drivers/cpufreq/cpufreq.c

   1 /*
   2  *  linux/drivers/cpufreq/cpufreq.c
   3  *
   4  *  Copyright (C) 2001 Russell King
   5  *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
   6  *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
   7  *
   8  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
   9  *      Added handling for CPU hotplug
  10  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
  11  *      Fix handling for CPU hotplug -- affected CPUs
  12  *
  13  * This program is free software; you can redistribute it and/or modify
  14  * it under the terms of the GNU General Public License version 2 as
  15  * published by the Free Software Foundation.
  16  */
  17
  18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  19
  20 #include <linux/cpu.h>
  21 #include <linux/cpufreq.h>
  22 #include <linux/delay.h>
  23 #include <linux/device.h>
  24 #include <linux/init.h>
  25 #include <linux/kernel_stat.h>
  26 #include <linux/module.h>
  27 #include <linux/mutex.h>
  28 #include <linux/slab.h>
  29 #include <linux/suspend.h>
  30 #include <linux/syscore_ops.h>
  31 #include <linux/tick.h>
  32 #include <trace/events/power.h>
  33
  34 static LIST_HEAD(cpufreq_policy_list);
  35
  36 static inline bool policy_is_inactive(struct cpufreq_policy *policy)
  37 {
  38         return cpumask_empty(policy->cpus);
  39 }
  40
  41 /* Macros to iterate over CPU policies */
  42 #define for_each_suitable_policy(__policy, __active)                     \
  43         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
  44                 if ((__active) == !policy_is_inactive(__policy))
  45
  46 #define for_each_active_policy(__policy)                \
  47         for_each_suitable_policy(__policy, true)
  48 #define for_each_inactive_policy(__policy)              \
  49         for_each_suitable_policy(__policy, false)
  50
  51 #define for_each_policy(__policy)                       \
  52         list_for_each_entry(__policy, &cpufreq_policy_list, policy_list)
  53
  54 /* Iterate over governors */
  55 static LIST_HEAD(cpufreq_governor_list);
  56 #define for_each_governor(__governor)                           \
  57         list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
  58
  59 /**
  60  * The "cpufreq driver" - the arch- or hardware-dependent low
  61  * level driver of CPUFreq support, and its spinlock. This lock
  62  * also protects the cpufreq_cpu_data array.
  63  */
  64 static struct cpufreq_driver *cpufreq_driver;
  65 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
  66 static DEFINE_RWLOCK(cpufreq_driver_lock);
  67
  68 /* Flag to suspend/resume CPUFreq governors */
  69 static bool cpufreq_suspended;
  70
  71 static inline bool has_target(void)
  72 {
  73         return cpufreq_driver->target_index || cpufreq_driver->target;
  74 }
  75
  76 /* internal prototypes */
  77 static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
  78 static int cpufreq_init_governor(struct cpufreq_policy *policy);
  79 static void cpufreq_exit_governor(struct cpufreq_policy *policy);
  80 static int cpufreq_start_governor(struct cpufreq_policy *policy);
  81 static void cpufreq_stop_governor(struct cpufreq_policy *policy);
  82 static void cpufreq_governor_limits(struct cpufreq_policy *policy);
  83
  84 /**
  85  * Two notifier lists: the "policy" list is involved in the
  86  * validation process for a new CPU frequency policy; the
  87  * "transition" list for kernel code that needs to handle
  88  * changes to devices when the CPU clock speed changes.
  89  * The mutex locks both lists.
  90  */
  91 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
  92 static struct srcu_notifier_head cpufreq_transition_notifier_list;
  93
  94 static bool init_cpufreq_transition_notifier_list_called;
  95 static int __init init_cpufreq_transition_notifier_list(void)
  96 {
  97         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
  98         init_cpufreq_transition_notifier_list_called = true;
  99         return 0;
 100 }
 101 pure_initcall(init_cpufreq_transition_notifier_list);
 102
 103 static int off __read_mostly;
 104 static int cpufreq_disabled(void)
 105 {
 106         return off;
 107 }
 108 void disable_cpufreq(void)
 109 {
 110         off = 1;
 111 }
 112 static DEFINE_MUTEX(cpufreq_governor_mutex);
 113
 114 bool have_governor_per_policy(void)
 115 {
 116         return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
 117 }
 118 EXPORT_SYMBOL_GPL(have_governor_per_policy);
 119
 120 struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
 121 {
 122         if (have_governor_per_policy())
 123                 return &policy->kobj;
 124         else
 125                 return cpufreq_global_kobject;
 126 }
 127 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
 128
 129 static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
 130 {
 131         u64 idle_time;
 132         u64 cur_wall_time;
 133         u64 busy_time;
 134
 135         cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
 136
 137         busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
 138         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
 139         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
 140         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
 141         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
 142         busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
 143
 144         idle_time = cur_wall_time - busy_time;
 145         if (wall)
 146                 *wall = cputime_to_usecs(cur_wall_time);
 147
 148         return cputime_to_usecs(idle_time);
 149 }
 150
 151 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
 152 {
 153         u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
 154
 155         if (idle_time == -1ULL)
 156                 return get_cpu_idle_time_jiffy(cpu, wall);
 157         else if (!io_busy)
 158                 idle_time += get_cpu_iowait_time_us(cpu, wall);
 159
 160         return idle_time;
 161 }
 162 EXPORT_SYMBOL_GPL(get_cpu_idle_time);
 163
 164 /*
 165  * This is a generic cpufreq init() routine which can be used by cpufreq
 166  * drivers of SMP systems. It will do following:
 167  * - validate & show freq table passed
 168  * - set policies transition latency
 169  * - policy->cpus with all possible CPUs
 170  */
 171 int cpufreq_generic_init(struct cpufreq_policy *policy,
 172                 struct cpufreq_frequency_table *table,
 173                 unsigned int transition_latency)
 174 {
 175         int ret;
 176
 177         ret = cpufreq_table_validate_and_show(policy, table);
 178         if (ret) {
 179                 pr_err("%s: invalid frequency table: %d\n", __func__, ret);
 180                 return ret;
 181         }
 182
 183         policy->cpuinfo.transition_latency = transition_latency;
 184
 185         /*
 186          * The driver only supports the SMP configuration where all processors
 187          * share the clock and voltage and clock.
 188          */
 189         cpumask_setall(policy->cpus);
 190
 191         return 0;
 192 }
 193 EXPORT_SYMBOL_GPL(cpufreq_generic_init);
 194
 195 struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
 196 {
 197         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
 198
 199         return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
 200 }
 201 EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
 202
 203 unsigned int cpufreq_generic_get(unsigned int cpu)
 204 {
 205         struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
 206
 207         if (!policy || IS_ERR(policy->clk)) {
 208                 pr_err("%s: No %s associated to cpu: %d\n",
 209                        __func__, policy ? "clk" : "policy", cpu);
 210                 return 0;
 211         }
 212
 213         return clk_get_rate(policy->clk) / 1000;
 214 }
 215 EXPORT_SYMBOL_GPL(cpufreq_generic_get);
 216
 217 /**
 218  * cpufreq_cpu_get: returns policy for a cpu and marks it busy.
 219  *
 220  * @cpu: cpu to find policy for.
 221  *
 222  * This returns policy for 'cpu', returns NULL if it doesn't exist.
 223  * It also increments the kobject reference count to mark it busy and so would
 224  * require a corresponding call to cpufreq_cpu_put() to decrement it back.
 225  * If corresponding call cpufreq_cpu_put() isn't made, the policy wouldn't be
 226  * freed as that depends on the kobj count.
 227  *
 228  * Return: A valid policy on success, otherwise NULL on failure.
 229  */
 230 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
 231 {
 232         struct cpufreq_policy *policy = NULL;
 233         unsigned long flags;
 234
 235         if (WARN_ON(cpu >= nr_cpu_ids))
 236                 return NULL;
 237
 238         /* get the cpufreq driver */
 239         read_lock_irqsave(&cpufreq_driver_lock, flags);
 240
 241         if (cpufreq_driver) {
 242                 /* get the CPU */
 243                 policy = cpufreq_cpu_get_raw(cpu);
 244                 if (policy)
 245                         kobject_get(&policy->kobj);
 246         }
 247
 248         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
 249
 250         return policy;
 251 }
 252 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 253
 254 /**
 255  * cpufreq_cpu_put: Decrements the usage count of a policy
 256  *
 257  * @policy: policy earlier returned by cpufreq_cpu_get().
 258  *
 259  * This decrements the kobject reference count incremented earlier by calling
 260  * cpufreq_cpu_get().
 261  */
 262 void cpufreq_cpu_put(struct cpufreq_policy *policy)
 263 {
 264         kobject_put(&policy->kobj);
 265 }
 266 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 267
 268 /*********************************************************************
 269  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 270  *********************************************************************/
 271
 272 /**
 273  * adjust_jiffies - adjust the system "loops_per_jiffy"
 274  *
 275  * This function alters the system "loops_per_jiffy" for the clock
 276  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 277  * systems as each CPU might be scaled differently. So, use the arch
 278  * per-CPU loops_per_jiffy value wherever possible.
