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                 WARN_ON(kobject_move(&policy->kobj, &dev->kobj));
1095         } else {
1096                 policy->cpu = cpu;
1097         }
1098
1099         policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
1100         cpumask_copy(policy->cpus, cpumask_of(cpu));
1101
1102         init_completion(&policy->kobj_unregister);
1103         INIT_WORK(&policy->update, handle_update);
1104
1105         /* call driver. From then on the cpufreq must be able
1106          * to accept all calls to ->verify and ->setpolicy for this CPU
1107          */
1108         ret = cpufreq_driver->init(policy);
1109         if (ret) {
1110                 pr_debug("initialization failed\n");
1111                 goto err_set_policy_cpu;
1112         }
1113
1114         /* related cpus should atleast have policy->cpus */
1115         cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
1116
1117         /*
1118          * affected cpus must always be the one, which are online. We aren't
1119          * managing offline cpus here.
1120          */
1121         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
1122
1123         if (!frozen) {
1124                 policy->user_policy.min = policy->min;
1125                 policy->user_policy.max = policy->max;
1126         }
1127
1128         down_write(&policy->rwsem);
1129         write_lock_irqsave(&cpufreq_driver_lock, flags);
1130         for_each_cpu(j, policy->cpus)
1131                 per_cpu(cpufreq_cpu_data, j) = policy;
1132         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1133
1134         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
1135                 policy->cur = cpufreq_driver->get(policy->cpu);
1136                 if (!policy->cur) {
1137                         pr_err("%s: ->get() failed\n", __func__);
1138                         goto err_get_freq;
1139                 }
1140         }
1141
1142         /*
1143          * Sometimes boot loaders set CPU frequency to a value outside of
1144          * frequency table present with cpufreq core. In such cases CPU might be
1145          * unstable if it has to run on that frequency for long duration of time
1146          * and so its better to set it to a frequency which is specified in
1147          * freq-table. This also makes cpufreq stats inconsistent as
1148          * cpufreq-stats would fail to register because current frequency of CPU
1149          * isn't found in freq-table.
1150          *
1151          * Because we don't want this change to effect boot process badly, we go
1152          * for the next freq which is >= policy->cur ('cur' must be set by now,
1153          * otherwise we will end up setting freq to lowest of the table as 'cur'
1154          * is initialized to zero).
1155          *
1156          * We are passing target-freq as "policy->cur - 1" otherwise
1157          * __cpufreq_driver_target() would simply fail, as policy->cur will be
1158          * equal to target-freq.
1159          */
1160         if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
1161             && has_target()) {
1162                 /* Are we running at unknown frequency ? */
1163                 ret = cpufreq_frequency_table_get_index(policy, policy->cur);
1164                 if (ret == -EINVAL) {
1165                         /* Warn user and fix it */
1166                         pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
1167                                 __func__, policy->cpu, policy->cur);
1168                         ret = __cpufreq_driver_target(policy, policy->cur - 1,
1169                                 CPUFREQ_RELATION_L);
1170
1171                         /*
1172                          * Reaching here after boot in a few seconds may not
1173                          * mean that system will remain stable at "unknown"
1174                          * frequency for longer duration. Hence, a BUG_ON().
1175                          */
1176                         BUG_ON(ret);
1177                         pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
1178                                 __func__, policy->cpu, policy->cur);
1179                 }
1180         }
1181
1182         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1183                                      CPUFREQ_START, policy);
1184
1185 #ifdef CONFIG_HOTPLUG_CPU
1186         gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
1187         if (gov) {
1188                 policy->governor = gov;
1189                 pr_debug("Restoring governor %s for cpu %d\n",
1190                        policy->governor->name, cpu);
1191         }
1192 #endif
1193
1194         if (!frozen) {
1195                 ret = cpufreq_add_dev_interface(policy, dev);
1196                 if (ret)
1197                         goto err_out_unregister;
1198                 blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1199                                 CPUFREQ_CREATE_POLICY, policy);
1200         }
1201
1202         write_lock_irqsave(&cpufreq_driver_lock, flags);
1203         list_add(&policy->policy_list, &cpufreq_policy_list);
1204         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1205
1206         cpufreq_init_policy(policy);
1207
1208         if (!frozen) {
1209                 policy->user_policy.policy = policy->policy;
1210                 policy->user_policy.governor = policy->governor;
1211         }
1212         up_write(&policy->rwsem);
1213
1214         kobject_uevent(&policy->kobj, KOBJ_ADD);
1215         up_read(&cpufreq_rwsem);
1216
1217         pr_debug("initialization complete\n");
1218
1219         return 0;
1220
1221 err_out_unregister:
1222 err_get_freq:
1223         write_lock_irqsave(&cpufreq_driver_lock, flags);
1224         for_each_cpu(j, policy->cpus)
1225                 per_cpu(cpufreq_cpu_data, j) = NULL;
1226         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1227
1228         if (cpufreq_driver->exit)
1229                 cpufreq_driver->exit(policy);
1230 err_set_policy_cpu:
1231         if (frozen) {
1232                 /* Do not leave stale fallback data behind. */
1233                 per_cpu(cpufreq_cpu_data_fallback, cpu) = NULL;
1234                 cpufreq_policy_put_kobj(policy);
1235         }
1236         cpufreq_policy_free(policy);
1237
1238 nomem_out:
1239         up_read(&cpufreq_rwsem);
1240
1241         return ret;
1242 }
1243
1244 /**
1245  * cpufreq_add_dev - add a CPU device
1246  *
1247  * Adds the cpufreq interface for a CPU device.
