linux/drivers/cpufreq/cpufreq.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/drivers/cpufreq/cpufreq.c
   4 *
   5 *  Copyright (C) 2001 Russell King
   6 *            (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
   7 *            (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
   8 *
   9 *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
  10 *      Added handling for CPU hotplug
  11 *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
  12 *      Fix handling for CPU hotplug -- affected CPUs
  13 */
  14
  15#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  16
  17#include <linux/cpu.h>
  18#include <linux/cpufreq.h>
  19#include <linux/cpu_cooling.h>
  20#include <linux/delay.h>
  21#include <linux/device.h>
  22#include <linux/init.h>
  23#include <linux/kernel_stat.h>
  24#include <linux/module.h>
  25#include <linux/mutex.h>
  26#include <linux/pm_qos.h>
  27#include <linux/slab.h>
  28#include <linux/suspend.h>
  29#include <linux/syscore_ops.h>
  30#include <linux/tick.h>
  31#include <trace/events/power.h>
  32
  33static LIST_HEAD(cpufreq_policy_list);
  34
  35/* Macros to iterate over CPU policies */
  36#define for_each_suitable_policy(__policy, __active)                     \
  37        list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
  38                if ((__active) == !policy_is_inactive(__policy))
  39
  40#define for_each_active_policy(__policy)                \
  41        for_each_suitable_policy(__policy, true)
  42#define for_each_inactive_policy(__policy)              \
  43        for_each_suitable_policy(__policy, false)
  44
  45/* Iterate over governors */
  46static LIST_HEAD(cpufreq_governor_list);
  47#define for_each_governor(__governor)                           \
  48        list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
  49
  50static char default_governor[CPUFREQ_NAME_LEN];
  51
  52/*
  53 * The "cpufreq driver" - the arch- or hardware-dependent low
  54 * level driver of CPUFreq support, and its spinlock. This lock
  55 * also protects the cpufreq_cpu_data array.
  56 */
  57static struct cpufreq_driver *cpufreq_driver;
  58static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
  59static DEFINE_RWLOCK(cpufreq_driver_lock);
  60
  61static DEFINE_STATIC_KEY_FALSE(cpufreq_freq_invariance);
  62bool cpufreq_supports_freq_invariance(void)
  63{
  64        return static_branch_likely(&cpufreq_freq_invariance);
  65}
  66
  67/* Flag to suspend/resume CPUFreq governors */
  68static bool cpufreq_suspended;
  69
  70static inline bool has_target(void)
  71{
  72        return cpufreq_driver->target_index || cpufreq_driver->target;
  73}
  74
  75/* internal prototypes */
  76static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
  77static int cpufreq_init_governor(struct cpufreq_policy *policy);
  78static void cpufreq_exit_governor(struct cpufreq_policy *policy);
  79static void cpufreq_governor_limits(struct cpufreq_policy *policy);
  80static int cpufreq_set_policy(struct cpufreq_policy *policy,
  81                              struct cpufreq_governor *new_gov,
  82                              unsigned int new_pol);
  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 */
  91static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
  92SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
  93
  94static int off __read_mostly;
  95static int cpufreq_disabled(void)
  96{
  97        return off;
  98}
  99void disable_cpufreq(void)
 100{
 101        off = 1;
 102}
 103static DEFINE_MUTEX(cpufreq_governor_mutex);
 104
 105bool have_governor_per_policy(void)
 106{
 107        return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
 108}
 109EXPORT_SYMBOL_GPL(have_governor_per_policy);
 110
 111static struct kobject *cpufreq_global_kobject;
 112
 113struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
 114{
 115        if (have_governor_per_policy())
 116                return &policy->kobj;
 117        else
 118                return cpufreq_global_kobject;
 119}
 120EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
 121
 122static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
 123{
 124        struct kernel_cpustat kcpustat;
 125        u64 cur_wall_time;
 126        u64 idle_time;
 127        u64 busy_time;
 128
 129        cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
 130
 131        kcpustat_cpu_fetch(&kcpustat, cpu);
 132
 133        busy_time = kcpustat.cpustat[CPUTIME_USER];
 134        busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
 135        busy_time += kcpustat.cpustat[CPUTIME_IRQ];
 136        busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
 137        busy_time += kcpustat.cpustat[CPUTIME_STEAL];
 138        busy_time += kcpustat.cpustat[CPUTIME_NICE];
 139
 140        idle_time = cur_wall_time - busy_time;
 141        if (wall)
 142                *wall = div_u64(cur_wall_time, NSEC_PER_USEC);
 143
 144        return div_u64(idle_time, NSEC_PER_USEC);
 145}
 146
 147u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
 148{
 149        u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
 150
 151        if (idle_time == -1ULL)
 152                return get_cpu_idle_time_jiffy(cpu, wall);
 153        else if (!io_busy)
 154                idle_time += get_cpu_iowait_time_us(cpu, wall);
 155
 156        return idle_time;
 157}
 158EXPORT_SYMBOL_GPL(get_cpu_idle_time);
 159
 160/*
 161 * This is a generic cpufreq init() routine which can be used by cpufreq
 162 * drivers of SMP systems. It will do following:
 163 * - validate & show freq table passed
 164 * - set policies transition latency
 165 * - policy->cpus with all possible CPUs
 166 */
 167void cpufreq_generic_init(struct cpufreq_policy *policy,
 168                struct cpufreq_frequency_table *table,
 169                unsigned int transition_latency)
 170{
 171        policy->freq_table = table;
 172        policy->cpuinfo.transition_latency = transition_latency;
 173
 174        /*
 175         * The driver only supports the SMP configuration where all processors
 176         * share the clock and voltage and clock.
 177         */
 178        cpumask_setall(policy->cpus);
 179}
 180EXPORT_SYMBOL_GPL(cpufreq_generic_init);
 181
 182struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
 183{
 184        struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
 185
 186        return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
 187}
 188EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
 189
 190unsigned int cpufreq_generic_get(unsigned int cpu)
 191{
 192        struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
 193
 194        if (!policy || IS_ERR(policy->clk)) {
 195                pr_err("%s: No %s associated to cpu: %d\n",
 196                       __func__, policy ? "clk" : "policy", cpu);
 197                return 0;
 198        }
 199
 200        return clk_get_rate(policy->clk) / 1000;
 201}
 202EXPORT_SYMBOL_GPL(cpufreq_generic_get);
 203
 204/**
 205 * cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
 206 * @cpu: CPU to find the policy for.
 207 *
 208 * Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
 209 * the kobject reference counter of that policy.  Return a valid policy on
 210 * success or NULL on failure.
 211 *
 212 * The policy returned by this function has to be released with the help of
 213 * cpufreq_cpu_put() to balance its kobject reference counter properly.
 214 */
 215struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
 216{
 217        struct cpufreq_policy *policy = NULL;
 218        unsigned long flags;
 219
 220        if (WARN_ON(cpu >= nr_cpu_ids))
 221                return NULL;
 222
 223        /* get the cpufreq driver */
 224        read_lock_irqsave(&cpufreq_driver_lock, flags);
 225
 226        if (cpufreq_driver) {
 227                /* get the CPU */
 228                policy = cpufreq_cpu_get_raw(cpu);
 229                if (policy)
 230                        kobject_get(&policy->kobj);
 231        }
 232
 233        read_unlock_irqrestore(&cpufreq_driver_lock, flags);
 234
 235        return policy;
 236}
 237EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 238
 239/**
 240 * cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
 241 * @policy: cpufreq policy returned by cpufreq_cpu_get().
 242 */
 243void cpufreq_cpu_put(struct cpufreq_policy *policy)
 244{
 245        kobject_put(&policy->kobj);
 246}
 247EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 248
 249/**
 250 * cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
 251 * @policy: cpufreq policy returned by cpufreq_cpu_acquire().
 252 */
 253void cpufreq_cpu_release(struct cpufreq_policy *policy)
 254{
 255        if (WARN_ON(!policy))
 256                return;
 257
 258        lockdep_assert_held(&policy->rwsem);
 259
 260        up_write(&policy->rwsem);
 261
 262        cpufreq_cpu_put(policy);
 263}
 264
 265/**
 266 * cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
 267 * @cpu: CPU to find the policy for.
 268 *
 269 * Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
 270 * if the policy returned by it is not NULL, acquire its rwsem for writing.
 271 * Return the policy if it is active or release it and return NULL otherwise.
 272 *
 273 * The policy returned by this function has to be released with the help of
 274 * cpufreq_cpu_release() in order to release its rwsem and balance its usage
 275 * counter properly.
 276 */
 277struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
 278{
 279        struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 280
 281        if (!policy)
 282                return NULL;
 283
 284        down_write(&policy->rwsem);
 285
 286        if (policy_is_inactive(policy)) {
 287                cpufreq_cpu_release(policy);
 288                return NULL;
 289        }
 290
 291        return policy;
 292}
 293
 294/*********************************************************************
 295 *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 296 *********************************************************************/
 297
 298/**
 299 * adjust_jiffies - Adjust the system "loops_per_jiffy".
 300 * @val: CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
 301 * @ci: Frequency change information.
 302 *
 303 * This function alters the system "loops_per_jiffy" for the clock
 304 * speed change. Note that loops_per_jiffy cannot be updated on SMP
 305 * systems as each CPU might be scaled differently. So, use the arch
 306 * per-CPU loops_per_jiffy value wherever possible.
 307 */
 308static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 309{
 310#ifndef CONFIG_SMP
 311        static unsigned long l_p_j_ref;
 312        static unsigned int l_p_j_ref_freq;
 313
 314        if (ci->flags & CPUFREQ_CONST_LOOPS)
 315                return;
 316
 317        if (!l_p_j_ref_freq) {
 318                l_p_j_ref = loops_per_jiffy;
 319                l_p_j_ref_freq = ci->old;
 320                pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
 321                         l_p_j_ref, l_p_j_ref_freq);
 322        }
 323        if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
 324                loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
 325                                                                ci->new);
 326                pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
 327                         loops_per_jiffy, ci->new);
 328        }
 329#endif
 330}
 331
 332/**
 333 * cpufreq_notify_transition - Notify frequency transition and adjust jiffies.
 334 * @policy: cpufreq policy to enable fast frequency switching for.
 335 * @freqs: contain details of the frequency update.
 336 * @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
 337 *
 338 * This function calls the transition notifiers and adjust_jiffies().
 339 *
 340 * It is called twice on all CPU frequency changes that have external effects.
 341 */
 342static void cpufreq_notify_transition(struct cpufreq_policy *policy,
 343                                      struct cpufreq_freqs *freqs,
 344                                      unsigned int state)
 345{
 346        int cpu;
 347
 348        BUG_ON(irqs_disabled());
 349
 350        if (cpufreq_disabled())
 351                return;
 352
 353        freqs->policy = policy;
 354        freqs->flags = cpufreq_driver->flags;
 355        pr_debug("notification %u of frequency transition to %u kHz\n",
 356                 state, freqs->new);
 357
 358        switch (state) {
 359        case CPUFREQ_PRECHANGE:
 360                /*
 361                 * Detect if the driver reported a value as "old frequency"
 362                 * which is not equal to what the cpufreq core thinks is
 363                 * "old frequency".