 279  */
 280 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 281 {
 282 #ifndef CONFIG_SMP
 283         static unsigned long l_p_j_ref;
 284         static unsigned int l_p_j_ref_freq;
 285
 286         if (ci->flags & CPUFREQ_CONST_LOOPS)
 287                 return;
 288
 289         if (!l_p_j_ref_freq) {
 290                 l_p_j_ref = loops_per_jiffy;
 291                 l_p_j_ref_freq = ci->old;
 292                 pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
 293                          l_p_j_ref, l_p_j_ref_freq);
 294         }
 295         if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
 296                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
 297                                                                 ci->new);
 298                 pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
 299                          loops_per_jiffy, ci->new);
 300         }
 301 #endif
 302 }
 303
 304 static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
 305                 struct cpufreq_freqs *freqs, unsigned int state)
 306 {
 307         BUG_ON(irqs_disabled());
 308
 309         if (cpufreq_disabled())
 310                 return;
 311
 312         freqs->flags = cpufreq_driver->flags;
 313         pr_debug("notification %u of frequency transition to %u kHz\n",
 314                  state, freqs->new);
 315
 316         switch (state) {
 317
 318         case CPUFREQ_PRECHANGE:
 319                 /* detect if the driver reported a value as "old frequency"
 320                  * which is not equal to what the cpufreq core thinks is
 321                  * "old frequency".
 322                  */
 323                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 324                         if ((policy) && (policy->cpu == freqs->cpu) &&
 325                             (policy->cur) && (policy->cur != freqs->old)) {
 326                                 pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
 327                                          freqs->old, policy->cur);
 328                                 freqs->old = policy->cur;
 329                         }
 330                 }
 331                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 332                                 CPUFREQ_PRECHANGE, freqs);
 333                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 334                 break;
 335
 336         case CPUFREQ_POSTCHANGE:
 337                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 338                 pr_debug("FREQ: %lu - CPU: %lu\n",
 339                          (unsigned long)freqs->new, (unsigned long)freqs->cpu);
 340                 trace_cpu_frequency(freqs->new, freqs->cpu);
 341                 cpufreq_stats_record_transition(policy, freqs->new);
 342                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 343                                 CPUFREQ_POSTCHANGE, freqs);
 344                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 345                         policy->cur = freqs->new;
 346                 break;
 347         }
 348 }
 349
 350 /**
 351  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 352  * on frequency transition.
 353  *
 354  * This function calls the transition notifiers and the "adjust_jiffies"
 355  * function. It is called twice on all CPU frequency changes that have
 356  * external effects.
 357  */
 358 static void cpufreq_notify_transition(struct cpufreq_policy *policy,
 359                 struct cpufreq_freqs *freqs, unsigned int state)
 360 {
 361         for_each_cpu(freqs->cpu, policy->cpus)
 362                 __cpufreq_notify_transition(policy, freqs, state);
 363 }
 364
 365 /* Do post notifications when there are chances that transition has failed */
 366 static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
 367                 struct cpufreq_freqs *freqs, int transition_failed)
 368 {
 369         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
 370         if (!transition_failed)
 371                 return;
 372
 373         swap(freqs->old, freqs->new);
 374         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
 375         cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
 376 }
 377
 378 void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
 379                 struct cpufreq_freqs *freqs)
 380 {
 381
 382         /*
 383          * Catch double invocations of _begin() which lead to self-deadlock.
 384          * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
 385          * doesn't invoke _begin() on their behalf, and hence the chances of
 386          * double invocations are very low. Moreover, there are scenarios
 387          * where these checks can emit false-positive warnings in these
 388          * drivers; so we avoid that by skipping them altogether.
 389          */
 390         WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
 391                                 && current == policy->transition_task);
 392
 393 wait:
 394         wait_event(policy->transition_wait, !policy->transition_ongoing);
 395
 396         spin_lock(&policy->transition_lock);
 397
 398         if (unlikely(policy->transition_ongoing)) {
 399                 spin_unlock(&policy->transition_lock);
 400                 goto wait;
 401         }
 402
 403         policy->transition_ongoing = true;
 404         policy->transition_task = current;
 405
 406         spin_unlock(&policy->transition_lock);
 407
 408         cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
 409 }
 410 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
 411
 412 void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
 413                 struct cpufreq_freqs *freqs, int transition_failed)
 414 {
 415         if (unlikely(WARN_ON(!policy->transition_ongoing)))
 416                 return;
 417
 418         cpufreq_notify_post_transition(policy, freqs, transition_failed);
 419
 420         policy->transition_ongoing = false;
 421         policy->transition_task = NULL;
 422
 423         wake_up(&policy->transition_wait);
 424 }
 425 EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
 426
 427 /*
 428  * Fast frequency switching status count.  Positive means "enabled", negative
 429  * means "disabled" and 0 means "not decided yet".
 430  */
 431 static int cpufreq_fast_switch_count;
 432 static DEFINE_MUTEX(cpufreq_fast_switch_lock);
 433
 434 static void cpufreq_list_transition_notifiers(void)
 435 {
 436         struct notifier_block *nb;
 437
 438         pr_info("Registered transition notifiers:\n");
 439
 440         mutex_lock(&cpufreq_transition_notifier_list.mutex);
 441
 442         for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
 443                 pr_info("%pF\n", nb->notifier_call);
 444
 445         mutex_unlock(&cpufreq_transition_notifier_list.mutex);
 446 }
 447
 448 /**
 449  * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
 450  * @policy: cpufreq policy to enable fast frequency switching for.
 451  *
 452  * Try to enable fast frequency switching for @policy.
 453  *
 454  * The attempt will fail if there is at least one transition notifier registered
 455  * at this point, as fast frequency switching is quite fundamentally at odds
 456  * with transition notifiers.  Thus if successful, it will make registration of
 457  * transition notifiers fail going forward.
 458  */
 459 void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
 460 {
 461         lockdep_assert_held(&policy->rwsem);
 462
 463         if (!policy->fast_switch_possible)
 464                 return;
 465
 466         mutex_lock(&cpufreq_fast_switch_lock);
 467         if (cpufreq_fast_switch_count >= 0) {
 468                 cpufreq_fast_switch_count++;
 469                 policy->fast_switch_enabled = true;
 470         } else {
 471                 pr_warn("CPU%u: Fast frequency switching not enabled\n",
 472                         policy->cpu);
 473                 cpufreq_list_transition_notifiers();
 474         }
 475         mutex_unlock(&cpufreq_fast_switch_lock);
 476 }
 477 EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
 478
 479 /**
 480  * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
 481  * @policy: cpufreq policy to disable fast frequency switching for.
 482  */
 483 void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
 484 {
 485         mutex_lock(&cpufreq_fast_switch_lock);
 486         if (policy->fast_switch_enabled) {
 487                 policy->fast_switch_enabled = false;
 488                 if (!WARN_ON(cpufreq_fast_switch_count <= 0))
 489                         cpufreq_fast_switch_count--;
 490         }
 491         mutex_unlock(&cpufreq_fast_switch_lock);
 492 }
 493 EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
 494
 495 /**
 496  * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
 497  * one.
 498  * @target_freq: target frequency to resolve.
 499  *
 500  * The target to driver frequency mapping is cached in the policy.
 501  *
 502  * Return: Lowest driver-supported frequency greater than or equal to the
 503  * given target_freq, subject to policy (min/max) and driver limitations.
 504  */
 505 unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
 506                                          unsigned int target_freq)
 507 {
 508         target_freq = clamp_val(target_freq, policy->min, policy->max);
 509         policy->cached_target_freq = target_freq;
 510
 511         if (cpufreq_driver->target_index) {
 512                 int idx;
 513
 514                 idx = cpufreq_frequency_table_target(policy, target_freq,
 515                                                      CPUFREQ_RELATION_L);
 516                 policy->cached_resolved_idx = idx;
 517                 return policy->freq_table[idx].frequency;
 518         }
 519
 520         if (cpufreq_driver->resolve_freq)
 521                 return cpufreq_driver->resolve_freq(policy, target_freq);
 522
 523         return target_freq;
 524 }
 525 EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
 526
 527 /*********************************************************************
 528  *                          SYSFS INTERFACE                          *
 529  *********************************************************************/
 530 static ssize_t show_boost(struct kobject *kobj,
 531                                  struct attribute *attr, char *buf)
 532 {
 533         return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
 534 }
 535
 536 static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
 537                                   const char *buf, size_t count)
 538 {
 539         int ret, enable;
 540
 541         ret = sscanf(buf, "%d", &enable);
 542         if (ret != 1 || enable < 0 || enable > 1)
 543                 return -EINVAL;
 544
 545         if (cpufreq_boost_trigger_state(enable)) {
 546                 pr_err("%s: Cannot %s BOOST!\n",
 547                        __func__, enable ? "enable" : "disable");
 548                 return -EINVAL;
 549         }
 550
 551         pr_debug("%s: cpufreq BOOST %s\n",
 552                  __func__, enable ? "enabled" : "disabled");
 553
 554         return count;
 555 }
 556 define_one_global_rw(boost);
 557
 558 static struct cpufreq_governor *find_governor(const char *str_governor)
 559 {
 560         struct cpufreq_governor *t;
 561
 562         for_each_governor(t)
 563                 if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
 564                         return t;
 565
 566         return NULL;
 567 }
 568
 569 /**
 570  * cpufreq_parse_governor - parse a governor string
 571  */
 572 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
 573                                 struct cpufreq_governor **governor)
 574 {
 575         int err = -EINVAL;
 576
 577         if (cpufreq_driver->setpolicy) {
 578                 if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 579                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 580                         err = 0;
 581                 } else if (!strncasecmp(str_governor, "powersave",
 582                                                 CPUFREQ_NAME_LEN)) {
 583                         *policy = CPUFREQ_POLICY_POWERSAVE;
 584                         err = 0;
 585                 }
 586         } else {
 587                 struct cpufreq_governor *t;
 588
 589                 mutex_lock(&cpufreq_governor_mutex);
 590
 591                 t = find_governor(str_governor);
 592
 593                 if (t == NULL) {
 594                         int ret;
 595
 596                         mutex_unlock(&cpufreq_governor_mutex);
 597                         ret = request_module("cpufreq_%s", str_governor);
 598                         mutex_lock(&cpufreq_governor_mutex);
 599
 600                         if (ret == 0)
 601                                 t = find_governor(str_governor);
 602                 }
 603
 604                 if (t != NULL) {
 605                         *governor = t;
 606                         err = 0;
 607                 }
 608
 609                 mutex_unlock(&cpufreq_governor_mutex);
 610         }
 611         return err;
 612 }
 613
 614 /**
 615  * cpufreq_per_cpu_attr_read() / show_##file_name() -
 616  * print out cpufreq information
 617  *
 618  * Write out information from cpufreq_driver->policy[cpu]; object must be
 619  * "unsigned int".
