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  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
   7  *
   8  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
   9  *      Added handling for CPU hotplug
  10  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
  11  *      Fix handling for CPU hotplug -- affected CPUs
  12  *
  13  * This program is free software; you can redistribute it and/or modify
  14  * it under the terms of the GNU General Public License version 2 as
  15  * published by the Free Software Foundation.
  16  */
  17
  18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  19
  20 #include <linux/cpu.h>
  21 #include <linux/cpufreq.h>
  22 #include <linux/delay.h>
  23 #include <linux/device.h>
  24 #include <linux/init.h>
  25 #include <linux/kernel_stat.h>
  26 #include <linux/module.h>
  27 #include <linux/mutex.h>
  28 #include <linux/slab.h>
  29 #include <linux/suspend.h>
  30 #include <linux/syscore_ops.h>
  31 #include <linux/tick.h>
  32 #include <trace/events/power.h>
  33
  34 static LIST_HEAD(cpufreq_policy_list);
  35
  36 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
  37 {
  38         return cpumask_empty(policy->cpus);
  39 }
  40
  41 /* Macros to iterate over CPU policies */
  42 #define for_each_suitable_policy(__policy, __active)                     \
  43         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
  44                 if ((__active) == !policy_is_inactive(__policy))
  45
  46 #define for_each_active_policy(__policy)                \
  47         for_each_suitable_policy(__policy, true)
  48 #define for_each_inactive_policy(__policy)              \
  49         for_each_suitable_policy(__policy, false)
  50
  51 #define for_each_policy(__policy)                       \
  52         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
  53
  54 /* Iterate over governors */
  55 static LIST_HEAD(cpufreq_governor_list);
  56 #define for_each_governor(__governor)                           \
  57         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
  58
  59 /**
  60  * The "cpufreq driver" - the arch- or hardware-dependent low
  61  * level driver of CPUFreq support, and its spinlock. This lock
  62  * also protects the cpufreq_cpu_data array.
  63  */
  64 static struct cpufreq_driver *cpufreq_driver;
  65 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
  66 static DEFINE_RWLOCK(cpufreq_driver_lock);
  67
  68 /* Flag to suspend/resume CPUFreq governors */
  69 static bool cpufreq_suspended;
  70
  71 static inline bool has_target(void)
  72 {
  73         return cpufreq_driver->target_index || cpufreq_driver->target;
  74 }
  75
  76 /* internal prototypes */
  77 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
  78 static int cpufreq_init_governor(struct cpufreq_policy *policy);
  79 static void cpufreq_exit_governor(struct cpufreq_policy *policy);
  80 static int cpufreq_start_governor(struct cpufreq_policy *policy);
  81 static void cpufreq_stop_governor(struct cpufreq_policy *policy);
  82 static void cpufreq_governor_limits(struct cpufreq_policy *policy);
  83
  84 /**
  85  * Two notifier lists: the "policy" list is involved in the
  86  * validation process for a new CPU frequency policy; the
  87  * "transition" list for kernel code that needs to handle
  88  * changes to devices when the CPU clock speed changes.
  89  * The mutex locks both lists.
  90  */
  91 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
  92 static struct srcu_notifier_head cpufreq_transition_notifier_list;
  93
  94 static bool init_cpufreq_transition_notifier_list_called;
  95 static int __init init_cpufreq_transition_notifier_list(void)
  96 {
  97         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
  98         init_cpufreq_transition_notifier_list_called = true;
  99         return 0;
 100 }
 101 pure_initcall(init_cpufreq_transition_notifier_list);
 102
 103 static int off __read_mostly;
 104 static int cpufreq_disabled(void)
 105 {
 106         return off;
 107 }
 108 void disable_cpufreq(void)
 109 {
 110         off = 1;
 111 }
 112 static DEFINE_MUTEX(cpufreq_governor_mutex);
 113
 114 bool have_governor_per_policy(void)
 115 {
 116         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
 117 }
 118 EXPORT_SYMBOL_GPL(have_governor_per_policy);
 119
 120 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
 121 {
 122         if (have_governor_per_policy())
 123                 return &policy->kobj;
 124         else
 125                 return cpufreq_global_kobject;
 126 }
 127 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
 128
 129 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
 130 {
 131         u64 idle_time;
 132         u64 cur_wall_time;
 133         u64 busy_time;
 134
 135         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
 136
 137         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
 138         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
 139         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
 140         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
 141         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
 142         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
 143
 144         idle_time = cur_wall_time - busy_time;
 145         if (wall)
 146                 *wall = cputime_to_usecs(cur_wall_time);
 147
 148         return cputime_to_usecs(idle_time);
 149 }
 150
 151 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
 152 {
 153         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
 154
 155         if (idle_time == -1ULL)
 156                 return get_cpu_idle_time_jiffy(cpu, wall);
 157         else if (!io_busy)
 158                 idle_time += get_cpu_iowait_time_us(cpu, wall);
 159
 160         return idle_time;
 161 }
 162 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
 163
 164 /*
 165  * This is a generic cpufreq init() routine which can be used by cpufreq
 166  * drivers of SMP systems. It will do following:
 167  * - validate & show freq table passed
 168  * - set policies transition latency
 169  * - policy->cpus with all possible CPUs
 170  */
 171 int cpufreq_generic_init(struct cpufreq_policy *policy,
 172                 struct cpufreq_frequency_table *table,
 173                 unsigned int transition_latency)
 174 {
 175         int ret;
 176
 177         ret = cpufreq_table_validate_and_show(policy, table);
 178         if (ret) {
 179                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
 180                 return ret;
 181         }
 182
 183         policy->cpuinfo.transition_latency = transition_latency;
 184
 185         /*
 186          * The driver only supports the SMP configuration where all processors
 187          * share the clock and voltage and clock.
 188          */
 189         cpumask_setall(policy->cpus);
 190
 191         return 0;
 192 }
 193 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
 194
 195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
 196 {
 197         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
 198
 199         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
 200 }
 201 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
 202
 203 unsigned int cpufreq_generic_get(unsigned int cpu)
 204 {
 205         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
 206
 207         if (!policy || IS_ERR(policy->clk)) {
 208                 pr_err("%s: No %s associated to cpu: %d\n",
 209                        __func__, policy ? "clk" : "policy", cpu);
 210                 return 0;
 211         }
 212
 213         return clk_get_rate(policy->clk) / 1000;
 214 }
 215 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
 216
 217 /**
 218  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
 219  *
 220  * @cpu: cpu to find policy for.
 221  *
 222  * This returns policy for 'cpu', returns NULL if it doesn't exist.
 223  * It also increments the kobject reference count to mark it busy and so would
 224  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
 225  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
 226  * freed as that depends on the kobj count.
 227  *
 228  * Return: A valid policy on success, otherwise NULL on failure.
 229  */
 230 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
 231 {
 232         struct cpufreq_policy *policy = NULL;
 233         unsigned long flags;
 234
 235         if (WARN_ON(cpu >= nr_cpu_ids))
 236                 return NULL;
 237
 238         /* get the cpufreq driver */
 239         read_lock_irqsave(&cpufreq_driver_lock, flags);
 240
 241         if (cpufreq_driver) {
 242                 /* get the CPU */
 243                 policy = cpufreq_cpu_get_raw(cpu);
 244                 if (policy)
 245                         kobject_get(&policy->kobj);
 246         }
 247
 248         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
 249
 250         return policy;
 251 }
 252 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 253
 254 /**
 255  * cpufreq_cpu_put: Decrements the usage count of a policy
 256  *
 257  * @policy: policy earlier returned by cpufreq_cpu_get().
 258  *
 259  * This decrements the kobject reference count incremented earlier by calling
 260  * cpufreq_cpu_get().
 261  */
 262 void cpufreq_cpu_put(struct cpufreq_policy *policy)
 263 {
 264         kobject_put(&policy->kobj);
 265 }
 266 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 267
 268 /*********************************************************************
 269  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 270  *********************************************************************/
 271
 272 /**
 273  * adjust_jiffies - adjust the system "loops_per_jiffy"
 274  *
 275  * This function alters the system "loops_per_jiffy" for the clock
 276  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 277  * systems as each CPU might be scaled differently. So, use the arch
 278  * per-CPU loops_per_jiffy value wherever possible.
 279  */
 280 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 281 {
 282 #ifndef CONFIG_SMP
 283         static unsigned long l_p_j_ref;
 284         static unsigned int l_p_j_ref_freq;
 285
 286         if (ci->flags & CPUFREQ_CONST_LOOPS)
 287                 return;
 288
 289         if (!l_p_j_ref_freq) {
 290                 l_p_j_ref = loops_per_jiffy;
 291                 l_p_j_ref_freq = ci->old;
 292                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
 293                          l_p_j_ref, l_p_j_ref_freq);
 294         }
 295         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
 296                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
 297                                                                 ci->new);
 298                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
 299                          loops_per_jiffy, ci->new);
 300         }
 301 #endif
 302 }
 303
 304 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
 305                 struct cpufreq_freqs *freqs, unsigned int state)
 306 {
 307         BUG_ON(irqs_disabled());
 308
 309         if (cpufreq_disabled())
 310                 return;
 311
 312         freqs->flags = cpufreq_driver->flags;
 313         pr_debug("notification %u of frequency transition to %u kHz\n",
 314                  state, freqs->new);
 315
 316         switch (state) {
 317
 318         case CPUFREQ_PRECHANGE:
 319                 /* detect if the driver reported a value as "old frequency"
 320                  * which is not equal to what the cpufreq core thinks is
 321                  * "old frequency".
