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 #include <linux/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/notifier.h>
  22 #include <linux/cpufreq.h>
  23 #include <linux/delay.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/device.h>
  27 #include <linux/slab.h>
  28 #include <linux/cpu.h>
  29 #include <linux/completion.h>
  30 #include <linux/mutex.h>
  31
  32 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
  33                                                 "cpufreq-core", msg)
  34
  35 /**
  36  * The "cpufreq driver" - the arch- or hardware-dependent low
  37  * level driver of CPUFreq support, and its spinlock. This lock
  38  * also protects the cpufreq_cpu_data array.
  39  */
  40 static struct cpufreq_driver *cpufreq_driver;
  41 static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
  42 static DEFINE_SPINLOCK(cpufreq_driver_lock);
  43
  44 /*
  45  * cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
  46  * all cpufreq/hotplug/workqueue/etc related lock issues.
  47  *
  48  * The rules for this semaphore:
  49  * - Any routine that wants to read from the policy structure will
  50  *   do a down_read on this semaphore.
  51  * - Any routine that will write to the policy structure and/or may take away
  52  *   the policy altogether (eg. CPU hotplug), will hold this lock in write
  53  *   mode before doing so.
  54  *
  55  * Additional rules:
  56  * - All holders of the lock should check to make sure that the CPU they
  57  *   are concerned with are online after they get the lock.
  58  * - Governor routines that can be called in cpufreq hotplug path should not
  59  *   take this sem as top level hotplug notifier handler takes this.
  60  */
  61 static DEFINE_PER_CPU(int, policy_cpu);
  62 static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
  63
  64 #define lock_policy_rwsem(mode, cpu)                                    \
  65 int lock_policy_rwsem_##mode                                            \
  66 (int cpu)                                                               \
  67 {                                                                       \
  68         int policy_cpu = per_cpu(policy_cpu, cpu);                      \
  69         BUG_ON(policy_cpu == -1);                                       \
  70         down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));            \
  71         if (unlikely(!cpu_online(cpu))) {                               \
  72                 up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu));      \
  73                 return -1;                                              \
  74         }                                                               \
  75                                                                         \
  76         return 0;                                                       \
  77 }
  78
  79 lock_policy_rwsem(read, cpu);
  80 EXPORT_SYMBOL_GPL(lock_policy_rwsem_read);
  81
  82 lock_policy_rwsem(write, cpu);
  83 EXPORT_SYMBOL_GPL(lock_policy_rwsem_write);
  84
  85 void unlock_policy_rwsem_read(int cpu)
  86 {
  87         int policy_cpu = per_cpu(policy_cpu, cpu);
  88         BUG_ON(policy_cpu == -1);
  89         up_read(&per_cpu(cpu_policy_rwsem, policy_cpu));
  90 }
  91 EXPORT_SYMBOL_GPL(unlock_policy_rwsem_read);
  92
  93 void unlock_policy_rwsem_write(int cpu)
  94 {
  95         int policy_cpu = per_cpu(policy_cpu, cpu);
  96         BUG_ON(policy_cpu == -1);
  97         up_write(&per_cpu(cpu_policy_rwsem, policy_cpu));
  98 }
  99 EXPORT_SYMBOL_GPL(unlock_policy_rwsem_write);
 100
 101
 102 /* internal prototypes */
 103 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
 104 static unsigned int __cpufreq_get(unsigned int cpu);
 105 static void handle_update(struct work_struct *work);
 106
 107 /**
 108  * Two notifier lists: the "policy" list is involved in the
 109  * validation process for a new CPU frequency policy; the
 110  * "transition" list for kernel code that needs to handle
 111  * changes to devices when the CPU clock speed changes.
 112  * The mutex locks both lists.
 113  */
 114 static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
 115 static struct srcu_notifier_head cpufreq_transition_notifier_list;
 116
 117 static int __init init_cpufreq_transition_notifier_list(void)
 118 {
 119         srcu_init_notifier_head(&cpufreq_transition_notifier_list);
 120         return 0;
 121 }
 122 pure_initcall(init_cpufreq_transition_notifier_list);
 123
 124 static LIST_HEAD(cpufreq_governor_list);
 125 static DEFINE_MUTEX (cpufreq_governor_mutex);
 126
 127 struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
 128 {
 129         struct cpufreq_policy *data;
 130         unsigned long flags;
 131
 132         if (cpu >= NR_CPUS)
 133                 goto err_out;
 134
 135         /* get the cpufreq driver */
 136         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 137
 138         if (!cpufreq_driver)
 139                 goto err_out_unlock;
 140
 141         if (!try_module_get(cpufreq_driver->owner))
 142                 goto err_out_unlock;
 143
 144
 145         /* get the CPU */
 146         data = cpufreq_cpu_data[cpu];
 147
 148         if (!data)
 149                 goto err_out_put_module;
 150
 151         if (!kobject_get(&data->kobj))
 152                 goto err_out_put_module;
 153
 154         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 155         return data;
 156
 157 err_out_put_module:
 158         module_put(cpufreq_driver->owner);
 159 err_out_unlock:
 160         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 161 err_out:
 162         return NULL;
 163 }
 164 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 165
 166
 167 void cpufreq_cpu_put(struct cpufreq_policy *data)
 168 {
 169         kobject_put(&data->kobj);
 170         module_put(cpufreq_driver->owner);
 171 }
 172 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 173
 174
 175 /*********************************************************************
 176  *                     UNIFIED DEBUG HELPERS                         *
 177  *********************************************************************/
 178 #ifdef CONFIG_CPU_FREQ_DEBUG
 179
 180 /* what part(s) of the CPUfreq subsystem are debugged? */
 181 static unsigned int debug;
 182
 183 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 184  * set or modify this value.
 185  */
 186 static unsigned int debug_ratelimit = 1;
 187
 188 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 189  * loading of a cpufreq driver, temporarily disabled when a new policy
 190  * is set, and disabled upon cpufreq driver removal
 191  */
 192 static unsigned int disable_ratelimit = 1;
 193 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 194
 195 static void cpufreq_debug_enable_ratelimit(void)
 196 {
 197         unsigned long flags;
 198
 199         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 200         if (disable_ratelimit)
 201                 disable_ratelimit--;
 202         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 203 }
 204
 205 static void cpufreq_debug_disable_ratelimit(void)
 206 {
 207         unsigned long flags;
 208
 209         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 210         disable_ratelimit++;
 211         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 212 }
 213
 214 void cpufreq_debug_printk(unsigned int type, const char *prefix,
 215                                                         const char *fmt, ...)
