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  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License version 2 as
  12  * published by the Free Software Foundation.
  13  *
  14  */
  15
  16 #include <linux/config.h>
  17 #include <linux/kernel.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/notifier.h>
  21 #include <linux/cpufreq.h>
  22 #include <linux/delay.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/device.h>
  26 #include <linux/slab.h>
  27 #include <linux/cpu.h>
  28 #include <linux/completion.h>
  29 #include <linux/mutex.h>
  30
  31 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
  32
  33 /**
  34  * The "cpufreq driver" - the arch- or hardware-dependend low
  35  * level driver of CPUFreq support, and its spinlock. This lock
  36  * also protects the cpufreq_cpu_data array.
  37  */
  38 static struct cpufreq_driver    *cpufreq_driver;
  39 static struct cpufreq_policy    *cpufreq_cpu_data[NR_CPUS];
  40 static DEFINE_SPINLOCK(cpufreq_driver_lock);
  41
  42 /* internal prototypes */
  43 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
  44 static void handle_update(void *data);
  45
  46 /**
  47  * Two notifier lists: the "policy" list is involved in the
  48  * validation process for a new CPU frequency policy; the
  49  * "transition" list for kernel code that needs to handle
  50  * changes to devices when the CPU clock speed changes.
  51  * The mutex locks both lists.
  52  */
  53 static struct notifier_block    *cpufreq_policy_notifier_list;
  54 static struct notifier_block    *cpufreq_transition_notifier_list;
  55 static DECLARE_RWSEM            (cpufreq_notifier_rwsem);
  56
  57
  58 static LIST_HEAD(cpufreq_governor_list);
  59 static DEFINE_MUTEX             (cpufreq_governor_mutex);
  60
  61 struct cpufreq_policy * cpufreq_cpu_get(unsigned int cpu)
  62 {
  63         struct cpufreq_policy *data;
  64         unsigned long flags;
  65
  66         if (cpu >= NR_CPUS)
  67                 goto err_out;
  68
  69         /* get the cpufreq driver */
  70         spin_lock_irqsave(&cpufreq_driver_lock, flags);
  71
  72         if (!cpufreq_driver)
  73                 goto err_out_unlock;
  74
  75         if (!try_module_get(cpufreq_driver->owner))
  76                 goto err_out_unlock;
  77
  78
  79         /* get the CPU */
  80         data = cpufreq_cpu_data[cpu];
  81
  82         if (!data)
  83                 goto err_out_put_module;
  84
  85         if (!kobject_get(&data->kobj))
  86                 goto err_out_put_module;
  87
  88
  89         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  90
  91         return data;
  92
  93  err_out_put_module:
  94         module_put(cpufreq_driver->owner);
  95  err_out_unlock:
  96         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
  97  err_out:
  98         return NULL;
  99 }
 100 EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
 101
 102 void cpufreq_cpu_put(struct cpufreq_policy *data)
 103 {
 104         kobject_put(&data->kobj);
 105         module_put(cpufreq_driver->owner);
 106 }
 107 EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
 108
 109
 110 /*********************************************************************
 111  *                     UNIFIED DEBUG HELPERS                         *
 112  *********************************************************************/
 113 #ifdef CONFIG_CPU_FREQ_DEBUG
 114
 115 /* what part(s) of the CPUfreq subsystem are debugged? */
 116 static unsigned int debug;
 117
 118 /* is the debug output ratelimit'ed using printk_ratelimit? User can
 119  * set or modify this value.
 120  */
 121 static unsigned int debug_ratelimit = 1;
 122
 123 /* is the printk_ratelimit'ing enabled? It's enabled after a successful
 124  * loading of a cpufreq driver, temporarily disabled when a new policy
 125  * is set, and disabled upon cpufreq driver removal
 126  */
 127 static unsigned int disable_ratelimit = 1;
 128 static DEFINE_SPINLOCK(disable_ratelimit_lock);
 129
 130 static void cpufreq_debug_enable_ratelimit(void)
 131 {
 132         unsigned long flags;
 133
 134         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 135         if (disable_ratelimit)
 136                 disable_ratelimit--;
 137         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 138 }
 139
 140 static void cpufreq_debug_disable_ratelimit(void)
 141 {
 142         unsigned long flags;
 143
 144         spin_lock_irqsave(&disable_ratelimit_lock, flags);
 145         disable_ratelimit++;
 146         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 147 }
 148
 149 void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
 150 {
 151         char s[256];
 152         va_list args;
 153         unsigned int len;
 154         unsigned long flags;
 155
 156         WARN_ON(!prefix);
 157         if (type & debug) {
 158                 spin_lock_irqsave(&disable_ratelimit_lock, flags);
 159                 if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
 160                         spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 161                         return;
 162                 }
 163                 spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
 164
 165                 len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
 166
 167                 va_start(args, fmt);
 168                 len += vsnprintf(&s[len], (256 - len), fmt, args);
 169                 va_end(args);
 170
 171                 printk(s);
 172
 173                 WARN_ON(len < 5);
 174         }
 175 }
 176 EXPORT_SYMBOL(cpufreq_debug_printk);
 177
 178
 179 module_param(debug, uint, 0644);
 180 MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
 181
 182 module_param(debug_ratelimit, uint, 0644);
 183 MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
 184
 185 #else /* !CONFIG_CPU_FREQ_DEBUG */
 186
 187 static inline void cpufreq_debug_enable_ratelimit(void) { return; }
 188 static inline void cpufreq_debug_disable_ratelimit(void) { return; }
 189
 190 #endif /* CONFIG_CPU_FREQ_DEBUG */
 191
 192
 193 /*********************************************************************
 194  *            EXTERNALLY AFFECTING FREQUENCY CHANGES                 *
 195  *********************************************************************/
 196
 197 /**
 198  * adjust_jiffies - adjust the system "loops_per_jiffy"
 199  *
 200  * This function alters the system "loops_per_jiffy" for the clock
 201  * speed change. Note that loops_per_jiffy cannot be updated on SMP
 202  * systems as each CPU might be scaled differently. So, use the arch
 203  * per-CPU loops_per_jiffy value wherever possible.