 322                  */
 323                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 324                         if ((policy) && (policy->cpu == freqs->cpu) &&
 325                             (policy->cur) && (policy->cur != freqs->old)) {
 326                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
 327                                          freqs->old, policy->cur);
 328                                 freqs->old = policy->cur;
 329                         }
 330                 }
 331                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 332                                 CPUFREQ_PRECHANGE, freqs);
 333                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 334                 break;
 335
 336         case CPUFREQ_POSTCHANGE:
 337                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 338                 pr_debug("FREQ: %lu - CPU: %lu\n",
 339                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
 340                 trace_cpu_frequency(freqs->new, freqs->cpu);
 341                 cpufreq_stats_record_transition(policy, freqs->new);
 342                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 343                                 CPUFREQ_POSTCHANGE, freqs);
 344                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 345                         policy->cur = freqs->new;
 346                 break;
 347         }
 348 }
 349
 350 /**
 351  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 352  * on frequency transition.
 353  *
 354  * This function calls the transition notifiers and the "adjust_jiffies"
 355  * function. It is called twice on all CPU frequency changes that have
 356  * external effects.
 357  */
 358 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
 359                 struct cpufreq_freqs *freqs, unsigned int state)
 360 {
 361         for_each_cpu(freqs->cpu, policy->cpus)
 362                 __cpufreq_notify_transition(policy, freqs, state);
 363 }
 364
 365 /* Do post notifications when there are chances that transition has failed */
 366 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
 367                 struct cpufreq_freqs *freqs, int transition_failed)
 368 {
 369         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
 370         if (!transition_failed)
 371                 return;
 372
 373         swap(freqs->old, freqs->new);
 374         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
 375         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
 376 }
 377
 378 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
 379                 struct cpufreq_freqs *freqs)
 380 {
 381
 382         /*
 383          * Catch double invocations of _begin() which lead to self-deadlock.
 384          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
 385          * doesn't invoke _begin() on their behalf, and hence the chances of
 386          * double invocations are very low. Moreover, there are scenarios
 387          * where these checks can emit false-positive warnings in these
 388          * drivers; so we avoid that by skipping them altogether.
 389          */
 390         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
 391                                 && current == policy->transition_task);
 392
 393 wait:
 394         wait_event(policy->transition_wait, !policy->transition_ongoing);
 395
 396         spin_lock(&policy->transition_lock);
 397
 398         if (unlikely(policy->transition_ongoing)) {
 399                 spin_unlock(&policy->transition_lock);
 400                 goto wait;
 401         }
 402
 403         policy->transition_ongoing = true;
 404         policy->transition_task = current;
 405
 406         spin_unlock(&policy->transition_lock);
 407
 408         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
 409 }
 410 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
 411
 412 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
 413                 struct cpufreq_freqs *freqs, int transition_failed)
 414 {
 415         if (unlikely(WARN_ON(!policy->transition_ongoing)))
 416                 return;
 417
 418         cpufreq_notify_post_transition(policy, freqs, transition_failed);
 419
 420         policy->transition_ongoing = false;
 421         policy->transition_task = NULL;
 422
 423         wake_up(&policy->transition_wait);
 424 }
 425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
 426
 427 /*
 428  * Fast frequency switching status count.  Positive means "enabled", negative
 429  * means "disabled" and 0 means "not decided yet".
 430  */
 431 static int cpufreq_fast_switch_count;
 432 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
 433
 434 static void cpufreq_list_transition_notifiers(void)
 435 {
 436         struct notifier_block *nb;
 437
 438         pr_info("Registered transition notifiers:\n");
 439
 440         mutex_lock(&cpufreq_transition_notifier_list.mutex);
 441
 442         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
 443                 pr_info("%pF\n", nb->notifier_call);
 444
 445         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
 446 }
 447
 448 /**
 449  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
 450  * @policy: cpufreq policy to enable fast frequency switching for.
 451  *
 452  * Try to enable fast frequency switching for @policy.
 453  *
 454  * The attempt will fail if there is at least one transition notifier registered
 455  * at this point, as fast frequency switching is quite fundamentally at odds
 456  * with transition notifiers.  Thus if successful, it will make registration of
 457  * transition notifiers fail going forward.
 458  */
 459 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
 460 {
 461         lockdep_assert_held(&policy->rwsem);
 462
 463         if (!policy->fast_switch_possible)
 464                 return;
 465
 466         mutex_lock(&cpufreq_fast_switch_lock);
 467         if (cpufreq_fast_switch_count >= 0) {
 468                 cpufreq_fast_switch_count++;
 469                 policy->fast_switch_enabled = true;
 470         } else {
 471                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
 472                         policy->cpu);
 473                 cpufreq_list_transition_notifiers();
 474         }
 475         mutex_unlock(&cpufreq_fast_switch_lock);
 476 }
 477 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
 478
 479 /**
 480  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
 481  * @policy: cpufreq policy to disable fast frequency switching for.
 482  */
 483 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
 484 {
 485         mutex_lock(&cpufreq_fast_switch_lock);
 486         if (policy->fast_switch_enabled) {
 487                 policy->fast_switch_enabled = false;
 488                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
 489                         cpufreq_fast_switch_count--;
 490         }
 491         mutex_unlock(&cpufreq_fast_switch_lock);
 492 }
 493 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
 494
 495 /**
 496  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
 497  * one.
 498  * @target_freq: target frequency to resolve.
 499  *
 500  * The target to driver frequency mapping is cached in the policy.
 501  *
 502  * Return: Lowest driver-supported frequency greater than or equal to the
 503  * given target_freq, subject to policy (min/max) and driver limitations.
 504  */
 505 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
 506                                          unsigned int target_freq)
 507 {
 508         target_freq = clamp_val(target_freq, policy->min, policy->max);
 509         policy->cached_target_freq = target_freq;
 510
 511         if (cpufreq_driver->target_index) {
 512                 int idx;
 513
 514                 idx = cpufreq_frequency_table_target(policy, target_freq,
 515                                                      CPUFREQ_RELATION_L);
 516                 policy->cached_resolved_idx = idx;
 517                 return policy->freq_table[idx].frequency;
 518         }
 519
 520         if (cpufreq_driver->resolve_freq)
 521                 return cpufreq_driver->resolve_freq(policy, target_freq);
 522
 523         return target_freq;
 524 }
 525 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
 526
 527 /*********************************************************************
 528  *                          SYSFS INTERFACE                          *
 529  *********************************************************************/
 530 static ssize_t show_boost(struct kobject *kobj,
 531                                  struct attribute *attr, char *buf)
 532 {
 533         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
 534 }
 535
 536 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
 537                                   const char *buf, size_t count)
 538 {
 539         int ret, enable;
 540
 541         ret = sscanf(buf, "%d", &enable);
 542         if (ret != 1 || enable < 0 || enable > 1)
 543                 return -EINVAL;
 544
 545         if (cpufreq_boost_trigger_state(enable)) {
 546                 pr_err("%s: Cannot %s BOOST!\n",
 547                        __func__, enable ? "enable" : "disable");
 548                 return -EINVAL;
 549         }
 550
 551         pr_debug("%s: cpufreq BOOST %s\n",
 552                  __func__, enable ? "enabled" : "disabled");
 553
 554         return count;
 555 }
 556 define_one_global_rw(boost);
 557
 558 static struct cpufreq_governor *find_governor(const char *str_governor)
 559 {
 560         struct cpufreq_governor *t;
 561
 562         for_each_governor(t)
 563                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
 564                         return t;
 565
 566         return NULL;
 567 }
 568
 569 /**
 570  * cpufreq_parse_governor - parse a governor string
 571  */
 572 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
 573                                 struct cpufreq_governor **governor)
 574 {
 575         int err = -EINVAL;
 576
 577         if (cpufreq_driver->setpolicy) {
 578                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 579                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 580                         err = 0;
 581                 } else if (!strncasecmp(str_governor, "powersave",
 582                                                 CPUFREQ_NAME_LEN)) {
 583                         *policy = CPUFREQ_POLICY_POWERSAVE;
 584                         err = 0;
 585                 }
 586         } else {
 587                 struct cpufreq_governor *t;
 588
 589                 mutex_lock(&cpufreq_governor_mutex);
 590
 591                 t = find_governor(str_governor);
 592
 593                 if (t == NULL) {
 594                         int ret;
 595
 596                         mutex_unlock(&cpufreq_governor_mutex);
 597                         ret = request_module("cpufreq_%s", str_governor);
 598                         mutex_lock(&cpufreq_governor_mutex);
 599
 600                         if (ret == 0)
 601                                 t = find_governor(str_governor);
 602                 }
 603
 604                 if (t != NULL) {
 605                         *governor = t;
 606                         err = 0;
 607                 }
 608
 609                 mutex_unlock(&cpufreq_governor_mutex);
 610         }
 611         return err;
 612 }
 613
 614 /**
 615  * cpufreq_per_cpu_attr_read() / show_##file_name() -
 616  * print out cpufreq information
 617  *
 618  * Write out information from cpufreq_driver->policy[cpu]; object must be
 619  * "unsigned int".
