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,
 920                                struct device *dev)
 921 {
 922         dev_dbg(dev, "%s: Adding symlink\n", __func__);
 923         return sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
 924 }
 925
 926 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
 927                                    struct device *dev)
 928 {
 929         dev_dbg(dev, "%s: Removing symlink\n", __func__);
 930         sysfs_remove_link(&dev->kobj, "cpufreq");
 931 }
 932
 933 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
 934 {
 935         struct freq_attr **drv_attr;
 936         int ret = 0;
 937
 938         /* set up files for this cpu device */
 939         drv_attr = cpufreq_driver->attr;
 940         while (drv_attr && *drv_attr) {
 941                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 942                 if (ret)
 943                         return ret;
 944                 drv_attr++;
 945         }
 946         if (cpufreq_driver->get) {
 947                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 948                 if (ret)
 949                         return ret;
 950         }
 951
 952         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 953         if (ret)
 954                 return ret;
 955
 956         if (cpufreq_driver->bios_limit) {
 957                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
 958                 if (ret)
 959                         return ret;
 960         }
 961
 962         return 0;
 963 }
 964
 965 __weak struct cpufreq_governor *cpufreq_default_governor(void)
 966 {
 967         return NULL;
 968 }
 969
 970 static int cpufreq_init_policy(struct cpufreq_policy *policy)
 971 {
 972         struct cpufreq_governor *gov = NULL;
 973         struct cpufreq_policy new_policy;
 974
 975         memcpy(&new_policy, policy, sizeof(*policy));
 976
 977         /* Update governor of new_policy to the governor used before hotplug */
 978         gov = find_governor(policy->last_governor);
 979         if (gov) {
 980                 pr_debug("Restoring governor %s for cpu %d\n",
 981                                 policy->governor->name, policy->cpu);
 982         } else {
 983                 gov = cpufreq_default_governor();
 984                 if (!gov)
 985                         return -ENODATA;
 986         }
 987
 988         new_policy.governor = gov;
 989
 990         /* Use the default policy if there is no last_policy. */
 991         if (cpufreq_driver->setpolicy) {
 992                 if (policy->last_policy)
 993                         new_policy.policy = policy->last_policy;
 994                 else
 995                         cpufreq_parse_governor(gov->name, &new_policy.policy,
 996                                                NULL);
 997         }
 998         /* set default policy */
 999         return cpufreq_set_policy(policy, &new_policy);
1000 }
1001
1002 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1003 {
1004         int ret = 0;
1005
1006         /* Has this CPU been taken care of already? */
1007         if (cpumask_test_cpu(cpu, policy->cpus))
1008                 return 0;
1009
1010         down_write(&policy->rwsem);
1011         if (has_target())
1012                 cpufreq_stop_governor(policy);
1013
1014         cpumask_set_cpu(cpu, policy->cpus);
1015
1016         if (has_target()) {
1017                 ret = cpufreq_start_governor(policy);
1018                 if (ret)
1019                         pr_err("%s: Failed to start governor\n", __func__);
1020         }
1021         up_write(&policy->rwsem);
1022         return ret;
1023 }
1024
1025 static void handle_update(struct work_struct *work)
1026 {
1027         struct cpufreq_policy *policy =
1028                 container_of(work, struct cpufreq_policy, update);
1029         unsigned int cpu = policy->cpu;
1030         pr_debug("handle_update for cpu %u called\n", cpu);
1031         cpufreq_update_policy(cpu);
1032 }
1033
1034 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1035 {
1036         struct cpufreq_policy *policy;
1037         int ret;
1038
1039         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1040         if (!policy)
1041                 return NULL;
1042
1043         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1044                 goto err_free_policy;
1045
1046         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1047                 goto err_free_cpumask;
1048
1049         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1050                 goto err_free_rcpumask;
1051
1052         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1053                                    cpufreq_global_kobject, "policy%u", cpu);
1054         if (ret) {
1055                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1056                 goto err_free_real_cpus;
1057         }
1058
1059         INIT_LIST_HEAD(&policy->policy_list);
1060         init_rwsem(&policy->rwsem);
1061         spin_lock_init(&policy->transition_lock);
1062         init_waitqueue_head(&policy->transition_wait);
1063         init_completion(&policy->kobj_unregister);
1064         INIT_WORK(&policy->update, handle_update);
1065
1066         policy->cpu = cpu;
1067         return policy;
1068
1069 err_free_real_cpus:
1070         free_cpumask_var(policy->real_cpus);
1071 err_free_rcpumask:
1072         free_cpumask_var(policy->related_cpus);
1073 err_free_cpumask:
1074         free_cpumask_var(policy->cpus);
1075 err_free_policy:
1076         kfree(policy);
1077
1078         return NULL;
1079 }
1080
1081 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1082 {
1083         struct kobject *kobj;
1084         struct completion *cmp;
1085
1086         if (notify)
1087                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1088                                              CPUFREQ_REMOVE_POLICY, policy);
1089
1090         down_write(&policy->rwsem);
1091         cpufreq_stats_free_table(policy);
1092         kobj = &policy->kobj;
1093         cmp = &policy->kobj_unregister;
1094         up_write(&policy->rwsem);
1095         kobject_put(kobj);
1096
1097         /*
1098          * We need to make sure that the underlying kobj is
1099          * actually not referenced anymore by anybody before we
1100          * proceed with unloading.
