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