 204  */
 205 #ifndef CONFIG_SMP
 206 static unsigned long l_p_j_ref;
 207 static unsigned int  l_p_j_ref_freq;
 208
 209 static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
 210 {
 211         if (ci->flags & CPUFREQ_CONST_LOOPS)
 212                 return;
 213
 214         if (!l_p_j_ref_freq) {
 215                 l_p_j_ref = loops_per_jiffy;
 216                 l_p_j_ref_freq = ci->old;
 217                 dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
 218         }
 219         if ((val == CPUFREQ_PRECHANGE  && ci->old < ci->new) ||
 220             (val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
 221             (val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
 222                 loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
 223                 dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
 224         }
 225 }
 226 #else
 227 static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
 228 #endif
 229
 230
 231 /**
 232  * cpufreq_notify_transition - call notifier chain and adjust_jiffies on frequency transition
 233  *
 234  * This function calls the transition notifiers and the "adjust_jiffies" function. It is called
 235  * twice on all CPU frequency changes that have external effects.
 236  */
 237 void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
 238 {
 239         BUG_ON(irqs_disabled());
 240
 241         freqs->flags = cpufreq_driver->flags;
 242         dprintk("notification %u of frequency transition to %u kHz\n", state, freqs->new);
 243
 244         down_read(&cpufreq_notifier_rwsem);
 245         switch (state) {
 246         case CPUFREQ_PRECHANGE:
 247                 /* detect if the driver reported a value as "old frequency" which
 248                  * is not equal to what the cpufreq core thinks is "old frequency".
 249                  */
 250                 if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
 251                         if ((likely(cpufreq_cpu_data[freqs->cpu])) &&
 252                             (likely(cpufreq_cpu_data[freqs->cpu]->cpu == freqs->cpu)) &&
 253                             (likely(cpufreq_cpu_data[freqs->cpu]->cur)) &&
 254                             (unlikely(freqs->old != cpufreq_cpu_data[freqs->cpu]->cur)))
 255                         {
 256                                 dprintk(KERN_WARNING "Warning: CPU frequency is %u, "
 257                                        "cpufreq assumed %u kHz.\n", freqs->old, cpufreq_cpu_data[freqs->cpu]->cur);
 258                                 freqs->old = cpufreq_cpu_data[freqs->cpu]->cur;
 259                         }
 260                 }
 261                 notifier_call_chain(&cpufreq_transition_notifier_list, CPUFREQ_PRECHANGE, freqs);
 262                 adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
 263                 break;
 264         case CPUFREQ_POSTCHANGE:
 265                 adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
 266                 notifier_call_chain(&cpufreq_transition_notifier_list, CPUFREQ_POSTCHANGE, freqs);
 267                 if ((likely(cpufreq_cpu_data[freqs->cpu])) &&
 268                     (likely(cpufreq_cpu_data[freqs->cpu]->cpu == freqs->cpu)))
 269                         cpufreq_cpu_data[freqs->cpu]->cur = freqs->new;
 270                 break;
 271         }
 272         up_read(&cpufreq_notifier_rwsem);
 273 }
 274 EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 275
 276
 277
 278 /*********************************************************************
 279  *                          SYSFS INTERFACE                          *
 280  *********************************************************************/
 281
 282 /**
 283  * cpufreq_parse_governor - parse a governor string
 284  */
 285 static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
 286                                 struct cpufreq_governor **governor)
 287 {
 288         if (!cpufreq_driver)
 289                 return -EINVAL;
 290         if (cpufreq_driver->setpolicy) {
 291                 if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
 292                         *policy = CPUFREQ_POLICY_PERFORMANCE;
 293                         return 0;
 294                 } else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
 295                         *policy = CPUFREQ_POLICY_POWERSAVE;
 296                         return 0;
 297                 }
 298                 return -EINVAL;
 299         } else {
 300                 struct cpufreq_governor *t;
 301                 mutex_lock(&cpufreq_governor_mutex);
 302                 if (!cpufreq_driver || !cpufreq_driver->target)
 303                         goto out;
 304                 list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 305                         if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN)) {
 306                                 *governor = t;
 307                                 mutex_unlock(&cpufreq_governor_mutex);
 308                                 return 0;
 309                         }
 310                 }
 311         out:
 312                 mutex_unlock(&cpufreq_governor_mutex);
 313         }
 314         return -EINVAL;
 315 }
 316 EXPORT_SYMBOL_GPL(cpufreq_parse_governor);
 317
 318
 319 /* drivers/base/cpu.c */
 320 extern struct sysdev_class cpu_sysdev_class;
 321
 322
 323 /**
 324  * cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
 325  *
 326  * Write out information from cpufreq_driver->policy[cpu]; object must be
 327  * "unsigned int".
 328  */
 329
 330 #define show_one(file_name, object)                                     \
 331 static ssize_t show_##file_name                                         \
 332 (struct cpufreq_policy * policy, char *buf)                             \
 333 {                                                                       \
 334         return sprintf (buf, "%u\n", policy->object);                   \
 335 }
 336
 337 show_one(cpuinfo_min_freq, cpuinfo.min_freq);
 338 show_one(cpuinfo_max_freq, cpuinfo.max_freq);
 339 show_one(scaling_min_freq, min);
 340 show_one(scaling_max_freq, max);
 341 show_one(scaling_cur_freq, cur);
 342
 343 /**
 344  * cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
 345  */
 346 #define store_one(file_name, object)                    \
 347 static ssize_t store_##file_name                                        \
 348 (struct cpufreq_policy * policy, const char *buf, size_t count)         \
 349 {                                                                       \
 350         unsigned int ret = -EINVAL;                                     \
 351         struct cpufreq_policy new_policy;                               \
 352                                                                         \
 353         ret = cpufreq_get_policy(&new_policy, policy->cpu);             \
 354         if (ret)                                                        \
 355                 return -EINVAL;                                         \
 356                                                                         \