1101          */
1102         pr_debug("waiting for dropping of refcount\n");
1103         wait_for_completion(cmp);
1104         pr_debug("wait complete\n");
1105 }
1106
1107 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1108 {
1109         unsigned long flags;
1110         int cpu;
1111
1112         /* Remove policy from list */
1113         write_lock_irqsave(&cpufreq_driver_lock, flags);
1114         list_del(&policy->policy_list);
1115
1116         for_each_cpu(cpu, policy->related_cpus)
1117                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1118         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1119
1120         cpufreq_policy_put_kobj(policy, notify);
1121         free_cpumask_var(policy->real_cpus);
1122         free_cpumask_var(policy->related_cpus);
1123         free_cpumask_var(policy->cpus);
1124         kfree(policy);
1125 }
1126
1127 static int cpufreq_online(unsigned int cpu)
1128 {
1129         struct cpufreq_policy *policy;
1130         bool new_policy;
1131         unsigned long flags;
1132         unsigned int j;
1133         int ret;
1134
1135         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1136
1137         /* Check if this CPU already has a policy to manage it */
1138         policy = per_cpu(cpufreq_cpu_data, cpu);
1139         if (policy) {
1140                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1141                 if (!policy_is_inactive(policy))
1142                         return cpufreq_add_policy_cpu(policy, cpu);
1143
1144                 /* This is the only online CPU for the policy.  Start over. */
1145                 new_policy = false;
1146                 down_write(&policy->rwsem);
1147                 policy->cpu = cpu;
1148                 policy->governor = NULL;
1149                 up_write(&policy->rwsem);
1150         } else {
1151                 new_policy = true;
1152                 policy = cpufreq_policy_alloc(cpu);
1153                 if (!policy)
1154                         return -ENOMEM;
1155         }
1156
1157         cpumask_copy(policy->cpus, cpumask_of(cpu));
1158
1159         /* call driver. From then on the cpufreq must be able
1160          * to accept all calls to ->verify and ->setpolicy for this CPU
1161          */
1162         ret = cpufreq_driver->init(policy);
1163         if (ret) {
1164                 pr_debug("initialization failed\n");
1165                 goto out_free_policy;
1166         }
1167
1168         down_write(&policy->rwsem);
1169
1170         if (new_policy) {
1171                 /* related_cpus should at least include policy->cpus. */
1172                 cpumask_copy(policy->related_cpus, policy->cpus);
1173                 /* Clear mask of registered CPUs */
1174                 cpumask_clear(policy->real_cpus);
1175         }
1176
1177         /*
1178          * affected cpus must always be the one, which are online. We aren't
1179          * managing offline cpus here.
1180          */
1181         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1182
1183         if (new_policy) {
1184                 policy->user_policy.min = policy->min;
1185                 policy->user_policy.max = policy->max;
1186
1187                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1188                 for_each_cpu(j, policy->related_cpus)
1189                         per_cpu(cpufreq_cpu_data, j) = policy;
1190                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1191         }
1192
1193         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1194                 policy->cur = cpufreq_driver->get(policy->cpu);
1195                 if (!policy->cur) {
1196                         pr_err("%s: ->get() failed\n", __func__);
1197                         goto out_exit_policy;
1198                 }
1199         }
1200
1201         /*
1202          * Sometimes boot loaders set CPU frequency to a value outside of
1203          * frequency table present with cpufreq core. In such cases CPU might be
1204          * unstable if it has to run on that frequency for long duration of time
1205          * and so its better to set it to a frequency which is specified in
1206          * freq-table. This also makes cpufreq stats inconsistent as
1207          * cpufreq-stats would fail to register because current frequency of CPU
1208          * isn't found in freq-table.
1209          *
1210          * Because we don't want this change to effect boot process badly, we go
1211          * for the next freq which is >= policy->cur ('cur' must be set by now,
1212          * otherwise we will end up setting freq to lowest of the table as 'cur'
1213          * is initialized to zero).
1214          *
1215          * We are passing target-freq as "policy->cur - 1" otherwise
1216          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1217          * equal to target-freq.
1218          */
1219         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1220             && has_target()) {
1221                 /* Are we running at unknown frequency ? */
1222                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1223                 if (ret == -EINVAL) {
1224                         /* Warn user and fix it */
1225                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1226                                 __func__, policy->cpu, policy->cur);
1227                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1228                                 CPUFREQ_RELATION_L);
1229
1230                         /*
1231                          * Reaching here after boot in a few seconds may not
1232                          * mean that system will remain stable at "unknown"
1233                          * frequency for longer duration. Hence, a BUG_ON().
1234                          */
1235                         BUG_ON(ret);
1236                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1237                                 __func__, policy->cpu, policy->cur);
1238                 }
1239         }
1240
1241         if (new_policy) {
1242                 ret = cpufreq_add_dev_interface(policy);
1243                 if (ret)
1244                         goto out_exit_policy;
1245
1246                 cpufreq_stats_create_table(policy);
1247                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1248                                 CPUFREQ_CREATE_POLICY, policy);
1249
1250                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1251                 list_add(&policy->policy_list, &cpufreq_policy_list);
1252                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1253         }
1254
1255         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1256                                      CPUFREQ_START, policy);
1257
1258         ret = cpufreq_init_policy(policy);
1259         if (ret) {
1260                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1261                        __func__, cpu, ret);
1262                 /* cpufreq_policy_free() will notify based on this */
1263                 new_policy = false;
1264                 goto out_exit_policy;
1265         }
1266
1267         up_write(&policy->rwsem);
1268
1269         kobject_uevent(&policy->kobj, KOBJ_ADD);
1270
1271         /* Callback for handling stuff after policy is ready */
1272         if (cpufreq_driver->ready)
1273                 cpufreq_driver->ready(policy);
1274
1275         pr_debug("initialization complete\n");
1276
1277         return 0;
1278
1279 out_exit_policy:
1280         up_write(&policy->rwsem);
1281
1282         if (cpufreq_driver->exit)
1283                 cpufreq_driver->exit(policy);
1284 out_free_policy:
1285         cpufreq_policy_free(policy, !new_policy);
1286         return ret;
1287 }
1288
1289 static int cpufreq_offline(unsigned int cpu);
1290
1291 /**
1292  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1293  * @dev: CPU device.
