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  *
   7  *  Oct 2005 - Ashok Raj <ashok.raj@intel.com>
   8  *      Added handling for CPU hotplug
   9  *  Feb 2006 - Jacob Shin <jacob.shin@amd.com>
  10  *      Fix handling for CPU hotplug -- affected CPUs
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17
  18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  19
  20 #include <linux/kernel.h>
  21 #include <linux/module.h>
  22 #include <linux/init.h>
  23 #include <linux/notifier.h>
  24 #include <linux/cpufreq.h>
  25 #include <linux/delay.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/device.h>
  29 #include <linux/slab.h>
  30 #include <linux/cpu.h>
  31 #include <linux/completion.h>
  32 #include <linux/mutex.h>
  33 #include <linux/syscore_ops.h>
  34
  35 #include <trace/events/power.h>
  36
  37 /**
  38  * The "cpufreq driver" - the arch- or hardware-dependent low
  39  * level driver of CPUFreq support, and its spinlock. This lock
  40  * also protects the cpufreq_cpu_data array.
  41  */
  42 static struct cpufreq_driver *cpufreq_driver;
  43 static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
  44 #ifdef CONFIG_HOTPLUG_CPU
  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 #endif
  48 static DEFINE_RWLOCK(cpufreq_driver_lock);
  49
  50 /*
  51  * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
  52  * all cpufreq/hotplug/workqueue/etc related lock issues.
  53  *
  54  * The rules for this semaphore:
  55  * - Any routine that wants to read from the policy structure will
  56  *   do a down_read on this semaphore.
  57  * - Any routine that will write to the policy structure and/or may take away
  58  *   the policy altogether (eg. CPU hotplug), will hold this lock in write
  59  *   mode before doing so.
  60  *
  61  * Additional rules:
  62  * - Governor routines that can be called in cpufreq hotplug path should not
  63  *   take this sem as top level hotplug notifier handler takes this.
  64  * - Lock should not be held across
  65  *     __cpufreq_governor(data, CPUFREQ_GOV_STOP);
  66  */
  67 static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
  68 static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
  69
  70 #define lock_policy_rwsem(mode, cpu)                                    \
  71 static int lock_policy_rwsem_##mode(int cpu)                            \
  72 {                                                                       \
  73         int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);              \
  74         BUG_ON(policy_cpu == -1);                                       \
  75         down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));            \
  76                                                                         \
  77         return 0;                                                       \
  78 }
  79
  80 lock_policy_rwsem(read, cpu);
  81 lock_policy_rwsem(write, cpu);
  82
  83 #define unlock_policy_rwsem(mode, cpu)                                  \
  84 static void unlock_policy_rwsem_##mode(int cpu)                         \
  85 {                                                                       \
  86         int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);              \
  87         BUG_ON(policy_cpu == -1);                                       \
  88         up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));              \
  89 }
  90
  91 unlock_policy_rwsem(read, cpu);
  92 unlock_policy_rwsem(write, cpu);
  93
  94 /* internal prototypes */
  95 static int __cpufreq_governor(struct cpufreq_policy *policy,
  96                 unsigned int event);
  97 static unsigned int __cpufreq_get(unsigned int cpu);
  98 static void handle_update(struct work_struct *work);
  99
 100 /**
 101  * Two notifier lists: the "policy" list is involved in the
 102  * validation process for a new CPU frequency policy; the
 103  * "transition" list for kernel code that needs to handle
 104  * changes to devices when the CPU clock speed changes.
 105  * The mutex locks both lists.
 106  */
 107 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
 108 static struct srcu_notifier_head cpufreq_transition_notifier_list;
 109
 110 static bool init_cpufreq_transition_notifier_list_called;
 111 static int __init init_cpufreq_transition_notifier_list(void)
 112 {
 113         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
 114         init_cpufreq_transition_notifier_list_called = true;
 115         return 0;
 116 }
 117 pure_initcall(init_cpufreq_transition_notifier_list);
 118
 119 static int off __read_mostly;
 120 static int cpufreq_disabled(void)
 121 {
 122         return off;
 123 }
 124 void disable_cpufreq(void)
 125 {
 126         off = 1;
 127 }
 128 static LIST_HEAD(cpufreq_governor_list);
 129 static DEFINE_MUTEX(cpufreq_governor_mutex);
 130
 131 bool have_governor_per_policy(void)
 132 {
 133         return cpufreq_driver->have_governor_per_policy;
 134 }
 135
 136 static struct cpufreq_policy *__cpufreq_cpu_get(unsigned int cpu, bool sysfs)
 137 {
 138         struct cpufreq_policy *data;
 139         unsigned long flags;
 140
 141         if (cpu >= nr_cpu_ids)
 142                 goto err_out;
 143
 144         /* get the cpufreq driver */
 145         read_lock_irqsave(&cpufreq_driver_lock, flags);
 146
 147         if (!cpufreq_driver)
 148                 goto err_out_unlock;
 149
 150         if (!try_module_get(cpufreq_driver->owner))
 151                 goto err_out_unlock;
 152
 153
 154         /* get the CPU */
 155         data = per_cpu(cpufreq_cpu_data, cpu);
 156
 157         if (!data)
 158                 goto err_out_put_module;
 159
 160         if (!sysfs && !kobject_get(&data->kobj))
 161                 goto err_out_put_module;
 162
 163         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
 164         return data;
 165
 166 err_out_put_module:
 167         module_put(cpufreq_driver->owner);
 168 err_out_unlock:
 169         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
 170 err_out:
 171         return NULL;
 172 }
 173
 174 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
 175 {
 176         if (cpufreq_disabled())
 177                 return NULL;
 178
 179         return __cpufreq_cpu_get(cpu, false);
 180 }
 181 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 182
 183 static struct cpufreq_policy *cpufreq_cpu_get_sysfs(unsigned int cpu)
 184 {
 185         return __cpufreq_cpu_get(cpu, true);
 186 }
 187
 188 static void __cpufreq_cpu_put(struct cpufreq_policy *data, bool sysfs)
 189 {
 190         if (!sysfs)
 191                 kobject_put(&data->kobj);
 192         module_put(cpufreq_driver->owner);
 193 }
 194
 195 void cpufreq_cpu_put(struct cpufreq_policy *data)
 196 {
 197         if (cpufreq_disabled())
 198                 return;
 199
 200         __cpufreq_cpu_put(data, false);
 201 }
 202 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 203
 204 static void cpufreq_cpu_put_sysfs(struct cpufreq_policy *data)
 205 {
 206         __cpufreq_cpu_put(data, true);
 207 }
 208
 209 /*********************************************************************
 210  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 211  *********************************************************************/
 212
 213 /**
 214  * adjust_jiffies - adjust the system "loops_per_jiffy"
 215  *
 216  * This function alters the system "loops_per_jiffy" for the clock
 217  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 218  * systems as each CPU might be scaled differently. So, use the arch
 219  * per-CPU loops_per_jiffy value wherever possible.
 220  */
 221 #ifndef CONFIG_SMP
 222 static unsigned long l_p_j_ref;
 223 static unsigned int  l_p_j_ref_freq;
 224
 225 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 226 {
 227         if (ci->flags & CPUFREQ_CONST_LOOPS)
 228                 return;
 229
 230         if (!l_p_j_ref_freq) {
 231                 l_p_j_ref = loops_per_jiffy;
 232                 l_p_j_ref_freq = ci->old;
 233                 pr_debug("saving %lu as reference value for loops_per_jiffy; "
 234                         "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 235         }
 236         if ((val == CPUFREQ_POSTCHANGE  && ci->old != ci->new) ||
 237             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 238                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
 239                                                                 ci->new);
 240                 pr_debug("scaling loops_per_jiffy to %lu "
 241                         "for frequency %u kHz\n", loops_per_jiffy, ci->new);
 242         }
 243 }
 244 #else
 245 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 246 {
 247         return;
 248 }
 249 #endif
 250
 251
 252 /**
 253  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 254  * on frequency transition.