1248  *
1249  * The Oracle says: try running cpufreq registration/unregistration concurrently
1250  * with with cpu hotplugging and all hell will break loose. Tried to clean this
1251  * mess up, but more thorough testing is needed. - Mathieu
1252  */
1253 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
1254 {
1255         return __cpufreq_add_dev(dev, sif, false);
1256 }
1257
1258 static int cpufreq_nominate_new_policy_cpu(struct cpufreq_policy *policy,
1259                                            unsigned int old_cpu)
1260 {
1261         struct device *cpu_dev;
1262         int ret;
1263
1264         /* first sibling now owns the new sysfs dir */
1265         cpu_dev = get_cpu_device(cpumask_any_but(policy->cpus, old_cpu));
1266
1267         sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
1268         ret = kobject_move(&policy->kobj, &cpu_dev->kobj);
1269         if (ret) {
1270                 pr_err("%s: Failed to move kobj: %d", __func__, ret);
1271
1272                 down_write(&policy->rwsem);
1273                 cpumask_set_cpu(old_cpu, policy->cpus);
1274                 up_write(&policy->rwsem);
1275
1276                 ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
1277                                         "cpufreq");
1278
1279                 return -EINVAL;
1280         }
1281
1282         return cpu_dev->id;
1283 }
1284
1285 static int __cpufreq_remove_dev_prepare(struct device *dev,
1286                                         struct subsys_interface *sif,
1287                                         bool frozen)
1288 {
1289         unsigned int cpu = dev->id, cpus;
1290         int new_cpu, ret;
1291         unsigned long flags;
1292         struct cpufreq_policy *policy;
1293
1294         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1295
1296         write_lock_irqsave(&cpufreq_driver_lock, flags);
1297
1298         policy = per_cpu(cpufreq_cpu_data, cpu);
1299
1300         /* Save the policy somewhere when doing a light-weight tear-down */
1301         if (frozen)
1302                 per_cpu(cpufreq_cpu_data_fallback, cpu) = policy;
1303
1304         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1305
1306         if (!policy) {
1307                 pr_debug("%s: No cpu_data found\n", __func__);
1308                 return -EINVAL;
1309         }
1310
1311         if (has_target()) {
1312                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
1313                 if (ret) {
1314                         pr_err("%s: Failed to stop governor\n", __func__);
1315                         return ret;
1316                 }
1317         }
1318
1319 #ifdef CONFIG_HOTPLUG_CPU
1320         if (!cpufreq_driver->setpolicy)
1321                 strncpy(per_cpu(cpufreq_cpu_governor, cpu),
1322                         policy->governor->name, CPUFREQ_NAME_LEN);
1323 #endif
1324
1325         down_read(&policy->rwsem);
1326         cpus = cpumask_weight(policy->cpus);
1327         up_read(&policy->rwsem);
1328
1329         if (cpu != policy->cpu) {
1330                 sysfs_remove_link(&dev->kobj, "cpufreq");
1331         } else if (cpus > 1) {
1332                 new_cpu = cpufreq_nominate_new_policy_cpu(policy, cpu);
1333                 if (new_cpu >= 0) {
1334                         update_policy_cpu(policy, new_cpu);
1335
1336                         if (!frozen) {
1337                                 pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
1338                                                 __func__, new_cpu, cpu);
1339                         }
1340                 }
1341         }
1342
1343         return 0;
1344 }
1345
1346 static int __cpufreq_remove_dev_finish(struct device *dev,
1347                                        struct subsys_interface *sif,
1348                                        bool frozen)
1349 {
1350         unsigned int cpu = dev->id, cpus;
1351         int ret;
1352         unsigned long flags;
1353         struct cpufreq_policy *policy;
1354
1355         read_lock_irqsave(&cpufreq_driver_lock, flags);
1356         policy = per_cpu(cpufreq_cpu_data, cpu);
1357         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1358
1359         if (!policy) {
1360                 pr_debug("%s: No cpu_data found\n", __func__);
1361                 return -EINVAL;
1362         }
1363
1364         down_write(&policy->rwsem);
1365         cpus = cpumask_weight(policy->cpus);
1366
1367         if (cpus > 1)
1368                 cpumask_clear_cpu(cpu, policy->cpus);
1369         up_write(&policy->rwsem);
1370
1371         /* If cpu is last user of policy, free policy */
1372         if (cpus == 1) {
1373                 if (has_target()) {
1374                         ret = __cpufreq_governor(policy,
1375                                         CPUFREQ_GOV_POLICY_EXIT);
1376                         if (ret) {
1377                                 pr_err("%s: Failed to exit governor\n",
1378                                                 __func__);
1379                                 return ret;
1380                         }
1381                 }
1382
1383                 if (!frozen)
1384                         cpufreq_policy_put_kobj(policy);
1385
1386                 /*
1387                  * Perform the ->exit() even during light-weight tear-down,
1388                  * since this is a core component, and is essential for the
1389                  * subsequent light-weight ->init() to succeed.