 357         ret = sscanf (buf, "%u", &new_policy.object);                   \
 358         if (ret != 1)                                                   \
 359                 return -EINVAL;                                         \
 360                                                                         \
 361         ret = cpufreq_set_policy(&new_policy);                          \
 362                                                                         \
 363         return ret ? ret : count;                                       \
 364 }
 365
 366 store_one(scaling_min_freq,min);
 367 store_one(scaling_max_freq,max);
 368
 369 /**
 370  * show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
 371  */
 372 static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
 373 {
 374         unsigned int cur_freq = cpufreq_get(policy->cpu);
 375         if (!cur_freq)
 376                 return sprintf(buf, "<unknown>");
 377         return sprintf(buf, "%u\n", cur_freq);
 378 }
 379
 380
 381 /**
 382  * show_scaling_governor - show the current policy for the specified CPU
 383  */
 384 static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
 385 {
 386         if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
 387                 return sprintf(buf, "powersave\n");
 388         else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
 389                 return sprintf(buf, "performance\n");
 390         else if (policy->governor)
 391                 return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
 392         return -EINVAL;
 393 }
 394
 395
 396 /**
 397  * store_scaling_governor - store policy for the specified CPU
 398  */
 399 static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
 400                                        const char *buf, size_t count)
 401 {
 402         unsigned int ret = -EINVAL;
 403         char    str_governor[16];
 404         struct cpufreq_policy new_policy;
 405
 406         ret = cpufreq_get_policy(&new_policy, policy->cpu);
 407         if (ret)
 408                 return ret;
 409
 410         ret = sscanf (buf, "%15s", str_governor);
 411         if (ret != 1)
 412                 return -EINVAL;
 413
 414         if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
 415                 return -EINVAL;
 416
 417         ret = cpufreq_set_policy(&new_policy);
 418
 419         return ret ? ret : count;
 420 }
 421
 422 /**
 423  * show_scaling_driver - show the cpufreq driver currently loaded
 424  */
 425 static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
 426 {
 427         return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
 428 }
 429
 430 /**
 431  * show_scaling_available_governors - show the available CPUfreq governors
 432  */
 433 static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
 434                                 char *buf)
 435 {
 436         ssize_t i = 0;
 437         struct cpufreq_governor *t;
 438
 439         if (!cpufreq_driver->target) {
 440                 i += sprintf(buf, "performance powersave");
 441                 goto out;
 442         }
 443
 444         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
 445                 if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
 446                         goto out;
 447                 i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
 448         }
 449  out:
 450         i += sprintf(&buf[i], "\n");
 451         return i;
 452 }
 453 /**
 454  * show_affected_cpus - show the CPUs affected by each transition
 455  */
 456 static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
 457 {
 458         ssize_t i = 0;
 459         unsigned int cpu;
 460
 461         for_each_cpu_mask(cpu, policy->cpus) {
 462                 if (i)
 463                         i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
 464                 i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
 465                 if (i >= (PAGE_SIZE - 5))
 466                     break;
 467         }
 468         i += sprintf(&buf[i], "\n");
 469         return i;
 470 }
 471
 472
 473 #define define_one_ro(_name) \
 474 static struct freq_attr _name = \
 475 __ATTR(_name, 0444, show_##_name, NULL)
 476
 477 #define define_one_ro0400(_name) \
 478 static struct freq_attr _name = \
 479 __ATTR(_name, 0400, show_##_name, NULL)
 480
 481 #define define_one_rw(_name) \
 482 static struct freq_attr _name = \
 483 __ATTR(_name, 0644, show_##_name, store_##_name)
 484
 485 define_one_ro0400(cpuinfo_cur_freq);
 486 define_one_ro(cpuinfo_min_freq);
 487 define_one_ro(cpuinfo_max_freq);
 488 define_one_ro(scaling_available_governors);
 489 define_one_ro(scaling_driver);
 490 define_one_ro(scaling_cur_freq);
 491 define_one_ro(affected_cpus);
 492 define_one_rw(scaling_min_freq);
 493 define_one_rw(scaling_max_freq);
 494 define_one_rw(scaling_governor);
 495
 496 static struct attribute * default_attrs[] = {
 497         &cpuinfo_min_freq.attr,
 498         &cpuinfo_max_freq.attr,
 499         &scaling_min_freq.attr,
 500         &scaling_max_freq.attr,
 501         &affected_cpus.attr,
 502         &scaling_governor.attr,
 503         &scaling_driver.attr,
 504         &scaling_available_governors.attr,
 505         NULL
 506 };
 507
 508 #define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
 509 #define to_attr(a) container_of(a,struct freq_attr,attr)
 510
 511 static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
 512 {
 513         struct cpufreq_policy * policy = to_policy(kobj);
 514         struct freq_attr * fattr = to_attr(attr);
 515         ssize_t ret;
 516         policy = cpufreq_cpu_get(policy->cpu);
 517         if (!policy)
 518                 return -EINVAL;
 519         ret = fattr->show ? fattr->show(policy,buf) : -EIO;
 520         cpufreq_cpu_put(policy);
 521         return ret;
 522 }
 523
 524 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 525                      const char * buf, size_t count)
 526 {
 527         struct cpufreq_policy * policy = to_policy(kobj);
 528         struct freq_attr * fattr = to_attr(attr);
 529         ssize_t ret;
 530         policy = cpufreq_cpu_get(policy->cpu);
 531         if (!policy)
 532                 return -EINVAL;
 533         ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
 534         cpufreq_cpu_put(policy);
 535         return ret;
 536 }
 537
 538 static void cpufreq_sysfs_release(struct kobject * kobj)
 539 {
 540         struct cpufreq_policy * policy = to_policy(kobj);
 541         dprintk("last reference is dropped\n");
 542         complete(&policy->kobj_unregister);
 543 }
 544
 545 static struct sysfs_ops sysfs_ops = {
 546         .show   = show,
 547         .store  = store,
 548 };
 549
 550 static struct kobj_type ktype_cpufreq = {
 551         .sysfs_ops      = &sysfs_ops,
 552         .default_attrs  = default_attrs,
 553         .release        = cpufreq_sysfs_release,
 554 };
 555
 556
 557 /**
 558  * cpufreq_add_dev - add a CPU device
 559  *
 560  * Adds the cpufreq interface for a CPU device.