 364                 */
 365                if (policy->cur && policy->cur != freqs->old) {
 366                        pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
 367                                 freqs->old, policy->cur);
 368                        freqs->old = policy->cur;
 369                }
 370
 371                srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 372                                         CPUFREQ_PRECHANGE, freqs);
 373
 374                adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 375                break;
 376
 377        case CPUFREQ_POSTCHANGE:
 378                adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 379                pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
 380                         cpumask_pr_args(policy->cpus));
 381
 382                for_each_cpu(cpu, policy->cpus)
 383                        trace_cpu_frequency(freqs->new, cpu);
 384
 385                srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 386                                         CPUFREQ_POSTCHANGE, freqs);
 387
 388                cpufreq_stats_record_transition(policy, freqs->new);
 389                policy->cur = freqs->new;
 390        }
 391}
 392
 393/* Do post notifications when there are chances that transition has failed */
 394static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
 395                struct cpufreq_freqs *freqs, int transition_failed)
 396{
 397        cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
 398        if (!transition_failed)
 399                return;
 400
 401        swap(freqs->old, freqs->new);
 402        cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
 403        cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
 404}
 405
 406void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
 407                struct cpufreq_freqs *freqs)
 408{
 409
 410        /*
 411         * Catch double invocations of _begin() which lead to self-deadlock.
 412         * ASYNC_NOTIFICATION drivers are left out because the cpufreq core
 413         * doesn't invoke _begin() on their behalf, and hence the chances of
 414         * double invocations are very low. Moreover, there are scenarios
 415         * where these checks can emit false-positive warnings in these
 416         * drivers; so we avoid that by skipping them altogether.
 417         */
 418        WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
 419                                && current == policy->transition_task);
 420
 421wait:
 422        wait_event(policy->transition_wait, !policy->transition_ongoing);
 423
 424        spin_lock(&policy->transition_lock);
 425
 426        if (unlikely(policy->transition_ongoing)) {
 427                spin_unlock(&policy->transition_lock);
 428                goto wait;
 429        }
 430
 431        policy->transition_ongoing = true;
 432        policy->transition_task = current;
 433
 434        spin_unlock(&policy->transition_lock);
 435
 436        cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
 437}
 438EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
 439
 440void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
 441                struct cpufreq_freqs *freqs, int transition_failed)
 442{
 443        if (WARN_ON(!policy->transition_ongoing))
 444                return;
 445
 446        cpufreq_notify_post_transition(policy, freqs, transition_failed);
 447
 448        arch_set_freq_scale(policy->related_cpus,
 449                            policy->cur,
 450                            policy->cpuinfo.max_freq);
 451
 452        policy->transition_ongoing = false;
 453        policy->transition_task = NULL;
 454
 455        wake_up(&policy->transition_wait);
 456}
 457EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
 458
 459/*
 460 * Fast frequency switching status count.  Positive means "enabled", negative
 461 * means "disabled" and 0 means "not decided yet".
 462 */
 463static int cpufreq_fast_switch_count;
 464static DEFINE_MUTEX(cpufreq_fast_switch_lock);
 465
 466static void cpufreq_list_transition_notifiers(void)
 467{
 468        struct notifier_block *nb;
 469
 470        pr_info("Registered transition notifiers:\n");
 471
 472        mutex_lock(&cpufreq_transition_notifier_list.mutex);
 473
 474        for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
 475                pr_info("%pS\n", nb->notifier_call);
 476
 477        mutex_unlock(&cpufreq_transition_notifier_list.mutex);
 478}
 479
 480/**
 481 * cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
 482 * @policy: cpufreq policy to enable fast frequency switching for.
 483 *
 484 * Try to enable fast frequency switching for @policy.
 485 *
 486 * The attempt will fail if there is at least one transition notifier registered
 487 * at this point, as fast frequency switching is quite fundamentally at odds
 488 * with transition notifiers.  Thus if successful, it will make registration of
 489 * transition notifiers fail going forward.
 490 */
 491void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
 492{
 493        lockdep_assert_held(&policy->rwsem);
 494
 495        if (!policy->fast_switch_possible)
 496                return;
 497
 498        mutex_lock(&cpufreq_fast_switch_lock);
 499        if (cpufreq_fast_switch_count >= 0) {
 500                cpufreq_fast_switch_count++;
 501                policy->fast_switch_enabled = true;
 502        } else {
 503                pr_warn("CPU%u: Fast frequency switching not enabled\n",
 504                        policy->cpu);
 505                cpufreq_list_transition_notifiers();
 506        }
 507        mutex_unlock(&cpufreq_fast_switch_lock);
 508}
 509EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
 510
 511/**
 512 * cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
 513 * @policy: cpufreq policy to disable fast frequency switching for.
 514 */
 515void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
 516{
 517        mutex_lock(&cpufreq_fast_switch_lock);
 518        if (policy->fast_switch_enabled) {
 519                policy->fast_switch_enabled = false;
 520                if (!WARN_ON(cpufreq_fast_switch_count <= 0))
 521                        cpufreq_fast_switch_count--;
 522        }
 523        mutex_unlock(&cpufreq_fast_switch_lock);
 524}
 525EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
 526
 527/**
 528 * cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
 529 * one.
 530 * @policy: associated policy to interrogate
 531 * @target_freq: target frequency to resolve.
 532 *
 533 * The target to driver frequency mapping is cached in the policy.
 534 *
 535 * Return: Lowest driver-supported frequency greater than or equal to the
 536 * given target_freq, subject to policy (min/max) and driver limitations.
 537 */
 538unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
 539                                         unsigned int target_freq)
 540{
 541        target_freq = clamp_val(target_freq, policy->min, policy->max);
 542        policy->cached_target_freq = target_freq;
 543
 544        if (cpufreq_driver->target_index) {
 545                unsigned int idx;
 546
 547                idx = cpufreq_frequency_table_target(policy, target_freq,
 548                                                     CPUFREQ_RELATION_L);
 549                policy->cached_resolved_idx = idx;
 550                return policy->freq_table[idx].frequency;
 551        }
 552
 553        if (cpufreq_driver->resolve_freq)
 554                return cpufreq_driver->resolve_freq(policy, target_freq);
 555
 556        return target_freq;
 557}
 558EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
 559
 560unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
 561{
 562        unsigned int latency;
 563
 564        if (policy->transition_delay_us)
 565                return policy->transition_delay_us;
 566
 567        latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
 568        if (latency) {
 569                /*
 570                 * For platforms that can change the frequency very fast (< 10
 571                 * us), the above formula gives a decent transition delay. But
 572                 * for platforms where transition_latency is in milliseconds, it
 573                 * ends up giving unrealistic values.
 574                 *
 575                 * Cap the default transition delay to 10 ms, which seems to be
 576                 * a reasonable amount of time after which we should reevaluate
 577                 * the frequency.
 578                 */
 579                return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
 580        }
 581
 582        return LATENCY_MULTIPLIER;
 583}
 584EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
 585
 586/*********************************************************************
 587 *                          SYSFS INTERFACE                          *
 588 *********************************************************************/
 589static ssize_t show_boost(struct kobject *kobj,
 590                          struct kobj_attribute *attr, char *buf)
 591{
 592        return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
 593}
 594
 595static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
 596                           const char *buf, size_t count)
 597{
 598        int ret, enable;
 599
 600        ret = sscanf(buf, "%d", &enable);
 601        if (ret != 1 || enable < 0 || enable > 1)
 602                return -EINVAL;
 603
 604        if (cpufreq_boost_trigger_state(enable)) {
 605                pr_err("%s: Cannot %s BOOST!\n",
 606                       __func__, enable ? "enable" : "disable");
 607                return -EINVAL;
 608        }
 609
 610        pr_debug("%s: cpufreq BOOST %s\n",
 611                 __func__, enable ? "enabled" : "disabled");
 612
 613        return count;
 614}
 615define_one_global_rw(boost);
 616
 617static struct cpufreq_governor *find_governor(const char *str_governor)
 618{
 619        struct cpufreq_governor *t;
 620
 621        for_each_governor(t)
 622                if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
 623                        return t;
 624
 625        return NULL;
 626}
 627
 628static struct cpufreq_governor *get_governor(const char *str_governor)
 629{
 630        struct cpufreq_governor *t;
 631
 632        mutex_lock(&cpufreq_governor_mutex);
 633        t = find_governor(str_governor);
 634        if (!t)
 635                goto unlock;
 636
 637        if (!try_module_get(t->owner))
 638                t = NULL;
 639
 640unlock:
 641        mutex_unlock(&cpufreq_governor_mutex);
 642
 643        return t;
 644}
 645
 646static unsigned int cpufreq_parse_policy(char *str_governor)
 647{
 648        if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
 649                return CPUFREQ_POLICY_PERFORMANCE;
 650
 651        if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
 652                return CPUFREQ_POLICY_POWERSAVE;
 653
 654        return CPUFREQ_POLICY_UNKNOWN;
 655}
 656
 657/**
 658 * cpufreq_parse_governor - parse a governor string only for has_target()
 659 * @str_governor: Governor name.
 660 */
 661static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
 662{
 663        struct cpufreq_governor *t;
 664
 665        t = get_governor(str_governor);
 666        if (t)
 667                return t;
 668
 669        if (request_module("cpufreq_%s", str_governor))
 670                return NULL;
 671
 672        return get_governor(str_governor);
 673}
 674
 675/*
 676 * cpufreq_per_cpu_attr_read() / show_##file_name() -
 677 * print out cpufreq information
 678 *
 679 * Write out information from cpufreq_driver->policy[cpu]; object must be
 680 * "unsigned int".