1390                  */
1391                 if (cpufreq_driver->exit)
1392                         cpufreq_driver->exit(policy);
1393
1394                 /* Remove policy from list of active policies */
1395                 write_lock_irqsave(&cpufreq_driver_lock, flags);
1396                 list_del(&policy->policy_list);
1397                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1398
1399                 if (!frozen)
1400                         cpufreq_policy_free(policy);
1401         } else {
1402                 if (has_target()) {
1403                         if ((ret = __cpufreq_governor(policy, CPUFREQ_GOV_START)) ||
1404                                         (ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))) {
1405                                 pr_err("%s: Failed to start governor\n",
1406                                                 __func__);
1407                                 return ret;
1408                         }
1409                 }
1410         }
1411
1412         per_cpu(cpufreq_cpu_data, cpu) = NULL;
1413         return 0;
1414 }
1415
1416 /**
1417  * cpufreq_remove_dev - remove a CPU device
1418  *
1419  * Removes the cpufreq interface for a CPU device.
1420  */
1421 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1422 {
1423         unsigned int cpu = dev->id;
1424         int ret;
1425
1426         if (cpu_is_offline(cpu))
1427                 return 0;
1428
1429         ret = __cpufreq_remove_dev_prepare(dev, sif, false);
1430
1431         if (!ret)
1432                 ret = __cpufreq_remove_dev_finish(dev, sif, false);
1433
1434         return ret;
1435 }
1436
1437 static void handle_update(struct work_struct *work)
1438 {
1439         struct cpufreq_policy *policy =
1440                 container_of(work, struct cpufreq_policy, update);
1441         unsigned int cpu = policy->cpu;
1442         pr_debug("handle_update for cpu %u called\n", cpu);
1443         cpufreq_update_policy(cpu);
1444 }
1445
1446 /**
1447  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
1448  *      in deep trouble.
1449  *      @cpu: cpu number
1450  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
1451  *      @new_freq: CPU frequency the CPU actually runs at
1452  *
1453  *      We adjust to current frequency first, and need to clean up later.
1454  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1455  */
1456 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1457                                 unsigned int new_freq)
1458 {
1459         struct cpufreq_policy *policy;
1460         struct cpufreq_freqs freqs;
1461         unsigned long flags;
1462
1463         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1464                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1465
1466         freqs.old = old_freq;
1467         freqs.new = new_freq;
1468
1469         read_lock_irqsave(&cpufreq_driver_lock, flags);
1470         policy = per_cpu(cpufreq_cpu_data, cpu);
1471         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
1472
1473         cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
1474         cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
1475 }
1476
1477 /**
1478  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1479  * @cpu: CPU number
1480  *
1481  * This is the last known freq, without actually getting it from the driver.
1482  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1483  */
1484 unsigned int cpufreq_quick_get(unsigned int cpu)
1485 {
1486         struct cpufreq_policy *policy;
1487         unsigned int ret_freq = 0;
1488
1489         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1490                 return cpufreq_driver->get(cpu);
1491
1492         policy = cpufreq_cpu_get(cpu);
1493         if (policy) {
1494                 ret_freq = policy->cur;
1495                 cpufreq_cpu_put(policy);
1496         }
1497
1498         return ret_freq;
1499 }
1500 EXPORT_SYMBOL(cpufreq_quick_get);
1501
1502 /**
1503  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1504  * @cpu: CPU number
1505  *
1506  * Just return the max possible frequency for a given CPU.
1507  */
1508 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1509 {
1510         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1511         unsigned int ret_freq = 0;
1512
1513         if (policy) {
1514                 ret_freq = policy->max;
1515                 cpufreq_cpu_put(policy);
1516         }
1517
1518         return ret_freq;
1519 }
1520 EXPORT_SYMBOL(cpufreq_quick_get_max);
1521
1522 static unsigned int __cpufreq_get(unsigned int cpu)
1523 {
1524         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1525         unsigned int ret_freq = 0;
1526
1527         if (!cpufreq_driver->get)
1528                 return ret_freq;
1529
1530         ret_freq = cpufreq_driver->get(cpu);
1531
1532         if (ret_freq && policy->cur &&
1533                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1534                 /* verify no discrepancy between actual and
1535                                         saved value exists */
1536                 if (unlikely(ret_freq != policy->cur)) {
1537                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1538                         schedule_work(&policy->update);
1539                 }
1540         }
1541
1542         return ret_freq;
1543 }
1544
1545 /**
1546  * cpufreq_get - get the current CPU frequency (in kHz)
1547  * @cpu: CPU number
1548  *
1549  * Get the CPU current (static) CPU frequency
1550  */
1551 unsigned int cpufreq_get(unsigned int cpu)
1552 {
1553         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1554         unsigned int ret_freq = 0;
1555
1556         if (policy) {
1557                 down_read(&policy->rwsem);
1558                 ret_freq = __cpufreq_get(cpu);
1559                 up_read(&policy->rwsem);
1560
1561                 cpufreq_cpu_put(policy);
1562         }
1563
1564         return ret_freq;
1565 }
1566 EXPORT_SYMBOL(cpufreq_get);
1567
1568 static struct subsys_interface cpufreq_interface = {
1569         .name           = "cpufreq",
1570         .subsys         = &cpu_subsys,
1571         .add_dev        = cpufreq_add_dev,
1572         .remove_dev     = cpufreq_remove_dev,
1573 };
1574
1575 /**
1576  * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1577  *
1578  * This function is only executed for the boot processor.  The other CPUs
1579  * have been put offline by means of CPU hotplug.
