = NULL; /* to assure that the starting sequence is
 749                                   * run in cpufreq_set_policy */
 750         mutex_unlock(&policy->lock);
 751
 752         /* set default policy */
 753         ret = cpufreq_set_policy(&new_policy);
 754         if (ret) {
 755                 dprintk("setting policy failed\n");
 756                 goto err_out_unregister;
 757         }
 758
 759         module_put(cpufreq_driver->owner);
 760         dprintk("initialization complete\n");
 761         cpufreq_debug_enable_ratelimit();
 762
 763         return 0;
 764
 765
 766 err_out_unregister:
 767         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 768         for_each_cpu_mask(j, policy->cpus)
 769                 cpufreq_cpu_data[j] = NULL;
 770         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 771
 772         kobject_unregister(&policy->kobj);
 773         wait_for_completion(&policy->kobj_unregister);
 774
 775 err_out_driver_exit:
 776         if (cpufreq_driver->exit)
 777                 cpufreq_driver->exit(policy);
 778
 779 err_out:
 780         kfree(policy);
 781
 782 nomem_out:
 783         module_put(cpufreq_driver->owner);
 784 module_out:
 785         cpufreq_debug_enable_ratelimit();
 786         return ret;
 787 }
 788
 789
 790 /**
 791  * cpufreq_remove_dev - remove a CPU device
 792  *
 793  * Removes the cpufreq interface for a CPU device.
 794  */
 795 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 796 {
 797         unsigned int cpu = sys_dev->id;
 798         unsigned long flags;
 799         struct cpufreq_policy *data;
 800 #ifdef CONFIG_SMP
 801         struct sys_device *cpu_sys_dev;
 802         unsigned int j;
 803 #endif
 804
 805         cpufreq_debug_disable_ratelimit();
 806         dprintk("unregistering CPU %u\n", cpu);
 807
 808         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 809         data = cpufreq_cpu_data[cpu];
 810
 811         if (!data) {
 812                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 813                 cpufreq_debug_enable_ratelimit();
 814                 return -EINVAL;
 815         }
 816         cpufreq_cpu_data[cpu] = NULL;
 817
 818
 819 #ifdef CONFIG_SMP
 820         /* if this isn't the CPU which is the parent of the kobj, we
 821          * only need to unlink, put and exit
 822          */
 823         if (unlikely(cpu != data->cpu)) {
 824                 dprintk("removing link\n");
 825                 cpu_clear(cpu, data->cpus);
 826                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 827                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 828                 cpufreq_cpu_put(data);
 829                 cpufreq_debug_enable_ratelimit();
 830                 return 0;
 831         }
 832 #endif
 833
 834
 835         if (!kobject_get(&data->kobj)) {
 836                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 837                 cpufreq_debug_enable_ratelimit();
 838                 return -EFAULT;
 839         }
 840
 841 #ifdef CONFIG_SMP
 842         /* if we have other CPUs still registered, we need to unlink them,
 843          * or else wait_for_completion below will lock up. Clean the
 844          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 845          * links afterwards.
 846          */
 847         if (unlikely(cpus_weight(data->cpus) > 1)) {
 848                 for_each_cpu_mask(j, data->cpus) {
 849                         if (j == cpu)
 850                                 continue;
 851                         cpufreq_cpu_data[j] = NULL;
 852                 }
 853         }
 854
 855         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 856
 857         if (unlikely(cpus_weight(data->cpus) > 1)) {
 858                 for_each_cpu_mask(j, data->cpus) {
 859                         if (j == cpu)
 860                                 continue;
 861                         dprintk("removing link for cpu %u\n", j);
 862                         cpu_sys_dev = get_cpu_sysdev(j);
 863                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 864                         cpufreq_cpu_put(data);
 865                 }
 866         }
 867 #else
 868         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 869 #endif
 870
 871         mutex_lock(&data->lock);
 872         if (cpufreq_driver->target)
 873                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 874         mutex_unlock(&data->lock);
 875
 876         kobject_unregister(&data->kobj);
 877
 878         kobject_put(&data->kobj);
 879
 880         /* we need to make sure that the underlying kobj is actually
 881          * not referenced anymore by anybody before we proceed with
 882          * unloading.