 681 */
 682
 683#define show_one(file_name, object)                     \
 684static ssize_t show_##file_name                         \
 685(struct cpufreq_policy *policy, char *buf)              \
 686{                                                       \
 687        return sprintf(buf, "%u\n", policy->object);    \
 688}
 689
 690show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 691show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 692show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
 693show_one(scaling_min_freq, min);
 694show_one(scaling_max_freq, max);
 695
 696__weak unsigned int arch_freq_get_on_cpu(int cpu)
 697{
 698        return 0;
 699}
 700
 701static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
 702{
 703        ssize_t ret;
 704        unsigned int freq;
 705
 706        freq = arch_freq_get_on_cpu(policy->cpu);
 707        if (freq)
 708                ret = sprintf(buf, "%u\n", freq);
 709        else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
 710                ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
 711        else
 712                ret = sprintf(buf, "%u\n", policy->cur);
 713        return ret;
 714}
 715
 716/*
 717 * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 718 */
 719#define store_one(file_name, object)                    \
 720static ssize_t store_##file_name                                        \
 721(struct cpufreq_policy *policy, const char *buf, size_t count)          \
 722{                                                                       \
 723        unsigned long val;                                              \
 724        int ret;                                                        \
 725                                                                        \
 726        ret = sscanf(buf, "%lu", &val);                                 \
 727        if (ret != 1)                                                   \
 728                return -EINVAL;                                         \
 729                                                                        \
 730        ret = freq_qos_update_request(policy->object##_freq_req, val);\
 731        return ret >= 0 ? count : ret;                                  \
 732}
 733
 734store_one(scaling_min_freq, min);
 735store_one(scaling_max_freq, max);
 736
 737/*
 738 * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 739 */
 740static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
 741                                        char *buf)
 742{
 743        unsigned int cur_freq = __cpufreq_get(policy);
 744
 745        if (cur_freq)
 746                return sprintf(buf, "%u\n", cur_freq);
 747
 748        return sprintf(buf, "<unknown>\n");
 749}
 750
 751/*
 752 * show_scaling_governor - show the current policy for the specified CPU
 753 */
 754static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
 755{
 756        if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
 757                return sprintf(buf, "powersave\n");
 758        else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 759                return sprintf(buf, "performance\n");
 760        else if (policy->governor)
 761                return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
 762                                policy->governor->name);
 763        return -EINVAL;
 764}
 765
 766/*
 767 * store_scaling_governor - store policy for the specified CPU
 768 */
 769static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
 770                                        const char *buf, size_t count)
 771{
 772        char str_governor[16];
 773        int ret;
 774
 775        ret = sscanf(buf, "%15s", str_governor);
 776        if (ret != 1)
 777                return -EINVAL;
 778
 779        if (cpufreq_driver->setpolicy) {
 780                unsigned int new_pol;
 781
 782                new_pol = cpufreq_parse_policy(str_governor);
 783                if (!new_pol)
 784                        return -EINVAL;
 785
 786                ret = cpufreq_set_policy(policy, NULL, new_pol);
 787        } else {
 788                struct cpufreq_governor *new_gov;
 789
 790                new_gov = cpufreq_parse_governor(str_governor);
 791                if (!new_gov)
 792                        return -EINVAL;
 793
 794                ret = cpufreq_set_policy(policy, new_gov,
 795                                         CPUFREQ_POLICY_UNKNOWN);
 796
 797                module_put(new_gov->owner);
 798        }
 799
 800        return ret ? ret : count;
 801}
 802
 803/*
 804 * show_scaling_driver - show the cpufreq driver currently loaded
 805 */
 806static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
 807{
 808        return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
 809}
 810
 811/*
 812 * show_scaling_available_governors - show the available CPUfreq governors
 813 */
 814static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
 815                                                char *buf)
 816{
 817        ssize_t i = 0;
 818        struct cpufreq_governor *t;
 819
 820        if (!has_target()) {
 821                i += sprintf(buf, "performance powersave");
 822                goto out;
 823        }
 824
 825        mutex_lock(&cpufreq_governor_mutex);
 826        for_each_governor(t) {
 827                if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
 828                    - (CPUFREQ_NAME_LEN + 2)))
 829                        break;
 830                i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
 831        }
 832        mutex_unlock(&cpufreq_governor_mutex);
 833out:
 834        i += sprintf(&buf[i], "\n");
 835        return i;
 836}
 837
 838ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
 839{
 840        ssize_t i = 0;
 841        unsigned int cpu;
 842
 843        for_each_cpu(cpu, mask) {
 844                if (i)
 845                        i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 846                i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 847                if (i >= (PAGE_SIZE - 5))
 848                        break;
 849        }
 850        i += sprintf(&buf[i], "\n");
 851        return i;
 852}
 853EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
 854
 855/*
 856 * show_related_cpus - show the CPUs affected by each transition even if
 857 * hw coordination is in use
 858 */
 859static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
 860{
 861        return cpufreq_show_cpus(policy->related_cpus, buf);
 862}
 863
 864/*
 865 * show_affected_cpus - show the CPUs affected by each transition
 866 */
 867static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
 868{
 869        return cpufreq_show_cpus(policy->cpus, buf);
 870}
 871
 872static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
 873                                        const char *buf, size_t count)
 874{
 875        unsigned int freq = 0;
 876        unsigned int ret;
 877
 878        if (!policy->governor || !policy->governor->store_setspeed)
 879                return -EINVAL;
 880
 881        ret = sscanf(buf, "%u", &freq);
 882        if (ret != 1)
 883                return -EINVAL;
 884
 885        policy->governor->store_setspeed(policy, freq);
 886
 887        return count;
 888}
 889
 890static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
 891{
 892        if (!policy->governor || !policy->governor->show_setspeed)
 893                return sprintf(buf, "<unsupported>\n");
 894
 895        return policy->governor->show_setspeed(policy, buf);
 896}
 897
 898/*
 899 * show_bios_limit - show the current cpufreq HW/BIOS limitation
 900 */
 901static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
 902{
 903        unsigned int limit;
 904        int ret;
 905        ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
 906        if (!ret)
 907                return sprintf(buf, "%u\n", limit);
 908        return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
 909}
 910
 911cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
 912cpufreq_freq_attr_ro(cpuinfo_min_freq);
 913cpufreq_freq_attr_ro(cpuinfo_max_freq);
 914cpufreq_freq_attr_ro(cpuinfo_transition_latency);
 915cpufreq_freq_attr_ro(scaling_available_governors);
 916cpufreq_freq_attr_ro(scaling_driver);
 917cpufreq_freq_attr_ro(scaling_cur_freq);
 918cpufreq_freq_attr_ro(bios_limit);
 919cpufreq_freq_attr_ro(related_cpus);
 920cpufreq_freq_attr_ro(affected_cpus);
 921cpufreq_freq_attr_rw(scaling_min_freq);
 922cpufreq_freq_attr_rw(scaling_max_freq);
 923cpufreq_freq_attr_rw(scaling_governor);
 924cpufreq_freq_attr_rw(scaling_setspeed);
 925
 926static struct attribute *default_attrs[] = {
 927        &cpuinfo_min_freq.attr,
 928        &cpuinfo_max_freq.attr,
 929        &cpuinfo_transition_latency.attr,
 930        &scaling_min_freq.attr,
 931        &scaling_max_freq.attr,
 932        &affected_cpus.attr,
 933        &related_cpus.attr,
 934        &scaling_governor.attr,
 935        &scaling_driver.attr,
 936        &scaling_available_governors.attr,
 937        &scaling_setspeed.attr,
 938        NULL
 939};
 940
 941#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
 942#define to_attr(a) container_of(a, struct freq_attr, attr)
 943
 944static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 945{
 946        struct cpufreq_policy *policy = to_policy(kobj);
 947        struct freq_attr *fattr = to_attr(attr);
 948        ssize_t ret;
 949
 950        if (!fattr->show)
 951                return -EIO;
 952
 953        down_read(&policy->rwsem);
 954        ret = fattr->show(policy, buf);
 955        up_read(&policy->rwsem);
 956
 957        return ret;
 958}
 959
 960static ssize_t store(struct kobject *kobj, struct attribute *attr,
 961                     const char *buf, size_t count)
 962{
 963        struct cpufreq_policy *policy = to_policy(kobj);
 964        struct freq_attr *fattr = to_attr(attr);
 965        ssize_t ret = -EINVAL;
 966
 967        if (!fattr->store)
 968                return -EIO;
 969
 970        /*
 971         * cpus_read_trylock() is used here to work around a circular lock
 972         * dependency problem with respect to the cpufreq_register_driver().
 973         */
 974        if (!cpus_read_trylock())
 975                return -EBUSY;
 976
 977        if (cpu_online(policy->cpu)) {
 978                down_write(&policy->rwsem);
 979                ret = fattr->store(policy, buf, count);
 980                up_write(&policy->rwsem);
 981        }
 982
 983        cpus_read_unlock();
 984
 985        return ret;
 986}
 987
 988static void cpufreq_sysfs_release(struct kobject *kobj)
 989{
 990        struct cpufreq_policy *policy = to_policy(kobj);
 991        pr_debug("last reference is dropped\n");
 992        complete(&policy->kobj_unregister);
 993}
 994
 995static const struct sysfs_ops sysfs_ops = {
 996        .show   = show,
 997        .store  = store,
 998};
 999
1000static struct kobj_type ktype_cpufreq = {
1001        .sysfs_ops      = &sysfs_ops,
1002        .default_attrs  = default_attrs,
1003        .release        = cpufreq_sysfs_release,
1004};
1005
1006static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu)
1007{
1008        struct device *dev = get_cpu_device(cpu);
1009
1010        if (unlikely(!dev))
1011                return;
1012
1013        if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
1014                return;
1015
1016        dev_dbg(dev, "%s: Adding symlink\n", __func__);
1017        if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
1018                dev_err(dev, "cpufreq symlink creation failed\n");
1019}
1020
1021static void remove_cpu_dev_symlink(struct cpufreq_policy *policy,
1022                                   struct device *dev)
1023{
1024        dev_dbg(dev, "%s: Removing symlink\n", __func__);
1025        sysfs_remove_link(&dev->kobj, "cpufreq");
1026}
1027
1028static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
1029{
1030        struct freq_attr **drv_attr;
1031        int ret = 0;
1032
1033        /* set up files for this cpu device */
1034        drv_attr = cpufreq_driver->attr;
1035        while (drv_attr && *drv_attr) {
1036                ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
1037                if (ret)
1038                        return ret;
1039                drv_attr++;
1040        }
1041        if (cpufreq_driver->get) {
1042                ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
1043                if (ret)
1044                        return ret;
1045        }
1046
1047        ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
1048        if (ret)
1049                return ret;
1050
1051        if (cpufreq_driver->bios_limit) {
1052                ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
1053                if (ret)
1054                        return ret;
1055        }
1056
1057        return 0;
1058}
1059
1060static int cpufreq_init_policy(struct cpufreq_policy *policy)
1061{
1062        struct cpufreq_governor *gov = NULL;
1063        unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
1064        int ret;
1065
1066        if (has_target()) {
1067                /* Update policy governor to the one used before hotplug. */
1068                gov = get_governor(policy->last_governor);
1069                if (gov) {
1070                        pr_debug("Restoring governor %s for cpu %d\n",
1071                                 gov->name, policy->cpu);
1072                } else {
1073                        gov = get_governor(default_governor);
1074                }
1075
1076                if (!gov) {
1077                        gov = cpufreq_default_governor();
1078                        __module_get(gov->owner);
1079                }
1080
1081        } else {
1082
1083                /* Use the default policy if there is no last_policy. */
1084                if (policy->last_policy) {
1085                        pol = policy->last_policy;
1086                } else {
1087                        pol = cpufreq_parse_policy(default_governor);
1088                        /*
1089                         * In case the default governor is neither "performance"
1090                         * nor "powersave", fall back to the initial policy
1091                         * value set by the driver.