1294  * @sif: Subsystem interface structure pointer (not used)
1295  */
1296 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1297 {
1298         struct cpufreq_policy *policy;
1299         unsigned cpu = dev->id;
1300         int ret;
1301
1302         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1303
1304         if (cpu_online(cpu)) {
1305                 ret = cpufreq_online(cpu);
1306                 if (ret)
1307                         return ret;
1308         }
1309
1310         /* Create sysfs link on CPU registration */
1311         policy = per_cpu(cpufreq_cpu_data, cpu);
1312         if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1313                 return 0;
1314
1315         ret = add_cpu_dev_symlink(policy, dev);
1316         if (ret) {
1317                 cpumask_clear_cpu(cpu, policy->real_cpus);
1318                 cpufreq_offline(cpu);
1319         }
1320
1321         return ret;
1322 }
1323
1324 static int cpufreq_offline(unsigned int cpu)
1325 {
1326         struct cpufreq_policy *policy;
1327         int ret;
1328
1329         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1330
1331         policy = cpufreq_cpu_get_raw(cpu);
1332         if (!policy) {
1333                 pr_debug("%s: No cpu_data found\n", __func__);
1334                 return 0;
1335         }
1336
1337         down_write(&policy->rwsem);
1338         if (has_target())
1339                 cpufreq_stop_governor(policy);
1340
1341         cpumask_clear_cpu(cpu, policy->cpus);
1342
1343         if (policy_is_inactive(policy)) {
1344                 if (has_target())
1345                         strncpy(policy->last_governor, policy->governor->name,
1346                                 CPUFREQ_NAME_LEN);
1347                 else
1348                         policy->last_policy = policy->policy;
1349         } else if (cpu == policy->cpu) {
1350                 /* Nominate new CPU */
1351                 policy->cpu = cpumask_any(policy->cpus);
1352         }
1353
1354         /* Start governor again for active policy */
1355         if (!policy_is_inactive(policy)) {
1356                 if (has_target()) {
1357                         ret = cpufreq_start_governor(policy);
1358                         if (ret)
1359                                 pr_err("%s: Failed to start governor\n", __func__);
1360                 }
1361
1362                 goto unlock;
1363         }
1364
1365         if (cpufreq_driver->stop_cpu)
1366                 cpufreq_driver->stop_cpu(policy);
1367
1368         if (has_target())
1369                 cpufreq_exit_governor(policy);
1370
1371         /*
1372          * Perform the ->exit() even during light-weight tear-down,
1373          * since this is a core component, and is essential for the
1374          * subsequent light-weight ->init() to succeed.
1375          */
1376         if (cpufreq_driver->exit) {
1377                 cpufreq_driver->exit(policy);
1378                 policy->freq_table = NULL;
1379         }
1380
1381 unlock:
1382         up_write(&policy->rwsem);
1383         return 0;
1384 }
1385
1386 /**
1387  * cpufreq_remove_dev - remove a CPU device
1388  *
1389  * Removes the cpufreq interface for a CPU device.
1390  */
1391 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1392 {
1393         unsigned int cpu = dev->id;
1394         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1395
1396         if (!policy)
1397                 return;
1398
1399         if (cpu_online(cpu))
1400                 cpufreq_offline(cpu);
1401
1402         cpumask_clear_cpu(cpu, policy->real_cpus);
1403         remove_cpu_dev_symlink(policy, dev);
1404
1405         if (cpumask_empty(policy->real_cpus))
1406                 cpufreq_policy_free(policy, true);
1407 }
1408
1409 /**
1410  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1411  *      in deep trouble.
1412  *      @policy: policy managing CPUs
1413  *      @new_freq: CPU frequency the CPU actually runs at
1414  *
1415  *      We adjust to current frequency first, and need to clean up later.
1416  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1417  */
1418 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1419                                 unsigned int new_freq)
1420 {
1421         struct cpufreq_freqs freqs;
1422
1423         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1424                  policy->cur, new_freq);
1425
1426         freqs.old = policy->cur;
1427         freqs.new = new_freq;
1428
1429         cpufreq_freq_transition_begin(policy, &freqs);
1430         cpufreq_freq_transition_end(policy, &freqs, 0);
1431 }
1432
1433 /**
1434  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1435  * @cpu: CPU number
1436  *
1437  * This is the last known freq, without actually getting it from the driver.
1438  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1439  */
1440 unsigned int cpufreq_quick_get(unsigned int cpu)
1441 {
1442         struct cpufreq_policy *policy;
1443         unsigned int ret_freq = 0;
1444         unsigned long flags;
1445
1446         read_lock_irqsave(&cpufreq_driver_lock, flags);
1447
1448         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1449                 ret_freq = cpufreq_driver->get(cpu);
1450                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1451                 return ret_freq;
1452         }
1453
1454         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1455
1456         policy = cpufreq_cpu_get(cpu);
1457         if (policy) {
1458                 ret_freq = policy->cur;
1459                 cpufreq_cpu_put(policy);
1460         }
1461
1462         return ret_freq;
1463 }
1464 EXPORT_SYMBOL(cpufreq_quick_get);
1465
1466 /**
1467  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1468  * @cpu: CPU number
1469  *
1470  * Just return the max possible frequency for a given CPU.
1471  */
1472 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1473 {
1474         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1475         unsigned int ret_freq = 0;
1476
1477         if (policy) {
1478                 ret_freq = policy->max;
1479                 cpufreq_cpu_put(policy);
1480         }
1481
1482         return ret_freq;
1483 }
1484 EXPORT_SYMBOL(cpufreq_quick_get_max);
1485
1486 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1487 {
1488         unsigned int ret_freq = 0;
1489
1490         if (!cpufreq_driver->get)
1491                 return ret_freq;
1492
1493         ret_freq = cpufreq_driver->get(policy->cpu);
1494
1495         /*
1496          * Updating inactive policies is invalid, so avoid doing that.  Also
1497          * if fast frequency switching is used with the given policy, the check
1498          * against policy->cur is pointless, so skip it in that case too.