 883          */
 884         dprintk("waiting for dropping of refcount\n");
 885         wait_for_completion(&data->kobj_unregister);
 886         dprintk("wait complete\n");
 887
 888         if (cpufreq_driver->exit)
 889                 cpufreq_driver->exit(data);
 890
 891         kfree(data);
 892
 893         cpufreq_debug_enable_ratelimit();
 894         return 0;
 895 }
 896
 897
 898 static void handle_update(void *data)
 899 {
 900         unsigned int cpu = (unsigned int)(long)data;
 901         dprintk("handle_update for cpu %u called\n", cpu);
 902         cpufreq_update_policy(cpu);
 903 }
 904
 905 /**
 906  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 907  *      @cpu: cpu number
 908  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 909  *      @new_freq: CPU frequency the CPU actually runs at
 910  *
 911  *      We adjust to current frequency first, and need to clean up later. So either call
 912  *      to cpufreq_update_policy() or schedule handle_update()).
 913  */
 914 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 915 {
 916         struct cpufreq_freqs freqs;
 917
 918         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
 919                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 920
 921         freqs.cpu = cpu;
 922         freqs.old = old_freq;
 923         freqs.new = new_freq;
 924         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 925         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 926 }
 927
 928
 929 /**
 930  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
 931  * @cpu: CPU number
 932  *
 933  * This is the last known freq, without actually getting it from the driver.
 934  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 935  */
 936 unsigned int cpufreq_quick_get(unsigned int cpu)
 937 {
 938         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 939         unsigned int ret = 0;
 940
 941         if (policy) {
 942                 mutex_lock(&policy->lock);
 943                 ret = policy->cur;
 944                 mutex_unlock(&policy->lock);
 945                 cpufreq_cpu_put(policy);
 946         }
 947
 948         return (ret);
 949 }
 950 EXPORT_SYMBOL(cpufreq_quick_get);
 951
 952
 953 /**
 954  * cpufreq_get - get the current CPU frequency (in kHz)
 955  * @cpu: CPU number
 956  *
 957  * Get the CPU current (static) CPU frequency
 958  */
 959 unsigned int cpufreq_get(unsigned int cpu)
 960 {
 961         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 962         unsigned int ret = 0;
 963
 964         if (!policy)
 965                 return 0;
 966
 967         if (!cpufreq_driver->get)
 968                 goto out;
 969
 970         mutex_lock(&policy->lock);
 971
 972         ret = cpufreq_driver->get(cpu);
 973
 974         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 975                 /* verify no discrepancy between actual and saved value exists */
 976                 if (unlikely(ret != policy->cur)) {
 977                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 978                         schedule_work(&policy->update);
 979                 }
 980         }
 981
 982         mutex_unlock(&policy->lock);
 983
 984 out:
 985         cpufreq_cpu_put(policy);
 986
 987         return (ret);
 988 }
 989 EXPORT_SYMBOL(cpufreq_get);
 990
 991
 992 /**
 993  *      cpufreq_suspend - let the low level driver prepare for suspend
 994  */
 995
 996 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 997 {
 998         int cpu = sysdev->id;
 999         unsigned int ret = 0;
1000         unsigned int cur_freq = 0;
1001         struct cpufreq_policy *cpu_policy;
1002
1003         dprintk("suspending cpu %u\n", cpu);
1004
1005         if (!cpu_online(cpu))
1006                 return 0;
1007
1008         /* we may be lax here as interrupts are off. Nonetheless
1009          * we need to grab the correct cpu policy, as to check
1010          * whether we really run on this CPU.