 216 {
 217         char s[256];
 218         va_list args;
 219         unsigned int len;
 220         unsigned long flags;
 221
 222         WARN_ON(!prefix);
 223         if (type & debug) {
 224                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 225                 if (!disable_ratelimit && debug_ratelimit
 226                                         && !printk_ratelimit()) {
 227                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 228                         return;
 229                 }
 230                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 231
 232                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 233
 234                 va_start(args, fmt);
 235                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 236                 va_end(args);
 237
 238                 printk(s);
 239
 240                 WARN_ON(len < 5);
 241         }
 242 }
 243 EXPORT_SYMBOL(cpufreq_debug_printk);
 244
 245
 246 module_param(debug, uint, 0644);
 247 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core,"
 248                         " 2 to debug drivers, and 4 to debug governors.");
 249
 250 module_param(debug_ratelimit, uint, 0644);
 251 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging:"
 252                                         " set to 0 to disable ratelimiting.");
 253
 254 #else /* !CONFIG_CPU_FREQ_DEBUG */
 255
 256 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 257 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 258
 259 #endif /* CONFIG_CPU_FREQ_DEBUG */
 260
 261
 262 /*********************************************************************
 263  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 264  *********************************************************************/
 265
 266 /**
 267  * adjust_jiffies - adjust the system "loops_per_jiffy"
 268  *
 269  * This function alters the system "loops_per_jiffy" for the clock
 270  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 271  * systems as each CPU might be scaled differently. So, use the arch
 272  * per-CPU loops_per_jiffy value wherever possible.
 273  */
 274 #ifndef CONFIG_SMP
 275 static unsigned long l_p_j_ref;
 276 static unsigned int  l_p_j_ref_freq;
 277
 278 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 279 {
 280         if (ci->flags & CPUFREQ_CONST_LOOPS)
 281                 return;
 282
 283         if (!l_p_j_ref_freq) {
 284                 l_p_j_ref = loops_per_jiffy;
 285                 l_p_j_ref_freq = ci->old;
 286                 dprintk("saving %lu as reference value for loops_per_jiffy;"
 287                         "freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 288         }
 289         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 290             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 291             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 292                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
 293                                                                 ci->new);
 294                 dprintk("scaling loops_per_jiffy to %lu"
 295                         "for frequency %u kHz\n", loops_per_jiffy, ci->new);
 296         }
 297 }
 298 #else
 299 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 300 {
 301         return;
 302 }
 303 #endif
 304
 305
 306 /**
 307  * cpufreq_notify_transition - call notifier chain and adjust_jiffies
 308  * on frequency transition.
 309  *
 310  * This function calls the transition notifiers and the "adjust_jiffies"
 311  * function. It is called twice on all CPU frequency changes that have
 312  * external effects.
 313  */
 314 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 315 {
 316         struct cpufreq_policy *policy;
 317
 318         BUG_ON(irqs_disabled());
 319
 320         freqs->flags = cpufreq_driver->flags;
 321         dprintk("notification %u of frequency transition to %u kHz\n",
 322                 state, freqs->new);
 323
 324         policy = cpufreq_cpu_data[freqs->cpu];
 325         switch (state) {
 326
 327         case CPUFREQ_PRECHANGE:
 328                 /* detect if the driver reported a value as "old frequency"
 329                  * which is not equal to what the cpufreq core thinks is
 330                  * "old frequency".
 331                  */
 332                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 333                         if ((policy) && (policy->cpu == freqs->cpu) &&
 334                             (policy->cur) && (policy->cur != freqs->old)) {
 335                                 dprintk("Warning: CPU frequency is"
 336                                         " %u, cpufreq assumed %u kHz.\n",
 337                                         freqs->old, policy->cur);
 338                                 freqs->old = policy->cur;
 339                         }
 340                 }
 341                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 342                                 CPUFREQ_PRECHANGE, freqs);
 343                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 344                 break;
 345
 346         case CPUFREQ_POSTCHANGE:
 347                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 348                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
 349                                 CPUFREQ_POSTCHANGE, freqs);
 350                 if (likely(policy) && likely(policy->cpu == freqs->cpu))
 351                         policy->cur = freqs->new;
 352                 break;
 353         }
 354 }
 355 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 356
 357
 358
 359 /*********************************************************************
 360  *                          SYSFS INTERFACE                          *
 361  *********************************************************************/
 362
 363 static struct cpufreq_governor *__find_governor(const char *str_governor)
 364 {
 365         struct cpufreq_governor *t;
 366
 367         list_for_each_entry(t, &cpufreq_governor_list, governor_list)
 368                 if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN))
 369                         return t;
 370
 371         return NULL;
 372 }
 373
 374 /**
 375  * cpufreq_parse_governor - parse a governor string
 376  */
 377 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 378                                 struct cpufreq_governor **governor)
 379 {
 380         int err = -EINVAL;
 381
 382         if (!cpufreq_driver)
 383                 goto out;
 384
 385         if (cpufreq_driver->setpolicy) {
 386                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 387                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 388                         err = 0;
 389                 } else if (!strnicmp(str_governor, "powersave",
 390                                                 CPUFREQ_NAME_LEN)) {
 391                         *policy = CPUFREQ_POLICY_POWERSAVE;
 392                         err = 0;
 393                 }
 394         } else if (cpufreq_driver->target) {
 395                 struct cpufreq_governor *t;
 396
 397                 mutex_lock(&cpufreq_governor_mutex);
 398
 399                 t = __find_governor(str_governor);
 400
 401                 if (t == NULL) {
 402                         char *name = kasprintf(GFP_KERNEL, "cpufreq_%s",
 403                                                                 str_governor);
 404
 405                         if (name) {
 406                                 int ret;
 407
 408                                 mutex_unlock(&cpufreq_governor_mutex);
 409                                 ret = request_module(name);
 410                                 mutex_lock(&cpufreq_governor_mutex);
 411
 412                                 if (ret == 0)
 413                                         t = __find_governor(str_governor);
 414                         }
 415
 416                         kfree(name);
 417                 }
 418
 419                 if (t != NULL) {
 420                         *governor = t;
 421                         err = 0;
 422                 }
 423
 424                 mutex_unlock(&cpufreq_governor_mutex);
 425         }
 426   out:
 427         return err;
 428 }
 429
 430
 431 /* drivers/base/cpu.c */
 432 extern struct sysdev_class cpu_sysdev_class;
 433
 434
 435 /**
 436  * cpufreq_per_cpu_attr_read() / show_##file_name() -
 437  * print out cpufreq information
 438  *
 439  * Write out information from cpufreq_driver->policy[cpu]; object must be
 440  * "unsigned int".