 620  */
 621
 622 #define show_one(file_name, object)                     \
 623 static ssize_t show_##file_name                         \
 624 (struct cpufreq_policy *policy, char *buf)              \
 625 {                                                       \
 626         return sprintf(buf, "%u\n", policy->object);    \
 627 }
 628
 629 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 630 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 631 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
 632 show_one(scaling_min_freq, min);
 633 show_one(scaling_max_freq, max);
 634
 635 static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
 636 {
 637         ssize_t ret;
 638
 639         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
 640                 ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
 641         else
 642                 ret = sprintf(buf, "%u\n", policy->cur);
 643         return ret;
 644 }
 645
 646 static int cpufreq_set_policy(struct cpufreq_policy *policy,
 647                                 struct cpufreq_policy *new_policy);
 648
 649 /**
 650  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 651  */
 652 #define store_one(file_name, object)                    \
 653 static ssize_t store_##file_name                                        \
 654 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
 655 {                                                                       \
 656         int ret, temp;                                                  \
 657         struct cpufreq_policy new_policy;                               \
 658                                                                         \
 659         memcpy(&new_policy, policy, sizeof(*policy));                   \
 660                                                                         \
 661         ret = sscanf(buf, "%u", &new_policy.object);                    \
 662         if (ret != 1)                                                   \
 663                 return -EINVAL;                                         \
 664                                                                         \
 665         temp = new_policy.object;                                       \
 666         ret = cpufreq_set_policy(policy, &new_policy);          \
 667         if (!ret)                                                       \
 668                 policy->user_policy.object = temp;                      \
 669                                                                         \
 670         return ret ? ret : count;                                       \
 671 }
 672
 673 store_one(scaling_min_freq, min);
 674 store_one(scaling_max_freq, max);
 675
 676 /**
 677  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 678  */
 679 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
 680                                         char *buf)
 681 {
 682         unsigned int cur_freq = __cpufreq_get(policy);
 683         if (!cur_freq)
 684                 return sprintf(buf, "<unknown>");
 685         return sprintf(buf, "%u\n", cur_freq);
 686 }
 687
 688 /**
 689  * show_scaling_governor - show the current policy for the specified CPU
 690  */
 691 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
 692 {
 693         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
 694                 return sprintf(buf, "powersave\n");
 695         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 696                 return sprintf(buf, "performance\n");
 697         else if (policy->governor)
 698                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
 699                                 policy->governor->name);
 700         return -EINVAL;
 701 }
 702
 703 /**
 704  * store_scaling_governor - store policy for the specified CPU
 705  */
 706 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
 707                                         const char *buf, size_t count)
 708 {
 709         int ret;
 710         char    str_governor[16];
 711         struct cpufreq_policy new_policy;
 712
 713         memcpy(&new_policy, policy, sizeof(*policy));
 714
 715         ret = sscanf(buf, "%15s", str_governor);
 716         if (ret != 1)
 717                 return -EINVAL;
 718
 719         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
 720                                                 &new_policy.governor))
 721                 return -EINVAL;
 722
 723         ret = cpufreq_set_policy(policy, &new_policy);
 724         return ret ? ret : count;
 725 }
 726
 727 /**
 728  * show_scaling_driver - show the cpufreq driver currently loaded
 729  */
 730 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
 731 {
 732         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
 733 }
 734
 735 /**
 736  * show_scaling_available_governors - show the available CPUfreq governors
 737  */
 738 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
 739                                                 char *buf)
 740 {
 741         ssize_t i = 0;
 742         struct cpufreq_governor *t;
 743
 744         if (!has_target()) {
 745                 i += sprintf(buf, "performance powersave");
 746                 goto out;
 747         }
 748
 749         for_each_governor(t) {
 750                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
 751                     - (CPUFREQ_NAME_LEN + 2)))
 752                         goto out;
 753                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
 754         }
 755 out:
 756         i += sprintf(&buf[i], "\n");
 757         return i;
 758 }
 759
 760 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
 761 {
 762         ssize_t i = 0;
 763         unsigned int cpu;
 764
 765         for_each_cpu(cpu, mask) {
 766                 if (i)
 767                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 768                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 769                 if (i >= (PAGE_SIZE - 5))
 770                         break;
 771         }
 772         i += sprintf(&buf[i], "\n");
 773         return i;
 774 }
 775 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
 776
 777 /**
 778  * show_related_cpus - show the CPUs affected by each transition even if
 779  * hw coordination is in use
 780  */
 781 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
 782 {
 783         return cpufreq_show_cpus(policy->related_cpus, buf);
 784 }
 785
 786 /**
 787  * show_affected_cpus - show the CPUs affected by each transition
 788  */
 789 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
 790 {
 791         return cpufreq_show_cpus(policy->cpus, buf);
 792 }
 793
 794 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
 795                                         const char *buf, size_t count)
 796 {
 797         unsigned int freq = 0;
 798         unsigned int ret;
 799
 800         if (!policy->governor || !policy->governor->store_setspeed)
 801                 return -EINVAL;
 802
 803         ret = sscanf(buf, "%u", &freq);
 804         if (ret != 1)
 805                 return -EINVAL;
 806
 807         policy->governor->store_setspeed(policy, freq);
 808
 809         return count;
 810 }
 811
 812 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
 813 {
 814         if (!policy->governor || !policy->governor->show_setspeed)
 815                 return sprintf(buf, "<unsupported>\n");
 816
 817         return policy->governor->show_setspeed(policy, buf);
 818 }
 819
 820 /**
 821  * show_bios_limit - show the current cpufreq HW/BIOS limitation
 822  */
 823 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
 824 {
 825         unsigned int limit;
 826         int ret;
 827         if (cpufreq_driver->bios_limit) {
 828                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
 829                 if (!ret)
 830                         return sprintf(buf, "%u\n", limit);
 831         }
 832         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
 833 }
 834
 835 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
 836 cpufreq_freq_attr_ro(cpuinfo_min_freq);
 837 cpufreq_freq_attr_ro(cpuinfo_max_freq);
 838 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
 839 cpufreq_freq_attr_ro(scaling_available_governors);
 840 cpufreq_freq_attr_ro(scaling_driver);
 841 cpufreq_freq_attr_ro(scaling_cur_freq);
 842 cpufreq_freq_attr_ro(bios_limit);
 843 cpufreq_freq_attr_ro(related_cpus);
 844 cpufreq_freq_attr_ro(affected_cpus);
 845 cpufreq_freq_attr_rw(scaling_min_freq);
 846 cpufreq_freq_attr_rw(scaling_max_freq);
 847 cpufreq_freq_attr_rw(scaling_governor);
 848 cpufreq_freq_attr_rw(scaling_setspeed);
 849
 850 static struct attribute *default_attrs[] = {
 851         &cpuinfo_min_freq.attr,
 852         &cpuinfo_max_freq.attr,
 853         &cpuinfo_transition_latency.attr,
 854         &scaling_min_freq.attr,
 855         &scaling_max_freq.attr,
 856         &affected_cpus.attr,
 857         &related_cpus.attr,
 858         &scaling_governor.attr,
 859         &scaling_driver.attr,
 860         &scaling_available_governors.attr,
 861         &scaling_setspeed.attr,
 862         NULL
 863 };
 864
 865 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
 866 #define to_attr(a) container_of(a, struct freq_attr, attr)
 867
 868 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 869 {
 870         struct cpufreq_policy *policy = to_policy(kobj);
 871         struct freq_attr *fattr = to_attr(attr);
 872         ssize_t ret;
 873
 874         down_read(&policy->rwsem);
 875         ret = fattr->show(policy, buf);
 876         up_read(&policy->rwsem);
 877
 878         return ret;
 879 }
 880
 881 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 882                      const char *buf, size_t count)
 883 {
 884         struct cpufreq_policy *policy = to_policy(kobj);
 885         struct freq_attr *fattr = to_attr(attr);
 886         ssize_t ret = -EINVAL;
 887
 888         get_online_cpus();
 889
 890         if (cpu_online(policy->cpu)) {
 891                 down_write(&policy->rwsem);
 892                 ret = fattr->store(policy, buf, count);
 893                 up_write(&policy->rwsem);
 894         }
 895
 896         put_online_cpus();
 897
 898         return ret;
 899 }
 900
 901 static void cpufreq_sysfs_release(struct kobject *kobj)
 902 {
 903         struct cpufreq_policy *policy = to_policy(kobj);
 904         pr_debug("last reference is dropped\n");
 905         complete(&policy->kobj_unregister);
 906 }
 907
 908 static const struct sysfs_ops sysfs_ops = {
 909         .show   = show,
 910         .store  = store,
 911 };
 912
 913 static struct kobj_type ktype_cpufreq = {
 914         .sysfs_ops      = &sysfs_ops,
 915         .default_attrs  = default_attrs,
 916         .release        = cpufreq_sysfs_release,
 917 };
 918
 919 static int add_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
 920 {
 921         struct device *cpu_dev;
 922
 923         pr_debug("%s: Adding symlink for CPU: %u\n", __func__, cpu);
 924
 925         if (!policy)
 926                 return 0;
 927
 928         cpu_dev = get_cpu_device(cpu);
 929         if (WARN_ON(!cpu_dev))
 930                 return 0;
 931
 932         return sysfs_create_link(&cpu_dev->kobj, &policy->kobj, "cpufreq");
 933 }
 934
 935 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu)
 936 {
 937         struct device *cpu_dev;
 938
 939         pr_debug("%s: Removing symlink for CPU: %u\n", __func__, cpu);
 940