 620  */
 621
 622 #define show_one(file_name, object)                     \
 623 static ssize_t show_##file_name                         \
 624 (struct cpufreq_policy *policy, char *buf)              \
 625 {                                                       \
 626         return sprintf(buf, "%u\n", policy->object);    \
 627 }
 628
 629 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 630 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 631 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
 632 show_one(scaling_min_freq, min);
 633 show_one(scaling_max_freq, max);
 634
 635 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
 636 {
 637         ssize_t ret;
 638
 639         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
 640                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
 641         else
 642                 ret = sprintf(buf, "%u\n", policy->cur);
 643         return ret;
 644 }
 645
 646 static int cpufreq_set_policy(struct cpufreq_policy *policy,
 647                                 struct cpufreq_policy *new_policy);
 648
 649 /**
 650  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 651  */
 652 #define store_one(file_name, object)                    \
 653 static ssize_t store_##file_name                                        \
 654 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
 655 {                                                                       \
 656         int ret, temp;                                                  \
 657         struct cpufreq_policy new_policy;                               \
 658                                                                         \
 659         memcpy(&new_policy, policy, sizeof(*policy));                   \
 660                                                                         \
 661         ret = sscanf(buf, "%u", &new_policy.object);                    \
 662         if (ret != 1)                                                   \
 663                 return -EINVAL;                                         \
 664                                                                         \
 665         temp = new_policy.object;                                       \
 666         ret = cpufreq_set_policy(policy, &new_policy);          \
 667         if (!ret)                                                       \
 668                 policy->user_policy.object = temp;                      \
 669                                                                         \
 670         return ret ? ret : count;                                       \
 671 }
 672
 673 store_one(scaling_min_freq, min);
 674 store_one(scaling_max_freq, max);
 675
 676 /**
 677  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 678  */
 679 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
 680                                         char *buf)
 681 {
 682         unsigned int cur_freq = __cpufreq_get(policy);
 683         if (!cur_freq)
 684                 return sprintf(buf, "<unknown>");
 685         return sprintf(buf, "%u\n", cur_freq);
 686 }
 687
 688 /**
 689  * show_scaling_governor - show the current policy for the specified CPU
 690  */
 691 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
 692 {
 693         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
 694                 return sprintf(buf, "powersave\n");
 695         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 696                 return sprintf(buf, "performance\n");
 697         else if (policy->governor)
 698                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
 699                                 policy->governor->name);
 700         return -EINVAL;
 701 }
 702
 703 /**
 704  * store_scaling_governor - store policy for the specified CPU
 705  */
 706 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
 707                                         const char *buf, size_t count)
 708 {
 709         int ret;
 710         char    str_governor[16];
 711         struct cpufreq_policy new_policy;
 712
 713         memcpy(&new_policy, policy, sizeof(*policy));
 714
 715         ret = sscanf(buf, "%15s", str_governor);
 716         if (ret != 1)
 717                 return -EINVAL;
 718
 719         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
 720                                                 &new_policy.governor))
 721                 return -EINVAL;
 722
 723         ret = cpufreq_set_policy(policy, &new_policy);
 724         return ret ? ret : count;
 725 }
 726
 727 /**
 728  * show_scaling_driver - show the cpufreq driver currently loaded
 729  */
 730 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
 731 {
 732         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
 733 }
 734
 735 /**
 736  * show_scaling_available_governors - show the available CPUfreq governors
 737  */
 738 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
 739                                                 char *buf)
 740 {
 741         ssize_t i = 0;
 742         struct cpufreq_governor *t;
 743
 744         if (!has_target()) {
 745                 i += sprintf(buf, "performance powersave");
 746                 goto out;
 747         }
 748
 749         for_each_governor(t) {
 750                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
 751                     - (CPUFREQ_NAME_LEN + 2)))
 752                         goto out;
 753                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
 754         }
 755 out:
 756         i += sprintf(&buf[i], "\n");
 757         return i;
 758 }
 759
 760 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
 761 {
 762         ssize_t i = 0;
 763         unsigned int cpu;
 764
 765         for_each_cpu(cpu, mask) {
 766                 if (i)
 767                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 768                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 769                 if (i >= (PAGE_SIZE - 5))
 770                         break;
 771         }
 772         i += sprintf(&buf[i], "\n");
 773         return i;
 774 }
 775 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
 776
 777 /**
 778  * show_related_cpus - show the CPUs affected by each transition even if
 779  * hw coordination is in use
 780  */
 781 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
 782 {
 783         return cpufreq_show_cpus(policy->related_cpus, buf);
 784 }
 785
 786 /**
 787  * show_affected_cpus - show the CPUs affected by each transition
 788  */
 789 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
 790 {
 791         return cpufreq_show_cpus(policy->cpus, buf);
 792 }
 793
 794 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
 795                                         const char *buf, size_t count)
 796 {
 797         unsigned int freq = 0;
 798         unsigned int ret;
 799
 800         if (!policy->governor || !policy->governor->store_setspeed)
 801                 return -EINVAL;
 802
 803         ret = sscanf(buf, "%u", &freq);
 804         if (ret != 1)
 805                 return -EINVAL;
 806
 807         policy->governor->store_setspeed(policy, freq);
 808
 809         return count;
 810 }
 811
 812 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
 813 {
 814         if (!policy->governor || !policy->governor->show_setspeed)
 815                 return sprintf(buf, "<unsupported>\n");
 816
 817         return policy->governor->show_setspeed(policy, buf);
 818 }
 819
 820 /**
 821  * show_bios_limit - show the current cpufreq HW/BIOS limitation
 822  */
 823 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
 824 {
 825         unsigned int limit;
 826         int ret;
 827         if (cpufreq_driver->bios_limit) {
 828                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
 829                 if (!ret)
 830                         return sprintf(buf, "%u\n", limit);
 831         }
 832         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
 833 }
 834
 835 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
 836 cpufreq_freq_attr_ro(cpuinfo_min_freq);
 837 cpufreq_freq_attr_ro(cpuinfo_max_freq);
 838 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
 839 cpufreq_freq_attr_ro(scaling_available_governors);
 840 cpufreq_freq_attr_ro(scaling_driver);
 841 cpufreq_freq_attr_ro(scaling_cur_freq);
 842 cpufreq_freq_attr_ro(bios_limit);
 843 cpufreq_freq_attr_ro(related_cpus);
 844 cpufreq_freq_attr_ro(affected_cpus);
 845 cpufreq_freq_attr_rw(scaling_min_freq);
 846 cpufreq_freq_attr_rw(scaling_max_freq);
 847 cpufreq_freq_attr_rw(scaling_governor);
 848 cpufreq_freq_attr_rw(scaling_setspeed);
 849
 850 static struct attribute *default_attrs[] = {
 851         &cpuinfo_min_freq.attr,
 852         &cpuinfo_max_freq.attr,
 853         &cpuinfo_transition_latency.attr,
 854         &scaling_min_freq.attr,
 855         &scaling_max_freq.attr,
 856         &affected_cpus.attr,
 857         &related_cpus.attr,
 858         &scaling_governor.attr,
 859         &scaling_driver.attr,
 860         &scaling_available_governors.attr,
 861         &scaling_setspeed.attr,
 862         NULL
 863 };
 864
 865 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
 866 #define to_attr(a) container_of(a, struct freq_attr, attr)
 867
 868 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 869 {
 870         struct cpufreq_policy *policy = to_policy(kobj);
 871         struct freq_attr *fattr = to_attr(attr);
 872         ssize_t ret;
 873
 874         down_read(&policy->rwsem);
 875         ret = fattr->show(policy, buf);
 876         up_read(&policy->rwsem);
 877
 878         return ret;
 879 }
 880
 881 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 882                      const char *buf, size_t count)
 883 {
 884         struct cpufreq_policy *policy = to_policy(kobj);
 885         struct freq_attr *fattr = to_attr(attr);
 886         ssize_t ret = -EINVAL;
 887
 888         get_online_cpus();
 889
 890         if (cpu_online(policy->cpu)) {
 891                 down_write(&policy->rwsem);
 892                 ret = fattr->store(policy, buf, count);
 893                 up_write(&policy->rwsem);
 894         }
 895
 896         put_online_cpus();
 897
 898         return ret;
 899 }
 900
 901 static void cpufreq_sysfs_release(struct kobject *kobj)
 902 {
 903         struct cpufreq_policy *policy = to_policy(kobj);
 904         pr_debug("last reference is dropped\n");
 905         complete(&policy->kobj_unregister);
 906 }
 907
 908 static const struct sysfs_ops sysfs_ops = {
 909         .show   = show,
 910         .store  = store,
 911 };
 912
 913 static struct kobj_type ktype_cpufreq = {
 914         .sysfs_ops      = &sysfs_ops,
 915         .default_attrs  = default_attrs,
 916         .release        = cpufreq_sysfs_release,
 917 };
 918
 919 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
 920 {
 921         struct device *cpu_dev;
 922
 923         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
 924
 925         if (!policy)
 926                 return 0;
 927
 928         cpu_dev = get_cpu_device(cpu);
 929         if (WARN_ON(!cpu_dev))
 930                 return 0;
 931
 932         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
 933 }
 934
 935 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
 936 {
 937         struct device *cpu_dev;
 938
 939         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
 940
 941         cpu_dev = get_cpu_device(cpu);
 942         if (WARN_ON(!cpu_dev))
 943                 return;
 944
 945         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
 946 }
 947
 948 /* Add/remove symlinks for all related CPUs */
 949 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
 950 {
 951         unsigned int j;
 952         int ret = 0;
 953
 954         /* Some related CPUs might not be present (physically hotplugged) */
 955         for_each_cpu(j, policy->real_cpus) {
 956                 ret = add_cpu_dev_symlink(policy, j);
 957                 if (ret)
 958                         break;
 959         }