 255  *
 256  * This function calls the transition notifiers and the "adjust_jiffies"
 257  * function. It is called twice on all CPU frequency changes that have
 258  * external effects.
 259  */
 260 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 261 {
 262         struct cpufreq_policy *policy;
 263         unsigned long flags;
 264
 265         BUG_ON(irqs_disabled());
 266
 267         if (cpufreq_disabled())
 268                 return;
 269
 270         freqs->flags = cpufreq_driver->flags;
 271         pr_debug("notification %u of frequency transition to %u kHz\n",
 272                 state, freqs->new);
 273
 274         read_lock_irqsave(&cpufreq_driver_lock, flags);
 275         policy = per_cpu(cpufreq_cpu_data, freqs->cpu);
 276         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
 277
 278         switch (state) {
 279
 280         case CPUFREQ_PRECHANGE:
 281                 /* detect if the driver reported a value as "old frequency"
 282                  * which is not equal to what the cpufreq core thinks is
 283                  * "old frequency".
 284                  */
 285                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 286                         if ((policy) && (policy->cpu == freqs->cpu) &&
 287                             (policy->cur) && (policy->cur != freqs->old)) {
 288                                 pr_debug("Warning: CPU frequency is"
 289                                         " %u, cpufreq assumed %u kHz.\n",
 290                                         freqs->old, policy->cur);
 291                                 freqs->old = policy->cur;
 292                         }
 293                 }
 294                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 295                                 CPUFREQ_PRECHANGE, freqs);
 296                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 297                 break;
 298
 299         case CPUFREQ_POSTCHANGE:
 300                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 301                 pr_debug("FREQ: %lu - CPU: %lu", (unsigned long)freqs->new,
 302                         (unsigned long)freqs->cpu);
 303                 trace_cpu_frequency(freqs->new, freqs->cpu);
 304                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 305                                 CPUFREQ_POSTCHANGE, freqs);
 306                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 307                         policy->cur = freqs->new;
 308                 break;
 309         }
 310 }
 311 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 312
 313
 314
 315 /*********************************************************************
 316  *                          SYSFS INTERFACE                          *
 317  *********************************************************************/
 318
 319 static struct cpufreq_governor *__find_governor(const char *str_governor)
 320 {
 321         struct cpufreq_governor *t;
 322
 323         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
 324                 if (!strnicmp(str_governor, t->name, CPUFREQ_NAME_LEN))
 325                         return t;
 326
 327         return NULL;
 328 }
 329
 330 /**
 331  * cpufreq_parse_governor - parse a governor string
 332  */
 333 static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
 334                                 struct cpufreq_governor **governor)
 335 {
 336         int err = -EINVAL;
 337
 338         if (!cpufreq_driver)
 339                 goto out;
 340
 341         if (cpufreq_driver->setpolicy) {
 342                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 343                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 344                         err = 0;
 345                 } else if (!strnicmp(str_governor, "powersave",
 346                                                 CPUFREQ_NAME_LEN)) {
 347                         *policy = CPUFREQ_POLICY_POWERSAVE;
 348                         err = 0;
 349                 }
 350         } else if (cpufreq_driver->target) {
 351                 struct cpufreq_governor *t;
 352
 353                 mutex_lock(&cpufreq_governor_mutex);
 354
 355                 t = __find_governor(str_governor);
 356
 357                 if (t == NULL) {
 358                         int ret;
 359
 360                         mutex_unlock(&cpufreq_governor_mutex);
 361                         ret = request_module("cpufreq_%s", str_governor);
 362                         mutex_lock(&cpufreq_governor_mutex);
 363
 364                         if (ret == 0)
 365                                 t = __find_governor(str_governor);
 366                 }
 367
 368                 if (t != NULL) {
 369                         *governor = t;
 370                         err = 0;
 371                 }
 372
 373                 mutex_unlock(&cpufreq_governor_mutex);
 374         }
 375 out:
 376         return err;
 377 }
 378
 379
 380 /**
 381  * cpufreq_per_cpu_attr_read() / show_##file_name() -
 382  * print out cpufreq information
 383  *
 384  * Write out information from cpufreq_driver->policy[cpu]; object must be
 385  * "unsigned int".
 386  */
 387
 388 #define show_one(file_name, object)                     \
 389 static ssize_t show_##file_name                         \
 390 (struct cpufreq_policy *policy, char *buf)              \
 391 {                                                       \
 392         return sprintf(buf, "%u\n", policy->object);    \
 393 }
 394
 395 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 396 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 397 show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
 398 show_one(scaling_min_freq, min);
 399 show_one(scaling_max_freq, max);
 400 show_one(scaling_cur_freq, cur);
 401
 402 static int __cpufreq_set_policy(struct cpufreq_policy *data,
 403                                 struct cpufreq_policy *policy);
 404
 405 /**
 406  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 407  */
 408 #define store_one(file_name, object)                    \
 409 static ssize_t store_##file_name                                        \
 410 (struct cpufreq_policy *policy, const char *buf, size_t count)          \
 411 {                                                                       \
 412         unsigned int ret;                                               \
 413         struct cpufreq_policy new_policy;                               \
 414                                                                         \
 415         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 416         if (ret)                                                        \
 417                 return -EINVAL;                                         \
 418                                                                         \
 419         ret = sscanf(buf, "%u", &new_policy.object);                    \
 420         if (ret != 1)                                                   \
 421                 return -EINVAL;                                         \
 422                                                                         \
 423         ret = __cpufreq_set_policy(policy, &new_policy);                \
 424         policy->user_policy.object = policy->object;                    \
 425                                                                         \
 426         return ret ? ret : count;                                       \
 427 }
 428
 429 store_one(scaling_min_freq, min);
 430 store_one(scaling_max_freq, max);
 431
 432 /**
 433  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 434  */
 435 static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
 436                                         char *buf)
 437 {
 438         unsigned int cur_freq = __cpufreq_get(policy->cpu);
 439         if (!cur_freq)
 440                 return sprintf(buf, "<unknown>");
 441         return sprintf(buf, "%u\n", cur_freq);
 442 }
 443
 444
 445 /**
 446  * show_scaling_governor - show the current policy for the specified CPU
 447  */
 448 static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
 449 {
 450         if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
 451                 return sprintf(buf, "powersave\n");
 452         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 453                 return sprintf(buf, "performance\n");
 454         else if (policy->governor)
 455                 return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
 456                                 policy->governor->name);
 457         return -EINVAL;
 458 }
 459
 460
 461 /**
 462  * store_scaling_governor - store policy for the specified CPU
 463  */
 464 static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
 465                                         const char *buf, size_t count)
 466 {
 467         unsigned int ret;
 468         char    str_governor[16];
 469         struct cpufreq_policy new_policy;
 470
 471         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 472         if (ret)
 473                 return ret;
 474
 475         ret = sscanf(buf, "%15s", str_governor);
 476         if (ret != 1)
