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