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_nsecs(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 = div_u64(cur_wall_time, NSEC_PER_USEC);
 147
 148         return div_u64(idle_time, NSEC_PER_USEC);
 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
 684         if (cur_freq)
 685                 return sprintf(buf, "%u\n", cur_freq);
 686
 687         return sprintf(buf, "<unknown>\n");
 688 }
 689
 690 /**
 691  * show_scaling_governor - show the current policy for the specified CPU
 692  */
 693 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
 694 {
 695         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
 696                 return sprintf(buf, "powersave\n");
 697         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 698                 return sprintf(buf, "performance\n");
 699         else if (policy->governor)
 700                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
 701                                 policy->governor->name);
 702         return -EINVAL;
 703 }
 704
 705 /**
 706  * store_scaling_governor - store policy for the specified CPU
 707  */
 708 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
 709                                         const char *buf, size_t count)
 710 {
 711         int ret;
 712         char    str_governor[16];
 713         struct cpufreq_policy new_policy;
 714
 715         memcpy(&new_policy, policy, sizeof(*policy));
 716
 717         ret = sscanf(buf, "%15s", str_governor);
 718         if (ret != 1)
 719                 return -EINVAL;
 720
 721         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
 722                                                 &new_policy.governor))
 723                 return -EINVAL;
 724
 725         ret = cpufreq_set_policy(policy, &new_policy);
 726         return ret ? ret : count;
 727 }
 728
 729 /**
 730  * show_scaling_driver - show the cpufreq driver currently loaded
 731  */
 732 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
 733 {
 734         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
 735 }
 736
 737 /**
 738  * show_scaling_available_governors - show the available CPUfreq governors
 739  */
 740 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
 741                                                 char *buf)
 742 {
 743         ssize_t i = 0;
 744         struct cpufreq_governor *t;
 745
 746         if (!has_target()) {
 747                 i += sprintf(buf, "performance powersave");
 748                 goto out;
 749         }
 750
 751         for_each_governor(t) {
 752                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
 753                     - (CPUFREQ_NAME_LEN + 2)))
 754                         goto out;
 755                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
 756         }
 757 out:
 758         i += sprintf(&buf[i], "\n");
 759         return i;
 760 }
 761
 762 ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
 763 {
 764         ssize_t i = 0;
 765         unsigned int cpu;
 766
 767         for_each_cpu(cpu, mask) {
 768                 if (i)
 769                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 770                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 771                 if (i >= (PAGE_SIZE - 5))
 772                         break;
 773         }
 774         i += sprintf(&buf[i], "\n");
 775         return i;
 776 }
 777 EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
 778
 779 /**
 780  * show_related_cpus - show the CPUs affected by each transition even if
 781  * hw coordination is in use
 782  */
 783 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
 784 {
 785         return cpufreq_show_cpus(policy->related_cpus, buf);
 786 }
 787
 788 /**
 789  * show_affected_cpus - show the CPUs affected by each transition
 790  */
 791 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
 792 {
 793         return cpufreq_show_cpus(policy->cpus, buf);
 794 }
 795
 796 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
 797                                         const char *buf, size_t count)
 798 {
 799         unsigned int freq = 0;
 800         unsigned int ret;
 801
 802         if (!policy->governor || !policy->governor->store_setspeed)
 803                 return -EINVAL;
 804
 805         ret = sscanf(buf, "%u", &freq);
 806         if (ret != 1)
 807                 return -EINVAL;
 808
 809         policy->governor->store_setspeed(policy, freq);
 810
 811         return count;
 812 }
 813
 814 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
 815 {
 816         if (!policy->governor || !policy->governor->show_setspeed)
 817                 return sprintf(buf, "<unsupported>\n");
 818
 819         return policy->governor->show_setspeed(policy, buf);
 820 }
 821
 822 /**
 823  * show_bios_limit - show the current cpufreq HW/BIOS limitation
 824  */
 825 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
 826 {
 827         unsigned int limit;
 828         int ret;
 829         if (cpufreq_driver->bios_limit) {
 830                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
 831                 if (!ret)
 832                         return sprintf(buf, "%u\n", limit);
 833         }
 834         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
 835 }
 836
 837 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
 838 cpufreq_freq_attr_ro(cpuinfo_min_freq);
 839 cpufreq_freq_attr_ro(cpuinfo_max_freq);
 840 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
 841 cpufreq_freq_attr_ro(scaling_available_governors);
 842 cpufreq_freq_attr_ro(scaling_driver);
 843 cpufreq_freq_attr_ro(scaling_cur_freq);
 844 cpufreq_freq_attr_ro(bios_limit);
 845 cpufreq_freq_attr_ro(related_cpus);
 846 cpufreq_freq_attr_ro(affected_cpus);
 847 cpufreq_freq_attr_rw(scaling_min_freq);
 848 cpufreq_freq_attr_rw(scaling_max_freq);
 849 cpufreq_freq_attr_rw(scaling_governor);
 850 cpufreq_freq_attr_rw(scaling_setspeed);
 851
 852 static struct attribute *default_attrs[] = {
 853         &cpuinfo_min_freq.attr,
 854         &cpuinfo_max_freq.attr,
 855         &cpuinfo_transition_latency.attr,
 856         &scaling_min_freq.attr,
 857         &scaling_max_freq.attr,
 858         &affected_cpus.attr,
 859         &related_cpus.attr,
 860         &scaling_governor.attr,
 861         &scaling_driver.attr,
 862         &scaling_available_governors.attr,
 863         &scaling_setspeed.attr,
 864         NULL
 865 };
 866
 867 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
 868 #define to_attr(a) container_of(a, struct freq_attr, attr)
 869
 870 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 871 {
 872         struct cpufreq_policy *policy = to_policy(kobj);
 873         struct freq_attr *fattr = to_attr(attr);
 874         ssize_t ret;
 875
 876         down_read(&policy->rwsem);
 877         ret = fattr->show(policy, buf);
 878         up_read(&policy->rwsem);
 879
 880         return ret;
 881 }
 882
 883 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 884                      const char *buf, size_t count)
 885 {
 886         struct cpufreq_policy *policy = to_policy(kobj);
 887         struct freq_attr *fattr = to_attr(attr);
 888         ssize_t ret = -EINVAL;
 889
 890         get_online_cpus();
 891
 892         if (cpu_online(policy->cpu)) {
 893                 down_write(&policy->rwsem);
 894                 ret = fattr->store(policy, buf, count);
 895                 up_write(&policy->rwsem);
 896         }
 897
 898         put_online_cpus();
 899
 900         return ret;
 901 }
 902
 903 static void cpufreq_sysfs_release(struct kobject *kobj)
 904 {
 905         struct cpufreq_policy *policy = to_policy(kobj);
 906         pr_debug("last reference is dropped\n");
 907         complete(&policy->kobj_unregister);
 908 }
 909
 910 static const struct sysfs_ops sysfs_ops = {
 911         .show   = show,
 912         .store  = store,
 913 };
 914
 915 static struct kobj_type ktype_cpufreq = {
 916         .sysfs_ops      = &sysfs_ops,
 917         .default_attrs  = default_attrs,
 918         .release        = cpufreq_sysfs_release,
 919 };
 920
 921 static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
 922 {
 923         struct device *dev = get_cpu_device(cpu);
 924
 925         if (!dev)
 926                 return;
 927
 928         if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
 929                 return;
 930
 931         dev_dbg(dev, "%s: Adding symlink\n", __func__);
 932         if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
 933                 dev_err(dev, "cpufreq symlink creation failed\n");
 934 }
 935
 936 static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