 941         cpu_dev = get_cpu_device(cpu);
 942         if (WARN_ON(!cpu_dev))
 943                 return;
 944
 945         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
 946 }
 947
 948 /* Add/remove symlinks for all related CPUs */
 949 static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
 950 {
 951         unsigned int j;
 952         int ret = 0;
 953
 954         /* Some related CPUs might not be present (physically hotplugged) */
 955         for_each_cpu(j, policy->real_cpus) {
 956                 ret = add_cpu_dev_symlink(policy, j);
 957                 if (ret)
 958                         break;
 959         }
 960
 961         return ret;
 962 }
 963
 964 static void cpufreq_remove_dev_symlink(struct cpufreq_policy *policy)
 965 {
 966         unsigned int j;
 967
 968         /* Some related CPUs might not be present (physically hotplugged) */
 969         for_each_cpu(j, policy->real_cpus)
 970                 remove_cpu_dev_symlink(policy, j);
 971 }
 972
 973 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
 974 {
 975         struct freq_attr **drv_attr;
 976         int ret = 0;
 977
 978         /* set up files for this cpu device */
 979         drv_attr = cpufreq_driver->attr;
 980         while (drv_attr && *drv_attr) {
 981                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 982                 if (ret)
 983                         return ret;
 984                 drv_attr++;
 985         }
 986         if (cpufreq_driver->get) {
 987                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 988                 if (ret)
 989                         return ret;
 990         }
 991
 992         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 993         if (ret)
 994                 return ret;
 995
 996         if (cpufreq_driver->bios_limit) {
 997                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
 998                 if (ret)
 999                         return ret;
1000         }
1001
1002         return cpufreq_add_dev_symlink(policy);
1003 }
1004
1005 __weak struct cpufreq_governor *cpufreq_default_governor(void)
1006 {
1007         return NULL;
1008 }
1009
1010 static int cpufreq_init_policy(struct cpufreq_policy *policy)
1011 {
1012         struct cpufreq_governor *gov = NULL;
1013         struct cpufreq_policy new_policy;
1014
1015         memcpy(&new_policy, policy, sizeof(*policy));
1016
1017         /* Update governor of new_policy to the governor used before hotplug */
1018         gov = find_governor(policy->last_governor);
1019         if (gov) {
1020                 pr_debug("Restoring governor %s for cpu %d\n",
1021                                 policy->governor->name, policy->cpu);
1022         } else {
1023                 gov = cpufreq_default_governor();
1024                 if (!gov)
1025                         return -ENODATA;
1026         }
1027
1028         new_policy.governor = gov;
1029
1030         /* Use the default policy if there is no last_policy. */
1031         if (cpufreq_driver->setpolicy) {
1032                 if (policy->last_policy)
1033                         new_policy.policy = policy->last_policy;
1034                 else
1035                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1036                                                NULL);
1037         }
1038         /* set default policy */
1039         return cpufreq_set_policy(policy, &new_policy);
1040 }
1041
1042 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1043 {
1044         int ret = 0;
1045
1046         /* Has this CPU been taken care of already? */
1047         if (cpumask_test_cpu(cpu, policy->cpus))
1048                 return 0;
1049
1050         down_write(&policy->rwsem);
1051         if (has_target())
1052                 cpufreq_stop_governor(policy);
1053
1054         cpumask_set_cpu(cpu, policy->cpus);
1055
1056         if (has_target()) {
1057                 ret = cpufreq_start_governor(policy);
1058                 if (ret)
1059                         pr_err("%s: Failed to start governor\n", __func__);
1060         }
1061         up_write(&policy->rwsem);
1062         return ret;
1063 }
1064
1065 static void handle_update(struct work_struct *work)
1066 {
1067         struct cpufreq_policy *policy =
1068                 container_of(work, struct cpufreq_policy, update);
1069         unsigned int cpu = policy->cpu;
1070         pr_debug("handle_update for cpu %u called\n", cpu);
1071         cpufreq_update_policy(cpu);
1072 }
1073
1074 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1075 {
1076         struct device *dev = get_cpu_device(cpu);
1077         struct cpufreq_policy *policy;
1078         int ret;
1079
1080         if (WARN_ON(!dev))
1081                 return NULL;
1082
1083         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1084         if (!policy)
1085                 return NULL;
1086
1087         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1088                 goto err_free_policy;
1089
1090         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1091                 goto err_free_cpumask;
1092
1093         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1094                 goto err_free_rcpumask;
1095
1096         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1097                                    cpufreq_global_kobject, "policy%u", cpu);
1098         if (ret) {
1099                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1100                 goto err_free_real_cpus;
1101         }
1102
1103         INIT_LIST_HEAD(&policy->policy_list);
1104         init_rwsem(&policy->rwsem);
1105         spin_lock_init(&policy->transition_lock);
1106         init_waitqueue_head(&policy->transition_wait);
1107         init_completion(&policy->kobj_unregister);
1108         INIT_WORK(&policy->update, handle_update);
1109
1110         policy->cpu = cpu;
1111         return policy;
1112
1113 err_free_real_cpus:
1114         free_cpumask_var(policy->real_cpus);
1115 err_free_rcpumask:
1116         free_cpumask_var(policy->related_cpus);
1117 err_free_cpumask:
1118         free_cpumask_var(policy->cpus);
1119 err_free_policy:
1120         kfree(policy);
1121
1122         return NULL;
1123 }
1124
1125 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy, bool notify)
1126 {
1127         struct kobject *kobj;
1128         struct completion *cmp;
1129
1130         if (notify)
1131                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1132                                              CPUFREQ_REMOVE_POLICY, policy);
1133
1134         down_write(&policy->rwsem);
1135         cpufreq_stats_free_table(policy);
1136         cpufreq_remove_dev_symlink(policy);
1137         kobj = &policy->kobj;
1138         cmp = &policy->kobj_unregister;
1139         up_write(&policy->rwsem);
1140         kobject_put(kobj);
1141
1142         /*
1143          * We need to make sure that the underlying kobj is
1144          * actually not referenced anymore by anybody before we
1145          * proceed with unloading.
1146          */
1147         pr_debug("waiting for dropping of refcount\n");
1148         wait_for_completion(cmp);
1149         pr_debug("wait complete\n");
1150 }
1151
1152 static void cpufreq_policy_free(struct cpufreq_policy *policy, bool notify)
1153 {
1154         unsigned long flags;
1155         int cpu;
1156
1157         /* Remove policy from list */
1158         write_lock_irqsave(&cpufreq_driver_lock, flags);
1159         list_del(&policy->policy_list);
1160
1161         for_each_cpu(cpu, policy->related_cpus)
1162                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1163         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1164
1165         cpufreq_policy_put_kobj(policy, notify);
1166         free_cpumask_var(policy->real_cpus);
1167         free_cpumask_var(policy->related_cpus);
1168         free_cpumask_var(policy->cpus);
1169         kfree(policy);
1170 }
1171
1172 static int cpufreq_online(unsigned int cpu)
1173 {
1174         struct cpufreq_policy *policy;
1175         bool new_policy;
1176         unsigned long flags;
1177         unsigned int j;
1178         int ret;
1179
1180         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1181
1182         /* Check if this CPU already has a policy to manage it */
1183         policy = per_cpu(cpufreq_cpu_data, cpu);
1184         if (policy) {
1185                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1186                 if (!policy_is_inactive(policy))
1187                         return cpufreq_add_policy_cpu(policy, cpu);
1188
1189                 /* This is the only online CPU for the policy.  Start over. */
1190                 new_policy = false;
1191                 down_write(&policy->rwsem);
1192                 policy->cpu = cpu;
1193                 policy->governor = NULL;
1194                 up_write(&policy->rwsem);
1195         } else {
1196                 new_policy = true;
1197                 policy = cpufreq_policy_alloc(cpu);
1198                 if (!policy)
1199                         return -ENOMEM;
1200         }
1201
1202         cpumask_copy(policy->cpus, cpumask_of(cpu));
1203
1204         /* call driver. From then on the cpufreq must be able
1205          * to accept all calls to ->verify and ->setpolicy for this CPU
1206          */
1207         ret = cpufreq_driver->init(policy);
1208         if (ret) {
1209                 pr_debug("initialization failed\n");
1210                 goto out_free_policy;
1211         }
1212
1213         down_write(&policy->rwsem);
1214
1215         if (new_policy) {
1216                 /* related_cpus should at least include policy->cpus. */
1217                 cpumask_copy(policy->related_cpus, policy->cpus);
1218                 /* Remember CPUs present at the policy creation time. */
1219                 cpumask_and(policy->real_cpus, policy->cpus, cpu_present_mask);
1220         }
1221
1222         /*
1223          * affected cpus must always be the one, which are online. We aren't
1224          * managing offline cpus here.
1225          */
1226         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1227
1228         if (new_policy) {
1229                 policy->user_policy.min = policy->min;
1230                 policy->user_policy.max = policy->max;
1231
1232                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1233                 for_each_cpu(j, policy->related_cpus)
1234                         per_cpu(cpufreq_cpu_data, j) = policy;
1235                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1236         }
1237
1238         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1239                 policy->cur = cpufreq_driver->get(policy->cpu);
1240                 if (!policy->cur) {
1241                         pr_err("%s: ->get() failed\n", __func__);
1242                         goto out_exit_policy;
1243                 }
1244         }
1245
1246         /*
1247          * Sometimes boot loaders set CPU frequency to a value outside of
1248          * frequency table present with cpufreq core. In such cases CPU might be
1249          * unstable if it has to run on that frequency for long duration of time
1250          * and so its better to set it to a frequency which is specified in
1251          * freq-table. This also makes cpufreq stats inconsistent as
1252          * cpufreq-stats would fail to register because current frequency of CPU
1253          * isn't found in freq-table.
