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