 441  */
 442
 443 #define show_one(file_name, object)                     \
 444 static ssize_t show_##file_name                         \
 445 (struct cpufreq_policy * policy, char *buf)             \
 446 {                                                       \
 447         return sprintf (buf, "%u\n", policy->object);   \
 448 }
 449
 450 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 451 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 452 show_one(scaling_min_freq, min);
 453 show_one(scaling_max_freq, max);
 454 show_one(scaling_cur_freq, cur);
 455
 456 static int __cpufreq_set_policy(struct cpufreq_policy *data,
 457                                 struct cpufreq_policy *policy);
 458
 459 /**
 460  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 461  */
 462 #define store_one(file_name, object)                    \
 463 static ssize_t store_##file_name                                        \
 464 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 465 {                                                                       \
 466         unsigned int ret = -EINVAL;                                     \
 467         struct cpufreq_policy new_policy;                               \
 468                                                                         \
 469         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 470         if (ret)                                                        \
 471                 return -EINVAL;                                         \
 472                                                                         \
 473         ret = sscanf (buf, "%u", &new_policy.object);                   \
 474         if (ret != 1)                                                   \
 475                 return -EINVAL;                                         \
 476                                                                         \
 477         ret = __cpufreq_set_policy(policy, &new_policy);                \
 478         policy->user_policy.object = policy->object;                    \
 479                                                                         \
 480         return ret ? ret : count;                                       \
 481 }
 482
 483 store_one(scaling_min_freq,min);
 484 store_one(scaling_max_freq,max);
 485
 486 /**
 487  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 488  */
 489 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy,
 490                                                         char *buf)
 491 {
 492         unsigned int cur_freq = __cpufreq_get(policy->cpu);
 493         if (!cur_freq)
 494                 return sprintf(buf, "<unknown>");
 495         return sprintf(buf, "%u\n", cur_freq);
 496 }
 497
 498
 499 /**
 500  * show_scaling_governor - show the current policy for the specified CPU
 501  */
 502 static ssize_t show_scaling_governor (struct cpufreq_policy * policy,
 503                                                         char *buf)
 504 {
 505         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 506                 return sprintf(buf, "powersave\n");
 507         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 508                 return sprintf(buf, "performance\n");
 509         else if (policy->governor)
 510                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 511         return -EINVAL;
 512 }
 513
 514
 515 /**
 516  * store_scaling_governor - store policy for the specified CPU
 517  */
 518 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 519                                        const char *buf, size_t count)
 520 {
 521         unsigned int ret = -EINVAL;
 522         char    str_governor[16];
 523         struct cpufreq_policy new_policy;
 524
 525         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 526         if (ret)
 527                 return ret;
 528
 529         ret = sscanf (buf, "%15s", str_governor);
 530         if (ret != 1)
 531                 return -EINVAL;
 532
 533         if (cpufreq_parse_governor(str_governor, &new_policy.policy,
 534                                                 &new_policy.governor))
 535                 return -EINVAL;
 536
 537         /* Do not use cpufreq_set_policy here or the user_policy.max
 538            will be wrongly overridden */
 539         ret = __cpufreq_set_policy(policy, &new_policy);
 540
 541         policy->user_policy.policy = policy->policy;
 542         policy->user_policy.governor = policy->governor;
 543
 544         if (ret)
 545                 return ret;
 546         else
 547                 return count;
 548 }
 549
 550 /**
 551  * show_scaling_driver - show the cpufreq driver currently loaded
 552  */
 553 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 554 {
 555         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 556 }
 557
 558 /**
 559  * show_scaling_available_governors - show the available CPUfreq governors
 560  */
 561 static ssize_t show_scaling_available_governors (struct cpufreq_policy *policy,
 562                                 char *buf)
 563 {
 564         ssize_t i = 0;
 565         struct cpufreq_governor *t;
 566
 567         if (!cpufreq_driver->target) {
 568                 i += sprintf(buf, "performance powersave");
 569                 goto out;
 570         }
 571
 572         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 573                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 574                         goto out;
 575                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 576         }
 577 out:
 578         i += sprintf(&buf[i], "\n");
 579         return i;
 580 }
 581 /**
 582  * show_affected_cpus - show the CPUs affected by each transition
 583  */
 584 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 585 {
 586         ssize_t i = 0;
 587         unsigned int cpu;
 588
 589         for_each_cpu_mask(cpu, policy->cpus) {
 590                 if (i)
 591                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 592                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 593                 if (i >= (PAGE_SIZE - 5))
 594                     break;
 595         }
 596         i += sprintf(&buf[i], "\n");
 597         return i;
 598 }
 599
 600
 601 #define define_one_ro(_name) \
 602 static struct freq_attr _name = \
 603 __ATTR(_name, 0444, show_##_name, NULL)
 604
 605 #define define_one_ro0400(_name) \
 606 static struct freq_attr _name = \
 607 __ATTR(_name, 0400, show_##_name, NULL)
 608
 609 #define define_one_rw(_name) \
 610 static struct freq_attr _name = \
 611 __ATTR(_name, 0644, show_##_name, store_##_name)
 612
 613 define_one_ro0400(cpuinfo_cur_freq);
 614 define_one_ro(cpuinfo_min_freq);
 615 define_one_ro(cpuinfo_max_freq);
 616 define_one_ro(scaling_available_governors);
 617 define_one_ro(scaling_driver);
 618 define_one_ro(scaling_cur_freq);
 619 define_one_ro(affected_cpus);
 620 define_one_rw(scaling_min_freq);
 621 define_one_rw(scaling_max_freq);
 622 define_one_rw(scaling_governor);
 623
 624 static struct attribute * default_attrs[] = {
 625         &cpuinfo_min_freq.attr,
 626         &cpuinfo_max_freq.attr,
 627         &scaling_min_freq.attr,
 628         &scaling_max_freq.attr,
 629         &affected_cpus.attr,
 630         &scaling_governor.attr,
 631         &scaling_driver.attr,
 632         &scaling_available_governors.attr,
 633         NULL
 634 };
 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;
 644         policy = cpufreq_cpu_get(policy->cpu);
 645         if (!policy)
 646                 return -EINVAL;
 647
 648         if (lock_policy_rwsem_read(policy->cpu) < 0)
 649                 return -EINVAL;
 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
 658         cpufreq_cpu_put(policy);
 659         return ret;
 660 }
 661
 662 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 663                      const char * buf, size_t count)
 664 {
 665         struct cpufreq_policy * policy = to_policy(kobj);
 666         struct freq_attr * fattr = to_attr(attr);
 667         ssize_t ret;
 668         policy = cpufreq_cpu_get(policy->cpu);
 669         if (!policy)
 670                 return -EINVAL;
 671
 672         if (lock_policy_rwsem_write(policy->cpu) < 0)
 673                 return -EINVAL;
 674
 675         if (fattr->store)
 676                 ret = fattr->store(policy, buf, count);
 677         else
 678                 ret = -EIO;
 679
 680         unlock_policy_rwsem_write(policy->cpu);
 681
 682         cpufreq_cpu_put(policy);
 683         return ret;
 684 }
 685
 686 static void cpufreq_sysfs_release(struct kobject * kobj)
 687 {
 688         struct cpufreq_policy * policy = to_policy(kobj);
 689         dprintk("last reference is dropped\n");
 690         complete(&policy->kobj_unregister);
 691 }
 692
 693 static struct sysfs_ops sysfs_ops = {
 694         .show   = show,
 695         .store  = store,
 696 };
 697
 698 static struct kobj_type ktype_cpufreq = {
 699         .sysfs_ops      = &sysfs_ops,
 700         .default_attrs  = default_attrs,
 701         .release        = cpufreq_sysfs_release,
 702 };
 703
 704
 705 /**
 706  * cpufreq_add_dev - add a CPU device
 707  *
 708  * Adds the cpufreq interface for a CPU device.