 477                 return -EINVAL;
 478
 479         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
 480                                                 &new_policy.governor))
 481                 return -EINVAL;
 482
 483         /* Do not use cpufreq_set_policy here or the user_policy.max
 484            will be wrongly overridden */
 485         ret = __cpufreq_set_policy(policy, &new_policy);
 486
 487         policy->user_policy.policy = policy->policy;
 488         policy->user_policy.governor = policy->governor;
 489
 490         if (ret)
 491                 return ret;
 492         else
 493                 return count;
 494 }
 495
 496 /**
 497  * show_scaling_driver - show the cpufreq driver currently loaded
 498  */
 499 static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
 500 {
 501         return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
 502 }
 503
 504 /**
 505  * show_scaling_available_governors - show the available CPUfreq governors
 506  */
 507 static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
 508                                                 char *buf)
 509 {
 510         ssize_t i = 0;
 511         struct cpufreq_governor *t;
 512
 513         if (!cpufreq_driver->target) {
 514                 i += sprintf(buf, "performance powersave");
 515                 goto out;
 516         }
 517
 518         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 519                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
 520                     - (CPUFREQ_NAME_LEN + 2)))
 521                         goto out;
 522                 i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
 523         }
 524 out:
 525         i += sprintf(&buf[i], "\n");
 526         return i;
 527 }
 528
 529 static ssize_t show_cpus(const struct cpumask *mask, char *buf)
 530 {
 531         ssize_t i = 0;
 532         unsigned int cpu;
 533
 534         for_each_cpu(cpu, mask) {
 535                 if (i)
 536                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 537                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 538                 if (i >= (PAGE_SIZE - 5))
 539                         break;
 540         }
 541         i += sprintf(&buf[i], "\n");
 542         return i;
 543 }
 544
 545 /**
 546  * show_related_cpus - show the CPUs affected by each transition even if
 547  * hw coordination is in use
 548  */
 549 static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
 550 {
 551         return show_cpus(policy->related_cpus, buf);
 552 }
 553
 554 /**
 555  * show_affected_cpus - show the CPUs affected by each transition
 556  */
 557 static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
 558 {
 559         return show_cpus(policy->cpus, buf);
 560 }
 561
 562 static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
 563                                         const char *buf, size_t count)
 564 {
 565         unsigned int freq = 0;
 566         unsigned int ret;
 567
 568         if (!policy->governor || !policy->governor->store_setspeed)
 569                 return -EINVAL;
 570
 571         ret = sscanf(buf, "%u", &freq);
 572         if (ret != 1)
 573                 return -EINVAL;
 574
 575         policy->governor->store_setspeed(policy, freq);
 576
 577         return count;
 578 }
 579
 580 static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
 581 {
 582         if (!policy->governor || !policy->governor->show_setspeed)
 583                 return sprintf(buf, "<unsupported>\n");
 584
 585         return policy->governor->show_setspeed(policy, buf);
 586 }
 587
 588 /**
 589  * show_bios_limit - show the current cpufreq HW/BIOS limitation
 590  */
 591 static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
 592 {
 593         unsigned int limit;
 594         int ret;
 595         if (cpufreq_driver->bios_limit) {
 596                 ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
 597                 if (!ret)
 598                         return sprintf(buf, "%u\n", limit);
 599         }
 600         return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
 601 }
 602
 603 cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
 604 cpufreq_freq_attr_ro(cpuinfo_min_freq);
 605 cpufreq_freq_attr_ro(cpuinfo_max_freq);
 606 cpufreq_freq_attr_ro(cpuinfo_transition_latency);
 607 cpufreq_freq_attr_ro(scaling_available_governors);
 608 cpufreq_freq_attr_ro(scaling_driver);
 609 cpufreq_freq_attr_ro(scaling_cur_freq);
 610 cpufreq_freq_attr_ro(bios_limit);
 611 cpufreq_freq_attr_ro(related_cpus);
 612 cpufreq_freq_attr_ro(affected_cpus);
 613 cpufreq_freq_attr_rw(scaling_min_freq);
 614 cpufreq_freq_attr_rw(scaling_max_freq);
 615 cpufreq_freq_attr_rw(scaling_governor);
 616 cpufreq_freq_attr_rw(scaling_setspeed);
 617
 618 static struct attribute *default_attrs[] = {
 619         &cpuinfo_min_freq.attr,
 620         &cpuinfo_max_freq.attr,
 621         &cpuinfo_transition_latency.attr,
 622         &scaling_min_freq.attr,
 623         &scaling_max_freq.attr,
 624         &affected_cpus.attr,
 625         &related_cpus.attr,
 626         &scaling_governor.attr,
 627         &scaling_driver.attr,
 628         &scaling_available_governors.attr,
 629         &scaling_setspeed.attr,
 630         NULL
 631 };
 632
 633 struct kobject *cpufreq_global_kobject;
 634 EXPORT_SYMBOL(cpufreq_global_kobject);
 635
 636 #define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
 637 #define to_attr(a) container_of(a, struct freq_attr, attr)
 638
 639 static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
 640 {
 641         struct cpufreq_policy *policy = to_policy(kobj);
 642         struct freq_attr *fattr = to_attr(attr);
 643         ssize_t ret = -EINVAL;
 644         policy = cpufreq_cpu_get_sysfs(policy->cpu);
 645         if (!policy)
 646                 goto no_policy;
 647
 648         if (lock_policy_rwsem_read(policy->cpu) < 0)
 649                 goto fail;
 650
 651         if (fattr->show)
 652                 ret = fattr->show(policy, buf);
 653         else
 654                 ret = -EIO;
 655
 656         unlock_policy_rwsem_read(policy->cpu);
 657 fail:
 658         cpufreq_cpu_put_sysfs(policy);
 659 no_policy:
 660         return ret;
 661 }
 662
 663 static ssize_t store(struct kobject *kobj, struct attribute *attr,
 664                      const char *buf, size_t count)
 665 {
 666         struct cpufreq_policy *policy = to_policy(kobj);
 667         struct freq_attr *fattr = to_attr(attr);
 668         ssize_t ret = -EINVAL;
 669         policy = cpufreq_cpu_get_sysfs(policy->cpu);
 670         if (!policy)
 671                 goto no_policy;
 672
 673         if (lock_policy_rwsem_write(policy->cpu) < 0)
 674                 goto fail;
 675
 676         if (fattr->store)
 677                 ret = fattr->store(policy, buf, count);
 678         else
 679                 ret = -EIO;
 680
 681         unlock_policy_rwsem_write(policy->cpu);
 682 fail:
 683         cpufreq_cpu_put_sysfs(policy);
 684 no_policy:
 685         return ret;
 686 }
 687
 688 static void cpufreq_sysfs_release(struct kobject *kobj)
 689 {
 690         struct cpufreq_policy *policy = to_policy(kobj);
 691         pr_debug("last reference is dropped\n");
 692         complete(&policy->kobj_unregister);
 693 }
 694
 695 static const struct sysfs_ops sysfs_ops = {
 696         .show   = show,
 697         .store  = store,
 698 };
 699
 700 static struct kobj_type ktype_cpufreq = {
 701         .sysfs_ops      = &sysfs_ops,
 702         .default_attrs  = default_attrs,
 703         .release        = cpufreq_sysfs_release,
 704 };
 705
 706 /* symlink affected CPUs */
 707 static int cpufreq_add_dev_symlink(unsigned int cpu,
 708                                    struct cpufreq_policy *policy)
 709 {
 710         unsigned int j;
 711         int ret = 0;
 712
 713         for_each_cpu(j, policy->cpus) {
 714                 struct cpufreq_policy *managed_policy;
 715                 struct device *cpu_dev;
 716
 717                 if (j == cpu)
 718                         continue;
 719
 720                 pr_debug("CPU %u already managed, adding link\n", j);
 721                 managed_policy = cpufreq_cpu_get(cpu);
 722                 cpu_dev = get_cpu_device(j);
 723                 ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
 724                                         "cpufreq");
 725                 if (ret) {
 726                         cpufreq_cpu_put(managed_policy);
 727                         return ret;
 728                 }
 729         }
 730         return ret;
 731 }
 732
 733 static int cpufreq_add_dev_interface(unsigned int cpu,
 734                                      struct cpufreq_policy *policy,
 735                                      struct device *dev)
 736 {
 737         struct cpufreq_policy new_policy;
 738         struct freq_attr **drv_attr;
 739         unsigned long flags;
 740         int ret = 0;
 741         unsigned int j;
 742
 743         /* prepare interface data */
 744         ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
 745                                    &dev->kobj, "cpufreq");
 746         if (ret)
 747                 return ret;