1499          */
1500         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1501                 return ret_freq;
1502
1503         if (ret_freq && policy->cur &&
1504                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1505                 /* verify no discrepancy between actual and
1506                                         saved value exists */
1507                 if (unlikely(ret_freq != policy->cur)) {
1508                         cpufreq_out_of_sync(policy, ret_freq);
1509                         schedule_work(&policy->update);
1510                 }
1511         }
1512
1513         return ret_freq;
1514 }
1515
1516 /**
1517  * cpufreq_get - get the current CPU frequency (in kHz)
1518  * @cpu: CPU number
1519  *
1520  * Get the CPU current (static) CPU frequency
1521  */
1522 unsigned int cpufreq_get(unsigned int cpu)
1523 {
1524         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1525         unsigned int ret_freq = 0;
1526
1527         if (policy) {
1528                 down_read(&policy->rwsem);
1529                 ret_freq = __cpufreq_get(policy);
1530                 up_read(&policy->rwsem);
1531
1532                 cpufreq_cpu_put(policy);
1533         }
1534
1535         return ret_freq;
1536 }
1537 EXPORT_SYMBOL(cpufreq_get);
1538
1539 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1540 {
1541         unsigned int new_freq;
1542
1543         new_freq = cpufreq_driver->get(policy->cpu);
1544         if (!new_freq)
1545                 return 0;
1546
1547         if (!policy->cur) {
1548                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1549                 policy->cur = new_freq;
1550         } else if (policy->cur != new_freq && has_target()) {
1551                 cpufreq_out_of_sync(policy, new_freq);
1552         }
1553
1554         return new_freq;
1555 }
1556
1557 static struct subsys_interface cpufreq_interface = {
1558         .name           = "cpufreq",
1559         .subsys         = &cpu_subsys,
1560         .add_dev        = cpufreq_add_dev,
1561         .remove_dev     = cpufreq_remove_dev,
1562 };
1563
1564 /*
1565  * In case platform wants some specific frequency to be configured
1566  * during suspend..
1567  */
1568 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1569 {
1570         int ret;
1571
1572         if (!policy->suspend_freq) {
1573                 pr_debug("%s: suspend_freq not defined\n", __func__);
1574                 return 0;
1575         }
1576
1577         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1578                         policy->suspend_freq);
1579
1580         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1581                         CPUFREQ_RELATION_H);
1582         if (ret)
1583                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1584                                 __func__, policy->suspend_freq, ret);
1585
1586         return ret;
1587 }
1588 EXPORT_SYMBOL(cpufreq_generic_suspend);
1589
1590 /**
1591  * cpufreq_suspend() - Suspend CPUFreq governors
1592  *
1593  * Called during system wide Suspend/Hibernate cycles for suspending governors
1594  * as some platforms can't change frequency after this point in suspend cycle.
1595  * Because some of the devices (like: i2c, regulators, etc) they use for
1596  * changing frequency are suspended quickly after this point.
1597  */
1598 void cpufreq_suspend(void)
1599 {
1600         struct cpufreq_policy *policy;
1601
1602         if (!cpufreq_driver)
1603                 return;
1604
1605         if (!has_target() && !cpufreq_driver->suspend)
1606                 goto suspend;
1607
1608         pr_debug("%s: Suspending Governors\n", __func__);
1609
1610         for_each_active_policy(policy) {
1611                 if (has_target()) {
1612                         down_write(&policy->rwsem);
1613                         cpufreq_stop_governor(policy);
1614                         up_write(&policy->rwsem);
1615                 }
1616
1617                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1618                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1619                                 policy);
1620         }
1621
1622 suspend:
1623         cpufreq_suspended = true;
1624 }
1625
1626 /**
1627  * cpufreq_resume() - Resume CPUFreq governors
1628  *
1629  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1630  * are suspended with cpufreq_suspend().
1631  */
1632 void cpufreq_resume(void)
1633 {
1634         struct cpufreq_policy *policy;
1635         int ret;
1636
1637         if (!cpufreq_driver)
1638                 return;
1639
1640         cpufreq_suspended = false;
1641
1642         if (!has_target() && !cpufreq_driver->resume)
1643                 return;
1644
1645         pr_debug("%s: Resuming Governors\n", __func__);
1646
1647         for_each_active_policy(policy) {
1648                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1649                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1650                                 policy);
1651                 } else if (has_target()) {
1652                         down_write(&policy->rwsem);
1653                         ret = cpufreq_start_governor(policy);
1654                         up_write(&policy->rwsem);
1655
1656                         if (ret)
1657                                 pr_err("%s: Failed to start governor for policy: %p\n",
1658                                        __func__, policy);
1659                 }
1660         }
1661 }
1662
1663 /**
1664  *      cpufreq_get_current_driver - return current driver's name
1665  *
1666  *      Return the name string of the currently loaded cpufreq driver
1667  *      or NULL, if none.
1668  */
1669 const char *cpufreq_get_current_driver(void)
1670 {
1671         if (cpufreq_driver)
1672                 return cpufreq_driver->name;
1673
1674         return NULL;
1675 }
1676 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1677
1678 /**
1679  *      cpufreq_get_driver_data - return current driver data
1680  *
1681  *      Return the private data of the currently loaded cpufreq
1682  *      driver, or NULL if no cpufreq driver is loaded.
1683  */
1684 void *cpufreq_get_driver_data(void)
1685 {
1686         if (cpufreq_driver)
1687                 return cpufreq_driver->driver_data;
1688
1689         return NULL;
1690 }
1691 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1692
1693 /*********************************************************************
1694  *                     NOTIFIER LISTS INTERFACE                      *
1695  *********************************************************************/
1696
1697 /**
1698  *      cpufreq_register_notifier - register a driver with cpufreq
1699  *      @nb: notifier function to register
1700  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1701  *
1702  *      Add a driver to one of two lists: either a list of drivers that
1703  *      are notified about clock rate changes (once before and once after
1704  *      the transition), or a list of drivers that are notified about
1705  *      changes in cpufreq policy.
1706  *
1707  *      This function may sleep, and has the same return conditions as
1708  *      blocking_notifier_chain_register.