1011          */
1012
1013         cpu_policy = cpufreq_cpu_get(cpu);
1014         if (!cpu_policy)
1015                 return -EINVAL;
1016
1017         /* only handle each CPU group once */
1018         if (unlikely(cpu_policy->cpu != cpu)) {
1019                 cpufreq_cpu_put(cpu_policy);
1020                 return 0;
1021         }
1022
1023         if (cpufreq_driver->suspend) {
1024                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1025                 if (ret) {
1026                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1027                                         "step on CPU %u\n", cpu_policy->cpu);
1028                         cpufreq_cpu_put(cpu_policy);
1029                         return ret;
1030                 }
1031         }
1032
1033
1034         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1035                 goto out;
1036
1037         if (cpufreq_driver->get)
1038                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1039
1040         if (!cur_freq || !cpu_policy->cur) {
1041                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1042                        "frequency is what timing core thinks it is.\n");
1043                 goto out;
1044         }
1045
1046         if (unlikely(cur_freq != cpu_policy->cur)) {
1047                 struct cpufreq_freqs freqs;
1048
1049                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1050                         dprintk("Warning: CPU frequency is %u, "
1051                                "cpufreq assumed %u kHz.\n",
1052                                cur_freq, cpu_policy->cur);
1053
1054                 freqs.cpu = cpu;
1055                 freqs.old = cpu_policy->cur;
1056                 freqs.new = cur_freq;
1057
1058                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
1059                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1060                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1061
1062                 cpu_policy->cur = cur_freq;
1063         }
1064
1065 out:
1066         cpufreq_cpu_put(cpu_policy);
1067         return 0;
1068 }
1069
1070 /**
1071  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1072  *
1073  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1074  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1075  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1076  *          restored.
1077  */
1078 static int cpufreq_resume(struct sys_device * sysdev)
1079 {
1080         int cpu = sysdev->id;
1081         unsigned int ret = 0;
1082         struct cpufreq_policy *cpu_policy;
1083
1084         dprintk("resuming cpu %u\n", cpu);
1085
1086         if (!cpu_online(cpu))
1087                 return 0;
1088
1089         /* we may be lax here as interrupts are off. Nonetheless
1090          * we need to grab the correct cpu policy, as to check
1091          * whether we really run on this CPU.
1092          */
1093
1094         cpu_policy = cpufreq_cpu_get(cpu);
1095         if (!cpu_policy)
1096                 return -EINVAL;
1097
1098         /* only handle each CPU group once */
1099         if (unlikely(cpu_policy->cpu != cpu)) {
1100                 cpufreq_cpu_put(cpu_policy);
1101                 return 0;
1102         }
1103
1104         if (cpufreq_driver->resume) {
1105                 ret = cpufreq_driver->resume(cpu_policy);
1106                 if (ret) {
1107                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1108                                         "step on CPU %u\n", cpu_policy->cpu);
1109                         cpufreq_cpu_put(cpu_policy);
1110                         return ret;
1111                 }
1112         }
1113
1114         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1115                 unsigned int cur_freq = 0;
1116
1117                 if (cpufreq_driver->get)
1118                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1119
1120                 if (!cur_freq || !cpu_policy->cur) {
1121                         printk(KERN_ERR "cpufreq: resume failed to assert "
1122                                         "current frequency is what timing core "
1123                                         "thinks it is.\n");
1124                         goto out;
1125                 }
1126
1127                 if (unlikely(cur_freq != cpu_policy->cur)) {
1128                         struct cpufreq_freqs freqs;
1129
1130                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1131                                 dprintk("Warning: CPU frequency"
1132                                        "is %u, cpufreq assumed %u kHz.\n",
1133                                        cur_freq, cpu_policy->cur);
1134
1135                         freqs.cpu = cpu;
1136                         freqs.old = cpu_policy->cur;
1137                         freqs.new = cur_freq;
1138
1139                         srcu_notifier_call_chain(
1140                                         &cpufreq_transition_notifier_list,
1141                                         CPUFREQ_RESUMECHANGE, &freqs);
1142                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1143
1144                         cpu_policy->cur = cur_freq;
1145                 }
1146         }
1147
1148 out:
1149         schedule_work(&cpu_policy->update);
1150         cpufreq_cpu_put(cpu_policy);
1151         return ret;
1152 }
1153
1154 static struct sysdev_driver cpufreq_sysdev_driver = {
1155         .add            = cpufreq_add_dev,
1156         .remove         = cpufreq_remove_dev,
1157         .suspend        = cpufreq_suspend,
1158         .resume         = cpufreq_resume,
1159 };
1160
1161
1162 /*********************************************************************
1163  *                     NOTIFIER LISTS INTERFACE                      *
1164  *********************************************************************/
1165
1166 /**
1167  *      cpufreq_register_notifier - register a driver with cpufreq
1168  *      @nb: notifier function to register
1169  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1170  *
1171  *      Add a driver to one of two lists: either a list of drivers that
1172  *      are notified about clock rate changes (once before and once after
1173  *      the transition), or a list of drivers that are notified about
1174  *      changes in cpufreq policy.