 960
 961         return ret;
 962 }
 963
 964 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
 965 {
 966         unsigned int j;
 967
 968         /* Some related CPUs might not be present (physically hotplugged) */
 969         for_each_cpu(j, policy->real_cpus)
 970                 remove_cpu_dev_symlink(policy, j);
 971 }
 972
 973 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
 974 {
 975         struct freq_attr **drv_attr;
 976         int ret = 0;
 977
 978         /* set up files for this cpu device */
 979         drv_attr = cpufreq_driver->attr;
 980         while (drv_attr && *drv_attr) {
 981                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 982                 if (ret)
 983                         return ret;
 984                 drv_attr++;
 985         }
 986         if (cpufreq_driver->get) {
 987                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 988                 if (ret)
 989                         return ret;
 990         }
 991
 992         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 993         if (ret)
 994                 return ret;
 995
 996         if (cpufreq_driver->bios_limit) {
 997                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
 998                 if (ret)
 999                         return ret;
1000         }
1001
1002         return cpufreq_add_dev_symlink(policy);
1003 }
1004
1005 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1006 {
1007         return NULL;
1008 }
1009
1010 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1011 {
1012         struct cpufreq_governor *gov = NULL;
1013         struct cpufreq_policy new_policy;
1014
1015         memcpy(&new_policy, policy, sizeof(*policy));
1016
1017         /* Update governor of new_policy to the governor used before hotplug */
1018         gov = find_governor(policy->last_governor);
1019         if (gov) {
1020                 pr_debug("Restoring governor %s for cpu %d\n",
1021                                 policy->governor->name, policy->cpu);
1022         } else {
1023                 gov = cpufreq_default_governor();
1024                 if (!gov)
1025                         return -ENODATA;
1026         }
1027
1028         new_policy.governor = gov;
1029
1030         /* Use the default policy if there is no last_policy. */
1031         if (cpufreq_driver->setpolicy) {
1032                 if (policy->last_policy)
1033                         new_policy.policy = policy->last_policy;
1034                 else
1035                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1036                                                NULL);
1037         }
1038         /* set default policy */
1039         return cpufreq_set_policy(policy, &new_policy);
1040 }
1041
1042 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1043 {
1044         int ret = 0;
1045
1046         /* Has this CPU been taken care of already? */
1047         if (cpumask_test_cpu(cpu, policy->cpus))
1048                 return 0;
1049
1050         down_write(&policy->rwsem);
1051         if (has_target())
1052                 cpufreq_stop_governor(policy);
1053
1054         cpumask_set_cpu(cpu, policy->cpus);
1055
1056         if (has_target()) {
1057                 ret = cpufreq_start_governor(policy);
1058                 if (ret)
1059                         pr_err("%s: Failed to start governor\n", __func__);
1060         }
1061         up_write(&policy->rwsem);
1062         return ret;
1063 }
1064
1065 static void handle_update(struct work_struct *work)
1066 {
1067         struct cpufreq_policy *policy =
1068                 container_of(work, struct cpufreq_policy, update);
1069         unsigned int cpu = policy->cpu;
1070         pr_debug("handle_update for cpu %u called\n", cpu);
1071         cpufreq_update_policy(cpu);
1072 }
1073
1074 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1075 {
1076         struct cpufreq_policy *policy;
1077         int ret;
1078
1079         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1080         if (!policy)
1081                 return NULL;
1082
1083         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1084                 goto err_free_policy;
1085
1086         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1087                 goto err_free_cpumask;
1088
1089         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1090                 goto err_free_rcpumask;
1091
1092         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1093                                    cpufreq_global_kobject, "policy%u", cpu);
1094         if (ret) {
1095                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1096                 goto err_free_real_cpus;
1097         }
1098
1099         INIT_LIST_HEAD(&policy->policy_list);
1100         init_rwsem(&policy->rwsem);
1101         spin_lock_init(&policy->transition_lock);
1102         init_waitqueue_head(&policy->transition_wait);
1103         init_completion(&policy->kobj_unregister);
1104         INIT_WORK(&policy->update, handle_update);
1105
1106         policy->cpu = cpu;
1107         return policy;
1108
1109 err_free_real_cpus:
1110         free_cpumask_var(policy->real_cpus);
1111 err_free_rcpumask:
1112         free_cpumask_var(policy->related_cpus);
1113 err_free_cpumask:
1114         free_cpumask_var(policy->cpus);
1115 err_free_policy:
1116         kfree(policy);
1117
1118         return NULL;
1119 }
1120
1121 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1122 {
1123         struct kobject *kobj;
1124         struct completion *cmp;
1125
1126         if (notify)
1127                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1128                                              CPUFREQ_REMOVE_POLICY, policy);
1129
1130         down_write(&policy->rwsem);
1131         cpufreq_stats_free_table(policy);
1132         cpufreq_remove_dev_symlink(policy);
1133         kobj = &policy->kobj;
1134         cmp = &policy->kobj_unregister;
1135         up_write(&policy->rwsem);
1136         kobject_put(kobj);
1137
1138         /*
1139          * We need to make sure that the underlying kobj is
1140          * actually not referenced anymore by anybody before we
1141          * proceed with unloading.
1142          */
1143         pr_debug("waiting for dropping of refcount\n");
1144         wait_for_completion(cmp);
1145         pr_debug("wait complete\n");
1146 }
1147
1148 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1149 {
1150         unsigned long flags;
1151         int cpu;
1152
1153         /* Remove policy from list */
1154         write_lock_irqsave(&cpufreq_driver_lock, flags);
1155         list_del(&policy->policy_list);
1156
1157         for_each_cpu(cpu, policy->related_cpus)
1158                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1159         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1160
1161         cpufreq_policy_put_kobj(policy, notify);
1162         free_cpumask_var(policy->real_cpus);
1163         free_cpumask_var(policy->related_cpus);
1164         free_cpumask_var(policy->cpus);
1165         kfree(policy);
1166 }
1167
1168 static int cpufreq_online(unsigned int cpu)
1169 {
1170         struct cpufreq_policy *policy;
1171         bool new_policy;
1172         unsigned long flags;
1173         unsigned int j;
1174         int ret;
1175
1176         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1177
1178         /* Check if this CPU already has a policy to manage it */
1179         policy = per_cpu(cpufreq_cpu_data, cpu);
1180         if (policy) {
1181                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1182                 if (!policy_is_inactive(policy))
1183                         return cpufreq_add_policy_cpu(policy, cpu);
1184
1185                 /* This is the only online CPU for the policy.  Start over. */
1186                 new_policy = false;
1187                 down_write(&policy->rwsem);
1188                 policy->cpu = cpu;
1189                 policy->governor = NULL;
1190                 up_write(&policy->rwsem);
1191         } else {
1192                 new_policy = true;
1193                 policy = cpufreq_policy_alloc(cpu);
1194                 if (!policy)
1195                         return -ENOMEM;
1196         }
1197
1198         cpumask_copy(policy->cpus, cpumask_of(cpu));
1199
1200         /* call driver. From then on the cpufreq must be able
1201          * to accept all calls to ->verify and ->setpolicy for this CPU
1202          */
1203         ret = cpufreq_driver->init(policy);
1204         if (ret) {
1205                 pr_debug("initialization failed\n");
1206                 goto out_free_policy;
1207         }
1208
1209         down_write(&policy->rwsem);
1210
1211         if (new_policy) {
1212                 /* related_cpus should at least include policy->cpus. */
1213                 cpumask_copy(policy->related_cpus, policy->cpus);
1214                 /* Remember CPUs present at the policy creation time. */
1215                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1216         }
1217
1218         /*
1219          * affected cpus must always be the one, which are online. We aren't
1220          * managing offline cpus here.
1221          */
1222         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1223
1224         if (new_policy) {
1225                 policy->user_policy.min = policy->min;
1226                 policy->user_policy.max = policy->max;
1227
1228                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1229                 for_each_cpu(j, policy->related_cpus)
1230                         per_cpu(cpufreq_cpu_data, j) = policy;
1231                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1232         }
1233
1234         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1235                 policy->cur = cpufreq_driver->get(policy->cpu);
1236                 if (!policy->cur) {
1237                         pr_err("%s: ->get() failed\n", __func__);
1238                         goto out_exit_policy;
1239                 }
1240         }
1241
1242         /*
1243          * Sometimes boot loaders set CPU frequency to a value outside of
1244          * frequency table present with cpufreq core. In such cases CPU might be
1245          * unstable if it has to run on that frequency for long duration of time
1246          * and so its better to set it to a frequency which is specified in
1247          * freq-table. This also makes cpufreq stats inconsistent as
1248          * cpufreq-stats would fail to register because current frequency of CPU
1249          * isn't found in freq-table.
1250          *
1251          * Because we don't want this change to effect boot process badly, we go
1252          * for the next freq which is >= policy->cur ('cur' must be set by now,
1253          * otherwise we will end up setting freq to lowest of the table as 'cur'
1254          * is initialized to zero).
1255          *
1256          * We are passing target-freq as "policy->cur - 1" otherwise
1257          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1258          * equal to target-freq.
1259          */
1260         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1261             && has_target()) {
1262                 /* Are we running at unknown frequency ? */
1263                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1264                 if (ret == -EINVAL) {
1265                         /* Warn user and fix it */
1266                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1267                                 __func__, policy->cpu, policy->cur);
1268                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1269                                 CPUFREQ_RELATION_L);
1270
1271                         /*
1272                          * Reaching here after boot in a few seconds may not
1273                          * mean that system will remain stable at "unknown"
1274                          * frequency for longer duration. Hence, a BUG_ON().