1709  */
1710 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1711 {
1712         int ret;
1713
1714         if (cpufreq_disabled())
1715                 return -EINVAL;
1716
1717         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1718
1719         switch (list) {
1720         case CPUFREQ_TRANSITION_NOTIFIER:
1721                 mutex_lock(&cpufreq_fast_switch_lock);
1722
1723                 if (cpufreq_fast_switch_count > 0) {
1724                         mutex_unlock(&cpufreq_fast_switch_lock);
1725                         return -EBUSY;
1726                 }
1727                 ret = srcu_notifier_chain_register(
1728                                 &cpufreq_transition_notifier_list, nb);
1729                 if (!ret)
1730                         cpufreq_fast_switch_count--;
1731
1732                 mutex_unlock(&cpufreq_fast_switch_lock);
1733                 break;
1734         case CPUFREQ_POLICY_NOTIFIER:
1735                 ret = blocking_notifier_chain_register(
1736                                 &cpufreq_policy_notifier_list, nb);
1737                 break;
1738         default:
1739                 ret = -EINVAL;
1740         }
1741
1742         return ret;
1743 }
1744 EXPORT_SYMBOL(cpufreq_register_notifier);
1745
1746 /**
1747  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1748  *      @nb: notifier block to be unregistered
1749  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1750  *
1751  *      Remove a driver from the CPU frequency notifier list.
1752  *
1753  *      This function may sleep, and has the same return conditions as
1754  *      blocking_notifier_chain_unregister.
1755  */
1756 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1757 {
1758         int ret;
1759
1760         if (cpufreq_disabled())
1761                 return -EINVAL;
1762
1763         switch (list) {
1764         case CPUFREQ_TRANSITION_NOTIFIER:
1765                 mutex_lock(&cpufreq_fast_switch_lock);
1766
1767                 ret = srcu_notifier_chain_unregister(
1768                                 &cpufreq_transition_notifier_list, nb);
1769                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1770                         cpufreq_fast_switch_count++;
1771
1772                 mutex_unlock(&cpufreq_fast_switch_lock);
1773                 break;
1774         case CPUFREQ_POLICY_NOTIFIER:
1775                 ret = blocking_notifier_chain_unregister(
1776                                 &cpufreq_policy_notifier_list, nb);
1777                 break;
1778         default:
1779                 ret = -EINVAL;
1780         }
1781
1782         return ret;
1783 }
1784 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1785
1786
1787 /*********************************************************************
1788  *                              GOVERNORS                            *
1789  *********************************************************************/
1790
1791 /**
1792  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1793  * @policy: cpufreq policy to switch the frequency for.
1794  * @target_freq: New frequency to set (may be approximate).
1795  *
1796  * Carry out a fast frequency switch without sleeping.
1797  *
1798  * The driver's ->fast_switch() callback invoked by this function must be
1799  * suitable for being called from within RCU-sched read-side critical sections
1800  * and it is expected to select the minimum available frequency greater than or
1801  * equal to @target_freq (CPUFREQ_RELATION_L).
1802  *
1803  * This function must not be called if policy->fast_switch_enabled is unset.
1804  *
1805  * Governors calling this function must guarantee that it will never be invoked
1806  * twice in parallel for the same policy and that it will never be called in
1807  * parallel with either ->target() or ->target_index() for the same policy.
1808  *
1809  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1810  * callback to indicate an error condition, the hardware configuration must be
1811  * preserved.
1812  */
1813 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1814                                         unsigned int target_freq)
1815 {
1816         target_freq = clamp_val(target_freq, policy->min, policy->max);
1817
1818         return cpufreq_driver->fast_switch(policy, target_freq);
1819 }
1820 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1821
1822 /* Must set freqs->new to intermediate frequency */
1823 static int __target_intermediate(struct cpufreq_policy *policy,
1824                                  struct cpufreq_freqs *freqs, int index)
1825 {
1826         int ret;
1827
1828         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1829
1830         /* We don't need to switch to intermediate freq */
1831         if (!freqs->new)
1832                 return 0;
1833
1834         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1835                  __func__, policy->cpu, freqs->old, freqs->new);
1836
1837         cpufreq_freq_transition_begin(policy, freqs);
1838         ret = cpufreq_driver->target_intermediate(policy, index);
1839         cpufreq_freq_transition_end(policy, freqs, ret);
1840
1841         if (ret)
1842                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1843                        __func__, ret);
1844
1845         return ret;
1846 }
1847
1848 static int __target_index(struct cpufreq_policy *policy, int index)
1849 {
1850         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1851         unsigned int intermediate_freq = 0;
1852         unsigned int newfreq = policy->freq_table[index].frequency;
1853         int retval = -EINVAL;
1854         bool notify;
1855
1856         if (newfreq == policy->cur)
1857                 return 0;
1858
1859         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1860         if (notify) {
1861                 /* Handle switching to intermediate frequency */
1862                 if (cpufreq_driver->get_intermediate) {
1863                         retval = __target_intermediate(policy, &freqs, index);
1864                         if (retval)
1865                                 return retval;
1866
1867                         intermediate_freq = freqs.new;
1868                         /* Set old freq to intermediate */
1869                         if (intermediate_freq)
1870                                 freqs.old = freqs.new;
1871                 }
1872
1873                 freqs.new = newfreq;
1874                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1875                          __func__, policy->cpu, freqs.old, freqs.new);
1876
1877                 cpufreq_freq_transition_begin(policy, &freqs);
1878         }
1879
1880         retval = cpufreq_driver->target_index(policy, index);
1881         if (retval)
1882                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1883                        retval);
1884
1885         if (notify) {
1886                 cpufreq_freq_transition_end(policy, &freqs, retval);
1887
1888                 /*
1889                  * Failed after setting to intermediate freq? Driver should have
1890                  * reverted back to initial frequency and so should we. Check
1891                  * here for intermediate_freq instead of get_intermediate, in
1892                  * case we haven't switched to intermediate freq at all.
