drivers/cpufreq/cpufreq.c

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