1275                          */
1276                         BUG_ON(ret);
1277                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1278                                 __func__, policy->cpu, policy->cur);
1279                 }
1280         }
1281
1282         if (new_policy) {
1283                 ret = cpufreq_add_dev_interface(policy);
1284                 if (ret)
1285                         goto out_exit_policy;
1286
1287                 cpufreq_stats_create_table(policy);
1288                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1289                                 CPUFREQ_CREATE_POLICY, policy);
1290
1291                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1292                 list_add(&policy->policy_list, &cpufreq_policy_list);
1293                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1294         }
1295
1296         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1297                                      CPUFREQ_START, policy);
1298
1299         ret = cpufreq_init_policy(policy);
1300         if (ret) {
1301                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1302                        __func__, cpu, ret);
1303                 /* cpufreq_policy_free() will notify based on this */
1304                 new_policy = false;
1305                 goto out_exit_policy;
1306         }
1307
1308         up_write(&policy->rwsem);
1309
1310         kobject_uevent(&policy->kobj, KOBJ_ADD);
1311
1312         /* Callback for handling stuff after policy is ready */
1313         if (cpufreq_driver->ready)
1314                 cpufreq_driver->ready(policy);
1315
1316         pr_debug("initialization complete\n");
1317
1318         return 0;
1319
1320 out_exit_policy:
1321         up_write(&policy->rwsem);
1322
1323         if (cpufreq_driver->exit)
1324                 cpufreq_driver->exit(policy);
1325 out_free_policy:
1326         cpufreq_policy_free(policy, !new_policy);
1327         return ret;
1328 }
1329
1330 /**
1331  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1332  * @dev: CPU device.
1333  * @sif: Subsystem interface structure pointer (not used)
1334  */
1335 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1336 {
1337         struct cpufreq_policy *policy;
1338         unsigned cpu = dev->id;
1339
1340         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1341
1342         if (cpu_online(cpu))
1343                 return cpufreq_online(cpu);
1344
1345         /*
1346          * A hotplug notifier will follow and we will handle it as CPU online
1347          * then.  For now, just create the sysfs link, unless there is no policy
1348          * or the link is already present.
1349          */
1350         policy = per_cpu(cpufreq_cpu_data, cpu);
1351         if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1352                 return 0;
1353
1354         return add_cpu_dev_symlink(policy, cpu);
1355 }
1356
1357 static void cpufreq_offline(unsigned int cpu)
1358 {
1359         struct cpufreq_policy *policy;
1360         int ret;
1361
1362         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1363
1364         policy = cpufreq_cpu_get_raw(cpu);
1365         if (!policy) {
1366                 pr_debug("%s: No cpu_data found\n", __func__);
1367                 return;
1368         }
1369
1370         down_write(&policy->rwsem);
1371         if (has_target())
1372                 cpufreq_stop_governor(policy);
1373
1374         cpumask_clear_cpu(cpu, policy->cpus);
1375
1376         if (policy_is_inactive(policy)) {
1377                 if (has_target())
1378                         strncpy(policy->last_governor, policy->governor->name,
1379                                 CPUFREQ_NAME_LEN);
1380                 else
1381                         policy->last_policy = policy->policy;
1382         } else if (cpu == policy->cpu) {
1383                 /* Nominate new CPU */
1384                 policy->cpu = cpumask_any(policy->cpus);
1385         }
1386
1387         /* Start governor again for active policy */
1388         if (!policy_is_inactive(policy)) {
1389                 if (has_target()) {
1390                         ret = cpufreq_start_governor(policy);
1391                         if (ret)
1392                                 pr_err("%s: Failed to start governor\n", __func__);
1393                 }
1394
1395                 goto unlock;
1396         }
1397
1398         if (cpufreq_driver->stop_cpu)
1399                 cpufreq_driver->stop_cpu(policy);
1400
1401         if (has_target())
1402                 cpufreq_exit_governor(policy);
1403
1404         /*
1405          * Perform the ->exit() even during light-weight tear-down,
1406          * since this is a core component, and is essential for the
1407          * subsequent light-weight ->init() to succeed.
1408          */
1409         if (cpufreq_driver->exit) {
1410                 cpufreq_driver->exit(policy);
1411                 policy->freq_table = NULL;
1412         }
1413
1414 unlock:
1415         up_write(&policy->rwsem);
1416 }
1417
1418 /**
1419  * cpufreq_remove_dev - remove a CPU device
1420  *
1421  * Removes the cpufreq interface for a CPU device.
1422  */
1423 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1424 {
1425         unsigned int cpu = dev->id;
1426         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1427
1428         if (!policy)
1429                 return;
1430
1431         if (cpu_online(cpu))
1432                 cpufreq_offline(cpu);
1433
1434         cpumask_clear_cpu(cpu, policy->real_cpus);
1435         remove_cpu_dev_symlink(policy, cpu);
1436
1437         if (cpumask_empty(policy->real_cpus))
1438                 cpufreq_policy_free(policy, true);
1439 }
1440
1441 /**
1442  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1443  *      in deep trouble.
1444  *      @policy: policy managing CPUs
1445  *      @new_freq: CPU frequency the CPU actually runs at
1446  *
1447  *      We adjust to current frequency first, and need to clean up later.
1448  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1449  */
1450 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1451                                 unsigned int new_freq)
1452 {
1453         struct cpufreq_freqs freqs;
1454
1455         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1456                  policy->cur, new_freq);
1457
1458         freqs.old = policy->cur;
1459         freqs.new = new_freq;
1460
1461         cpufreq_freq_transition_begin(policy, &freqs);
1462         cpufreq_freq_transition_end(policy, &freqs, 0);
1463 }
1464
1465 /**
1466  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1467  * @cpu: CPU number
1468  *
1469  * This is the last known freq, without actually getting it from the driver.
1470  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1471  */
1472 unsigned int cpufreq_quick_get(unsigned int cpu)
1473 {
1474         struct cpufreq_policy *policy;
1475         unsigned int ret_freq = 0;
1476         unsigned long flags;
1477
1478         read_lock_irqsave(&cpufreq_driver_lock, flags);
1479
1480         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1481                 ret_freq = cpufreq_driver->get(cpu);
1482                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1483                 return ret_freq;
1484         }
1485
1486         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1487
1488         policy = cpufreq_cpu_get(cpu);
1489         if (policy) {
1490                 ret_freq = policy->cur;
1491                 cpufreq_cpu_put(policy);
1492         }
1493
1494         return ret_freq;
1495 }
1496 EXPORT_SYMBOL(cpufreq_quick_get);
1497
1498 /**
1499  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1500  * @cpu: CPU number
1501  *
1502  * Just return the max possible frequency for a given CPU.
1503  */
1504 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1505 {
1506         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1507         unsigned int ret_freq = 0;
1508
1509         if (policy) {
1510                 ret_freq = policy->max;
1511                 cpufreq_cpu_put(policy);
1512         }
1513
1514         return ret_freq;
1515 }
1516 EXPORT_SYMBOL(cpufreq_quick_get_max);
1517
1518 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1519 {
1520         unsigned int ret_freq = 0;
1521
1522         if (!cpufreq_driver->get)
1523                 return ret_freq;
1524
1525         ret_freq = cpufreq_driver->get(policy->cpu);
1526
1527         /*
1528          * Updating inactive policies is invalid, so avoid doing that.  Also
1529          * if fast frequency switching is used with the given policy, the check
1530          * against policy->cur is pointless, so skip it in that case too.
1531          */
1532         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1533                 return ret_freq;
1534
1535         if (ret_freq && policy->cur &&
1536                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1537                 /* verify no discrepancy between actual and
1538                                         saved value exists */
1539                 if (unlikely(ret_freq != policy->cur)) {
1540                         cpufreq_out_of_sync(policy, ret_freq);
1541                         schedule_work(&policy->update);
1542                 }
1543         }
1544
1545         return ret_freq;
1546 }
1547
1548 /**
1549  * cpufreq_get - get the current CPU frequency (in kHz)
1550  * @cpu: CPU number
1551  *
1552  * Get the CPU current (static) CPU frequency
1553  */
1554 unsigned int cpufreq_get(unsigned int cpu)
1555 {
1556         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1557         unsigned int ret_freq = 0;
1558
1559         if (policy) {
1560                 down_read(&policy->rwsem);
1561                 ret_freq = __cpufreq_get(policy);
1562                 up_read(&policy->rwsem);
1563
1564                 cpufreq_cpu_put(policy);
1565         }
1566
1567         return ret_freq;
1568 }
1569 EXPORT_SYMBOL(cpufreq_get);
1570
1571 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1572 {
1573         unsigned int new_freq;
1574
1575         new_freq = cpufreq_driver->get(policy->cpu);
1576         if (!new_freq)
1577                 return 0;
1578
1579         if (!policy->cur) {
1580                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1581                 policy->cur = new_freq;
1582         } else if (policy->cur != new_freq && has_target()) {
1583                 cpufreq_out_of_sync(policy, new_freq);
1584         }
1585
1586         return new_freq;
1587 }
1588
1589 static struct subsys_interface cpufreq_interface = {
1590         .name           = "cpufreq",
1591         .subsys         = &cpu_subsys,
1592         .add_dev        = cpufreq_add_dev,
1593         .remove_dev     = cpufreq_remove_dev,
1594 };
1595
1596 /*
1597  * In case platform wants some specific frequency to be configured
1598  * during suspend..