1580  */
1581 static int cpufreq_bp_suspend(void)
1582 {
1583         int ret = 0;
1584
1585         int cpu = smp_processor_id();
1586         struct cpufreq_policy *policy;
1587
1588         pr_debug("suspending cpu %u\n", cpu);
1589
1590         /* If there's no policy for the boot CPU, we have nothing to do. */
1591         policy = cpufreq_cpu_get(cpu);
1592         if (!policy)
1593                 return 0;
1594
1595         if (cpufreq_driver->suspend) {
1596                 ret = cpufreq_driver->suspend(policy);
1597                 if (ret)
1598                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1599                                         "step on CPU %u\n", policy->cpu);
1600         }
1601
1602         cpufreq_cpu_put(policy);
1603         return ret;
1604 }
1605
1606 /**
1607  * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1608  *
1609  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1610  *      2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1611  *          restored. It will verify that the current freq is in sync with
1612  *          what we believe it to be. This is a bit later than when it
1613  *          should be, but nonethteless it's better than calling
1614  *          cpufreq_driver->get() here which might re-enable interrupts...
1615  *
1616  * This function is only executed for the boot CPU.  The other CPUs have not
1617  * been turned on yet.
1618  */
1619 static void cpufreq_bp_resume(void)
1620 {
1621         int ret = 0;
1622
1623         int cpu = smp_processor_id();
1624         struct cpufreq_policy *policy;
1625
1626         pr_debug("resuming cpu %u\n", cpu);
1627
1628         /* If there's no policy for the boot CPU, we have nothing to do. */
1629         policy = cpufreq_cpu_get(cpu);
1630         if (!policy)
1631                 return;
1632
1633         if (cpufreq_driver->resume) {
1634                 ret = cpufreq_driver->resume(policy);
1635                 if (ret) {
1636                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1637                                         "step on CPU %u\n", policy->cpu);
1638                         goto fail;
1639                 }
1640         }
1641
1642         schedule_work(&policy->update);
1643
1644 fail:
1645         cpufreq_cpu_put(policy);
1646 }
1647
1648 static struct syscore_ops cpufreq_syscore_ops = {
1649         .suspend        = cpufreq_bp_suspend,
1650         .resume         = cpufreq_bp_resume,
1651 };
1652
1653 /**
1654  *      cpufreq_get_current_driver - return current driver's name
1655  *
1656  *      Return the name string of the currently loaded cpufreq driver
1657  *      or NULL, if none.
1658  */
1659 const char *cpufreq_get_current_driver(void)
1660 {
1661         if (cpufreq_driver)
1662                 return cpufreq_driver->name;
1663
1664         return NULL;
1665 }
1666 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1667
1668 /*********************************************************************
1669  *                     NOTIFIER LISTS INTERFACE                      *
1670  *********************************************************************/
1671
1672 /**
1673  *      cpufreq_register_notifier - register a driver with cpufreq
1674  *      @nb: notifier function to register
1675  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1676  *
1677  *      Add a driver to one of two lists: either a list of drivers that
1678  *      are notified about clock rate changes (once before and once after
1679  *      the transition), or a list of drivers that are notified about
1680  *      changes in cpufreq policy.
1681  *
1682  *      This function may sleep, and has the same return conditions as
1683  *      blocking_notifier_chain_register.
1684  */
1685 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1686 {
1687         int ret;
1688
1689         if (cpufreq_disabled())
1690                 return -EINVAL;
1691
1692         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1693
1694         switch (list) {
1695         case CPUFREQ_TRANSITION_NOTIFIER:
1696                 ret = srcu_notifier_chain_register(
1697                                 &cpufreq_transition_notifier_list, nb);
1698                 break;
1699         case CPUFREQ_POLICY_NOTIFIER:
1700                 ret = blocking_notifier_chain_register(
1701                                 &cpufreq_policy_notifier_list, nb);
1702                 break;
1703         default:
1704                 ret = -EINVAL;
1705         }
1706
1707         return ret;
1708 }
1709 EXPORT_SYMBOL(cpufreq_register_notifier);
1710
1711 /**
1712  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1713  *      @nb: notifier block to be unregistered
1714  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1715  *
1716  *      Remove a driver from the CPU frequency notifier list.
1717  *
1718  *      This function may sleep, and has the same return conditions as
1719  *      blocking_notifier_chain_unregister.
1720  */
1721 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1722 {
1723         int ret;
1724
1725         if (cpufreq_disabled())
1726                 return -EINVAL;
1727
1728         switch (list) {
1729         case CPUFREQ_TRANSITION_NOTIFIER:
1730                 ret = srcu_notifier_chain_unregister(
1731                                 &cpufreq_transition_notifier_list, nb);
1732                 break;
1733         case CPUFREQ_POLICY_NOTIFIER:
1734                 ret = blocking_notifier_chain_unregister(
1735                                 &cpufreq_policy_notifier_list, nb);
1736                 break;
1737         default:
1738                 ret = -EINVAL;
1739         }
1740
1741         return ret;
1742 }
1743 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1744
1745
1746 /*********************************************************************
1747  *                              GOVERNORS                            *
1748  *********************************************************************/
1749
1750 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1751                             unsigned int target_freq,
1752                             unsigned int relation)
1753 {
1754         int retval = -EINVAL;
1755         unsigned int old_target_freq = target_freq;
1756
1757         if (cpufreq_disabled())
1758                 return -ENODEV;
1759
1760         /* Make sure that target_freq is within supported range */
1761         if (target_freq > policy->max)
1762                 target_freq = policy->max;
1763         if (target_freq < policy->min)
1764                 target_freq = policy->min;
1765
1766         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1767                         policy->cpu, target_freq, relation, old_target_freq);
1768
1769         /*
1770          * This might look like a redundant call as we are checking it again
1771          * after finding index. But it is left intentionally for cases where
1772          * exactly same freq is called again and so we can save on few function
1773          * calls.