1893                  */
1894                 if (unlikely(retval && intermediate_freq)) {
1895                         freqs.old = intermediate_freq;
1896                         freqs.new = policy->restore_freq;
1897                         cpufreq_freq_transition_begin(policy, &freqs);
1898                         cpufreq_freq_transition_end(policy, &freqs, 0);
1899                 }
1900         }
1901
1902         return retval;
1903 }
1904
1905 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1906                             unsigned int target_freq,
1907                             unsigned int relation)
1908 {
1909         unsigned int old_target_freq = target_freq;
1910         int index;
1911
1912         if (cpufreq_disabled())
1913                 return -ENODEV;
1914
1915         /* Make sure that target_freq is within supported range */
1916         target_freq = clamp_val(target_freq, policy->min, policy->max);
1917
1918         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1919                  policy->cpu, target_freq, relation, old_target_freq);
1920
1921         /*
1922          * This might look like a redundant call as we are checking it again
1923          * after finding index. But it is left intentionally for cases where
1924          * exactly same freq is called again and so we can save on few function
1925          * calls.
1926          */
1927         if (target_freq == policy->cur)
1928                 return 0;
1929
1930         /* Save last value to restore later on errors */
1931         policy->restore_freq = policy->cur;
1932
1933         if (cpufreq_driver->target)
1934                 return cpufreq_driver->target(policy, target_freq, relation);
1935
1936         if (!cpufreq_driver->target_index)
1937                 return -EINVAL;
1938
1939         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1940
1941         return __target_index(policy, index);
1942 }
1943 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1944
1945 int cpufreq_driver_target(struct cpufreq_policy *policy,
1946                           unsigned int target_freq,
1947                           unsigned int relation)
1948 {
1949         int ret = -EINVAL;
1950
1951         down_write(&policy->rwsem);
1952
1953         ret = __cpufreq_driver_target(policy, target_freq, relation);
1954
1955         up_write(&policy->rwsem);
1956
1957         return ret;
1958 }
1959 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1960
1961 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1962 {
1963         return NULL;
1964 }
1965
1966 static int cpufreq_init_governor(struct cpufreq_policy *policy)
1967 {
1968         int ret;
1969
1970         /* Don't start any governor operations if we are entering suspend */
1971         if (cpufreq_suspended)
1972                 return 0;
1973         /*
1974          * Governor might not be initiated here if ACPI _PPC changed
1975          * notification happened, so check it.
1976          */
1977         if (!policy->governor)
1978                 return -EINVAL;
1979
1980         if (policy->governor->max_transition_latency &&
1981             policy->cpuinfo.transition_latency >
1982             policy->governor->max_transition_latency) {
1983                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
1984
1985                 if (gov) {
1986                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
1987                                 policy->governor->name, gov->name);
1988                         policy->governor = gov;
1989                 } else {
1990                         return -EINVAL;
1991                 }
1992         }
1993
1994         if (!try_module_get(policy->governor->owner))
1995                 return -EINVAL;
1996
1997         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
1998
1999         if (policy->governor->init) {
2000                 ret = policy->governor->init(policy);
2001                 if (ret) {
2002                         module_put(policy->governor->owner);
2003                         return ret;
2004                 }
2005         }
2006
2007         return 0;
2008 }
2009
2010 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2011 {
2012         if (cpufreq_suspended || !policy->governor)
2013                 return;
2014
2015         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2016
2017         if (policy->governor->exit)
2018                 policy->governor->exit(policy);
2019
2020         module_put(policy->governor->owner);
2021 }
2022
2023 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2024 {
2025         int ret;
2026
2027         if (cpufreq_suspended)
2028                 return 0;
2029
2030         if (!policy->governor)
2031                 return -EINVAL;
2032
2033         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2034
2035         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2036                 cpufreq_update_current_freq(policy);
2037
2038         if (policy->governor->start) {
2039                 ret = policy->governor->start(policy);
2040                 if (ret)
2041                         return ret;
2042         }
2043
2044         if (policy->governor->limits)
2045                 policy->governor->limits(policy);
2046
2047         return 0;
2048 }
2049
2050 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2051 {
2052         if (cpufreq_suspended || !policy->governor)
2053                 return;
2054
2055         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2056
2057         if (policy->governor->stop)
2058                 policy->governor->stop(policy);
2059 }
2060
2061 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2062 {
2063         if (cpufreq_suspended || !policy->governor)
2064                 return;
2065
2066         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2067
2068         if (policy->governor->limits)
2069                 policy->governor->limits(policy);
2070 }
2071
2072 int cpufreq_register_governor(struct cpufreq_governor *governor)
2073 {
2074         int err;
2075
2076         if (!governor)
2077                 return -EINVAL;
2078
2079         if (cpufreq_disabled())
2080                 return -ENODEV;
2081
2082         mutex_lock(&cpufreq_governor_mutex);
2083
2084         err = -EBUSY;
2085         if (!find_governor(governor->name)) {
2086                 err = 0;
2087                 list_add(&governor->governor_list, &cpufreq_governor_list);
2088         }
2089
2090         mutex_unlock(&cpufreq_governor_mutex);
2091         return err;
2092 }
2093 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2094
2095 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2096 {
2097         struct cpufreq_policy *policy;
2098         unsigned long flags;
2099
2100         if (!governor)
2101                 return;
2102
2103         if (cpufreq_disabled())
2104                 return;
2105
2106         /* clear last_governor for all inactive policies */
2107         read_lock_irqsave(&cpufreq_driver_lock, flags);
2108         for_each_inactive_policy(policy) {
2109                 if (!strcmp(policy->last_governor, governor->name)) {
2110                         policy->governor = NULL;
2111                         strcpy(policy->last_governor, "\0");
2112                 }
2113         }
2114         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2115
2116         mutex_lock(&cpufreq_governor_mutex);
2117         list_del(&governor->governor_list);
2118         mutex_unlock(&cpufreq_governor_mutex);
2119         return;
2120 }
2121 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2122
2123
2124 /*********************************************************************
2125  *                          POLICY INTERFACE                         *
2126  *********************************************************************/
2127
2128 /**
2129  * cpufreq_get_policy - get the current cpufreq_policy
2130  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2131  *      is written
2132  *
2133  * Reads the current cpufreq policy.