 937                                    struct device *dev)
 938 {
 939         dev_dbg(dev, "%s: Removing symlink\n", __func__);
 940         sysfs_remove_link(&dev->kobj, "cpufreq");
 941 }
 942
 943 static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
 944 {
 945         struct freq_attr **drv_attr;
 946         int ret = 0;
 947
 948         /* set up files for this cpu device */
 949         drv_attr = cpufreq_driver->attr;
 950         while (drv_attr && *drv_attr) {
 951                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 952                 if (ret)
 953                         return ret;
 954                 drv_attr++;
 955         }
 956         if (cpufreq_driver->get) {
 957                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 958                 if (ret)
 959                         return ret;
 960         }
 961
 962         ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 963         if (ret)
 964                 return ret;
 965
 966         if (cpufreq_driver->bios_limit) {
 967                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
 968                 if (ret)
 969                         return ret;
 970         }
 971
 972         return 0;
 973 }
 974
 975 __weak struct cpufreq_governor *cpufreq_default_governor(void)
 976 {
 977         return NULL;
 978 }
 979
 980 static int cpufreq_init_policy(struct cpufreq_policy *policy)
 981 {
 982         struct cpufreq_governor *gov = NULL;
 983         struct cpufreq_policy new_policy;
 984
 985         memcpy(&new_policy, policy, sizeof(*policy));
 986
 987         /* Update governor of new_policy to the governor used before hotplug */
 988         gov = find_governor(policy->last_governor);
 989         if (gov) {
 990                 pr_debug("Restoring governor %s for cpu %d\n",
 991                                 policy->governor->name, policy->cpu);
 992         } else {
 993                 gov = cpufreq_default_governor();
 994                 if (!gov)
 995                         return -ENODATA;
 996         }
 997
 998         new_policy.governor = gov;
 999
1000         /* Use the default policy if there is no last_policy. */
1001         if (cpufreq_driver->setpolicy) {
1002                 if (policy->last_policy)
1003                         new_policy.policy = policy->last_policy;
1004                 else
1005                         cpufreq_parse_governor(gov->name, &new_policy.policy,
1006                                                NULL);
1007         }
1008         /* set default policy */
1009         return cpufreq_set_policy(policy, &new_policy);
1010 }
1011
1012 static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1013 {
1014         int ret = 0;
1015
1016         /* Has this CPU been taken care of already? */
1017         if (cpumask_test_cpu(cpu, policy->cpus))
1018                 return 0;
1019
1020         down_write(&policy->rwsem);
1021         if (has_target())
1022                 cpufreq_stop_governor(policy);
1023
1024         cpumask_set_cpu(cpu, policy->cpus);
1025
1026         if (has_target()) {
1027                 ret = cpufreq_start_governor(policy);
1028                 if (ret)
1029                         pr_err("%s: Failed to start governor\n", __func__);
1030         }
1031         up_write(&policy->rwsem);
1032         return ret;
1033 }
1034
1035 static void handle_update(struct work_struct *work)
1036 {
1037         struct cpufreq_policy *policy =
1038                 container_of(work, struct cpufreq_policy, update);
1039         unsigned int cpu = policy->cpu;
1040         pr_debug("handle_update for cpu %u called\n", cpu);
1041         cpufreq_update_policy(cpu);
1042 }
1043
1044 static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1045 {
1046         struct cpufreq_policy *policy;
1047         int ret;
1048
1049         policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1050         if (!policy)
1051                 return NULL;
1052
1053         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1054                 goto err_free_policy;
1055
1056         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1057                 goto err_free_cpumask;
1058
1059         if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1060                 goto err_free_rcpumask;
1061
1062         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1063                                    cpufreq_global_kobject, "policy%u", cpu);
1064         if (ret) {
1065                 pr_err("%s: failed to init policy->kobj: %d\n", __func__, ret);
1066                 goto err_free_real_cpus;
1067         }
1068
1069         INIT_LIST_HEAD(&policy->policy_list);
1070         init_rwsem(&policy->rwsem);
1071         spin_lock_init(&policy->transition_lock);
1072         init_waitqueue_head(&policy->transition_wait);
1073         init_completion(&policy->kobj_unregister);
1074         INIT_WORK(&policy->update, handle_update);
1075
1076         policy->cpu = cpu;
1077         return policy;
1078
1079 err_free_real_cpus:
1080         free_cpumask_var(policy->real_cpus);
1081 err_free_rcpumask:
1082         free_cpumask_var(policy->related_cpus);
1083 err_free_cpumask:
1084         free_cpumask_var(policy->cpus);
1085 err_free_policy:
1086         kfree(policy);
1087
1088         return NULL;
1089 }
1090
1091 static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1092 {
1093         struct kobject *kobj;
1094         struct completion *cmp;
1095
1096         down_write(&policy->rwsem);
1097         cpufreq_stats_free_table(policy);
1098         kobj = &policy->kobj;
1099         cmp = &policy->kobj_unregister;
1100         up_write(&policy->rwsem);
1101         kobject_put(kobj);
1102
1103         /*
1104          * We need to make sure that the underlying kobj is
1105          * actually not referenced anymore by anybody before we
1106          * proceed with unloading.
1107          */
1108         pr_debug("waiting for dropping of refcount\n");
1109         wait_for_completion(cmp);
1110         pr_debug("wait complete\n");
1111 }
1112
1113 static void cpufreq_policy_free(struct cpufreq_policy *policy)
1114 {
1115         unsigned long flags;
1116         int cpu;
1117
1118         /* Remove policy from list */
1119         write_lock_irqsave(&cpufreq_driver_lock, flags);
1120         list_del(&policy->policy_list);
1121
1122         for_each_cpu(cpu, policy->related_cpus)
1123                 per_cpu(cpufreq_cpu_data, cpu) = NULL;
1124         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1125
1126         cpufreq_policy_put_kobj(policy);
1127         free_cpumask_var(policy->real_cpus);
1128         free_cpumask_var(policy->related_cpus);
1129         free_cpumask_var(policy->cpus);
1130         kfree(policy);
1131 }
1132
1133 static int cpufreq_online(unsigned int cpu)
1134 {
1135         struct cpufreq_policy *policy;
1136         bool new_policy;
1137         unsigned long flags;
1138         unsigned int j;
1139         int ret;
1140
1141         pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1142
1143         /* Check if this CPU already has a policy to manage it */
1144         policy = per_cpu(cpufreq_cpu_data, cpu);
1145         if (policy) {
1146                 WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1147                 if (!policy_is_inactive(policy))
1148                         return cpufreq_add_policy_cpu(policy, cpu);
1149
1150                 /* This is the only online CPU for the policy.  Start over. */
1151                 new_policy = false;
1152                 down_write(&policy->rwsem);
1153                 policy->cpu = cpu;
1154                 policy->governor = NULL;
1155                 up_write(&policy->rwsem);
1156         } else {
1157                 new_policy = true;
1158                 policy = cpufreq_policy_alloc(cpu);
1159                 if (!policy)
1160                         return -ENOMEM;
1161         }
1162
1163         cpumask_copy(policy->cpus, cpumask_of(cpu));
1164
1165         /* call driver. From then on the cpufreq must be able
1166          * to accept all calls to ->verify and ->setpolicy for this CPU
1167          */
1168         ret = cpufreq_driver->init(policy);
1169         if (ret) {
1170                 pr_debug("initialization failed\n");
1171                 goto out_free_policy;
1172         }
1173
1174         down_write(&policy->rwsem);
1175
1176         if (new_policy) {
1177                 /* related_cpus should at least include policy->cpus. */
1178                 cpumask_copy(policy->related_cpus, policy->cpus);
1179         }
1180
1181         /*
1182          * affected cpus must always be the one, which are online. We aren't
1183          * managing offline cpus here.
