drivers/cpufreq/cpufreq.c

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