1175  *
1176  *      This function may sleep, and has the same return conditions as
1177  *      blocking_notifier_chain_register.
1178  */
1179 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1180 {
1181         int ret;
1182
1183         switch (list) {
1184         case CPUFREQ_TRANSITION_NOTIFIER:
1185                 ret = srcu_notifier_chain_register(
1186                                 &cpufreq_transition_notifier_list, nb);
1187                 break;
1188         case CPUFREQ_POLICY_NOTIFIER:
1189                 ret = blocking_notifier_chain_register(
1190                                 &cpufreq_policy_notifier_list, nb);
1191                 break;
1192         default:
1193                 ret = -EINVAL;
1194         }
1195
1196         return ret;
1197 }
1198 EXPORT_SYMBOL(cpufreq_register_notifier);
1199
1200
1201 /**
1202  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1203  *      @nb: notifier block to be unregistered
1204  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1205  *
1206  *      Remove a driver from the CPU frequency notifier list.
1207  *
1208  *      This function may sleep, and has the same return conditions as
1209  *      blocking_notifier_chain_unregister.
1210  */
1211 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1212 {
1213         int ret;
1214
1215         switch (list) {
1216         case CPUFREQ_TRANSITION_NOTIFIER:
1217                 ret = srcu_notifier_chain_unregister(
1218                                 &cpufreq_transition_notifier_list, nb);
1219                 break;
1220         case CPUFREQ_POLICY_NOTIFIER:
1221                 ret = blocking_notifier_chain_unregister(
1222                                 &cpufreq_policy_notifier_list, nb);
1223                 break;
1224         default:
1225                 ret = -EINVAL;
1226         }
1227
1228         return ret;
1229 }
1230 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1231
1232
1233 /*********************************************************************
1234  *                              GOVERNORS                            *
1235  *********************************************************************/
1236
1237
1238 /* Must be called with lock_cpu_hotplug held */
1239 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1240                             unsigned int target_freq,
1241                             unsigned int relation)
1242 {
1243         int retval = -EINVAL;
1244
1245         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1246                 target_freq, relation);
1247         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1248                 retval = cpufreq_driver->target(policy, target_freq, relation);
1249
1250         return retval;
1251 }
1252 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1253
1254 int cpufreq_driver_target(struct cpufreq_policy *policy,
1255                           unsigned int target_freq,
1256                           unsigned int relation)
1257 {
1258         int ret;
1259
1260         policy = cpufreq_cpu_get(policy->cpu);
1261         if (!policy)
1262                 return -EINVAL;
1263
1264         lock_cpu_hotplug();
1265         mutex_lock(&policy->lock);
1266
1267         ret = __cpufreq_driver_target(policy, target_freq, relation);
1268
1269         mutex_unlock(&policy->lock);
1270         unlock_cpu_hotplug();
1271
1272         cpufreq_cpu_put(policy);
1273         return ret;
1274 }
1275 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1276
1277 /*
1278  * Locking: Must be called with the lock_cpu_hotplug() lock held
1279  * when "event" is CPUFREQ_GOV_LIMITS
1280  */
1281
1282 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1283 {
1284         int ret;
1285
1286         if (!try_module_get(policy->governor->owner))
1287                 return -EINVAL;
1288
1289         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1290         ret = policy->governor->governor(policy, event);
1291
1292         /* we keep one module reference alive for each CPU governed by this CPU */
1293         if ((event != CPUFREQ_GOV_START) || ret)
1294                 module_put(policy->governor->owner);
1295         if ((event == CPUFREQ_GOV_STOP) && !ret)
1296                 module_put(policy->governor->owner);
1297
1298         return ret;
1299 }
1300
1301
1302 int cpufreq_register_governor(struct cpufreq_governor *governor)
1303 {
1304         int err;
1305
1306         if (!governor)
1307                 return -EINVAL;
1308
1309         mutex_lock(&cpufreq_governor_mutex);