1184          */
1185         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1186
1187         if (new_policy) {
1188                 policy->user_policy.min = policy->min;
1189                 policy->user_policy.max = policy->max;
1190
1191                 for_each_cpu(j, policy->related_cpus) {
1192                         per_cpu(cpufreq_cpu_data, j) = policy;
1193                         add_cpu_dev_symlink(policy, j);
1194                 }
1195         } else {
1196                 policy->min = policy->user_policy.min;
1197                 policy->max = policy->user_policy.max;
1198         }
1199
1200         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1201                 policy->cur = cpufreq_driver->get(policy->cpu);
1202                 if (!policy->cur) {
1203                         pr_err("%s: ->get() failed\n", __func__);
1204                         goto out_exit_policy;
1205                 }
1206         }
1207
1208         /*
1209          * Sometimes boot loaders set CPU frequency to a value outside of
1210          * frequency table present with cpufreq core. In such cases CPU might be
1211          * unstable if it has to run on that frequency for long duration of time
1212          * and so its better to set it to a frequency which is specified in
1213          * freq-table. This also makes cpufreq stats inconsistent as
1214          * cpufreq-stats would fail to register because current frequency of CPU
1215          * isn't found in freq-table.
1216          *
1217          * Because we don't want this change to effect boot process badly, we go
1218          * for the next freq which is >= policy->cur ('cur' must be set by now,
1219          * otherwise we will end up setting freq to lowest of the table as 'cur'
1220          * is initialized to zero).
1221          *
1222          * We are passing target-freq as "policy->cur - 1" otherwise
1223          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1224          * equal to target-freq.
1225          */
1226         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1227             && has_target()) {
1228                 /* Are we running at unknown frequency ? */
1229                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1230                 if (ret == -EINVAL) {
1231                         /* Warn user and fix it */
1232                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1233                                 __func__, policy->cpu, policy->cur);
1234                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1235                                 CPUFREQ_RELATION_L);
1236
1237                         /*
1238                          * Reaching here after boot in a few seconds may not
1239                          * mean that system will remain stable at "unknown"
1240                          * frequency for longer duration. Hence, a BUG_ON().
1241                          */
1242                         BUG_ON(ret);
1243                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1244                                 __func__, policy->cpu, policy->cur);
1245                 }
1246         }
1247
1248         if (new_policy) {
1249                 ret = cpufreq_add_dev_interface(policy);
1250                 if (ret)
1251                         goto out_exit_policy;
1252
1253                 cpufreq_stats_create_table(policy);
1254
1255                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1256                 list_add(&policy->policy_list, &cpufreq_policy_list);
1257                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1258         }
1259
1260         ret = cpufreq_init_policy(policy);
1261         if (ret) {
1262                 pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1263                        __func__, cpu, ret);
1264                 /* cpufreq_policy_free() will notify based on this */
1265                 new_policy = false;
1266                 goto out_exit_policy;
1267         }
1268
1269         up_write(&policy->rwsem);
1270
1271         kobject_uevent(&policy->kobj, KOBJ_ADD);
1272
1273         /* Callback for handling stuff after policy is ready */
1274         if (cpufreq_driver->ready)
1275                 cpufreq_driver->ready(policy);
1276
1277         pr_debug("initialization complete\n");
1278
1279         return 0;
1280
1281 out_exit_policy:
1282         up_write(&policy->rwsem);
1283
1284         if (cpufreq_driver->exit)
1285                 cpufreq_driver->exit(policy);
1286
1287         for_each_cpu(j, policy->real_cpus)
1288                 remove_cpu_dev_symlink(policy, get_cpu_device(j));
1289
1290 out_free_policy:
1291         cpufreq_policy_free(policy);
1292         return ret;
1293 }
1294
1295 /**
1296  * cpufreq_add_dev - the cpufreq interface for a CPU device.
1297  * @dev: CPU device.
1298  * @sif: Subsystem interface structure pointer (not used)
1299  */
1300 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1301 {
1302         struct cpufreq_policy *policy;
1303         unsigned cpu = dev->id;
1304         int ret;
1305
1306         dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1307
1308         if (cpu_online(cpu)) {
1309                 ret = cpufreq_online(cpu);
1310                 if (ret)
1311                         return ret;
1312         }
1313
1314         /* Create sysfs link on CPU registration */
1315         policy = per_cpu(cpufreq_cpu_data, cpu);
1316         if (policy)
1317                 add_cpu_dev_symlink(policy, cpu);
1318
1319         return 0;
1320 }
1321
1322 static int cpufreq_offline(unsigned int cpu)
1323 {
1324         struct cpufreq_policy *policy;
1325         int ret;
1326
1327         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1328
1329         policy = cpufreq_cpu_get_raw(cpu);
1330         if (!policy) {
1331                 pr_debug("%s: No cpu_data found\n", __func__);
1332                 return 0;
1333         }
1334
1335         down_write(&policy->rwsem);
1336         if (has_target())
1337                 cpufreq_stop_governor(policy);
1338
1339         cpumask_clear_cpu(cpu, policy->cpus);
1340
1341         if (policy_is_inactive(policy)) {
1342                 if (has_target())
1343                         strncpy(policy->last_governor, policy->governor->name,
1344                                 CPUFREQ_NAME_LEN);
1345                 else
1346                         policy->last_policy = policy->policy;
1347         } else if (cpu == policy->cpu) {
1348                 /* Nominate new CPU */
1349                 policy->cpu = cpumask_any(policy->cpus);
1350         }
1351
1352         /* Start governor again for active policy */
1353         if (!policy_is_inactive(policy)) {
1354                 if (has_target()) {
1355                         ret = cpufreq_start_governor(policy);
1356                         if (ret)
1357                                 pr_err("%s: Failed to start governor\n", __func__);
1358                 }
1359
1360                 goto unlock;
1361         }
1362
1363         if (cpufreq_driver->stop_cpu)
1364                 cpufreq_driver->stop_cpu(policy);
1365
1366         if (has_target())
1367                 cpufreq_exit_governor(policy);
1368
1369         /*
1370          * Perform the ->exit() even during light-weight tear-down,
1371          * since this is a core component, and is essential for the
1372          * subsequent light-weight ->init() to succeed.