1254          *
1255          * Because we don't want this change to effect boot process badly, we go
1256          * for the next freq which is >= policy->cur ('cur' must be set by now,
1257          * otherwise we will end up setting freq to lowest of the table as 'cur'
1258          * is initialized to zero).
1259          *
1260          * We are passing target-freq as "policy->cur - 1" otherwise
1261          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1262          * equal to target-freq.
1263          */
1264         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1265             && has_target()) {
1266                 /* Are we running at unknown frequency ? */
1267                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1268                 if (ret == -EINVAL) {
1269                         /* Warn user and fix it */
1270                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1271                                 __func__, policy->cpu, policy->cur);
1272                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1273                                 CPUFREQ_RELATION_L);
1274
1275                         /*
1276                          * Reaching here after boot in a few seconds may not
1277                          * mean that system will remain stable at "unknown"
1278                          * frequency for longer duration. Hence, a BUG_ON().
1279                          */
1280                         BUG_ON(ret);
1281                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1282                                 __func__, policy->cpu, policy->cur);
1283                 }
1284         }
1285
1286         if (new_policy) {
1287                 ret = cpufreq_add_dev_interface(policy);
1288                 if (ret)
1289                         goto out_exit_policy;
1290
1291                 cpufreq_stats_create_table(policy);
1292                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1293                                 CPUFREQ_CREATE_POLICY, policy);
1294
1295                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1296                 list_add(&policy->policy_list, &cpufreq_policy_list);
1297                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1298         }
1299
1300         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1301                                      CPUFREQ_START, policy);
1302
1303         ret = cpufreq_init_policy(policy);
1304         if (ret) {
1305                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1306                        __func__, cpu, ret);
1307                 /* cpufreq_policy_free() will notify based on this */
1308                 new_policy = false;
1309                 goto out_exit_policy;
1310         }
1311
1312         up_write(&policy->rwsem);
1313
1314         kobject_uevent(&policy->kobj, KOBJ_ADD);
1315
1316         /* Callback for handling stuff after policy is ready */
1317         if (cpufreq_driver->ready)
1318                 cpufreq_driver->ready(policy);
1319
1320         pr_debug("initialization complete\n");
1321
1322         return 0;
1323
1324 out_exit_policy:
1325         up_write(&policy->rwsem);
1326
1327         if (cpufreq_driver->exit)
1328                 cpufreq_driver->exit(policy);
1329 out_free_policy:
1330         cpufreq_policy_free(policy, !new_policy);
1331         return ret;
1332 }
1333
1334 /**
1335  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1336  * @dev: CPU device.
1337  * @sif: Subsystem interface structure pointer (not used)
1338  */
1339 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1340 {
1341         struct cpufreq_policy *policy;
1342         unsigned cpu = dev->id;
1343
1344         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1345
1346         if (cpu_online(cpu))
1347                 return cpufreq_online(cpu);
1348
1349         /*
1350          * A hotplug notifier will follow and we will handle it as CPU online
1351          * then.  For now, just create the sysfs link, unless there is no policy
1352          * or the link is already present.
1353          */
1354         policy = per_cpu(cpufreq_cpu_data, cpu);
1355         if (!policy || cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1356                 return 0;
1357
1358         return add_cpu_dev_symlink(policy, cpu);
1359 }
1360
1361 static int cpufreq_offline(unsigned int cpu)
1362 {
1363         struct cpufreq_policy *policy;
1364         int ret;
1365
1366         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1367
1368         policy = cpufreq_cpu_get_raw(cpu);
1369         if (!policy) {
1370                 pr_debug("%s: No cpu_data found\n", __func__);
1371                 return 0;
1372         }
1373
1374         down_write(&policy->rwsem);
1375         if (has_target())
1376                 cpufreq_stop_governor(policy);
1377
1378         cpumask_clear_cpu(cpu, policy->cpus);
1379
1380         if (policy_is_inactive(policy)) {
1381                 if (has_target())
1382                         strncpy(policy->last_governor, policy->governor->name,
1383                                 CPUFREQ_NAME_LEN);
1384                 else
1385                         policy->last_policy = policy->policy;
1386         } else if (cpu == policy->cpu) {
1387                 /* Nominate new CPU */
1388                 policy->cpu = cpumask_any(policy->cpus);
1389         }
1390
1391         /* Start governor again for active policy */
1392         if (!policy_is_inactive(policy)) {
1393                 if (has_target()) {
1394                         ret = cpufreq_start_governor(policy);
1395                         if (ret)
1396                                 pr_err("%s: Failed to start governor\n", __func__);
1397                 }
1398
1399                 goto unlock;
1400         }
1401
1402         if (cpufreq_driver->stop_cpu)
1403                 cpufreq_driver->stop_cpu(policy);
1404
1405         if (has_target())
1406                 cpufreq_exit_governor(policy);
1407
1408         /*
1409          * Perform the ->exit() even during light-weight tear-down,
1410          * since this is a core component, and is essential for the
1411          * subsequent light-weight ->init() to succeed.
1412          */
1413         if (cpufreq_driver->exit) {
1414                 cpufreq_driver->exit(policy);
1415                 policy->freq_table = NULL;
1416         }
1417
1418 unlock:
1419         up_write(&policy->rwsem);
1420         return 0;
1421 }
1422
1423 /**
1424  * cpufreq_remove_dev - remove a CPU device
1425  *
1426  * Removes the cpufreq interface for a CPU device.
1427  */
1428 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1429 {
1430         unsigned int cpu = dev->id;
1431         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1432
1433         if (!policy)
1434                 return;
1435
1436         if (cpu_online(cpu))
1437                 cpufreq_offline(cpu);
1438
1439         cpumask_clear_cpu(cpu, policy->real_cpus);
1440         remove_cpu_dev_symlink(policy, cpu);
1441
1442         if (cpumask_empty(policy->real_cpus))
1443                 cpufreq_policy_free(policy, true);
1444 }
1445
1446 /**
1447  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1448  *      in deep trouble.
1449  *      @policy: policy managing CPUs
1450  *      @new_freq: CPU frequency the CPU actually runs at
1451  *
1452  *      We adjust to current frequency first, and need to clean up later.
1453  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1454  */
1455 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1456                                 unsigned int new_freq)
1457 {
1458         struct cpufreq_freqs freqs;
1459
1460         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1461                  policy->cur, new_freq);
1462
1463         freqs.old = policy->cur;
1464         freqs.new = new_freq;
1465
1466         cpufreq_freq_transition_begin(policy, &freqs);
1467         cpufreq_freq_transition_end(policy, &freqs, 0);
1468 }
1469
1470 /**
1471  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1472  * @cpu: CPU number
1473  *
1474  * This is the last known freq, without actually getting it from the driver.
1475  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1476  */
1477 unsigned int cpufreq_quick_get(unsigned int cpu)
1478 {
1479         struct cpufreq_policy *policy;
1480         unsigned int ret_freq = 0;
1481         unsigned long flags;
1482
1483         read_lock_irqsave(&cpufreq_driver_lock, flags);
1484
1485         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1486                 ret_freq = cpufreq_driver->get(cpu);
1487                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1488                 return ret_freq;
1489         }
1490
1491         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1492
1493         policy = cpufreq_cpu_get(cpu);
1494         if (policy) {
1495                 ret_freq = policy->cur;
1496                 cpufreq_cpu_put(policy);
1497         }
1498
1499         return ret_freq;
1500 }
1501 EXPORT_SYMBOL(cpufreq_quick_get);
1502
1503 /**
1504  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1505  * @cpu: CPU number
1506  *
1507  * Just return the max possible frequency for a given CPU.
1508  */
1509 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1510 {
1511         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1512         unsigned int ret_freq = 0;
1513
1514         if (policy) {
1515                 ret_freq = policy->max;
1516                 cpufreq_cpu_put(policy);
1517         }
1518
1519         return ret_freq;
1520 }
1521 EXPORT_SYMBOL(cpufreq_quick_get_max);
1522
1523 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1524 {
1525         unsigned int ret_freq = 0;
1526
1527         if (!cpufreq_driver->get)
1528                 return ret_freq;
1529
1530         ret_freq = cpufreq_driver->get(policy->cpu);
1531
1532         /*
1533          * Updating inactive policies is invalid, so avoid doing that.  Also
1534          * if fast frequency switching is used with the given policy, the check
1535          * against policy->cur is pointless, so skip it in that case too.
1536          */
1537         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1538                 return ret_freq;
1539
1540         if (ret_freq && policy->cur &&
1541                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1542                 /* verify no discrepancy between actual and
1543                                         saved value exists */
1544                 if (unlikely(ret_freq != policy->cur)) {
1545                         cpufreq_out_of_sync(policy, ret_freq);
1546                         schedule_work(&policy->update);
1547                 }
1548         }
1549
1550         return ret_freq;
1551 }
1552
1553 /**
1554  * cpufreq_get - get the current CPU frequency (in kHz)
1555  * @cpu: CPU number
1556  *
1557  * Get the CPU current (static) CPU frequency
1558  */
1559 unsigned int cpufreq_get(unsigned int cpu)
1560 {
1561         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1562         unsigned int ret_freq = 0;
1563
1564         if (policy) {
1565                 down_read(&policy->rwsem);
1566                 ret_freq = __cpufreq_get(policy);
1567                 up_read(&policy->rwsem);
1568
1569                 cpufreq_cpu_put(policy);
1570         }
1571
1572         return ret_freq;
1573 }
1574 EXPORT_SYMBOL(cpufreq_get);
1575
1576 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1577 {
1578         unsigned int new_freq;
1579
1580         new_freq = cpufreq_driver->get(policy->cpu);
1581         if (!new_freq)
1582                 return 0;
1583
1584         if (!policy->cur) {
1585                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1586                 policy->cur = new_freq;
1587         } else if (policy->cur != new_freq && has_target()) {
1588                 cpufreq_out_of_sync(policy, new_freq);
1589         }
1590
1591         return new_freq;
1592 }
1593
1594 static struct subsys_interface cpufreq_interface = {
1595         .name           = "cpufreq",
1596         .subsys         = &cpu_subsys,
1597         .add_dev        = cpufreq_add_dev,
1598         .remove_dev     = cpufreq_remove_dev,
1599 };
1600
1601 /*
1602  * In case platform wants some specific frequency to be configured
1603  * during suspend..