1092                         */
1093                        if (pol == CPUFREQ_POLICY_UNKNOWN)
1094                                pol = policy->policy;
1095                }
1096                if (pol != CPUFREQ_POLICY_PERFORMANCE &&
1097                    pol != CPUFREQ_POLICY_POWERSAVE)
1098                        return -ENODATA;
1099        }
1100
1101        ret = cpufreq_set_policy(policy, gov, pol);
1102        if (gov)
1103                module_put(gov->owner);
1104
1105        return ret;
1106}
1107
1108static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
1109{
1110        int ret = 0;
1111
1112        /* Has this CPU been taken care of already? */
1113        if (cpumask_test_cpu(cpu, policy->cpus))
1114                return 0;
1115
1116        down_write(&policy->rwsem);
1117        if (has_target())
1118                cpufreq_stop_governor(policy);
1119
1120        cpumask_set_cpu(cpu, policy->cpus);
1121
1122        if (has_target()) {
1123                ret = cpufreq_start_governor(policy);
1124                if (ret)
1125                        pr_err("%s: Failed to start governor\n", __func__);
1126        }
1127        up_write(&policy->rwsem);
1128        return ret;
1129}
1130
1131void refresh_frequency_limits(struct cpufreq_policy *policy)
1132{
1133        if (!policy_is_inactive(policy)) {
1134                pr_debug("updating policy for CPU %u\n", policy->cpu);
1135
1136                cpufreq_set_policy(policy, policy->governor, policy->policy);
1137        }
1138}
1139EXPORT_SYMBOL(refresh_frequency_limits);
1140
1141static void handle_update(struct work_struct *work)
1142{
1143        struct cpufreq_policy *policy =
1144                container_of(work, struct cpufreq_policy, update);
1145
1146        pr_debug("handle_update for cpu %u called\n", policy->cpu);
1147        down_write(&policy->rwsem);
1148        refresh_frequency_limits(policy);
1149        up_write(&policy->rwsem);
1150}
1151
1152static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
1153                                void *data)
1154{
1155        struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
1156
1157        schedule_work(&policy->update);
1158        return 0;
1159}
1160
1161static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
1162                                void *data)
1163{
1164        struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
1165
1166        schedule_work(&policy->update);
1167        return 0;
1168}
1169
1170static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
1171{
1172        struct kobject *kobj;
1173        struct completion *cmp;
1174
1175        down_write(&policy->rwsem);
1176        cpufreq_stats_free_table(policy);
1177        kobj = &policy->kobj;
1178        cmp = &policy->kobj_unregister;
1179        up_write(&policy->rwsem);
1180        kobject_put(kobj);
1181
1182        /*
1183         * We need to make sure that the underlying kobj is
1184         * actually not referenced anymore by anybody before we
1185         * proceed with unloading.
1186         */
1187        pr_debug("waiting for dropping of refcount\n");
1188        wait_for_completion(cmp);
1189        pr_debug("wait complete\n");
1190}
1191
1192static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
1193{
1194        struct cpufreq_policy *policy;
1195        struct device *dev = get_cpu_device(cpu);
1196        int ret;
1197
1198        if (!dev)
1199                return NULL;
1200
1201        policy = kzalloc(sizeof(*policy), GFP_KERNEL);
1202        if (!policy)
1203                return NULL;
1204
1205        if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
1206                goto err_free_policy;
1207
1208        if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
1209                goto err_free_cpumask;
1210
1211        if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
1212                goto err_free_rcpumask;
1213
1214        ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
1215                                   cpufreq_global_kobject, "policy%u", cpu);
1216        if (ret) {
1217                dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
1218                /*
1219                 * The entire policy object will be freed below, but the extra
1220                 * memory allocated for the kobject name needs to be freed by
1221                 * releasing the kobject.
1222                 */
1223                kobject_put(&policy->kobj);
1224                goto err_free_real_cpus;
1225        }
1226
1227        freq_constraints_init(&policy->constraints);
1228
1229        policy->nb_min.notifier_call = cpufreq_notifier_min;
1230        policy->nb_max.notifier_call = cpufreq_notifier_max;
1231
1232        ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
1233                                    &policy->nb_min);
1234        if (ret) {
1235                dev_err(dev, "Failed to register MIN QoS notifier: %d (%*pbl)\n",
1236                        ret, cpumask_pr_args(policy->cpus));
1237                goto err_kobj_remove;
1238        }
1239
1240        ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
1241                                    &policy->nb_max);
1242        if (ret) {
1243                dev_err(dev, "Failed to register MAX QoS notifier: %d (%*pbl)\n",
1244                        ret, cpumask_pr_args(policy->cpus));
1245                goto err_min_qos_notifier;
1246        }
1247
1248        INIT_LIST_HEAD(&policy->policy_list);
1249        init_rwsem(&policy->rwsem);
1250        spin_lock_init(&policy->transition_lock);
1251        init_waitqueue_head(&policy->transition_wait);
1252        init_completion(&policy->kobj_unregister);
1253        INIT_WORK(&policy->update, handle_update);
1254
1255        policy->cpu = cpu;
1256        return policy;
1257
1258err_min_qos_notifier:
1259        freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1260                                 &policy->nb_min);
1261err_kobj_remove:
1262        cpufreq_policy_put_kobj(policy);
1263err_free_real_cpus:
1264        free_cpumask_var(policy->real_cpus);
1265err_free_rcpumask:
1266        free_cpumask_var(policy->related_cpus);
1267err_free_cpumask:
1268        free_cpumask_var(policy->cpus);
1269err_free_policy:
1270        kfree(policy);
1271
1272        return NULL;
1273}
1274
1275static void cpufreq_policy_free(struct cpufreq_policy *policy)
1276{
1277        unsigned long flags;
1278        int cpu;
1279
1280        /* Remove policy from list */
1281        write_lock_irqsave(&cpufreq_driver_lock, flags);
1282        list_del(&policy->policy_list);
1283
1284        for_each_cpu(cpu, policy->related_cpus)
1285                per_cpu(cpufreq_cpu_data, cpu) = NULL;
1286        write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1287
1288        freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
1289                                 &policy->nb_max);
1290        freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
1291                                 &policy->nb_min);
1292
1293        /* Cancel any pending policy->update work before freeing the policy. */
1294        cancel_work_sync(&policy->update);
1295
1296        if (policy->max_freq_req) {
1297                /*
1298                 * CPUFREQ_CREATE_POLICY notification is sent only after
1299                 * successfully adding max_freq_req request.
1300                 */
1301                blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1302                                             CPUFREQ_REMOVE_POLICY, policy);
1303                freq_qos_remove_request(policy->max_freq_req);
1304        }
1305
1306        freq_qos_remove_request(policy->min_freq_req);
1307        kfree(policy->min_freq_req);
1308
1309        cpufreq_policy_put_kobj(policy);
1310        free_cpumask_var(policy->real_cpus);
1311        free_cpumask_var(policy->related_cpus);
1312        free_cpumask_var(policy->cpus);
1313        kfree(policy);
1314}
1315
1316static int cpufreq_online(unsigned int cpu)
1317{
1318        struct cpufreq_policy *policy;
1319        bool new_policy;
1320        unsigned long flags;
1321        unsigned int j;
1322        int ret;
1323
1324        pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
1325
1326        /* Check if this CPU already has a policy to manage it */
1327        policy = per_cpu(cpufreq_cpu_data, cpu);
1328        if (policy) {
1329                WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
1330                if (!policy_is_inactive(policy))
1331                        return cpufreq_add_policy_cpu(policy, cpu);
1332
1333                /* This is the only online CPU for the policy.  Start over. */
1334                new_policy = false;
1335                down_write(&policy->rwsem);
1336                policy->cpu = cpu;
1337                policy->governor = NULL;
1338                up_write(&policy->rwsem);
1339        } else {
1340                new_policy = true;
1341                policy = cpufreq_policy_alloc(cpu);
1342                if (!policy)
1343                        return -ENOMEM;
1344        }
1345
1346        if (!new_policy && cpufreq_driver->online) {
1347                ret = cpufreq_driver->online(policy);
1348                if (ret) {
1349                        pr_debug("%s: %d: initialization failed\n", __func__,
1350                                 __LINE__);
1351                        goto out_exit_policy;
1352                }
1353
1354                /* Recover policy->cpus using related_cpus */
1355                cpumask_copy(policy->cpus, policy->related_cpus);
1356        } else {
1357                cpumask_copy(policy->cpus, cpumask_of(cpu));
1358
1359                /*
1360                 * Call driver. From then on the cpufreq must be able
1361                 * to accept all calls to ->verify and ->setpolicy for this CPU.
1362                 */
1363                ret = cpufreq_driver->init(policy);
1364                if (ret) {
1365                        pr_debug("%s: %d: initialization failed\n", __func__,
1366                                 __LINE__);
1367                        goto out_free_policy;
1368                }
1369
1370                /*
1371                 * The initialization has succeeded and the policy is online.
1372                 * If there is a problem with its frequency table, take it
1373                 * offline and drop it.
1374                 */
1375                ret = cpufreq_table_validate_and_sort(policy);
1376                if (ret)
1377                        goto out_offline_policy;
1378
1379                /* related_cpus should at least include policy->cpus. */
1380                cpumask_copy(policy->related_cpus, policy->cpus);
1381        }
1382
1383        down_write(&policy->rwsem);
1384        /*
1385         * affected cpus must always be the one, which are online. We aren't
1386         * managing offline cpus here.
1387         */
1388        cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1389
1390        if (new_policy) {
1391                for_each_cpu(j, policy->related_cpus) {
1392                        per_cpu(cpufreq_cpu_data, j) = policy;
1393                        add_cpu_dev_symlink(policy, j);
1394                }
1395
1396                policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
1397                                               GFP_KERNEL);
1398                if (!policy->min_freq_req) {
1399                        ret = -ENOMEM;
1400                        goto out_destroy_policy;
1401                }
1402
1403                ret = freq_qos_add_request(&policy->constraints,
1404                                           policy->min_freq_req, FREQ_QOS_MIN,
1405                                           policy->min);
1406                if (ret < 0) {
1407                        /*
1408                         * So we don't call freq_qos_remove_request() for an
1409                         * uninitialized request.
1410                         */
1411                        kfree(policy->min_freq_req);
1412                        policy->min_freq_req = NULL;
1413                        goto out_destroy_policy;
1414                }
1415
1416                /*
1417                 * This must be initialized right here to avoid calling
1418                 * freq_qos_remove_request() on uninitialized request in case
1419                 * of errors.
1420                 */
1421                policy->max_freq_req = policy->min_freq_req + 1;
1422
1423                ret = freq_qos_add_request(&policy->constraints,
1424                                           policy->max_freq_req, FREQ_QOS_MAX,
1425                                           policy->max);
1426                if (ret < 0) {
1427                        policy->max_freq_req = NULL;
1428                        goto out_destroy_policy;
1429                }
1430
1431                blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1432                                CPUFREQ_CREATE_POLICY, policy);
1433        }
1434
1435        if (cpufreq_driver->get && has_target()) {
1436                policy->cur = cpufreq_driver->get(policy->cpu);
1437                if (!policy->cur) {
1438                        ret = -EIO;
1439                        pr_err("%s: ->get() failed\n", __func__);
1440                        goto out_destroy_policy;
1441                }
1442        }
1443
1444        /*
1445         * Sometimes boot loaders set CPU frequency to a value outside of
1446         * frequency table present with cpufreq core. In such cases CPU might be
1447         * unstable if it has to run on that frequency for long duration of time
1448         * and so its better to set it to a frequency which is specified in
1449         * freq-table. This also makes cpufreq stats inconsistent as
1450         * cpufreq-stats would fail to register because current frequency of CPU
1451         * isn't found in freq-table.