1373          */
1374         if (cpufreq_driver->exit) {
1375                 cpufreq_driver->exit(policy);
1376                 policy->freq_table = NULL;
1377         }
1378
1379 unlock:
1380         up_write(&policy->rwsem);
1381         return 0;
1382 }
1383
1384 /**
1385  * cpufreq_remove_dev - remove a CPU device
1386  *
1387  * Removes the cpufreq interface for a CPU device.
1388  */
1389 static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1390 {
1391         unsigned int cpu = dev->id;
1392         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1393
1394         if (!policy)
1395                 return;
1396
1397         if (cpu_online(cpu))
1398                 cpufreq_offline(cpu);
1399
1400         cpumask_clear_cpu(cpu, policy->real_cpus);
1401         remove_cpu_dev_symlink(policy, dev);
1402
1403         if (cpumask_empty(policy->real_cpus))
1404                 cpufreq_policy_free(policy);
1405 }
1406
1407 /**
1408  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1409  *      in deep trouble.
1410  *      @policy: policy managing CPUs
1411  *      @new_freq: CPU frequency the CPU actually runs at
1412  *
1413  *      We adjust to current frequency first, and need to clean up later.
1414  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1415  */
1416 static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1417                                 unsigned int new_freq)
1418 {
1419         struct cpufreq_freqs freqs;
1420
1421         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1422                  policy->cur, new_freq);
1423
1424         freqs.old = policy->cur;
1425         freqs.new = new_freq;
1426
1427         cpufreq_freq_transition_begin(policy, &freqs);
1428         cpufreq_freq_transition_end(policy, &freqs, 0);
1429 }
1430
1431 /**
1432  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1433  * @cpu: CPU number
1434  *
1435  * This is the last known freq, without actually getting it from the driver.
1436  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1437  */
1438 unsigned int cpufreq_quick_get(unsigned int cpu)
1439 {
1440         struct cpufreq_policy *policy;
1441         unsigned int ret_freq = 0;
1442         unsigned long flags;
1443
1444         read_lock_irqsave(&cpufreq_driver_lock, flags);
1445
1446         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1447                 ret_freq = cpufreq_driver->get(cpu);
1448                 read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1449                 return ret_freq;
1450         }
1451
1452         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1453
1454         policy = cpufreq_cpu_get(cpu);
1455         if (policy) {
1456                 ret_freq = policy->cur;
1457                 cpufreq_cpu_put(policy);
1458         }
1459
1460         return ret_freq;
1461 }
1462 EXPORT_SYMBOL(cpufreq_quick_get);
1463
1464 /**
1465  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1466  * @cpu: CPU number
1467  *
1468  * Just return the max possible frequency for a given CPU.
1469  */
1470 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1471 {
1472         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1473         unsigned int ret_freq = 0;
1474
1475         if (policy) {
1476                 ret_freq = policy->max;
1477                 cpufreq_cpu_put(policy);
1478         }
1479
1480         return ret_freq;
1481 }
1482 EXPORT_SYMBOL(cpufreq_quick_get_max);
1483
1484 static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1485 {
1486         unsigned int ret_freq = 0;
1487
1488         if (!cpufreq_driver->get)
1489                 return ret_freq;
1490
1491         ret_freq = cpufreq_driver->get(policy->cpu);
1492
1493         /*
1494          * Updating inactive policies is invalid, so avoid doing that.  Also
1495          * if fast frequency switching is used with the given policy, the check
1496          * against policy->cur is pointless, so skip it in that case too.
1497          */
1498         if (unlikely(policy_is_inactive(policy)) || policy->fast_switch_enabled)
1499                 return ret_freq;
1500
1501         if (ret_freq && policy->cur &&
1502                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1503                 /* verify no discrepancy between actual and
1504                                         saved value exists */
1505                 if (unlikely(ret_freq != policy->cur)) {
1506                         cpufreq_out_of_sync(policy, ret_freq);
1507                         schedule_work(&policy->update);
1508                 }
1509         }
1510
1511         return ret_freq;
1512 }
1513
1514 /**
1515  * cpufreq_get - get the current CPU frequency (in kHz)
1516  * @cpu: CPU number
1517  *
1518  * Get the CPU current (static) CPU frequency
1519  */
1520 unsigned int cpufreq_get(unsigned int cpu)
1521 {
1522         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1523         unsigned int ret_freq = 0;
1524
1525         if (policy) {
1526                 down_read(&policy->rwsem);
1527
1528                 if (!policy_is_inactive(policy))
1529                         ret_freq = __cpufreq_get(policy);
1530
1531                 up_read(&policy->rwsem);
1532
1533                 cpufreq_cpu_put(policy);
1534         }
1535
1536         return ret_freq;
1537 }
1538 EXPORT_SYMBOL(cpufreq_get);
1539
1540 static unsigned int cpufreq_update_current_freq(struct cpufreq_policy *policy)
1541 {
1542         unsigned int new_freq;
1543
1544         new_freq = cpufreq_driver->get(policy->cpu);
1545         if (!new_freq)
1546                 return 0;
1547
1548         if (!policy->cur) {
1549                 pr_debug("cpufreq: Driver did not initialize current freq\n");
1550                 policy->cur = new_freq;
1551         } else if (policy->cur != new_freq && has_target()) {
1552                 cpufreq_out_of_sync(policy, new_freq);
1553         }
1554
1555         return new_freq;
1556 }
1557
1558 static struct subsys_interface cpufreq_interface = {
1559         .name           = "cpufreq",
1560         .subsys         = &cpu_subsys,
1561         .add_dev        = cpufreq_add_dev,
1562         .remove_dev     = cpufreq_remove_dev,
1563 };
1564
1565 /*
1566  * In case platform wants some specific frequency to be configured
1567  * during suspend..
1568  */
1569 int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1570 {
1571         int ret;
1572
1573         if (!policy->suspend_freq) {
1574                 pr_debug("%s: suspend_freq not defined\n", __func__);
1575                 return 0;
1576         }
1577
1578         pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1579                         policy->suspend_freq);
1580
1581         ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1582                         CPUFREQ_RELATION_H);
1583         if (ret)
1584                 pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1585                                 __func__, policy->suspend_freq, ret);
1586
1587         return ret;
1588 }
1589 EXPORT_SYMBOL(cpufreq_generic_suspend);
1590
1591 /**
1592  * cpufreq_suspend() - Suspend CPUFreq governors
1593  *
1594  * Called during system wide Suspend/Hibernate cycles for suspending governors
1595  * as some platforms can't change frequency after this point in suspend cycle.
1596  * Because some of the devices (like: i2c, regulators, etc) they use for
1597  * changing frequency are suspended quickly after this point.