 748
 749         /* set up files for this cpu device */
 750         drv_attr = cpufreq_driver->attr;
 751         while ((drv_attr) && (*drv_attr)) {
 752                 ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 753                 if (ret)
 754                         goto err_out_kobj_put;
 755                 drv_attr++;
 756         }
 757         if (cpufreq_driver->get) {
 758                 ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 759                 if (ret)
 760                         goto err_out_kobj_put;
 761         }
 762         if (cpufreq_driver->target) {
 763                 ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 764                 if (ret)
 765                         goto err_out_kobj_put;
 766         }
 767         if (cpufreq_driver->bios_limit) {
 768                 ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
 769                 if (ret)
 770                         goto err_out_kobj_put;
 771         }
 772
 773         write_lock_irqsave(&cpufreq_driver_lock, flags);
 774         for_each_cpu(j, policy->cpus) {
 775                 per_cpu(cpufreq_cpu_data, j) = policy;
 776                 per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
 777         }
 778         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
 779
 780         ret = cpufreq_add_dev_symlink(cpu, policy);
 781         if (ret)
 782                 goto err_out_kobj_put;
 783
 784         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 785         /* assure that the starting sequence is run in __cpufreq_set_policy */
 786         policy->governor = NULL;
 787
 788         /* set default policy */
 789         ret = __cpufreq_set_policy(policy, &new_policy);
 790         policy->user_policy.policy = policy->policy;
 791         policy->user_policy.governor = policy->governor;
 792
 793         if (ret) {
 794                 pr_debug("setting policy failed\n");
 795                 if (cpufreq_driver->exit)
 796                         cpufreq_driver->exit(policy);
 797         }
 798         return ret;
 799
 800 err_out_kobj_put:
 801         kobject_put(&policy->kobj);
 802         wait_for_completion(&policy->kobj_unregister);
 803         return ret;
 804 }
 805
 806 #ifdef CONFIG_HOTPLUG_CPU
 807 static int cpufreq_add_policy_cpu(unsigned int cpu, unsigned int sibling,
 808                                   struct device *dev)
 809 {
 810         struct cpufreq_policy *policy;
 811         int ret = 0;
 812         unsigned long flags;
 813
 814         policy = cpufreq_cpu_get(sibling);
 815         WARN_ON(!policy);
 816
 817         __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
 818
 819         lock_policy_rwsem_write(sibling);
 820
 821         write_lock_irqsave(&cpufreq_driver_lock, flags);
 822
 823         cpumask_set_cpu(cpu, policy->cpus);
 824         per_cpu(cpufreq_policy_cpu, cpu) = policy->cpu;
 825         per_cpu(cpufreq_cpu_data, cpu) = policy;
 826         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
 827
 828         unlock_policy_rwsem_write(sibling);
 829
 830         __cpufreq_governor(policy, CPUFREQ_GOV_START);
 831         __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
 832
 833         ret = sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
 834         if (ret) {
 835                 cpufreq_cpu_put(policy);
 836                 return ret;
 837         }
 838
 839         return 0;
 840 }
 841 #endif
 842
 843 /**
 844  * cpufreq_add_dev - add a CPU device
 845  *
 846  * Adds the cpufreq interface for a CPU device.
 847  *
 848  * The Oracle says: try running cpufreq registration/unregistration concurrently
 849  * with with cpu hotplugging and all hell will break loose. Tried to clean this
 850  * mess up, but more thorough testing is needed. - Mathieu
 851  */
 852 static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
 853 {
 854         unsigned int j, cpu = dev->id;
 855         int ret = -ENOMEM;
 856         struct cpufreq_policy *policy;
 857         unsigned long flags;
 858 #ifdef CONFIG_HOTPLUG_CPU
 859         struct cpufreq_governor *gov;
 860         int sibling;
 861 #endif
 862
 863         if (cpu_is_offline(cpu))
 864                 return 0;
 865
 866         pr_debug("adding CPU %u\n", cpu);
 867
 868 #ifdef CONFIG_SMP
 869         /* check whether a different CPU already registered this
 870          * CPU because it is in the same boat. */
 871         policy = cpufreq_cpu_get(cpu);
 872         if (unlikely(policy)) {
 873                 cpufreq_cpu_put(policy);
 874                 return 0;
 875         }
 876
 877 #ifdef CONFIG_HOTPLUG_CPU
 878         /* Check if this cpu was hot-unplugged earlier and has siblings */
 879         read_lock_irqsave(&cpufreq_driver_lock, flags);
 880         for_each_online_cpu(sibling) {
 881                 struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
 882                 if (cp && cpumask_test_cpu(cpu, cp->related_cpus)) {
 883                         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
 884                         return cpufreq_add_policy_cpu(cpu, sibling, dev);
 885                 }
 886         }
 887         read_unlock_irqrestore(&cpufreq_driver_lock, flags);
 888 #endif
 889 #endif
 890
 891         if (!try_module_get(cpufreq_driver->owner)) {
 892                 ret = -EINVAL;
 893                 goto module_out;
 894         }
 895
 896         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 897         if (!policy)
 898                 goto nomem_out;
 899
 900         if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
 901                 goto err_free_policy;
 902
 903         if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
 904                 goto err_free_cpumask;
 905
 906         policy->cpu = cpu;
 907         policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
 908         cpumask_copy(policy->cpus, cpumask_of(cpu));
 909
 910         /* Initially set CPU itself as the policy_cpu */
 911         per_cpu(cpufreq_policy_cpu, cpu) = cpu;
 912
 913         init_completion(&policy->kobj_unregister);
 914         INIT_WORK(&policy->update, handle_update);
 915
 916         /* call driver. From then on the cpufreq must be able
 917          * to accept all calls to ->verify and ->setpolicy for this CPU
 918          */
 919         ret = cpufreq_driver->init(policy);
 920         if (ret) {
 921                 pr_debug("initialization failed\n");
 922                 goto err_set_policy_cpu;
 923         }
 924
 925         /* related cpus should atleast have policy->cpus */
 926         cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
 927
 928         /*
 929          * affected cpus must always be the one, which are online. We aren't
 930          * managing offline cpus here.
 931          */
 932         cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
 933
 934         policy->user_policy.min = policy->min;
 935         policy->user_policy.max = policy->max;
 936
 937         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
 938                                      CPUFREQ_START, policy);
 939
 940 #ifdef CONFIG_HOTPLUG_CPU
 941         gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
 942         if (gov) {
 943                 policy->governor = gov;
 944                 pr_debug("Restoring governor %s for cpu %d\n",
 945                        policy->governor->name, cpu);
 946         }
 947 #endif
 948
 949         ret = cpufreq_add_dev_interface(cpu, policy, dev);
 950         if (ret)
 951                 goto err_out_unregister;
 952
 953         kobject_uevent(&policy->kobj, KOBJ_ADD);
 954         module_put(cpufreq_driver->owner);
 955         pr_debug("initialization complete\n");
 956
 957         return 0;
 958
 959 err_out_unregister:
 960         write_lock_irqsave(&cpufreq_driver_lock, flags);
 961         for_each_cpu(j, policy->cpus)
 962                 per_cpu(cpufreq_cpu_data, j) = NULL;
 963         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
 964
 965         kobject_put(&policy->kobj);
 966         wait_for_completion(&policy->kobj_unregister);
 967
 968 err_set_policy_cpu:
 969         per_cpu(cpufreq_policy_cpu, cpu) = -1;
 970         free_cpumask_var(policy->related_cpus);
 971 err_free_cpumask:
 972         free_cpumask_var(policy->cpus);
 973 err_free_policy:
 974         kfree(policy);
 975 nomem_out:
 976         module_put(cpufreq_driver->owner);
 977 module_out:
 978         return ret;
 979 }
 980
 981 static void update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
 982 {
 983         int j;
 984
 985         policy->last_cpu = policy->cpu;
 986         policy->cpu = cpu;
 987
 988         for_each_cpu(j, policy->cpus)
 989                 per_cpu(cpufreq_policy_cpu, j) = cpu;
 990
 991 #ifdef CONFIG_CPU_FREQ_TABLE
 992         cpufreq_frequency_table_update_policy_cpu(policy);
 993 #endif
 994         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
 995                         CPUFREQ_UPDATE_POLICY_CPU, policy);
 996 }
 997
 998 /**
 999  * __cpufreq_remove_dev - remove a CPU device
1000  *
1001  * Removes the cpufreq interface for a CPU device.