1452         *
1453         * Because we don't want this change to effect boot process badly, we go
1454         * for the next freq which is >= policy->cur ('cur' must be set by now,
1455         * otherwise we will end up setting freq to lowest of the table as 'cur'
1456         * is initialized to zero).
1457         *
1458         * We are passing target-freq as "policy->cur - 1" otherwise
1459         * __cpufreq_driver_target() would simply fail, as policy->cur will be
1460         * equal to target-freq.
1461         */
1462        if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1463            && has_target()) {
1464                unsigned int old_freq = policy->cur;
1465
1466                /* Are we running at unknown frequency ? */
1467                ret = cpufreq_frequency_table_get_index(policy, old_freq);
1468                if (ret == -EINVAL) {
1469                        ret = __cpufreq_driver_target(policy, old_freq - 1,
1470                                                      CPUFREQ_RELATION_L);
1471
1472                        /*
1473                         * Reaching here after boot in a few seconds may not
1474                         * mean that system will remain stable at "unknown"
1475                         * frequency for longer duration. Hence, a BUG_ON().
1476                         */
1477                        BUG_ON(ret);
1478                        pr_info("%s: CPU%d: Running at unlisted initial frequency: %u KHz, changing to: %u KHz\n",
1479                                __func__, policy->cpu, old_freq, policy->cur);
1480                }
1481        }
1482
1483        if (new_policy) {
1484                ret = cpufreq_add_dev_interface(policy);
1485                if (ret)
1486                        goto out_destroy_policy;
1487
1488                cpufreq_stats_create_table(policy);
1489
1490                write_lock_irqsave(&cpufreq_driver_lock, flags);
1491                list_add(&policy->policy_list, &cpufreq_policy_list);
1492                write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1493        }
1494
1495        ret = cpufreq_init_policy(policy);
1496        if (ret) {
1497                pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
1498                       __func__, cpu, ret);
1499                goto out_destroy_policy;
1500        }
1501
1502        up_write(&policy->rwsem);
1503
1504        kobject_uevent(&policy->kobj, KOBJ_ADD);
1505
1506        /* Callback for handling stuff after policy is ready */
1507        if (cpufreq_driver->ready)
1508                cpufreq_driver->ready(policy);
1509
1510        if (cpufreq_thermal_control_enabled(cpufreq_driver))
1511                policy->cdev = of_cpufreq_cooling_register(policy);
1512
1513        pr_debug("initialization complete\n");
1514
1515        return 0;
1516
1517out_destroy_policy:
1518        for_each_cpu(j, policy->real_cpus)
1519                remove_cpu_dev_symlink(policy, get_cpu_device(j));
1520
1521        up_write(&policy->rwsem);
1522
1523out_offline_policy:
1524        if (cpufreq_driver->offline)
1525                cpufreq_driver->offline(policy);
1526
1527out_exit_policy:
1528        if (cpufreq_driver->exit)
1529                cpufreq_driver->exit(policy);
1530
1531out_free_policy:
1532        cpufreq_policy_free(policy);
1533        return ret;
1534}
1535
1536/**
1537 * cpufreq_add_dev - the cpufreq interface for a CPU device.
1538 * @dev: CPU device.
1539 * @sif: Subsystem interface structure pointer (not used)
1540 */
1541static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1542{
1543        struct cpufreq_policy *policy;
1544        unsigned cpu = dev->id;
1545        int ret;
1546
1547        dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
1548
1549        if (cpu_online(cpu)) {
1550                ret = cpufreq_online(cpu);
1551                if (ret)
1552                        return ret;
1553        }
1554
1555        /* Create sysfs link on CPU registration */
1556        policy = per_cpu(cpufreq_cpu_data, cpu);
1557        if (policy)
1558                add_cpu_dev_symlink(policy, cpu);
1559
1560        return 0;
1561}
1562
1563static int cpufreq_offline(unsigned int cpu)
1564{
1565        struct cpufreq_policy *policy;
1566        int ret;
1567
1568        pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1569
1570        policy = cpufreq_cpu_get_raw(cpu);
1571        if (!policy) {
1572                pr_debug("%s: No cpu_data found\n", __func__);
1573                return 0;
1574        }
1575
1576        down_write(&policy->rwsem);
1577        if (has_target())
1578                cpufreq_stop_governor(policy);
1579
1580        cpumask_clear_cpu(cpu, policy->cpus);
1581
1582        if (policy_is_inactive(policy)) {
1583                if (has_target())
1584                        strncpy(policy->last_governor, policy->governor->name,
1585                                CPUFREQ_NAME_LEN);
1586                else
1587                        policy->last_policy = policy->policy;
1588        } else if (cpu == policy->cpu) {
1589                /* Nominate new CPU */
1590                policy->cpu = cpumask_any(policy->cpus);
1591        }
1592
1593        /* Start governor again for active policy */
1594        if (!policy_is_inactive(policy)) {
1595                if (has_target()) {
1596                        ret = cpufreq_start_governor(policy);
1597                        if (ret)
1598                                pr_err("%s: Failed to start governor\n", __func__);
1599                }
1600
1601                goto unlock;
1602        }
1603
1604        if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
1605                cpufreq_cooling_unregister(policy->cdev);
1606                policy->cdev = NULL;
1607        }
1608
1609        if (cpufreq_driver->stop_cpu)
1610                cpufreq_driver->stop_cpu(policy);
1611
1612        if (has_target())
1613                cpufreq_exit_governor(policy);
1614
1615        /*
1616         * Perform the ->offline() during light-weight tear-down, as
1617         * that allows fast recovery when the CPU comes back.
1618         */
1619        if (cpufreq_driver->offline) {
1620                cpufreq_driver->offline(policy);
1621        } else if (cpufreq_driver->exit) {
1622                cpufreq_driver->exit(policy);
1623                policy->freq_table = NULL;
1624        }
1625
1626unlock:
1627        up_write(&policy->rwsem);
1628        return 0;
1629}
1630
1631/*
1632 * cpufreq_remove_dev - remove a CPU device
1633 *
1634 * Removes the cpufreq interface for a CPU device.
1635 */
1636static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1637{
1638        unsigned int cpu = dev->id;
1639        struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1640
1641        if (!policy)
1642                return;
1643
1644        if (cpu_online(cpu))
1645                cpufreq_offline(cpu);
1646
1647        cpumask_clear_cpu(cpu, policy->real_cpus);
1648        remove_cpu_dev_symlink(policy, dev);
1649
1650        if (cpumask_empty(policy->real_cpus)) {
1651                /* We did light-weight exit earlier, do full tear down now */
1652                if (cpufreq_driver->offline)
1653                        cpufreq_driver->exit(policy);
1654
1655                cpufreq_policy_free(policy);
1656        }
1657}
1658
1659/**
1660 * cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
1661 * @policy: Policy managing CPUs.
1662 * @new_freq: New CPU frequency.
1663 *
1664 * Adjust to the current frequency first and clean up later by either calling
1665 * cpufreq_update_policy(), or scheduling handle_update().
1666 */
1667static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
1668                                unsigned int new_freq)
1669{
1670        struct cpufreq_freqs freqs;
1671
1672        pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
1673                 policy->cur, new_freq);
1674
1675        freqs.old = policy->cur;
1676        freqs.new = new_freq;
1677
1678        cpufreq_freq_transition_begin(policy, &freqs);
1679        cpufreq_freq_transition_end(policy, &freqs, 0);
1680}
1681
1682static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
1683{
1684        unsigned int new_freq;
1685
1686        new_freq = cpufreq_driver->get(policy->cpu);
1687        if (!new_freq)
1688                return 0;
1689
1690        /*
1691         * If fast frequency switching is used with the given policy, the check
1692         * against policy->cur is pointless, so skip it in that case.
1693         */
1694        if (policy->fast_switch_enabled || !has_target())
1695                return new_freq;
1696
1697        if (policy->cur != new_freq) {
1698                cpufreq_out_of_sync(policy, new_freq);
1699                if (update)
1700                        schedule_work(&policy->update);
1701        }
1702
1703        return new_freq;
1704}
1705
1706/**
1707 * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1708 * @cpu: CPU number
1709 *
1710 * This is the last known freq, without actually getting it from the driver.
1711 * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1712 */
1713unsigned int cpufreq_quick_get(unsigned int cpu)
1714{
1715        struct cpufreq_policy *policy;
1716        unsigned int ret_freq = 0;
1717        unsigned long flags;
1718
1719        read_lock_irqsave(&cpufreq_driver_lock, flags);
1720
1721        if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
1722                ret_freq = cpufreq_driver->get(cpu);
1723                read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1724                return ret_freq;
1725        }
1726
1727        read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1728
1729        policy = cpufreq_cpu_get(cpu);
1730        if (policy) {
1731                ret_freq = policy->cur;
1732                cpufreq_cpu_put(policy);
1733        }
1734
1735        return ret_freq;
1736}
1737EXPORT_SYMBOL(cpufreq_quick_get);
1738
1739/**
1740 * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1741 * @cpu: CPU number
1742 *
1743 * Just return the max possible frequency for a given CPU.
1744 */
1745unsigned int cpufreq_quick_get_max(unsigned int cpu)
1746{
1747        struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1748        unsigned int ret_freq = 0;
1749
1750        if (policy) {
1751                ret_freq = policy->max;
1752                cpufreq_cpu_put(policy);
1753        }
1754
1755        return ret_freq;
1756}
1757EXPORT_SYMBOL(cpufreq_quick_get_max);
1758
1759/**
1760 * cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
1761 * @cpu: CPU number
1762 *
1763 * The default return value is the max_freq field of cpuinfo.
1764 */
1765__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
1766{
1767        struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1768        unsigned int ret_freq = 0;
1769
1770        if (policy) {
1771                ret_freq = policy->cpuinfo.max_freq;
1772                cpufreq_cpu_put(policy);
1773        }
1774
1775        return ret_freq;
1776}
1777EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
1778
1779static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
1780{
1781        if (unlikely(policy_is_inactive(policy)))
1782                return 0;
1783
1784        return cpufreq_verify_current_freq(policy, true);
1785}
1786
1787/**
1788 * cpufreq_get - get the current CPU frequency (in kHz)
1789 * @cpu: CPU number
1790 *
1791 * Get the CPU current (static) CPU frequency
1792 */
1793unsigned int cpufreq_get(unsigned int cpu)
1794{
1795        struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1796        unsigned int ret_freq = 0;
1797
1798        if (policy) {
1799                down_read(&policy->rwsem);
1800                if (cpufreq_driver->get)
1801                        ret_freq = __cpufreq_get(policy);
1802                up_read(&policy->rwsem);
1803
1804                cpufreq_cpu_put(policy);
1805        }
1806
1807        return ret_freq;
1808}
1809EXPORT_SYMBOL(cpufreq_get);
1810
1811static struct subsys_interface cpufreq_interface = {
1812        .name           = "cpufreq",
1813        .subsys         = &cpu_subsys,
1814        .add_dev        = cpufreq_add_dev,
1815        .remove_dev     = cpufreq_remove_dev,
1816};
1817
1818/*
1819 * In case platform wants some specific frequency to be configured
1820 * during suspend..