1598  */
1599 void cpufreq_suspend(void)
1600 {
1601         struct cpufreq_policy *policy;
1602
1603         if (!cpufreq_driver)
1604                 return;
1605
1606         if (!has_target() && !cpufreq_driver->suspend)
1607                 goto suspend;
1608
1609         pr_debug("%s: Suspending Governors\n", __func__);
1610
1611         for_each_active_policy(policy) {
1612                 if (has_target()) {
1613                         down_write(&policy->rwsem);
1614                         cpufreq_stop_governor(policy);
1615                         up_write(&policy->rwsem);
1616                 }
1617
1618                 if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1619                         pr_err("%s: Failed to suspend driver: %p\n", __func__,
1620                                 policy);
1621         }
1622
1623 suspend:
1624         cpufreq_suspended = true;
1625 }
1626
1627 /**
1628  * cpufreq_resume() - Resume CPUFreq governors
1629  *
1630  * Called during system wide Suspend/Hibernate cycle for resuming governors that
1631  * are suspended with cpufreq_suspend().
1632  */
1633 void cpufreq_resume(void)
1634 {
1635         struct cpufreq_policy *policy;
1636         int ret;
1637
1638         if (!cpufreq_driver)
1639                 return;
1640
1641         cpufreq_suspended = false;
1642
1643         if (!has_target() && !cpufreq_driver->resume)
1644                 return;
1645
1646         pr_debug("%s: Resuming Governors\n", __func__);
1647
1648         for_each_active_policy(policy) {
1649                 if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1650                         pr_err("%s: Failed to resume driver: %p\n", __func__,
1651                                 policy);
1652                 } else if (has_target()) {
1653                         down_write(&policy->rwsem);
1654                         ret = cpufreq_start_governor(policy);
1655                         up_write(&policy->rwsem);
1656
1657                         if (ret)
1658                                 pr_err("%s: Failed to start governor for policy: %p\n",
1659                                        __func__, policy);
1660                 }
1661         }
1662 }
1663
1664 /**
1665  *      cpufreq_get_current_driver - return current driver's name
1666  *
1667  *      Return the name string of the currently loaded cpufreq driver
1668  *      or NULL, if none.
1669  */
1670 const char *cpufreq_get_current_driver(void)
1671 {
1672         if (cpufreq_driver)
1673                 return cpufreq_driver->name;
1674
1675         return NULL;
1676 }
1677 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1678
1679 /**
1680  *      cpufreq_get_driver_data - return current driver data
1681  *
1682  *      Return the private data of the currently loaded cpufreq
1683  *      driver, or NULL if no cpufreq driver is loaded.
1684  */
1685 void *cpufreq_get_driver_data(void)
1686 {
1687         if (cpufreq_driver)
1688                 return cpufreq_driver->driver_data;
1689
1690         return NULL;
1691 }
1692 EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1693
1694 /*********************************************************************
1695  *                     NOTIFIER LISTS INTERFACE                      *
1696  *********************************************************************/
1697
1698 /**
1699  *      cpufreq_register_notifier - register a driver with cpufreq
1700  *      @nb: notifier function to register
1701  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1702  *
1703  *      Add a driver to one of two lists: either a list of drivers that
1704  *      are notified about clock rate changes (once before and once after
1705  *      the transition), or a list of drivers that are notified about
1706  *      changes in cpufreq policy.
1707  *
1708  *      This function may sleep, and has the same return conditions as
1709  *      blocking_notifier_chain_register.
1710  */
1711 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1712 {
1713         int ret;
1714
1715         if (cpufreq_disabled())
1716                 return -EINVAL;
1717
1718         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1719
1720         switch (list) {
1721         case CPUFREQ_TRANSITION_NOTIFIER:
1722                 mutex_lock(&cpufreq_fast_switch_lock);
1723
1724                 if (cpufreq_fast_switch_count > 0) {
1725                         mutex_unlock(&cpufreq_fast_switch_lock);
1726                         return -EBUSY;
1727                 }
1728                 ret = srcu_notifier_chain_register(
1729                                 &cpufreq_transition_notifier_list, nb);
1730                 if (!ret)
1731                         cpufreq_fast_switch_count--;
1732
1733                 mutex_unlock(&cpufreq_fast_switch_lock);
1734                 break;
1735         case CPUFREQ_POLICY_NOTIFIER:
1736                 ret = blocking_notifier_chain_register(
1737                                 &cpufreq_policy_notifier_list, nb);
1738                 break;
1739         default:
1740                 ret = -EINVAL;
1741         }
1742
1743         return ret;
1744 }
1745 EXPORT_SYMBOL(cpufreq_register_notifier);
1746
1747 /**
1748  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1749  *      @nb: notifier block to be unregistered
1750  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1751  *
1752  *      Remove a driver from the CPU frequency notifier list.
1753  *
1754  *      This function may sleep, and has the same return conditions as
1755  *      blocking_notifier_chain_unregister.
1756  */
1757 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1758 {
1759         int ret;
1760
1761         if (cpufreq_disabled())
1762                 return -EINVAL;
1763
1764         switch (list) {
1765         case CPUFREQ_TRANSITION_NOTIFIER:
1766                 mutex_lock(&cpufreq_fast_switch_lock);
1767
1768                 ret = srcu_notifier_chain_unregister(
1769                                 &cpufreq_transition_notifier_list, nb);
1770                 if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
1771                         cpufreq_fast_switch_count++;
1772
1773                 mutex_unlock(&cpufreq_fast_switch_lock);
1774                 break;
1775         case CPUFREQ_POLICY_NOTIFIER:
1776                 ret = blocking_notifier_chain_unregister(
1777                                 &cpufreq_policy_notifier_list, nb);
1778                 break;
1779         default:
1780                 ret = -EINVAL;
1781         }
1782
1783         return ret;
1784 }
1785 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1786
1787
1788 /*********************************************************************
1789  *                              GOVERNORS                            *
1790  *********************************************************************/
1791
1792 /**
1793  * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
1794  * @policy: cpufreq policy to switch the frequency for.
1795  * @target_freq: New frequency to set (may be approximate).
1796  *
1797  * Carry out a fast frequency switch without sleeping.
1798  *
1799  * The driver's ->fast_switch() callback invoked by this function must be
1800  * suitable for being called from within RCU-sched read-side critical sections
1801  * and it is expected to select the minimum available frequency greater than or
1802  * equal to @target_freq (CPUFREQ_RELATION_L).
1803  *
1804  * This function must not be called if policy->fast_switch_enabled is unset.
1805  *
1806  * Governors calling this function must guarantee that it will never be invoked
1807  * twice in parallel for the same policy and that it will never be called in
1808  * parallel with either ->target() or ->target_index() for the same policy.
1809  *
1810  * If CPUFREQ_ENTRY_INVALID is returned by the driver's ->fast_switch()
1811  * callback to indicate an error condition, the hardware configuration must be
1812  * preserved.