 709  */
 710 static int cpufreq_add_dev (struct sys_device * sys_dev)
 711 {
 712         unsigned int cpu = sys_dev->id;
 713         int ret = 0;
 714         struct cpufreq_policy new_policy;
 715         struct cpufreq_policy *policy;
 716         struct freq_attr **drv_attr;
 717         struct sys_device *cpu_sys_dev;
 718         unsigned long flags;
 719         unsigned int j;
 720 #ifdef CONFIG_SMP
 721         struct cpufreq_policy *managed_policy;
 722 #endif
 723
 724         if (cpu_is_offline(cpu))
 725                 return 0;
 726
 727         cpufreq_debug_disable_ratelimit();
 728         dprintk("adding CPU %u\n", cpu);
 729
 730 #ifdef CONFIG_SMP
 731         /* check whether a different CPU already registered this
 732          * CPU because it is in the same boat. */
 733         policy = cpufreq_cpu_get(cpu);
 734         if (unlikely(policy)) {
 735                 cpufreq_cpu_put(policy);
 736                 cpufreq_debug_enable_ratelimit();
 737                 return 0;
 738         }
 739 #endif
 740
 741         if (!try_module_get(cpufreq_driver->owner)) {
 742                 ret = -EINVAL;
 743                 goto module_out;
 744         }
 745
 746         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 747         if (!policy) {
 748                 ret = -ENOMEM;
 749                 goto nomem_out;
 750         }
 751
 752         policy->cpu = cpu;
 753         policy->cpus = cpumask_of_cpu(cpu);
 754
 755         /* Initially set CPU itself as the policy_cpu */
 756         per_cpu(policy_cpu, cpu) = cpu;
 757         lock_policy_rwsem_write(cpu);
 758
 759         init_completion(&policy->kobj_unregister);
 760         INIT_WORK(&policy->update, handle_update);
 761
 762         /* call driver. From then on the cpufreq must be able
 763          * to accept all calls to ->verify and ->setpolicy for this CPU
 764          */
 765         ret = cpufreq_driver->init(policy);
 766         if (ret) {
 767                 dprintk("initialization failed\n");
 768                 unlock_policy_rwsem_write(cpu);
 769                 goto err_out;
 770         }
 771         policy->user_policy.min = policy->cpuinfo.min_freq;
 772         policy->user_policy.max = policy->cpuinfo.max_freq;
 773         policy->user_policy.governor = policy->governor;
 774
 775 #ifdef CONFIG_SMP
 776         for_each_cpu_mask(j, policy->cpus) {
 777                 if (cpu == j)
 778                         continue;
 779
 780                 /* check for existing affected CPUs.  They may not be aware
 781                  * of it due to CPU Hotplug.
 782                  */
 783                 managed_policy = cpufreq_cpu_get(j);
 784                 if (unlikely(managed_policy)) {
 785
 786                         /* Set proper policy_cpu */
 787                         unlock_policy_rwsem_write(cpu);
 788                         per_cpu(policy_cpu, cpu) = managed_policy->cpu;
 789
 790                         if (lock_policy_rwsem_write(cpu) < 0)
 791                                 goto err_out_driver_exit;
 792
 793                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 794                         managed_policy->cpus = policy->cpus;
 795                         cpufreq_cpu_data[cpu] = managed_policy;
 796                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 797
 798                         dprintk("CPU already managed, adding link\n");
 799                         ret = sysfs_create_link(&sys_dev->kobj,
 800                                                 &managed_policy->kobj,
 801                                                 "cpufreq");
 802                         if (ret) {
 803                                 unlock_policy_rwsem_write(cpu);
 804                                 goto err_out_driver_exit;
 805                         }
 806
 807                         cpufreq_debug_enable_ratelimit();
 808                         ret = 0;
 809                         unlock_policy_rwsem_write(cpu);
 810                         goto err_out_driver_exit; /* call driver->exit() */
 811                 }
 812         }
 813 #endif
 814         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 815
 816         /* prepare interface data */
 817         policy->kobj.parent = &sys_dev->kobj;
 818         policy->kobj.ktype = &ktype_cpufreq;
 819         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 820
 821         ret = kobject_register(&policy->kobj);
 822         if (ret) {
 823                 unlock_policy_rwsem_write(cpu);
 824                 goto err_out_driver_exit;
 825         }
 826         /* set up files for this cpu device */
 827         drv_attr = cpufreq_driver->attr;
 828         while ((drv_attr) && (*drv_attr)) {
 829                 if (sysfs_create_file(&policy->kobj, &((*drv_attr)->attr)))
 830                         goto err_out_driver_exit;
 831                 drv_attr++;
 832         }
 833         if (cpufreq_driver->get){
 834                 if (sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr))
 835                         goto err_out_driver_exit;
 836         }
 837         if (cpufreq_driver->target){
 838                 if (sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr))
 839                         goto err_out_driver_exit;
 840         }
 841
 842         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 843         for_each_cpu_mask(j, policy->cpus) {
 844                 cpufreq_cpu_data[j] = policy;
 845                 per_cpu(policy_cpu, j) = policy->cpu;
 846         }
 847         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 848
 849         /* symlink affected CPUs */
 850         for_each_cpu_mask(j, policy->cpus) {
 851                 if (j == cpu)
 852                         continue;
 853                 if (!cpu_online(j))
 854                         continue;
 855
 856                 dprintk("CPU %u already managed, adding link\n", j);
 857                 cpufreq_cpu_get(cpu);
 858                 cpu_sys_dev = get_cpu_sysdev(j);
 859                 ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
 860                                         "cpufreq");
 861                 if (ret) {
 862                         unlock_policy_rwsem_write(cpu);
 863                         goto err_out_unregister;
 864                 }
 865         }
 866
 867         policy->governor = NULL; /* to assure that the starting sequence is
 868                                   * run in cpufreq_set_policy */
 869
 870         /* set default policy */
 871         ret = __cpufreq_set_policy(policy, &new_policy);
 872         policy->user_policy.policy = policy->policy;
 873
 874         unlock_policy_rwsem_write(cpu);
 875
 876         if (ret) {
 877                 dprintk("setting policy failed\n");
 878                 goto err_out_unregister;
 879         }
 880
 881         module_put(cpufreq_driver->owner);
 882         dprintk("initialization complete\n");
 883         cpufreq_debug_enable_ratelimit();
 884
 885         return 0;
 886
 887
 888 err_out_unregister:
 889         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 890         for_each_cpu_mask(j, policy->cpus)
 891                 cpufreq_cpu_data[j] = NULL;
 892         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 893
 894         kobject_unregister(&policy->kobj);
 895         wait_for_completion(&policy->kobj_unregister);
 896
 897 err_out_driver_exit:
 898         if (cpufreq_driver->exit)
 899                 cpufreq_driver->exit(policy);
 900
 901 err_out:
 902         kfree(policy);
 903
 904 nomem_out:
 905         module_put(cpufreq_driver->owner);
 906 module_out:
 907         cpufreq_debug_enable_ratelimit();
 908         return ret;
 909 }
 910
 911
 912 /**
 913  * __cpufreq_remove_dev - remove a CPU device
 914  *
 915  * Removes the cpufreq interface for a CPU device.