 561  */
 562 static int cpufreq_add_dev (struct sys_device * sys_dev)
 563 {
 564         unsigned int cpu = sys_dev->id;
 565         int ret = 0;
 566         struct cpufreq_policy new_policy;
 567         struct cpufreq_policy *policy;
 568         struct freq_attr **drv_attr;
 569         unsigned long flags;
 570         unsigned int j;
 571
 572         if (cpu_is_offline(cpu))
 573                 return 0;
 574
 575         cpufreq_debug_disable_ratelimit();
 576         dprintk("adding CPU %u\n", cpu);
 577
 578 #ifdef CONFIG_SMP
 579         /* check whether a different CPU already registered this
 580          * CPU because it is in the same boat. */
 581         policy = cpufreq_cpu_get(cpu);
 582         if (unlikely(policy)) {
 583                 dprintk("CPU already managed, adding link\n");
 584                 sysfs_create_link(&sys_dev->kobj, &policy->kobj, "cpufreq");
 585                 cpufreq_debug_enable_ratelimit();
 586                 return 0;
 587         }
 588 #endif
 589
 590         if (!try_module_get(cpufreq_driver->owner)) {
 591                 ret = -EINVAL;
 592                 goto module_out;
 593         }
 594
 595         policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
 596         if (!policy) {
 597                 ret = -ENOMEM;
 598                 goto nomem_out;
 599         }
 600
 601         policy->cpu = cpu;
 602         policy->cpus = cpumask_of_cpu(cpu);
 603
 604         mutex_init(&policy->lock);
 605         mutex_lock(&policy->lock);
 606         init_completion(&policy->kobj_unregister);
 607         INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
 608
 609         /* call driver. From then on the cpufreq must be able
 610          * to accept all calls to ->verify and ->setpolicy for this CPU
 611          */
 612         ret = cpufreq_driver->init(policy);
 613         if (ret) {
 614                 dprintk("initialization failed\n");
 615                 mutex_unlock(&policy->lock);
 616                 goto err_out;
 617         }
 618
 619         memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
 620
 621         /* prepare interface data */
 622         policy->kobj.parent = &sys_dev->kobj;
 623         policy->kobj.ktype = &ktype_cpufreq;
 624         strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
 625
 626         ret = kobject_register(&policy->kobj);
 627         if (ret) {
 628                 mutex_unlock(&policy->lock);
 629                 goto err_out_driver_exit;
 630         }
 631         /* set up files for this cpu device */
 632         drv_attr = cpufreq_driver->attr;
 633         while ((drv_attr) && (*drv_attr)) {
 634                 sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
 635                 drv_attr++;
 636         }
 637         if (cpufreq_driver->get)
 638                 sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
 639         if (cpufreq_driver->target)
 640                 sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
 641
 642         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 643         for_each_cpu_mask(j, policy->cpus)
 644                 cpufreq_cpu_data[j] = policy;
 645         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 646         policy->governor = NULL; /* to assure that the starting sequence is
 647                                   * run in cpufreq_set_policy */
 648         mutex_unlock(&policy->lock);
 649
 650         /* set default policy */
 651
 652         ret = cpufreq_set_policy(&new_policy);
 653         if (ret) {
 654                 dprintk("setting policy failed\n");
 655                 goto err_out_unregister;
 656         }
 657
 658         module_put(cpufreq_driver->owner);
 659         dprintk("initialization complete\n");
 660         cpufreq_debug_enable_ratelimit();
 661
 662         return 0;
 663
 664
 665 err_out_unregister:
 666         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 667         for_each_cpu_mask(j, policy->cpus)
 668                 cpufreq_cpu_data[j] = NULL;
 669         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 670
 671         kobject_unregister(&policy->kobj);
 672         wait_for_completion(&policy->kobj_unregister);
 673
 674 err_out_driver_exit:
 675         if (cpufreq_driver->exit)
 676                 cpufreq_driver->exit(policy);
 677
 678 err_out:
 679         kfree(policy);
 680
 681 nomem_out:
 682         module_put(cpufreq_driver->owner);
 683 module_out:
 684         cpufreq_debug_enable_ratelimit();
 685         return ret;
 686 }
 687
 688
 689 /**
 690  * cpufreq_remove_dev - remove a CPU device
 691  *
 692  * Removes the cpufreq interface for a CPU device.
 693  */
 694 static int cpufreq_remove_dev (struct sys_device * sys_dev)
 695 {
 696         unsigned int cpu = sys_dev->id;
 697         unsigned long flags;
 698         struct cpufreq_policy *data;
 699 #ifdef CONFIG_SMP
 700         struct sys_device *cpu_sys_dev;
 701         unsigned int j;
 702 #endif
 703
 704         cpufreq_debug_disable_ratelimit();
 705         dprintk("unregistering CPU %u\n", cpu);
 706
 707         spin_lock_irqsave(&cpufreq_driver_lock, flags);
 708         data = cpufreq_cpu_data[cpu];
 709
 710         if (!data) {
 711                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 712                 cpufreq_debug_enable_ratelimit();
 713                 return -EINVAL;
 714         }
 715         cpufreq_cpu_data[cpu] = NULL;
 716
 717
 718 #ifdef CONFIG_SMP
 719         /* if this isn't the CPU which is the parent of the kobj, we
 720          * only need to unlink, put and exit
 721          */
 722         if (unlikely(cpu != data->cpu)) {
 723                 dprintk("removing link\n");
 724                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 725                 sysfs_remove_link(&sys_dev->kobj, "cpufreq");
 726                 cpufreq_cpu_put(data);
 727                 cpufreq_debug_enable_ratelimit();
 728                 return 0;
 729         }
 730 #endif
 731
 732
 733         if (!kobject_get(&data->kobj)) {
 734                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 735                 cpufreq_debug_enable_ratelimit();
 736                 return -EFAULT;
 737         }
 738
 739 #ifdef CONFIG_SMP
 740         /* if we have other CPUs still registered, we need to unlink them,
 741          * or else wait_for_completion below will lock up. Clean the
 742          * cpufreq_cpu_data[] while holding the lock, and remove the sysfs
 743          * links afterwards.
 744          */
 745         if (unlikely(cpus_weight(data->cpus) > 1)) {
 746                 for_each_cpu_mask(j, data->cpus) {
 747                         if (j == cpu)
 748                                 continue;
 749                         cpufreq_cpu_data[j] = NULL;
 750                 }
 751         }
 752
 753         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 754
 755         if (unlikely(cpus_weight(data->cpus) > 1)) {
 756                 for_each_cpu_mask(j, data->cpus) {
 757                         if (j == cpu)
 758                                 continue;
 759                         dprintk("removing link for cpu %u\n", j);
 760                         cpu_sys_dev = get_cpu_sysdev(j);
 761                         sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
 762                         cpufreq_cpu_put(data);
 763                 }
 764         }
 765 #else
 766         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 767 #endif
 768
 769         mutex_lock(&data->lock);
 770         if (cpufreq_driver->target)
 771                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
 772         mutex_unlock(&data->lock);
 773
 774         kobject_unregister(&data->kobj);
 775
 776         kobject_put(&data->kobj);
 777
 778         /* we need to make sure that the underlying kobj is actually
 779          * not referenced anymore by anybody before we proceed with
 780          * unloading.