1813  */
1814 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
1815                                         unsigned int target_freq)
1816 {
1817         target_freq = clamp_val(target_freq, policy->min, policy->max);
1818
1819         return cpufreq_driver->fast_switch(policy, target_freq);
1820 }
1821 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
1822
1823 /* Must set freqs->new to intermediate frequency */
1824 static int __target_intermediate(struct cpufreq_policy *policy,
1825                                  struct cpufreq_freqs *freqs, int index)
1826 {
1827         int ret;
1828
1829         freqs->new = cpufreq_driver->get_intermediate(policy, index);
1830
1831         /* We don't need to switch to intermediate freq */
1832         if (!freqs->new)
1833                 return 0;
1834
1835         pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
1836                  __func__, policy->cpu, freqs->old, freqs->new);
1837
1838         cpufreq_freq_transition_begin(policy, freqs);
1839         ret = cpufreq_driver->target_intermediate(policy, index);
1840         cpufreq_freq_transition_end(policy, freqs, ret);
1841
1842         if (ret)
1843                 pr_err("%s: Failed to change to intermediate frequency: %d\n",
1844                        __func__, ret);
1845
1846         return ret;
1847 }
1848
1849 static int __target_index(struct cpufreq_policy *policy, int index)
1850 {
1851         struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
1852         unsigned int intermediate_freq = 0;
1853         unsigned int newfreq = policy->freq_table[index].frequency;
1854         int retval = -EINVAL;
1855         bool notify;
1856
1857         if (newfreq == policy->cur)
1858                 return 0;
1859
1860         notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1861         if (notify) {
1862                 /* Handle switching to intermediate frequency */
1863                 if (cpufreq_driver->get_intermediate) {
1864                         retval = __target_intermediate(policy, &freqs, index);
1865                         if (retval)
1866                                 return retval;
1867
1868                         intermediate_freq = freqs.new;
1869                         /* Set old freq to intermediate */
1870                         if (intermediate_freq)
1871                                 freqs.old = freqs.new;
1872                 }
1873
1874                 freqs.new = newfreq;
1875                 pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1876                          __func__, policy->cpu, freqs.old, freqs.new);
1877
1878                 cpufreq_freq_transition_begin(policy, &freqs);
1879         }
1880
1881         retval = cpufreq_driver->target_index(policy, index);
1882         if (retval)
1883                 pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
1884                        retval);
1885
1886         if (notify) {
1887                 cpufreq_freq_transition_end(policy, &freqs, retval);
1888
1889                 /*
1890                  * Failed after setting to intermediate freq? Driver should have
1891                  * reverted back to initial frequency and so should we. Check
1892                  * here for intermediate_freq instead of get_intermediate, in
1893                  * case we haven't switched to intermediate freq at all.
1894                  */
1895                 if (unlikely(retval && intermediate_freq)) {
1896                         freqs.old = intermediate_freq;
1897                         freqs.new = policy->restore_freq;
1898                         cpufreq_freq_transition_begin(policy, &freqs);
1899                         cpufreq_freq_transition_end(policy, &freqs, 0);
1900                 }
1901         }
1902
1903         return retval;
1904 }
1905
1906 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1907                             unsigned int target_freq,
1908                             unsigned int relation)
1909 {
1910         unsigned int old_target_freq = target_freq;
1911         int index;
1912
1913         if (cpufreq_disabled())
1914                 return -ENODEV;
1915
1916         /* Make sure that target_freq is within supported range */
1917         target_freq = clamp_val(target_freq, policy->min, policy->max);
1918
1919         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1920                  policy->cpu, target_freq, relation, old_target_freq);
1921
1922         /*
1923          * This might look like a redundant call as we are checking it again
1924          * after finding index. But it is left intentionally for cases where
1925          * exactly same freq is called again and so we can save on few function
1926          * calls.
1927          */
1928         if (target_freq == policy->cur)
1929                 return 0;
1930
1931         /* Save last value to restore later on errors */
1932         policy->restore_freq = policy->cur;
1933
1934         if (cpufreq_driver->target)
1935                 return cpufreq_driver->target(policy, target_freq, relation);
1936
1937         if (!cpufreq_driver->target_index)
1938                 return -EINVAL;
1939
1940         index = cpufreq_frequency_table_target(policy, target_freq, relation);
1941
1942         return __target_index(policy, index);
1943 }
1944 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1945
1946 int cpufreq_driver_target(struct cpufreq_policy *policy,
1947                           unsigned int target_freq,
1948                           unsigned int relation)
1949 {
1950         int ret = -EINVAL;
1951
1952         down_write(&policy->rwsem);
1953
1954         ret = __cpufreq_driver_target(policy, target_freq, relation);
1955
1956         up_write(&policy->rwsem);
1957
1958         return ret;
1959 }
1960 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1961
1962 __weak struct cpufreq_governor *cpufreq_fallback_governor(void)
1963 {
1964         return NULL;
1965 }
1966
1967 static int cpufreq_init_governor(struct cpufreq_policy *policy)
1968 {
1969         int ret;
1970
1971         /* Don't start any governor operations if we are entering suspend */
1972         if (cpufreq_suspended)
1973                 return 0;
1974         /*
1975          * Governor might not be initiated here if ACPI _PPC changed
1976          * notification happened, so check it.
1977          */
1978         if (!policy->governor)
1979                 return -EINVAL;
1980
1981         if (policy->governor->max_transition_latency &&
1982             policy->cpuinfo.transition_latency >
1983             policy->governor->max_transition_latency) {
1984                 struct cpufreq_governor *gov = cpufreq_fallback_governor();
1985
1986                 if (gov) {
1987                         pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
1988                                 policy->governor->name, gov->name);
1989                         policy->governor = gov;
1990                 } else {
1991                         return -EINVAL;
1992                 }
1993         }
1994
1995         if (!try_module_get(policy->governor->owner))
1996                 return -EINVAL;
1997
1998         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
1999
2000         if (policy->governor->init) {
2001                 ret = policy->governor->init(policy);
2002                 if (ret) {
2003                         module_put(policy->governor->owner);
2004                         return ret;
2005                 }
2006         }
2007
2008         return 0;
2009 }
2010
2011 static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2012 {
2013         if (cpufreq_suspended || !policy->governor)
2014                 return;
2015
2016         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2017
2018         if (policy->governor->exit)
2019                 policy->governor->exit(policy);
2020
2021         module_put(policy->governor->owner);
2022 }
2023
2024 static int cpufreq_start_governor(struct cpufreq_policy *policy)
2025 {
2026         int ret;
2027
2028         if (cpufreq_suspended)
2029                 return 0;
2030
2031         if (!policy->governor)
2032                 return -EINVAL;
2033
2034         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2035
2036         if (cpufreq_driver->get && !cpufreq_driver->setpolicy)
2037                 cpufreq_update_current_freq(policy);
2038
2039         if (policy->governor->start) {
2040                 ret = policy->governor->start(policy);
2041                 if (ret)
2042                         return ret;
2043         }
2044
2045         if (policy->governor->limits)
2046                 policy->governor->limits(policy);
2047
2048         return 0;
2049 }
2050
2051 static void cpufreq_stop_governor(struct cpufreq_policy *policy)
2052 {
2053         if (cpufreq_suspended || !policy->governor)
2054                 return;
2055
2056         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2057
2058         if (policy->governor->stop)
2059                 policy->governor->stop(policy);
2060 }
2061
2062 static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2063 {
2064         if (cpufreq_suspended || !policy->governor)
2065                 return;
2066
2067         pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2068
2069         if (policy->governor->limits)
2070                 policy->governor->limits(policy);
2071 }
2072
2073 int cpufreq_register_governor(struct cpufreq_governor *governor)
2074 {
2075         int err;
2076
2077         if (!governor)
2078                 return -EINVAL;
2079
2080         if (cpufreq_disabled())
2081                 return -ENODEV;
2082
2083         mutex_lock(&cpufreq_governor_mutex);
2084
2085         err = -EBUSY;
2086         if (!find_governor(governor->name)) {
2087                 err = 0;
2088                 list_add(&governor->governor_list, &cpufreq_governor_list);
2089         }
2090
2091         mutex_unlock(&cpufreq_governor_mutex);
2092         return err;
2093 }
2094 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2095
2096 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2097 {
2098         struct cpufreq_policy *policy;
2099         unsigned long flags;
2100
2101         if (!governor)
2102                 return;
2103
2104         if (cpufreq_disabled())
2105                 return;
2106
2107         /* clear last_governor for all inactive policies */
2108         read_lock_irqsave(&cpufreq_driver_lock, flags);
2109         for_each_inactive_policy(policy) {
2110                 if (!strcmp(policy->last_governor, governor->name)) {
2111                         policy->governor = NULL;
2112                         strcpy(policy->last_governor, "\0");
2113                 }
2114         }
2115         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2116
2117         mutex_lock(&cpufreq_governor_mutex);
2118         list_del(&governor->governor_list);
2119         mutex_unlock(&cpufreq_governor_mutex);
2120         return;
2121 }
2122 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2123
2124
2125 /*********************************************************************
2126  *                          POLICY INTERFACE                         *
2127  *********************************************************************/
2128
2129 /**
2130  * cpufreq_get_policy - get the current cpufreq_policy
2131  * @policy: struct cpufreq_policy into which the current cpufreq_policy
2132  *      is written
2133  *
2134  * Reads the current cpufreq policy.