1310
1311         err = -EBUSY;
1312         if (__find_governor(governor->name) == NULL) {
1313                 err = 0;
1314                 list_add(&governor->governor_list, &cpufreq_governor_list);
1315         }
1316
1317         mutex_unlock(&cpufreq_governor_mutex);
1318         return err;
1319 }
1320 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1321
1322
1323 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1324 {
1325         if (!governor)
1326                 return;
1327
1328         mutex_lock(&cpufreq_governor_mutex);
1329         list_del(&governor->governor_list);
1330         mutex_unlock(&cpufreq_governor_mutex);
1331         return;
1332 }
1333 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1334
1335
1336
1337 /*********************************************************************
1338  *                          POLICY INTERFACE                         *
1339  *********************************************************************/
1340
1341 /**
1342  * cpufreq_get_policy - get the current cpufreq_policy
1343  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1344  *
1345  * Reads the current cpufreq policy.
1346  */
1347 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1348 {
1349         struct cpufreq_policy *cpu_policy;
1350         if (!policy)
1351                 return -EINVAL;
1352
1353         cpu_policy = cpufreq_cpu_get(cpu);
1354         if (!cpu_policy)
1355                 return -EINVAL;
1356
1357         mutex_lock(&cpu_policy->lock);
1358         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1359         mutex_unlock(&cpu_policy->lock);
1360
1361         cpufreq_cpu_put(cpu_policy);
1362         return 0;
1363 }
1364 EXPORT_SYMBOL(cpufreq_get_policy);
1365
1366
1367 /*
1368  * Locking: Must be called with the lock_cpu_hotplug() lock held
1369  */
1370 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1371 {
1372         int ret = 0;
1373
1374         cpufreq_debug_disable_ratelimit();
1375         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1376                 policy->min, policy->max);
1377
1378         memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
1379
1380         if (policy->min > data->min && policy->min > policy->max) {
1381                 ret = -EINVAL;
1382                 goto error_out;
1383         }
1384
1385         /* verify the cpu speed can be set within this limit */
1386         ret = cpufreq_driver->verify(policy);
1387         if (ret)
1388                 goto error_out;
1389
1390         /* adjust if necessary - all reasons */
1391         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1392                         CPUFREQ_ADJUST, policy);
1393
1394         /* adjust if necessary - hardware incompatibility*/
1395         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1396                         CPUFREQ_INCOMPATIBLE, policy);
1397
1398         /* verify the cpu speed can be set within this limit,
1399            which might be different to the first one */
1400         ret = cpufreq_driver->verify(policy);
1401         if (ret)
1402                 goto error_out;
1403
1404         /* notification of the new policy */
1405         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1406                         CPUFREQ_NOTIFY, policy);
1407
1408         data->min = policy->min;
1409         data->max = policy->max;
1410
1411         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1412
1413         if (cpufreq_driver->setpolicy) {
1414                 data->policy = policy->policy;
1415                 dprintk("setting range\n");
1416                 ret = cpufreq_driver->setpolicy(policy);
1417         } else {
1418                 if (policy->governor != data->governor) {
1419                         /* save old, working values */
1420                         struct cpufreq_governor *old_gov = data->governor;
1421
1422                         dprintk("governor switch\n");
1423
1424                         /* end old governor */
1425                         if (data->governor)
1426                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1427
1428                         /* start new governor */
1429                         data->governor = policy->governor;
1430                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1431                                 /* new governor failed, so re-start old one */
1432                                 dprintk("starting governor %s failed\n", data->governor->name);
1433                                 if (old_gov) {