1002  * Caller should already have policy_rwsem in write mode for this CPU.
1003  * This routine frees the rwsem before returning.
1004  */
1005 static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1006 {
1007         unsigned int cpu = dev->id, ret, cpus;
1008         unsigned long flags;
1009         struct cpufreq_policy *data;
1010         struct kobject *kobj;
1011         struct completion *cmp;
1012         struct device *cpu_dev;
1013
1014         pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
1015
1016         write_lock_irqsave(&cpufreq_driver_lock, flags);
1017
1018         data = per_cpu(cpufreq_cpu_data, cpu);
1019         per_cpu(cpufreq_cpu_data, cpu) = NULL;
1020
1021         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1022
1023         if (!data) {
1024                 pr_debug("%s: No cpu_data found\n", __func__);
1025                 return -EINVAL;
1026         }
1027
1028         if (cpufreq_driver->target)
1029                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1030
1031 #ifdef CONFIG_HOTPLUG_CPU
1032         if (!cpufreq_driver->setpolicy)
1033                 strncpy(per_cpu(cpufreq_cpu_governor, cpu),
1034                         data->governor->name, CPUFREQ_NAME_LEN);
1035 #endif
1036
1037         WARN_ON(lock_policy_rwsem_write(cpu));
1038         cpus = cpumask_weight(data->cpus);
1039         cpumask_clear_cpu(cpu, data->cpus);
1040         unlock_policy_rwsem_write(cpu);
1041
1042         if (cpu != data->cpu) {
1043                 sysfs_remove_link(&dev->kobj, "cpufreq");
1044         } else if (cpus > 1) {
1045                 /* first sibling now owns the new sysfs dir */
1046                 cpu_dev = get_cpu_device(cpumask_first(data->cpus));
1047                 sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
1048                 ret = kobject_move(&data->kobj, &cpu_dev->kobj);
1049                 if (ret) {
1050                         pr_err("%s: Failed to move kobj: %d", __func__, ret);
1051
1052                         WARN_ON(lock_policy_rwsem_write(cpu));
1053                         cpumask_set_cpu(cpu, data->cpus);
1054
1055                         write_lock_irqsave(&cpufreq_driver_lock, flags);
1056                         per_cpu(cpufreq_cpu_data, cpu) = data;
1057                         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1058
1059                         unlock_policy_rwsem_write(cpu);
1060
1061                         ret = sysfs_create_link(&cpu_dev->kobj, &data->kobj,
1062                                         "cpufreq");
1063                         return -EINVAL;
1064                 }
1065
1066                 WARN_ON(lock_policy_rwsem_write(cpu));
1067                 update_policy_cpu(data, cpu_dev->id);
1068                 unlock_policy_rwsem_write(cpu);
1069                 pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
1070                                 __func__, cpu_dev->id, cpu);
1071         }
1072
1073         pr_debug("%s: removing link, cpu: %d\n", __func__, cpu);
1074         cpufreq_cpu_put(data);
1075
1076         /* If cpu is last user of policy, free policy */
1077         if (cpus == 1) {
1078                 __cpufreq_governor(data, CPUFREQ_GOV_POLICY_EXIT);
1079
1080                 lock_policy_rwsem_read(cpu);
1081                 kobj = &data->kobj;
1082                 cmp = &data->kobj_unregister;
1083                 unlock_policy_rwsem_read(cpu);
1084                 kobject_put(kobj);
1085
1086                 /* we need to make sure that the underlying kobj is actually
1087                  * not referenced anymore by anybody before we proceed with
1088                  * unloading.
1089                  */
1090                 pr_debug("waiting for dropping of refcount\n");
1091                 wait_for_completion(cmp);
1092                 pr_debug("wait complete\n");
1093
1094                 if (cpufreq_driver->exit)
1095                         cpufreq_driver->exit(data);
1096
1097                 free_cpumask_var(data->related_cpus);
1098                 free_cpumask_var(data->cpus);
1099                 kfree(data);
1100         } else if (cpufreq_driver->target) {
1101                 __cpufreq_governor(data, CPUFREQ_GOV_START);
1102                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1103         }
1104
1105         per_cpu(cpufreq_policy_cpu, cpu) = -1;
1106         return 0;
1107 }
1108
1109
1110 static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
1111 {
1112         unsigned int cpu = dev->id;
1113         int retval;
1114
1115         if (cpu_is_offline(cpu))
1116                 return 0;
1117
1118         retval = __cpufreq_remove_dev(dev, sif);
1119         return retval;
1120 }
1121
1122
1123 static void handle_update(struct work_struct *work)
1124 {
1125         struct cpufreq_policy *policy =
1126                 container_of(work, struct cpufreq_policy, update);
1127         unsigned int cpu = policy->cpu;
1128         pr_debug("handle_update for cpu %u called\n", cpu);
1129         cpufreq_update_policy(cpu);
1130 }
1131
1132 /**
1133  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1134  *      @cpu: cpu number
1135  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
1136  *      @new_freq: CPU frequency the CPU actually runs at
1137  *
1138  *      We adjust to current frequency first, and need to clean up later.
1139  *      So either call to cpufreq_update_policy() or schedule handle_update()).
1140  */
1141 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1142                                 unsigned int new_freq)
1143 {
1144         struct cpufreq_freqs freqs;
1145
1146         pr_debug("Warning: CPU frequency out of sync: cpufreq and timing "
1147                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1148
1149         freqs.cpu = cpu;
1150         freqs.old = old_freq;
1151         freqs.new = new_freq;
1152         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1153         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1154 }
1155
1156
1157 /**
1158  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1159  * @cpu: CPU number
1160  *
1161  * This is the last known freq, without actually getting it from the driver.
1162  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1163  */
1164 unsigned int cpufreq_quick_get(unsigned int cpu)
1165 {
1166         struct cpufreq_policy *policy;
1167         unsigned int ret_freq = 0;
1168
1169         if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
1170                 return cpufreq_driver->get(cpu);
1171
1172         policy = cpufreq_cpu_get(cpu);
1173         if (policy) {
1174                 ret_freq = policy->cur;
1175                 cpufreq_cpu_put(policy);
1176         }
1177
1178         return ret_freq;
1179 }
1180 EXPORT_SYMBOL(cpufreq_quick_get);
1181
1182 /**
1183  * cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
1184  * @cpu: CPU number
1185  *
1186  * Just return the max possible frequency for a given CPU.