1821 */
1822int cpufreq_generic_suspend(struct cpufreq_policy *policy)
1823{
1824        int ret;
1825
1826        if (!policy->suspend_freq) {
1827                pr_debug("%s: suspend_freq not defined\n", __func__);
1828                return 0;
1829        }
1830
1831        pr_debug("%s: Setting suspend-freq: %u\n", __func__,
1832                        policy->suspend_freq);
1833
1834        ret = __cpufreq_driver_target(policy, policy->suspend_freq,
1835                        CPUFREQ_RELATION_H);
1836        if (ret)
1837                pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
1838                                __func__, policy->suspend_freq, ret);
1839
1840        return ret;
1841}
1842EXPORT_SYMBOL(cpufreq_generic_suspend);
1843
1844/**
1845 * cpufreq_suspend() - Suspend CPUFreq governors.
1846 *
1847 * Called during system wide Suspend/Hibernate cycles for suspending governors
1848 * as some platforms can't change frequency after this point in suspend cycle.
1849 * Because some of the devices (like: i2c, regulators, etc) they use for
1850 * changing frequency are suspended quickly after this point.
1851 */
1852void cpufreq_suspend(void)
1853{
1854        struct cpufreq_policy *policy;
1855
1856        if (!cpufreq_driver)
1857                return;
1858
1859        if (!has_target() && !cpufreq_driver->suspend)
1860                goto suspend;
1861
1862        pr_debug("%s: Suspending Governors\n", __func__);
1863
1864        for_each_active_policy(policy) {
1865                if (has_target()) {
1866                        down_write(&policy->rwsem);
1867                        cpufreq_stop_governor(policy);
1868                        up_write(&policy->rwsem);
1869                }
1870
1871                if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
1872                        pr_err("%s: Failed to suspend driver: %s\n", __func__,
1873                                cpufreq_driver->name);
1874        }
1875
1876suspend:
1877        cpufreq_suspended = true;
1878}
1879
1880/**
1881 * cpufreq_resume() - Resume CPUFreq governors.
1882 *
1883 * Called during system wide Suspend/Hibernate cycle for resuming governors that
1884 * are suspended with cpufreq_suspend().
1885 */
1886void cpufreq_resume(void)
1887{
1888        struct cpufreq_policy *policy;
1889        int ret;
1890
1891        if (!cpufreq_driver)
1892                return;
1893
1894        if (unlikely(!cpufreq_suspended))
1895                return;
1896
1897        cpufreq_suspended = false;
1898
1899        if (!has_target() && !cpufreq_driver->resume)
1900                return;
1901
1902        pr_debug("%s: Resuming Governors\n", __func__);
1903
1904        for_each_active_policy(policy) {
1905                if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
1906                        pr_err("%s: Failed to resume driver: %p\n", __func__,
1907                                policy);
1908                } else if (has_target()) {
1909                        down_write(&policy->rwsem);
1910                        ret = cpufreq_start_governor(policy);
1911                        up_write(&policy->rwsem);
1912
1913                        if (ret)
1914                                pr_err("%s: Failed to start governor for policy: %p\n",
1915                                       __func__, policy);
1916                }
1917        }
1918}
1919
1920/**
1921 * cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
1922 * @flags: Flags to test against the current cpufreq driver's flags.
1923 *
1924 * Assumes that the driver is there, so callers must ensure that this is the
1925 * case.
1926 */
1927bool cpufreq_driver_test_flags(u16 flags)
1928{
1929        return !!(cpufreq_driver->flags & flags);
1930}
1931
1932/**
1933 * cpufreq_get_current_driver - Return the current driver's name.
1934 *
1935 * Return the name string of the currently registered cpufreq driver or NULL if
1936 * none.
1937 */
1938const char *cpufreq_get_current_driver(void)
1939{
1940        if (cpufreq_driver)
1941                return cpufreq_driver->name;
1942
1943        return NULL;
1944}
1945EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1946
1947/**
1948 * cpufreq_get_driver_data - Return current driver data.
1949 *
1950 * Return the private data of the currently registered cpufreq driver, or NULL
1951 * if no cpufreq driver has been registered.
1952 */
1953void *cpufreq_get_driver_data(void)
1954{
1955        if (cpufreq_driver)
1956                return cpufreq_driver->driver_data;
1957
1958        return NULL;
1959}
1960EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
1961
1962/*********************************************************************
1963 *                     NOTIFIER LISTS INTERFACE                      *
1964 *********************************************************************/
1965
1966/**
1967 * cpufreq_register_notifier - Register a notifier with cpufreq.
1968 * @nb: notifier function to register.
1969 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
1970 *
1971 * Add a notifier to one of two lists: either a list of notifiers that run on
1972 * clock rate changes (once before and once after every transition), or a list
1973 * of notifiers that ron on cpufreq policy changes.
1974 *
1975 * This function may sleep and it has the same return values as
1976 * blocking_notifier_chain_register().
1977 */
1978int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1979{
1980        int ret;
1981
1982        if (cpufreq_disabled())
1983                return -EINVAL;
1984
1985        switch (list) {
1986        case CPUFREQ_TRANSITION_NOTIFIER:
1987                mutex_lock(&cpufreq_fast_switch_lock);
1988
1989                if (cpufreq_fast_switch_count > 0) {
1990                        mutex_unlock(&cpufreq_fast_switch_lock);
1991                        return -EBUSY;
1992                }
1993                ret = srcu_notifier_chain_register(
1994                                &cpufreq_transition_notifier_list, nb);
1995                if (!ret)
1996                        cpufreq_fast_switch_count--;
1997
1998                mutex_unlock(&cpufreq_fast_switch_lock);
1999                break;
2000        case CPUFREQ_POLICY_NOTIFIER:
2001                ret = blocking_notifier_chain_register(
2002                                &cpufreq_policy_notifier_list, nb);
2003                break;
2004        default:
2005                ret = -EINVAL;
2006        }
2007
2008        return ret;
2009}
2010EXPORT_SYMBOL(cpufreq_register_notifier);
2011
2012/**
2013 * cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
2014 * @nb: notifier block to be unregistered.
2015 * @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
2016 *
2017 * Remove a notifier from one of the cpufreq notifier lists.
2018 *
2019 * This function may sleep and it has the same return values as
2020 * blocking_notifier_chain_unregister().
2021 */
2022int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
2023{
2024        int ret;
2025
2026        if (cpufreq_disabled())
2027                return -EINVAL;
2028
2029        switch (list) {
2030        case CPUFREQ_TRANSITION_NOTIFIER:
2031                mutex_lock(&cpufreq_fast_switch_lock);
2032
2033                ret = srcu_notifier_chain_unregister(
2034                                &cpufreq_transition_notifier_list, nb);
2035                if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
2036                        cpufreq_fast_switch_count++;
2037
2038                mutex_unlock(&cpufreq_fast_switch_lock);
2039                break;
2040        case CPUFREQ_POLICY_NOTIFIER:
2041                ret = blocking_notifier_chain_unregister(
2042                                &cpufreq_policy_notifier_list, nb);
2043                break;
2044        default:
2045                ret = -EINVAL;
2046        }
2047
2048        return ret;
2049}
2050EXPORT_SYMBOL(cpufreq_unregister_notifier);
2051
2052
2053/*********************************************************************
2054 *                              GOVERNORS                            *
2055 *********************************************************************/
2056
2057/**
2058 * cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
2059 * @policy: cpufreq policy to switch the frequency for.
2060 * @target_freq: New frequency to set (may be approximate).
2061 *
2062 * Carry out a fast frequency switch without sleeping.
2063 *
2064 * The driver's ->fast_switch() callback invoked by this function must be
2065 * suitable for being called from within RCU-sched read-side critical sections
2066 * and it is expected to select the minimum available frequency greater than or
2067 * equal to @target_freq (CPUFREQ_RELATION_L).
2068 *
2069 * This function must not be called if policy->fast_switch_enabled is unset.
2070 *
2071 * Governors calling this function must guarantee that it will never be invoked
2072 * twice in parallel for the same policy and that it will never be called in
2073 * parallel with either ->target() or ->target_index() for the same policy.
2074 *
2075 * Returns the actual frequency set for the CPU.
2076 *
2077 * If 0 is returned by the driver's ->fast_switch() callback to indicate an
2078 * error condition, the hardware configuration must be preserved.
2079 */
2080unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
2081                                        unsigned int target_freq)
2082{
2083        unsigned int freq;
2084        int cpu;
2085
2086        target_freq = clamp_val(target_freq, policy->min, policy->max);
2087        freq = cpufreq_driver->fast_switch(policy, target_freq);
2088
2089        if (!freq)
2090                return 0;
2091
2092        policy->cur = freq;
2093        arch_set_freq_scale(policy->related_cpus, freq,
2094                            policy->cpuinfo.max_freq);
2095        cpufreq_stats_record_transition(policy, freq);
2096
2097        if (trace_cpu_frequency_enabled()) {
2098                for_each_cpu(cpu, policy->cpus)
2099                        trace_cpu_frequency(freq, cpu);
2100        }
2101
2102        return freq;
2103}
2104EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
2105
2106/**
2107 * cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
2108 * @cpu: Target CPU.
2109 * @min_perf: Minimum (required) performance level (units of @capacity).
2110 * @target_perf: Target (desired) performance level (units of @capacity).
2111 * @capacity: Capacity of the target CPU.
2112 *
2113 * Carry out a fast performance level switch of @cpu without sleeping.
2114 *
2115 * The driver's ->adjust_perf() callback invoked by this function must be
2116 * suitable for being called from within RCU-sched read-side critical sections
2117 * and it is expected to select a suitable performance level equal to or above
2118 * @min_perf and preferably equal to or below @target_perf.
2119 *
2120 * This function must not be called if policy->fast_switch_enabled is unset.
2121 *
2122 * Governors calling this function must guarantee that it will never be invoked
2123 * twice in parallel for the same CPU and that it will never be called in
2124 * parallel with either ->target() or ->target_index() or ->fast_switch() for
2125 * the same CPU.
2126 */
2127void cpufreq_driver_adjust_perf(unsigned int cpu,
2128                                 unsigned long min_perf,
2129                                 unsigned long target_perf,
2130                                 unsigned long capacity)
2131{
2132        cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
2133}
2134
2135/**
2136 * cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
2137 *
2138 * Return 'true' if the ->adjust_perf callback is present for the
2139 * current driver or 'false' otherwise.