2135  */
2136 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2137 {
2138         struct cpufreq_policy *cpu_policy;
2139         if (!policy)
2140                 return -EINVAL;
2141
2142         cpu_policy = cpufreq_cpu_get(cpu);
2143         if (!cpu_policy)
2144                 return -EINVAL;
2145
2146         memcpy(policy, cpu_policy, sizeof(*policy));
2147
2148         cpufreq_cpu_put(cpu_policy);
2149         return 0;
2150 }
2151 EXPORT_SYMBOL(cpufreq_get_policy);
2152
2153 /*
2154  * policy : current policy.
2155  * new_policy: policy to be set.
2156  */
2157 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2158                                 struct cpufreq_policy *new_policy)
2159 {
2160         struct cpufreq_governor *old_gov;
2161         int ret;
2162
2163         pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2164                  new_policy->cpu, new_policy->min, new_policy->max);
2165
2166         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2167
2168         /*
2169         * This check works well when we store new min/max freq attributes,
2170         * because new_policy is a copy of policy with one field updated.
2171         */
2172         if (new_policy->min > new_policy->max)
2173                 return -EINVAL;
2174
2175         /* verify the cpu speed can be set within this limit */
2176         ret = cpufreq_driver->verify(new_policy);
2177         if (ret)
2178                 return ret;
2179
2180         /* adjust if necessary - all reasons */
2181         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2182                         CPUFREQ_ADJUST, new_policy);
2183
2184         /*
2185          * verify the cpu speed can be set within this limit, which might be
2186          * different to the first one
2187          */
2188         ret = cpufreq_driver->verify(new_policy);
2189         if (ret)
2190                 return ret;
2191
2192         /* notification of the new policy */
2193         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2194                         CPUFREQ_NOTIFY, new_policy);
2195
2196         policy->min = new_policy->min;
2197         policy->max = new_policy->max;
2198
2199         policy->cached_target_freq = UINT_MAX;
2200
2201         pr_debug("new min and max freqs are %u - %u kHz\n",
2202                  policy->min, policy->max);
2203
2204         if (cpufreq_driver->setpolicy) {
2205                 policy->policy = new_policy->policy;
2206                 pr_debug("setting range\n");
2207                 return cpufreq_driver->setpolicy(new_policy);
2208         }
2209
2210         if (new_policy->governor == policy->governor) {
2211                 pr_debug("cpufreq: governor limits update\n");
2212                 cpufreq_governor_limits(policy);
2213                 return 0;
2214         }
2215
2216         pr_debug("governor switch\n");
2217
2218         /* save old, working values */
2219         old_gov = policy->governor;
2220         /* end old governor */
2221         if (old_gov) {
2222                 cpufreq_stop_governor(policy);
2223                 cpufreq_exit_governor(policy);
2224         }
2225
2226         /* start new governor */
2227         policy->governor = new_policy->governor;
2228         ret = cpufreq_init_governor(policy);
2229         if (!ret) {
2230                 ret = cpufreq_start_governor(policy);
2231                 if (!ret) {
2232                         pr_debug("cpufreq: governor change\n");
2233                         return 0;
2234                 }
2235                 cpufreq_exit_governor(policy);
2236         }
2237
2238         /* new governor failed, so re-start old one */
2239         pr_debug("starting governor %s failed\n", policy->governor->name);
2240         if (old_gov) {
2241                 policy->governor = old_gov;
2242                 if (cpufreq_init_governor(policy))
2243                         policy->governor = NULL;
2244                 else
2245                         cpufreq_start_governor(policy);
2246         }
2247
2248         return ret;
2249 }
2250
2251 /**
2252  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2253  *      @cpu: CPU which shall be re-evaluated
2254  *
2255  *      Useful for policy notifiers which have different necessities
2256  *      at different times.