1774          */
1775         if (target_freq == policy->cur)
1776                 return 0;
1777
1778         if (cpufreq_driver->target)
1779                 retval = cpufreq_driver->target(policy, target_freq, relation);
1780         else if (cpufreq_driver->target_index) {
1781                 struct cpufreq_frequency_table *freq_table;
1782                 struct cpufreq_freqs freqs;
1783                 bool notify;
1784                 int index;
1785
1786                 freq_table = cpufreq_frequency_get_table(policy->cpu);
1787                 if (unlikely(!freq_table)) {
1788                         pr_err("%s: Unable to find freq_table\n", __func__);
1789                         goto out;
1790                 }
1791
1792                 retval = cpufreq_frequency_table_target(policy, freq_table,
1793                                 target_freq, relation, &index);
1794                 if (unlikely(retval)) {
1795                         pr_err("%s: Unable to find matching freq\n", __func__);
1796                         goto out;
1797                 }
1798
1799                 if (freq_table[index].frequency == policy->cur) {
1800                         retval = 0;
1801                         goto out;
1802                 }
1803
1804                 notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
1805
1806                 if (notify) {
1807                         freqs.old = policy->cur;
1808                         freqs.new = freq_table[index].frequency;
1809                         freqs.flags = 0;
1810
1811                         pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
1812                                         __func__, policy->cpu, freqs.old,
1813                                         freqs.new);
1814
1815                         cpufreq_notify_transition(policy, &freqs,
1816                                         CPUFREQ_PRECHANGE);
1817                 }
1818
1819                 retval = cpufreq_driver->target_index(policy, index);
1820                 if (retval)
1821                         pr_err("%s: Failed to change cpu frequency: %d\n",
1822                                         __func__, retval);
1823
1824                 if (notify)
1825                         cpufreq_notify_post_transition(policy, &freqs, retval);
1826         }
1827
1828 out:
1829         return retval;
1830 }
1831 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1832
1833 int cpufreq_driver_target(struct cpufreq_policy *policy,
1834                           unsigned int target_freq,
1835                           unsigned int relation)
1836 {
1837         int ret = -EINVAL;
1838
1839         down_write(&policy->rwsem);
1840
1841         ret = __cpufreq_driver_target(policy, target_freq, relation);
1842
1843         up_write(&policy->rwsem);
1844
1845         return ret;
1846 }
1847 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1848
1849 /*
1850  * when "event" is CPUFREQ_GOV_LIMITS
1851  */
1852
1853 static int __cpufreq_governor(struct cpufreq_policy *policy,
1854                                         unsigned int event)
1855 {
1856         int ret;
1857
1858         /* Only must be defined when default governor is known to have latency
1859            restrictions, like e.g. conservative or ondemand.
1860            That this is the case is already ensured in Kconfig
1861         */
1862 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1863         struct cpufreq_governor *gov = &cpufreq_gov_performance;
1864 #else
1865         struct cpufreq_governor *gov = NULL;
1866 #endif
1867
1868         if (policy->governor->max_transition_latency &&
1869             policy->cpuinfo.transition_latency >
1870             policy->governor->max_transition_latency) {
1871                 if (!gov)
1872                         return -EINVAL;
1873                 else {
1874                         printk(KERN_WARNING "%s governor failed, too long"
1875                                " transition latency of HW, fallback"
1876                                " to %s governor\n",
1877                                policy->governor->name,
1878                                gov->name);
1879                         policy->governor = gov;
1880                 }
1881         }
1882
1883         if (event == CPUFREQ_GOV_POLICY_INIT)
1884                 if (!try_module_get(policy->governor->owner))
1885                         return -EINVAL;
1886
1887         pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1888                                                 policy->cpu, event);
1889
1890         mutex_lock(&cpufreq_governor_lock);
1891         if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
1892             || (!policy->governor_enabled
1893             && (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
1894                 mutex_unlock(&cpufreq_governor_lock);
1895                 return -EBUSY;
1896         }
1897
1898         if (event == CPUFREQ_GOV_STOP)
1899                 policy->governor_enabled = false;
1900         else if (event == CPUFREQ_GOV_START)
1901                 policy->governor_enabled = true;
1902
1903         mutex_unlock(&cpufreq_governor_lock);
1904
1905         ret = policy->governor->governor(policy, event);
1906
1907         if (!ret) {
1908                 if (event == CPUFREQ_GOV_POLICY_INIT)
1909                         policy->governor->initialized++;
1910                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
1911                         policy->governor->initialized--;
1912         } else {
1913                 /* Restore original values */
1914                 mutex_lock(&cpufreq_governor_lock);
1915                 if (event == CPUFREQ_GOV_STOP)
1916                         policy->governor_enabled = true;
1917                 else if (event == CPUFREQ_GOV_START)
1918                         policy->governor_enabled = false;
1919                 mutex_unlock(&cpufreq_governor_lock);
1920         }
1921
1922         if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
1923                         ((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
1924                 module_put(policy->governor->owner);
1925
1926         return ret;
1927 }
1928
1929 int cpufreq_register_governor(struct cpufreq_governor *governor)
1930 {
1931         int err;
1932
1933         if (!governor)
1934                 return -EINVAL;
1935
1936         if (cpufreq_disabled())
1937                 return -ENODEV;
1938
1939         mutex_lock(&cpufreq_governor_mutex);
1940
1941         governor->initialized = 0;
1942         err = -EBUSY;
1943         if (__find_governor(governor->name) == NULL) {
1944                 err = 0;
1945                 list_add(&governor->governor_list, &cpufreq_governor_list);
1946         }
1947
1948         mutex_unlock(&cpufreq_governor_mutex);
1949         return err;
1950 }
1951 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1952