1187  */
1188 unsigned int cpufreq_quick_get_max(unsigned int cpu)
1189 {
1190         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1191         unsigned int ret_freq = 0;
1192
1193         if (policy) {
1194                 ret_freq = policy->max;
1195                 cpufreq_cpu_put(policy);
1196         }
1197
1198         return ret_freq;
1199 }
1200 EXPORT_SYMBOL(cpufreq_quick_get_max);
1201
1202
1203 static unsigned int __cpufreq_get(unsigned int cpu)
1204 {
1205         struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
1206         unsigned int ret_freq = 0;
1207
1208         if (!cpufreq_driver->get)
1209                 return ret_freq;
1210
1211         ret_freq = cpufreq_driver->get(cpu);
1212
1213         if (ret_freq && policy->cur &&
1214                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1215                 /* verify no discrepancy between actual and
1216                                         saved value exists */
1217                 if (unlikely(ret_freq != policy->cur)) {
1218                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1219                         schedule_work(&policy->update);
1220                 }
1221         }
1222
1223         return ret_freq;
1224 }
1225
1226 /**
1227  * cpufreq_get - get the current CPU frequency (in kHz)
1228  * @cpu: CPU number
1229  *
1230  * Get the CPU current (static) CPU frequency
1231  */
1232 unsigned int cpufreq_get(unsigned int cpu)
1233 {
1234         unsigned int ret_freq = 0;
1235         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1236
1237         if (!policy)
1238                 goto out;
1239
1240         if (unlikely(lock_policy_rwsem_read(cpu)))
1241                 goto out_policy;
1242
1243         ret_freq = __cpufreq_get(cpu);
1244
1245         unlock_policy_rwsem_read(cpu);
1246
1247 out_policy:
1248         cpufreq_cpu_put(policy);
1249 out:
1250         return ret_freq;
1251 }
1252 EXPORT_SYMBOL(cpufreq_get);
1253
1254 static struct subsys_interface cpufreq_interface = {
1255         .name           = "cpufreq",
1256         .subsys         = &cpu_subsys,
1257         .add_dev        = cpufreq_add_dev,
1258         .remove_dev     = cpufreq_remove_dev,
1259 };
1260
1261
1262 /**
1263  * cpufreq_bp_suspend - Prepare the boot CPU for system suspend.
1264  *
1265  * This function is only executed for the boot processor.  The other CPUs
1266  * have been put offline by means of CPU hotplug.
1267  */
1268 static int cpufreq_bp_suspend(void)
1269 {
1270         int ret = 0;
1271
1272         int cpu = smp_processor_id();
1273         struct cpufreq_policy *cpu_policy;
1274
1275         pr_debug("suspending cpu %u\n", cpu);
1276
1277         /* If there's no policy for the boot CPU, we have nothing to do. */
1278         cpu_policy = cpufreq_cpu_get(cpu);
1279         if (!cpu_policy)
1280                 return 0;
1281
1282         if (cpufreq_driver->suspend) {
1283                 ret = cpufreq_driver->suspend(cpu_policy);
1284                 if (ret)
1285                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1286                                         "step on CPU %u\n", cpu_policy->cpu);
1287         }
1288
1289         cpufreq_cpu_put(cpu_policy);
1290         return ret;
1291 }
1292
1293 /**
1294  * cpufreq_bp_resume - Restore proper frequency handling of the boot CPU.
1295  *
1296  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1297  *      2.) schedule call cpufreq_update_policy() ASAP as interrupts are
1298  *          restored. It will verify that the current freq is in sync with
1299  *          what we believe it to be. This is a bit later than when it
1300  *          should be, but nonethteless it's better than calling
1301  *          cpufreq_driver->get() here which might re-enable interrupts...
1302  *
1303  * This function is only executed for the boot CPU.  The other CPUs have not
1304  * been turned on yet.
1305  */
1306 static void cpufreq_bp_resume(void)
1307 {
1308         int ret = 0;
1309
1310         int cpu = smp_processor_id();
1311         struct cpufreq_policy *cpu_policy;
1312
1313         pr_debug("resuming cpu %u\n", cpu);
1314
1315         /* If there's no policy for the boot CPU, we have nothing to do. */
1316         cpu_policy = cpufreq_cpu_get(cpu);
1317         if (!cpu_policy)
1318                 return;
1319
1320         if (cpufreq_driver->resume) {
1321                 ret = cpufreq_driver->resume(cpu_policy);
1322                 if (ret) {
1323                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1324                                         "step on CPU %u\n", cpu_policy->cpu);
1325                         goto fail;
1326                 }
1327         }
1328
1329         schedule_work(&cpu_policy->update);
1330
1331 fail:
1332         cpufreq_cpu_put(cpu_policy);
1333 }
1334
1335 static struct syscore_ops cpufreq_syscore_ops = {
1336         .suspend        = cpufreq_bp_suspend,
1337         .resume         = cpufreq_bp_resume,
1338 };
1339
1340 /**
1341  *      cpufreq_get_current_driver - return current driver's name
1342  *
1343  *      Return the name string of the currently loaded cpufreq driver
1344  *      or NULL, if none.
1345  */
1346 const char *cpufreq_get_current_driver(void)
1347 {
1348         if (cpufreq_driver)
1349                 return cpufreq_driver->name;
1350
1351         return NULL;
1352 }
1353 EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
1354
1355 /*********************************************************************
1356  *                     NOTIFIER LISTS INTERFACE                      *
1357  *********************************************************************/
1358
1359 /**
1360  *      cpufreq_register_notifier - register a driver with cpufreq
1361  *      @nb: notifier function to register
1362  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1363  *
1364  *      Add a driver to one of two lists: either a list of drivers that
1365  *      are notified about clock rate changes (once before and once after
1366  *      the transition), or a list of drivers that are notified about
1367  *      changes in cpufreq policy.
1368  *
1369  *      This function may sleep, and has the same return conditions as
1370  *      blocking_notifier_chain_register.
1371  */
1372 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1373 {
1374         int ret;
1375
1376         if (cpufreq_disabled())
1377                 return -EINVAL;
1378
1379         WARN_ON(!init_cpufreq_transition_notifier_list_called);
1380
1381         switch (list) {
1382         case CPUFREQ_TRANSITION_NOTIFIER:
1383                 ret = srcu_notifier_chain_register(
1384                                 &cpufreq_transition_notifier_list, nb);
1385                 break;
1386         case CPUFREQ_POLICY_NOTIFIER:
1387                 ret = blocking_notifier_chain_register(
1388                                 &cpufreq_policy_notifier_list, nb);
1389                 break;
1390         default:
1391                 ret = -EINVAL;
1392         }
1393
1394         return ret;
1395 }
1396 EXPORT_SYMBOL(cpufreq_register_notifier);
1397
1398
1399 /**
1400  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1401  *      @nb: notifier block to be unregistered
1402  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1403  *
1404  *      Remove a driver from the CPU frequency notifier list.
1405  *
1406  *      This function may sleep, and has the same return conditions as
1407  *      blocking_notifier_chain_unregister.