1604  */
1605 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1606 {
1607         int ret;
1608
1609         if (!policy->suspend_freq) {
1610                 pr_debug("%s: suspend_freq not defined\n", __func__);
1611                 return 0;
1612         }
1613
1614         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1615                         policy->suspend_freq);
1616
1617         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1618                         CPUFREQ_RELATION_H);
1619         if (ret)
1620                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1621                                 __func__, policy->suspend_freq, ret);
1622
1623         return ret;
1624 }
1625 EXPORT_SYMBOL(cpufreq_generic_suspend);
1626
1627 /**
1628  * cpufreq_suspend() - Suspend CPUFreq governors
1629  *
1630  * Called during system wide Suspend/Hibernate cycles for suspending governors
1631  * as some platforms can't change frequency after this point in suspend cycle.
1632  * Because some of the devices (like: i2c, regulators, etc) they use for
1633  * changing frequency are suspended quickly after this point.
1634  */
1635 void cpufreq_suspend(void)
1636 {
1637         struct cpufreq_policy *policy;
1638
1639         if (!cpufreq_driver)
1640                 return;
1641
1642         if (!has_target() && !cpufreq_driver->suspend)
1643                 goto suspend;
1644
1645         pr_debug("%s: Suspending Governors\n", __func__);
1646
1647         for_each_active_policy(policy) {
1648                 if (has_target()) {
1649                         down_write(&policy->rwsem);
1650                         cpufreq_stop_governor(policy);
1651                         up_write(&policy->rwsem);
1652                 }
1653
1654                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1655                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1656                                 policy);
1657         }
1658
1659 suspend:
1660         cpufreq_suspended = true;
1661 }
1662
1663 /**
1664  * cpufreq_resume() - Resume CPUFreq governors
1665  *
1666  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1667  * are suspended with cpufreq_suspend().
1668  */
1669 void cpufreq_resume(void)
1670 {
1671         struct cpufreq_policy *policy;
1672         int ret;
1673
1674         if (!cpufreq_driver)
1675                 return;
1676
1677         cpufreq_suspended = false;
1678
1679         if (!has_target() && !cpufreq_driver->resume)
1680                 return;
1681
1682         pr_debug("%s: Resuming Governors\n", __func__);
1683
1684         for_each_active_policy(policy) {
1685                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1686                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1687                                 policy);
1688                 } else if (has_target()) {
1689                         down_write(&policy->rwsem);
1690                         ret = cpufreq_start_governor(policy);
1691                         up_write(&policy->rwsem);
1692
1693                         if (ret)
1694                                 pr_err("%s: Failed to start governor for policy: %p\n",
1695                                        __func__, policy);
1696                 }
1697         }
1698 }
1699
1700 /**
1701  *      cpufreq_get_current_driver - return current driver's name
1702  *
1703  *      Return the name string of the currently loaded cpufreq driver
1704  *      or NULL, if none.
1705  */
1706 const char *cpufreq_get_current_driver(void)
1707 {
1708         if (cpufreq_driver)
1709                 return cpufreq_driver->name;
1710
1711         return NULL;
1712 }
1713 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1714
1715 /**
1716  *      cpufreq_get_driver_data - return current driver data
1717  *
1718  *      Return the private data of the currently loaded cpufreq
1719  *      driver, or NULL if no cpufreq driver is loaded.
1720  */
1721 void *cpufreq_get_driver_data(void)
1722 {
1723         if (cpufreq_driver)
1724                 return cpufreq_driver->driver_data;
1725
1726         return NULL;
1727 }
1728 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1729
1730 /*********************************************************************
1731  *                     NOTIFIER LISTS INTERFACE                      *
1732  *********************************************************************/
1733
1734 /**
1735  *      cpufreq_register_notifier - register a driver with cpufreq
1736  *      @nb: notifier function to register
1737  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1738  *
1739  *      Add a driver to one of two lists: either a list of drivers that
1740  *      are notified about clock rate changes (once before and once after
1741  *      the transition), or a list of drivers that are notified about
1742  *      changes in cpufreq policy.
1743  *
1744  *      This function may sleep, and has the same return conditions as
1745  *      blocking_notifier_chain_register.
1746  */
1747 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1748 {
1749         int ret;
1750
1751         if (cpufreq_disabled())
1752                 return -EINVAL;
1753
1754         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1755
1756         switch (list) {
1757         case CPUFREQ_TRANSITION_NOTIFIER:
1758                 mutex_lock(&cpufreq_fast_switch_lock);
1759
1760                 if (cpufreq_fast_switch_count > 0) {
1761                         mutex_unlock(&cpufreq_fast_switch_lock);
1762                         return -EBUSY;
1763                 }
1764                 ret = srcu_notifier_chain_register(
1765                                 &cpufreq_transition_notifier_list, nb);
1766                 if (!ret)
1767                         cpufreq_fast_switch_count--;
1768
1769                 mutex_unlock(&cpufreq_fast_switch_lock);
1770                 break;
1771         case CPUFREQ_POLICY_NOTIFIER:
1772                 ret = blocking_notifier_chain_register(
1773                                 &cpufreq_policy_notifier_list, nb);
1774                 break;
1775         default:
1776                 ret = -EINVAL;
1777         }
1778
1779         return ret;
1780 }
1781 EXPORT_SYMBOL(cpufreq_register_notifier);
1782
1783 /**
1784  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1785  *      @nb: notifier block to be unregistered
1786  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1787  *
1788  *      Remove a driver from the CPU frequency notifier list.
1789  *
1790  *      This function may sleep, and has the same return conditions as
1791  *      blocking_notifier_chain_unregister.
1792  */
1793 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1794 {
1795         int ret;
1796
1797         if (cpufreq_disabled())
1798                 return -EINVAL;
1799
1800         switch (list) {
1801         case CPUFREQ_TRANSITION_NOTIFIER:
1802                 mutex_lock(&cpufreq_fast_switch_lock);
1803
1804                 ret = srcu_notifier_chain_unregister(
1805                                 &cpufreq_transition_notifier_list, nb);
1806                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1807                         cpufreq_fast_switch_count++;
1808
1809                 mutex_unlock(&cpufreq_fast_switch_lock);
1810                 break;
1811         case CPUFREQ_POLICY_NOTIFIER:
1812                 ret = blocking_notifier_chain_unregister(
1813                                 &cpufreq_policy_notifier_list, nb);
1814                 break;
1815         default:
1816                 ret = -EINVAL;
1817         }
1818
1819         return ret;
1820 }
1821 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1822
1823
1824 /*********************************************************************
1825  *                              GOVERNORS                            *
1826  *********************************************************************/
1827
1828 /**
1829  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1830  * @policy: cpufreq policy to switch the frequency for.
1831  * @target_freq: New frequency to set (may be approximate).
1832  *
1833  * Carry out a fast frequency switch without sleeping.
1834  *
1835  * The driver's ->fast_switch() callback invoked by this function must be
1836  * suitable for being called from within RCU-sched read-side critical sections
1837  * and it is expected to select the minimum available frequency greater than or
1838  * equal to @target_freq (CPUFREQ_RELATION_L).
1839  *
1840  * This function must not be called if policy->fast_switch_enabled is unset.
1841  *
1842  * Governors calling this function must guarantee that it will never be invoked
1843  * twice in parallel for the same policy and that it will never be called in
1844  * parallel with either ->target() or ->target_index() for the same policy.
1845  *
1846  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1847  * callback to indicate an error condition, the hardware configuration must be
1848  * preserved.
1849  */
1850 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1851                                         unsigned int target_freq)
1852 {
1853         target_freq = clamp_val(target_freq, policy->min, policy->max);
1854
1855         return cpufreq_driver->fast_switch(policy, target_freq);
1856 }
1857 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1858
1859 /* Must set freqs->new to intermediate frequency */
1860 static int __target_intermediate(struct cpufreq_policy *policy,
1861                                  struct cpufreq_freqs *freqs, int index)
1862 {
1863         int ret;
1864
1865         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1866
1867         /* We don't need to switch to intermediate freq */
1868         if (!freqs->new)
1869                 return 0;
1870
1871         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1872                  __func__, policy->cpu, freqs->old, freqs->new);
1873
1874         cpufreq_freq_transition_begin(policy, freqs);
1875         ret = cpufreq_driver->target_intermediate(policy, index);
1876         cpufreq_freq_transition_end(policy, freqs, ret);
1877
1878         if (ret)
1879                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1880                        __func__, ret);
1881
1882         return ret;
1883 }
1884
1885 static int __target_index(struct cpufreq_policy *policy, int index)
1886 {
1887         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1888         unsigned int intermediate_freq = 0;
1889         unsigned int newfreq = policy->freq_table[index].frequency;
1890         int retval = -EINVAL;
1891         bool notify;
1892
1893         if (newfreq == policy->cur)
1894                 return 0;
1895
1896         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1897         if (notify) {
1898                 /* Handle switching to intermediate frequency */
1899                 if (cpufreq_driver->get_intermediate) {
1900                         retval = __target_intermediate(policy, &freqs, index);
1901                         if (retval)
1902                                 return retval;
1903
1904                         intermediate_freq = freqs.new;
1905                         /* Set old freq to intermediate */
1906                         if (intermediate_freq)
1907                                 freqs.old = freqs.new;
1908                 }
1909
1910                 freqs.new = newfreq;
1911                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1912                          __func__, policy->cpu, freqs.old, freqs.new);
1913
1914                 cpufreq_freq_transition_begin(policy, &freqs);
1915         }
1916
1917         retval = cpufreq_driver->target_index(policy, index);
1918         if (retval)
1919                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1920                        retval);
1921
1922         if (notify) {
1923                 cpufreq_freq_transition_end(policy, &freqs, retval);
1924
1925                 /*
1926                  * Failed after setting to intermediate freq? Driver should have
1927                  * reverted back to initial frequency and so should we. Check
1928                  * here for intermediate_freq instead of get_intermediate, in
1929                  * case we haven't switched to intermediate freq at all.