1953 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1954 {
1955 #ifdef CONFIG_HOTPLUG_CPU
1956         int cpu;
1957 #endif
1958
1959         if (!governor)
1960                 return;
1961
1962         if (cpufreq_disabled())
1963                 return;
1964
1965 #ifdef CONFIG_HOTPLUG_CPU
1966         for_each_present_cpu(cpu) {
1967                 if (cpu_online(cpu))
1968                         continue;
1969                 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1970                         strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1971         }
1972 #endif
1973
1974         mutex_lock(&cpufreq_governor_mutex);
1975         list_del(&governor->governor_list);
1976         mutex_unlock(&cpufreq_governor_mutex);
1977         return;
1978 }
1979 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1980
1981
1982 /*********************************************************************
1983  *                          POLICY INTERFACE                         *
1984  *********************************************************************/
1985
1986 /**
1987  * cpufreq_get_policy - get the current cpufreq_policy
1988  * @policy: struct cpufreq_policy into which the current cpufreq_policy
1989  *      is written
1990  *
1991  * Reads the current cpufreq policy.
1992  */
1993 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1994 {
1995         struct cpufreq_policy *cpu_policy;
1996         if (!policy)
1997                 return -EINVAL;
1998
1999         cpu_policy = cpufreq_cpu_get(cpu);
2000         if (!cpu_policy)
2001                 return -EINVAL;
2002
2003         memcpy(policy, cpu_policy, sizeof(*policy));
2004
2005         cpufreq_cpu_put(cpu_policy);
2006         return 0;
2007 }
2008 EXPORT_SYMBOL(cpufreq_get_policy);
2009
2010 /*
2011  * policy : current policy.
2012  * new_policy: policy to be set.
2013  */
2014 static int cpufreq_set_policy(struct cpufreq_policy *policy,
2015                                 struct cpufreq_policy *new_policy)
2016 {
2017         int ret = 0, failed = 1;
2018
2019         pr_debug("setting new policy for CPU %u: %u - %u kHz\n", new_policy->cpu,
2020                 new_policy->min, new_policy->max);
2021
2022         memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
2023
2024         if (new_policy->min > policy->max || new_policy->max < policy->min) {
2025                 ret = -EINVAL;
2026                 goto error_out;
2027         }
2028
2029         /* verify the cpu speed can be set within this limit */
2030         ret = cpufreq_driver->verify(new_policy);
2031         if (ret)
2032                 goto error_out;
2033
2034         /* adjust if necessary - all reasons */
2035         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2036                         CPUFREQ_ADJUST, new_policy);
2037
2038         /* adjust if necessary - hardware incompatibility*/
2039         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2040                         CPUFREQ_INCOMPATIBLE, new_policy);
2041
2042         /*
2043          * verify the cpu speed can be set within this limit, which might be
2044          * different to the first one
2045          */
2046         ret = cpufreq_driver->verify(new_policy);
2047         if (ret)
2048                 goto error_out;
2049
2050         /* notification of the new policy */
2051         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
2052                         CPUFREQ_NOTIFY, new_policy);
2053
2054         policy->min = new_policy->min;
2055         policy->max = new_policy->max;
2056
2057         pr_debug("new min and max freqs are %u - %u kHz\n",
2058                                         policy->min, policy->max);
2059
2060         if (cpufreq_driver->setpolicy) {
2061                 policy->policy = new_policy->policy;
2062                 pr_debug("setting range\n");
2063                 ret = cpufreq_driver->setpolicy(new_policy);
2064         } else {
2065                 if (new_policy->governor != policy->governor) {
2066                         /* save old, working values */
2067                         struct cpufreq_governor *old_gov = policy->governor;
2068
2069                         pr_debug("governor switch\n");
2070
2071                         /* end old governor */
2072                         if (policy->governor) {
2073                                 __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
2074                                 up_write(&policy->rwsem);
2075                                 __cpufreq_governor(policy,
2076                                                 CPUFREQ_GOV_POLICY_EXIT);
2077                                 down_write(&policy->rwsem);
2078                         }
2079
2080                         /* start new governor */
2081                         policy->governor = new_policy->governor;
2082                         if (!__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT)) {
2083                                 if (!__cpufreq_governor(policy, CPUFREQ_GOV_START)) {
2084                                         failed = 0;
2085                                 } else {
2086                                         up_write(&policy->rwsem);
2087                                         __cpufreq_governor(policy,
2088                                                         CPUFREQ_GOV_POLICY_EXIT);
2089                                         down_write(&policy->rwsem);
2090                                 }
2091                         }
2092
2093                         if (failed) {
2094                                 /* new governor failed, so re-start old one */
2095                                 pr_debug("starting governor %s failed\n",
2096                                                         policy->governor->name);
2097                                 if (old_gov) {
2098                                         policy->governor = old_gov;
2099                                         __cpufreq_governor(policy,
2100                                                         CPUFREQ_GOV_POLICY_INIT);
2101                                         __cpufreq_governor(policy,
2102                                                            CPUFREQ_GOV_START);
2103                                 }
2104                                 ret = -EINVAL;
2105                                 goto error_out;
2106                         }
2107                         /* might be a policy change, too, so fall through */
2108                 }
2109                 pr_debug("governor: change or update limits\n");
2110                 ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2111         }
2112
2113 error_out:
2114         return ret;
2115 }
2116
2117 /**
2118  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
2119  *      @cpu: CPU which shall be re-evaluated
2120  *
2121  *      Useful for policy notifiers which have different necessities
2122  *      at different times.