 781          */
 782         dprintk("waiting for dropping of refcount\n");
 783         wait_for_completion(&data->kobj_unregister);
 784         dprintk("wait complete\n");
 785
 786         if (cpufreq_driver->exit)
 787                 cpufreq_driver->exit(data);
 788
 789         kfree(data);
 790
 791         cpufreq_debug_enable_ratelimit();
 792
 793         return 0;
 794 }
 795
 796
 797 static void handle_update(void *data)
 798 {
 799         unsigned int cpu = (unsigned int)(long)data;
 800         dprintk("handle_update for cpu %u called\n", cpu);
 801         cpufreq_update_policy(cpu);
 802 }
 803
 804 /**
 805  *      cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
 806  *      @cpu: cpu number
 807  *      @old_freq: CPU frequency the kernel thinks the CPU runs at
 808  *      @new_freq: CPU frequency the CPU actually runs at
 809  *
 810  *      We adjust to current frequency first, and need to clean up later. So either call
 811  *      to cpufreq_update_policy() or schedule handle_update()).
 812  */
 813 static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
 814 {
 815         struct cpufreq_freqs freqs;
 816
 817         dprintk(KERN_WARNING "Warning: CPU frequency out of sync: cpufreq and timing "
 818                "core thinks of %u, is %u kHz.\n", old_freq, new_freq);
 819
 820         freqs.cpu = cpu;
 821         freqs.old = old_freq;
 822         freqs.new = new_freq;
 823         cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 824         cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 825 }
 826
 827
 828 /**
 829  * cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
 830  * @cpu: CPU number
 831  *
 832  * This is the last known freq, without actually getting it from the driver.
 833  * Return value will be same as what is shown in scaling_cur_freq in sysfs.
 834  */
 835 unsigned int cpufreq_quick_get(unsigned int cpu)
 836 {
 837         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 838         unsigned int ret = 0;
 839
 840         if (policy) {
 841                 mutex_lock(&policy->lock);
 842                 ret = policy->cur;
 843                 mutex_unlock(&policy->lock);
 844                 cpufreq_cpu_put(policy);
 845         }
 846
 847         return (ret);
 848 }
 849 EXPORT_SYMBOL(cpufreq_quick_get);
 850
 851
 852 /**
 853  * cpufreq_get - get the current CPU frequency (in kHz)
 854  * @cpu: CPU number
 855  *
 856  * Get the CPU current (static) CPU frequency
 857  */
 858 unsigned int cpufreq_get(unsigned int cpu)
 859 {
 860         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
 861         unsigned int ret = 0;
 862
 863         if (!policy)
 864                 return 0;
 865
 866         if (!cpufreq_driver->get)
 867                 goto out;
 868
 869         mutex_lock(&policy->lock);
 870
 871         ret = cpufreq_driver->get(cpu);
 872
 873         if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS))
 874         {
 875                 /* verify no discrepancy between actual and saved value exists */
 876                 if (unlikely(ret != policy->cur)) {
 877                         cpufreq_out_of_sync(cpu, policy->cur, ret);
 878                         schedule_work(&policy->update);
 879                 }
 880         }
 881
 882         mutex_unlock(&policy->lock);
 883
 884  out:
 885         cpufreq_cpu_put(policy);
 886
 887         return (ret);
 888 }
 889 EXPORT_SYMBOL(cpufreq_get);
 890
 891
 892 /**
 893  *      cpufreq_suspend - let the low level driver prepare for suspend
 894  */
 895
 896 static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
 897 {
 898         int cpu = sysdev->id;
 899         unsigned int ret = 0;
 900         unsigned int cur_freq = 0;
 901         struct cpufreq_policy *cpu_policy;
 902
 903         dprintk("resuming cpu %u\n", cpu);
 904
 905         if (!cpu_online(cpu))
 906                 return 0;
 907
 908         /* we may be lax here as interrupts are off. Nonetheless
 909          * we need to grab the correct cpu policy, as to check
 910          * whether we really run on this CPU.
 911          */
 912
 913         cpu_policy = cpufreq_cpu_get(cpu);
 914         if (!cpu_policy)
 915                 return -EINVAL;
 916
 917         /* only handle each CPU group once */
 918         if (unlikely(cpu_policy->cpu != cpu)) {
 919                 cpufreq_cpu_put(cpu_policy);
 920                 return 0;
 921         }
 922
 923         if (cpufreq_driver->suspend) {
 924                 ret = cpufreq_driver->suspend(cpu_policy, pmsg);
 925                 if (ret) {
 926                         printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
 927                                         "step on CPU %u\n", cpu_policy->cpu);
 928                         cpufreq_cpu_put(cpu_policy);
 929                         return ret;
 930                 }
 931         }
 932
 933
 934         if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
 935                 goto out;
 936
 937         if (cpufreq_driver->get)
 938                 cur_freq = cpufreq_driver->get(cpu_policy->cpu);
 939
 940         if (!cur_freq || !cpu_policy->cur) {
 941                 printk(KERN_ERR "cpufreq: suspend failed to assert current "
 942                        "frequency is what timing core thinks it is.\n");
 943                 goto out;
 944         }
 945
 946         if (unlikely(cur_freq != cpu_policy->cur)) {
 947                 struct cpufreq_freqs freqs;
 948
 949                 if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
 950                         dprintk(KERN_DEBUG "Warning: CPU frequency is %u, "
 951                                "cpufreq assumed %u kHz.\n",
 952                                cur_freq, cpu_policy->cur);
 953
 954                 freqs.cpu = cpu;
 955                 freqs.old = cpu_policy->cur;
 956                 freqs.new = cur_freq;
 957
 958                 notifier_call_chain(&cpufreq_transition_notifier_list,
 959                                     CPUFREQ_SUSPENDCHANGE, &freqs);
 960                 adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
 961
 962                 cpu_policy->cur = cur_freq;
 963         }
 964
 965  out:
 966         cpufreq_cpu_put(cpu_policy);
 967         return 0;
 968 }
 969
 970 /**
 971  *      cpufreq_resume -  restore proper CPU frequency handling after resume
 972  *
 973  *      1.) resume CPUfreq hardware support (cpufreq_driver->resume())
 974  *      2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
 975  *      3.) schedule call cpufreq_update_policy() ASAP as interrupts are
 976  *          restored.
 977  */
 978 static int cpufreq_resume(struct sys_device * sysdev)
 979 {
 980         int cpu = sysdev->id;
 981         unsigned int ret = 0;
 982         struct cpufreq_policy *cpu_policy;
 983
 984         dprintk("resuming cpu %u\n", cpu);
 985
 986         if (!cpu_online(cpu))
 987                 return 0;
 988
 989         /* we may be lax here as interrupts are off. Nonetheless
 990          * we need to grab the correct cpu policy, as to check
 991          * whether we really run on this CPU.