1930                  */
1931                 if (unlikely(retval && intermediate_freq)) {
1932                         freqs.old = intermediate_freq;
1933                         freqs.new = policy->restore_freq;
1934                         cpufreq_freq_transition_begin(policy, &freqs);
1935                         cpufreq_freq_transition_end(policy, &freqs, 0);
1936                 }
1937         }
1938
1939         return retval;
1940 }
1941
1942 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1943                             unsigned int target_freq,
1944                             unsigned int relation)
1945 {
1946         unsigned int old_target_freq = target_freq;
1947         int index;
1948
1949         if (cpufreq_disabled())
1950                 return -ENODEV;
1951
1952         /* Make sure that target_freq is within supported range */
1953         target_freq = clamp_val(target_freq, policy->min, policy->max);
1954
1955         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1956                  policy->cpu, target_freq, relation, old_target_freq);
1957
1958         /*
1959          * This might look like a redundant call as we are checking it again
1960          * after finding index. But it is left intentionally for cases where
1961          * exactly same freq is called again and so we can save on few function
1962          * calls.
1963          */
1964         if (target_freq == policy->cur)
1965                 return 0;
1966
1967         /* Save last value to restore later on errors */
1968         policy->restore_freq = policy->cur;
1969
1970         if (cpufreq_driver->target)
1971                 return cpufreq_driver->target(policy, target_freq, relation);
1972
1973         if (!cpufreq_driver->target_index)
1974                 return -EINVAL;
1975
1976         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1977
1978         return __target_index(policy, index);
1979 }
1980 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1981
1982 int cpufreq_driver_target(struct cpufreq_policy *policy,
1983                           unsigned int target_freq,
1984                           unsigned int relation)
1985 {
1986         int ret = -EINVAL;
1987
1988         down_write(&policy->rwsem);
1989
1990         ret = __cpufreq_driver_target(policy, target_freq, relation);
1991
1992         up_write(&policy->rwsem);
1993
1994         return ret;
1995 }
1996 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1997
1998 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1999 {
2000         return NULL;
2001 }
2002
2003 static int cpufreq_init_governor(struct cpufreq_policy *policy)
2004 {
2005         int ret;
2006
2007         /* Don't start any governor operations if we are entering suspend */
2008         if (cpufreq_suspended)
2009                 return 0;
2010         /*
2011          * Governor might not be initiated here if ACPI _PPC changed
2012          * notification happened, so check it.
2013          */
2014         if (!policy->governor)
2015                 return -EINVAL;
2016
2017         if (policy->governor->max_transition_latency &&
2018             policy->cpuinfo.transition_latency >
2019             policy->governor->max_transition_latency) {
2020                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
2021
2022                 if (gov) {
2023                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
2024                                 policy->governor->name, gov->name);
2025                         policy->governor = gov;
2026                 } else {
2027                         return -EINVAL;
2028                 }
2029         }
2030
2031         if (!try_module_get(policy->governor->owner))
2032                 return -EINVAL;
2033
2034         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2035
2036         if (policy->governor->init) {
2037                 ret = policy->governor->init(policy);
2038                 if (ret) {
2039                         module_put(policy->governor->owner);
2040                         return ret;
2041                 }
2042         }
2043
2044         return 0;
2045 }
2046
2047 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2048 {
2049         if (cpufreq_suspended || !policy->governor)
2050                 return;
2051
2052         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2053
2054         if (policy->governor->exit)
2055                 policy->governor->exit(policy);
2056
2057         module_put(policy->governor->owner);
2058 }
2059
2060 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2061 {
2062         int ret;
2063
2064         if (cpufreq_suspended)
2065                 return 0;
2066
2067         if (!policy->governor)
2068                 return -EINVAL;
2069
2070         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2071
2072         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2073                 cpufreq_update_current_freq(policy);
2074
2075         if (policy->governor->start) {
2076                 ret = policy->governor->start(policy);
2077                 if (ret)
2078                         return ret;
2079         }
2080
2081         if (policy->governor->limits)
2082                 policy->governor->limits(policy);
2083
2084         return 0;
2085 }
2086
2087 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2088 {
2089         if (cpufreq_suspended || !policy->governor)
2090                 return;
2091
2092         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2093
2094         if (policy->governor->stop)
2095                 policy->governor->stop(policy);
2096 }
2097
2098 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2099 {
2100         if (cpufreq_suspended || !policy->governor)
2101                 return;
2102
2103         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2104
2105         if (policy->governor->limits)
2106                 policy->governor->limits(policy);
2107 }
2108
2109 int cpufreq_register_governor(struct cpufreq_governor *governor)
2110 {
2111         int err;
2112
2113         if (!governor)
2114                 return -EINVAL;
2115
2116         if (cpufreq_disabled())
2117                 return -ENODEV;
2118
2119         mutex_lock(&cpufreq_governor_mutex);
2120
2121         err = -EBUSY;
2122         if (!find_governor(governor->name)) {
2123                 err = 0;
2124                 list_add(&governor->governor_list, &cpufreq_governor_list);
2125         }
2126
2127         mutex_unlock(&cpufreq_governor_mutex);
2128         return err;
2129 }
2130 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2131
2132 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2133 {
2134         struct cpufreq_policy *policy;
2135         unsigned long flags;
2136
2137         if (!governor)
2138                 return;
2139
2140         if (cpufreq_disabled())
2141                 return;
2142
2143         /* clear last_governor for all inactive policies */
2144         read_lock_irqsave(&cpufreq_driver_lock, flags);
2145         for_each_inactive_policy(policy) {
2146                 if (!strcmp(policy->last_governor, governor->name)) {
2147                         policy->governor = NULL;
2148                         strcpy(policy->last_governor, "\0");
2149                 }
2150         }
2151         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2152
2153         mutex_lock(&cpufreq_governor_mutex);
2154         list_del(&governor->governor_list);
2155         mutex_unlock(&cpufreq_governor_mutex);
2156         return;
2157 }
2158 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2159
2160
2161 /*********************************************************************
2162  *                          POLICY INTERFACE                         *
2163  *********************************************************************/
2164
2165 /**
2166  * cpufreq_get_policy - get the current cpufreq_policy
2167  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2168  *      is written
2169  *
2170  * Reads the current cpufreq policy.
2171  */
2172 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2173 {
2174         struct cpufreq_policy *cpu_policy;
2175         if (!policy)
2176                 return -EINVAL;
2177
2178         cpu_policy = cpufreq_cpu_get(cpu);
2179         if (!cpu_policy)
2180                 return -EINVAL;
2181
2182         memcpy(policy, cpu_policy, sizeof(*policy));
2183
2184         cpufreq_cpu_put(cpu_policy);
2185         return 0;
2186 }
2187 EXPORT_SYMBOL(cpufreq_get_policy);
2188
2189 /*
2190  * policy : current policy.
2191  * new_policy: policy to be set.
2192  */
2193 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2194                                 struct cpufreq_policy *new_policy)
2195 {
2196         struct cpufreq_governor *old_gov;
2197         int ret;
2198
2199         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2200                  new_policy->cpu, new_policy->min, new_policy->max);
2201
2202         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2203
2204         /*
2205         * This check works well when we store new min/max freq attributes,
2206         * because new_policy is a copy of policy with one field updated.
