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