2123  */
2124 int cpufreq_update_policy(unsigned int cpu)
2125 {
2126         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
2127         struct cpufreq_policy new_policy;
2128         int ret;
2129
2130         if (!policy) {
2131                 ret = -ENODEV;
2132                 goto no_policy;
2133         }
2134
2135         down_write(&policy->rwsem);
2136
2137         pr_debug("updating policy for CPU %u\n", cpu);
2138         memcpy(&new_policy, policy, sizeof(*policy));
2139         new_policy.min = policy->user_policy.min;
2140         new_policy.max = policy->user_policy.max;
2141         new_policy.policy = policy->user_policy.policy;
2142         new_policy.governor = policy->user_policy.governor;
2143
2144         /*
2145          * BIOS might change freq behind our back
2146          * -> ask driver for current freq and notify governors about a change
2147          */
2148         if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
2149                 new_policy.cur = cpufreq_driver->get(cpu);
2150                 if (!policy->cur) {
2151                         pr_debug("Driver did not initialize current freq");
2152                         policy->cur = new_policy.cur;
2153                 } else {
2154                         if (policy->cur != new_policy.cur && has_target())
2155                                 cpufreq_out_of_sync(cpu, policy->cur,
2156                                                                 new_policy.cur);
2157                 }
2158         }
2159
2160         ret = cpufreq_set_policy(policy, &new_policy);
2161
2162         up_write(&policy->rwsem);
2163
2164         cpufreq_cpu_put(policy);
2165 no_policy:
2166         return ret;
2167 }
2168 EXPORT_SYMBOL(cpufreq_update_policy);
2169
2170 static int cpufreq_cpu_callback(struct notifier_block *nfb,
2171                                         unsigned long action, void *hcpu)
2172 {
2173         unsigned int cpu = (unsigned long)hcpu;
2174         struct device *dev;
2175         bool frozen = false;
2176
2177         dev = get_cpu_device(cpu);
2178         if (dev) {
2179
2180                 if (action & CPU_TASKS_FROZEN)
2181                         frozen = true;
2182
2183                 switch (action & ~CPU_TASKS_FROZEN) {
2184                 case CPU_ONLINE:
2185                         __cpufreq_add_dev(dev, NULL, frozen);
2186                         cpufreq_update_policy(cpu);
2187                         break;
2188
2189                 case CPU_DOWN_PREPARE:
2190                         __cpufreq_remove_dev_prepare(dev, NULL, frozen);
2191                         break;
2192
2193                 case CPU_POST_DEAD:
2194                         __cpufreq_remove_dev_finish(dev, NULL, frozen);
2195                         break;
2196
2197                 case CPU_DOWN_FAILED:
2198                         __cpufreq_add_dev(dev, NULL, frozen);
2199                         break;
2200                 }
2201         }
2202         return NOTIFY_OK;
2203 }
2204
2205 static struct notifier_block __refdata cpufreq_cpu_notifier = {
2206         .notifier_call = cpufreq_cpu_callback,
2207 };
2208
2209 /*********************************************************************
2210  *               BOOST                                               *
2211  *********************************************************************/
2212 static int cpufreq_boost_set_sw(int state)
2213 {
2214         struct cpufreq_frequency_table *freq_table;
2215         struct cpufreq_policy *policy;
2216         int ret = -EINVAL;
2217
2218         list_for_each_entry(policy, &cpufreq_policy_list, policy_list) {
2219                 freq_table = cpufreq_frequency_get_table(policy->cpu);
2220                 if (freq_table) {
2221                         ret = cpufreq_frequency_table_cpuinfo(policy,
2222                                                         freq_table);
2223                         if (ret) {
2224                                 pr_err("%s: Policy frequency update failed\n",
2225                                        __func__);
2226                                 break;
2227                         }
2228                         policy->user_policy.max = policy->max;
2229                         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
2230                 }
2231         }
2232
2233         return ret;
2234 }
2235
2236 int cpufreq_boost_trigger_state(int state)
2237 {
2238         unsigned long flags;
2239         int ret = 0;
2240
2241         if (cpufreq_driver->boost_enabled == state)
2242                 return 0;
2243
2244         write_lock_irqsave(&cpufreq_driver_lock, flags);
2245         cpufreq_driver->boost_enabled = state;
2246         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2247
2248         ret = cpufreq_driver->set_boost(state);
2249         if (ret) {
2250                 write_lock_irqsave(&cpufreq_driver_lock, flags);
2251                 cpufreq_driver->boost_enabled = !state;
2252                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2253
2254                 pr_err("%s: Cannot %s BOOST\n", __func__,
2255                        state ? "enable" : "disable");
2256         }
2257
2258         return ret;
2259 }
2260
2261 int cpufreq_boost_supported(void)
2262 {
2263         if (likely(cpufreq_driver))
2264                 return cpufreq_driver->boost_supported;
2265
2266         return 0;
2267 }
2268 EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
2269
2270 int cpufreq_boost_enabled(void)
2271 {
2272         return cpufreq_driver->boost_enabled;
2273 }
2274 EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
2275
2276 /*********************************************************************
2277  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
2278  *********************************************************************/
2279
2280 /**
2281  * cpufreq_register_driver - register a CPU Frequency driver
2282  * @driver_data: A struct cpufreq_driver containing the values#
2283  * submitted by the CPU Frequency driver.