1599  */
1600 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1601 {
1602         int ret;
1603
1604         if (!policy->suspend_freq) {
1605                 pr_debug("%s: suspend_freq not defined\n", __func__);
1606                 return 0;
1607         }
1608
1609         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1610                         policy->suspend_freq);
1611
1612         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1613                         CPUFREQ_RELATION_H);
1614         if (ret)
1615                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1616                                 __func__, policy->suspend_freq, ret);
1617
1618         return ret;
1619 }
1620 EXPORT_SYMBOL(cpufreq_generic_suspend);
1621
1622 /**
1623  * cpufreq_suspend() - Suspend CPUFreq governors
1624  *
1625  * Called during system wide Suspend/Hibernate cycles for suspending governors
1626  * as some platforms can't change frequency after this point in suspend cycle.
1627  * Because some of the devices (like: i2c, regulators, etc) they use for
1628  * changing frequency are suspended quickly after this point.
1629  */
1630 void cpufreq_suspend(void)
1631 {
1632         struct cpufreq_policy *policy;
1633
1634         if (!cpufreq_driver)
1635                 return;
1636
1637         if (!has_target() && !cpufreq_driver->suspend)
1638                 goto suspend;
1639
1640         pr_debug("%s: Suspending Governors\n", __func__);
1641
1642         for_each_active_policy(policy) {
1643                 if (has_target()) {
1644                         down_write(&policy->rwsem);
1645                         cpufreq_stop_governor(policy);
1646                         up_write(&policy->rwsem);
1647                 }
1648
1649                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1650                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1651                                 policy);
1652         }
1653
1654 suspend:
1655         cpufreq_suspended = true;
1656 }
1657
1658 /**
1659  * cpufreq_resume() - Resume CPUFreq governors
1660  *
1661  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1662  * are suspended with cpufreq_suspend().
1663  */
1664 void cpufreq_resume(void)
1665 {
1666         struct cpufreq_policy *policy;
1667         int ret;
1668
1669         if (!cpufreq_driver)
1670                 return;
1671
1672         cpufreq_suspended = false;
1673
1674         if (!has_target() && !cpufreq_driver->resume)
1675                 return;
1676
1677         pr_debug("%s: Resuming Governors\n", __func__);
1678
1679         for_each_active_policy(policy) {
1680                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1681                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1682                                 policy);
1683                 } else if (has_target()) {
1684                         down_write(&policy->rwsem);
1685                         ret = cpufreq_start_governor(policy);
1686                         up_write(&policy->rwsem);
1687
1688                         if (ret)
1689                                 pr_err("%s: Failed to start governor for policy: %p\n",
1690                                        __func__, policy);
1691                 }
1692         }
1693 }
1694
1695 /**
1696  *      cpufreq_get_current_driver - return current driver's name
1697  *
1698  *      Return the name string of the currently loaded cpufreq driver
1699  *      or NULL, if none.
1700  */
1701 const char *cpufreq_get_current_driver(void)
1702 {
1703         if (cpufreq_driver)
1704                 return cpufreq_driver->name;
1705
1706         return NULL;
1707 }
1708 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1709
1710 /**
1711  *      cpufreq_get_driver_data - return current driver data
1712  *
1713  *      Return the private data of the currently loaded cpufreq
1714  *      driver, or NULL if no cpufreq driver is loaded.
1715  */
1716 void *cpufreq_get_driver_data(void)
1717 {
1718         if (cpufreq_driver)
1719                 return cpufreq_driver->driver_data;
1720
1721         return NULL;
1722 }
1723 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1724
1725 /*********************************************************************
1726  *                     NOTIFIER LISTS INTERFACE                      *
1727  *********************************************************************/
1728
1729 /**
1730  *      cpufreq_register_notifier - register a driver with cpufreq
1731  *      @nb: notifier function to register
1732  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1733  *
1734  *      Add a driver to one of two lists: either a list of drivers that
1735  *      are notified about clock rate changes (once before and once after
1736  *      the transition), or a list of drivers that are notified about
1737  *      changes in cpufreq policy.
1738  *
1739  *      This function may sleep, and has the same return conditions as
1740  *      blocking_notifier_chain_register.
1741  */
1742 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1743 {
1744         int ret;
1745
1746         if (cpufreq_disabled())
1747                 return -EINVAL;
1748
1749         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1750
1751         switch (list) {
1752         case CPUFREQ_TRANSITION_NOTIFIER:
1753                 mutex_lock(&cpufreq_fast_switch_lock);
1754
1755                 if (cpufreq_fast_switch_count > 0) {
1756                         mutex_unlock(&cpufreq_fast_switch_lock);
1757                         return -EBUSY;
1758                 }
1759                 ret = srcu_notifier_chain_register(
1760                                 &cpufreq_transition_notifier_list, nb);
1761                 if (!ret)
1762                         cpufreq_fast_switch_count--;
1763
1764                 mutex_unlock(&cpufreq_fast_switch_lock);
1765                 break;
1766         case CPUFREQ_POLICY_NOTIFIER:
1767                 ret = blocking_notifier_chain_register(
1768                                 &cpufreq_policy_notifier_list, nb);
1769                 break;
1770         default:
1771                 ret = -EINVAL;
1772         }
1773
1774         return ret;
1775 }
1776 EXPORT_SYMBOL(cpufreq_register_notifier);
1777
1778 /**
1779  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1780  *      @nb: notifier block to be unregistered
1781  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1782  *
1783  *      Remove a driver from the CPU frequency notifier list.
1784  *
1785  *      This function may sleep, and has the same return conditions as
1786  *      blocking_notifier_chain_unregister.
1787  */
1788 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1789 {
1790         int ret;
1791
1792         if (cpufreq_disabled())
1793                 return -EINVAL;
1794
1795         switch (list) {
1796         case CPUFREQ_TRANSITION_NOTIFIER:
1797                 mutex_lock(&cpufreq_fast_switch_lock);
1798
1799                 ret = srcu_notifier_chain_unregister(
1800                                 &cpufreq_transition_notifier_list, nb);
1801                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1802                         cpufreq_fast_switch_count++;
1803
1804                 mutex_unlock(&cpufreq_fast_switch_lock);
1805                 break;
1806         case CPUFREQ_POLICY_NOTIFIER:
1807                 ret = blocking_notifier_chain_unregister(
1808                                 &cpufreq_policy_notifier_list, nb);
1809                 break;
1810         default:
1811                 ret = -EINVAL;
1812         }
1813
1814         return ret;
1815 }
1816 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1817
1818
1819 /*********************************************************************
1820  *                              GOVERNORS                            *
1821  *********************************************************************/
1822
1823 /**
1824  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1825  * @policy: cpufreq policy to switch the frequency for.
1826  * @target_freq: New frequency to set (may be approximate).
1827  *
1828  * Carry out a fast frequency switch without sleeping.
1829  *
1830  * The driver's ->fast_switch() callback invoked by this function must be
1831  * suitable for being called from within RCU-sched read-side critical sections
1832  * and it is expected to select the minimum available frequency greater than or
1833  * equal to @target_freq (CPUFREQ_RELATION_L).
1834  *
1835  * This function must not be called if policy->fast_switch_enabled is unset.
1836  *
1837  * Governors calling this function must guarantee that it will never be invoked
1838  * twice in parallel for the same policy and that it will never be called in
1839  * parallel with either ->target() or ->target_index() for the same policy.
1840  *
1841  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1842  * callback to indicate an error condition, the hardware configuration must be
1843  * preserved.
1844  */
1845 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1846                                         unsigned int target_freq)
1847 {
1848         target_freq = clamp_val(target_freq, policy->min, policy->max);
1849
1850         return cpufreq_driver->fast_switch(policy, target_freq);
1851 }
1852 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1853
1854 /* Must set freqs->new to intermediate frequency */
1855 static int __target_intermediate(struct cpufreq_policy *policy,
1856                                  struct cpufreq_freqs *freqs, int index)
1857 {
1858         int ret;
1859
1860         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1861
1862         /* We don't need to switch to intermediate freq */
1863         if (!freqs->new)
1864                 return 0;
1865
1866         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1867                  __func__, policy->cpu, freqs->old, freqs->new);
1868
1869         cpufreq_freq_transition_begin(policy, freqs);
1870         ret = cpufreq_driver->target_intermediate(policy, index);
1871         cpufreq_freq_transition_end(policy, freqs, ret);
1872
1873         if (ret)
1874                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1875                        __func__, ret);
1876
1877         return ret;
1878 }
1879
1880 static int __target_index(struct cpufreq_policy *policy, int index)
1881 {
1882         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1883         unsigned int intermediate_freq = 0;
1884         unsigned int newfreq = policy->freq_table[index].frequency;
1885         int retval = -EINVAL;
1886         bool notify;
1887
1888         if (newfreq == policy->cur)
1889                 return 0;
1890
1891         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1892         if (notify) {
1893                 /* Handle switching to intermediate frequency */
1894                 if (cpufreq_driver->get_intermediate) {
1895                         retval = __target_intermediate(policy, &freqs, index);
1896                         if (retval)
1897                                 return retval;
1898
1899                         intermediate_freq = freqs.new;
1900                         /* Set old freq to intermediate */
1901                         if (intermediate_freq)
1902                                 freqs.old = freqs.new;
1903                 }
1904
1905                 freqs.new = newfreq;
1906                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1907                          __func__, policy->cpu, freqs.old, freqs.new);
1908
1909                 cpufreq_freq_transition_begin(policy, &freqs);
1910         }
1911
1912         retval = cpufreq_driver->target_index(policy, index);
1913         if (retval)
1914                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1915                        retval);
1916
1917         if (notify) {
1918                 cpufreq_freq_transition_end(policy, &freqs, retval);
1919
1920                 /*
1921                  * Failed after setting to intermediate freq? Driver should have
1922                  * reverted back to initial frequency and so should we. Check
1923                  * here for intermediate_freq instead of get_intermediate, in
1924                  * case we haven't switched to intermediate freq at all.