2134  */
2135 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2136 {
2137         struct cpufreq_policy *cpu_policy;
2138         if (!policy)
2139                 return -EINVAL;
2140
2141         cpu_policy = cpufreq_cpu_get(cpu);
2142         if (!cpu_policy)
2143                 return -EINVAL;
2144
2145         memcpy(policy, cpu_policy, sizeof(*policy));
2146
2147         cpufreq_cpu_put(cpu_policy);
2148         return 0;
2149 }
2150 EXPORT_SYMBOL(cpufreq_get_policy);
2151
2152 /*
2153  * policy : current policy.
2154  * new_policy: policy to be set.
2155  */
2156 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2157                                 struct cpufreq_policy *new_policy)
2158 {
2159         struct cpufreq_governor *old_gov;
2160         int ret;
2161
2162         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2163                  new_policy->cpu, new_policy->min, new_policy->max);
2164
2165         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2166
2167         /*
2168         * This check works well when we store new min/max freq attributes,
2169         * because new_policy is a copy of policy with one field updated.
2170         */
2171         if (new_policy->min > new_policy->max)
2172                 return -EINVAL;
2173
2174         /* verify the cpu speed can be set within this limit */
2175         ret = cpufreq_driver->verify(new_policy);
2176         if (ret)
2177                 return ret;
2178
2179         /* adjust if necessary - all reasons */
2180         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2181                         CPUFREQ_ADJUST, new_policy);
2182
2183         /*
2184          * verify the cpu speed can be set within this limit, which might be
2185          * different to the first one
2186          */
2187         ret = cpufreq_driver->verify(new_policy);
2188         if (ret)
2189                 return ret;
2190
2191         /* notification of the new policy */
2192         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2193                         CPUFREQ_NOTIFY, new_policy);
2194
2195         policy->min = new_policy->min;
2196         policy->max = new_policy->max;
2197
2198         policy->cached_target_freq = UINT_MAX;
2199
2200         pr_debug("new min and max freqs are %u - %u kHz\n",
2201                  policy->min, policy->max);
2202
2203         if (cpufreq_driver->setpolicy) {
2204                 policy->policy = new_policy->policy;
2205                 pr_debug("setting range\n");
2206                 return cpufreq_driver->setpolicy(new_policy);
2207         }
2208
2209         if (new_policy->governor == policy->governor) {
2210                 pr_debug("cpufreq: governor limits update\n");
2211                 cpufreq_governor_limits(policy);
2212                 return 0;
2213         }
2214
2215         pr_debug("governor switch\n");
2216
2217         /* save old, working values */
2218         old_gov = policy->governor;
2219         /* end old governor */
2220         if (old_gov) {
2221                 cpufreq_stop_governor(policy);
2222                 cpufreq_exit_governor(policy);
2223         }
2224
2225         /* start new governor */
2226         policy->governor = new_policy->governor;
2227         ret = cpufreq_init_governor(policy);
2228         if (!ret) {
2229                 ret = cpufreq_start_governor(policy);
2230                 if (!ret) {
2231                         pr_debug("cpufreq: governor change\n");
2232                         return 0;
2233                 }
2234                 cpufreq_exit_governor(policy);
2235         }
2236
2237         /* new governor failed, so re-start old one */
2238         pr_debug("starting governor %s failed\n", policy->governor->name);
2239         if (old_gov) {
2240                 policy->governor = old_gov;
2241                 if (cpufreq_init_governor(policy))
2242                         policy->governor = NULL;
2243                 else
2244                         cpufreq_start_governor(policy);
2245         }
2246
2247         return ret;
2248 }
2249
2250 /**
2251  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2252  *      @cpu: CPU which shall be re-evaluated
2253  *
2254  *      Useful for policy notifiers which have different necessities
2255  *      at different times.
2256  */
2257 int cpufreq_update_policy(unsigned int cpu)
2258 {
2259         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2260         struct cpufreq_policy new_policy;
2261         int ret;
2262
2263         if (!policy)
2264                 return -ENODEV;
2265
2266         down_write(&policy->rwsem);
2267
2268         pr_debug("updating policy for CPU %u\n", cpu);
2269         memcpy(&new_policy, policy, sizeof(*policy));
2270         new_policy.min = policy->user_policy.min;
2271         new_policy.max = policy->user_policy.max;
2272
2273         /*
2274          * BIOS might change freq behind our back
2275          * -> ask driver for current freq and notify governors about a change
2276          */
2277         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2278                 if (cpufreq_suspended) {
2279                         ret = -EAGAIN;
2280                         goto unlock;
2281                 }
2282                 new_policy.cur = cpufreq_update_current_freq(policy);
2283                 if (WARN_ON(!new_policy.cur)) {
2284                         ret = -EIO;
2285                         goto unlock;
2286                 }
2287         }
2288
2289         ret = cpufreq_set_policy(policy, &new_policy);
2290
2291 unlock:
2292         up_write(&policy->rwsem);
2293
2294         cpufreq_cpu_put(policy);
2295         return ret;
2296 }
2297 EXPORT_SYMBOL(cpufreq_update_policy);
2298
2299 /*********************************************************************
2300  *               BOOST                                               *
2301  *********************************************************************/
2302 static int cpufreq_boost_set_sw(int state)
2303 {
2304         struct cpufreq_policy *policy;
2305         int ret = -EINVAL;
2306
2307         for_each_active_policy(policy) {
2308                 if (!policy->freq_table)
2309                         continue;
2310
2311                 ret = cpufreq_frequency_table_cpuinfo(policy,
2312                                                       policy->freq_table);
2313                 if (ret) {
2314                         pr_err("%s: Policy frequency update failed\n",
2315                                __func__);
2316                         break;
2317                 }
2318
2319                 down_write(&policy->rwsem);
2320                 policy->user_policy.max = policy->max;
2321                 cpufreq_governor_limits(policy);
2322                 up_write(&policy->rwsem);
2323         }
2324
2325         return ret;
2326 }
2327
2328 int cpufreq_boost_trigger_state(int state)
2329 {
2330         unsigned long flags;
2331         int ret = 0;
2332
2333         if (cpufreq_driver->boost_enabled == state)
2334                 return 0;
2335
2336         write_lock_irqsave(&cpufreq_driver_lock, flags);
2337         cpufreq_driver->boost_enabled = state;