1408  */
1409 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1410 {
1411         int ret;
1412
1413         if (cpufreq_disabled())
1414                 return -EINVAL;
1415
1416         switch (list) {
1417         case CPUFREQ_TRANSITION_NOTIFIER:
1418                 ret = srcu_notifier_chain_unregister(
1419                                 &cpufreq_transition_notifier_list, nb);
1420                 break;
1421         case CPUFREQ_POLICY_NOTIFIER:
1422                 ret = blocking_notifier_chain_unregister(
1423                                 &cpufreq_policy_notifier_list, nb);
1424                 break;
1425         default:
1426                 ret = -EINVAL;
1427         }
1428
1429         return ret;
1430 }
1431 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1432
1433
1434 /*********************************************************************
1435  *                              GOVERNORS                            *
1436  *********************************************************************/
1437
1438
1439 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1440                             unsigned int target_freq,
1441                             unsigned int relation)
1442 {
1443         int retval = -EINVAL;
1444         unsigned int old_target_freq = target_freq;
1445
1446         if (cpufreq_disabled())
1447                 return -ENODEV;
1448
1449         /* Make sure that target_freq is within supported range */
1450         if (target_freq > policy->max)
1451                 target_freq = policy->max;
1452         if (target_freq < policy->min)
1453                 target_freq = policy->min;
1454
1455         pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
1456                         policy->cpu, target_freq, relation, old_target_freq);
1457
1458         if (target_freq == policy->cur)
1459                 return 0;
1460
1461         if (cpufreq_driver->target)
1462                 retval = cpufreq_driver->target(policy, target_freq, relation);
1463
1464         return retval;
1465 }
1466 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1467
1468 int cpufreq_driver_target(struct cpufreq_policy *policy,
1469                           unsigned int target_freq,
1470                           unsigned int relation)
1471 {
1472         int ret = -EINVAL;
1473
1474         policy = cpufreq_cpu_get(policy->cpu);
1475         if (!policy)
1476                 goto no_policy;
1477
1478         if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1479                 goto fail;
1480
1481         ret = __cpufreq_driver_target(policy, target_freq, relation);
1482
1483         unlock_policy_rwsem_write(policy->cpu);
1484
1485 fail:
1486         cpufreq_cpu_put(policy);
1487 no_policy:
1488         return ret;
1489 }
1490 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1491
1492 int __cpufreq_driver_getavg(struct cpufreq_policy *policy, unsigned int cpu)
1493 {
1494         int ret = 0;
1495
1496         if (cpufreq_disabled())
1497                 return ret;
1498
1499         if (!cpufreq_driver->getavg)
1500                 return 0;
1501
1502         policy = cpufreq_cpu_get(policy->cpu);
1503         if (!policy)
1504                 return -EINVAL;
1505
1506         ret = cpufreq_driver->getavg(policy, cpu);
1507
1508         cpufreq_cpu_put(policy);
1509         return ret;
1510 }
1511 EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1512
1513 /*
1514  * when "event" is CPUFREQ_GOV_LIMITS
1515  */
1516
1517 static int __cpufreq_governor(struct cpufreq_policy *policy,
1518                                         unsigned int event)
1519 {
1520         int ret;
1521
1522         /* Only must be defined when default governor is known to have latency
1523            restrictions, like e.g. conservative or ondemand.
1524            That this is the case is already ensured in Kconfig
1525         */
1526 #ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
1527         struct cpufreq_governor *gov = &cpufreq_gov_performance;
1528 #else
1529         struct cpufreq_governor *gov = NULL;
1530 #endif
1531
1532         if (policy->governor->max_transition_latency &&
1533             policy->cpuinfo.transition_latency >
1534             policy->governor->max_transition_latency) {
1535                 if (!gov)
1536                         return -EINVAL;
1537                 else {
1538                         printk(KERN_WARNING "%s governor failed, too long"
1539                                " transition latency of HW, fallback"
1540                                " to %s governor\n",
1541                                policy->governor->name,
1542                                gov->name);
1543                         policy->governor = gov;
1544                 }
1545         }
1546
1547         if (!try_module_get(policy->governor->owner))
1548                 return -EINVAL;
1549
1550         pr_debug("__cpufreq_governor for CPU %u, event %u\n",
1551                                                 policy->cpu, event);
1552         ret = policy->governor->governor(policy, event);
1553
1554         if (!ret) {
1555                 if (event == CPUFREQ_GOV_POLICY_INIT)
1556                         policy->governor->initialized++;
1557                 else if (event == CPUFREQ_GOV_POLICY_EXIT)
1558                         policy->governor->initialized--;
1559         }
1560
1561         /* we keep one module reference alive for
1562                         each CPU governed by this CPU */
1563         if ((event != CPUFREQ_GOV_START) || ret)
1564                 module_put(policy->governor->owner);
1565         if ((event == CPUFREQ_GOV_STOP) && !ret)
1566                 module_put(policy->governor->owner);
1567
1568         return ret;
1569 }
1570
1571
1572 int cpufreq_register_governor(struct cpufreq_governor *governor)
1573 {
1574         int err;
1575
1576         if (!governor)
1577                 return -EINVAL;
1578
1579         if (cpufreq_disabled())
1580                 return -ENODEV;
1581
1582         mutex_lock(&cpufreq_governor_mutex);
1583
1584         governor->initialized = 0;
1585         err = -EBUSY;
1586         if (__find_governor(governor->name) == NULL) {
1587                 err = 0;
1588                 list_add(&governor->governor_list, &cpufreq_governor_list);
1589         }
1590
1591         mutex_unlock(&cpufreq_governor_mutex);
1592         return err;
1593 }
1594 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1595
1596
1597 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1598 {
1599 #ifdef CONFIG_HOTPLUG_CPU
1600         int cpu;
1601 #endif
1602
1603         if (!governor)
1604                 return;
1605
1606         if (cpufreq_disabled())
1607                 return;
1608
1609 #ifdef CONFIG_HOTPLUG_CPU
1610         for_each_present_cpu(cpu) {
1611                 if (cpu_online(cpu))
1612                         continue;
1613                 if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
1614                         strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
1615         }
1616 #endif
1617
1618         mutex_lock(&cpufreq_governor_mutex);
1619         list_del(&governor->governor_list);
1620         mutex_unlock(&cpufreq_governor_mutex);
1621         return;
1622 }
1623 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1624
1625
1626
1627 /*********************************************************************
1628  *                          POLICY INTERFACE                         *
1629  *********************************************************************/
1630
1631 /**
1632  * cpufreq_get_policy - get the current cpufreq_policy
1633  * @policy: struct cpufreq_policy into which the current cpufreq_policy
1634  *      is written
1635  *
1636  * Reads the current cpufreq policy.
1637  */
1638 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1639 {
1640         struct cpufreq_policy *cpu_policy;
1641         if (!policy)
1642                 return -EINVAL;
1643
1644         cpu_policy = cpufreq_cpu_get(cpu);
1645         if (!cpu_policy)
1646                 return -EINVAL;
1647
1648         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1649
1650         cpufreq_cpu_put(cpu_policy);
1651         return 0;
1652 }
1653 EXPORT_SYMBOL(cpufreq_get_policy);
1654
1655
1656 /*
1657  * data   : current policy.
1658  * policy : policy to be set.
1659  */
1660 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1661                                 struct cpufreq_policy *policy)
1662 {
1663         int ret = 0, failed = 1;
1664
1665         pr_debug("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1666                 policy->min, policy->max);
1667
1668         memcpy(&policy->cpuinfo, &data->cpuinfo,
1669                                 sizeof(struct cpufreq_cpuinfo));
1670
1671         if (policy->min > data->max || policy->max < data->min) {
1672                 ret = -EINVAL;
1673                 goto error_out;
1674         }
1675
1676         /* verify the cpu speed can be set within this limit */
1677         ret = cpufreq_driver->verify(policy);
1678         if (ret)
1679                 goto error_out;
1680
1681         /* adjust if necessary - all reasons */
1682         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1683                         CPUFREQ_ADJUST, policy);
1684
1685         /* adjust if necessary - hardware incompatibility*/
1686         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1687                         CPUFREQ_INCOMPATIBLE, policy);
1688
1689         /* verify the cpu speed can be set within this limit,
1690            which might be different to the first one */
1691         ret = cpufreq_driver->verify(policy);
1692         if (ret)
1693                 goto error_out;
1694
1695         /* notification of the new policy */
1696         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1697                         CPUFREQ_NOTIFY, policy);
1698
1699         data->min = policy->min;
1700         data->max = policy->max;
1701
1702         pr_debug("new min and max freqs are %u - %u kHz\n",
1703                                         data->min, data->max);
1704
1705         if (cpufreq_driver->setpolicy) {
1706                 data->policy = policy->policy;
1707                 pr_debug("setting range\n");
1708                 ret = cpufreq_driver->setpolicy(policy);
1709         } else {
1710                 if (policy->governor != data->governor) {
1711                         /* save old, working values */
1712                         struct cpufreq_governor *old_gov = data->governor;
1713
1714                         pr_debug("governor switch\n");
1715
1716                         /* end old governor */
1717                         if (data->governor) {
1718                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1719                                 __cpufreq_governor(data,
1720                                                 CPUFREQ_GOV_POLICY_EXIT);
1721                         }
1722
1723                         /* start new governor */
1724                         data->governor = policy->governor;
1725                         if (!__cpufreq_governor(data, CPUFREQ_GOV_POLICY_INIT)) {
1726                                 if (!__cpufreq_governor(data, CPUFREQ_GOV_START))
1727                                         failed = 0;
1728                                 else
1729                                         __cpufreq_governor(data,
1730                                                         CPUFREQ_GOV_POLICY_EXIT);
1731                         }
1732
1733                         if (failed) {
1734                                 /* new governor failed, so re-start old one */
1735                                 pr_debug("starting governor %s failed\n",
1736                                                         data->governor->name);
1737                                 if (old_gov) {
1738                                         data->governor = old_gov;
1739                                         __cpufreq_governor(data,
1740                                                         CPUFREQ_GOV_POLICY_INIT);
1741                                         __cpufreq_governor(data,
1742                                                            CPUFREQ_GOV_START);
1743                                 }
1744                                 ret = -EINVAL;
1745                                 goto error_out;
1746                         }
1747                         /* might be a policy change, too, so fall through */
1748                 }
1749                 pr_debug("governor: change or update limits\n");
1750                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1751         }
1752
1753 error_out:
1754         return ret;
1755 }
1756
1757 /**
1758  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1759  *      @cpu: CPU which shall be re-evaluated
1760  *
1761  *      Useful for policy notifiers which have different necessities
1762  *      at different times.