2257  */
2258 void cpufreq_update_policy(unsigned int cpu)
2259 {
2260         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2261         struct cpufreq_policy new_policy;
2262
2263         if (!policy)
2264                 return;
2265
2266         down_write(&policy->rwsem);
2267
2268         if (policy_is_inactive(policy))
2269                 goto unlock;
2270
2271         pr_debug("updating policy for CPU %u\n", cpu);
2272         memcpy(&new_policy, policy, sizeof(*policy));
2273         new_policy.min = policy->user_policy.min;
2274         new_policy.max = policy->user_policy.max;
2275
2276         /*
2277          * BIOS might change freq behind our back
2278          * -> ask driver for current freq and notify governors about a change
2279          */
2280         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2281                 if (cpufreq_suspended)
2282                         goto unlock;
2283
2284                 new_policy.cur = cpufreq_update_current_freq(policy);
2285                 if (WARN_ON(!new_policy.cur))
2286                         goto unlock;
2287         }
2288
2289         cpufreq_set_policy(policy, &new_policy);
2290
2291 unlock:
2292         up_write(&policy->rwsem);
2293
2294         cpufreq_cpu_put(policy);
2295 }
2296 EXPORT_SYMBOL(cpufreq_update_policy);
2297
2298 /*********************************************************************
2299  *               BOOST                                               *
2300  *********************************************************************/
2301 static int cpufreq_boost_set_sw(int state)
2302 {
2303         struct cpufreq_policy *policy;
2304         int ret = -EINVAL;
2305
2306         for_each_active_policy(policy) {
2307                 if (!policy->freq_table)
2308                         continue;
2309
2310                 ret = cpufreq_frequency_table_cpuinfo(policy,
2311                                                       policy->freq_table);
2312                 if (ret) {
2313                         pr_err("%s: Policy frequency update failed\n",
2314                                __func__);
2315                         break;
2316                 }
2317
2318                 down_write(&policy->rwsem);
2319                 policy->user_policy.max = policy->max;
2320                 cpufreq_governor_limits(policy);
2321                 up_write(&policy->rwsem);
2322         }
2323
2324         return ret;
2325 }
2326
2327 int cpufreq_boost_trigger_state(int state)
2328 {
2329         unsigned long flags;
2330         int ret = 0;
2331
2332         if (cpufreq_driver->boost_enabled == state)
2333                 return 0;
2334
2335         write_lock_irqsave(&cpufreq_driver_lock, flags);
2336         cpufreq_driver->boost_enabled = state;
2337         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2338
2339         ret = cpufreq_driver->set_boost(state);
2340         if (ret) {
2341                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2342                 cpufreq_driver->boost_enabled = !state;
2343                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2344
2345                 pr_err("%s: Cannot %s BOOST\n",
2346                        __func__, state ? "enable" : "disable");
2347         }
2348
2349         return ret;
2350 }
2351
2352 static bool cpufreq_boost_supported(void)
2353 {
2354         return likely(cpufreq_driver) && cpufreq_driver->set_boost;
2355 }
2356
2357 static int create_boost_sysfs_file(void)
2358 {
2359         int ret;
2360
2361         ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2362         if (ret)
2363                 pr_err("%s: cannot register global BOOST sysfs file\n",
2364                        __func__);
2365
2366         return ret;
2367 }
2368
2369 static void remove_boost_sysfs_file(void)
2370 {
2371         if (cpufreq_boost_supported())
2372                 sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2373 }
2374
2375 int cpufreq_enable_boost_support(void)
2376 {
2377         if (!cpufreq_driver)
2378                 return -EINVAL;
2379
2380         if (cpufreq_boost_supported())
2381                 return 0;
2382
2383         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2384
2385         /* This will get removed on driver unregister */
2386         return create_boost_sysfs_file();
2387 }
2388 EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2389
2390 int cpufreq_boost_enabled(void)
2391 {
2392         return cpufreq_driver->boost_enabled;
2393 }
2394 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2395
2396 /*********************************************************************
2397  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2398  *********************************************************************/
2399 static enum cpuhp_state hp_online;
2400
2401 static int cpuhp_cpufreq_online(unsigned int cpu)
2402 {
2403         cpufreq_online(cpu);
2404
2405         return 0;
2406 }
2407
2408 static int cpuhp_cpufreq_offline(unsigned int cpu)
2409 {
2410         cpufreq_offline(cpu);
2411
2412         return 0;
2413 }
2414
2415 /**
2416  * cpufreq_register_driver - register a CPU Frequency driver
2417  * @driver_data: A struct cpufreq_driver containing the values#
2418  * submitted by the CPU Frequency driver.
2419  *
2420  * Registers a CPU Frequency driver to this core code. This code
2421  * returns zero on success, -EEXIST when another driver got here first
2422  * (and isn't unregistered in the meantime).
2423  *
2424  */
2425 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2426 {
2427         unsigned long flags;
2428         int ret;
2429
2430         if (cpufreq_disabled())
2431                 return -ENODEV;
2432
2433         if (!driver_data || !driver_data->verify || !driver_data->init ||
2434             !(driver_data->setpolicy || driver_data->target_index ||
2435                     driver_data->target) ||
2436              (driver_data->setpolicy && (driver_data->target_index ||
2437                     driver_data->target)) ||
2438              (!!driver_data->get_intermediate != !!driver_data->target_intermediate))
2439                 return -EINVAL;
2440
2441         pr_debug("trying to register driver %s\n", driver_data->name);
2442
2443         /* Protect against concurrent CPU online/offline. */
2444         get_online_cpus();
2445
2446         write_lock_irqsave(&cpufreq_driver_lock, flags);
2447         if (cpufreq_driver) {
2448                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2449                 ret = -EEXIST;
2450                 goto out;
2451         }
2452         cpufreq_driver = driver_data;
2453         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2454
2455         if (driver_data->setpolicy)
2456                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2457
2458         if (cpufreq_boost_supported()) {
2459                 ret = create_boost_sysfs_file();
2460                 if (ret)
2461                         goto err_null_driver;
2462         }
2463
2464         ret = subsys_interface_register(&cpufreq_interface);
2465         if (ret)
2466                 goto err_boost_unreg;
2467
2468         if (!(cpufreq_driver->flags & CPUFREQ_STICKY) &&
2469             list_empty(&cpufreq_policy_list)) {
2470                 /* if all ->init() calls failed, unregister */
2471                 pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2472                          driver_data->name);
2473                 goto err_if_unreg;
2474         }
2475
2476         ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "cpufreq:online",
2477                                         cpuhp_cpufreq_online,
2478                                         cpuhp_cpufreq_offline);
2479         if (ret < 0)
2480                 goto err_if_unreg;
2481         hp_online = ret;
2482         ret = 0;
2483
2484         pr_debug("driver %s up and running\n", driver_data->name);
2485         goto out;
2486
2487 err_if_unreg:
2488         subsys_interface_unregister(&cpufreq_interface);
2489 err_boost_unreg:
2490         remove_boost_sysfs_file();
2491 err_null_driver:
2492         write_lock_irqsave(&cpufreq_driver_lock, flags);
2493         cpufreq_driver = NULL;
2494         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2495 out:
2496         put_online_cpus();
2497         return ret;
2498 }
2499 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2500
2501 /**
2502  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2503  *
2504  * Unregister the current CPUFreq driver. Only call this if you have
2505  * the right to do so, i.e. if you have succeeded in initialising before!
2506  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2507  * currently not initialised.
2508  */
2509 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2510 {
2511         unsigned long flags;
2512
2513         if (!cpufreq_driver || (driver != cpufreq_driver))
2514                 return -EINVAL;
2515
2516         pr_debug("unregistering driver %s\n", driver->name);
2517
2518         /* Protect against concurrent cpu hotplug */
2519         get_online_cpus();
2520         subsys_interface_unregister(&cpufreq_interface);
2521         remove_boost_sysfs_file();
2522         cpuhp_remove_state_nocalls(hp_online);
2523
2524         write_lock_irqsave(&cpufreq_driver_lock, flags);
2525
2526         cpufreq_driver = NULL;
2527
2528         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2529         put_online_cpus();
2530
2531         return 0;
2532 }
2533 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2534
2535 /*
2536  * Stop cpufreq at shutdown to make sure it isn't holding any locks
2537  * or mutexes when secondary CPUs are halted.
2538  */
2539 static struct syscore_ops cpufreq_syscore_ops = {
2540         .shutdown = cpufreq_suspend,
2541 };
2542
2543 struct kobject *cpufreq_global_kobject;
2544 EXPORT_SYMBOL(cpufreq_global_kobject);
2545
2546 static int __init cpufreq_core_init(void)
2547 {
2548         if (cpufreq_disabled())
2549                 return -ENODEV;
2550
2551         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2552         BUG_ON(!cpufreq_global_kobject);
2553
2554         register_syscore_ops(&cpufreq_syscore_ops);
2555
2556         return 0;
2557 }
2558 module_param(off, int, 0444);
2559 core_initcall(cpufreq_core_init);