2338         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2339
2340         ret = cpufreq_driver->set_boost(state);
2341         if (ret) {
2342                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2343                 cpufreq_driver->boost_enabled = !state;
2344                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2345
2346                 pr_err("%s: Cannot %s BOOST\n",
2347                        __func__, state ? "enable" : "disable");
2348         }
2349
2350         return ret;
2351 }
2352
2353 static bool cpufreq_boost_supported(void)
2354 {
2355         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2356 }
2357
2358 static int create_boost_sysfs_file(void)
2359 {
2360         int ret;
2361
2362         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2363         if (ret)
2364                 pr_err("%s: cannot register global BOOST sysfs file\n",
2365                        __func__);
2366
2367         return ret;
2368 }
2369
2370 static void remove_boost_sysfs_file(void)
2371 {
2372         if (cpufreq_boost_supported())
2373                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2374 }
2375
2376 int cpufreq_enable_boost_support(void)
2377 {
2378         if (!cpufreq_driver)
2379                 return -EINVAL;
2380
2381         if (cpufreq_boost_supported())
2382                 return 0;
2383
2384         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2385
2386         /* This will get removed on driver unregister */
2387         return create_boost_sysfs_file();
2388 }
2389 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2390
2391 int cpufreq_boost_enabled(void)
2392 {
2393         return cpufreq_driver->boost_enabled;
2394 }
2395 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2396
2397 /*********************************************************************
2398  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2399  *********************************************************************/
2400 static enum cpuhp_state hp_online;
2401
2402 /**
2403  * cpufreq_register_driver - register a CPU Frequency driver
2404  * @driver_data: A struct cpufreq_driver containing the values#
2405  * submitted by the CPU Frequency driver.
2406  *
2407  * Registers a CPU Frequency driver to this core code. This code
2408  * returns zero on success, -EEXIST when another driver got here first
2409  * (and isn't unregistered in the meantime).
2410  *
2411  */
2412 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2413 {
2414         unsigned long flags;
2415         int ret;
2416
2417         if (cpufreq_disabled())
2418                 return -ENODEV;
2419
2420         if (!driver_data || !driver_data->verify || !driver_data->init ||
2421             !(driver_data->setpolicy || driver_data->target_index ||
2422                     driver_data->target) ||
2423              (driver_data->setpolicy && (driver_data->target_index ||
2424                     driver_data->target)) ||
2425              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2426                 return -EINVAL;
2427
2428         pr_debug("trying to register driver %s\n", driver_data->name);
2429
2430         /* Protect against concurrent CPU online/offline. */
2431         get_online_cpus();
2432
2433         write_lock_irqsave(&cpufreq_driver_lock, flags);
2434         if (cpufreq_driver) {
2435                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2436                 ret = -EEXIST;
2437                 goto out;
2438         }
2439         cpufreq_driver = driver_data;
2440         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2441
2442         if (driver_data->setpolicy)
2443                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2444
2445         if (cpufreq_boost_supported()) {
2446                 ret = create_boost_sysfs_file();
2447                 if (ret)
2448                         goto err_null_driver;
2449         }
2450
2451         ret = subsys_interface_register(&cpufreq_interface);
2452         if (ret)
2453                 goto err_boost_unreg;
2454
2455         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2456             list_empty(&cpufreq_policy_list)) {
2457                 /* if all ->init() calls failed, unregister */
2458                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2459                          driver_data->name);
2460                 goto err_if_unreg;
2461         }
2462
2463         ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "cpufreq:online",
2464                                         cpufreq_online,
2465                                         cpufreq_offline);
2466         if (ret < 0)
2467                 goto err_if_unreg;
2468         hp_online = ret;
2469         ret = 0;
2470
2471         pr_debug("driver %s up and running\n", driver_data->name);
2472         goto out;
2473
2474 err_if_unreg:
2475         subsys_interface_unregister(&cpufreq_interface);
2476 err_boost_unreg:
2477         remove_boost_sysfs_file();
2478 err_null_driver:
2479         write_lock_irqsave(&cpufreq_driver_lock, flags);
2480         cpufreq_driver = NULL;
2481         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2482 out:
2483         put_online_cpus();
2484         return ret;
2485 }
2486 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2487
2488 /**
2489  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2490  *
2491  * Unregister the current CPUFreq driver. Only call this if you have
2492  * the right to do so, i.e. if you have succeeded in initialising before!
2493  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2494  * currently not initialised.
2495  */
2496 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2497 {
2498         unsigned long flags;
2499
2500         if (!cpufreq_driver || (driver != cpufreq_driver))
2501                 return -EINVAL;
2502
2503         pr_debug("unregistering driver %s\n", driver->name);
2504
2505         /* Protect against concurrent cpu hotplug */
2506         get_online_cpus();
2507         subsys_interface_unregister(&cpufreq_interface);
2508         remove_boost_sysfs_file();
2509         cpuhp_remove_state_nocalls(hp_online);
2510
2511         write_lock_irqsave(&cpufreq_driver_lock, flags);
2512
2513         cpufreq_driver = NULL;
2514
2515         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2516         put_online_cpus();
2517
2518         return 0;
2519 }
2520 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2521
2522 /*
2523  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2524  * or mutexes when secondary CPUs are halted.
2525  */
2526 static struct syscore_ops cpufreq_syscore_ops = {
2527         .shutdown = cpufreq_suspend,
2528 };
2529
2530 struct kobject *cpufreq_global_kobject;
2531 EXPORT_SYMBOL(cpufreq_global_kobject);
2532
2533 static int __init cpufreq_core_init(void)
2534 {
2535         if (cpufreq_disabled())
2536                 return -ENODEV;
2537
2538         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2539         BUG_ON(!cpufreq_global_kobject);
2540
2541         register_syscore_ops(&cpufreq_syscore_ops);
2542
2543         return 0;
2544 }
2545 core_initcall(cpufreq_core_init);