 916  * Caller should already have policy_rwsem in write mode for this CPU.
 917  * This routine frees the rwsem before returning.
 918  */
 919 static int __cpufreq_remove_dev (struct sys_device * sys_dev)
 920 {
 921         unsigned int cpu = sys_dev->id;
 922         unsigned long flags;
 923         struct cpufreq_policy *data;
 924 #ifdef CONFIG_SMP
 925         struct sys_device *cpu_sys_dev;
 926         unsigned int j;
 927 #endif
 928
 929         cpufreq_debug_disable_ratelimit();
 930         dprintk("unregistering CPU %u\n", cpu);
 931
 932         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 933         data = cpufreq_cpu_data[cpu];
 934
 935         if (!data) {
 936                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 937                 cpufreq_debug_enable_ratelimit();
 938                 unlock_policy_rwsem_write(cpu);
 939                 return -EINVAL;
 940         }
 941         cpufreq_cpu_data[cpu] = NULL;
 942
 943
 944 #ifdef CONFIG_SMP
 945         /* if this isn't the CPU which is the parent of the kobj, we
 946          * only need to unlink, put and exit
 947          */
 948         if (unlikely(cpu != data->cpu)) {
 949                 dprintk("removing link\n");
 950                 cpu_clear(cpu, data->cpus);
 951                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 952                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 953                 cpufreq_cpu_put(data);
 954                 cpufreq_debug_enable_ratelimit();
 955                 unlock_policy_rwsem_write(cpu);
 956                 return 0;
 957         }
 958 #endif
 959
 960
 961         if (!kobject_get(&data->kobj)) {
 962                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 963                 cpufreq_debug_enable_ratelimit();
 964                 unlock_policy_rwsem_write(cpu);
 965                 return -EFAULT;
 966         }
 967
 968 #ifdef CONFIG_SMP
 969         /* if we have other CPUs still registered, we need to unlink them,
 970          * or else wait_for_completion below will lock up. Clean the
 971          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 972          * links afterwards.
 973          */
 974         if (unlikely(cpus_weight(data->cpus) > 1)) {
 975                 for_each_cpu_mask(j, data->cpus) {
 976                         if (j == cpu)
 977                                 continue;
 978                         cpufreq_cpu_data[j] = NULL;
 979                 }
 980         }
 981
 982         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 983
 984         if (unlikely(cpus_weight(data->cpus) > 1)) {
 985                 for_each_cpu_mask(j, data->cpus) {
 986                         if (j == cpu)
 987                                 continue;
 988                         dprintk("removing link for cpu %u\n", j);
 989                         cpu_sys_dev = get_cpu_sysdev(j);
 990                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 991                         cpufreq_cpu_put(data);
 992                 }
 993         }
 994 #else
 995         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 996 #endif
 997
 998         if (cpufreq_driver->target)
 999                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1000
1001         unlock_policy_rwsem_write(cpu);
1002
1003         kobject_unregister(&data->kobj);
1004
1005         kobject_put(&data->kobj);
1006
1007         /* we need to make sure that the underlying kobj is actually
1008          * not referenced anymore by anybody before we proceed with
1009          * unloading.
1010          */
1011         dprintk("waiting for dropping of refcount\n");
1012         wait_for_completion(&data->kobj_unregister);
1013         dprintk("wait complete\n");
1014
1015         if (cpufreq_driver->exit)
1016                 cpufreq_driver->exit(data);
1017
1018         kfree(data);
1019
1020         cpufreq_debug_enable_ratelimit();
1021         return 0;
1022 }
1023
1024
1025 static int cpufreq_remove_dev (struct sys_device * sys_dev)
1026 {
1027         unsigned int cpu = sys_dev->id;
1028         int retval;
1029
1030         if (cpu_is_offline(cpu))
1031                 return 0;
1032
1033         if (unlikely(lock_policy_rwsem_write(cpu)))
1034                 BUG();
1035
1036         retval = __cpufreq_remove_dev(sys_dev);
1037         return retval;
1038 }
1039
1040
1041 static void handle_update(struct work_struct *work)
1042 {
1043         struct cpufreq_policy *policy =
1044                 container_of(work, struct cpufreq_policy, update);
1045         unsigned int cpu = policy->cpu;
1046         dprintk("handle_update for cpu %u called\n", cpu);
1047         cpufreq_update_policy(cpu);
1048 }
1049
1050 /**
1051  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
1052  *      @cpu: cpu number
1053  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
1054  *      @new_freq: CPU frequency the CPU actually runs at
1055  *
1056  *      We adjust to current frequency first, and need to clean up later. So either call
1057  *      to cpufreq_update_policy() or schedule handle_update()).
1058  */
1059 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq,
1060                                 unsigned int new_freq)
1061 {
1062         struct cpufreq_freqs freqs;
1063
1064         dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
1065                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
1066
1067         freqs.cpu = cpu;
1068         freqs.old = old_freq;
1069         freqs.new = new_freq;
1070         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
1071         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
1072 }
1073
1074
1075 /**
1076  * cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
1077  * @cpu: CPU number
1078  *
1079  * This is the last known freq, without actually getting it from the driver.
1080  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
1081  */
1082 unsigned int cpufreq_quick_get(unsigned int cpu)
1083 {
1084         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1085         unsigned int ret_freq = 0;
1086
1087         if (policy) {
1088                 if (unlikely(lock_policy_rwsem_read(cpu)))
1089                         return ret_freq;
1090
1091                 ret_freq = policy->cur;
1092
1093                 unlock_policy_rwsem_read(cpu);
1094                 cpufreq_cpu_put(policy);
1095         }
1096
1097         return (ret_freq);
1098 }
1099 EXPORT_SYMBOL(cpufreq_quick_get);
1100
1101
1102 static unsigned int __cpufreq_get(unsigned int cpu)
1103 {
1104         struct cpufreq_policy *policy = cpufreq_cpu_data[cpu];
1105         unsigned int ret_freq = 0;
1106
1107         if (!cpufreq_driver->get)
1108                 return (ret_freq);
1109
1110         ret_freq = cpufreq_driver->get(cpu);
1111
1112         if (ret_freq && policy->cur &&
1113                 !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1114                 /* verify no discrepancy between actual and
1115                                         saved value exists */
1116                 if (unlikely(ret_freq != policy->cur)) {
1117                         cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
1118                         schedule_work(&policy->update);
1119                 }
1120         }
1121
1122         return (ret_freq);
1123 }
1124
1125 /**
1126  * cpufreq_get - get the current CPU frequency (in kHz)
1127  * @cpu: CPU number
1128  *
1129  * Get the CPU current (static) CPU frequency
1130  */
1131 unsigned int cpufreq_get(unsigned int cpu)
1132 {
1133         unsigned int ret_freq = 0;
1134         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1135
1136         if (!policy)
1137                 goto out;
1138
1139         if (unlikely(lock_policy_rwsem_read(cpu)))
1140                 goto out_policy;
1141
1142         ret_freq = __cpufreq_get(cpu);
1143
1144         unlock_policy_rwsem_read(cpu);
1145
1146 out_policy:
1147         cpufreq_cpu_put(policy);
1148 out:
1149         return (ret_freq);
1150 }
1151 EXPORT_SYMBOL(cpufreq_get);
1152
1153
1154 /**
1155  *      cpufreq_suspend - let the low level driver prepare for suspend
1156  */
1157
1158 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
1159 {
1160         int cpu = sysdev->id;
1161         int ret = 0;
1162         unsigned int cur_freq = 0;
1163         struct cpufreq_policy *cpu_policy;
1164
1165         dprintk("suspending cpu %u\n", cpu);
1166
1167         if (!cpu_online(cpu))
1168                 return 0;
1169
1170         /* we may be lax here as interrupts are off. Nonetheless
1171          * we need to grab the correct cpu policy, as to check
1172          * whether we really run on this CPU.
