drivers/cpufreq/cpufreq.c

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