1434                                         data->governor = old_gov;
1435                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1436                                 }
1437                                 ret = -EINVAL;
1438                                 goto error_out;
1439                         }
1440                         /* might be a policy change, too, so fall through */
1441                 }
1442                 dprintk("governor: change or update limits\n");
1443                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1444         }
1445
1446 error_out:
1447         cpufreq_debug_enable_ratelimit();
1448         return ret;
1449 }
1450
1451 /**
1452  *      cpufreq_set_policy - set a new CPUFreq policy
1453  *      @policy: policy to be set.
1454  *
1455  *      Sets a new CPU frequency and voltage scaling policy.
1456  */
1457 int cpufreq_set_policy(struct cpufreq_policy *policy)
1458 {
1459         int ret = 0;
1460         struct cpufreq_policy *data;
1461
1462         if (!policy)
1463                 return -EINVAL;
1464
1465         data = cpufreq_cpu_get(policy->cpu);
1466         if (!data)
1467                 return -EINVAL;
1468
1469         lock_cpu_hotplug();
1470
1471         /* lock this CPU */
1472         mutex_lock(&data->lock);
1473
1474         ret = __cpufreq_set_policy(data, policy);
1475         data->user_policy.min = data->min;
1476         data->user_policy.max = data->max;
1477         data->user_policy.policy = data->policy;
1478         data->user_policy.governor = data->governor;
1479
1480         mutex_unlock(&data->lock);
1481
1482         unlock_cpu_hotplug();
1483         cpufreq_cpu_put(data);
1484
1485         return ret;
1486 }
1487 EXPORT_SYMBOL(cpufreq_set_policy);
1488
1489
1490 /**
1491  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1492  *      @cpu: CPU which shall be re-evaluated
1493  *
1494  *      Usefull for policy notifiers which have different necessities
1495  *      at different times.
1496  */
1497 int cpufreq_update_policy(unsigned int cpu)
1498 {
1499         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1500         struct cpufreq_policy policy;
1501         int ret = 0;
1502
1503         if (!data)
1504                 return -ENODEV;
1505
1506         lock_cpu_hotplug();
1507         mutex_lock(&data->lock);
1508
1509         dprintk("updating policy for CPU %u\n", cpu);
1510         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1511         policy.min = data->user_policy.min;
1512         policy.max = data->user_policy.max;
1513         policy.policy = data->user_policy.policy;
1514         policy.governor = data->user_policy.governor;
1515
1516         /* BIOS might change freq behind our back
1517           -> ask driver for current freq and notify governors about a change */
1518         if (cpufreq_driver->get) {
1519                 policy.cur = cpufreq_driver->get(cpu);
1520                 if (!data->cur) {
1521                         dprintk("Driver did not initialize current freq");
1522                         data->cur = policy.cur;
1523                 } else {
1524                         if (data->cur != policy.cur)
1525                                 cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1526                 }
1527         }
1528
1529         ret = __cpufreq_set_policy(data, &policy);
1530
1531         mutex_unlock(&data->lock);
1532         unlock_cpu_hotplug();
1533         cpufreq_cpu_put(data);
1534         return ret;
1535 }
1536 EXPORT_SYMBOL(cpufreq_update_policy);
1537
1538 #ifdef CONFIG_HOTPLUG_CPU
1539 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1540                                         unsigned long action, void *hcpu)
1541 {
1542         unsigned int cpu = (unsigned long)hcpu;
1543         struct cpufreq_policy *policy;
1544         struct sys_device *sys_dev;
1545
1546         sys_dev = get_cpu_sysdev(cpu);
1547
1548         if (sys_dev) {
1549                 switch (action) {
1550                 case CPU_ONLINE:
1551                         cpufreq_add_dev(sys_dev);
1552                         break;
1553                 case CPU_DOWN_PREPARE:
1554                         /*
1555                          * We attempt to put this cpu in lowest frequency
1556                          * possible before going down. This will permit
1557                          * hardware-managed P-State to switch other related
1558                          * threads to min or higher speeds if possible.