1173          */
1174
1175         cpu_policy = cpufreq_cpu_get(cpu);
1176         if (!cpu_policy)
1177                 return -EINVAL;
1178
1179         /* only handle each CPU group once */
1180         if (unlikely(cpu_policy->cpu != cpu)) {
1181                 cpufreq_cpu_put(cpu_policy);
1182                 return 0;
1183         }
1184
1185         if (cpufreq_driver->suspend) {
1186                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
1187                 if (ret) {
1188                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
1189                                         "step on CPU %u\n", cpu_policy->cpu);
1190                         cpufreq_cpu_put(cpu_policy);
1191                         return ret;
1192                 }
1193         }
1194
1195
1196         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
1197                 goto out;
1198
1199         if (cpufreq_driver->get)
1200                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1201
1202         if (!cur_freq || !cpu_policy->cur) {
1203                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
1204                        "frequency is what timing core thinks it is.\n");
1205                 goto out;
1206         }
1207
1208         if (unlikely(cur_freq != cpu_policy->cur)) {
1209                 struct cpufreq_freqs freqs;
1210
1211                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1212                         dprintk("Warning: CPU frequency is %u, "
1213                                "cpufreq assumed %u kHz.\n",
1214                                cur_freq, cpu_policy->cur);
1215
1216                 freqs.cpu = cpu;
1217                 freqs.old = cpu_policy->cur;
1218                 freqs.new = cur_freq;
1219
1220                 srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
1221                                     CPUFREQ_SUSPENDCHANGE, &freqs);
1222                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
1223
1224                 cpu_policy->cur = cur_freq;
1225         }
1226
1227 out:
1228         cpufreq_cpu_put(cpu_policy);
1229         return 0;
1230 }
1231
1232 /**
1233  *      cpufreq_resume -  restore proper CPU frequency handling after resume
1234  *
1235  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
1236  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
1237  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
1238  *          restored.
1239  */
1240 static int cpufreq_resume(struct sys_device * sysdev)
1241 {
1242         int cpu = sysdev->id;
1243         int ret = 0;
1244         struct cpufreq_policy *cpu_policy;
1245
1246         dprintk("resuming cpu %u\n", cpu);
1247
1248         if (!cpu_online(cpu))
1249                 return 0;
1250
1251         /* we may be lax here as interrupts are off. Nonetheless
1252          * we need to grab the correct cpu policy, as to check
1253          * whether we really run on this CPU.
1254          */
1255
1256         cpu_policy = cpufreq_cpu_get(cpu);
1257         if (!cpu_policy)
1258                 return -EINVAL;
1259
1260         /* only handle each CPU group once */
1261         if (unlikely(cpu_policy->cpu != cpu)) {
1262                 cpufreq_cpu_put(cpu_policy);
1263                 return 0;
1264         }
1265
1266         if (cpufreq_driver->resume) {
1267                 ret = cpufreq_driver->resume(cpu_policy);
1268                 if (ret) {
1269                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1270                                         "step on CPU %u\n", cpu_policy->cpu);
1271                         cpufreq_cpu_put(cpu_policy);
1272                         return ret;
1273                 }
1274         }
1275
1276         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1277                 unsigned int cur_freq = 0;
1278
1279                 if (cpufreq_driver->get)
1280                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1281
1282                 if (!cur_freq || !cpu_policy->cur) {
1283                         printk(KERN_ERR "cpufreq: resume failed to assert "
1284                                         "current frequency is what timing core "
1285                                         "thinks it is.\n");
1286                         goto out;
1287                 }
1288
1289                 if (unlikely(cur_freq != cpu_policy->cur)) {
1290                         struct cpufreq_freqs freqs;
1291
1292                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1293                                 dprintk("Warning: CPU frequency"
1294                                        "is %u, cpufreq assumed %u kHz.\n",
1295                                        cur_freq, cpu_policy->cur);
1296
1297                         freqs.cpu = cpu;
1298                         freqs.old = cpu_policy->cur;
1299                         freqs.new = cur_freq;
1300
1301                         srcu_notifier_call_chain(
1302                                         &cpufreq_transition_notifier_list,
1303                                         CPUFREQ_RESUMECHANGE, &freqs);
1304                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1305
1306                         cpu_policy->cur = cur_freq;
1307                 }
1308         }
1309
1310 out:
1311         schedule_work(&cpu_policy->update);
1312         cpufreq_cpu_put(cpu_policy);
1313         return ret;
1314 }
1315
1316 static struct sysdev_driver cpufreq_sysdev_driver = {
1317         .add            = cpufreq_add_dev,
1318         .remove         = cpufreq_remove_dev,
1319         .suspend        = cpufreq_suspend,
1320         .resume         = cpufreq_resume,
1321 };
1322
1323
1324 /*********************************************************************
1325  *                     NOTIFIER LISTS INTERFACE                      *
1326  *********************************************************************/
1327
1328 /**
1329  *      cpufreq_register_notifier - register a driver with cpufreq
1330  *      @nb: notifier function to register
1331  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1332  *
1333  *      Add a driver to one of two lists: either a list of drivers that
1334  *      are notified about clock rate changes (once before and once after
1335  *      the transition), or a list of drivers that are notified about
1336  *      changes in cpufreq policy.
1337  *
1338  *      This function may sleep, and has the same return conditions as
1339  *      blocking_notifier_chain_register.
1340  */
1341 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1342 {
1343         int ret;
1344
1345         switch (list) {
1346         case CPUFREQ_TRANSITION_NOTIFIER:
1347                 ret = srcu_notifier_chain_register(
1348                                 &cpufreq_transition_notifier_list, nb);
1349                 break;
1350         case CPUFREQ_POLICY_NOTIFIER:
1351                 ret = blocking_notifier_chain_register(
1352                                 &cpufreq_policy_notifier_list, nb);
1353                 break;
1354         default:
1355                 ret = -EINVAL;
1356         }
1357
1358         return ret;
1359 }
1360 EXPORT_SYMBOL(cpufreq_register_notifier);
1361
1362
1363 /**
1364  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1365  *      @nb: notifier block to be unregistered
1366  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1367  *
1368  *      Remove a driver from the CPU frequency notifier list.
