drivers/cpufreq/cpufreq.c

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