2284  *
2285  * Registers a CPU Frequency driver to this core code. This code
2286  * returns zero on success, -EBUSY when another driver got here first
2287  * (and isn't unregistered in the meantime).
2288  *
2289  */
2290 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
2291 {
2292         unsigned long flags;
2293         int ret;
2294
2295         if (cpufreq_disabled())
2296                 return -ENODEV;
2297
2298         if (!driver_data || !driver_data->verify || !driver_data->init ||
2299             !(driver_data->setpolicy || driver_data->target_index ||
2300                     driver_data->target))
2301                 return -EINVAL;
2302
2303         pr_debug("trying to register driver %s\n", driver_data->name);
2304
2305         if (driver_data->setpolicy)
2306                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
2307
2308         write_lock_irqsave(&cpufreq_driver_lock, flags);
2309         if (cpufreq_driver) {
2310                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2311                 return -EEXIST;
2312         }
2313         cpufreq_driver = driver_data;
2314         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2315
2316         if (cpufreq_boost_supported()) {
2317                 /*
2318                  * Check if driver provides function to enable boost -
2319                  * if not, use cpufreq_boost_set_sw as default
2320                  */
2321                 if (!cpufreq_driver->set_boost)
2322                         cpufreq_driver->set_boost = cpufreq_boost_set_sw;
2323
2324                 ret = cpufreq_sysfs_create_file(&boost.attr);
2325                 if (ret) {
2326                         pr_err("%s: cannot register global BOOST sysfs file\n",
2327                                 __func__);
2328                         goto err_null_driver;
2329                 }
2330         }
2331
2332         ret = subsys_interface_register(&cpufreq_interface);
2333         if (ret)
2334                 goto err_boost_unreg;
2335
2336         if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
2337                 int i;
2338                 ret = -ENODEV;
2339
2340                 /* check for at least one working CPU */
2341                 for (i = 0; i < nr_cpu_ids; i++)
2342                         if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
2343                                 ret = 0;
2344                                 break;
2345                         }
2346
2347                 /* if all ->init() calls failed, unregister */
2348                 if (ret) {
2349                         pr_debug("no CPU initialized for driver %s\n",
2350                                                         driver_data->name);
2351                         goto err_if_unreg;
2352                 }
2353         }
2354
2355         register_hotcpu_notifier(&cpufreq_cpu_notifier);
2356         pr_debug("driver %s up and running\n", driver_data->name);
2357
2358         return 0;
2359 err_if_unreg:
2360         subsys_interface_unregister(&cpufreq_interface);
2361 err_boost_unreg:
2362         if (cpufreq_boost_supported())
2363                 cpufreq_sysfs_remove_file(&boost.attr);
2364 err_null_driver:
2365         write_lock_irqsave(&cpufreq_driver_lock, flags);
2366         cpufreq_driver = NULL;
2367         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2368         return ret;
2369 }
2370 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
2371
2372 /**
2373  * cpufreq_unregister_driver - unregister the current CPUFreq driver
2374  *
2375  * Unregister the current CPUFreq driver. Only call this if you have
2376  * the right to do so, i.e. if you have succeeded in initialising before!
2377  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
2378  * currently not initialised.
2379  */
2380 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
2381 {
2382         unsigned long flags;
2383
2384         if (!cpufreq_driver || (driver != cpufreq_driver))
2385                 return -EINVAL;
2386
2387         pr_debug("unregistering driver %s\n", driver->name);
2388
2389         subsys_interface_unregister(&cpufreq_interface);
2390         if (cpufreq_boost_supported())
2391                 cpufreq_sysfs_remove_file(&boost.attr);
2392
2393         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
2394
2395         down_write(&cpufreq_rwsem);
2396         write_lock_irqsave(&cpufreq_driver_lock, flags);
2397
2398         cpufreq_driver = NULL;
2399
2400         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
2401         up_write(&cpufreq_rwsem);
2402
2403         return 0;
2404 }
2405 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
2406
2407 static int __init cpufreq_core_init(void)
2408 {
2409         if (cpufreq_disabled())
2410                 return -ENODEV;
2411
2412         cpufreq_global_kobject = kobject_create();
2413         BUG_ON(!cpufreq_global_kobject);
2414         register_syscore_ops(&cpufreq_syscore_ops);
2415
2416         return 0;
2417 }
2418 core_initcall(cpufreq_core_init);