2207         */
2208         if (new_policy->min > new_policy->max)
2209                 return -EINVAL;
2210
2211         /* verify the cpu speed can be set within this limit */
2212         ret = cpufreq_driver->verify(new_policy);
2213         if (ret)
2214                 return ret;
2215
2216         /* adjust if necessary - all reasons */
2217         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2218                         CPUFREQ_ADJUST, new_policy);
2219
2220         /*
2221          * verify the cpu speed can be set within this limit, which might be
2222          * different to the first one
2223          */
2224         ret = cpufreq_driver->verify(new_policy);
2225         if (ret)
2226                 return ret;
2227
2228         /* notification of the new policy */
2229         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2230                         CPUFREQ_NOTIFY, new_policy);
2231
2232         policy->min = new_policy->min;
2233         policy->max = new_policy->max;
2234
2235         policy->cached_target_freq = UINT_MAX;
2236
2237         pr_debug("new min and max freqs are %u - %u kHz\n",
2238                  policy->min, policy->max);
2239
2240         if (cpufreq_driver->setpolicy) {
2241                 policy->policy = new_policy->policy;
2242                 pr_debug("setting range\n");
2243                 return cpufreq_driver->setpolicy(new_policy);
2244         }
2245
2246         if (new_policy->governor == policy->governor) {
2247                 pr_debug("cpufreq: governor limits update\n");
2248                 cpufreq_governor_limits(policy);
2249                 return 0;
2250         }
2251
2252         pr_debug("governor switch\n");
2253
2254         /* save old, working values */
2255         old_gov = policy->governor;
2256         /* end old governor */
2257         if (old_gov) {
2258                 cpufreq_stop_governor(policy);
2259                 cpufreq_exit_governor(policy);
2260         }
2261
2262         /* start new governor */
2263         policy->governor = new_policy->governor;
2264         ret = cpufreq_init_governor(policy);
2265         if (!ret) {
2266                 ret = cpufreq_start_governor(policy);
2267                 if (!ret) {
2268                         pr_debug("cpufreq: governor change\n");
2269                         return 0;
2270                 }
2271                 cpufreq_exit_governor(policy);
2272         }
2273
2274         /* new governor failed, so re-start old one */
2275         pr_debug("starting governor %s failed\n", policy->governor->name);
2276         if (old_gov) {
2277                 policy->governor = old_gov;
2278                 if (cpufreq_init_governor(policy))
2279                         policy->governor = NULL;
2280                 else
2281                         cpufreq_start_governor(policy);
2282         }
2283
2284         return ret;
2285 }
2286
2287 /**
2288  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2289  *      @cpu: CPU which shall be re-evaluated
2290  *
2291  *      Useful for policy notifiers which have different necessities
2292  *      at different times.
2293  */
2294 int cpufreq_update_policy(unsigned int cpu)
2295 {
2296         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2297         struct cpufreq_policy new_policy;
2298         int ret;
2299
2300         if (!policy)
2301                 return -ENODEV;
2302
2303         down_write(&policy->rwsem);
2304
2305         pr_debug("updating policy for CPU %u\n", cpu);
2306         memcpy(&new_policy, policy, sizeof(*policy));
2307         new_policy.min = policy->user_policy.min;
2308         new_policy.max = policy->user_policy.max;
2309
2310         /*
2311          * BIOS might change freq behind our back
2312          * -> ask driver for current freq and notify governors about a change
2313          */
2314         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2315                 if (cpufreq_suspended) {
2316                         ret = -EAGAIN;
2317                         goto unlock;
2318                 }
2319                 new_policy.cur = cpufreq_update_current_freq(policy);
2320                 if (WARN_ON(!new_policy.cur)) {
2321                         ret = -EIO;
2322                         goto unlock;
2323                 }
2324         }
2325
2326         ret = cpufreq_set_policy(policy, &new_policy);
2327
2328 unlock:
2329         up_write(&policy->rwsem);
2330
2331         cpufreq_cpu_put(policy);
2332         return ret;
2333 }
2334 EXPORT_SYMBOL(cpufreq_update_policy);
2335
2336 /*********************************************************************
2337  *               BOOST                                               *
2338  *********************************************************************/
2339 static int cpufreq_boost_set_sw(int state)
2340 {
2341         struct cpufreq_policy *policy;
2342         int ret = -EINVAL;
2343
2344         for_each_active_policy(policy) {
2345                 if (!policy->freq_table)
2346                         continue;
2347
2348                 ret = cpufreq_frequency_table_cpuinfo(policy,
2349                                                       policy->freq_table);
2350                 if (ret) {
2351                         pr_err("%s: Policy frequency update failed\n",
2352                                __func__);
2353                         break;
2354                 }
2355
2356                 down_write(&policy->rwsem);
2357                 policy->user_policy.max = policy->max;
2358                 cpufreq_governor_limits(policy);
2359                 up_write(&policy->rwsem);
2360         }
2361
2362         return ret;
2363 }
2364
2365 int cpufreq_boost_trigger_state(int state)
2366 {
2367         unsigned long flags;
2368         int ret = 0;
2369
2370         if (cpufreq_driver->boost_enabled == state)
2371                 return 0;
2372
2373         write_lock_irqsave(&cpufreq_driver_lock, flags);
2374         cpufreq_driver->boost_enabled = state;
2375         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2376
2377         ret = cpufreq_driver->set_boost(state);
2378         if (ret) {
2379                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2380                 cpufreq_driver->boost_enabled = !state;
2381                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2382
2383                 pr_err("%s: Cannot %s BOOST\n",
2384                        __func__, state ? "enable" : "disable");
2385         }
2386
2387         return ret;
2388 }
2389
2390 static bool cpufreq_boost_supported(void)
2391 {
2392         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2393 }
2394
2395 static int create_boost_sysfs_file(void)
2396 {
2397         int ret;
2398
2399         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2400         if (ret)
2401                 pr_err("%s: cannot register global BOOST sysfs file\n",
2402                        __func__);
2403
2404         return ret;
2405 }
2406
2407 static void remove_boost_sysfs_file(void)
2408 {
2409         if (cpufreq_boost_supported())
2410                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2411 }
2412
2413 int cpufreq_enable_boost_support(void)
2414 {
2415         if (!cpufreq_driver)
2416                 return -EINVAL;
2417
2418         if (cpufreq_boost_supported())
2419                 return 0;
2420
2421         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2422
2423         /* This will get removed on driver unregister */
2424         return create_boost_sysfs_file();
2425 }
2426 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2427
2428 int cpufreq_boost_enabled(void)
2429 {
2430         return cpufreq_driver->boost_enabled;
2431 }
2432 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2433
2434 /*********************************************************************
2435  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2436  *********************************************************************/
2437 static enum cpuhp_state hp_online;
2438
2439 /**
2440  * cpufreq_register_driver - register a CPU Frequency driver
2441  * @driver_data: A struct cpufreq_driver containing the values#
2442  * submitted by the CPU Frequency driver.
2443  *
2444  * Registers a CPU Frequency driver to this core code. This code
2445  * returns zero on success, -EEXIST when another driver got here first
2446  * (and isn't unregistered in the meantime).
2447  *
2448  */
2449 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2450 {
2451         unsigned long flags;
2452         int ret;
2453
2454         if (cpufreq_disabled())
2455                 return -ENODEV;
2456
2457         if (!driver_data || !driver_data->verify || !driver_data->init ||
2458             !(driver_data->setpolicy || driver_data->target_index ||
2459                     driver_data->target) ||
2460              (driver_data->setpolicy && (driver_data->target_index ||
2461                     driver_data->target)) ||
2462              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2463                 return -EINVAL;
2464
2465         pr_debug("trying to register driver %s\n", driver_data->name);
2466
2467         /* Protect against concurrent CPU online/offline. */
2468         get_online_cpus();
2469
2470         write_lock_irqsave(&cpufreq_driver_lock, flags);
2471         if (cpufreq_driver) {
2472                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2473                 ret = -EEXIST;
2474                 goto out;
2475         }
2476         cpufreq_driver = driver_data;
2477         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2478
2479         if (driver_data->setpolicy)
2480                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2481
2482         if (cpufreq_boost_supported()) {
2483                 ret = create_boost_sysfs_file();
2484                 if (ret)
2485                         goto err_null_driver;
2486         }
2487
2488         ret = subsys_interface_register(&cpufreq_interface);
2489         if (ret)
2490                 goto err_boost_unreg;
2491
2492         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2493             list_empty(&cpufreq_policy_list)) {
2494                 /* if all ->init() calls failed, unregister */
2495                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2496                          driver_data->name);
2497                 goto err_if_unreg;
2498         }
2499
2500         ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "cpufreq:online",
2501                                         cpufreq_online,
2502                                         cpufreq_offline);
2503         if (ret < 0)
2504                 goto err_if_unreg;
2505         hp_online = ret;
2506         ret = 0;
2507
2508         pr_debug("driver %s up and running\n", driver_data->name);
2509         goto out;
2510
2511 err_if_unreg:
2512         subsys_interface_unregister(&cpufreq_interface);
2513 err_boost_unreg:
2514         remove_boost_sysfs_file();
2515 err_null_driver:
2516         write_lock_irqsave(&cpufreq_driver_lock, flags);
2517         cpufreq_driver = NULL;
2518         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2519 out:
2520         put_online_cpus();
2521         return ret;
2522 }
2523 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2524
2525 /**
2526  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2527  *
2528  * Unregister the current CPUFreq driver. Only call this if you have
2529  * the right to do so, i.e. if you have succeeded in initialising before!
2530  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2531  * currently not initialised.
2532  */
2533 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2534 {
2535         unsigned long flags;
2536
2537         if (!cpufreq_driver || (driver != cpufreq_driver))
2538                 return -EINVAL;
2539
2540         pr_debug("unregistering driver %s\n", driver->name);
2541
2542         /* Protect against concurrent cpu hotplug */
2543         get_online_cpus();
2544         subsys_interface_unregister(&cpufreq_interface);
2545         remove_boost_sysfs_file();
2546         cpuhp_remove_state_nocalls(hp_online);
2547
2548         write_lock_irqsave(&cpufreq_driver_lock, flags);
2549
2550         cpufreq_driver = NULL;
2551
2552         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2553         put_online_cpus();
2554
2555         return 0;
2556 }
2557 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2558
2559 /*
2560  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2561  * or mutexes when secondary CPUs are halted.
2562  */
2563 static struct syscore_ops cpufreq_syscore_ops = {
2564         .shutdown = cpufreq_suspend,
2565 };
2566
2567 struct kobject *cpufreq_global_kobject;
2568 EXPORT_SYMBOL(cpufreq_global_kobject);
2569
2570 static int __init cpufreq_core_init(void)
2571 {
2572         if (cpufreq_disabled())
2573                 return -ENODEV;
2574
2575         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2576         BUG_ON(!cpufreq_global_kobject);
2577
2578         register_syscore_ops(&cpufreq_syscore_ops);
2579
2580         return 0;
2581 }
2582 core_initcall(cpufreq_core_init);