1763  */
1764 int cpufreq_update_policy(unsigned int cpu)
1765 {
1766         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1767         struct cpufreq_policy policy;
1768         int ret;
1769
1770         if (!data) {
1771                 ret = -ENODEV;
1772                 goto no_policy;
1773         }
1774
1775         if (unlikely(lock_policy_rwsem_write(cpu))) {
1776                 ret = -EINVAL;
1777                 goto fail;
1778         }
1779
1780         pr_debug("updating policy for CPU %u\n", cpu);
1781         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1782         policy.min = data->user_policy.min;
1783         policy.max = data->user_policy.max;
1784         policy.policy = data->user_policy.policy;
1785         policy.governor = data->user_policy.governor;
1786
1787         /* BIOS might change freq behind our back
1788           -> ask driver for current freq and notify governors about a change */
1789         if (cpufreq_driver->get) {
1790                 policy.cur = cpufreq_driver->get(cpu);
1791                 if (!data->cur) {
1792                         pr_debug("Driver did not initialize current freq");
1793                         data->cur = policy.cur;
1794                 } else {
1795                         if (data->cur != policy.cur && cpufreq_driver->target)
1796                                 cpufreq_out_of_sync(cpu, data->cur,
1797                                                                 policy.cur);
1798                 }
1799         }
1800
1801         ret = __cpufreq_set_policy(data, &policy);
1802
1803         unlock_policy_rwsem_write(cpu);
1804
1805 fail:
1806         cpufreq_cpu_put(data);
1807 no_policy:
1808         return ret;
1809 }
1810 EXPORT_SYMBOL(cpufreq_update_policy);
1811
1812 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1813                                         unsigned long action, void *hcpu)
1814 {
1815         unsigned int cpu = (unsigned long)hcpu;
1816         struct device *dev;
1817
1818         dev = get_cpu_device(cpu);
1819         if (dev) {
1820                 switch (action) {
1821                 case CPU_ONLINE:
1822                 case CPU_ONLINE_FROZEN:
1823                         cpufreq_add_dev(dev, NULL);
1824                         break;
1825                 case CPU_DOWN_PREPARE:
1826                 case CPU_DOWN_PREPARE_FROZEN:
1827                         __cpufreq_remove_dev(dev, NULL);
1828                         break;
1829                 case CPU_DOWN_FAILED:
1830                 case CPU_DOWN_FAILED_FROZEN:
1831                         cpufreq_add_dev(dev, NULL);
1832                         break;
1833                 }
1834         }
1835         return NOTIFY_OK;
1836 }
1837
1838 static struct notifier_block __refdata cpufreq_cpu_notifier = {
1839     .notifier_call = cpufreq_cpu_callback,
1840 };
1841
1842 /*********************************************************************
1843  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1844  *********************************************************************/
1845
1846 /**
1847  * cpufreq_register_driver - register a CPU Frequency driver
1848  * @driver_data: A struct cpufreq_driver containing the values#
1849  * submitted by the CPU Frequency driver.
1850  *
1851  *   Registers a CPU Frequency driver to this core code. This code
1852  * returns zero on success, -EBUSY when another driver got here first
1853  * (and isn't unregistered in the meantime).
1854  *
1855  */
1856 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1857 {
1858         unsigned long flags;
1859         int ret;
1860
1861         if (cpufreq_disabled())
1862                 return -ENODEV;
1863
1864         if (!driver_data || !driver_data->verify || !driver_data->init ||
1865             ((!driver_data->setpolicy) && (!driver_data->target)))
1866                 return -EINVAL;
1867
1868         pr_debug("trying to register driver %s\n", driver_data->name);
1869
1870         if (driver_data->setpolicy)
1871                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1872
1873         write_lock_irqsave(&cpufreq_driver_lock, flags);
1874         if (cpufreq_driver) {
1875                 write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1876                 return -EBUSY;
1877         }
1878         cpufreq_driver = driver_data;
1879         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1880
1881         ret = subsys_interface_register(&cpufreq_interface);
1882         if (ret)
1883                 goto err_null_driver;
1884
1885         if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1886                 int i;
1887                 ret = -ENODEV;
1888
1889                 /* check for at least one working CPU */
1890                 for (i = 0; i < nr_cpu_ids; i++)
1891                         if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
1892                                 ret = 0;
1893                                 break;
1894                         }
1895
1896                 /* if all ->init() calls failed, unregister */
1897                 if (ret) {
1898                         pr_debug("no CPU initialized for driver %s\n",
1899                                                         driver_data->name);
1900                         goto err_if_unreg;
1901                 }
1902         }
1903
1904         register_hotcpu_notifier(&cpufreq_cpu_notifier);
1905         pr_debug("driver %s up and running\n", driver_data->name);
1906
1907         return 0;
1908 err_if_unreg:
1909         subsys_interface_unregister(&cpufreq_interface);
1910 err_null_driver:
1911         write_lock_irqsave(&cpufreq_driver_lock, flags);
1912         cpufreq_driver = NULL;
1913         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1914         return ret;
1915 }
1916 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1917
1918
1919 /**
1920  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1921  *
1922  *    Unregister the current CPUFreq driver. Only call this if you have
1923  * the right to do so, i.e. if you have succeeded in initialising before!
1924  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1925  * currently not initialised.
1926  */
1927 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1928 {
1929         unsigned long flags;
1930
1931         if (!cpufreq_driver || (driver != cpufreq_driver))
1932                 return -EINVAL;
1933
1934         pr_debug("unregistering driver %s\n", driver->name);
1935
1936         subsys_interface_unregister(&cpufreq_interface);
1937         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1938
1939         write_lock_irqsave(&cpufreq_driver_lock, flags);
1940         cpufreq_driver = NULL;
1941         write_unlock_irqrestore(&cpufreq_driver_lock, flags);
1942
1943         return 0;
1944 }
1945 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1946
1947 static int __init cpufreq_core_init(void)
1948 {
1949         int cpu;
1950
1951         if (cpufreq_disabled())
1952                 return -ENODEV;
1953
1954         for_each_possible_cpu(cpu) {
1955                 per_cpu(cpufreq_policy_cpu, cpu) = -1;
1956                 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1957         }
1958
1959         cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
1960         BUG_ON(!cpufreq_global_kobject);
1961         register_syscore_ops(&cpufreq_syscore_ops);
1962
1963         return 0;
1964 }
1965 core_initcall(cpufreq_core_init);