1925                  */
1926                 if (unlikely(retval && intermediate_freq)) {
1927                         freqs.old = intermediate_freq;
1928                         freqs.new = policy->restore_freq;
1929                         cpufreq_freq_transition_begin(policy, &freqs);
1930                         cpufreq_freq_transition_end(policy, &freqs, 0);
1931                 }
1932         }
1933
1934         return retval;
1935 }
1936
1937 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1938                             unsigned int target_freq,
1939                             unsigned int relation)
1940 {
1941         unsigned int old_target_freq = target_freq;
1942         int index;
1943
1944         if (cpufreq_disabled())
1945                 return -ENODEV;
1946
1947         /* Make sure that target_freq is within supported range */
1948         target_freq = clamp_val(target_freq, policy->min, policy->max);
1949
1950         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1951                  policy->cpu, target_freq, relation, old_target_freq);
1952
1953         /*
1954          * This might look like a redundant call as we are checking it again
1955          * after finding index. But it is left intentionally for cases where
1956          * exactly same freq is called again and so we can save on few function
1957          * calls.
1958          */
1959         if (target_freq == policy->cur)
1960                 return 0;
1961
1962         /* Save last value to restore later on errors */
1963         policy->restore_freq = policy->cur;
1964
1965         if (cpufreq_driver->target)
1966                 return cpufreq_driver->target(policy, target_freq, relation);
1967
1968         if (!cpufreq_driver->target_index)
1969                 return -EINVAL;
1970
1971         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1972
1973         return __target_index(policy, index);
1974 }
1975 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1976
1977 int cpufreq_driver_target(struct cpufreq_policy *policy,
1978                           unsigned int target_freq,
1979                           unsigned int relation)
1980 {
1981         int ret = -EINVAL;
1982
1983         down_write(&policy->rwsem);
1984
1985         ret = __cpufreq_driver_target(policy, target_freq, relation);
1986
1987         up_write(&policy->rwsem);
1988
1989         return ret;
1990 }
1991 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1992
1993 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1994 {
1995         return NULL;
1996 }
1997
1998 static int cpufreq_init_governor(struct cpufreq_policy *policy)
1999 {
2000         int ret;
2001
2002         /* Don't start any governor operations if we are entering suspend */
2003         if (cpufreq_suspended)
2004                 return 0;
2005         /*
2006          * Governor might not be initiated here if ACPI _PPC changed
2007          * notification happened, so check it.
2008          */
2009         if (!policy->governor)
2010                 return -EINVAL;
2011
2012         if (policy->governor->max_transition_latency &&
2013             policy->cpuinfo.transition_latency >
2014             policy->governor->max_transition_latency) {
2015                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2016
2017                 if (gov) {
2018                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2019                                 policy->governor->name, gov->name);
2020                         policy->governor = gov;
2021                 } else {
2022                         return -EINVAL;
2023                 }
2024         }
2025
2026         if (!try_module_get(policy->governor->owner))
2027                 return -EINVAL;
2028
2029         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2030
2031         if (policy->governor->init) {
2032                 ret = policy->governor->init(policy);
2033                 if (ret) {
2034                         module_put(policy->governor->owner);
2035                         return ret;
2036                 }
2037         }
2038
2039         return 0;
2040 }
2041
2042 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2043 {
2044         if (cpufreq_suspended || !policy->governor)
2045                 return;
2046
2047         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2048
2049         if (policy->governor->exit)
2050                 policy->governor->exit(policy);
2051
2052         module_put(policy->governor->owner);
2053 }
2054
2055 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2056 {
2057         int ret;
2058
2059         if (cpufreq_suspended)
2060                 return 0;
2061
2062         if (!policy->governor)
2063                 return -EINVAL;
2064
2065         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2066
2067         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2068                 cpufreq_update_current_freq(policy);
2069
2070         if (policy->governor->start) {
2071                 ret = policy->governor->start(policy);
2072                 if (ret)
2073                         return ret;
2074         }
2075
2076         if (policy->governor->limits)
2077                 policy->governor->limits(policy);
2078
2079         return 0;
2080 }
2081
2082 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2083 {
2084         if (cpufreq_suspended || !policy->governor)
2085                 return;
2086
2087         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2088
2089         if (policy->governor->stop)
2090                 policy->governor->stop(policy);
2091 }
2092
2093 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2094 {
2095         if (cpufreq_suspended || !policy->governor)
2096                 return;
2097
2098         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2099
2100         if (policy->governor->limits)
2101                 policy->governor->limits(policy);
2102 }
2103
2104 int cpufreq_register_governor(struct cpufreq_governor *governor)
2105 {
2106         int err;
2107
2108         if (!governor)
2109                 return -EINVAL;
2110
2111         if (cpufreq_disabled())
2112                 return -ENODEV;
2113
2114         mutex_lock(&cpufreq_governor_mutex);
2115
2116         err = -EBUSY;
2117         if (!find_governor(governor->name)) {
2118                 err = 0;
2119                 list_add(&governor->governor_list, &cpufreq_governor_list);
2120         }
2121
2122         mutex_unlock(&cpufreq_governor_mutex);
2123         return err;
2124 }
2125 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2126
2127 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2128 {
2129         struct cpufreq_policy *policy;
2130         unsigned long flags;
2131
2132         if (!governor)
2133                 return;
2134
2135         if (cpufreq_disabled())
2136                 return;
2137
2138         /* clear last_governor for all inactive policies */
2139         read_lock_irqsave(&cpufreq_driver_lock, flags);
2140         for_each_inactive_policy(policy) {
2141                 if (!strcmp(policy->last_governor, governor->name)) {
2142                         policy->governor = NULL;
2143                         strcpy(policy->last_governor, "\0");
2144                 }
2145         }
2146         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2147
2148         mutex_lock(&cpufreq_governor_mutex);
2149         list_del(&governor->governor_list);
2150         mutex_unlock(&cpufreq_governor_mutex);
2151         return;
2152 }
2153 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2154
2155
2156 /*********************************************************************
2157  *                          POLICY INTERFACE                         *
2158  *********************************************************************/
2159
2160 /**
2161  * cpufreq_get_policy - get the current cpufreq_policy
2162  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2163  *      is written
2164  *
2165  * Reads the current cpufreq policy.
2166  */
2167 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2168 {
2169         struct cpufreq_policy *cpu_policy;
2170         if (!policy)
2171                 return -EINVAL;
2172
2173         cpu_policy = cpufreq_cpu_get(cpu);
2174         if (!cpu_policy)
2175                 return -EINVAL;
2176
2177         memcpy(policy, cpu_policy, sizeof(*policy));
2178
2179         cpufreq_cpu_put(cpu_policy);
2180         return 0;
2181 }
2182 EXPORT_SYMBOL(cpufreq_get_policy);
2183
2184 /*
2185  * policy : current policy.
2186  * new_policy: policy to be set.
2187  */
2188 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2189                                 struct cpufreq_policy *new_policy)
2190 {
2191         struct cpufreq_governor *old_gov;
2192         int ret;
2193
2194         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2195                  new_policy->cpu, new_policy->min, new_policy->max);
2196
2197         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2198
2199         /*
2200         * This check works well when we store new min/max freq attributes,
2201         * because new_policy is a copy of policy with one field updated.
2202         */
2203         if (new_policy->min > new_policy->max)
2204                 return -EINVAL;
2205
2206         /* verify the cpu speed can be set within this limit */
2207         ret = cpufreq_driver->verify(new_policy);
2208         if (ret)
2209                 return ret;
2210
2211         /* adjust if necessary - all reasons */
2212         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2213                         CPUFREQ_ADJUST, new_policy);
2214
2215         /*
2216          * verify the cpu speed can be set within this limit, which might be
2217          * different to the first one
2218          */
2219         ret = cpufreq_driver->verify(new_policy);
2220         if (ret)
2221                 return ret;
2222
2223         /* notification of the new policy */
2224         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2225                         CPUFREQ_NOTIFY, new_policy);
2226
2227         policy->min = new_policy->min;
2228         policy->max = new_policy->max;
2229
2230         policy->cached_target_freq = UINT_MAX;
2231
2232         pr_debug("new min and max freqs are %u - %u kHz\n",
2233                  policy->min, policy->max);
2234
2235         if (cpufreq_driver->setpolicy) {
2236                 policy->policy = new_policy->policy;
2237                 pr_debug("setting range\n");
2238                 return cpufreq_driver->setpolicy(new_policy);
2239         }
2240
2241         if (new_policy->governor == policy->governor) {
2242                 pr_debug("cpufreq: governor limits update\n");
2243                 cpufreq_governor_limits(policy);
2244                 return 0;
2245         }
2246
2247         pr_debug("governor switch\n");
2248
2249         /* save old, working values */
2250         old_gov = policy->governor;
2251         /* end old governor */
2252         if (old_gov) {
2253                 cpufreq_stop_governor(policy);
2254                 cpufreq_exit_governor(policy);
2255         }
2256
2257         /* start new governor */
2258         policy->governor = new_policy->governor;
2259         ret = cpufreq_init_governor(policy);
2260         if (!ret) {
2261                 ret = cpufreq_start_governor(policy);
2262                 if (!ret) {
2263                         pr_debug("cpufreq: governor change\n");
2264                         return 0;
2265                 }
2266                 cpufreq_exit_governor(policy);
2267         }
2268
2269         /* new governor failed, so re-start old one */
2270         pr_debug("starting governor %s failed\n", policy->governor->name);
2271         if (old_gov) {
2272                 policy->governor = old_gov;
2273                 if (cpufreq_init_governor(policy))
2274                         policy->governor = NULL;
2275                 else
2276                         cpufreq_start_governor(policy);
2277         }
2278
2279         return ret;
2280 }
2281
2282 /**
2283  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2284  *      @cpu: CPU which shall be re-evaluated
2285  *
2286  *      Useful for policy notifiers which have different necessities
2287  *      at different times.