2140 */
2141bool cpufreq_driver_has_adjust_perf(void)
2142{
2143        return !!cpufreq_driver->adjust_perf;
2144}
2145
2146/* Must set freqs->new to intermediate frequency */
2147static int __target_intermediate(struct cpufreq_policy *policy,
2148                                 struct cpufreq_freqs *freqs, int index)
2149{
2150        int ret;
2151
2152        freqs->new = cpufreq_driver->get_intermediate(policy, index);
2153
2154        /* We don't need to switch to intermediate freq */
2155        if (!freqs->new)
2156                return 0;
2157
2158        pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
2159                 __func__, policy->cpu, freqs->old, freqs->new);
2160
2161        cpufreq_freq_transition_begin(policy, freqs);
2162        ret = cpufreq_driver->target_intermediate(policy, index);
2163        cpufreq_freq_transition_end(policy, freqs, ret);
2164
2165        if (ret)
2166                pr_err("%s: Failed to change to intermediate frequency: %d\n",
2167                       __func__, ret);
2168
2169        return ret;
2170}
2171
2172static int __target_index(struct cpufreq_policy *policy, int index)
2173{
2174        struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
2175        unsigned int restore_freq, intermediate_freq = 0;
2176        unsigned int newfreq = policy->freq_table[index].frequency;
2177        int retval = -EINVAL;
2178        bool notify;
2179
2180        if (newfreq == policy->cur)
2181                return 0;
2182
2183        /* Save last value to restore later on errors */
2184        restore_freq = policy->cur;
2185
2186        notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
2187        if (notify) {
2188                /* Handle switching to intermediate frequency */
2189                if (cpufreq_driver->get_intermediate) {
2190                        retval = __target_intermediate(policy, &freqs, index);
2191                        if (retval)
2192                                return retval;
2193
2194                        intermediate_freq = freqs.new;
2195                        /* Set old freq to intermediate */
2196                        if (intermediate_freq)
2197                                freqs.old = freqs.new;
2198                }
2199
2200                freqs.new = newfreq;
2201                pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
2202                         __func__, policy->cpu, freqs.old, freqs.new);
2203
2204                cpufreq_freq_transition_begin(policy, &freqs);
2205        }
2206
2207        retval = cpufreq_driver->target_index(policy, index);
2208        if (retval)
2209                pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
2210                       retval);
2211
2212        if (notify) {
2213                cpufreq_freq_transition_end(policy, &freqs, retval);
2214
2215                /*
2216                 * Failed after setting to intermediate freq? Driver should have
2217                 * reverted back to initial frequency and so should we. Check
2218                 * here for intermediate_freq instead of get_intermediate, in
2219                 * case we haven't switched to intermediate freq at all.
2220                 */
2221                if (unlikely(retval && intermediate_freq)) {
2222                        freqs.old = intermediate_freq;
2223                        freqs.new = restore_freq;
2224                        cpufreq_freq_transition_begin(policy, &freqs);
2225                        cpufreq_freq_transition_end(policy, &freqs, 0);
2226                }
2227        }
2228
2229        return retval;
2230}
2231
2232int __cpufreq_driver_target(struct cpufreq_policy *policy,
2233                            unsigned int target_freq,
2234                            unsigned int relation)
2235{
2236        unsigned int old_target_freq = target_freq;
2237        int index;
2238
2239        if (cpufreq_disabled())
2240                return -ENODEV;
2241
2242        /* Make sure that target_freq is within supported range */
2243        target_freq = clamp_val(target_freq, policy->min, policy->max);
2244
2245        pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
2246                 policy->cpu, target_freq, relation, old_target_freq);
2247
2248        /*
2249         * This might look like a redundant call as we are checking it again
2250         * after finding index. But it is left intentionally for cases where
2251         * exactly same freq is called again and so we can save on few function
2252         * calls.
2253         */
2254        if (target_freq == policy->cur &&
2255            !(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
2256                return 0;
2257
2258        if (cpufreq_driver->target)
2259                return cpufreq_driver->target(policy, target_freq, relation);
2260
2261        if (!cpufreq_driver->target_index)
2262                return -EINVAL;
2263
2264        index = cpufreq_frequency_table_target(policy, target_freq, relation);
2265
2266        return __target_index(policy, index);
2267}
2268EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
2269
2270int cpufreq_driver_target(struct cpufreq_policy *policy,
2271                          unsigned int target_freq,
2272                          unsigned int relation)
2273{
2274        int ret;
2275
2276        down_write(&policy->rwsem);
2277
2278        ret = __cpufreq_driver_target(policy, target_freq, relation);
2279
2280        up_write(&policy->rwsem);
2281
2282        return ret;
2283}
2284EXPORT_SYMBOL_GPL(cpufreq_driver_target);
2285
2286__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
2287{
2288        return NULL;
2289}
2290
2291static int cpufreq_init_governor(struct cpufreq_policy *policy)
2292{
2293        int ret;
2294
2295        /* Don't start any governor operations if we are entering suspend */
2296        if (cpufreq_suspended)
2297                return 0;
2298        /*
2299         * Governor might not be initiated here if ACPI _PPC changed
2300         * notification happened, so check it.
2301         */
2302        if (!policy->governor)
2303                return -EINVAL;
2304
2305        /* Platform doesn't want dynamic frequency switching ? */
2306        if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
2307            cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
2308                struct cpufreq_governor *gov = cpufreq_fallback_governor();
2309
2310                if (gov) {
2311                        pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
2312                                policy->governor->name, gov->name);
2313                        policy->governor = gov;
2314                } else {
2315                        return -EINVAL;
2316                }
2317        }
2318
2319        if (!try_module_get(policy->governor->owner))
2320                return -EINVAL;
2321
2322        pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2323
2324        if (policy->governor->init) {
2325                ret = policy->governor->init(policy);
2326                if (ret) {
2327                        module_put(policy->governor->owner);
2328                        return ret;
2329                }
2330        }
2331
2332        policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
2333
2334        return 0;
2335}
2336
2337static void cpufreq_exit_governor(struct cpufreq_policy *policy)
2338{
2339        if (cpufreq_suspended || !policy->governor)
2340                return;
2341
2342        pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2343
2344        if (policy->governor->exit)
2345                policy->governor->exit(policy);
2346
2347        module_put(policy->governor->owner);
2348}
2349
2350int cpufreq_start_governor(struct cpufreq_policy *policy)
2351{
2352        int ret;
2353
2354        if (cpufreq_suspended)
2355                return 0;
2356
2357        if (!policy->governor)
2358                return -EINVAL;
2359
2360        pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2361
2362        if (cpufreq_driver->get)
2363                cpufreq_verify_current_freq(policy, false);
2364
2365        if (policy->governor->start) {
2366                ret = policy->governor->start(policy);
2367                if (ret)
2368                        return ret;
2369        }
2370
2371        if (policy->governor->limits)
2372                policy->governor->limits(policy);
2373
2374        return 0;
2375}
2376
2377void cpufreq_stop_governor(struct cpufreq_policy *policy)
2378{
2379        if (cpufreq_suspended || !policy->governor)
2380                return;
2381
2382        pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2383
2384        if (policy->governor->stop)
2385                policy->governor->stop(policy);
2386}
2387
2388static void cpufreq_governor_limits(struct cpufreq_policy *policy)
2389{
2390        if (cpufreq_suspended || !policy->governor)
2391                return;
2392
2393        pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
2394
2395        if (policy->governor->limits)
2396                policy->governor->limits(policy);
2397}
2398
2399int cpufreq_register_governor(struct cpufreq_governor *governor)
2400{
2401        int err;
2402
2403        if (!governor)
2404                return -EINVAL;
2405
2406        if (cpufreq_disabled())
2407                return -ENODEV;
2408
2409        mutex_lock(&cpufreq_governor_mutex);
2410
2411        err = -EBUSY;
2412        if (!find_governor(governor->name)) {
2413                err = 0;
2414                list_add(&governor->governor_list, &cpufreq_governor_list);
2415        }
2416
2417        mutex_unlock(&cpufreq_governor_mutex);
2418        return err;
2419}
2420EXPORT_SYMBOL_GPL(cpufreq_register_governor);
2421
2422void cpufreq_unregister_governor(struct cpufreq_governor *governor)
2423{
2424        struct cpufreq_policy *policy;
2425        unsigned long flags;
2426
2427        if (!governor)
2428                return;
2429
2430        if (cpufreq_disabled())
2431                return;
2432
2433        /* clear last_governor for all inactive policies */
2434        read_lock_irqsave(&cpufreq_driver_lock, flags);
2435        for_each_inactive_policy(policy) {
2436                if (!strcmp(policy->last_governor, governor->name)) {
2437                        policy->governor = NULL;
2438                        strcpy(policy->last_governor, "\0");
2439                }
2440        }
2441        read_unlock_irqrestore(&cpufreq_driver_lock, flags);
2442
2443        mutex_lock(&cpufreq_governor_mutex);
2444        list_del(&governor->governor_list);
2445        mutex_unlock(&cpufreq_governor_mutex);
2446}
2447EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
2448
2449
2450/*********************************************************************
2451 *                          POLICY INTERFACE                         *
2452 *********************************************************************/
2453
2454/**
2455 * cpufreq_get_policy - get the current cpufreq_policy
2456 * @policy: struct cpufreq_policy into which the current cpufreq_policy
2457 *      is written
2458 * @cpu: CPU to find the policy for
2459 *
2460 * Reads the current cpufreq policy.
2461 */
2462int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
2463{
2464        struct cpufreq_policy *cpu_policy;
2465        if (!policy)
2466                return -EINVAL;
2467
2468        cpu_policy = cpufreq_cpu_get(cpu);
2469        if (!cpu_policy)
2470                return -EINVAL;
2471
2472        memcpy(policy, cpu_policy, sizeof(*policy));
2473
2474        cpufreq_cpu_put(cpu_policy);
2475        return 0;
2476}
2477EXPORT_SYMBOL(cpufreq_get_policy);
2478
2479/**
2480 * cpufreq_set_policy - Modify cpufreq policy parameters.
2481 * @policy: Policy object to modify.
2482 * @new_gov: Policy governor pointer.
2483 * @new_pol: Policy value (for drivers with built-in governors).
2484 *
2485 * Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
2486 * limits to be set for the policy, update @policy with the verified limits
2487 * values and either invoke the driver's ->setpolicy() callback (if present) or
2488 * carry out a governor update for @policy.  That is, run the current governor's
2489 * ->limits() callback (if @new_gov points to the same object as the one in
2490 * @policy) or replace the governor for @policy with @new_gov.
2491 *
2492 * The cpuinfo part of @policy is not updated by this function.
2493 */
2494static int cpufreq_set_policy(struct cpufreq_policy *policy,
2495                              struct cpufreq_governor *new_gov,
2496                              unsigned int new_pol)
2497{
2498        struct cpufreq_policy_data new_data;
2499        struct cpufreq_governor *old_gov;
2500        int ret;
2501
2502        memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2503        new_data.freq_table = policy->freq_table;
2504        new_data.cpu = policy->cpu;
2505        /*
2506         * PM QoS framework collects all the requests from users and provide us
2507         * the final aggregated value here.
2508         */
2509        new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
2510        new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
2511
2512        pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
2513                 new_data.cpu, new_data.min, new_data.max);
2514
2515        /*
2516         * Verify that the CPU speed can be set within these limits and make sure
2517         * that min <= max.