1369  *
1370  *      This function may sleep, and has the same return conditions as
1371  *      blocking_notifier_chain_unregister.
1372  */
1373 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1374 {
1375         int ret;
1376
1377         switch (list) {
1378         case CPUFREQ_TRANSITION_NOTIFIER:
1379                 ret = srcu_notifier_chain_unregister(
1380                                 &cpufreq_transition_notifier_list, nb);
1381                 break;
1382         case CPUFREQ_POLICY_NOTIFIER:
1383                 ret = blocking_notifier_chain_unregister(
1384                                 &cpufreq_policy_notifier_list, nb);
1385                 break;
1386         default:
1387                 ret = -EINVAL;
1388         }
1389
1390         return ret;
1391 }
1392 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1393
1394
1395 /*********************************************************************
1396  *                              GOVERNORS                            *
1397  *********************************************************************/
1398
1399
1400 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1401                             unsigned int target_freq,
1402                             unsigned int relation)
1403 {
1404         int retval = -EINVAL;
1405
1406         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1407                 target_freq, relation);
1408         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1409                 retval = cpufreq_driver->target(policy, target_freq, relation);
1410
1411         return retval;
1412 }
1413 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1414
1415 int cpufreq_driver_target(struct cpufreq_policy *policy,
1416                           unsigned int target_freq,
1417                           unsigned int relation)
1418 {
1419         int ret;
1420
1421         policy = cpufreq_cpu_get(policy->cpu);
1422         if (!policy)
1423                 return -EINVAL;
1424
1425         if (unlikely(lock_policy_rwsem_write(policy->cpu)))
1426                 return -EINVAL;
1427
1428         ret = __cpufreq_driver_target(policy, target_freq, relation);
1429
1430         unlock_policy_rwsem_write(policy->cpu);
1431
1432         cpufreq_cpu_put(policy);
1433         return ret;
1434 }
1435 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1436
1437 int __cpufreq_driver_getavg(struct cpufreq_policy *policy)
1438 {
1439         int ret = 0;
1440
1441         policy = cpufreq_cpu_get(policy->cpu);
1442         if (!policy)
1443                 return -EINVAL;
1444
1445         if (cpu_online(policy->cpu) && cpufreq_driver->getavg)
1446                 ret = cpufreq_driver->getavg(policy->cpu);
1447
1448         cpufreq_cpu_put(policy);
1449         return ret;
1450 }
1451 EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
1452
1453 /*
1454  * when "event" is CPUFREQ_GOV_LIMITS
1455  */
1456
1457 static int __cpufreq_governor(struct cpufreq_policy *policy,
1458                                         unsigned int event)
1459 {
1460         int ret;
1461
1462         if (!try_module_get(policy->governor->owner))
1463                 return -EINVAL;
1464
1465         dprintk("__cpufreq_governor for CPU %u, event %u\n",
1466                                                 policy->cpu, event);
1467         ret = policy->governor->governor(policy, event);
1468
1469         /* we keep one module reference alive for
1470                         each CPU governed by this CPU */
1471         if ((event != CPUFREQ_GOV_START) || ret)
1472                 module_put(policy->governor->owner);
1473         if ((event == CPUFREQ_GOV_STOP) && !ret)
1474                 module_put(policy->governor->owner);
1475
1476         return ret;
1477 }
1478
1479
1480 int cpufreq_register_governor(struct cpufreq_governor *governor)
1481 {
1482         int err;
1483
1484         if (!governor)
1485                 return -EINVAL;
1486
1487         mutex_lock(&cpufreq_governor_mutex);
1488
1489         err = -EBUSY;
1490         if (__find_governor(governor->name) == NULL) {
1491                 err = 0;
1492                 list_add(&governor->governor_list, &cpufreq_governor_list);
1493         }
1494
1495         mutex_unlock(&cpufreq_governor_mutex);
1496         return err;
1497 }
1498 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1499
1500
1501 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1502 {
1503         if (!governor)
1504                 return;
1505
1506         mutex_lock(&cpufreq_governor_mutex);
1507         list_del(&governor->governor_list);
1508         mutex_unlock(&cpufreq_governor_mutex);
1509         return;
1510 }
1511 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1512
1513
1514
1515 /*********************************************************************
1516  *                          POLICY INTERFACE                         *
1517  *********************************************************************/
1518
1519 /**
1520  * cpufreq_get_policy - get the current cpufreq_policy
1521  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1522  *
1523  * Reads the current cpufreq policy.
1524  */
1525 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1526 {
1527         struct cpufreq_policy *cpu_policy;
1528         if (!policy)
1529                 return -EINVAL;
1530
1531         cpu_policy = cpufreq_cpu_get(cpu);
1532         if (!cpu_policy)
1533                 return -EINVAL;
1534
1535         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1536
1537         cpufreq_cpu_put(cpu_policy);
1538         return 0;
1539 }
1540 EXPORT_SYMBOL(cpufreq_get_policy);
1541
1542
1543 /*
1544  * data   : current policy.
1545  * policy : policy to be set.
1546  */
1547 static int __cpufreq_set_policy(struct cpufreq_policy *data,
1548                                 struct cpufreq_policy *policy)
1549 {
1550         int ret = 0;
1551
1552         cpufreq_debug_disable_ratelimit();
1553         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1554                 policy->min, policy->max);
1555
1556         memcpy(&policy->cpuinfo, &data->cpuinfo,
1557                                 sizeof(struct cpufreq_cpuinfo));
1558
1559         if (policy->min > data->min && policy->min > policy->max) {
1560                 ret = -EINVAL;
1561                 goto error_out;
1562         }
1563
1564         /* verify the cpu speed can be set within this limit */
1565         ret = cpufreq_driver->verify(policy);
1566         if (ret)
1567                 goto error_out;
1568
1569         /* adjust if necessary - all reasons */
1570         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1571                         CPUFREQ_ADJUST, policy);
1572
1573         /* adjust if necessary - hardware incompatibility*/
1574         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1575                         CPUFREQ_INCOMPATIBLE, policy);
1576
1577         /* verify the cpu speed can be set within this limit,
1578            which might be different to the first one */
1579         ret = cpufreq_driver->verify(policy);
1580         if (ret)
1581                 goto error_out;
1582
1583         /* notification of the new policy */
1584         blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
1585                         CPUFREQ_NOTIFY, policy);
1586
1587         data->min = policy->min;
1588         data->max = policy->max;
1589
1590         dprintk("new min and max freqs are %u - %u kHz\n",
1591                                         data->min, data->max);
1592
1593         if (cpufreq_driver->setpolicy) {
1594                 data->policy = policy->policy;
1595                 dprintk("setting range\n");
1596                 ret = cpufreq_driver->setpolicy(policy);
1597         } else {
1598                 if (policy->governor != data->governor) {
1599                         /* save old, working values */
1600                         struct cpufreq_governor *old_gov = data->governor;
1601
1602                         dprintk("governor switch\n");
1603
1604                         /* end old governor */
1605                         if (data->governor)
1606                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1607
1608                         /* start new governor */
1609                         data->governor = policy->governor;
1610                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1611                                 /* new governor failed, so re-start old one */
1612                                 dprintk("starting governor %s failed\n",
1613                                                         data->governor->name);
1614                                 if (old_gov) {
1615                                         data->governor = old_gov;
1616                                         __cpufreq_governor(data,
1617                                                            CPUFREQ_GOV_START);
1618                                 }
1619                                 ret = -EINVAL;
1620                                 goto error_out;
1621                         }
1622                         /* might be a policy change, too, so fall through */
1623                 }
1624                 dprintk("governor: change or update limits\n");
1625                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1626         }
1627
1628 error_out:
1629         cpufreq_debug_enable_ratelimit();
1630         return ret;
1631 }
1632
1633 /**
1634  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1635  *      @cpu: CPU which shall be re-evaluated
1636  *
1637  *      Usefull for policy notifiers which have different necessities
1638  *      at different times.