 992          */
 993
 994         cpu_policy = cpufreq_cpu_get(cpu);
 995         if (!cpu_policy)
 996                 return -EINVAL;
 997
 998         /* only handle each CPU group once */
 999         if (unlikely(cpu_policy->cpu != cpu)) {
1000                 cpufreq_cpu_put(cpu_policy);
1001                 return 0;
1002         }
1003
1004         if (cpufreq_driver->resume) {
1005                 ret = cpufreq_driver->resume(cpu_policy);
1006                 if (ret) {
1007                         printk(KERN_ERR "cpufreq: resume failed in ->resume "
1008                                         "step on CPU %u\n", cpu_policy->cpu);
1009                         cpufreq_cpu_put(cpu_policy);
1010                         return ret;
1011                 }
1012         }
1013
1014         if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
1015                 unsigned int cur_freq = 0;
1016
1017                 if (cpufreq_driver->get)
1018                         cur_freq = cpufreq_driver->get(cpu_policy->cpu);
1019
1020                 if (!cur_freq || !cpu_policy->cur) {
1021                         printk(KERN_ERR "cpufreq: resume failed to assert "
1022                                         "current frequency is what timing core "
1023                                         "thinks it is.\n");
1024                         goto out;
1025                 }
1026
1027                 if (unlikely(cur_freq != cpu_policy->cur)) {
1028                         struct cpufreq_freqs freqs;
1029
1030                         if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
1031                                 dprintk(KERN_WARNING "Warning: CPU frequency"
1032                                        "is %u, cpufreq assumed %u kHz.\n",
1033                                        cur_freq, cpu_policy->cur);
1034
1035                         freqs.cpu = cpu;
1036                         freqs.old = cpu_policy->cur;
1037                         freqs.new = cur_freq;
1038
1039                         notifier_call_chain(&cpufreq_transition_notifier_list,
1040                                         CPUFREQ_RESUMECHANGE, &freqs);
1041                         adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
1042
1043                         cpu_policy->cur = cur_freq;
1044                 }
1045         }
1046
1047 out:
1048         schedule_work(&cpu_policy->update);
1049         cpufreq_cpu_put(cpu_policy);
1050         return ret;
1051 }
1052
1053 static struct sysdev_driver cpufreq_sysdev_driver = {
1054         .add            = cpufreq_add_dev,
1055         .remove         = cpufreq_remove_dev,
1056         .suspend        = cpufreq_suspend,
1057         .resume         = cpufreq_resume,
1058 };
1059
1060
1061 /*********************************************************************
1062  *                     NOTIFIER LISTS INTERFACE                      *
1063  *********************************************************************/
1064
1065 /**
1066  *      cpufreq_register_notifier - register a driver with cpufreq
1067  *      @nb: notifier function to register
1068  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1069  *
1070  *      Add a driver to one of two lists: either a list of drivers that
1071  *      are notified about clock rate changes (once before and once after
1072  *      the transition), or a list of drivers that are notified about
1073  *      changes in cpufreq policy.
1074  *
1075  *      This function may sleep, and has the same return conditions as
1076  *      notifier_chain_register.
1077  */
1078 int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
1079 {
1080         int ret;
1081
1082         down_write(&cpufreq_notifier_rwsem);
1083         switch (list) {
1084         case CPUFREQ_TRANSITION_NOTIFIER:
1085                 ret = notifier_chain_register(&cpufreq_transition_notifier_list, nb);
1086                 break;
1087         case CPUFREQ_POLICY_NOTIFIER:
1088                 ret = notifier_chain_register(&cpufreq_policy_notifier_list, nb);
1089                 break;
1090         default:
1091                 ret = -EINVAL;
1092         }
1093         up_write(&cpufreq_notifier_rwsem);
1094
1095         return ret;
1096 }
1097 EXPORT_SYMBOL(cpufreq_register_notifier);
1098
1099
1100 /**
1101  *      cpufreq_unregister_notifier - unregister a driver with cpufreq
1102  *      @nb: notifier block to be unregistered
1103  *      @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
1104  *
1105  *      Remove a driver from the CPU frequency notifier list.
1106  *
1107  *      This function may sleep, and has the same return conditions as
1108  *      notifier_chain_unregister.
1109  */
1110 int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
1111 {
1112         int ret;
1113
1114         down_write(&cpufreq_notifier_rwsem);
1115         switch (list) {
1116         case CPUFREQ_TRANSITION_NOTIFIER:
1117                 ret = notifier_chain_unregister(&cpufreq_transition_notifier_list, nb);
1118                 break;
1119         case CPUFREQ_POLICY_NOTIFIER:
1120                 ret = notifier_chain_unregister(&cpufreq_policy_notifier_list, nb);
1121                 break;
1122         default:
1123                 ret = -EINVAL;
1124         }
1125         up_write(&cpufreq_notifier_rwsem);
1126
1127         return ret;
1128 }
1129 EXPORT_SYMBOL(cpufreq_unregister_notifier);
1130
1131
1132 /*********************************************************************
1133  *                              GOVERNORS                            *
1134  *********************************************************************/
1135
1136
1137 int __cpufreq_driver_target(struct cpufreq_policy *policy,
1138                             unsigned int target_freq,
1139                             unsigned int relation)
1140 {
1141         int retval = -EINVAL;
1142
1143         lock_cpu_hotplug();
1144         dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
1145                 target_freq, relation);
1146         if (cpu_online(policy->cpu) && cpufreq_driver->target)
1147                 retval = cpufreq_driver->target(policy, target_freq, relation);
1148
1149         unlock_cpu_hotplug();
1150
1151         return retval;
1152 }
1153 EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
1154
1155 int cpufreq_driver_target(struct cpufreq_policy *policy,
1156                           unsigned int target_freq,
1157                           unsigned int relation)
1158 {
1159         int ret;
1160
1161         policy = cpufreq_cpu_get(policy->cpu);
1162         if (!policy)
1163                 return -EINVAL;
1164
1165         mutex_lock(&policy->lock);
1166
1167         ret = __cpufreq_driver_target(policy, target_freq, relation);
1168
1169         mutex_unlock(&policy->lock);
1170
1171         cpufreq_cpu_put(policy);
1172
1173         return ret;
1174 }
1175 EXPORT_SYMBOL_GPL(cpufreq_driver_target);
1176
1177
1178 static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
1179 {