2288  */
2289 int cpufreq_update_policy(unsigned int cpu)
2290 {
2291         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2292         struct cpufreq_policy new_policy;
2293         int ret;
2294
2295         if (!policy)
2296                 return -ENODEV;
2297
2298         down_write(&policy->rwsem);
2299
2300         pr_debug("updating policy for CPU %u\n", cpu);
2301         memcpy(&new_policy, policy, sizeof(*policy));
2302         new_policy.min = policy->user_policy.min;
2303         new_policy.max = policy->user_policy.max;
2304
2305         /*
2306          * BIOS might change freq behind our back
2307          * -> ask driver for current freq and notify governors about a change
2308          */
2309         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2310                 if (cpufreq_suspended) {
2311                         ret = -EAGAIN;
2312                         goto unlock;
2313                 }
2314                 new_policy.cur = cpufreq_update_current_freq(policy);
2315                 if (WARN_ON(!new_policy.cur)) {
2316                         ret = -EIO;
2317                         goto unlock;
2318                 }
2319         }
2320
2321         ret = cpufreq_set_policy(policy, &new_policy);
2322
2323 unlock:
2324         up_write(&policy->rwsem);
2325
2326         cpufreq_cpu_put(policy);
2327         return ret;
2328 }
2329 EXPORT_SYMBOL(cpufreq_update_policy);
2330
2331 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2332                                         unsigned long action, void *hcpu)
2333 {
2334         unsigned int cpu = (unsigned long)hcpu;
2335
2336         switch (action & ~CPU_TASKS_FROZEN) {
2337         case CPU_ONLINE:
2338         case CPU_DOWN_FAILED:
2339                 cpufreq_online(cpu);
2340                 break;
2341
2342         case CPU_DOWN_PREPARE:
2343                 cpufreq_offline(cpu);
2344                 break;
2345         }
2346         return NOTIFY_OK;
2347 }
2348
2349 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2350         .notifier_call = cpufreq_cpu_callback,
2351 };
2352
2353 /*********************************************************************
2354  *               BOOST                                               *
2355  *********************************************************************/
2356 static int cpufreq_boost_set_sw(int state)
2357 {
2358         struct cpufreq_policy *policy;
2359         int ret = -EINVAL;
2360
2361         for_each_active_policy(policy) {
2362                 if (!policy->freq_table)
2363                         continue;
2364
2365                 ret = cpufreq_frequency_table_cpuinfo(policy,
2366                                                       policy->freq_table);
2367                 if (ret) {
2368                         pr_err("%s: Policy frequency update failed\n",
2369                                __func__);
2370                         break;
2371                 }
2372
2373                 down_write(&policy->rwsem);
2374                 policy->user_policy.max = policy->max;
2375                 cpufreq_governor_limits(policy);
2376                 up_write(&policy->rwsem);
2377         }
2378
2379         return ret;
2380 }
2381
2382 int cpufreq_boost_trigger_state(int state)
2383 {
2384         unsigned long flags;
2385         int ret = 0;
2386
2387         if (cpufreq_driver->boost_enabled == state)
2388                 return 0;
2389
2390         write_lock_irqsave(&cpufreq_driver_lock, flags);
2391         cpufreq_driver->boost_enabled = state;
2392         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2393
2394         ret = cpufreq_driver->set_boost(state);
2395         if (ret) {
2396                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2397                 cpufreq_driver->boost_enabled = !state;
2398                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2399
2400                 pr_err("%s: Cannot %s BOOST\n",
2401                        __func__, state ? "enable" : "disable");
2402         }
2403
2404         return ret;
2405 }
2406
2407 static bool cpufreq_boost_supported(void)
2408 {
2409         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2410 }
2411
2412 static int create_boost_sysfs_file(void)
2413 {
2414         int ret;
2415
2416         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2417         if (ret)
2418                 pr_err("%s: cannot register global BOOST sysfs file\n",
2419                        __func__);
2420
2421         return ret;
2422 }
2423
2424 static void remove_boost_sysfs_file(void)
2425 {
2426         if (cpufreq_boost_supported())
2427                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2428 }
2429
2430 int cpufreq_enable_boost_support(void)
2431 {
2432         if (!cpufreq_driver)
2433                 return -EINVAL;
2434
2435         if (cpufreq_boost_supported())
2436                 return 0;
2437
2438         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2439
2440         /* This will get removed on driver unregister */
2441         return create_boost_sysfs_file();
2442 }
2443 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2444
2445 int cpufreq_boost_enabled(void)
2446 {
2447         return cpufreq_driver->boost_enabled;
2448 }
2449 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2450
2451 /*********************************************************************
2452  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2453  *********************************************************************/
2454
2455 /**
2456  * cpufreq_register_driver - register a CPU Frequency driver
2457  * @driver_data: A struct cpufreq_driver containing the values#
2458  * submitted by the CPU Frequency driver.
2459  *
2460  * Registers a CPU Frequency driver to this core code. This code
2461  * returns zero on success, -EEXIST when another driver got here first
2462  * (and isn't unregistered in the meantime).
2463  *
2464  */
2465 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2466 {
2467         unsigned long flags;
2468         int ret;
2469
2470         if (cpufreq_disabled())
2471                 return -ENODEV;
2472
2473         if (!driver_data || !driver_data->verify || !driver_data->init ||
2474             !(driver_data->setpolicy || driver_data->target_index ||
2475                     driver_data->target) ||
2476              (driver_data->setpolicy && (driver_data->target_index ||
2477                     driver_data->target)) ||
2478              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2479                 return -EINVAL;
2480
2481         pr_debug("trying to register driver %s\n", driver_data->name);
2482
2483         /* Protect against concurrent CPU online/offline. */
2484         get_online_cpus();
2485
2486         write_lock_irqsave(&cpufreq_driver_lock, flags);
2487         if (cpufreq_driver) {
2488                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2489                 ret = -EEXIST;
2490                 goto out;
2491         }
2492         cpufreq_driver = driver_data;
2493         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2494
2495         if (driver_data->setpolicy)
2496                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2497
2498         if (cpufreq_boost_supported()) {
2499                 ret = create_boost_sysfs_file();
2500                 if (ret)
2501                         goto err_null_driver;
2502         }
2503
2504         ret = subsys_interface_register(&cpufreq_interface);
2505         if (ret)
2506                 goto err_boost_unreg;
2507
2508         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2509             list_empty(&cpufreq_policy_list)) {
2510                 /* if all ->init() calls failed, unregister */
2511                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2512                          driver_data->name);
2513                 goto err_if_unreg;
2514         }
2515
2516         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2517         pr_debug("driver %s up and running\n", driver_data->name);
2518         goto out;
2519
2520 err_if_unreg:
2521         subsys_interface_unregister(&cpufreq_interface);
2522 err_boost_unreg:
2523         remove_boost_sysfs_file();
2524 err_null_driver:
2525         write_lock_irqsave(&cpufreq_driver_lock, flags);
2526         cpufreq_driver = NULL;
2527         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2528 out:
2529         put_online_cpus();
2530         return ret;
2531 }
2532 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2533
2534 /**
2535  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2536  *
2537  * Unregister the current CPUFreq driver. Only call this if you have
2538  * the right to do so, i.e. if you have succeeded in initialising before!
2539  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2540  * currently not initialised.
2541  */
2542 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2543 {
2544         unsigned long flags;
2545
2546         if (!cpufreq_driver || (driver != cpufreq_driver))
2547                 return -EINVAL;
2548
2549         pr_debug("unregistering driver %s\n", driver->name);
2550
2551         /* Protect against concurrent cpu hotplug */
2552         get_online_cpus();
2553         subsys_interface_unregister(&cpufreq_interface);
2554         remove_boost_sysfs_file();
2555         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2556
2557         write_lock_irqsave(&cpufreq_driver_lock, flags);
2558
2559         cpufreq_driver = NULL;
2560
2561         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2562         put_online_cpus();
2563
2564         return 0;
2565 }
2566 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2567
2568 /*
2569  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2570  * or mutexes when secondary CPUs are halted.
2571  */
2572 static struct syscore_ops cpufreq_syscore_ops = {
2573         .shutdown = cpufreq_suspend,
2574 };
2575
2576 struct kobject *cpufreq_global_kobject;
2577 EXPORT_SYMBOL(cpufreq_global_kobject);
2578
2579 static int __init cpufreq_core_init(void)
2580 {
2581         if (cpufreq_disabled())
2582                 return -ENODEV;
2583
2584         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2585         BUG_ON(!cpufreq_global_kobject);
2586
2587         register_syscore_ops(&cpufreq_syscore_ops);
2588
2589         return 0;
2590 }
2591 core_initcall(cpufreq_core_init);