drivers/cpufreq/cpufreq.c

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