1180         int ret;
1181
1182         if (!try_module_get(policy->governor->owner))
1183                 return -EINVAL;
1184
1185         dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
1186         ret = policy->governor->governor(policy, event);
1187
1188         /* we keep one module reference alive for each CPU governed by this CPU */
1189         if ((event != CPUFREQ_GOV_START) || ret)
1190                 module_put(policy->governor->owner);
1191         if ((event == CPUFREQ_GOV_STOP) && !ret)
1192                 module_put(policy->governor->owner);
1193
1194         return ret;
1195 }
1196
1197
1198 int cpufreq_governor(unsigned int cpu, unsigned int event)
1199 {
1200         int ret = 0;
1201         struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1202
1203         if (!policy)
1204                 return -EINVAL;
1205
1206         mutex_lock(&policy->lock);
1207         ret = __cpufreq_governor(policy, event);
1208         mutex_unlock(&policy->lock);
1209
1210         cpufreq_cpu_put(policy);
1211
1212         return ret;
1213 }
1214 EXPORT_SYMBOL_GPL(cpufreq_governor);
1215
1216
1217 int cpufreq_register_governor(struct cpufreq_governor *governor)
1218 {
1219         struct cpufreq_governor *t;
1220
1221         if (!governor)
1222                 return -EINVAL;
1223
1224         mutex_lock(&cpufreq_governor_mutex);
1225
1226         list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
1227                 if (!strnicmp(governor->name,t->name,CPUFREQ_NAME_LEN)) {
1228                         mutex_unlock(&cpufreq_governor_mutex);
1229                         return -EBUSY;
1230                 }
1231         }
1232         list_add(&governor->governor_list, &cpufreq_governor_list);
1233
1234         mutex_unlock(&cpufreq_governor_mutex);
1235
1236         return 0;
1237 }
1238 EXPORT_SYMBOL_GPL(cpufreq_register_governor);
1239
1240
1241 void cpufreq_unregister_governor(struct cpufreq_governor *governor)
1242 {
1243         if (!governor)
1244                 return;
1245
1246         mutex_lock(&cpufreq_governor_mutex);
1247         list_del(&governor->governor_list);
1248         mutex_unlock(&cpufreq_governor_mutex);
1249         return;
1250 }
1251 EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
1252
1253
1254
1255 /*********************************************************************
1256  *                          POLICY INTERFACE                         *
1257  *********************************************************************/
1258
1259 /**
1260  * cpufreq_get_policy - get the current cpufreq_policy
1261  * @policy: struct cpufreq_policy into which the current cpufreq_policy is written
1262  *
1263  * Reads the current cpufreq policy.
1264  */
1265 int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
1266 {
1267         struct cpufreq_policy *cpu_policy;
1268         if (!policy)
1269                 return -EINVAL;
1270
1271         cpu_policy = cpufreq_cpu_get(cpu);
1272         if (!cpu_policy)
1273                 return -EINVAL;
1274
1275         mutex_lock(&cpu_policy->lock);
1276         memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
1277         mutex_unlock(&cpu_policy->lock);
1278
1279         cpufreq_cpu_put(cpu_policy);
1280
1281         return 0;
1282 }
1283 EXPORT_SYMBOL(cpufreq_get_policy);
1284
1285
1286 static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
1287 {
1288         int ret = 0;
1289
1290         cpufreq_debug_disable_ratelimit();
1291         dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
1292                 policy->min, policy->max);
1293
1294         memcpy(&policy->cpuinfo,
1295                &data->cpuinfo,
1296                sizeof(struct cpufreq_cpuinfo));
1297
1298         /* verify the cpu speed can be set within this limit */
1299         ret = cpufreq_driver->verify(policy);
1300         if (ret)
1301                 goto error_out;
1302
1303         down_read(&cpufreq_notifier_rwsem);
1304
1305         /* adjust if necessary - all reasons */
1306         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_ADJUST,
1307                             policy);
1308
1309         /* adjust if necessary - hardware incompatibility*/
1310         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_INCOMPATIBLE,
1311                             policy);
1312
1313         /* verify the cpu speed can be set within this limit,
1314            which might be different to the first one */
1315         ret = cpufreq_driver->verify(policy);
1316         if (ret) {
1317                 up_read(&cpufreq_notifier_rwsem);
1318                 goto error_out;
1319         }
1320
1321         /* notification of the new policy */
1322         notifier_call_chain(&cpufreq_policy_notifier_list, CPUFREQ_NOTIFY,
1323                             policy);
1324
1325         up_read(&cpufreq_notifier_rwsem);
1326
1327         data->min    = policy->min;
1328         data->max    = policy->max;
1329
1330         dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
1331
1332         if (cpufreq_driver->setpolicy) {
1333                 data->policy = policy->policy;
1334                 dprintk("setting range\n");
1335                 ret = cpufreq_driver->setpolicy(policy);
1336         } else {
1337                 if (policy->governor != data->governor) {
1338                         /* save old, working values */
1339                         struct cpufreq_governor *old_gov = data->governor;
1340
1341                         dprintk("governor switch\n");
1342
1343                         /* end old governor */
1344                         if (data->governor)
1345                                 __cpufreq_governor(data, CPUFREQ_GOV_STOP);
1346
1347                         /* start new governor */
1348                         data->governor = policy->governor;
1349                         if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
1350                                 /* new governor failed, so re-start old one */
1351                                 dprintk("starting governor %s failed\n", data->governor->name);
1352                                 if (old_gov) {
1353                                         data->governor = old_gov;
1354                                         __cpufreq_governor(data, CPUFREQ_GOV_START);
1355                                 }
1356                                 ret = -EINVAL;
1357                                 goto error_out;
1358                         }
1359                         /* might be a policy change, too, so fall through */
1360                 }
1361                 dprintk("governor: change or update limits\n");
1362                 __cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
1363         }
1364
1365  error_out:
1366         cpufreq_debug_enable_ratelimit();
1367         return ret;
1368 }
1369
1370 /**
1371  *      cpufreq_set_policy - set a new CPUFreq policy
1372  *      @policy: policy to be set.
1373  *
1374  *      Sets a new CPU frequency and voltage scaling policy.