1559                          */
1560                         policy = cpufreq_cpu_data[cpu];
1561                         if (policy) {
1562                                 cpufreq_driver_target(policy, policy->min,
1563                                                 CPUFREQ_RELATION_H);
1564                         }
1565                         break;
1566                 case CPU_DEAD:
1567                         cpufreq_remove_dev(sys_dev);
1568                         break;
1569                 }
1570         }
1571         return NOTIFY_OK;
1572 }
1573
1574 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1575 {
1576     .notifier_call = cpufreq_cpu_callback,
1577 };
1578 #endif /* CONFIG_HOTPLUG_CPU */
1579
1580 /*********************************************************************
1581  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1582  *********************************************************************/
1583
1584 /**
1585  * cpufreq_register_driver - register a CPU Frequency driver
1586  * @driver_data: A struct cpufreq_driver containing the values#
1587  * submitted by the CPU Frequency driver.
1588  *
1589  *   Registers a CPU Frequency driver to this core code. This code
1590  * returns zero on success, -EBUSY when another driver got here first
1591  * (and isn't unregistered in the meantime).
1592  *
1593  */
1594 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1595 {
1596         unsigned long flags;
1597         int ret;
1598
1599         if (!driver_data || !driver_data->verify || !driver_data->init ||
1600             ((!driver_data->setpolicy) && (!driver_data->target)))
1601                 return -EINVAL;
1602
1603         dprintk("trying to register driver %s\n", driver_data->name);
1604
1605         if (driver_data->setpolicy)
1606                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1607
1608         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1609         if (cpufreq_driver) {
1610                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1611                 return -EBUSY;
1612         }
1613         cpufreq_driver = driver_data;
1614         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1615
1616         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1617
1618         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1619                 int i;
1620                 ret = -ENODEV;
1621
1622                 /* check for at least one working CPU */
1623                 for (i=0; i<NR_CPUS; i++)
1624                         if (cpufreq_cpu_data[i])
1625                                 ret = 0;
1626
1627                 /* if all ->init() calls failed, unregister */
1628                 if (ret) {
1629                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1630                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1631
1632                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1633                         cpufreq_driver = NULL;
1634                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1635                 }
1636         }
1637
1638         if (!ret) {
1639                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1640                 dprintk("driver %s up and running\n", driver_data->name);
1641                 cpufreq_debug_enable_ratelimit();
1642         }
1643
1644         return (ret);
1645 }
1646 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1647
1648
1649 /**
1650  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1651  *
1652  *    Unregister the current CPUFreq driver. Only call this if you have
1653  * the right to do so, i.e. if you have succeeded in initialising before!
1654  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1655  * currently not initialised.
1656  */
1657 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1658 {
1659         unsigned long flags;
1660
1661         cpufreq_debug_disable_ratelimit();
1662
1663         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1664                 cpufreq_debug_enable_ratelimit();
1665                 return -EINVAL;
1666         }
1667
1668         dprintk("unregistering driver %s\n", driver->name);
1669
1670         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1671         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1672
1673         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1674         cpufreq_driver = NULL;
1675         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1676
1677         return 0;
1678 }
1679 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);