1639  */
1640 int cpufreq_update_policy(unsigned int cpu)
1641 {
1642         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1643         struct cpufreq_policy policy;
1644         int ret = 0;
1645
1646         if (!data)
1647                 return -ENODEV;
1648
1649         if (unlikely(lock_policy_rwsem_write(cpu)))
1650                 return -EINVAL;
1651
1652         dprintk("updating policy for CPU %u\n", cpu);
1653         memcpy(&policy, data, sizeof(struct cpufreq_policy));
1654         policy.min = data->user_policy.min;
1655         policy.max = data->user_policy.max;
1656         policy.policy = data->user_policy.policy;
1657         policy.governor = data->user_policy.governor;
1658
1659         /* BIOS might change freq behind our back
1660           -> ask driver for current freq and notify governors about a change */
1661         if (cpufreq_driver->get) {
1662                 policy.cur = cpufreq_driver->get(cpu);
1663                 if (!data->cur) {
1664                         dprintk("Driver did not initialize current freq");
1665                         data->cur = policy.cur;
1666                 } else {
1667                         if (data->cur != policy.cur)
1668                                 cpufreq_out_of_sync(cpu, data->cur,
1669                                                                 policy.cur);
1670                 }
1671         }
1672
1673         ret = __cpufreq_set_policy(data, &policy);
1674
1675         unlock_policy_rwsem_write(cpu);
1676
1677         cpufreq_cpu_put(data);
1678         return ret;
1679 }
1680 EXPORT_SYMBOL(cpufreq_update_policy);
1681
1682 static int cpufreq_cpu_callback(struct notifier_block *nfb,
1683                                         unsigned long action, void *hcpu)
1684 {
1685         unsigned int cpu = (unsigned long)hcpu;
1686         struct sys_device *sys_dev;
1687         struct cpufreq_policy *policy;
1688
1689         sys_dev = get_cpu_sysdev(cpu);
1690         if (sys_dev) {
1691                 switch (action) {
1692                 case CPU_ONLINE:
1693                 case CPU_ONLINE_FROZEN:
1694                         cpufreq_add_dev(sys_dev);
1695                         break;
1696                 case CPU_DOWN_PREPARE:
1697                 case CPU_DOWN_PREPARE_FROZEN:
1698                         if (unlikely(lock_policy_rwsem_write(cpu)))
1699                                 BUG();
1700
1701                         policy = cpufreq_cpu_data[cpu];
1702                         if (policy) {
1703                                 __cpufreq_driver_target(policy, policy->min,
1704                                                 CPUFREQ_RELATION_H);
1705                         }
1706                         __cpufreq_remove_dev(sys_dev);
1707                         break;
1708                 case CPU_DOWN_FAILED:
1709                 case CPU_DOWN_FAILED_FROZEN:
1710                         cpufreq_add_dev(sys_dev);
1711                         break;
1712                 }
1713         }
1714         return NOTIFY_OK;
1715 }
1716
1717 static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
1718 {
1719     .notifier_call = cpufreq_cpu_callback,
1720 };
1721
1722 /*********************************************************************
1723  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1724  *********************************************************************/
1725
1726 /**
1727  * cpufreq_register_driver - register a CPU Frequency driver
1728  * @driver_data: A struct cpufreq_driver containing the values#
1729  * submitted by the CPU Frequency driver.
1730  *
1731  *   Registers a CPU Frequency driver to this core code. This code
1732  * returns zero on success, -EBUSY when another driver got here first
1733  * (and isn't unregistered in the meantime).
1734  *
1735  */
1736 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1737 {
1738         unsigned long flags;
1739         int ret;
1740
1741         if (!driver_data || !driver_data->verify || !driver_data->init ||
1742             ((!driver_data->setpolicy) && (!driver_data->target)))
1743                 return -EINVAL;
1744
1745         dprintk("trying to register driver %s\n", driver_data->name);
1746
1747         if (driver_data->setpolicy)
1748                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1749
1750         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1751         if (cpufreq_driver) {
1752                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1753                 return -EBUSY;
1754         }
1755         cpufreq_driver = driver_data;
1756         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1757
1758         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1759
1760         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1761                 int i;
1762                 ret = -ENODEV;
1763
1764                 /* check for at least one working CPU */
1765                 for (i=0; i<NR_CPUS; i++)
1766                         if (cpufreq_cpu_data[i])
1767                                 ret = 0;
1768
1769                 /* if all ->init() calls failed, unregister */
1770                 if (ret) {
1771                         dprintk("no CPU initialized for driver %s\n",
1772                                                         driver_data->name);
1773                         sysdev_driver_unregister(&cpu_sysdev_class,
1774                                                 &cpufreq_sysdev_driver);
1775
1776                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1777                         cpufreq_driver = NULL;
1778                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1779                 }
1780         }
1781
1782         if (!ret) {
1783                 register_hotcpu_notifier(&cpufreq_cpu_notifier);
1784                 dprintk("driver %s up and running\n", driver_data->name);
1785                 cpufreq_debug_enable_ratelimit();
1786         }
1787
1788         return (ret);
1789 }
1790 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1791
1792
1793 /**
1794  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1795  *
1796  *    Unregister the current CPUFreq driver. Only call this if you have
1797  * the right to do so, i.e. if you have succeeded in initialising before!
1798  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1799  * currently not initialised.
1800  */
1801 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1802 {
1803         unsigned long flags;
1804
1805         cpufreq_debug_disable_ratelimit();
1806
1807         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1808                 cpufreq_debug_enable_ratelimit();
1809                 return -EINVAL;
1810         }
1811
1812         dprintk("unregistering driver %s\n", driver->name);
1813
1814         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1815         unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
1816
1817         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1818         cpufreq_driver = NULL;
1819         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1820
1821         return 0;
1822 }
1823 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
1824
1825 static int __init cpufreq_core_init(void)
1826 {
1827         int cpu;
1828
1829         for_each_possible_cpu(cpu) {
1830                 per_cpu(policy_cpu, cpu) = -1;
1831                 init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
1832         }
1833         return 0;
1834 }
1835
1836 core_initcall(cpufreq_core_init);