1375  */
1376 int cpufreq_set_policy(struct cpufreq_policy *policy)
1377 {
1378         int ret = 0;
1379         struct cpufreq_policy *data;
1380
1381         if (!policy)
1382                 return -EINVAL;
1383
1384         data = cpufreq_cpu_get(policy->cpu);
1385         if (!data)
1386                 return -EINVAL;
1387
1388         /* lock this CPU */
1389         mutex_lock(&data->lock);
1390
1391         ret = __cpufreq_set_policy(data, policy);
1392         data->user_policy.min = data->min;
1393         data->user_policy.max = data->max;
1394         data->user_policy.policy = data->policy;
1395         data->user_policy.governor = data->governor;
1396
1397         mutex_unlock(&data->lock);
1398         cpufreq_cpu_put(data);
1399
1400         return ret;
1401 }
1402 EXPORT_SYMBOL(cpufreq_set_policy);
1403
1404
1405 /**
1406  *      cpufreq_update_policy - re-evaluate an existing cpufreq policy
1407  *      @cpu: CPU which shall be re-evaluated
1408  *
1409  *      Usefull for policy notifiers which have different necessities
1410  *      at different times.
1411  */
1412 int cpufreq_update_policy(unsigned int cpu)
1413 {
1414         struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
1415         struct cpufreq_policy policy;
1416         int ret = 0;
1417
1418         if (!data)
1419                 return -ENODEV;
1420
1421         mutex_lock(&data->lock);
1422
1423         dprintk("updating policy for CPU %u\n", cpu);
1424         memcpy(&policy,
1425                data,
1426                sizeof(struct cpufreq_policy));
1427         policy.min = data->user_policy.min;
1428         policy.max = data->user_policy.max;
1429         policy.policy = data->user_policy.policy;
1430         policy.governor = data->user_policy.governor;
1431
1432         /* BIOS might change freq behind our back
1433           -> ask driver for current freq and notify governors about a change */
1434         if (cpufreq_driver->get) {
1435                 policy.cur = cpufreq_driver->get(cpu);
1436                 if (data->cur != policy.cur)
1437                         cpufreq_out_of_sync(cpu, data->cur, policy.cur);
1438         }
1439
1440         ret = __cpufreq_set_policy(data, &policy);
1441
1442         mutex_unlock(&data->lock);
1443
1444         cpufreq_cpu_put(data);
1445         return ret;
1446 }
1447 EXPORT_SYMBOL(cpufreq_update_policy);
1448
1449 static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
1450                                         unsigned long action, void *hcpu)
1451 {
1452         unsigned int cpu = (unsigned long)hcpu;
1453         struct cpufreq_policy *policy;
1454         struct sys_device *sys_dev;
1455
1456         sys_dev = get_cpu_sysdev(cpu);
1457
1458         if (sys_dev) {
1459                 switch (action) {
1460                 case CPU_ONLINE:
1461                         cpufreq_add_dev(sys_dev);
1462                         break;
1463                 case CPU_DOWN_PREPARE:
1464                         /*
1465                          * We attempt to put this cpu in lowest frequency
1466                          * possible before going down. This will permit
1467                          * hardware-managed P-State to switch other related
1468                          * threads to min or higher speeds if possible.
1469                          */
1470                         policy = cpufreq_cpu_data[cpu];
1471                         if (policy) {
1472                                 cpufreq_driver_target(policy, policy->min,
1473                                                 CPUFREQ_RELATION_H);
1474                         }
1475                         break;
1476                 case CPU_DEAD:
1477                         cpufreq_remove_dev(sys_dev);
1478                         break;
1479                 }
1480         }
1481         return NOTIFY_OK;
1482 }
1483
1484 static struct notifier_block cpufreq_cpu_notifier =
1485 {
1486     .notifier_call = cpufreq_cpu_callback,
1487 };
1488
1489 /*********************************************************************
1490  *               REGISTER / UNREGISTER CPUFREQ DRIVER                *
1491  *********************************************************************/
1492
1493 /**
1494  * cpufreq_register_driver - register a CPU Frequency driver
1495  * @driver_data: A struct cpufreq_driver containing the values#
1496  * submitted by the CPU Frequency driver.
1497  *
1498  *   Registers a CPU Frequency driver to this core code. This code
1499  * returns zero on success, -EBUSY when another driver got here first
1500  * (and isn't unregistered in the meantime).
1501  *
1502  */
1503 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
1504 {
1505         unsigned long flags;
1506         int ret;
1507
1508         if (!driver_data || !driver_data->verify || !driver_data->init ||
1509             ((!driver_data->setpolicy) && (!driver_data->target)))
1510                 return -EINVAL;
1511
1512         dprintk("trying to register driver %s\n", driver_data->name);
1513
1514         if (driver_data->setpolicy)
1515                 driver_data->flags |= CPUFREQ_CONST_LOOPS;
1516
1517         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1518         if (cpufreq_driver) {
1519                 spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1520                 return -EBUSY;
1521         }
1522         cpufreq_driver = driver_data;
1523         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1524
1525         ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
1526
1527         if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
1528                 int i;
1529                 ret = -ENODEV;
1530
1531                 /* check for at least one working CPU */
1532                 for (i=0; i<NR_CPUS; i++)
1533                         if (cpufreq_cpu_data[i])
1534                                 ret = 0;
1535
1536                 /* if all ->init() calls failed, unregister */
1537                 if (ret) {
1538                         dprintk("no CPU initialized for driver %s\n", driver_data->name);
1539                         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1540
1541                         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1542                         cpufreq_driver = NULL;
1543                         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1544                 }
1545         }
1546
1547         if (!ret) {
1548                 register_cpu_notifier(&cpufreq_cpu_notifier);
1549                 dprintk("driver %s up and running\n", driver_data->name);
1550                 cpufreq_debug_enable_ratelimit();
1551         }
1552
1553         return (ret);
1554 }
1555 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
1556
1557
1558 /**
1559  * cpufreq_unregister_driver - unregister the current CPUFreq driver
1560  *
1561  *    Unregister the current CPUFreq driver. Only call this if you have
1562  * the right to do so, i.e. if you have succeeded in initialising before!
1563  * Returns zero if successful, and -EINVAL if the cpufreq_driver is
1564  * currently not initialised.
1565  */
1566 int cpufreq_unregister_driver(struct cpufreq_driver *driver)
1567 {
1568         unsigned long flags;
1569
1570         cpufreq_debug_disable_ratelimit();
1571
1572         if (!cpufreq_driver || (driver != cpufreq_driver)) {
1573                 cpufreq_debug_enable_ratelimit();
1574                 return -EINVAL;
1575         }
1576
1577         dprintk("unregistering driver %s\n", driver->name);
1578
1579         sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
1580         unregister_cpu_notifier(&cpufreq_cpu_notifier);
1581
1582         spin_lock_irqsave(&cpufreq_driver_lock, flags);
1583         cpufreq_driver = NULL;
1584         spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
1585
1586         return 0;
1587 }
1588 EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);