2518         */
2519        ret = cpufreq_driver->verify(&new_data);
2520        if (ret)
2521                return ret;
2522
2523        policy->min = new_data.min;
2524        policy->max = new_data.max;
2525        trace_cpu_frequency_limits(policy);
2526
2527        policy->cached_target_freq = UINT_MAX;
2528
2529        pr_debug("new min and max freqs are %u - %u kHz\n",
2530                 policy->min, policy->max);
2531
2532        if (cpufreq_driver->setpolicy) {
2533                policy->policy = new_pol;
2534                pr_debug("setting range\n");
2535                return cpufreq_driver->setpolicy(policy);
2536        }
2537
2538        if (new_gov == policy->governor) {
2539                pr_debug("governor limits update\n");
2540                cpufreq_governor_limits(policy);
2541                return 0;
2542        }
2543
2544        pr_debug("governor switch\n");
2545
2546        /* save old, working values */
2547        old_gov = policy->governor;
2548        /* end old governor */
2549        if (old_gov) {
2550                cpufreq_stop_governor(policy);
2551                cpufreq_exit_governor(policy);
2552        }
2553
2554        /* start new governor */
2555        policy->governor = new_gov;
2556        ret = cpufreq_init_governor(policy);
2557        if (!ret) {
2558                ret = cpufreq_start_governor(policy);
2559                if (!ret) {
2560                        pr_debug("governor change\n");
2561                        sched_cpufreq_governor_change(policy, old_gov);
2562                        return 0;
2563                }
2564                cpufreq_exit_governor(policy);
2565        }
2566
2567        /* new governor failed, so re-start old one */
2568        pr_debug("starting governor %s failed\n", policy->governor->name);
2569        if (old_gov) {
2570                policy->governor = old_gov;
2571                if (cpufreq_init_governor(policy))
2572                        policy->governor = NULL;
2573                else
2574                        cpufreq_start_governor(policy);
2575        }
2576
2577        return ret;
2578}
2579
2580/**
2581 * cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
2582 * @cpu: CPU to re-evaluate the policy for.
2583 *
2584 * Update the current frequency for the cpufreq policy of @cpu and use
2585 * cpufreq_set_policy() to re-apply the min and max limits, which triggers the
2586 * evaluation of policy notifiers and the cpufreq driver's ->verify() callback
2587 * for the policy in question, among other things.
2588 */
2589void cpufreq_update_policy(unsigned int cpu)
2590{
2591        struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
2592
2593        if (!policy)
2594                return;
2595
2596        /*
2597         * BIOS might change freq behind our back
2598         * -> ask driver for current freq and notify governors about a change
2599         */
2600        if (cpufreq_driver->get && has_target() &&
2601            (cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
2602                goto unlock;
2603
2604        refresh_frequency_limits(policy);
2605
2606unlock:
2607        cpufreq_cpu_release(policy);
2608}
2609EXPORT_SYMBOL(cpufreq_update_policy);
2610
2611/**
2612 * cpufreq_update_limits - Update policy limits for a given CPU.
2613 * @cpu: CPU to update the policy limits for.
2614 *
2615 * Invoke the driver's ->update_limits callback if present or call
2616 * cpufreq_update_policy() for @cpu.
2617 */
2618void cpufreq_update_limits(unsigned int cpu)
2619{
2620        if (cpufreq_driver->update_limits)
2621                cpufreq_driver->update_limits(cpu);
2622        else
2623                cpufreq_update_policy(cpu);
2624}
2625EXPORT_SYMBOL_GPL(cpufreq_update_limits);
2626
2627/*********************************************************************
2628 *               BOOST                                               *
2629 *********************************************************************/
2630static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
2631{
2632        int ret;
2633
2634        if (!policy->freq_table)
2635                return -ENXIO;
2636
2637        ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
2638        if (ret) {
2639                pr_err("%s: Policy frequency update failed\n", __func__);
2640                return ret;
2641        }
2642
2643        ret = freq_qos_update_request(policy->max_freq_req, policy->max);
2644        if (ret < 0)
2645                return ret;
2646
2647        return 0;
2648}
2649
2650int cpufreq_boost_trigger_state(int state)
2651{
2652        struct cpufreq_policy *policy;
2653        unsigned long flags;
2654        int ret = 0;
2655
2656        if (cpufreq_driver->boost_enabled == state)
2657                return 0;
2658
2659        write_lock_irqsave(&cpufreq_driver_lock, flags);
2660        cpufreq_driver->boost_enabled = state;
2661        write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2662
2663        get_online_cpus();
2664        for_each_active_policy(policy) {
2665                ret = cpufreq_driver->set_boost(policy, state);
2666                if (ret)
2667                        goto err_reset_state;
2668        }
2669        put_online_cpus();
2670
2671        return 0;
2672
2673err_reset_state:
2674        put_online_cpus();
2675
2676        write_lock_irqsave(&cpufreq_driver_lock, flags);
2677        cpufreq_driver->boost_enabled = !state;
2678        write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2679
2680        pr_err("%s: Cannot %s BOOST\n",
2681               __func__, state ? "enable" : "disable");
2682
2683        return ret;
2684}
2685
2686static bool cpufreq_boost_supported(void)
2687{
2688        return cpufreq_driver->set_boost;
2689}
2690
2691static int create_boost_sysfs_file(void)
2692{
2693        int ret;
2694
2695        ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
2696        if (ret)
2697                pr_err("%s: cannot register global BOOST sysfs file\n",
2698                       __func__);
2699
2700        return ret;
2701}
2702
2703static void remove_boost_sysfs_file(void)
2704{
2705        if (cpufreq_boost_supported())
2706                sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
2707}
2708
2709int cpufreq_enable_boost_support(void)
2710{
2711        if (!cpufreq_driver)
2712                return -EINVAL;
2713
2714        if (cpufreq_boost_supported())
2715                return 0;
2716
2717        cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2718
2719        /* This will get removed on driver unregister */
2720        return create_boost_sysfs_file();
2721}
2722EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
2723
2724int cpufreq_boost_enabled(void)
2725{
2726        return cpufreq_driver->boost_enabled;
2727}
2728EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2729
2730/*********************************************************************
2731 *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2732 *********************************************************************/
2733static enum cpuhp_state hp_online;
2734
2735static int cpuhp_cpufreq_online(unsigned int cpu)
2736{
2737        cpufreq_online(cpu);
2738
2739        return 0;
2740}
2741
2742static int cpuhp_cpufreq_offline(unsigned int cpu)
2743{
2744        cpufreq_offline(cpu);
2745
2746        return 0;
2747}
2748
2749/**
2750 * cpufreq_register_driver - register a CPU Frequency driver
2751 * @driver_data: A struct cpufreq_driver containing the values#
2752 * submitted by the CPU Frequency driver.
2753 *
2754 * Registers a CPU Frequency driver to this core code. This code
2755 * returns zero on success, -EEXIST when another driver got here first
2756 * (and isn't unregistered in the meantime).
2757 *
2758 */
2759int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2760{
2761        unsigned long flags;
2762        int ret;
2763
2764        if (cpufreq_disabled())
2765                return -ENODEV;
2766
2767        /*
2768         * The cpufreq core depends heavily on the availability of device
2769         * structure, make sure they are available before proceeding further.
2770         */
2771        if (!get_cpu_device(0))
2772                return -EPROBE_DEFER;
2773
2774        if (!driver_data || !driver_data->verify || !driver_data->init ||
2775            !(driver_data->setpolicy || driver_data->target_index ||
2776                    driver_data->target) ||
2777             (driver_data->setpolicy && (driver_data->target_index ||
2778                    driver_data->target)) ||
2779             (!driver_data->get_intermediate != !driver_data->target_intermediate) ||
2780             (!driver_data->online != !driver_data->offline))
2781                return -EINVAL;
2782
2783        pr_debug("trying to register driver %s\n", driver_data->name);
2784
2785        /* Protect against concurrent CPU online/offline. */
2786        cpus_read_lock();
2787
2788        write_lock_irqsave(&cpufreq_driver_lock, flags);
2789        if (cpufreq_driver) {
2790                write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2791                ret = -EEXIST;
2792                goto out;
2793        }
2794        cpufreq_driver = driver_data;
2795        write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2796
2797        /*
2798         * Mark support for the scheduler's frequency invariance engine for
2799         * drivers that implement target(), target_index() or fast_switch().
2800         */
2801        if (!cpufreq_driver->setpolicy) {
2802                static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
2803                pr_debug("supports frequency invariance");
2804        }
2805
2806        if (driver_data->setpolicy)
2807                driver_data->flags |= CPUFREQ_CONST_LOOPS;
2808
2809        if (cpufreq_boost_supported()) {
2810                ret = create_boost_sysfs_file();
2811                if (ret)
2812                        goto err_null_driver;
2813        }
2814
2815        ret = subsys_interface_register(&cpufreq_interface);
2816        if (ret)
2817                goto err_boost_unreg;
2818
2819        if (unlikely(list_empty(&cpufreq_policy_list))) {
2820                /* if all ->init() calls failed, unregister */
2821                ret = -ENODEV;
2822                pr_debug("%s: No CPU initialized for driver %s\n", __func__,
2823                         driver_data->name);
2824                goto err_if_unreg;
2825        }
2826
2827        ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
2828                                                   "cpufreq:online",
2829                                                   cpuhp_cpufreq_online,
2830                                                   cpuhp_cpufreq_offline);
2831        if (ret < 0)
2832                goto err_if_unreg;
2833        hp_online = ret;
2834        ret = 0;
2835
2836        pr_debug("driver %s up and running\n", driver_data->name);
2837        goto out;
2838
2839err_if_unreg:
2840        subsys_interface_unregister(&cpufreq_interface);
2841err_boost_unreg:
2842        remove_boost_sysfs_file();
2843err_null_driver:
2844        write_lock_irqsave(&cpufreq_driver_lock, flags);
2845        cpufreq_driver = NULL;
2846        write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2847out:
2848        cpus_read_unlock();
2849        return ret;
2850}
2851EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2852
2853/*
2854 * cpufreq_unregister_driver - unregister the current CPUFreq driver
2855 *
2856 * Unregister the current CPUFreq driver. Only call this if you have
2857 * the right to do so, i.e. if you have succeeded in initialising before!
2858 * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2859 * currently not initialised.
2860 */
2861int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2862{
2863        unsigned long flags;
2864
2865        if (!cpufreq_driver || (driver != cpufreq_driver))
2866                return -EINVAL;
2867
2868        pr_debug("unregistering driver %s\n", driver->name);
2869
2870        /* Protect against concurrent cpu hotplug */
2871        cpus_read_lock();
2872        subsys_interface_unregister(&cpufreq_interface);
2873        remove_boost_sysfs_file();
2874        static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
2875        cpuhp_remove_state_nocalls_cpuslocked(hp_online);
2876
2877        write_lock_irqsave(&cpufreq_driver_lock, flags);
2878
2879        cpufreq_driver = NULL;
2880
2881        write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2882        cpus_read_unlock();
2883
2884        return 0;
2885}
2886EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2887
2888static int __init cpufreq_core_init(void)
2889{
2890        struct cpufreq_governor *gov = cpufreq_default_governor();
2891
2892        if (cpufreq_disabled())
2893                return -ENODEV;
2894
2895        cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
2896        BUG_ON(!cpufreq_global_kobject);
2897
2898        if (!strlen(default_governor))
2899                strncpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
2900
2901        return 0;
2902}
2903module_param(off, int, 0444);
2904module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
2905core_initcall(cpufreq_core_init);
2906